Thallium sorption and speciation in soils: Role of micaceous clay minerals and manganese oxides

Sorption processes control the solubility of toxic thallium (Tl) in soils and thereby its potential leaching into groundwater or uptake by plants. Micaceous clay minerals and Mn oxides are considered to be key sorbents for Tl in soils. We studied the sorption and speciation of Tl in 36 geogenically Tl-rich topsoil materials from the Swiss Jura Mountains by combining chemical extractions, isotope exchange experiments, adsorption experiments, X-ray absorption spectroscopy (XAS), and sorption modelling. We demonstrate that the relation between exchangeable and soluble Tl determined in batch extractions of soils with only geogenic and with freshly spiked Tl matches adsorption isotherms of freshly spiked Tl, and that this relation can be described with a published 3-site cation exchange model for Tl adsorption onto illite. Complemented with XAS data, the results show that micaceous clay minerals control the short-term solubility of Tl via cation exchange, but also the long-term sequestration of most geogenic soil Tl (>90%) via structural ﬁxation. Adsorption competition with K + and NH 4+ at the frayed edges of micaceous clay minerals greatly aﬀects Tl solubility. Increases in the dissolved concentrations of K and NH 4 in soil pore water may therefore lead to the release of Tl into solution. The fractions of geogenic Tl associated with Mn oxides were about half as high as the fractions of exchangeable Tl. This Mn-associated Tl is not readily mobilized by cation exchange, but could be released during periodic water logging and soil reduction. In (periodically) reducing environments, the potential of Mn oxides for long-term Tl sequestration is therefore limited. In conclusion, the results from this study high-light the importance of micaceous clay minerals for Tl cycling in soils and sediments, and suggest that concepts developed to assess the sorption of (radio)caesium onto micaceous clay minerals in soils and sediments are transferable to Tl.

The mobility and bioavailability of Tl in soils is controlled by its speciation and sorption onto soil components. Therefore, qualitative and quantitative information on the speciation of Tl and on key sorption processes is essential to assess risks arising from Tl contaminated soils. In the common pH-Eh range of soils, Tl mainly occurs as relatively well-soluble monovalent Tl(I), and to a lesser extent as highly insoluble trivalent Tl(III) (Nriagu, 1998;Peter and Viraraghavan, 2005). The Tl + cation has a similar ionic radius and a similar low hydration enthalpy as the alkali metal cations potassium (K + ), rubidium (Rb + ), and caesium (Cs + ). In analogy to Cs and Rb, Tl can therefore substitute K in minerals such as K-feldspars or replace K in the interlayer of mica (Heinrichs et al., 1980;Jović, 1993;Nriagu, 1998;Shannon, 1976;Wick et al., 2018).
Clay minerals have long been postulated to be key sorbents for Tl(I) in soils and sediments, based on the alkali metal-like behaviour of Tl(I) (Matthews and Riley, 1970;Tremel et al., 1997b). As for Cs and Rb (Cremers et al., 1988;Evans et al., 1983;Sawhney, 1972), especially micaceous clay minerals like illite are commonly assumed to strongly bind Tl in soils (Jacobson et al., 2005a;Nriagu, 1998;Tremel et al., 1997b). This assumption was confirmed in a recent spectroscopic study on geogenically Tl-rich soils (Voegelin et al., 2015), which indicated that Tl in topsoil materials was dominantly Tl(I) associated with illite (or with other micaceous clay minerals). Illite is a nonexpandable 2:1 phyllosilicate with a basal spacing of 10 Å and dehydrated K + in the collapsed interlayer. In soils, illite is either inherited from sedimentary parent material or formed by the weathering of muscovite (Meunier and Velde, 2004). Weathering involves the release of K at particles edges, resulting in frayed particle edges with an extended basal spacing of 14 Å (Fuller et al., 2015;Jackson, 1968;Meunier and Velde, 2004). At the wedge between the collapsed interlayer and the frayed edges of illite platelets, the alkali metal cations Cs + and Rb + adsorb with a very high affinity and in competition with K + , ammonium (NH 4 + ), and, to some extent, barium (Ba 2+ ) (Brouwer et al., 1983;Comans et al., 1991;Lee et al., 2017;Meunier and Velde, 2004;Poinssot et al., 1999). The competitive sorption of Cs and Rb onto illite has been successfully modelled using a 3-site cation exchange model (Bradbury and Baeyens, 2000;Poinssot et al., 1999). A low fraction of high-affinity sorption sites (frayed edge sites, FES) and an intermediate fraction of medium-affinity sorp-tion sites (type-2 sites, T2S) account for the specific adsorption of Cs and Rb at the frayed edges of illite platelets, and a large fraction of low-affinity cation exchange sites for non-specific adsorption of Cs and Rb on the planar surfaces of illite (planar sites, PS). In a recent study on the adsorption of Tl onto purified illite, we showed that the adsorption of Tl + onto purified Na-, Ca-, K-and NH 4 -exchanged illite over wide Tl concentration ranges can be adequately described with the 3-site cation exchange model (Wick et al., 2018). The corresponding cation exchange selectivitiy coefficients showed that Tl adsorbs very strongly onto illite (as confirmed in another recent sorption study (Martin et al., 2018)), with an adsorption affinity between the affinities of Cs and Rb. X-ray absorption near edge structure (XANES) spectra indicated that FES and T2S model sites served to describe the adsorption of (dehydrated) Tl at illite particle edges (Wick et al., 2018). The parameterized cation exchange model allows describing the adsorption of Tl onto purified illite in the presence of competing cations. To date, however, the ability of this model to predict the adsorption of Tl onto illite or, more generally, onto micaceous clay minerals in soils has not been evaluated.
Manganese (Mn) oxides can accumulate high levels of Tl in marine and terrestrial deposits (Crittenden et al., 1962;Koschinsky and Hein, 2003;Nriagu, 1998;Peacock and Moon, 2012;Rehkämper and Nielsen, 2004), and may also be important for Tl retention in soils (Tremel et al., 1997b;Vaněk et al., 2011;Voegelin et al., 2015). In soils, Mn oxides form by microbial or autocatalytic oxidation of Mn 2+ (Gilkes and McKenzie, 1988;Tebo et al., 2004) and can occur as local accumulations referred to as Mn-concretions (Blume et al., 2016). Strong Tl sorption by Mn oxides has been attributed to three uptake mechanisms: (i) Oxidation of Tl(I) and complexation of Tl(III) by vacancy-containing birnessite (Cruz-Hernández et al., 2019;Peacock and Moon, 2012;Wick et al., 2019), (ii) sorption of hydrated Tl(I) by, e.g., triclinic birnessite or todorokite (Peacock and Moon, 2012;Wick et al., 2019), or (iii) structural incorporation of dehydrated Tl + , for example in cryptomelane (Crittenden et al., 1962;Wick et al., 2019). Laboratory sorption experiments have shown that oxidative Tl scavenging by vacancy-containing d-Mn(IV)O 2 , a synthetic analogue of biogenic Mn oxides (Post, 1999;Tebo et al., 2004), can lead to very high Tl loadings at very low dissolved Tl concentrations (Wick et al., 2019). The partial reduction and structural transformation of fresh d-MnO 2 inhibits further oxidative Tl uptake (Wick et al., 2019). Nevertheless, partly-reduced and/or transformed d-MnO 2 , as well as triclinic birnessite, todorokite, and cryptomelane also exhibit a high sorption affinity for Tl(I) relative to illite (Wick et al., 2019). Birnessite additions to Tl-containing soil have been shown to effectively reduce Tl uptake by white mustard (Vaněk et al., 2011), and significant fractions of soil Tl were observed to be extractable in the reductive step of sequential extractions supposed to dissolve Mn oxides (Antić-Mladenović et al., 2017;Gomez-Gonzalez et al., 2015;Vaněk et al., 2009;Vaněk et al., 2011;Yang et al., 2005). These results thus point to the relevance of Mn oxides for Tl retention in soils. In a spectroscopic study on geogenically Tl-rich soils, Tl(III) associated with soil Mn concretions was identified by point analyses. Bulk analyses on corresponding topsoil samples, on the other hand, indicated that most geogenic Tl was Tl(I) associated with micaceous clay minerals (Voegelin et al., 2015). Factors that can limit the relevance of Mn oxides for Tl retention in soils include their usually low abundance relative to micaceous clay minerals, their susceptibility to reductive transformation or dissolution (Wick et al., 2019), and strong sorption competition with other major and trace metal cations (Wick et al., 2019). To date, the importance of Mn oxides for Tl retention in soils still remains uncertain.
The aim of the present study is to gain new insights into the relevance of micaceous clay minerals and Mn oxides for Tl sorption in soils and a better understanding of their impacts on dissolved Tl concentrations. We therefore studied the sorption and speciation of Tl in geogenically Tl-rich topsoils formed on mineralized dolomitic rocks in the Swiss Jura mountains (Herrmann et al., 2018;Truninger, 1922;Voegelin et al., 2015). We used XANES spectroscopy to characterize Tl speciation, chemical extractions and isotopic exchange to quantify the pool of exchangeable Tl, and sequential extractions to estimate the fraction of Tl associated with Mn oxides. The relationship between exchangeable and soluble Tl in geogenically Tl-rich soils and adsorption isotherms determined on freshly Tl-spiked soils were predicted and interpreted using the cation exchange model for Tl sorption onto illite (Wick et al., 2018) and sorption isotherms for Tl sorption onto Mn oxides (Wick et al., 2019).

