Determination of polyoxymethylene (POM) water partition coefficients for DDT and its degradation products, with inter-laboratory comparison of the passive sampling methodology and bioaccumulation in earthworm (Eisenia fetida)

K_POM was determined for 10 DDX: o,p′-DDT, p,p′-DDT, o,p′-DDD, p,p′-DDD, o,p′-DDE, p,p′-DDE, p,p′-DDMU, p,p′-DDM, p,p′-DBP and dicofol (Supplementary Table S1). Fortification standards of native DDX compounds were prepared in methanol to obtain four mixtures (C_f1–C_f4; Supplementary Table S2). The aqueous solutions for the POM–water experiments were prepared using Milli-Q Ultrapure water, NaN₃ and salts of NaH₂PO₄ and Na₂HPO₄. Two internal standard (IS) toluene solutions were prepared: (i) ¹³C-labelled o,p′-DDT, p,p′-DDT, o,p′-DDD, p,p′-DDD, o,p′-DDE and p,p′-DDE and (ii) deuterium-labelled dicofol (Supplementary Table S3). ¹³C-labelled PCB101 was used as a recovery standard (RS) (Supplementary Table S3).

The influence of analyte concentrations on K_POM was tested using 28-day end-over-end tumbling (Gerhardt Laboshake, Germany) with four different DDX concentration levels, chosen to cover the expected range of concentrations to be found in soils at historical contaminated sites but not exceeding aqueous solubilities: C1, C2, C3 and C4 in methanol (Table 1). The time required for the analytes to reach equilibrium between the aqueous phase and the POM (at pH 7.0), and the potential effect of pH (at 28 days tumbling), were investigated using the C2-level concentration, at tumbling times of 3, 7, 14, 28 and 56 days, and in buffer solutions titrated to pH 4.1 and pH 8.4 with 3 mol L^–1 HCl and 1 mol L^–1 NaOH (Table 1). All tests were performed in triplicate.

In brief, 76-μm-thick POM (CS Hyde, Lake Villa, IL, USA) was cut into 4- × 5-cm strips, weighed and placed in a 100-mL bottle with 100 g of 1 g L^–1 NaN₃ and 10 mmol L^–1 PO₄-buffer solution. For tests using concentration levels C1 and C2, 50 µL of the DDX fortification mixture (C_f1–C_f2; Supplementary Table S2) was added. For concentration levels C3 and C4, a staggered fortification approach was used to avoid exceedance of water solubilities. Once fortified, bottles were tumbled end-over-end. At the end of each test, POM strips were removed, wiped dry and placed in glass amber vials. POM strips and sample solutions were stored at +4°C until analysis.

Water samples were extracted twice with n-hexane using liquid–liquid extraction, with IS added prior to extraction. Extracts were combined, evaporated to 2 mL, treated with anhydrous Na₂SO₄ and concentrated to 100 µL. RS solution (Supplementary Table S3) was added, the solvent exchanged to toluene, and the sample stored in gas chromatography (GC) vials at −20°C for gas chromatography–mass spectrometry (GC-MS) analysis. POM strips were extracted twice with 40 mL of acetone–n-hexane (1:1, v/v) using ultra-sonication, with IS added beforehand. Combined extracts were evaporated to ~0.5 mL and extracts of low concentrations (C1, C2) were cleaned up with deactivated silica. All extracts were concentrated to 100 µL, solvent was exchanged to toluene, RS was added, and the sample stored at −20°C for GC-MS analysis. Further details are in ‘POM Concentrations, C_POM (Part 1)’ in the Supplementary material.

All measurements were performed in the selected ion monitoring mode. Identification and quantification of the target compounds were done using quantification standard solutions that included all compounds, in addition to IS and RS. More details are given in ‘POM Concentrations, C_POM (Part 1)’ in the Supplementary material.

Target compounds were quantified by use of a 10-point calibration curve. If a calibration point was below 80% or above 120% of expected linearity, the point was excluded to a minimum of eight calibration points. Quantification standards were analysed after every tenth sample to monitor instrumental performance and ensure calibration validity, with solvent blanks run at the start and end of each sequence to check for carry-over effects. Liners were replaced after 50 injections or upon sign of decomposition. Concentrations were calculated using isotope dilution. Owing to lack of an IS for p,p′-DDMU, p,p′-DDM and p,p′-DBP, calculations were based on 13C-o,p′-DDE peak areas (Supplementary Table S3). The limit of detection (LOD) was defined as the mean blank concentration plus 3× standard deviation (s.d.) and, in the absence of blank contamination, the LOD was set to the lowest concentration of the calibration curve. Samples with concentrations exceeding the range of the calibration curve were diluted, re-spiked and reanalysed. Sample blanks (n = 3) and process blanks (aqueous and POM) were processed in parallel to the experiments.

