Science of the Total Environment 517 (2015) 48–56

Contents lists available at ScienceDirect

Science of the Total Environment journal homepage:

Spatial distribution and partitioning behavior of selected poly- and perfluoroalkyl substances in freshwater ecosystems: A French nationwide survey Gabriel Munoz a, Jean-Luc Giraudel a, Fabrizio Botta c, François Lestremau c, Marie-Hélène Dévier a, Hélène Budzinski b, Pierre Labadie b,⁎ a b c

University of Bordeaux, EPOC, UMR 5805, LPTC, 351 Cours de la Libération, F-33400 Talence, France. CNRS, EPOC, UMR 5805, LPTC, 351 Cours de la Libération, F-33400 Talence, France. INERIS, Parc Technologique Alata, BP2, 60550 Verneuil-en-Halatte, France.




• A large-scale survey of PFASs in 133 French rivers and lakes is reported. • Descriptive statistics, correlations and partitioning coefficients were determined. • Non-detects were taken into account using functions from the NADA Rpackage. • Hot spots of PFAS contamination were found near large urban and industrial areas. • Sediment levels were partly controlled by grain size and organic carbon content.

a r t i c l e

i n f o

Article history: Received 7 December 2014 Received in revised form 12 February 2015 Accepted 12 February 2015 Available online 23 February 2015 Editor: D. Barcelo Keywords: Perfluoroalkyl substances Water Sediment Partitioning Artificial neural networks Non-detects

a b s t r a c t The spatial distribution and partitioning of 22 poly- and perfluoroalkyl substances (PFASs) in 133 selected rivers and lakes were investigated at a nationwide scale in mainland France. ΣPFASs was in the range b LOD–725 ng L−1 in the dissolved phase (median: 7.9 ng L−1) and bLOD–25 ng g−1 dry weight (dw) in the sediment (median: 0.48 ng g−1 dw); dissolved PFAS levels were significantly lower at “reference” sites than at urban, rural or industrial sites. Although perfluorooctane sulfonate (PFOS) was found to be the prevalent compound on average, a multivariate analysis based on neural networks revealed noteworthy trends for other compounds at specific locations and, in some cases, at watershed scale. For instance, several sites along the Rhône River displayed a peculiar PFAS signature, perfluoroalkyl carboxylates (PFCAs) often dominating the PFAS profile (e.g., PFCAs N 99% of ΣPFASs in the sediment, likely as a consequence of industrial point source discharge). Several treatments for data below detection limits (non-detects) were used to compute descriptive statistics, differences among groups, and correlations between congeners, as well as log Kd and log Koc partition coefficients; in that respect, the Regression on Order Statistics (robust ROS) method was preferred for descriptive statistics computation while the Akritas–Theil–Sen estimator was used for regression and correlation analyses. Multiple regression results

⁎ Corresponding author. E-mail address: [email protected] (P. Labadie). 0048-9697/© 2015 Elsevier B.V. All rights reserved.

G. Munoz et al. / Science of the Total Environment 517 (2015) 48–56


suggest that PFAS levels in the dissolved phase and sediment characteristics (organic carbon fraction and grain size) may be significant controlling factors of PFAS levels in the sediment. © 2015 Elsevier B.V. All rights reserved.

