Effect of pH, surface charge and soil properties on the solid – solution partitioning of perfluoroalkyl substances (PFASs) in a wide range of temperate soils

), was investigated for 11 temperate mineral soils and related to soil properties such as organic carbon content (0.2 – 3%), concentrations of Fe and Al (hydr)oxides, and texture. PFAS sorption was positively related to the perfluorocarbon chain length of the molecule, and inversely related to solution pH for all substances. The negative slope between log K d and pH became steeper with increasing per-fluorocarbon chain length of the PFAS ( r 2 = 0.75, p ≤ 0.05). Organic carbon (OC) alone was a poor predictor of the partitioning for all PFASs, except for FOSA ( r 2 = 0.71), and the OC-normalized PFAS partitioning, as derived from organic soil materials, underestimated PFAS sorption to the soils. Multiple linear regression suggested sorption contributions ( p ≤ 0.05) from OC for perfluorooctane sulfonate (PFOS) and FOSA

• pH-dependent partitioning of six PFASs was studied for eleven temperate mineral soils.The pH-dependent soil-water partitioning of six perfluoroalkyl substances (PFASs) of environmental concern (PFOA, PFDA, PFUnDA, PFHxS, PFOS and FOSA), was investigated for 11 temperate mineral soils and related to soil properties such as organic carbon content (0.2-3%), concentrations of Fe and Al (hydr)oxides, and texture.PFAS sorption was positively related to the perfluorocarbon chain length of the molecule, and inversely related to solution pH for all substances.The negative slope between log K d and pH became steeper with increasing perfluorocarbon chain length of the PFAS (r 2 = 0.75, p ≤ 0.05).Organic carbon (OC) alone was a poor predictor of the partitioning for all PFASs, except for FOSA (r 2 = 0.71), and the OC-normalized PFAS partitioning, as derived from organic soil materials, underestimated PFAS sorption to the soils.Multiple linear regression suggested sorption contributions (p ≤ 0.05) from OC for perfluorooctane sulfonate (PFOS) and FOSA, and Fe/Al (hydr) oxides for PFOS, FOSA, and perfluorodecanoate (PFDA).FOSA was the only substance under study for which there was a statistically significant correlation between its binding and soil texture (silt + clay).To predict PFAS sorption, the surface net charge of the soil organic matter fraction of all soils was calculated using the Stockholm Humic Model.When calibrated against charge-dependent PFAS sorption to a peat (Oe) material, the derived model significantly underestimated the measured K d values for 10 out of 11 soils.To conclude, additional