No DDX compounds were detected in the blanks, except for p,p′-DDMU (found in both sample blanks and process blanks), traces of p,p′-DDM (in one sample blank) and p,p′-DBP (in one process blank) (Supplementary Table S4). If the amount in the blank was >10% of the amount in the corresponding sample, the data for this sample was excluded from further evaluation (Supplementary Table S5).

Since the concentrations of DDX in both the POM and the water phase were measured, K_POM can be obtained even if there was a mass loss of the analyte during the experiment (e.g. volatilisation, transformation, glassware sorption). Nevertheless, to ensure good data quality, mass balance calculations were performed (amount in water + amount in POM) ÷ added amount), and if the recovery was <50 or >130%, the data were excluded (Supplementary Table S6). To check for potential degradation of DDT (to DDD) and dicofol (to DBP), which can occur in the GC injector (Foreman and Gates 1997; Fujii et al. 2011), mass balance calculations (recovery, i.e. added v. measured amounts) were compared between p,p′-DDT and ∑p,p′-DDT/D, o,p′-DDT and ∑o,p′-DDT/D, and dicofol and ∑p,p′-dicofol + DBP. These calculations indicated very low, or acceptably low, degradation in all experiments, except for dicofol at pH 8.4 as expected, due to the instability of this substance under alkaline conditions (Yin et al. 2017). The recovery was generally higher for ∑DDT/D than DDT, but the difference between DDT and ∑DDT/D recoveries were on average only 7% (s.d. ± 3) and 10% (s.d. ± 3) for p,p′- and o,p′-DDT/D respectively, which is well below the 20% threshold set by the United States Environmental Protection Agency (1995) guidelines for methods for the determination of organic compounds in drinking water. The difference between the recovery of dicofol and ∑p,p′-dicofol + DBP were on average 16% (s.d. ± 16), indicating a higher degradation of dicofol to DBP than DDT to DDD.

Statistical comparisons between treatments were performed in JMP (ver. 17, JMP Statistical Discovery LLC, NC), using ANOVA followed by Tukey’s HSD test at α = 0.05. Values given in the text are arithmetic means ± s.d.

Soil samples were collected at a depth of 0–20 cm. The samples were sieved (<2 mm) and homogenised in the field before dividing into two subsamples sent to Lab A and B. After arrival at the lab, the soil sample was further homogenised and divided into subsamples for POM tests and analyses for soil characterisation. Details on the sampling, sample preparation and determination of soil properties (total organic carbon, TOC, content; pH; water holding capacity, WHC; and grain size distribution) and levels of potential soil contaminants other than DDX (pesticides and metals) are described in section ‘S1.3.1 Soil sampling and analyses of soils’ in the Supplementary material. Soil properties are listed in Table 2. None of the soils contained elevated levels of inorganic contaminants, but low to medium levels of pesticides other than DDT were found at some sites (Supplementary Table S8).

The tests were performed the same way in Lab A and B. In brief, 76-μm POM strips (4 × 5 cm) were placed in amber glass bottles together with 25 g of homogenised soil (calculated as dry weight, DW) and then mixed with 93 mL of water containing CaCl₂ and NaN₃. Vials were tumbled end-over-end for 28 days in the dark. Thereafter, the POM strips were removed, rinsed with ultrapure water, and wiped dry before being placed in clean scintillation vials and frozen (−20°C) until extraction.

To study variance caused by differences in the extraction and analysis steps of the two labs and thereby exclude any other variance (e.g. caused by heterogeneous contaminant dispersion in the soil samples and differences in conducting the shaking experiment), six of the POM strips at each lab were cut into halves (‘I’ and ‘II’) after the shake-test (in total 24 halves). The ‘I’ halves (n = 12) were then extracted and analysed by Lab B whereas Lab A extracted and analysed the ‘II’ halves (n = 12) (Supplementary Table S9).

The C_W,free (ng L^–1) of the DDX were calculated using the derived K_POM (L kg^–1 POM) (part 1 of this study) and Eqn 2:

where C_POM is the concentration of DDX in the POM strip (ng kg^–1), calculated from the amount of DDX in the strip divided by the weight of the POM strip (Supplementary Table S9).

