Thallium adsorption onto phyllosilicate minerals

The adsorption of thallium (Tl) onto phyllosilicate minerals plays a critical role in the retention of Tl in soils and sediments and the potential transfer of Tl into plants and groundwater. Especially micaceous minerals are thought to strongly bind monovalent Tl(i), in analogy to their strong binding of Cs. To advance the understanding of Tl(i) adsorption onto phyllosilicate minerals, we studied the adsorption of Tl(i) onto Na- and K-saturated illite and Na-saturated smectite, two muscovites, two vermiculites and a naturally Tl-enriched soil clay mineral fraction. Macroscopic adsorption isotherms were combined with the characterization of the adsorbed Tl by X-ray absorption spectroscopy (XAS). In combination, the results suggest that the adsorption of Tl(i) onto phyllosilicate minerals can be interpreted in terms of three major uptake paths: (i) highest-affinity inner-sphere adsorption of dehydrated Tl+ on a very low number of adsorption sites at the wedge of frayed particle edges of illite and around collapsed zones in vermiculite interlayers through complexation between two siloxane cavities, (ii) intermediate-affinity inner-sphere adsorption of partially dehydrated Tl+ on the planar surfaces of illite and muscovite through complexation onto siloxane cavities, (iii) low-affinity adsorption of hydrated Tl+, especially in the hydrated interlayers of smectite and expanded vermiculite. At the frayed edges of illite particles and in the vermiculite interlayer, Tl uptake can lead to the formation of new wedge sites that enable further adsorption of dehydrated Tl+. On the soil clay fraction, a shift in Tl(i) uptake from frayed edge sites (on illite) to planar sites (on illite and muscovite) was observed with increasing Tl(i) loading. The results from this study show that the adsorption of Tl(i) onto phyllosilicate minerals follows the same trends as reported for Cs and Rb and thus suggests that concepts to describe the retention of (radio)cesium by different types of phyllosilicate minerals in soils, sediments and rocks are also applicable to Tl(i).


X-ray diffraction (XRD)
For analysis by X-ray diffraction, about 20 mg of the samples Musc-A, Musc-B, Verm-A and Verm-B were suspended in ethanol, deposited on low-background Si wafers and air-dried. For Musc-A, Musc-B, and Verm-A, sample aliquots as prepared for adsorption experiments were used, for Verm-B, an unmilled sample aliquot was used. XRD patterns were recorded from 5° to 95° 2theta (step-size of 0.017°, 2 or 5.5 h per pattern, Co Kα radiation, X'Pert Powder diffractometer with XCelerator detector, Malvern Panalytical B.V., Almelo, The Netherlands).
The patterns of the muscovites and vermiculites are shown in Fig. S1. Qualitatively, they matched to muscovite and vermiculite reference patterns from the Crystallography Open Database (COD) (not shown), confirming phase identity. For the vermiculites, spectral differences between Verm-A (milled) and Verm-B (unmilled) were in line with the reported effect of milling on the XRD pattern of vermiculite. 1 For XRD analysis, the SCF (soil clay fraction) was exchanged with Mg or K and prepared as oriented mounts on low-background Si slides. After measurement at room temperature, the Mgexchanged SCF was reacted with ethylene glycol (5.5 h at 60°C) and analyzed again. The Kexchanged SCF was heated to temperatures of 100, 300 and 550°C (1 h). After each heat treatment, the sample was allowed to cool and an XRD pattern was recorded.
The XRD patterns of the oriented mounts are shown in Fig. S2. The pattern of the Mgexchanged SCF indicated illite/muscovite (10-Å peak; ~10.2°), hydroxy-interlayered vermiculite (broad peak between 10-and 14-Å peaks), and smectite or vermiculite (14-Å peak; 7.2°). EG treatment caused some interlayer expansion (lower-angle shoulder in the 14-Å peak at ~7.2°), indicative of the presence of smectite. K-exchange induced the (partial) shift of the 14-Å (7.2°) peak to 10 Å (10.2°), in line with the presence of (hydrox-interlayered) vermiculite. Heating to 550°C caused the disappearance of the 7-Å (14.3°) peak, in line with the destruction of kaolinite and only a minor fraction of chlorite. Overall, these results are in good agreement with earlier XRD results for the clay fraction of the soil P1 00-20 from which the SCF was isolated, which indicated that illite/muscovite and hydroxy-interlayered-vermiculite/chlorite were the dominant phylloscilicates, with minor fractions of illite-smectite and some kaolinite.

