Worst-case character of the railway soils
The soils used in this study exhibited high sand (61-84%) and low clay (5-15%) contents and, consequently, low water holding capacities (0.2-0.3 g/g, Table 1), properties which lead to rapid vertical transport of water. Similar soil textures were also assumed in three scenarios that were developed for authorization of herbicides on railway tracks in Germany [25]. These scenarios were implemented in the computer model PELMO that simulates possible leaching of pesticides in soil.
The pH-values of our railway soils were high (7.5-7.7) so that weak acids (e.g., quizalofop-P-acid, 2,4-D, or 2,4-DCP) were present in their more mobile, anionic form. Sorption is therefore expected to be weaker than in acidic soils. Furthermore, the organic carbon contents of <0.06-0.31% (Table 1) were among the lowest values reported for subsoils below railway tracks (0.06-2.3% for the fine material) [15, 25-27]. Low organic carbon and low clay contents are expected to result in weak sorption of most organic compounds.
Finally, the basal respiration, a measure for the microbial activity of soils, varied from 0.07 to 0.11 mg CO2-C kg-1h-1 (Table 1) and was thus in the range of values reported for Swedish railway soils (0.01-0.32 mg CO2-C kg-1h-1) [27], but clearly lower than in agricultural topsoils [28].
Overall, the soils selected represent a realistic worst-case regarding sorption and degradation of organic compounds.
Degradation of 2,4-D and its metabolites in railway soils
Degradation of 2,4-D differed considerably in the four soils (Fig. 2, left). In soil MR and the crushed sand (CS), a pronounced lag-phase was observed with no or very slow initial degradation, followed by a sharp decline in concentration after about 100 and 50 d, respectively. An initial phase with slower degradation was also found in soil MBS (and, to a certain degree, in EB).
Bi-phasic degradation of pesticides in soil incubation studies normally goes along with a deceleration over time [23]. Occasionally, also an increase of the degradation rate is observed [29, 30]. Microorganisms may require a certain adaptation time to cope with contaminants such as 2,4-D. When transferring the soils from the railway tracks into the laboratory environment, the conditions for microorganisms are changed. In particular, the soil structure is disturbed by sieving, with a possible negative impact on microorganisms. After some time, however, microorganisms may recover and eventually be capable of degrading 2,4-D.
On the other hand, the observed sharp decline may also be the result of another laboratory artifact. Microorganisms, which were originally not present in the soil, may have been introduced during collection, sieving, sampling, or moisturizing, which may have altered the degradation behavior. It is thus not clear whether such a pronounced increase in the degradation rate of 2,4-D would also occur under natural conditions in soils below railway tracks.
In the kinetic analysis, we considered the slow initial phase as representative for natural conditions, which resulted in conservative estimates of the DT50 and DT90 values in the respective soils (Table 2, top; DTx is defined as the time required for x% dissipation). The DT50 values in this initial phase ranged from 39 d in MBS to 203 d in CS, while in MR, no degradation was observed during the initial 84 d (>1000 d). For soil EB, even though visually bi-phasic, an SFO DT50 of the entire experimental duration was considered adequate.
Dissipation of 2,4-DCP, the primary metabolite of 2,4-D, when applied as test substance, was consistently faster, particularly in soils MBS and MR (Fig. 2, second panel). For 2,4-DCP, a bi-phasic decline was observed as well, but with the more typical deceleration over time. The dissipation curves were well described with the bi-phasic DFOP or FOMC model, resulting in DT50 values of 1-66 d (Table 2, top). DT90 values ranged from 7 to 519 d, indicating the clearly slower degradation at later time points.
DT50 values of 2,4-DCA, the secondary metabolite of 2,4-D, also varied considerably between soils, with very fast elimination in MR (DT50, 0.3 d), but slow elimination in CS (101 d). 2,4-DCA concentrations could be adequately described by the SFO model, except for soil MR, where a better fit was obtained with the DFOP model (Fig. 2, right; Table 2, top). DT90 values were between 2 and 335 d.