To avoid underestimation of C_W,free during equilibrium passive sampling, a criterion of <5% depletion of the HOC from the soil or sediment has been recommended (Mayer et al. 2014). With the exception of dicofol, the highest depletion (defined in the Supplementary material, see ‘S1.3.3 Control of depletion of DDX from the soil during the POM-test’) reported for Lab A and B was 31 and 36%, both found for o,p′-DDE, for which the C_soil was close to the limit of quantification (LOQ). Soils with TOC content >2% came close to meeting the 5% criterion for all DDX (Supplementary Table S10), except for dicofol analysed by Lab A, for which the depletion was 35%. Lab A also obtained very high depletion of dicofol (85 to >100%) for four forest plant nursery soils, whereas Lab B received ≤17% for all soils. Dicofol in these four samples was thus excluded from further evaluation in both Parts 2 and 3.

POM strips were extracted as described above (Lab A) with minor adjustments for Lab B; using 24 h of shaking at low speed instead of sonication, and isooctane instead of toluene as final solvent. Lab A performed clean up of all POM extracts using deactivated silica. More details are given in ‘S1.3 Interlaboratory comparison of the POM-method (Part 2)’ in the Supplementary material.

Lab A analysed soils and POM extracts using GC-MS as described in Part 1. Lab B analysed soil and POM extracts using GC-MS/MS and multiple reaction monitoring (MRM) mode. Identification and quantification were performed using authentic reference standards (Supplementary Table S3). For data evaluation, the software Agilent MassHunter Quantitative Analysis (for QQQ) was used.

In Lab A, the QA+QC of the GC-MS analysis is described in Part 1. POM sample blanks (i.e. POM in aqueous solution, with no soil added; n = 3) and a process blank per nine POM samples were included in the sample preparation stage and run together with the POM samples. Three process blanks were included in the preparation and analysis of the soil samples. The LOQ was defined as the mean blank concentration plus 10× s.d. and, in the absence of blank contamination, the LOQ was set to a signal-to-noise (S/N) ratio of 10. The average + s.d. of the recoveries for IS added to the POM samples was 98 ± 24% (n = 252). For IS added to soil samples, the average + s.d. of the recovery was 103 ± 18% (n = 63).

In Lab B, POM sample blanks (i.e. POM in aqueous solution, with no soil added; n = 3) and a process blank per six POM samples were included in the sample preparation stage and run together with the POM samples. Target compounds were quantified using an 11-point calibration curve. If a calibration point was below 80% or above 120% of expected linearity, the point was excluded to a minimum of seven calibration points. Concentrations of target compounds were calculated as described for Lab A. If the amount in the blank was 10% of the amount in the corresponding sample, the sample was excluded from further evaluation. Recoveries for IS added to the POM samples were occasionally found to be high, with six exceptional values. Excluding these six, the average ± s.d. was 106 ± 55 (n = 246). The average ± s.d. of the recovery for the soil samples was 104 ± 58 (n = 98). When samples were below the LOQ, replacement values were used for statistical analysis. The LOQ replacement values were half of the lowest point used on the calibration curve. If the compound was below the limit of detection (S/N ratio <3), the replacement value was three times the value reported, divided by the S/N ratio.

The statistical evaluation was performed using Excel (ver. 2407 Build 16.0.17830.20210, Microsoft) and the Excel add-in statistical program Data Analysis Toolpack.

No correlation between pH and log K_POM was observed (Supplementary Fig. S1). There was an increase in mean K_POM between pH 7.0 and 8.4 for p,p′-DDM and p,p′-DDT, but the reverse was found for p,p′-DDD, and no other significant differences were observed at P < 0.05 (Supplementary Table S14). The pH effect was not evaluated for p,p′-DDMU and dicofol due to lack of data (blank/sample ratios >10% and C_W < LOD respectively). Given the small sample size (n = 3) and varying results, no trend was deemed strong enough to separate the data set. Thus, data from the tests at the three pH levels (pH 4.1, 7.0 and 8.4) were combined into one data set for the evaluation of equilibrium time and the potential effect of analyte concentrations on K_POM.