1. Introduction Poly- and perfluoroalkyl substances (PFASs) are surfactants that have been used since the 1950s in manifold industrial applications, including metal plating, fluoropolymer processing aids, textiles, and firefighting foams (Prevedouros et al., 2006; Ahrens and Bundschuh, 2014). However, it was not until the last decade that their environmental fate and ecotoxicology aroused the interest of the scientific community (Kannan, 2011). Since then, PFASs have been reported in biotic and abiotic compartments worldwide, including remote polar areas (Houde et al., 2011). Of all the PFASs currently investigated, perfluorooctane sulfonate (PFOS) remains the most emblematic, due to its ubiquitous character and adverse effects (Lau et al., 2007). Numerous papers have reported on the highly bioaccumulative potential of PFOS in biota (Houde et al., 2011) as well as on its persistence (Olsen et al., 2007). These properties led to the classification of PFOS as a Persistent Organic Pollutant (POP) in 2009, under the framework of the Stockholm Convention (UNEP-POPS-COP.4-SC-4-17). Since then, the use of PFOS-containing products has been drastically restricted by the EU (e.g., PFOS-based aqueous film forming foams being banned since June 2011). However, a number of industrial sectors such as electroplating, photolithography or hydraulic fluids for aviation still benefit from derogations until a non-toxic substitute is available (2010/757/EU Commission Regulation). Other PFASs, which have not been added yet to the listing of POPs in the Stockholm Convention, may also be cause for concern, especially medium and long-chain carboxylates (Wolf et al., 2012; Buhrke et al., 2013). Until now, only a few surveys have addressed PFAS environmental contamination at nationwide or larger spatial scale. For instance, Loos et al. (2009) targeted 122 rivers in Europe, although the focus was not put exclusively on PFASs. However, by compiling these data with other values from the literature, a first estimate of PFOS and perfluorooctanoate (PFOA) European discharges was calculated (Pistocchi and Loos, 2009). Clara et al. (2009) evaluated the occurrence of perfluoroalkyl sulfonates (PFSAs), carboxylates (PFCAs) and sulfonamides at national level, including sediments from 7 Austrian lakes and the river Danube, while Kwadijk et al. (2010) investigated the spatial distribution, sediment–water distribution coefficient, and bioaccumulation factor of 15 selected PFASs across 21 locations in The Netherlands. Boiteux et al. (2012) provided a first nationwide review of PFAS contamination in French raw and treated water for human consumption. In France, PFASs have been reported at relatively high levels in urban hydrosystems such as the River Seine and the River Orge (mean ∑ PFASs = 55 ng L − 1 and 73 ng L − 1 , respectively) (Labadie and Chevreuil, 2011a,b), perfluorohexane sulfonate (PFHxS) and PFOS being the dominant congeners at these sites, and perfluorohexanoate (PFHxA) and PFOA the main PFCAs, as was also observed near the mouth of the river Seine (McLachlan et al., 2007). Labadie and Chevreuil (2011a) reported on the bioaccumulation propensity and tissue-distribution of a wide range of PFASs in fish (European chub), while Munschy et al. (2013) investigated PFAS spatial distribution in shellfish along French coasts, pointing to contrasting PFAS patterns between the Atlantic and Mediterranean coasts. Human exposure to PFAS was assessed in a 2007 French survey, PFOS and PFOA being among the most frequently reported PFASs (detection frequency N 90%) in breast milk, at times reaching levels above 300 ng L−1 (Antignac et al., 2013). In this context, the present study aimed at investigating PFAS occurrence and partitioning in mainland France surface water bodies. As part of the implementation of a national action plan on aquatic environment pollution (October 2010), the French Ministry of Ecology decided to launch an innovative and comprehensive approach under the Water