Sampling site
The locality Erzmatt (''ore meadow") is a meadow near the village of Buus in the Swiss Jura Mountains (47°3 0 0 20 00 N, 7°51 0 07 00 E). The soils on the Erzmatt show high levels of Tl, arsenic (As), and iron (Fe) of geogenic origin (Truninger, 1922;Voegelin et al., 2015). The soil parent material at the site consists of dolomitized sediments (Stamberg member (former name: Trigonodus dolomite), Ladinian as a part of the Schinznach formation (former name Upper Muschelkalk), Triassic) (Adams and Diamond, 2019;Pietsch et al., 2016;Schauer and Aigner, 1997) that are locally mineralized (Truninger, 1922). This hydrothermal Tl-As-Fe mineralization is most probably linked to Upper Rhine Graben tectonics (Gonzalez, 1990) that triggered uprising fluids during Oligocene and Miocene time. Still active hot springs in the northern Upper Rhine Graben (Germany) may be considered as a recent analogue of such hydrothermal systems (Loges et al., 2012). The observation of Tl-and As-bearing jarosite and avicennite in soil samples from the Erzmatt suggest that the unweathered ore body consisted of Tl-As-Fe sulfides (Herrmann et al., 2018;Truninger, 1922;Voegelin et al., 2015). The Erzmatt was not covered by ice during the last glacial maximum, suggesting that soils at the site may be more than 10,000 years old. The soils belong to the Cambisol group, exhibit weak redoximorphic features and have a neutral to slightly acidic pH (Fig. S1). The climate in the area is temperate (mean annual air temperature approx. 9°C; mean annual precipitation approx. 960 mm) (Voegelin et al., 2015).

Soil sampling
For the present study, topsoil samples from the Erzmatt were kindly provided by the cantonal authorities (Amt fü r Umwelt und Energie (AUE), Kanton Basel-Landschaft), which performed an extensive monitoring campaign between 2012 and 2014 to determine the extension of the geochemical anomaly on the Erzmatt. At about 150 sites, four topsoil cores (0-20 cm depth) were taken, pooled, and air-dried at 40°C. The pooled samples were sieved to 2 mm and stored in polyethylene (PE) containers. Based on acid-extractable element contents and soil pH values determined by the cantonal authorities, 36 topsoil samples were selected for this study. These samples cover geogenic Tl contents from $3 to $1000 mg/kg and also vary with respect to Mn, Fe, calcium (Ca), and magnesium (Mg) contents as well as pH values. Based on acid-extractable Tl contents, the 36 topsoil samples were arranged in six groups of six samples each (A1-A6 with $3 mg/kg, B1-B6 with $10 mg/kg, C1-C6 with $30 mg/kg, D1-D6 with $100 mg/kg, E1-E6 with $300 mg/kg, and F1-F6 with $1000 mg/kg Tl).
A topsoil sample from an earlier study on the Erzmatt (P1 00-20; 100 mg/kg Tl (Voegelin et al., 2015)) was used to extract the clay mineral fraction. First organic matter was decomposed (hydrogen peroxide, 80°C), followed by the removal of Fe and Mn oxides (dithionite, 80°C). After sodium (Na) exchange followed by dialysis, the clay fraction was separated by centrifugation. Subsequently, the clay minerals were flocculated with Na, concentrated by settling, and stored as suspension in the dark at 4°C.

Bulk soil analyses
Total element contents were determined using an energy-dispersive X-ray fluorescence spectrometer (XRF; Xepos+, SPECTRO Analytical Instruments GmbH, Germany) with a built-in calibration for geological samples. For XRF analysis, 4.0 g of soil material was ground in a ball mill, mixed with wax and pressed into a 32-mm diameter pellet. Total contents of carbon (C) and nitrogen (N) (CNS total element analyser, EuroVector) and the contents of inorganic C (Coulomat, UIC, Inc.) were measured, and the contents of organic C calculated as the difference between total and inorganic C contents. Soil texture was measured using a laser diffraction grain sizer (Hydro 2000S, Mastersizer 2000, Malvern Instruments Ltd, U. K.), after overnight dispersion of the soils in 0.1% calgon solution.
The mineralogy of six soil samples with different total Tl loadings (A6, B4, C5, D1, D6, E6) was determined on randomly oriented powder samples by X-ray diffraction (XRD) analysis (details in the Appendix). Samples were ground (<20 mm) and filled into frontloading sample holders. X-ray diffraction patterns were measured with a Bragg-Brentano diffractometer (BRUKER AXS D8) using Co K a radiation. The qualitative phase composition was determined with the software DIFFRAC.EVA (BRUKER AXS) and the quantitative determination of the mineralogical composition was carried out by Rietveld analysis using the Rietveld program Profex/BGMN Dö belin and Kleeberg, 2015;Omotoso et al., 2006).

Exchangeable cations
All soils were extracted with 1 M NH 4 -acetate (pH 6.8) to quantify the pools of exchangeable cations. For the extraction, 1.5 g of soil were weighed into 50 mL PE tubes and extracted in two consecutive steps with 12.5 mL 1 M NH 4 -acetate each (2 Â 30 min; horizontal table shaker; 350 rpm). After each step, the samples were centrifuged (5 min; 2700 g) and 10 mL of supernatant filtered (0.2 or 0.45 lm nylon membranes). Subsequently, the two filtered extracts were pooled and acidified (65% HNO 3 , 1% v/v). The combined extracts were diluted with 1% HNO 3 and analysed for Mg, K, Ba, Ca, Mn, Rb, Tl, and further elements using an inductively coupled plasma mass spectrometer (ICP-MS; Agilent 7500ce or Agilent Triple Quadrupole 8900). For each soil, the extraction was performed at least in duplicates and the determined amounts of exchangeable cations were averaged. To assess the completeness of the two-step extraction, six soil samples (A5, B2, C3, D5, E5, F5) were also extracted in four consecutive steps and the extracts from each step were analysed by ICP-MS individually. To determine exchangeable NH 4 + , 1.0 g of the soil materials was extracted in 5 mL of 2 M KCl for 30 min on a table shaker (350 rpm). After reaction, the suspensions were centrifuged and filtered (0.2 lm nylon membranes) and the filtrates analysed for NH 4 + using colorimetric photometry (indophenol blue method; Cary 50-3 photometer, or Hach-Lange cuvette test (LCK304)).