At 28 days, after shaking POM with DDX-spiked water (using the C2 concentration), a steady state indicating chemical equilibrium (i.e. no apparent change in K_POM values over time) was achieved for all the studied compounds (Fig. 1). Previous kinetic studies on POM of 76-µm thickness have shown that HOC with log K_OW < 4.5 can reach equilibrium with POM after 14 days of shaking, whereas compounds with higher hydrophobicity (log K_OW between 4.5 and 6.8) may need approximately double the time (Hawthorne et al. 2011; Josefsson et al. 2015). The same trend was observed in our experiment, where compounds with high K_OW values seemed to need longer to reach equilibrium, although most of the compounds appeared to have reached equilibrium already at 14 days. For example, p,p′-DDM, with the second lowest K_OW value of the studied DDX (log K_OW 5.3), may have reached equilibrium already after 7 days, whereas p,p′-DDE, which has the highest K_OW value (log K_OW 6.9) needed at least 28 days (statistics shown in Supplementary Table S15). There was also no difference between mean log K_POM values for p,p′-DDD, and o,p′- and p,p′-DDT derived at 14 and 28 days, indicating that 14 days of shaking will be enough for these compounds too. Dicofol (log K_OW 5.8) and DDMU (log K_OW 6.2) could not be statistically evaluated due to lack of data at several or all time points (Supplementary Fig. S2), but as the hydrophobicity of these compounds are lower than many of the other DDX compounds studied, they are assumed to reach equilibrium within the same time frame (i.e. within 28 days). Consequently, it was concluded that an exposure time of 28 days should be sufficient to reach equilibrium for all the investigated DDX. For evaluation of the potential effect of analyte concentrations on K_POM, the replicates with equilibration times of 28 and 56 days were used.

Fig. 1.

Mean log K_POM for the DDX as a function of tumbling time (days) derived for the C2 concentration level. Error bars represent the standard deviation (n = 3). Green circles, blue triangles, and red crosses indicate the pH of the test; 7.0, 8.4 and 4.1 respectively. Dicofol was excluded due to lack of data. DDMU is shown in Supplementary material; Fig. S2.

There were no consistent trends in K_POM values for three of the four tested concentrations (using the 28- and 56-day replicates; Supplementary Fig. S3 and Supplementary Table S16). The lowest concentration level (C1) gave, however, a noticeably lower mean log K_POM for the DDDs and DDEs but not for the DDTs (the other DDX could not be evaluated at this level due to data not meeting quality requirements or having C_W < LOD). These lower K_POM values at the C1 level for some compounds were likely due to overestimation of areas for peaks with low signal-to-noise ratios. The DDTs had 50% higher starting concentrations than DDDs and DDEs in C1 (Table 1) and should theoretically reach a higher C_W at equilibrium than the compounds with higher hydrophobicity (i.e. o,p′- and p,p′-DDE). Since the opposite was observed (C_Wo,p′-DDE > C_Wo,p′-DDT and C_Wp,p′-DDE ≈ C_Wp,p′-DDT (Supplementary Table S11), this indicates overestimation of peak area, leading to overestimation of C_W at this very low concentration level. Therefore, it was decided to exclude C1 data for DDE and DDD in the final determination of log K_POM.

The mean K_POM values obtained at C3 did not differ significantly from those obtained at C4 for all compounds that could be evaluated at these two levels (i.e. dicofol, p,p′-DDM, and p,p′- and o,p′-DDE). Between C2 and C3, o,p′- and p,p′-DDE had an increased log K_POM by 0.38 and 0.39 log units respectively. Between C2 and C4, there was a slightly lower mean log K_POM for p,p′-DDM (−0.24 log unit), but a slightly higher value for p,p′-DDE (+0.25 log unit). Regressions of log C_POM v. log C_W for the compounds with valid data (i.e. data that met our stated quality criteria) from three C levels (Supplementary Fig. S4) gave linear isotherms, with r² values of 0.99, 0.97 and 0.96, for p,p′‑DDM, p,p′‑DDE and o,p′‑DDE respectively. Hence, we concluded that K_POM is not dependant on the analyte concentrations within our investigated interval. Since no general trend with concentration levels was observed, data from all concentration levels were used to calculate the final K_POM values.

The final log K_POM values for the 10 DDX compounds studied are listed in Table 3. Our measurements of log K_POM were highly repeatable (relative standard deviation, r.s.d., was 0.9–3.1%) except for p,p′-DDMU (r.s.d. = 16%), for which the final mean log K_POM was based on only three valid measurements. The final K_POM values for p,p′-DDT and p,p′-DDE are in the same range as values previously reported by Endo et al. (2011), with a deviation of −0.3 and +0.3 log units (i.e. a factor 0.5 and 2) respectively, but with better or equal r.s.d. (Table 3).