Framework Directive (WFD) (2000/60/EC), in order to provide relevant information to update the lists of substances to be included in future monitoring schemes. A vast prospective campaign took place in spring–autumn 2012, which focused not only on potentially contaminated sites (i.e. urban or industrial) but also on supposedly pristine reference sites. Descriptive statistics of PFAS levels or molecular patterns were calculated for this comprehensive dataset which included 333 water and 129 sediment samples. The information was then summarized with the help of artificial neural networks via a Kohonen mapping (Giraudel and Lek, 2001). So far, the PFAS-focused literature has only yielded a few papers dealing with data below detection limits (non-detects) (De Solla et al., 2012; Jaspers et al., 2013; Rigét et al., 2013; Lam et al., 2014). Given the substantial number of observations that fell below detection limits in the present work, specific statistical treatments were implemented to compute descriptive statistics and to examine correlations between PFASs. An alternative method to determine log Kd and log Koc partitioning coefficients taking into account non-detects is also reported in this paper, along with the investigation of factors controlling PFAS sediment levels. 2. Materials and methods 2.1. Investigated compounds Four different groups of PFASs were targeted: perfluoroalkyl carboxylates, sulfonates, sulfonamides and sulfonamide acetic acids, as well as one fluorotelomer. A total of 22 individual molecules and a cluster of branched PFOS isomers, hereafter referred to as Br-PFOS, were therefore determined. Note that, in this paper, L-PFOS refers to the linear isomer, while “PFOS” refers to the sum of L-PFOS and Br-PFOS. Target analytes, as well as full details on chemicals, standards and consumables are indicated in the Supplementary information (SI). 2.2. Sampling strategy Water and sediment samples were collected at 133 locations, including 115 sampling points located in rivers and 18 in lakes (Fig. 1). River sampling sites were classified into five main types (SI Table S1) by the Direction de l'Eau et de la Biodiversité (DEB, French Ministry of Ecology): reference, farmland, industrial, urban, and “poor ecological status”, the latter reflecting low occurrence of aquatic plants and biota. Three campaigns were set up to collect water samples from rivers, leading to a number of 315 water samples. The first campaign took place in April–June 2012, the second one in September 2012, and the third one in November–December 2012. In contrast, a unique water sample was collected for each lake (June 2012). At each sampling site, a 1 L high density polyethylene (HDPE) bottle previously washed in the laboratory was rinsed 3 times with the site surface water, filled to the brim, sealed, and stored in a cooling box (5 ± 3 °C), pending shipment to the laboratory within 24 h. Sediment samples (n = 129) were collected during a single sampling campaign (August–November 2012). Only the top layer (1 to 5 cm depth) of the sediment was sampled, in agreement with guidance document #25 of the WFD. 2.3. Sample reception and pre-treatment Water samples were passed through GF/F (0.7 μm) Whatman glass microfiber filters (previously baked at 450 °C for 6 h) using Nalgene® polyethylene filtration units and the filtrate was divided into two 500 mL aliquots stored in HDPE bottles; filtrates were kept at −20 °C until analysis.


G. Munoz et al. / Science of the Total Environment 517 (2015) 48–56

Fig. 1. Map showing the 115 river sampling sites (dots) and 18 lake sampling sites (stars). River sites were color-coded according to pressure typology (green = reference; orange = urban; purple = poor ecological status; red = industrial; yellow = farmland).

Sediment samples were wet sieved at 2 mm and grinded at b 200 μm before characterization of Total Organic Carbon (TOC), dry matter level, and fine fraction content (mass percent fraction of the b63 μm fraction over the b2 mm fraction). Particle size was determined by laser granulometry using a Mastersizer 2000 (Malvern Instruments SARL, Orsay, France) and TOC by a TOC-VCSH analyzer (Shimadzu France SAS, Noisiel, France). Sediment samples were freeze-dried, conditioned in 15 mL HDPE tubes and stored in a cooling box. After reception, all samples were stored at 4 °C in a cold room until analysis.

2.4. Extraction and analysis Phenomenex Strata X-AW cartridges (200 mg/6 mL) fitted with glass wool cotton were conditioned with 8 mL of MeOH/NH4OH 0.2% in water (v/v) and rinsed with 5 mL of Milli-Q water. Water samples (500 mL) were spiked with internal standards (IS, 1 ng each, see Table S3), and then pumped through the X-AW cartridges at a rate of approximately 10 mL min− 1. After sample loading, cartridges were rinsed with 5 mL of Milli-Q water, dried under vacuum for 45 min and finally centrifuged (3 min, 5000 rpm). Analytes were eluted with 8 mL of MeOH/NH4OH 0.2% in water (v/v), the resulting extracts being concentrated to 500 μL at 42 °C under a N2 stream. Following transfer to a polypropylene injection vial, extracts were concentrated to 300 μL, before storage at −20 °C until analysis.