Comparison with agricultural soils
The European “FOrum for the Coordination of pesticide fate models and their USe” (FOCUS) recommends that, in case of bi-phasic degradation, only the slow phase should be considered when deriving DT50 values for groundwater assessments [23]. Furthermore, DT50 values shall be normalized to a reference temperature of 20°C and a reference moisture of pF2, and the geometric mean of the normalized DT50 values shall be used in pesticide leaching models [31]. Following these recommendations, we determined conservative geomean DT50 values of 115, 20, and 22 d for 2,4-D, 2,4-DCP, and 2,4-DCA (Table 2, top).
These values were then compared with DT50 values typically found in agricultural soils. For 2,4-D, numerous studies are available in public literature [18]. However, for a robust comparison, we used kinetic endpoints from studies that were performed for registration of 2,4-D in Europe [11] as these studies were evaluated according to the above cited recommendations of FOCUS. Degradation of 2,4-D was considerably slower in our railway soils than in these agricultural soils (on average by a factor of 26, Table 2, top), which was expected based on the fact that the microbial activity (approximated by the measured basal respiration) was at least one order of magnitude lower in the railway soils (Table 1) than in agricultural soils. Nonetheless, 2,4-D showed a remarkably slow degradation in our soils. To what extent this may be the consequence of the experimental conditions in the laboratory, cannot be answered here (see the above discussion on the lag-phase). Elimination of the metabolites 2,4-DCP and 2,4-DCA was, however, only 2-3× slower in the railway soils than in agricultural soils (again based on geomean DT50 values, Table 2, top).
Formation of 2,4-D metabolites
Highest amounts of 2,4-D metabolites were observed in soil EB, where 2,4-DCP reached 14% after 49 d and 2,4-DCA 2.6% after 120 d (Fig. 3, left; formation expressed in “parent” equivalents). In the other soils, metabolites were found in much lower amounts of ≈1% (2,4-DCP) and <0.5% (2,4-DCA). In incubation experiments with 2,4-DCP, the secondary metabolite 2,4-DCA was observed in amounts of up to 7% in EB (Fig. 3, second panel), 5% in MBS, and 1% in MR and CS. However, the methylation of 2,4-DCP to 2,4-DCA was reversible, particularly in soils MR and CS, with considerable formation of 2,4-DCP in incubation experiments with 2,4-DCA (up to 55 and 22%, respectively, Fig. 3, right).
Kinetic parameters for formation and further degradation of metabolites were fitted assuming sequential reactions (2,4-D → 2,4-DCP → 2,4-DCA, 2,4-DCP → 2,4-DCA, or 2,4-DCA → 2,4-DCP). Statistically significant DT50 values and formation fractions could be determined in a number of experiments and corresponding results are given in Table 2 (bottom). Some fits of experiments with notable metabolite formation are depicted in Fig. 3.
For the transformation from 2,4-D to 2,4-DCP, kinetic formation fractions of 34% and 11% were obtained for soils EB and MBS, respectively. The subsequent transformation from 2,4-DCP to 2,4-DCA occurred with highly differing formation fractions of 2-32%, again with the highest formation in soil EB. For the back reaction from 2,4-DCA to 2,4-DCP, only one statistically significant formation fraction could be derived, which was 79% in soil MR. Note that 2,4-DCP and 2,4-DCA formed from the respective precursors may have volatilized to some extent (see supplementary information) and formation fractions including volatilization may have been higher.
DT50 values resulting from sequential fitting generally were in good agreement with DT50 values determined in experiments with direct incubation of the respective compounds (Table 2, top and bottom). Differences were primarily observed when a bi-phasic model was used in direct incubation experiments and SFO kinetics in experiments with a precursor compound. Such differences are commonly observed, even if the same kinetic model is used for fitting. Half-lives may indeed be different, whether a compound is directly spiked to a soil or formed in soil from a precursor [32, 33].