The relationship between the final log K_POM and the log K_OW of the DDX (DDMU excluded) conformed to a linear regression (log K_POM = 0.44 × log K_OW + 2.6) with an explanatory power of r² = 0.81 (Supplementary Fig. S5). This regression has a lower slope (and different intercept) than the single-parameter linear free energy relationship (SP-LFER) model derived by Endo et al. (2011), which was based on log K_POM values for 43 compounds (not including any organochlorine pesticides (OCPs). Endo et al. also derived K_POM for 13 OCPs, including p,p′-DDT and p,p′-DDE, and noted that these deviated from the SP-LFER model; their experimentally determined K_POM values for p,p′-DDT and p,p′-DDE were lower (Table 3). However, their determined K_POM values matched well with our final values, and our regression between log K_POM and log K_OW displayed a good linearity. The K_OW values used here are from the SPARC model (ARCHem, see http://www.archemcalc.com/HTML/index.html). Few reliable experimental log K_OW data exist for DDX (Pontolillo and Eganhouse 2001). The choice of using SPARC is based on the results in Eganhouse et al. (2018), who evaluated five of the most widely used and currently available computational methods and concluded that SPARC provided results that best matched the few available controlled experimental DDX data. However, we note that values derived from SPARC in 2024 gave higher values for o,p′-DDT (+0.8 log unit) and p,p-DDT (+0.9 log unit) compared to those reported by Eganhouse et al. (2018).

As experimental data are generally considered more reliable than predicted data, we recommend using the experimental final log K_POM values rather than the ones derived from log K_POM − log K_OW regressions. The K_POM for DDMU should, though, be used with caution as the value is based on only n = 3 and the r.s.d. is high.

Other equilibrium passive sampler materials have been used for the determination of C_W,free of DDX, such as low-density polyethylene (PE) (Hale et al. 2009; Borrelli et al. 2018) and poly(dimethylsiloxane) (PDMS) (Maruya et al. 2009; Xing et al. 2009; Eganhouse 2016). The precision we obtained (s.d. < 0.18 and s.e. < 0.05 log units, excluding DDMU), was of equal quality to data compiled in the study by Eganhouse (2016), where s.d. was generally <0.2 for log K_PDMS/water derived from PDMS between 7 and 100 µm thick. It also compared well to data from Hale et al. (2009) where s.e. <0.16 for log K_PE/water, using 51-µm-thick PE. The higher s.d. of our DDMU data, along with the much lower K_POM value than K_{polymer/water} values reported by both Eganhouse (2016) and Hale et al. (2009) for this compound, indicates a potential underestimation. A comparison of our log K_POM data against log K_PDMS and log K_PE data from these studies is shown in Supplementary Fig. S6. In addition, all three materials are also regarded equivalent in terms of the affinity for hydrophobic non-polar chemicals (Endo et al. 2011). However, POM offers benefits when aiming to extract polar compounds from environmental phases since POM has H-bond acceptor sites in its molecular structure in contrast to PDMS and PE, and thus strong H-bond donor compounds generally have higher K_POM/water than K_PDMS/water and K_PE/water (Endo et al. 2011). Hence, POM is expected to be a more sensitive sorbent than PDMS and PE for, e.g. dicofol, which possesses an OH group, and thus has H-bond donor properties.

Soil concentrations of ∑p,p′-DDT/D, ∑o,p′-DDT/D, p,p′-DDE, o,p′-DDE, dicofol and p,p′-DBP are shown in Fig. 2 (C_soil for all individual compounds can be found in Supplementary Tables S17–S18). The C_soil values of p,p′-DDMU and p,p′-DDM were below the LOQ for all soils in Lab B’s analyses, and just above the LOQ in Lab A’s analyses. These two compounds were thus excluded from further evaluation. The mean C_soil (μg g^–1 DW) of ∑DDX of Lab A and B ranged from ~5 to ~30 μg g^–1 DW for eight of the nine studied sites, whereas one site (Åby) showed significantly higher concentrations (110 μg g^–1 DW). The most dominant DDX in all soil samples was ∑p,p′-DDT/D, which accounted for 60–76% of the total concentration, followed by ∑o,p′-DDT/D (10–23%), p,p′-DDE (2.1–18%), dicofol (0.77–8.0%), p,p′-DBP (0.30–3.7%) and o,p′-DDE (0.0006–1.9%) (Supplementary Table S19).

Fig. 2.

Concentrations (μg g^–1 DW) of DDX in soil samples from nine forest plant nurseries analysed by Lab A and Lab B. Åby concentrations as shown are divided by 10. Deje Nord in Lab B was analysed in triplicate (s.d. = 0.34 μg g^–1 DW).