Pseudo-porewater extracts
All soils were extracted with 0.01 M Ca-chloride (CaCl 2 ) to determine the soil pH value and dissolved cation concentrations in pseudo-porewater (Houba et al., 2000). In 50 mL PE tubes, 2.0 g of soil were extracted with 20 mL of CaCl 2 (solid-to-solution ratio 0.1 kg/L) for 2 h (horizontal table shaker; 350 rpm). After extraction and centrifugation (5 min; 2700 g), 10 mL of the supernatant were filtered (0.2 or 0.45 lm nylon membranes) and acidified (65% HNO 3 , 1% v/v). The pH value was measured in the remaining re-suspended extract. The samples were diluted with 1% HNO 3 and analysed for Na, K, Ba, Mg, Ca, Mn, Rb, Tl, and further elements using ICP-MS. NH 4 + was determined by colorimetric photometry (Cary-50-3 and LCK304). Each soil was extracted at least in duplicates and cation concentrations were averaged. A subset of samples (A5, B2, C3, D5, E5, F5) was equilibrated with 0.01 M CaCl 2 four times in sequence to assess changes in dissolved Tl over multiple equilibrations.

Solid-liquid distribution coefficients for exchangeable geogenic Tl
Solid-liquid distribution coefficients for exchangeable geogenic Tl ( Tl K d,geo , in [L/kg]) were calculated according to Eq. (1): Q Tl,NH4 corresponds to the amount of Tl exchanged with NH 4 -acetate (in [mol/kg]) and C Tl,Ca to the concentration of Tl in the CaCl 2 pseudo-porewater extracts (in [mol/ L]). Distribution coefficients for other cations (K, Rb, Ba, Mg) were calculated analogously. From the distribution coefficients of monovalent Tl, Rb and K, cation exchange selectivity coefficients were calculated as follows (e.g. for Tl-Rb exchange in Eq. (2)):

Isotope exchange experiment
With six selected soils (A6, B4, C5, D1, D6, E6), a kinetic isotope exchange experiment with radioactive 204 Tl was performed in 0.01 M CaCl 2 electrolyte to estimate independently the pool of (exchangeable) Tl that interacts with the solution. For this purpose, 1.0 g of dried soil was suspended in 10 mL 0.01 M CaCl 2 in 40 mL polypropylene centrifuge tubes (Beckman Coulter). After 2 h of equilibration, 204 Tl was spiked to the suspensions (total $9 Â 10 À12 mol Tl added with a 204 Tl activity of $660 Bq; more information on the 204 Tl tracer is given in Wick et al., 2018). The samples were subsequently reacted for 4 h and 1, 3, 7, 14, 30, and 90 days (d). For sampling, the suspensions were centrifuged (5 min, 2700 g) and 2 mL of the supernatant were diluted with 3 mL 0.01 M CaCl 2 and 15 mL scintillation cocktail (UltimaGold X/R, Merck). Solution pH was measured in the remaining supernatant. The 204 Tl activity was analysed using a liquid scintillation counter (Packard Tri-Carb 2750 TRILL). From the decrease in the 204 Tl activity, the distribution coefficients Tl-204 K d (in [L/kg]) were calculated according to Eq. (3): C Tl-204,init and C Tl-204,final are the initial and final 204 Tl activity in solution, respectively, V the solution volume (in [L]) and m the mass of soil (in [kg]). Each experiment was run in duplicates, and the resulting Tl-204 K d values were averaged.
In parallel to the experiment with radioactive 204 Tl, the same experiment was also performed without 204 Tl tracer spike. Solutions from this inactive experiment were filtered, acidified and analysed for Na, Mg, K, Ca, Rb, and Tl using ICP-MS and photometrically (Cary 50-3) for NH 4 . The experiment was performed in duplicates and measured concentrations were averaged. From Tl concentrations measured in the inactive samples (C Tl,Ca (t), in [mol/L]) and the distributions coefficients of 204 Tl ( Tl-204 K d (t), in [L/kg]) after equilibration time t, the pools of radiolabile soil Tl (E Tl (t), in [mol/kg]) were calculated using Eq. (4): 2.8. Sequential soil extraction 18 soil samples (A1, A2, A6, B1, B2, B6, C1, C2, C6, D1, D2, D6, E1, E2, E6, F1, F2, and F6) were extracted in a two-step sequential extraction to quantify the pools of exchangeable cations and Tl associated with Mn oxides. For the extraction, 1.0 g of soil was weighed into a dark 50 mL polypropylene tube. In the first step, the soils were extracted three times with 30 mL 1 M NH 4 -acetate (pH 6.8) for 30 min. In the second step, the soils were extracted three times with 30 mL of 0.1 M hydroxylamine hydrochloride (NH 3 OH Â HCl) and 1 M NH 4 -acetate (pH 6.1) (adapted from Zeien and Brü mmer (1989)) for 30 min. In both steps, the individual extracts were collected by centrifugation, filtered, acidified and diluted as required for analysis by ICP-MS.

Tl adsorption / re-extraction experiments
Six soil samples (A6, B4, C5, D1, D6, and E6) with geogenic Tl contents of $3 to $350 mg/kg were selected for Tl adsorption / re-extraction experiments. The adsorption and re-extraction experiments were performed three times with subsets of soils, using nearly the same protocol with minor variations (see Appendix for details). In general, 2.0 g of dried soil were weighed into 50 mL polypropylene tubes and conditioned two times with 20 mL 0.01 M Ca(NO 3 ) 2 -Â 4H 2 O / 0.0001 M KNO 3 solution for 2 h, recording the pH and measuring element concentrations in the conditioning solutions. For Tl adsorption, 0.01 M Ca(NO 3 ) 2 solutions were spiked with seven Tl concentrations (from $3.5 Â 10 À7 to $4.5 Â 10 À4 M) using 0.01 M or 0.001 M TlNO 3 (Sigma Aldrich) stock solutions. The wet soil materials (2.0 g of soil and $1.6 g conditioning solution) were then suspended in 20 mL of the Tl-containing electrolyte solutions and reacted for seven days on a table shaker (350 rpm) at room temperature. After reaction, the suspensions were centrifuged and the supernatant filtered (0.2 mm nylon membrane) and acidified (1% v/v HNO 3 suprapur) for the analysis of Na, Mg, K, Ca, Rb, Ba and Tl using ICP-MS. The solution pH was recorded in the suspensions after solution sampling and resuspension. The residual wet soil was kept in the fridge in the dark. The difference between spiked and residual dissolved Tl was assumed to correspond to Tl adsorbed to the soil. The total amount of adsorbed Tl was calculated by adding the amounts of NH 4 -exchangeable geogenic Tl in the individual soils prior to Tl adsorption to the amounts of freshly sorbed Tl (based on NH 4 -acetate extraction of soils from sorption experiments without Tl spike).
After Tl adsorption, the reacted soils were first extracted with 20 mL 0.01 M CaCl 2 to determine pseudo-porewater concentrations (as described above for geogenic Tl). In a second step, the soils were extracted twice with 20 mL 1 M NH 4 -acetate to determine the amounts of NH 4 -exchangeable Tl (as described above for geogenic Tl). After sorption and between the two extraction steps, supernatants were decanted as completely as possible and the wet soils were stored at 4°C in the dark. Carryover of Tl or other elements from residual solution into the next extraction step and dilution of the extracts by residual solution was calculated from solution composition and the gravimetrically determined volume of residual solution.
Data extraction and normalization were performed with Athena (Ravel and Newville, 2005). A linear curve fit to the pre-edge region (À100 to À30 eV relative to E 0 set to 12,668 eV) was subtracted for background removal and a cubic function fit to the post-edge region (+40 to +280 eV relative to E 0 ) was used to normalize the edge step to unity (at E 0 ) and to flatten the post-edge region. Analysis of the normalized XANES spectra over the energy range 12,648-12,768 eV by linear combination fitting (LCF) was performed with software from beamline 10.3.2 at the Advance Light Source (Berkeley, USA) (Marcus et al., 2004). Out of a larger set of reference spectra (Fig. S4), four reference spectra were selected for LCF analysis, as further detailed in the Appendix: (i) Tl(I)-illite (illite with Tl loading of 3800 mg/kg Tl from Wick et al. (2018)), representing dehydrated Tl + adsorbed at the frayed edges of illite or other micaceous clay minerals, (ii) Tl(I)-muscovite (Ba-rich muscovite (''oellacherite") from Lengenbach (Switzerland) with $200 mg/kg Tl(I)), representing structural Tl in the interlayer of micaceous clay minerals, (iii) dissolved aqueous Tl + (0.01 M TlNO 3 ) as proxy for (hydrated) Tl + adsorbed onto clay minerals (Wick et al., 2018), Mn oxides such as triclinic birnessite or todorokite (Wick et al., 2019), soil organic matter or Fe oxides and (iv) Tl(III) sorbed onto d-MnO 2 (spectrum Na-d-MnO 2 with 0.021 Tl/Mn from Wick et al. (2019)) as proxy for Tl(III) sorbed onto Mn oxides (or in Tl 2 O 3 ). Starting with the best one-component fit, the number of fit components was increased as long as the normalized sum of the squared residuals (NSSR = P (data i À fit i ) 2 / P data i 2 ) of the best fit decreased by at least 20%.