Introduction
The persistence, large environmental mobility, and toxicity of perfluoroalkyl substances (PFASs) highlight a need for more accurate environmental risk assessments.However, to perform robust long-term risk assessments, it is necessary to have a better understanding of the binding mechanisms in soils and to what extent PFAS mobility changes with geochemical conditions.It is now well established that the perfluorocarbon chain length in PFAS molecules, which reflects the intrinsic hydrophobicity, is closely related to the overall sorption of the substance (Guelfo et al., 2021;Guelfo and Higgins, 2013;Nguyen et al., 2020).Further, the soil solution pH has a substantial impact on PFAS binding due to the pH-dependent charge of common sorbents such as organic matter and Fe and Al (hydr)oxides (e.g., Nguyen et al., 2020;Higgins and Luthy, 2006;Campos-Pereira et al., 2020, 2022).This has been attributed to charge interactions between the particulate phase and PFAS (Nguyen et al., 2020;Higgins and Luthy, 2007).However, the mechanisms governing the magnitude of pH-and charge-dependent PFAS binding are not yet fully understood.Moreover, the pH-dependent sorption of PFASs, such as perfluorooctane sulfonamide (FOSA), which change their protonation state over the environmentally relevant pH range, has been described only to a limited extent in the current literature (Mejia-Avendaño et al., 2020;Nguyen et al., 2020;Zhi and Liu, 2018).This is despite the fact that the change in protonation state may substantially alter the binding, leaching and fate of FOSA in the environment.
In many studies, organic matter (measured as organic carbon, OC) has been found to be an important sorbent for PFASs (e.g., Umeh et al., 2021;Nguyen et al., 2020).However, OC alone has rarely been reported to strongly correlate with soil-pore water distribution coefficients (K d ) for PFAS (Umeh et al., 2021;Li et al., 2018), although OC has been found to be a better predictor for the sorption of longer-chained as compared to that of shorter-chained PFASs (Nguyen et al., 2020).The relative roles of soil components such as organic carbon (Campos-Pereira et al., 2022;Fabregat-Palau et al., 2022;Zhi and Liu, 2018), Fe-and Al (hydr)oxides (Campos-Pereira et al., 2020;Oliver et al., 2019;Tang et al., 2010;Wang and Shih, 2011), and clay minerals (Xiao et al., 2011;Jeon et al., 2011;Luft et al., 2022) remain to be fully understood, especially with regards to how their relative contributions change when the soil pH changes in natural and engineered systems (e.g., upon soil washing).While the sorption of many PFASs onto organic matter and metal (hydr)oxides is strongly pH-dependent, the pH-dependent PFAS binding to surfaces with little or no variable charge, such as silica or clay minerals, is significantly smaller or even absent (Tang et al., 2010;Zhao et al., 2014).Possibly, this is due to significant contributions from hydrophobic interactions to the overall sorption.These may involve siloxane patches (≡Si-O-Si≡) on the mineral surfaces (Shafique et al., 2017).Hence, in soils where hydrophobic interactions with silicates are important, the binding of PFASs would be expected to be less pH-dependent compared to soils where organic matter and metal (hydr) oxides are the major sorbents.
The complex sorption behavior of PFASs results from their unique chemical structure, which consists of an ionizable hydrophilic head group and a hydrophobic organofluorine tail.Therefore, common models for predicting the sorption of non-ionizable, hydrophobic organic chemicals are not suitable for describing the sorption of PFASs to soil or other geosorbents, as electrostatic effects between sorbate and sorbent are not considered (e.g.Karickhoff et al., 1979).To address this issue, Higgins and Luthy (2007) derived a mechanistic model for PFAS binding to organic matter in sediment materials, which considers both hydrophobic and electrostatic interactions.In their approach, the K d value was related to the PFAS-specific hydrophobicity, the organic carbon content of the sorbent, and the coulombic (electrostatic) free energy contribution, which was expressed as a Donnan potential between the organic matter fraction and the aqueous phase.The electrostatic contribution was calculated as a function of specific aqueous conditions, i.e., pH and dissolved ionic concentrations using a non-ideal competitive adsorption (NICA) Donnan model.Although theoretically sound, the model has, to the best of our knowledge, not yet been applied on soil materials.In previous work, a similar approach was used to relate the binding of PFASs to "pure" organic soil materials (O horizon and Sphagnum peat) to their net surface charge under varying soil solution conditions (Campos-Pereira et al., 2022, 2018).The net surface charge of the organic matter was calculated using the Stockholm Humic Model, SHM (Gustafsson, 2001).In mineral soils, the presence of cations such as Fe 3+ and Al 3+ will strongly affect the surface charge of the organic component; this effect can be estimated using SHM.Thus, in the current work we use SHM to explore the possibility that PFAS partitioning in a diverse set of mineral soils is determined by the net surface charge of the organic matter.To our knowledge this is the first time that PFAS partitioning of mineral soils has been predicted by a state-of-the-art model for proton and cation binding using relationships between log K d and net surface charge as obtained for pure organic matter samples.
The specific objectives were i) to determine the pH-dependent binding of six PFASs of environmental concern that are commonly detected in soils (PFOA, PFDA, PFUnDA, PFHxS, PFOS and FOSA), ii) to ealuate how specific and multiple soil properties influence the binding, and iii) to investigate whether it is possible to model PFAS binding to mineral soils using surface charge-dependent organic carbonnormalized soil-water partitioning coefficients (K OC values) determined for soil organic matter.11 temperate soils having a wide range of key soil properties (pH, OC, clay and Fe/Al (hydr)oxides) are used in this investigation.