Scanning electron microscopy -energy-dispersive X-ray (SEM-EDX) analyses
For SEM-EDX analyses, 2 mg aliquots of the samples Musc-A 870, Musc-B 860, Verm-A 820 and Verm-B 630 (numbers refer to Tl loading in mg/kg) were dispersed in 10 mL of doubly deionized (DDI) water, subsequently deposited onto Nucleopore filters by vacuum filtration, dried, and C-coated (Leica ACE600). The soil clay fraction was analogously prepared after diluting 50 µL of the SCF suspension (with 53 g/L solids) into 10 mL DDI water. The SEM (NanoSEM 230, FEI) was operated at an acceleration voltage of 15 kV. EDX data (X-MAX 80 detector, Oxford Instruments) were collected and quantitatively evaluated using the INCA software package (Version 4.15, Oxford Instruments).
Backscattered electron (BSE) images of the muscovites and vermiculites are shown in Fig. S3, the results from EDX point analyses are provided in Table S1. All samples exhibited the typical platy morphology of phyllosilicate minerals. For the muscovites, the data pointed to a larger share of finer material in the Musc-B sample. For the vermiculites, the results revealed that the sample Verm-A contained a much larger share of fine material than in Verm-A, and showed that larger vermiculite platelets dominated in Verm-B. The elemental composition data was in line with expectations for muscovites (high K and Al, low Mg) and vermiculites (high Mg and Fe, low K). BSE images of SCF are shown in Fig. S4, the results from EDX analyses on a range of points are listed in Table S2. Both sample morphology and EDX results attest to the heterogeneity of the soil clay fraction. The EDX results could be associated to 5 major groups based on element abundances (probable mineral phase association): (i) high Si (quartz), (ii) high Ti (anatase), (iii) high K / low Mg & Fe (illite, muscovite), (iv) low K / high Mg and Fe (chlorite), (v) intermediate K, Mg & Fe / high Al (hydroxy-interlayered vermiculite). The probable mineral phase associations are based on earlier XRD results for the soil P1 00-20 2 from which the SCF was isolated, and on XRD results for the SCF shown in the electronic supplementary information (ESI) section 1.1.

Cation exchange capacity (CEC) determinations
The CEC of the different solids listed in Table 1 were obtained by different methods: For the illite IdP, a CEC of 263 mmolc/kg had previously been identified for Cs-exchanged IdP using the 137 Cs isotope exchange technique, 3 and a CEC of 260 mmolc/kg for Tl-exchanged IdP using the same isotope exchange approach with 204 Tl instead of 137 Cs as radioisotope. 4 The CEC of 260 molc/kg was taken for Na-IdP and K-IdP.
For the smectite SWy used in this study, a CEC of 860 mmolc/kg has been determined using the 137 Cs isotope technique. For similar types of smectites, a CEC of 870 mmolc/kg has previously been determined using 22 Na isotope dilution, 5 and a CEC of 0.89 mmolc/kg using Cs/K exchange. 6 For the two muscovites, the 137 Cs isotope technique was used to measure the CEC on the residual solids from the Tl adsorption experiments with ~1000 mg/kg adsorbed Tl (as the available material mmolc/kg has been reported for raw material from the same source 1 based on the ammonium acetate method. 9 The CEC values for both Verm-A and Verm-B fall into the common range of CEC values of trioctahedral vermiculites. 7 For SCF, a CEC of 290 mmolc/kg was determined using the 137 Cs isotope exchange technique, 3 A similar value had been obtained by Ca-Mg / Ba-Ca exchange (as described for Verm-A above). Table S3. Adsorption data: mineral name, sample name as used in figures and text (mineral followed by Tl loading in mg/kg), suspension volume, mineral mass, total Tl in suspension, equilibration time, final pH, final dissolved Tl (C_Tl) and final adsorbed Tl loading (Q_Tl), adsorbed Tl equivalent fraction (NTl), distribution coefficient (log(Kd/(L/kg))) and conditional Tl-Na cation exchange selectivity coefficient (log(Kc,Tl-Na); not applicable to K-IdP, as adsorption was m in 10 mM KCl electrolyte).

Overview over XAS samples and measurements
An overview over the XAS samples and measurements used in this study is given in Table S4.
Most samples were Tl-loaded solids from the adsorption experiments of this study, few samples were from own earlier work. 4,10 The sample names of Tl-loaded phyllosilicates consist of the mineral name followed by the Tl loading in mg/kg. The XAS measurement type (XANES, EXAFS) is indicated together with measurement temperature in parentheses (RT = room temperature, 20 K = 20 Kelvin). Further details on the XAS measurements are given in the Materials and Methods section of the manuscript.