Modelling of Tl adsorption onto micaceous clay minerals
The adsorption of Tl onto micaceous clay minerals was predicted for geogenic Tl in all studied soils as well as for freshly adsorbed Tl on the six soils used for the adsorption experiment. The 3-site cation exchange model for the adsorption of Tl, Rb, NH 4 , K, Na, and Ca onto illite (Wick et al., 2018) was implemented in the geochemical equilibrium code PhreeqC (Parkhurst and Appelo, 1999). For Mg, the same cation exchange coefficients as for Ca were assumed. For the solution speciation, the database Minteq V4 was used. For Tl(I), free Tl + and the complexes TlNO 3 , TlCl and TlCl 2 À were taken into account. Oxidation of Tl(I) to Tl(III) and precipitation of Tl(I)-phases were excluded in the calculations.
For geogenic Tl, a first calculation was run for different dissolved Tl concentrations in the CaCl 2 extracts, but at constant concentrations of Rb, NH 4 , K, Na, Mg, and Ca corresponding to the average (or the 5-percentile or 95percentile) of their concentrations in the CaCl 2 extracts from all soils. The cation concentrations were chargebalanced with nitrate (NO 3 À ). Comparison of model calculations with Cl À instead of NO 3 À as charge-balancing anion showed that results for free Tl + and sorbed Tl(I) varied by less than 2%; reflecting the fact that neither nitrate nor chloride form strong complexes with Tl + or the other modelled cations. The modelled cation loadings on illite were scaled by assuming an average content of illite and mica (dioctahedral micaceous phyllosilicates) of 16.8% for all soils (details in the Appendix). A second calculation for geogenic Tl was run with the individual pseudo-porewater cation concentrations for each soil charge-balanced by NO 3 À . The resulting cation loadings on illite were either scaled by the same average clay content, or by the clay content of each soil individually (details in the Appendix). For the calculation of the adsorption isotherms for the soils A6, B4, C5, D1, D6, and E6, the concentrations of Rb, NH 4 , K, Na, Mg, and Ca were derived from the individual pseudo-porewater extracts, and the clay content was set to the fraction of mica and illite determined by XRD analysis (details in the Appendix).

Tl sorption onto Mn/Fe-enriched and depleted soil material from an acidic pseudogley soil
To gain further insights into the Tl adsorption behaviour in soils and the potential role of Mn oxides therein, Tl adsorption experiments were also performed with Mn/Fe-enriched and Mn/Fe-depleted soil material from an acidic pseudogley soil (stagnosol group), in analogy to the sorption experiment with soils from the Erzmatt described above. Tl L III -edge XANES spectra of the Tl-sorbed soil materials were collected at the SAMBA beamline (Synchrotron Soleil, Gif-sur-Yvette, France). Details on the soil material (characterization) and on the Tl adsorption experiment are given in the Appendix.

Physicochemical soil characterization
In Table 1, selected physicochemical properties are listed for the six soil materials used for Tl adsorption reextraction experiments (A6, B4, C5, D1, D6, and E6) together with the average and range for all 36 topsoil samples. The complete characterization results are given in Tables S1-S3. The soil samples had a pH CaCl2 between 5.0 and 7.6, and contained between 1.0% and 8.4% organic C. Their Fe content ranged between $8000 and $72,000 mg/kg and their Mn content between $260 and $1700 mg/kg. The mineralogical analysis of the six samples used in the adsorption / re-extraction and isotope exchange experiments (Table S2) indicated that phyllosilicates (chlorite, illite, kaolinite, muscovite) accounted for 20-29% of the soil mass, and that micaceous clay minerals (illite and Table 1 Soil pH, organic and inorganic C content, clay size fraction, XRD results (illite, muscovite, sum of phyllosilicates), total Mn and Tl (XRF) and extractable element contents for the six soil samples used in the adsorption / re-extraction and isotope exchange experiments and the corresponding average values, standard deviations and ranges for all analyzed soil samples. muscovite) represented the dominant phyllosilicate fraction (16-20%). The non-clay fraction was dominated by quartz (32-42%), K-feldspar (11-26%), and variable proportions of dolomite.
The total Tl concentration in the soil samples ranged from 2.2 to 1030 mg/kg (Table S3). The correlation analysis of total element contents (Table S4) revealed a very close correlation between Tl and As (r = 0.99) confirming their common origin from a hydrothermal Tl-As-Fe-sulfide mineralization, which was also reflected in the correlation of Tl (and As) with Fe (r = 0.78), and Sr (r = 0.75). The correlation between Ca and Mg (r = 0.86) pointed to dolomite as the major Ca and Mg source. Finally, the correlations of Al with Si (r = 0.96), K (0.87), Ba (0.88), and Rb (0.93) was attributed to the presence of these elements in primary sili-cates (quartz, feldspar) and phyllosilicates inherited from the soil parent material.

Exchangeable cations
Exchangeable Tl determined by the two-step 1 M NH 4acetate extraction ranged from 0.05 to 55 mg/kg (Table 1). The complete extraction results are given in Table S5. Exchangeable Tl closely correlated with total soil Tl (r = 0.98 for log-transformed Tl contents; Fig. 1A), and on average 3.6 ± 1.6% of the total soil Tl were exchangeable. Regression analysis indicated that the fraction of exchangeable Tl slightly decreased with increasing pH ((Tl NH4 /Tl tot ) = (12.5-1.32 Â pH CaCl2 )%). Averaged over all soils, exchangeable Ca accounted for 78 ± 9%, Mg for   19 ± 8%, and K for 2 ± 1% of the sum of exchangeable charge equivalents.
A test with four consecutive 1 M NH 4 -acetate extraction steps performed with six selected soil samples showed that 2.6 ± 0.9% of the total Tl was extracted in the first step, 1.1 ± 0.3% in the second step, 0.5 ± 0.2% in the third step and 0.3 ± 0.1%, in the fourth step (Table S6). This indicated that the two-step extraction use for all soils extracted $83% of the exchangeable Tl.

Dissolved cations in the pseudo-porewater extracts
Dissolved concentrations of Tl in the CaCl 2 extracts of six selected soils as well as the average and range for all soils are listed in Table 1 (complete results given in Table S7). Dissolved Tl CaCl2 ranged from 0.1 to 270 mg/L and was closely correlated with total Tl (r = 0.97 for log-transformed data; Fig. 1B) and exchangeable Tl (r = 0.99 for logtransformed data; Fig. 1C).
Tests with six selected soils showed that over four sequential equilibrations in 0.01 M CaCl 2 , dissolved Tl concentrations in the fourth equilibration corresponded to 47 ± 7% of the concentrations in the first equilibration (Table S8). This decrease most probably reflects a gradual depletion of the most readily equilibrating fraction of the exchangeable Tl pool. Alternatively, the decrease could be linked to decreasing concentrations of dissolved K over repeated CaCl 2 extractions and a resulting decrease in sorption competition between K and Tl for more specific cation exchange sites.