Spiking and aging of soils
A sample of field-moist soil, weighing 40 g on a dry weight (dw) basis, was added to 250 mL high-density polyethylene (HDPE) containers.To each soil sample, a mixture of the studied PFASs dissolved in analytical grade methanol (MeOH) was added dropwise (in total 2 mL) to yield soil concentrations of 104 ± 27 μg kg − 1 (dw) each of PFOA, PFHxS, PFOS, and FOSA, and 33 ± 7.5 μg kg − 1 (dw) of PFDA and PFUnDA (as shown in Table A.3).The spiking concentrations were low enough to ensure that the binding of all PFASs occurred in the linear range in the experiment (Milinovic et al., 2015;Campos-Pereira et al., 2022).After spiking, each container with soil was left open to air to allow the MeOH to evaporate, after which the material was thoroughly/mixed and the containers were tightly capped.The spiked soils were aged for nine months in darkness at +4 • C before the start of the experiment.
In addition to the laboratory-spiked soils, one field-contaminated soil (soil S3, Arboga (field-contaminated); 63 μg PFOS kg − 1 dw, 6.4 μg PFHxS kg − 1 dw) was included in the experiment (Table A.4).To allow for comparison of PFAS sorption between the field and the laboratory, an uncontaminated soil (S5 Arboga) was sampled in close proximity (~30 m) to the field-contaminated site (Table 1) and subsequently spiked as described above.
For the soil samples, total soil PFAS concentrations were determined by two commercial laboratories (Eurofins, ALS Scandinavia).The extracted total concentrations were, on average, 98 (±16)% of that of the methanolic spike used for quantification of the partitioning in the batch experiments (Table A3).However, for the soil with the lowest OC content (soil S11, Riddarhyttan 10-20 cm), only 67% of the spike concentration, on average, was recovered in the soil extracts.This was likely due to the low retention capacity of this soil, which was caused by its low OC content and high sand content (see Results and Discussion section).As a result, the partitioning of PFASs in this soil was quantified based on its extracted total concentrations after aging.For soils S1 and S9 (Kungsängen 0-10 cm, and Kungsängen 70-80 cm, respectively), all of the soil material was used in the sorption experiments, which is why no experimental extraction of total PFAS concentrations was performed on these soils.

pH-dependent batch experiments
2.00 g dw soil was added to 50 mL polypropylene (PP) Corning Falcon reactors (nonpyrogenic).The soils were suspended in a mixture of Milli-Q water (analytical grade, filtered through a powdered activated carbon filter) and NaNO 3 solution.Various amounts of acid (HNO 3 ) or base (NaOH, prepared on the same day) were then added to reach a number of evenly distributed solution pH values that ranged from 3.1 to 8.6 after equilibration.The concentrations added ranged from 0 to 70 mmol kg − 1 as HNO 3 , and from 0 to 50 mmol kg − 1 as NaOH.The additions of NaNO 3 stock solution (0.03 M) were adjusted to maintain the nitrate (NO 3 − ) concentration at 10 mM across all pH values.All batch reactors were set up in duplicate, with a total solution volume of 20 mL and a liquid-to-soil (L/S) ratio of 10 mL g − 1 dw soil.The reactors were equilibrated end-over-end at 45 rpm and 22 • C for 7 days, a time that has been shown to be sufficient to reach sorption equilibrium (Mejia-Avendaño et al., 2020;Zhi and Liu, 2018;Johnson et al., 2007).
After equilibration, the reactors were centrifuged at 2500g.After centrifugation, aqueous-phase aliquots for PFAS analysis were collected and stabilized with MeOH.Subsequently, pH was measured in the supernatant using a Red Rod Ag/AgCl electrode.After filtration (0.45 μm), the aqueous concentrations of inorganic constituents (Al, Fe, Ca, Mg, Mn, K) were determined using inductively coupled plasma optical emission spectroscopy (ICP-OES, PerkinElmer Avio 200).Further details on the PFAS analysis can be found in sections SI-1 and SI-2 in Appendix A.

Data analysis
The binding of PFASs to soils was determined as the difference between added PFASs and PFAS concentrations in the aqueous phase after equilibration (desorption).The soil-water partitioning coefficient K d (mL g − 1 dw; Table A.7) was calculated according to Eq. (1): where C s,i denotes the initial soil concentration (ng g − 1 dw) prior to desorption (i.e., the spike concentration, except for soil S11), LS is the liquid-to-soil ratio of the reactor (10 mL g − 1 dw) and C W [ng mL − 1 ] is the aqueous PFAS concentration measured after equilibration.The organic carbon-normalized soil-water partitioning coefficient K OC (mL g − 1 ) was calculated from K d and the fraction of soil organic carbon, f OC (unitless), according to Eq. ( 2) (see results in Table A.8): According to the Shapiro-Wilk text (Shapiro and Wilk, 1965), the logarithms of K d and K OC were generally normally distributed.Therefore, descriptive statistics of the partitioning data were calculated based on the log-transformed partitioning coefficients.The average pH value of all analyzed samples was 5.7, which is close to the average pH(H 2 O) of the soils (6.2).
Single (SLR) and multiple linear regression (MLR) were used to examine relationships between PFAS binding and soil properties at a fixed pH value of 5.6 ± 0.1, which reflects the average pH of all samples.The MLR models initially included the candidate predictors of OC, silt + clay content, Fe Ox , and Al Ox-Pyp , and non-significant predictors (p > 0.05) were sequentially removed, starting with the one with the highest p value.The final MLR models only include predictors with partial slopes that are significantly different from zero (p ≤ 0.05).Further details on the MLR methods can be found in section SI-5 and in Table A.12.