Tl LIII-edge XAS of Tl(III) and Tl(I) reference compounds
In            Table S5.  Table S5.

LCF analysis of XANES spectra of Tl-loaded vermiculites
The XANES spectra of the Tl-loaded vermiculites Verm-A and Verm-B showed a spectral shift with increasing Tl loading (Figs. S12 and S13) that was evaluated by LCF. As reference spectra for LCF analysis, the spectrum Na-SWy 1700 and the spectrum of the highest loaded vermiculite sample of the respective series (Verm-A 28000 or Verm-B 28000) were used. The LCF results are listed in Table S5, the LCF reconstructions are shown in Figs. S12 and S13. The results show that the spectral trends can be adequately described by a decreasing fraction of the reference spectrum Na-SWy 1700 with increasing Tl loading, which we interpret as a decrease in the fraction of hydrated Tl + adsorbed in the expanded vermiculite interlayer and a shift towards dominant uptake of dehydrated Tl + in collapsed zones of the vermiculite interlayer.

Ba-muscovite
EXAFS spectra of the Tl-rich Ba-muscovite sample were recorded at room temperature (RT) and 20 K. The k 2 -weighted spectra and the corresponding Fourier-transforms (k-range 3-7 Å -1 ; Kaiser-Bessel windows with dk = 2 Å -1 ) together with shell-fits of the peak around 3 Å are shown in Fig. S17. For shell-fit analysis, theoretical scattering paths were calculated using Feff 6.0 11 as implemented within the Demeter software package 12 , based on the structure of illite 13  We therefore used shell fitting only to test if the marked second-shell peak in the Fouriertransformed spectra of Ba-muscovite could be fitted with a second-shell Tl-Si path. In the fit, the coordination number of second-shell Si was fixed to 12, the expected value for Tl in the interlayer of muscovite, and the amplitude reduction factor to 1. The same interatomic distance and energy shift was fitted for both spectra, whereas separate Debye Waller factors were fitted to account for variations in thermal disorder with temperature. The fit results are listed in Table S6. The fitted Tl-Si distance of 3.84 Å was reasonable, considering an average K-Si distance of 3.77 in illite 13 and that the ionic radius for 12-fold coordinated Tl + (1.70 Å) is 0.06 Å larger than for K + (1.64 Å) (ionic radii from 14 ). Accordingly, we attributed the marked second-shell peak in the EXAFS spectra of Tl in Ba-muscovite to Tl-Si contributions and considered it as diagnostic spectral feature of Tl bound in the siloxane cavities of micaceous clay minerals.

Tl adsorbed onto Na-IdP
In Figure S18, the XANES and EXAFS spectra of samples Na-IdP with 50, 300 and 3800 mg/kg adsorbed Tl are compared (samples from ref 4 ). No marked changes are observed among the three XANES spectra, and also the EXAFS spectra of the Na-IdP with 300 and 3800 mg/kg adsorbed Tl do not reveal marked differences. The EXAFS spectrum Na-IdP 50 is very noisy, but as far as this can be judged, does not seem to vary from the spectra of higher loaded samples. In the 3-site cation exchange model parameterized for Tl adsorption onto Na-IdP, the FES concentration has been set to 0.65 mmol/kg. 4 The sample Na-IdP 50 has a Tl loading of 0.25 mmol/kg Tl, significantly less than the FES concentration in the model, and the sample Na-IdP a Tl loading of 1.5 mmol/kg, 2.3 times the FES concentration. The absence of marked spectral changes from the sample Na-IdP 50 over the sample Na-IdP 300 to the sample Na-IdP 3800 with a much higher Tl loading of 19 mmol/kg suggested that up to 19 mmol/kg Tl could be adsorbed with similar local Tl coordination as low levels of Tl whose adsorption was attributed to highaffinity FES in the 3-site cation exchange model.
The EXAFS spectra of Tl adsorbed onto Na-IdP at loadings of 3800 mg/kg (recorded at room temperature; sample from ref 4 ) and 3200 mg/kg (spectrum recorded at 20 K) are shown in Fig. S19.
The spectra exhibit a second-shell peak at the same position as Tl-rich Ba-muscovite (Fig. S17).
Assuming that Tl adsorbed onto the Na-IdP by complexation between two siloxane cavities, a Tl-Si coordination number of 12 would be expected as for Tl in Ba-muscovite. The much lower intensity of the second-shell peak of the Na-IdP spectra than the Ba-muscovite spectra might thus point to higher structural disorder of freshly adsorbed than structurally incorporated Tl. A shell-fit analysis in analogy to the one for Tl in Ba-muscovite was thus performed to test if the lower peak amplitudes could be attributed to higher disorder. The fit parameters are listed in Table S6,  Due to the limited fit range, however, the fitted parameters were relatively poorly constrained.
Furthermore, the fitted peak positions exhibit slight displacements from the experimental peak position, possibly pointing to spectral contributions from other scattering paths.