Sediment samples were treated with microwave-assisted solvent extraction. Briefly, 1 g of sediment was weighed in the microwave vessel, and spiked with IS (1 ng each), before the addition of 10 mL of MeOH. The vessels were then sealed and placed in a START E microwave oven (Milestone SRL, Sorisole, Italy). The extraction time was set at 10 min: 5 min to reach the temperature set point (70 °C) and a 5-min holding time at 70 °C. Following filtration on glass fiber cotton, extracts were concentrated to approximately 2 mL. A purification step was performed using ENVI-Carb graphite cartridges (250 mg/6 mL) previously conditioned with 5 mL of MeOH. After sample loading, cartridges were rinsed with 5 mL of MeOH, and the resulting extracts were concentrated and transferred into injection vials as previously mentioned. PFAS analyses were performed by ultra-high performance liquid chromatography negative electrospray ionization coupled with tandem mass spectrometry (HPLC–ESI(-)-MS/MS) with an Agilent 1200 LC coupled to an Agilent 6460 triple quadrupole mass spectrometer, both from Agilent Technologies (Massy, France). For details on chemical analyses, see the supplementary material. 2.5. Quality control All instrumental blanks remained exempt from analytes and IS. Procedural blanks, performed with 500 mL of Vittel mineral water for water samples and with 10 mL of MeOH for sediment samples, were run every ten samples. In water procedural blanks (n = 51), the most recurrent

G. Munoz et al. / Science of the Total Environment 517 (2015) 48–56

analytes were PFPeA and PFBA (0.15 ± 0.08 and 0.04 ± 0.05 ng, respectively), while in sediment procedural blanks (n = 11), PFHxA and PFPeA were systematically detected (0.2 ± 0.5 and 0.2 ± 0.1 ng, respectively) (Table S5). When analytes were found in procedural blanks, the method limit of detection (LOD) was calculated as the standard deviation of blanks corrected by the tn-1,95 student coefficient, n being the number of blank replicates. Otherwise, the LOD was classically derived from the signal-to-noise ratio (SNR) observed in low-contaminated samples or in spiked matrices. When two transitions were followed (quantitation and confirmation), the transition with the lowest SNR was chosen for LOD calculation. Method detection limits were in the range 0.01–0.27 ng L−1 and 0.001–0.18 ng g−1 dry weight (dw) in the dissolved phase and the sediment, respectively (Table 1). Short-chain carboxylates (C4–C6) were not targeted in the sediment, due to the higher variability of blank levels (PFHxA), recurrent blank contamination (PFPeA), or poor chromatographic performances (PFBA). For spiked water samples, analyte recovery rates were in the range 53–134% and relative standard deviations remained generally below 15%, except for short-chain carboxylates. Recoveries were in the 56–86% range in spiked reference sand samples. In addition to recovery tests, trueness tests were performed, IS being added at the start of the experiment along with native PFASs. Trueness rates ranged from 78 to 144% for 1 ng L−1 spiked Vittel mineral water samples (n = 12) and from 86 to 117% for 0.5 ng g−1 spiked reference sand samples (n = 3). The relatively high value (151%) observed for 6:2 FTSA when quantified against 13 C-PFOS suggests that this compound may be overestimated in the sediment. Moreover, a strong matrix interference hampered the qualifier peak integration of 6:2 FTSA, which was not quantified in the sediment. For the same reasons, PFBS was not targeted in the sediment matrix. Full details on blank results, instrumental detection limits, LODs, as well as trueness and recovery tests are available in the SI (Tables S5–S7). 2.6. Statistics and GIS Statistical tests were performed with the R statistical software (R version 2.15.3, R Core Team, 2013). Statistical significance was set at a 0.05 p-value cutoff. Due to the high number of non-detects, functions from the NADA R-package were used to calculate descriptive statistics, differences between groups and correlations (Helsel, 2012; Lee, 2013)