Both labs quantified similar C_soil for ∑p,p′-DDT/D, ∑o,p′-DDT/D, p,p′-DDE and p,p′-DBP, with concentration ratios of the labs (C_soil,LabA/C_soil,LabB) between 0.5 and 2 (Supplementary Table S20). Higher ratios, up to 4.8, were found for dicofol, but with a mean of 2.3. For this compound, Lab A found mostly higher levels (9 out of 10 sites), whereas for p,p′-DBP, Lab B found higher levels in 4 of 10 sites. This implies that a higher level of degradation in the GC injector of dicofol to p,p′-DBP occurred in Lab B. However, Lab B seemed to have had less conversion of DDT into DDD compared to Lab A. Differences in extraction procedure, clean up method, reference compound solutions and instrumental analysis can also be reasons for the different C_soil, which are expanded upon in Materials and methods, Part 2. For o,p′-DDE, one outlier was identified (negligible levels detected at Lab B) and excluded from the comparison. Overall, results showed relatively good agreement with mean ratios (Supplementary Table S20) ranging from 0.8 to 2.3, and the precision of the analysis (Lab B triplicate) was high, with r.s.d. ranging from 2.2 to 11%, with a mean of 6.4%.

The levels of DDX in the POM strips (C_POM) are shown in Fig. 3, all C_POM data, means and standard deviations are in Supplementary Table S21, and ratios between C_POM analysed by Lab A and Lab B are in Supplementary Table S22. Lab A consistently quantified equal or higher C_POM than Lab B for all compounds. For the POM analyses, both labs experienced major challenges with very high MS signals. This required dilution of the extracts, further addition of IS and reanalysis, often more than once. This procedure is not optimal for good accuracy and is likely to have caused some of the observed differences between the labs. The C_POM mean ratios between labs were below 2, except for dicofol and p,p′-DBP with C_POM mean ratios of 35 and 10 respectively. Although the degradation of dicofol to DBP in the injector was deemed to be at an acceptably low level for soil samples, the ratios for POM suggest that this could have been an issue in these analyses. This highlights the importance of monitoring the degradation or adjusting the injection method to avoid such degradation, e.g. as suggested by Yin et al. (2017). Another potential explanation for the differences could be that ¹³C-labelled internal standards were not available for these two compounds; a deuterated internal standard was used for dicofol, and DBP was quantified using ¹³C-o,p′-DDE. In spite of analytical challenges, the precision of the analyses was good for both labs, with a mean r.s.d. of 5.1 and 11% respectively for Lab A and B (Supplementary Table S21).

Fig. 3.

C_POM values and s.d. (n = 3 except for Ljungaskog Lab B, n = 2) for nine different soils analysed by Lab A and Lab B. Åby values as shown are divided by 10 for easier visualisation.

The strips shaken in soil from Klockatorp and Ljungaskog (by both labs) were cut in half and sent to the other lab for DDX analysis (n = 3 for each soil except Klockatorp shaken by Lab B, n = 2). This way, there were no differences in homogenisation and subsampling of the soil, and the comparison was focused on the analytical procedure from POM extraction onwards. Concentrations, standard deviations and lab data ratios are shown in Fig. 4 and Supplementary Table S22. The results showed that analysing the same POM strip resulted in similar precision, with a mean r.s.d. of 4.9 and 11% for Lab A and B respectively (Supplementary Table S23) (5.1 and 11% in the full C_POM comparison). Lab ratios improved, however, some high ratios were still found. The C_POM results for both isomers of ΣDDT/D and DDE were not significantly different between labs (t-test: P > 0.05), with all ratios between labs ranging from 0.8 to 1.4 (as compared with 1.2–2.6). C_POM values for dicofol and p,p′-DBP, however, were significantly different for every strip for nearly every site, with Lab A consistently quantifying higher values than Lab B, i.e. the same trend as for C_soil. This could be attributed to the analytical procedure from POM extraction onwards. As the agreement between labs was good for most substances (ratios close to one), and only unacceptably high for the low level compounds, dicofol and DPB, this points to the instrumental analysis as the major cause for the differences, e.g. the need for several dilution steps and reanalysis, lack of some ¹³C-labelled IS, and the potential degradation of dicofol to DPB in the GC injector. The somewhat higher precision of Lab A might partly be attributed to well established homogenisation and subsampling procedures.

Fig. 4.

Mean and standard deviation of DDX concentrations in the POM (C_POM) in the focused interlaboratory comparison study. Samples were analysed in triplicate, except where indicated with an asterisk (*), where n = 2.