In the kinetic analysis of the experiments with 2,4-D and its metabolites, the reversibility of the reaction from 2,4-DCP to 2,4-DCA should, in principle, be considered. However, in many experiments, formation of at least one of the two metabolites was too low for such a kinetic analysis. In other experiments, the kinetic analysis did not result in statistically significant fitting parameters – for example, in incubation experiments with 2,4-DCP and 2,4-DCA in soil MR. Note that the model Aquasim [34] was used to consider the interconversion of the two metabolites, with a similar approach as in [32, 35].
Degradation of quizalofop-P-ethyl and its metabolites in railway soils
Degradation curves of quizalofop-P-ethyl (QE), quizalofop-P-acid (QA), 3-OH-QA, and 3-OH-CQO in the four railway soils are shown in Fig. 4. Most experiments were better fitted by a bi-phasic model, except for 3-OH-QA where SFO fits were acceptable. The resulting DT50 and DT90 values are listed in Table 3 (top).
Degradation of QE was extremely fast in soils MBS, MR, and CS with DT50 values <1 d, but clearly slower in soil EB with a DT50 value of 21 d (Fig. 4, left). The bi-phasic degradation was most pronounced in soils EB and CS with high DT90 values of >1000 d and 68 d, respectively. Degradation of the three metabolites was consistently slower. The primary metabolite QA was degraded with a DT50 value of ≈100 d in soils EB and MBS. In CS, the DT50 value was ≈1 year, and in MR, no degradation was observed (Fig. 4, second panel). Degradation curves of the secondary metabolite 3-OH-QA were qualitatively similar to QA, but half-lives were 2-3× lower (35-630 d, Fig. 4, third panel). Finally, the tertiary metabolite 3-OH-CQO was degraded with DT50 values of 24-474 d. In comparison to the other metabolites, a slower degradation was observed in soil EB (Fig. 4, right).
As for 2,4-D and its metabolites, no clear correlation could be identified between soil parameters and rate of degradation. Degradation of all QE metabolites was slowest in soil MR (Fig. 4). Astonishingly, degradation of the parent compound itself was fastest in this soil. Abiotic hydrolysis of QE to QA may have been important in this soil. The geometric means listed in Table 3 (top) were again calculated from normalized DT50 values, considering only the slow phase in case of bi-phasic kinetics. These DT50 values were 4-18× higher than those reported for agricultural soils [14].
Formation of quizalofop-P-ethyl metabolites
Quizalofop-P-ethyl is a so-called pro-herbicide, which is, due to its lipophilicity, more readily taken up through the cuticle of leaves. As with other “FOP” herbicides [20], the compound is rapidly transformed in plants to the corresponding acid, which is the herbicidally active substance.
In the railway soils, quizalofop-P-ethyl was also quantitatively transformed to quizalofop-acid (Fig. 5, left). However, the secondary and tertiary metabolites, 3-OH-QA and 3-OH-CQO, were formed in much lower quantities (≤1%, shown for soil EB in Fig. 5, second panel), also in experiments with QA (third panel). Only in experiments with incubation of 3-OH-QA, 3-OH-CQO reached amounts of up to 29% after 4 months (Fig. 5, right).
Sequential fitting confirmed the quantitative transformation of QE to QA (formation fractions, 98-100%). For the subsequent transformation of QA to 3-OH-QA, formation fractions of 0.5% and 3.0% were determined in soil EB and MBS, respectively (Table 3, bottom). For the transformation of 3-OH-QA to 3-OH-CQO, no statistically significant formation fraction could be determined, but in soil EB, it must have been >29% (maximum formation at the end of incubation).
In degradation studies with agricultural soils, formation fractions of 70-100%, 32-76%, and 100%, respectively, were determined for the reaction sequence QE→QA→3-OH-QA→3-OH-CQO [14]. Hydroxylation of QA to 3-OH-QA thus seems to be clearly less important in railway soils than in agricultural soils.
Adsorption to railway soils
Adsorption experiments were evaluated with the Freundlich model [36],
where cw is the concentration in the aqueous phase and cs the concentration in soil. The Freundlich adsorption coefficients (KF) and the Freundlich exponents (1/n) were determined from linear regressions of log cw vs log cs. These so-called Freundlich isotherms for 2,4-D, QE, and their metabolites in the four railway soils are shown in Figs. 6 and 7, and corresponding KF and 1/n values are listed in Tables 4 and 5.