(see SI for details). To compare these results with those achieved with substitution techniques, two additional data set were created, in which all data that fell below LOD were classically substituted by either 0.5 × LOD or 0 × LOD. In the case of the Kohonen mapping, logarithm transformed PFAS concentrations were used, the self-organizing map (SOM) algorithm being adapted from a program file written by Giraudel and Lek (2001). Quantum GIS 1.8.0 “Lisbona” (QGIS) was used as a geographic information system, and base maps were downloaded from Natural Earth (URL: bhttp://www.naturalearthdata.comN). 3. Results and discussion 3.1. Importance of the statistical treatment of non-detects Non-detects are trace-level concentrations that are known to fall below an analytical threshold such as the LOD, and are therefore commonly reported as “bLOD”. As such, incorporating non-detects in the estimation of descriptive statistics (e.g., mean, standard deviation, median), differences among groups or correlations may turn out problematic (Helsel, 2006). A common practice consists in replacing all non-detects by a single arbitrary value, usually 0.5 × LOD. However, substitution techniques have no theoretical basis and may result in poor descriptive statistics and correlation estimates (Huston and Juarez-Colunga, 2009; Helsel, 2012). Alternatives to substitution techniques to determine descriptive statistics include, for instance, Kaplan Meier (KM) and Regression on Order Statistics (robust ROS) approaches. The KM estimator is a nonparametric method, recommended for small sample sizes (b50 observations) with up to 50% censoring percentage (Helsel, 2012). In contrast, robust ROS is a parametric method, assuming a lognormal distribution to impute unknown observations prior to the computation of summary statistics (Huston and Juarez-Colunga, 2009; Helsel, 2012). Robust ROS allows more flexibility in the size of the data set than does KM, and may provide accurate descriptive statistics up to 80% censoring percentage (Huston and Juarez-Colunga, 2009). The descriptive statistics generated from the present survey were therefore based on the robust ROS estimator. When censoring percentages were higher than 80%, mean and median were not calculated and were simply reported as “NC” (not calculated). Note that descriptive statistics computed with

Table 1 PFAS occurrence and concentration ranges observed in water and sediment (mean and median values are indicated in the SI). Dissolved phase (n = 333) −1


b0.17–11 b0.27–35 b0.10–86 b0.05–16 b0.08–36 b0.04–30 b0.07–10 b0.05–1.3 b0.07–0.94 b0.02–0.29 b0.07 b0.02–29 b0.02–217 b0.02–17 b0.01–197 b0.06–173 b0.04–0.21 NA NA b0.06–0.73 b0.09–0.46 b0.06–0.81 b0.05–95 bLOD–725


Sediment (n = 129) −1

Detection limit (ng L 0.17 0.27 0.10 0.05 0.08 0.04 0.07 0.05 0.07 0.02 0.07 0.02 0.02 0.02 0.01 0.06 0.04 NA NA 0.06 0.09 0.06 0.05


Detection frequency (%)

Range (ng g−1 dw)

Detection limit (ng g−1 dw)

Detection frequency (%)

49.5 57.7 71.5 70.0 83.8 65.8 54.7 16.5 3.9 0.6 0 65.2 81.1 41.7 85.9 84.7 1.5 NA NA 20.1 0.9 1.2 51.7

NQ NQ NQ b0.12–0.37 b0.11–1.1 b0.04–0.97 b0.05–2.4 b0.01–2.6 b0.06–1.9 b0.02–5.0 b0.02–1.3 NQ b0.02–0.63 b0.03–0.15 b0.04–2.7 b0.01–17 b0.18–0.20 b0.01–7.3 b0.01–2.3 b0.003–0.82 b0.01–0.067 b0.001–0.033 NQ bLOD–25

– – – 0.12 0.11 0.04 0.05 0.01 0.06 0.02 0.02 – 0.02 0.03 0.04 0.01 0.18 0.01 0.01 0.003 0.01 0.001 –

– – – 7.8 16.3 24.8 45.7 47.3 42.6 36.4 32.6 – 7.0 0.8 41.1 74.4 1.6 22.5 10.1 67.4 1.6 3.9 –

dw: dry weight; NQ: analyte not targeted; bLOD: below detection limit; NC: the mean was not computed (not applicable: censoring percentage N 80%); NA: not available.