Geochemical modeling
To further investigate the role of soil organic matter (SOM) for the sorption of PFASs, the Stockholm Humic Model (SHM) (Gustafsson, 2001) was used to estimate the pH-dependent surface net charge of the SOM fraction for each soil.These calculations were carried out using the chemical equilibrium software Visual MINTEQ (Gustafsson, 2022), with the assumption that only SOM was contributing to PFAS binding.The procedures for calculating the SOM net charge and the assumptions used are detailed in Appendix A, section SI-4.The calculated SOM net charge was then related to the charge-dependent PFAS sorption (K OC ), as found for the peat Oe material of Campos- Pereira et al. (2022).By assuming that the charge-K OC relationship for the studied soils was the same as in the peat Oe sample, which consisted almost exclusively of SOM, we were able to predict charge-corrected K OC values (K OC, pred ) for soil S1-S11.The compound-specific partitioning data and regression statistics (i.e., the relationship between SOM net charge and log K OC ) are provided in Table A.14 and Fig. A.1.The predicted K d values (K d, pred ) for soils S1-S11 were obtained from K d, pred = K OC, pred ⋅ f OC , where f OC is the fraction of organic carbon of the soil.

Effect of solution pH on partitioning
The partitioning of all analyzed substances was inversely related to solution pH in the majority of the individual soils (p < 0.05), with the average regressed linear slope ΔK d for each substance-soil pair being − 0.29 log units per pH unit (Fig. 1, Table A10).Acidic pH values increased the partitioning towards the solid phase for all studied PFASs, while neutral or alkaline pH values (i.e., pH 7-8.6) resulted in lower K d values (for more details see Table A7 and Figs.A2-A8).The magnitude of the effect of pH on partitioning varied significantly depending on the substance and soil in question (Fig. 1).The largest effect of pH on PFAS partitioning was observed in the sandy OC-poor soil S11 (average slope across all substances: 0.63 log units per unit pH) followed by the alkaline calcareous silt loam soil S4 (average slope: 0.50 log units per unit pH).
The linear regression slope between log K d and pH also depended on the PFAS, becoming steeper (i.e., more negative) with increasing perfluorocarbon chain length (p < 0.05, r 2 = 0.75) (Fig. 2, Table A10), similar to the observations of Nguyen et al. (2020).As none of the investigated PFASs except FOSA dissociates over the studied pH range (~3-9), the pH-dependent nature of PFAS sorption was attributed to the variable charge of the solid phase (i.e., the lower the net negative charge, the stronger the sorption).The steeper log K d -pH slope for longer-chained PFASs is consistent with the stronger contribution of electrostatic interactions between the electronegative fluorines of the tail part of the molecules and the soil particles (Erkoç and Erkoç, 2001;Johnson et al., 2007;Xiao et al., 2011;Park et al., 2020).The proposed nature of the pH dependence is also consistent with previous work on the pH-dependent sorption of PFASs to organic soil materials and to the iron (hydr)oxide ferrihydrite (Campos-Pereira et al. 2020, 2022Campos-Pereira et al. 2022).However, the log K d -pH slope is steeper for these "pure" materials compared to those obtained for the mineral soils in the present study, which might indicate a contribution from an unknown solid phase to PFAS binding, probably through hydrophobic interactions.
In contrast to the other PFASs under study, the sorption of FOSA was influenced by its dissociation near its pK a value (acid dissociation constant), which is in agreement with the study conducted by Nguyen et al. (2020).The pK a of FOSA has been estimated to be 6.24 (Rayne and Forest, 2009) A.9) in its undissociated form.However, at pH values above the pK a , FOSA was predominantly present in its dissociated form (C 8 F 17 SO 2 NH − ) and exhibited similar behavior to that of its anionic C 8 analog PFOS (Fig. 1