The compounds exhibited strongest adsorption to the construction material from site EB and weakest adsorption to subsoil from MR, except for 3-OH-CQO, which showed strongest adsorption to subsoil MBS and weakest adsorption to the crushed sand (CS) (Figs. 6 and 7). Adsorption of QE (KF, 1.5-66 mL/g) was about two orders of magnitude stronger than adsorption of 2,4-D (KF, 0.04-0.28 mL/g), which may be assumed to be present as a carboxylate anion in the railway soils (pKa value, 3.4) [11]. KF values of the metabolites were in between those of the two parent compounds. The weakest adsorption of 2,4-D was found in soil MR, where the adsorbed fraction represented only 3-18%, depending on the concentration level. In such cases, determination of the concentration in soil is less accurate, resulting in higher confidence intervals (Tables S6 and S7). For a precise determination of KF values, a higher fraction of adsorbed test compounds would have been desirable.
Geometric mean KF values are currently used as input values for groundwater assessments in the context of registration in Europe [31]. These geomean adsorption coefficients can be compared with typical values for agricultural soils (Tables 4 and 5). For this comparison, we only relied on studies submitted in the context of registration [11, 14]. Based on geomean KF, adsorption in the railway soils was clearly weaker than in agricultural soils, by a factor of 3-19.
Adsorption was non-linear for all compounds, with mean Freundlich exponents of 0.62-0.80 (Tables 4 and 5), indicating weaker adsorption at higher concentrations. The mean 1/n values were consistently lower in railway soils than in agricultural soils, i.e., non-linearity of adsorption was more pronounced.
Estimation of adsorption and degradation in railway soils from data with agricultural soils
Since adsorption and degradation data are usually not available for railway soils, the question may be posed whether such data can be estimated by extrapolation from agricultural soils. Sorption to organic matter is often considered as the predominant sorption mechanism for organic compounds in soils [37], i.e., it is assumed that KF values are roughly proportional to the organic carbon content of soils (Corg). KF values for railway soils (r) may thus simply be estimated from KF values for agricultural soils (a), correcting for Corg:
With this formula, we calculated KF(r) values for the seven test substances in this study (2,4-D, QE, and their metabolites), based on the geometric means of KF(a), Corg(a), and Corg(r) (Tables S6 and S7). These estimates were generally lower (1.5-7.6×) and thus more conservative than our measured adsorption coefficients, except for 2,4-DCP, where estimated and measured KF values were almost equal. In the railway soils, sorption to mineral substrates may be important as well and the above formula does not account for their contribution to sorption.
Freundlich exponents (1/n) for agricultural soils were consistently higher than those determined in this study for railway soils (Tables 4 and 5), i.e., the use of 1/n data from agricultural soils would also be more conservative regarding the prediction of leaching in soils. Overall, the estimation of sorption in railway soils from data with agricultural soils using the above approach seems to be conservative, in some cases very conservative. Sorption parameters measured specifically in railway soils may thus not be essential for a first tier groundwater assessment. However, it needs to be highlighted that this conclusion is based on only seven test substances.
Degradation half-lives in railway soils may be estimated in a similar way by extrapolation of data from agricultural soils, assuming that biological activity is linked to the organic matter content of a soil. This approach was, for example, implemented in the HardSPEC model, a simple tool developed in the UK for estimating surface water and groundwater exposure resulting from herbicides applied to hard surfaces [38]:
DT50(r) values estimated with this formula (excluding 2,4-D) yielded 1.5-8.2× higher values than actually measured (Tables S4 and S5), i.e., as for adsorption, this approach tends to be conservative as well. For 2,4-D, the estimated mean half-life was ≈2× lower, when considering only the initial slow phase of the degradation curves.
Concerning formation of metabolites, our experiments with railway soils showed substantial differences to agricultural soils. Generally, metabolites were formed in lower amounts. In particular, formation of the secondary and tertiary metabolite of QE (3-OH-QA and 3-OH-CQO) seems to be unimportant in railway soils.