Tl adsorbed onto K-IdP, Na-SWy, Musc-A and Musc-B and Verm-A and Verm-B
In Fig. S20, the EXAFS spectra of Tl adsorbed on to K-IdP (4200), Na-SWy (1700), Musc-A (2400), Musc-B (7400), Verm-A (2700) and Verm-B (28000) (with loadings in mg/kg) are shown together with shell-fits based on a single Tl-Si path to reproduce the experimental second-shell peak. The fit results are listed in Table S6. These shell-fits served to test if the experimental data could be reasonably reproduced by model fits in which the Tl-Si coordination numbers were set to the ones inferred from the interpretation of the combined macroscopic (adsorption data) and spectroscopic results. Owing to the limited EXAFS data range and the correlation between Tl-Si coordination number and Debye Waller factor, the fits were poorly constrained and fit results need to be interpreted with caution. In the manuscript, the EXAFS spectra shown in Fig. S20 (Table S6, fits 3 and 5) reasonably reproduced the experimental data (Fig. S20). The Tl-Si distances fell into the range of those of Ba-muscovite and Na-IdP, and the Debye Waller factors were similar to those of Tl adsorbed onto Na-IdP 3800.
In the case of Tl adsorbed onto Na-SWy, the Tl-Si coordination number was refined and the Debye Waller parameter was arbitrarily set to the one obtained for the spectrum Na-IdP 3800. The modelled Tl-Si peak closely reproduced the experimental peak.
For Verm-A 2700 and Verm-B 28000, the macroscopic and spectroscopic data suggested that Tl was mainly adsorbed as dehydrated Tl + complexed between two siloxane cavities in the collapsed interlayer. Accordingly, the CN of the Tl-Si path was set to 12 in the model fits, as for Na-IdP (Table S6). Especially for Verm-B 28000, the difference between modelled and experimental peak shape and amplitude was larger than for Na-IdP 3800 (Fig. S19, S20). Using a single Tl-Si path, the discrepancies could not be further reduced, as a higher Debye Waller factor would have been required to broaden the modelled peaks, but a lower one to obtain a higher peak amplitude. This points to additional spectral contributions, potentially arising from two distinct Tl-Si shells, from second-nearest O atoms, or from multiple scattering paths. However, the limited EXAFS data range did not allow to further test these possibilities because the number of adjustable fit parameters could not be further increased. Table S6. Shell-fit analysis of selected EXAFS spectra using a Tl-Si backscattering path (CN = coordination number; δ 2 = Debye Waller factor, R = interatomic distance, dE0 = energy shift, r-factor = normalized sum of squared residuals of fit). Spectra with same Fit. Nr. (and same background color) were fit simultaneously with some parameters constrained to be the same for both spectra. Values given without fit uncertainty (in parentheses) were fixed. The amplitude reduction factor was set to 1 in all fits. Fits were conducted on the Fourier-transformed k 2 -weighted EXAFS spectra (k-range 3-7 Å -1 ; Kaiser Bessel window parameter 2 Å -1 ) over the r-range 2.5-3.8 Å. The number of independent datapoints was 6.3 in fits including two spectra, and 3.2 in fits of one spectrum. The sample spectra and fits are shown in Figs. S17, S19 and S20.
Fit 1: Spectra of Ba-muscovite recorded at room temperature (RT) and 20 K. CN set to 12 assuming Tl to substitute K between two siloxane cavities. Individual δ 2 (temperature-dependence), but same R for both spectra. Fit 2: Spectra of 3800 / 3200 mg/kg Tl adsorbed onto Na-IdP recorded at RT and 20 K. CN set to 12 assuming Tl to be complexed between two siloxane cavities. Individual δ 2 (temperaturedependence), but same R for both spectra.   Table S6).  with the fit parameters listed in Table S6.