Effects of soil properties on partitioning
In general, the PFASs under study tended to be sorbed more strongly to soils that had higher OC content, higher silt and clay contents, higher extractable Fe Ox and Al Ox-Pyp , and were surface rather than subsurface horizons, which is in line with previous research and demonstrates the complex behavior of PFAS sorption (Knight et al., 2019(Knight et al., , 2021;;Nguyen et al., 2020;Umeh et al., 2021).However, only a few of the aforementioned K d -soil property relationships were statistically significant (p ≤ 0.05) for individual PFASs in the current study (Table A12).
Single linear regression (SLR) at pH 5.6 ± 0.1 showed that several soil properties had significant (p ≤ 0.05) positive correlations with the K d value for some of the PFASs, including OC (FOSA), oxalateextractable Fe, i.e., Fe Ox (FOSA), Al Ox-Pyp (PFOS, PFDA), the sum of Fe Ox and Al Ox-Pyp (PFOS, PFDA), CEC (FOSA), the combined content of silt and clay (FOSA), and total N content (FOSA) (Table A12).Although these relationships were significant, their predictive strength was weak to moderate (average r 2 = 0.58).Normalizing PFAS partitioning against OC (K OC or log K OC ) did not successfully explain the differences in PFAS sorption between soils (Table A.8;Fig. A.11 and A.12).Additionally, the experimentally determined K OC value (assuming that organic matter was the only sorbent) was significantly greater for the mineral soils (often >1 log unit greater for PFHxS, PFOS, PFOA and FOSA) than for the organic soil materials (Fig. 3   anionic PFASs including PFOS.For both PFOS and FOSA, OC was positively correlated with the K d value (p ≤ 0.05), as previously found by Nguyen et al. (2020) and Umeh et al. (2021).Both the SLR and MLR analysis suggested that Al Ox-Pyp played a role in the binding of PFOS, while Fe Ox played a role in the binding of FOSA.Together with OC, these parameters explained 79% and 84% of the variance in the measured K d values of PFOS and FOSA, respectively (Fig. 4).In contrast, Nguyen et al. (2020) did not identify any significant predictor associated with the mineral fraction for the sorption of PFOS, FOSA, and PFDA, with the exception of the soil micropore volume.Instead, these authors reported OC as the only significant predictor for the sorption of these three substances.A strong role of OC concerning the sorption of PFASs was also reported by other studies (e.g., Knight et al., 2019;Nguyen et al., 2020;Oliver et al., 2020).In our study, OC was not a significant predictor of sorption for any of the three investigated PFCAs (i.e., PFOA, PFDA, PFUnDA), nor for PFHxS.Li et al. (2019) proposed that the protein component of soil organic matter might be a better predictor of PFAS sorption than OC.However, our results indicate that total N was not a better predictor than OC for PFAS sorption (Fig. A12), which does not support this hypothesis.
PFOS was sorbed approximately twice as strongly in the fieldcontaminated soil S3 (Arboga field-contaminated) compared to in the laboratory-spiked soil S5 (Arboga; Fig. 1, A5 and A6).Considering the somewhat higher organic carbon and clay contents in the fieldcontaminated soil (Table 1), the results suggest that the effect of aging in the field was not a major factor affecting the soil-water partitioning of PFOS.

Partitioning as predicted from modeled SOM net charge
The modeling approach of applying surface charge-dependent K OC values for a peat (Oe) material to account for pH-dependent PFAS binding (i.e., Δlog K d per pH unit) in mineral soils resulted in satisfactory agreement in only a few cases (Fig. 5,Fig. A.14).FOSA was followed by PFOA as the PFASs for which the model exhibited the best overall performance.However, in the large majority of soils, the K d values were considerably underestimated.The best model fit was observed for soil S2 (Paskalampa Bs); predicted K d values were, on averae, 2.5 (±2.5) times lower than those measured for PFOA, PFUnDA, PFHxS, PFOS, and FOSA (PFDA sorption was not predicted, as calibration data for the peat Oe material was not available).The largest deviations from the modeled partitioning were observed for the Arboga clay soils S3 (impacted by aqueous film-forming foam, AFFF) and S5.For these two soils, the model underestimated the partitioning by a factor of 40 (±2.5).The fact that the model underestimated the measured partitioning for all soils to varying degrees indicates that there is an important retention mechanism, or possibly an unknown interaction effect, that is not captured by the model.One such mechanism could be the involvement of hydrophobic sorption to silicate minerals, as reported for C 7 -C 11 PFCAs and PFOS in isolated systems of silica particles (Shafique et al., 2017;Tang et al., 2010).To explore this possibility, for PFOA and PFHxS, we included a hypothetical pH-independent sorption contribution of a silicate phase with K d = 4.0 ± 2.0 L kg soil − 1 , where the exact value was allowed to depend on the soil and the PFAS compound.This greatly improved the overall model fit for these two substances (dashed lines in Fig. 5,Fig. A.15).However, without direct evidence, the assignment of a silicate sorption contribution remains speculative.Alternatively, the possible presence of a high-affinity sorbent component, such as black carbon or other pyrogenic carbonaceous materials (Zhi and Liu, 2018), may produce the observed difference between predicted and measured partitioning.However, this would require that such high-affinity phases were present in all mineral soils (possibly excluding soil S2), but not in the peat (Oe) material used for calibrating the model, which is less likely.