NH 4 -exchangeable versus CaCl 2 -dissolved Tl
In Fig. 1C, log-transformed exchangeable Tl NH4 contents are plotted against log-transformed dissolved Tl CaCl2 for all 36 soil samples. In a broad sense, this plot can be interpreted as a (de)sorption isotherm for exchangeable geogenic Tl. Linear regression of this pseudo-isotherm returned a slope close to unity (y = 2.13 + 0.94x; R 2 = 0.96). As shown in Fig. 1D, the same data can be plotted as distribution coefficients ( Tl K d,geo ; Eq. (1)) versus dissolved Tl. This plot indicates a small decrease in the log ( Tl K d,geo /(L/kg)) over a nearly three orders of magnitude increase in Tl CaCl2 and total Tl. From all soils, an average Tl K d,geo of 2.60 ± 0.20 is calculated. K d,geo values for Rb, K, Ba, and Mg were calculated analogously to Tl K d,geo (Table 9). Over the 36 soils, a close correlation was observed between the log K d values of Tl and Rb (r = 0.82), K (r = 0.76) and Ba (r = 0.75), but only a low correlation with Mg (r = 0.35), indicating that exchangeable Tl strongly interacted with Rb as well as K and Ba, but hardly with Mg (Table S10).

Kinetic 204 Tl isotope exchange
The results of the isotope exchange experiments with six soils are displayed in Fig. 2 and listed in Table S11. On average, log ( Tl-204 K d /(L/kg)) values after 4 h equilibration with 204 Tl were 0.2 ± 0.1 higher than log ( Tl K d,geo /(L/kg)) values derived from NH 4 -acetate exchangeable and CaCl 2 -soluble Tl. Accordingly, the pools of radiolabile Tl after 4 h equilibration were about 1.6 ± 0.4 times higher than the pools of exchangeable Tl determined by the two step 1 M NH 4 -acetate extraction. This finding is partly in line with the observation that two extraction steps with 1 M NH 4 -acetate extracted up to 20% less Tl than four steps. Furthermore, it is possible that the 1 M NH 4acetate used in the extracts may have caused the partial collapse of the frayed edges of illite and the trapping of exchangeably adsorbed Tl, as has been suggested for Cs (de Koning and Comans, 2004).
Interestingly, NH 4 concentrations in the CaCl 2 -extracts of all six soils rapidly increased over the first seven days of equilibration and decreased again thereafter, most probably due to microbial ammonification (N-mineralization) that liberates NH 4 + (Blume et al., 2016). After 30 d of reaction, NH 4 concentrations in all soils were below 7 mmol/L. In parallel to these changes in dissolved NH 4 + concentrations, the concentrations of Tl and Rb in solution markedly increased and decreased again, pointing to their release by highly specific sorption competition with NH 4 + , in line with Cs + mobilization from sediments due to ammonification (Comans et al., 1989;Evans et al., 1983). In addition, the close co-evolution of dissolved Tl and Rb concentrations during equilibration showed that the same type of adsorption process affected both cations.
The log Tl-204 K d values decreased with increasing NH 4 + concentrations, before they increased again with disappearing dissolved NH 4 + . These coupled trends suggested that NH 4 + temporarily also displaced initially adsorbed 204 Tl. The log Tl-204 K d values after 90-d equilibration were 0.6 ± 0.1 higher than after 4-h equilibration, indicating that 204 Tl at the end of the experiment had become more strongly sorbed than initially. The dissolved Tl concentrations in the CaCl 2 -extracts after 90-d equilibration, on the other hand, were 69 ± 7% lower than after 4-h equilibration. Consequently, the pools of radiolabile Tl were on average only 1.2 ± 0.2 times higher after 90-d equilibration, suggesting that fixation processes were too slow to be clearly reflected after 90 d, possibly also because of the confounding impacts of the temporary increase in dissolved NH 4 + concentrations, which appeared to result in an increase in 204 Tl sorption strength. Cs studies also revealed strong effects of NH 4 + on sorption reversibility when concentrations exceed 10 À4 M NH 4 + (de Koning and Comans, 2004).

Sequential extraction of exchangeable Tl and reductively mobilizable Tl and Mn
From the sequential extraction of exchangeable Tl and reductively mobilizable Tl and Mn, the results for six of the 18 analysed soil samples and the averages over all analysed soil samples are given in Table 2, the complete data is given in Table S12. Both extractions were performed in three consecutive extraction steps. Decreasing concentrations of Tl over the three extracts with 1 M NH 4 -acetate (pH 6.8) indicated that the exchangeable pool was depleted (Fig. S2). The first extraction with 1 M NH 4 -acetate + 0.1 M NH 3 OH Â HCl (pH 6.1) resulted in a clear increase in the extracted Tl fraction. Since the pH was not much lower than in the proceeding steps, the increase in extracted Tl was attributed to the reductant hydroxylamine. The decrease in the Tl content over the 3 extraction steps with hydroxylamine indicated that the respective pool got also depleted. On average, 58% of the total soil Mn were extracted by hydroxylamine at pH 6.1 (Table S12), but less than 1% of the total Fe. Therefore, and because the NH 4 -acetate in the hydroxylamine extract minimized the re-adsorption of reductively released Tl(I), the hydroxylamine extract was considered to adequately represent (non-exchangeable) Tl associated with reducible Mn oxides.
Over the 18 analysed soils, the fraction of Tl associated with Mn oxides (2.0 ± 1.3%) corresponded to about 59% of the exchangeable Tl fraction (3.7 ± 1.6%). Taking also the reducible Mn into account, the molar Tl/Mn ratios in the hydroxylamine extract and the corresponding loading of Mn oxides could be calculated based on the assumption that all Tl and Mn in the reducible fraction were associated with Mn oxides. Furthermore, assuming that the Tl sorbed onto Mn oxides was in equilibrium with the porewater, and using the Tl concentrations in the pseudo-porewater extract, log-transformed distribution coefficients for the sorption of Tl onto soil Mn oxides (log ( Tl K d,MnO2 / (L/kg)) were calculated, which varied over a rather narrow range for the 18 analysed soils (5.2 ± 0.3).

Tl adsorption and re-extraction
In Fig. 3, the adsorption isotherms for six soils together with the results from the subsequent re-extraction with CaCl 2 (dissolved Tl) and with NH 4 -acetate (exchangeable Tl) are shown. The analogous plots as log transformed distribution coefficients versus dissolved Tl are shown in Figure S3. The entire adsorption and re-extraction data, as well as the evolution of pH and Tl, Rb, K, Ba, Mg, and Ca in a blank sample are summarized in the Appendix (Table S13 and S14). In general, the isotherms derived from the NH 4 -acetate and CaCl 2 extracts (in analogy to the isotherm for geogenic Tl in Fig. 1) fell onto the same line as the adsorption isotherms of freshly adsorbed Tl. This close match suggested that the two extracts adequately described the relationship between exchangeably adsorbed and dissolved Tl. However, on average, only 68 ± 14% of the freshly adsorbed Tl was re-extracted as NH 4 -exchangeable  (Table S7) and scaled by the content of micaceous clay minerals (illite and musovite) in the individual soils (Table S2).
Tl, and dissolved Tl concentrations in the CaCl 2 extracts on average corresponded to 57 ± 5% of residual dissolved Tl in the adsorption experiments. These shifts suggested that part of the freshly adsorbed Tl had become fixed in lessexchangeable form (presumably in micaceous clay minerals), or that the NH 4 -acetate and CaCl 2 extracts underestimated the respective quantities for methodological reasons.