Based on the concentrations in the POM strips (C_POM), the C_W,free values for the nine field sites were determined (Supplementary Table S24 and Fig. 5) using K_POM established in Part 1 of this study (Table 3). The compound pairs p,p′-DDT/p,p′-DDD and o,p′-DDT/o,p′-DDD have different K_POM values, with DDTs having higher values. In our conversions of C_POM into C_W,free, we opted for using individual levels of DDT and DDD and their respective K_POM. Consequently, the following C_W,free results for DDT and DDD are approximations, with potential underestimation of DDT and overestimation of DDD. As we cannot be sure of the exact deviation from a true ΣDDT/D, the quantification results are to be seen as relatively valid with respect to each other, but not accurate. For every compound, Åby showed, as expected, significantly higher C_W,free than the other soils, due to the much higher C_soil, which reflects the land use history of the sampling area as a DDT dipping station for saplings. This, however, also resulted in higher variation in the analysis due to multiple dilution of the sample being necessary for the value to be within the range of the calibration curve. Åby also generally showed bad lab ratios for C_soil and C_POM. Although there was some variation in reported concentrations, the C_W,free data were significantly correlated between labs (P < 0.05, r² > 0.96 for log–log relationships) (Fig. 5), excluding p,p′-DBP (r² = 0.93) and dicofol (r² = 0.43) (Supplementary Fig. S5). Overall, the r.s.d. remained low for both labs (mean r.s.d. Lab A = 3.9%; mean r.s.d. Lab B = 10%).

Fig. 5.

Comparison of freely dissolved pore water concentrations (log values) of DDX compounds from nine sampling sites analysed by two labs.

Normalising worm DDX concentrations (C_worm) to the individual fat content of the worms resulted in lipid concentrations (C_{worm_lipid}) ranging for ∑DDX-10 from 558 to 3840 µmol kg^–1 lipid with Jakobsbyn at the low end and Sya at the top (Table 4). As mortality was 100% in the soil with the highest C_soil (Åby), no worms could be analysed for this soil. All C_worm and C_{worm_lipid} data are presented in Supplementary Tables S25 and S26.

log–log correlations between C_{worm_lipid} v. C_POM, C_soil and C_TOC (i.e. the soil concentration normalised to the fraction of TOC content; C_soil/f_TOC), were conducted for the ∑DDX-10 (Fig. 6), the individual DDX, ∑p,p′-DDT/D and ∑o,p′-DDT/D (Supplementary Table S27). The worst log–log correlations were observed for C_{worm_lipid} v. C_soil (r² of 0.60–0.93), and the best for C_{worm_lipid} v. C_POM (r² of 0.82–0.97) and C_{worm_lipid} v. C_TOC (r² from 0.89 to 0.97), excluding results for p,p′-DBP, the compound with the lowest K_OW value, which had r² of 0.84, 0.24 and 0.00012 for C_POM, C_TOC and C_soil respectively.

Fig. 6.

Correlations of log C_{worm_lipid} with various chemical measurements, including log C_POM (top left), log C_soil (top right) and log C_TOC (bottom left).

In addition to the strong linear log–log correlations between C_{worm_lipid} and C_POM, the slopes of these relationships were also close to one (between 0.80 and 1.05) for all DDX, except for dicofol and o,p′-DDD, which had slightly lower slopes of 0.66 and 0.77 respectively (Fig. 6 and Supplementary Table S27). The fact that the linear regression slopes were near one and exhibited better r² values than log–log correlations of C_{worm_lipid} v. C_soil demonstrates that the POM method has the capacity to accurately predict the bioavailability of DDX in soil and is superior for predicting bioaccumulation compared to analysing total DDX content in the soil.

The strong correlations and feasibility of using POM as a prediction tool for bioaccumulation align with other studies on POM and HOC bioaccumulation in earthworms. Arp et al. (2014) reported a log–log correlation (r² = 0.94) for PAH bioaccumulation by Enchytraeus crypticus using 76 μm POM membranes, as in this study. Similarly, Wang et al. (2018) found strong correlations for DDX bioaccumulation by Eisenia fetida using PE (r² = 0.93) and solid-phase microextraction (SPME) (r² = 0.91). By contrast, poorer r² values were observed for bioaccessibility methods such as Tenax (r² = 0.79), isotope dilution (r² = 0.77) and supercritical fluid extraction (r² = 0.47). The only study, to our knowledge, that has investigated the correlation between POM and earthworms is the study by Denyes et al. (2016), who also used 76 μm POM to predict E. fetida bioaccumulation of p,p′-DDT and p,p′-DDE. They did not report any r² values but concluded that their mean worm bioaccumulation factors (BAFs) were within 50% of the POM-derived BAFs.