Conclusions
The overall sorption of PFASs was positively correlated with the perfluorocarbon chain length of the molecule, indicating that hydrophobic sorption to the solid phase was the main mechanism responsible for PFAS binding.Furthermore, the binding of all PFASs was inversely proportional to the solution pH, indicating that electrostatic interactions with the solid phase play a significant role in their binding.The pH dependence of sorption was greater for longer-chained PFASs compared to shorter-chained substances, suggesting that the perfluorinated tail, as well as the head group, affects PFAS sorption through electrostatic interactions with the sorbent.
The sorption of PFASs to mineral soils was considerably underestimated when extrapolated from organic soil materials using OC content or SOM net charge.Thus, a model that predicts PFAS sorption solely from OC characteristics is unlikely to be successful for mineral soils.Minerogenic components also appear to contribute significantly to PFAS sorption in mineral soils, but the magnitude and mechanisms require further investigation.More research is needed to understand the PFAS sorption mechanisms to enable the development of geochemical models capable of simulating and predicting the binding of PFASs in mineral soils.

Credit author statement
Hugo Campos-Pereira: Methodology, Investigation, Writingoriginal draft, Visualization, Formal analysis, Writingreview and editing; Dan B. Kleja: Conceptualization, Writingreview and editing, Supervision.Lutz Ahrens: Methodology, Resources, Writingreview and editing, Supervision; Anja Enell: Investigation, Writingreview and editing; Johannes Kikuchi: Investigation, Writingreview and Fig. 5. Measured soil-water partitioning (points) for selected soils and partitioning predicted from SOM net charge (solid lines).For PFOA and PFHxS, partitioning was also predicted from the sum of SOM net charge-normalized sorption and a pH-independent sorption contribution (dashed lines).

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.

•
The pH dependence of sorption increased with the length of the perfluorocarbon chain.•Organic carbon (OC) alone was a poor predictor of PFAS sorption.• Multiple regression suggested sorption contributions from OC and metal oxides.• A model predicting PFAS sorption from net organic charge underestimated K d .
and Fig. A.12), indicating that extrapolating K OC values from organic soil materials would likely underestimate the binding of the PFASs to mineral soils.For PFOS and FOSA, multiple linear regression (MLR) models explained a larger portion of the variance in the K d values at pH 5.6 ± 0.1 (Fig. 4, Fig A.14).Again, this suggests that multiple soil properties play a role in the sorption of PFOS and FOSA, as previously shown by Nguyen et al. (2020), Umeh et al. (2021) and Knight et al. (2021) for

Fig. 2 .
Fig. 2. Slope of the log K d -pH relationship as influenced by the length of the perfluorinated carbon chain.For each PFAS, the average of the regression slope (Δlog K d per pH unit) for all 10 laboratory-spiked soils is shown.Error bars represent the standard error of the mean.*p ≤ 0.05.

Fig. 4 .
Fig. 4. A) t-values of soil properties of significant multiple linear regression (MLR) models describing the K d values at pH 5.6 ± 0.1 in 10 laboratory-spiked soils.Adjusted r 2 values for MLR models are given in parenthesis.*p ≤ 0.05, **p ≤ 0.01, ***p ≤ 0.001.All the partial slopes of the MLR models were significantly different from zero (p ≤ 0.05).No significant relationships were found for PFOA, PFUnDA and PFHxS, why these substances are not included in figure.B) Fits of statistical models (log-log scale, pH 5.6 ± 0.1), shown as predicted K d plotted against measured K d for PFDA, PFOS and FOSA.

Table 1
Selected properties of soils S1-S11 that were used for the experiments.