Speciation of geogenic and adsorbed Tl by Tl L III -edge XANES spectroscopy
In Fig. 4, the spectra of selected soil samples, of a soil clay mineral fraction, and of soils with freshly adsorbed Tl are compared to the four reference spectra selected for LCF analysis. A larger set of reference spectra (Fig. S4), the rationale for the selection of the four reference spectra shown in Fig. 4 for LCF analysis, LCF results (Table S15) and all sample spectra together with their LCF reconstructions (Fig. S5) are provided in the Appendix.
The overlay plot of three representative topsoil spectra with geogenic Tl contents of 10 mg/kg (B5), 102 mg/kg (D5), and 856 mg/kg (F5) reveals no marked spectral differences, indicating that Tl speciation was similar irrespective of total soil Tl contents. The LCF analysis (Table S15) of the soils with 7.6-1026 mg/kg geogenic Tl showed that Tl speciation was dominated ($98% on average) by Tl(I) associated with illite or other micaceous clay minerals. Remarkably, essentially the same spectrum was also recorded on a soil clay mineral fraction isolated from topsoil after the removal of soil organic matter and Fe and Mn oxides (P1 00-20 clay, Fig. 4), confirming the conclusion that Tl was associated with micaceous clay minerals. Because of the close similarity of the reference spectra Tl(I)-illite (Tl(I) adsorbed at the frayed edges of illite) and Tl(I)-muscovite (structural Tl(I) in the interlayer) (Fig. 4), however, XANES spectroscopy did not allow differentiating between Tl adsorption onto or structural incorporation into micaceous clay minerals.
The LCF analysis of some spectra pointed to a minor fraction of Tl(III). The fitted small Tl(III) fractions were associated with substantial uncertainty and could also have served to account for minor variations in energy calibration and instrumental resolution among spectra collected at different beamlines. On the other hand, if real, the small Tl(III) fractions were most probably associated with Mn oxides, noting that Tl(III) in Mn concretions has previously been identified in Erzmatt soils (Voegelin et al., 2015).
Avicennite (Tl 2 O 3 ) and Tl(I)-jarosite have previously been identified as secondary minerals in mineralized Tl-rich subsoil horizons on the Erzmatt (Herrmann et al., 2018;Voegelin et al., 2015). Considering that the XANES spectrum of Tl(I)-jarosite is fairly distinctive (Fig. S4), the soil XANES spectra showed that this secondary mineral played a minor role in the topsoil samples, and that soil Tl speciation was dominated by pedogenic Tl species.
The overlay plot of the spectra of three soil samples with freshly adsorbed Tl at levels that exceed geogenic Tl contents by about ten to hundred times also closely matched each other. LCF analysis indicated that these spectra could be described by a major fraction of 65 ± 1% of Tl(I) associated with micaceous clay minerals (adsorbed or incorporated), a minor but uncertain fraction of Tl(III), as well as 27% ± 1% of hydrated Tl + . This latter fraction may account for (hydrated) Tl + associated with illite or other clay minerals at higher loadings (Wick et al., 2018), Tl + adsorbed to certain Mn oxides (Peacock and Moon, 2012;Wick et al., 2019), Fe oxides or organic matter (Fig. S4). Thus, the spectra confirmed that most freshly adsorbed Tl was retained by micaceous clay minerals, but also pointed to Tl sorption to other soil components or less specific Tl sorption onto clay minerals at higher Tl loadings.

Model calculations for Tl adsorption onto micaceous soil clay minerals
The relationship between NH 4 -acetate exchangeable and dissolved Tl in CaCl 2 extracts over all 36 soils was interpreted as an isotherm for exchangeable geogenic Tl. In Fig. 1C, this isotherm is compared to a prediction for Tl (I) adsorption onto micaceous soil clay minerals calculated with the 3-site cation exchange model for Tl, Rb, K, NH 4 , Na, Mg and Ca adsorption onto illite (Wick et al., 2018). For this prediction, the average concentrations of Rb, K, Fig. 4. Tl L III -edge XANES spectra of soil samples with geogenic Tl loadings of 10 mg/kg (B5), 102 mg/kg (D5), and 856 mg/kg (F5), of geogenic Tl in a soil clay mineral fraction (P1 00-20 clay), and of soils (A6 and D1) with $100-1000 mg/kg freshly sorbed Tl in comparison to four reference spectra (dehydrated Tl + adsorbed onto FES of illite (Tl(I)-illite) and in the interlayer of Ba-rich muscovite (Tl(I)-muscovite); Tl(III) sorbed onto d-MnO 2 (Tl(III)d-MnO 2 ); hydrated aqueous Tl + ). See Appendix for further reference spectra and details on XAS data interpretation. NH 4 , Mg, and Ca in the CaCl 2 extracts of all 36 soils were used (Table S16) and only the Tl concentration was varied. Finally, the calculated exchanger composition was scaled to the average content of dioctahedral phyllosilicate (illite and muscovite) in the six soils examined by XRD (16.8 ± 2.0%). The predicted adsorption isotherm closely matched the relationship between exchangeable and dissolved Tl. Additional calculations based on the 5-and 95-percentile concentrations of competing cations in the CaCl 2 extracts (Table S16) were calculated to visualize potential variations in adsorbed amounts as a result of adsorption competition. Additional predictions were calculated based on the effective concentrations of all cations in the CaCl 2 extract of each soil (Table S7), which resulted in a slight improvement in fit quality (Fig. S6).
In the model calculations, adsorbed Tl in most soils was primarily allocated to the FES. Thus, Tl in these soils was assumed to be primarily adsorbed at the frayed edges of micaceous clay minerals where Tl + adsorption is mainly affected by adsorption competition with K + and NH 4 + . Although Rb + sorbs more strongly onto FES than K + or NH 4 + (Wick et al., 2018) competition with Rb + was not relevant in the model calculations because of the low soil Rb concentrations relative to K and NH 4 . In the soils with the highest geogenic Tl contents, the T2S contributed most to the modelled adsorbed Tl. From the experimental distributions coefficients of Tl, Rb and K, overall cation exchange selectivity coefficients for Tl-Rb and Tl-K exchange in the bulk soil materials were derived (Eq. (2)). Remarkably, model predictions for the log-transformed Tl-K exchange coefficient (0.95 ± 0.10) closely matched experimental values (0.97 ± 0.14). On the other hand, the model underestimated the adsorption of Rb by about 0.5 log units. By adjusting the cation exchange coefficient for Rb-Na exchange on FES in the model, the fits of sorbed Rb could be improved without affecting the model outcome for Tl + , reflecting that Tl + adsorption is mainly affected by adsorption competition with K + and NH 4 + , rather than competition with the trace metal Rb + .
The model predictions for Tl adsorption in the six soils used for the adsorption / re-extraction experiments also closely matched experimental adsorption as well as reextraction data (Fig. 3 and S3). The highest dissolved Tl concentrations and exchangeably adsorbed Tl contents in the Tl-spiked soils were up to two orders of magnitude higher than the highest levels of geogenic Tl. For these highest Tl spike levels, the model calculations attributed most adsorbed Tl to the T2S as the FES were already saturated with Tl at lower dissolved Tl concentrations.

Tl adsorption onto Mn/Fe-enriched and Mn/Fe-depleted acidic soil material
Tl adsorption experiments with Mn/Fe-enriched and Mn/Fe-depleted soil material from the Sd horizon of an acidic pseudogley soil showed similar extents of Tl adsorption for both materials, i.e., no detectable effect of Mn oxides on Tl adsorption. Also the Tl L III -edge XANES spectra provided no evidence for Tl association with Mn oxides, but indicated that essentially all Tl was retained by adsorp-tion onto micaceous clay minerals, with illite and muscovite together accounting for $11% of the soil mass (see section 11 in Appendix for further details).

DISCUSSION
The combined interpretation of all experimental results provides new insights into the importance of micaceous clay minerals for long-term Tl fixation and for short-term Tl adsorption via cation exchange in soils, as well as into the relevance of Mn oxides and other soil sorbent phases for Tl retention. These aspects and their environmental relevance are discussed in detail in this discussion section.