It can be argued, based on the good correlation between C_{worm_lipid} and C_TOC (Fig. 6), that this analysis can be used instead of POM analyses. One possibility, however, is that the good correlation can be attributed to the fact that all soils included in the study originated from plant nurseries situated in rural–agrarian regions with comparable NOM and similar contaminant age. Inclusion of soils from more urban environments would likely have resulted in greater variation and a poorer correlation, as anthropogenic carbon, such as soot, has been demonstrated to have a significantly greater sorption capacity for HOC than NOM (Cornelissen et al. 2005). The presence of types of black carbon, such as soot and charcoal, may therefore reduce the predictive power of C_TOC and lead to overestimation of bioavailability. It is therefore recommended that POM analysis and the here derived C_{worm_lipid} v. C_POM correlations are used in preference to C_soil and TOC determinations to predict the bioaccumulation of DDX in worms.

As illustrated in Fig. 6, C_POM can serve as a prediction tool for C_{worm_lipid}. Consequently, passive sampling with POM may also be employed to predict ecotoxicity, provided that a correlation exists between the organism’s bioaccumulation of DDX and the toxic response. However, the endpoints investigated here (growth, reproduction and mortality) were not sufficiently sensitive for the concentration range investigated. No significant correlation was found for mortality or reproduction against C_{worm_lipid} or C_POM. It should be noted, though, that the sample with the highest expected C_{worm_lipid} (Åby; with C_soil of 134 mg DDX kg^–1 DW), was not analysed, due to 100% mortality of worms, and that the C_soil of Σp,p′-DDX showed a moderate and significant (P < 0.05) linear correlation with mortality percentage at 28 days (Pearsons correlation: 0.72, r² = 0.52).

DDT is an insecticide, and low levels of DDT in soil are expected to have toxic effects on terrestrial organisms, for example, an acute 50% lethal concentration (LC₅₀) for the cricket Gryllus pennsylvanicus has been reported to be 10 mg DDT kg^–1 DW in soil with 10% NOM. There are, however, few data for terrestrial organisms available in ecological databases. Available data for Eisenia fetida in the scientific literature suggest that this species can tolerate high DDX concentrations, which is in line with our results (Table 4); the half maximal effective concentration (EC₅₀) for reproduction has been reported to be 588 mg DDT kg^–1 DW, in a sandy soil with 1% TOC (Hund-Rinke and Simon 2005) and Shi et al. (2016) reported an acute LC₅₀ of 274 mg DDT kg^–1 DW (14 days exposure) and a chronic LC₅₀ of 146 mg kg^–1 for historical DDT-contaminated soils. In addition, Shi et al. (2016) showed that DDT in concentrations lower than 100 mg kg^–1, in both historical contaminated soils and DDT-spiked natural soils, had minimal effects on the mortality of mature earthworms (Eisenia fetida), whereas few survivors were found in spiked artificial soils at DDT soil concentrations >200 mg kg^–1. Further investigations with more sensitive species or endpoints are thus required to determine whether C_POM can be used to predict toxicity.

POM can also be a tool to investigate the feasibility of remediation options and to check the achievement of objectives after remedial actions have been taken, especially in the case of stabilisation measures, which involve the sorption of contaminants onto a sorbent, e.g. biochar (Denyes et al. 2016). As the amount of contaminant is not removed from the soil, but stabilised, it is not possible to measure the effectiveness of remediation by C_soil analysis, and methods are needed to demonstrate how well the sorbent is working and whether the solubility, bioavailability and dispersion of the contaminants have been reduced to acceptable levels (Wang et al. 2018). Complementary measurements of C_W,free and TOC enable calculations of soil–porewater distribution coefficients (K_d) and TOC–normalised soil–water distribution coefficients (K_TOC), see Supplementary Eqn S3, which are useful for this purpose, i.e. to assess how the sorption of DDX can improve after stabilisation measures, but also for risk assessment purposes, e.g. in modelling transport between different environmental compartments.

The K_TOC values of this study can be found in Supplementary Table S28, and ratios of log K_TOC derived by Lab A and Lab B are presented in Supplementary Table S28. The ratios were close to one in experiments where depletion was <10%, with the exception of the thermolabile compounds (mainly dicofol and its degradation product p,p′-DBP). The thermoinstability and potential matrix effect (Foreman and Gates 1997) may result in overestimations and underestimations of log K_TOC, which is reflected in the absence of good linear correlations between log K_TOC and log K_OW in our study (examples given in Supplementary Fig. S7).

In conclusion, this study shows that POM is a reliable tool for measuring C_W,free, and can be used for further calculations of bioaccumulation, using the here derived correlations for earthworms. The importance of controlling for degradation during GC analysis to accurately determine individual DDX analytes is highlighted, which is important when deriving site and compound-specific distribution coefficients needed for dispersion modelling.