Fixation of Tl in the interlayers of micaceous clay minerals
The characterisation of the geogenic Tl by XAS indicated that nearly all soil Tl was associated with micaceous clay minerals. On the other hand, less than 10% of the total soil Tl was exchangeable with NH 4 + or in isotope exchange with 204 Tl over up to 90 days reaction time. Accordingly, 90% or more of the total soil Tl most probably were fixed in the interlayers of micaceous clay minerals. This finding supports earlier studies on naturally Tl-enriched soils, in which high residual Tl fractions in sequential extractions or the correlation of Tl concentrations with soil clay content were attributed to the association of Tl with primary silicates including mica, or secondary illite (Tremel et al., 1997b;Vaněk et al., 2009). Considering that the Erzmatt soils developed on mineralized carbonate rock, and that hydrothermal muscovite can form as accessory mineral in hydrothermal brines (Heinrich and Candela, 2014;Runyon et al., 2019), hydrothermal Tl-enriched muscovite could have been a source of Tl-bearing clay minerals in the Erzmatt soils. However, total soil Tl tightly correlated with As, pointing to their co-release from hydrothermal pyrite. Furthermore, hydrothermal waters assumed to be analogous to the waters that led to Tl-As-Fe mineralization on the Erzmatt have temperatures below 100°C and contain elevated concentrations of Rb in addition to Tl (Loges et al., 2012). Much higher temperatures, however, are required for the formation of hydrothermal muscovite, and the accumulation of Rb and Tl in hydrothermal muscovite would be expected to result in a tight correlation between total Tl and Rb contents in the Erzmatt soils. Consequently, we postulate that the fixation of Tl in micaceous minerals in topsoils on the Erzmatt occurred upon the release of Tl from weathering primary and secondary Tlbearing minerals over the course of soil formation.
In the case of freshly adsorbed Tl, exchangeable contents and soluble concentrations of Tl determined after adsorption point to the potential fixation of a fraction of the freshly added Tl, in line with studies on the kinetics of Cs adsorption onto and desorption from illite (Comans, 1999;de Koning and Comans, 2004;Durrant et al., 2017). Importantly, although it has been shown that Cs gets fixed in illite over periods of weeks to months, desorption data indicate that fixed Cs should not be considered permanently immobilized, but that it may be slowly released again over timescales of years to decades (Comans, 1999;Fuller et al., 2015). Similarly, Tl bound in the interlayers of phyllosilicates in soils from the Erzmatt may not be exchangeable over time scales relevant for soil cultivation and plant growth, but may be leached over longer time periods. Indeed, the observation that the fraction of exchangeable Tl does not vary greatly over nearly three orders of magnitude in total Tl in the Erzmatt soils possibly points to a (slow) equilibration between fixed and rapidly exchangeable Tl. The persistent high Tl levels in topsoils from the Erzmatt, on the other hand, indicate that Tl fixation in micaceous clay minerals substantially reduces Tl leaching, in line with the accumulation of Cs and Rb in weathered micaceous minerals observed in soils formed over thousands of years (Zaunbrecher et al., 2015b).

Adsorption of geogenic and spiked Tl onto micaceous clay minerals
Exchangeable geogenic Tl accounted for only 3.6 ± 1.6% of the total soil Tl contents, and could therefore not be characterized by bulk XAS. The XAS analysis of soils containing mainly freshly adsorbed Tl, however, showed that most of the added Tl adsorbed onto micaceous clay minerals (Fig. 4). For geogenic and freshly adsorbed Tl, the same relationship between NH 4 -acetateexchangeable Tl and dissolved Tl in pseudo-porewater extracts was observed, and this relationship matched the adsorption isotherms of the six soils (Fig. 5A). Furthermore, model predictions based on the cation exchange model for illite (Wick et al., 2018), on dissolved cation concentrations in pseudo-porewater extracts, and on the content of micaceous clay minerals correctly predicted the amounts of exchangeable geogenic and freshly adsorbed Tl (Figs. 1, 3, and 5). In combination, the macroscopic Tl extraction and adsorption data, spectroscopic and modelling results unequivocally show that micaceous clay minerals determine the solubility of Tl in the studied soils, and that this sorption process can be described with the cation exchange model for illite (Wick et al., 2018). This finding is independently underpinned by the close correlations between geogenic Tl and Rb distribution coefficients derived from soil extractions and by the co-variation of dissolved Tl and Rb concentrations in response to changing NH 4 + concentrations in the time-resolved CaCl 2 extractions (Fig. 2).
In the model calculations, geogenic Tl and Rb were nearly exclusively associated with the FES, and T2S became only relevant for freshly adsorbed Tl at the highest spike levels, where the FES were saturated with Tl. At lower and more common soil Tl concentrations, on the other hand, the FES were mainly saturated with K ($2/3) and NH 4 ($1/3). Changes in the concentration of these cations may therefore strongly affect Tl solubility, as has been shown in the past for Cs (Comans et al., 1989;de Koning and Comans, 2004;Evans et al., 1983). Ca and Mg, which mainly sorb onto the planar sites of micaceous clay minerals, on the other hand, may indirectly influence Tl solubility by displacing K and NH 4 from PS, thereby enhancing their competition with Tl for sorption onto T2S and especially FES.
The LCF analysis of the XANES spectra of freshly adsorbed Tl returned about 27% hydrated Tl. This hydrated Tl could represent less specifically adsorbed Tl on clay minerals, or hydrated Tl + adsorbed onto Mn oxides (Wick et al., 2019), Fe oxides (Casiot et al., 2011;Coup and Swedlund, 2015), or soil organic matter (Liu et al., 2011). Since freshly spiked and geogenic Tl exhibited the same relationship between exchangeable and dissolved Tl (Fig. 4), we speculate that the exchangeable geogenic Tl pool (3.6 ± 1.6% of the total geogenic Tl) may include a similar percentage of hydrated Tl (27%). This fraction of less specifically adsorbed exchangeable Tl would thus correspond to $1.0 ± 0.4% of total geogenic soil Tl.

Sorption of Tl onto soil Mn oxides
Using micro-focused X-ray fluorescence spectrometry (m-XRF), elevated concentrations of Tl in Mn-concretions in topsoil samples from the Erzmatt have previously been documented, and Tl in one concretion was identified as Tl(III) sorbed onto birnessite by m-XANES and extended X-ray absorption fine structure (EXAFS) spectroscopy (Voegelin et al., 2015). Bulk XANES data collected in the present study (Fig. 4) did not allow to quantify the fractions of Tl(I) or Tl(III) associated with Mn oxides because the spectra were dominated by the large fraction of Tl associated with micaceous clay minerals.
The chemical fractionation of Tl in 18 soils by sequential extraction, however, returned Mn oxide-associated Tl fractions that were about half as high as the exchangeable frac-tions (Table S12). Thus, the fraction of Tl associated with Mn oxides via strong sorption can be of similar magnitude as the fraction associated with micaceous clay minerals via fast cation exchange, as reflected in the similarity of the pseudo-isotherms for exchangeable and Mn oxide associated Tl in the Erzmatt soils expressed in terms of sorbed Tl per soil mass (Fig. 5AB). This interpretation is in agreement with studies showing that freshly adsorbed Tl in soils is mostly associated with the ''exchangeable" or ''reducible" fractions (Lehn and Schoer, 1987;Vaněk et al., 2010a;Vaněk et al., 2010b;Vaněk et al., 2011), the latter fraction often being assumed to represent Tl sorbed onto Mn oxides.
From the sequential extraction data, the Tl loading of Mn oxides in the soils was estimated (Tables 2 and S12). These loadings (expressed per sorbent mass) were much higher than the amounts of exchangeable Tl associated with micaceous clay minerals (Fig. 5C), reflecting the fact that Mn oxides exhibit a higher affinity for Tl sorption than illite (Wick et al., 2019). The log K d values derived from the chemical extraction data closely matched with log K d values for nonoxidative Tl sorption onto triclinic birnessite or todorokite, but were much lower than log K d values for oxidative Tl scavenging by d-MnO 2 (Fig. 5BC). This observation seems to be in conflict with the earlier identification of Tl(III) in a Mn concretion in an Erzmatt soil (Voegelin et al., 2015). However, it is possible that only a minor fraction of the Mn oxides in the Erzmatt soils can sorb Tl oxidatively, or that natural birnessites in the studied soils have a much lower content of uncapped vacancy sites than freshly synthesized d-MnO 2 . Additional experiments with Mn/Fe-enriched and -depleted soil materials from an acidic pseudogley soil showed that Tl mainly adsorbed onto micaceous clay minerals, and provided no evidence for substantial Tl sorption onto Mn oxides. In combination, these results suggest that highly specific oxidative Tl uptake by soil Mn oxides may often be limited, possibly because (freshly formed) reactive Mn oxides rapidly transform into less reactive Mn oxides that cannot bind Tl oxidatively (Wick et al., 2019). Oxidative Tl uptake may therefore be limited to the timeframe of biological or autocatalytic Mn oxide formation in redox-dynamic soils, but less relevant for aged Mn oxides in soils.

Adsorption of Tl onto Fe oxides and soil organic matter
Fe oxides and soil organic matter can be alternative sorbents for Tl in soils (Casiot et al., 2011;Coup and Swedlund, 2015;Jacobson et al., 2005b;Liu et al., 2011;Martin et al., 2020;Vaněk et al., 2009). Quantitative estimates for the adsorption of Tl onto these sorbents were based on published adsorption isotherms recorded in Na or K electrolytes, and on estimates for the amounts of ferrihydrite and humic acid in the Erzmatt topsoils (Table S21). The estimated amounts of Tl adsorbed onto Fe oxides correspond to only 6 ± 4% of the NH 4 -exchangeable geogenic Tl pool, the estimated amounts of Tl complexed by humic acids to only 11 ± 7%. These estimates roughly match with the interpretation of the XANES data, which suggested that about 27% of the exchangeable geogenic Tl may be hydrated Tl, which we assume to include Tl sorbed onto Fe oxides and organic matter (Section 4.2). However, the estimates of the amounts of Tl adsorbed onto Fe oxides or complexed by organic matter most probably represent upper limits, because sorption competition with major bivalent or trace metal cations and lower soil pH values than pH values in the sorption experiments were not taken into consideration. The estimates thus confirm the conclusion that exchangeable geogenic Tl in topsoils from the Erzmatt was mainly adsorbed onto micaceous clay minerals.
Although studies on the sorption of Tl onto ferrihydrite and humic acid have indicated that Tl + sorbs with a relatively low affinity comparable to K + (Coup and Swedlund, 2015;Martin et al., 2020), Fe oxides and organic matter may still be relevant sorbents for Tl under certain conditions. For example, adsorption of Tl onto organic matter may be important in organic-rich soil horizons or peats, as previously inferred from chemical extraction results (Vaněk et al., 2009), and in line with findings for Cs sorption to soils (Delvaux et al., 2001;Valcke and Cremers, 1994). Indeed, in a recent study on two soil profiles from the Erzmatt, elevated fractions of exchangeable Tl were observed in organic surface horizons and were accompanied by increased fractions of hydrated Tl + in the LCF analysis of corresponding XANES spectra (Vaněk et al., 2020), indicating that high levels of organic matter can sorb significant amounts of Tl in readily exchangeable form.

Environmental implications
This study showed that illite and other micaceous clay minerals play a key role in controlling the solubility of Tl in (contaminated) soils. In the long term, micaceous clay minerals can effectively sequester Tl by fixation in their interlayers. At the same time, micaceous clay minerals are important as sorbents of exchangeable Tl. The model developed for Tl adsorption onto illite (Wick et al., 2018) can be used to quantify the fast and reversible adsorption of Tl onto micaceous clay minerals in soils. The currently available data suggest that Tl in soils and sediments reacts largely in analogy to Cs and Rb (Wick et al., 2018). Accordingly, concepts developed to describe the reversible and irreversible sorption of Cs on illite (Brouwer et al., 1983;Comans, 1999;Comans et al., 1991;de Koning and Comans, 2004;Fuller et al., 2015) are transferable to Tl, but should be validated experimentally. Other clay minerals such as weathered biotite or vermiculite that have been shown to be important sorbents for Cs in other types of soils (Delvaux et al., 2001;Nakao et al., 2008;Valcke and Cremers, 1994;Zachara et al., 2002;Zaunbrecher et al., 2015a;Zaunbrecher et al., 2015b) can be expected to also strongly sorb Tl, with adsorption affinities intermediate between Cs and Rb.
The solubility of Tl adsorbed onto FES is strongly dependent on the dynamics of the main competing cations K + and NH 4 + . Therefore, agricultural practices that affect the concentrations of these cations, such as inorganic fertilizer or manure application, may have marked effects on the leaching and possibly also the plant uptake of Tl, although the latter also depends on the effects of K on Tl uptake into plants, as has been shown for Tl + uptake by algae (Hassler et al., 2007), and in analogy to the effects of K on the uptake of Cs by plants (Delvaux et al., 2000;Delvaux et al., 2001;Smolders et al., 1997). Also ammonification and denitrification processes and related changes in the NH 4 + concentrations in the soil porewater may affect Tl adsorption onto FES in soils.
In the Erzmatt soils, Mn oxides play a minor role as sorbents for Tl. In anthropogenically Tl contaminated soils with a shorter contamination history, however, larger fractions of soil Tl could be associated with Mn oxides, as inferred from sequential extractions in earlier work (Antić-Mladenović et al., 2017;Gomez-Gonzalez et al., 2015;Vaněk et al., 2009;Vaněk et al., 2011;Yang et al., 2005). Owing to the much higher sorption affinity of oxidative Tl uptake by reactive Mn oxides than of cation exchange on micaceous clay minerals, oxidatively sorbed Tl may be less accessible through simple desorption. On the other hand, Tl associated with Mn oxides may become mobilized in waterlogged soil via reductive Mn oxide dissolution (and the reduction of poorly soluble Tl(III) into more soluble Tl(I)). Thus, Mn oxides can be highly effective sorbents for Tl, but with a limited chemical stability range. Furthermore, the highly selective sorption of a range of other trace metals onto Mn oxides (Grangeon et al., 2012;Manceau et al., 2002;Peacock and Sherman, 2007;Peñ a et al., 2015;Post, 1999;Simanova and Peñ a, 2015) may further limit their ability to sequester Tl, in contrast to micaceous clay minerals where K and NH 4 are the dominant competing cations in soils (Wick et al., 2018).
The present study provides a solid basis for a more quantitative and mechanistic understanding of Tl sorption processes in soils. In continuation of this work, aspects that warrant further consideration include: (i) the relevance of different Tl sorption processes over a wider range of soils with contrasting physicochemical characteristics (soil pH, organic C, Mn oxide and Fe oxide content, clay mineralogy), (iii) the kinetics of Tl adsorption and desorption processes in soils, (iii) consequences of reductive dissolution of Mn oxides, ammonification/denitrification, and sulfate reduction on the dynamics of dissolved Tl in periodically waterlogged soils, or (iv) the impact of soil fertilization with N and K on the solubility and soil-plant transfer of Tl in agricultural soils.

Declaration of Competing Interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.
Niederhauser, Febin Ambalathattil, Irene Brunner, Marco Fleischmann (all Eawag) and Elmotaz Eltayeb (PSI) for their support in the laboratory. Prof. Dr. Christoph Heinrich (ETH Zü rich) is acknowledged for insightful comments about the genesis of ore deposits. The Swiss Light Source (Paul Scherrer Institute (PSI), Villigen, Switzerland), the Dutch Belgian Beamline (DUBBLE; European Synchrotron Radiation Facility (ESRF), Grenoble, France), and the French National Synchrotron Soleil (Gif-sur-Yvette, France) are kindly acknowledged for the allocation of beam time, and Olga Safonova and Maarten Nachtegaal (SuperXAS, PSI), Dipanjan Banerjee (DUBBLE, ESRF), and Gautier Landrot (SAMBA, Soleil) for support during XAS data collection. This work was financially supported by the Swiss National Science Foundation under contract No. 200021-162364.