G. Munoz et al. / Science of the Total Environment 517 (2015) 48–56

other statistical treatments for non-detects (0 × LOD, 0.5 × LOD, KM) are provided in the SI. When examining descriptive statistics, the results obtained with a substitution technique (0.5 × LOD or 0 × LOD) fell close to those obtained with a KM or a ROS treatment of non-detects (Tables S8–S9). The 0.5 × LOD substitution method produced a slight upward drift in the determination of the mean ΣPFASs compared to the 0 × LOD method (i.e. 28.6 ng L−1 Vs 28.3 ng L−1). This may seem inconsequential; however, the results might have differed significantly if we had reported higher detection limits. Obviously, this is of even higher concern for the calculation of molecular patterns. The relative abundance is poorly estimated when all nondetects are replaced by a fraction of the compound's detection limit, since an artificial signal which was not really present is added (Helsel, 2006); this can, in turn, lead to misinterpretations. For instance, the 0.5 × LOD substitution technique returned a mean abundance profile of 0.9% for PFTeDA in the dissolved phase, although this compound was never detected in water (Table S10). As a result, relatively high percentages will be found for compounds that are seldom detected, while major compounds will exhibit artificially lower percentages. To illustrate, the mean abundance of L-PFOS in the sediment (Table S11) given by the 0.5 × LOD substitution technique fell to 24% — about half the result obtained by a 0 × LOD substitution method — while the figure for PFDS inflated to an unrealistic 11%, due to a higher LOD (0.18 ng g− 1). In order to evade the problem of non-detects, it can be decided to consider only a subset of the original data — for example, the 50% most contaminated sites. Although this information is interesting in its own right, it leads to a loss of information. Clearly, none of the three substitution options discussed above is entirely adequate; however, the 0.5 × LOD substitution method may produce especially poor estimates of mean relative abundances and should therefore be avoided for this purpose. In this work, the robust ROS method was preferred to substitution approaches to compute descriptive statistics. 3.2. Levels in water and sediment The most frequently detected compounds in water (n = 333) were Br-PFOS and L-PFOS (Table 1). The linear and branched isomers of PFOS were detected simultaneously in 82% of samples, while 89% of samples showed the presence of either L-PFOS or Br-PFOS. Other recurrent compounds were generally short to medium perfluoroalkyl chain-length compounds such as PFHxA, PFOA or PFHxS, while the detection frequency of C10–C13 carboxylates followed an exponential decrease (R2 = 0.989, p = 0.005) with increasing perfluoroalkyl chain length. Equally noteworthy was the occurrence of 6:2 FTSA in a substantial number of samples (52%). On average, ΣPFASs in the dissolved phase (n = 333) was 28 ng L− 1 (median: 7.9 ng L−1) and ranged from bLOD at some pristine reference sites up to 725 ng L− 1 in the Ru d'Ancoeuil near the town of Melun. L-PFOS amounted to 5.1 ng L− 1 on average (median: 0.90 ng L−1), a much lower value than the 39 ng L−1 reported by Loos et al. (2009) in a survey of 122 European rivers. The levels of other key molecules in the dissolved phase (e.g. PFHxA, PFOA, PFHxS and Br-PFOS) were in the range 2.5–4.7 ng L−1 on average. Reference sites significantly differed from other sites (e.g., L-PFOS, Fig. 2), ΣPFASs always remaining below 3.7 ng L−1 and the main compound found being PFBA (range: b0.17–1.0 ng L−1) at such sites. No significant difference could be detected between the other 4 sampling site types (farmland, industrial, poor ecological status and urban), except in the case of L-PFOS (industrial = urban N farmland) and FOSA (urban N farmland). Also note that industrial locations displayed markedly higher levels of PFHxA, PFHxS and PFOS on average (Fig. S1). In sediment samples, L-PFOS and FOSA were often detected (74 and 67%, respectively), as well as C10–C12 carboxylates (43–47%). The mean ΣPFASs was 1.8 ng g− 1 dw (range: b LOD–25 ng g− 1 dw; median: 0.48 ng g−1 dw), while individual compounds were generally found in

Fig. 2. PFOS levels in rivers (n = 315, dissolved phase) according to sampling site type: farmland, industrial, PES (poor ecological status), reference and urban. The horizontal line represents the censoring threshold.

the low ng g−1 range, with a mean value of 0.72 ng g−1 dw reported for L-PFOS (median: 0.15 ng g−1 dw), while other relevant compounds (e.g., Br-PFOS, C10–C13 carboxylates, MeFOSAA) remained below 0.20 ng g− 1 dw on average. L-PFOS ranged from bLOD to 17 ng g− 1 dw, which is consistent with findings by Kwadijk et al. (2010) in The Netherlands (range 0.5–8.7 ng g−1 dw) or by Hloušková et al. (2014) in the Czech Republic (0.2–17.7 ng g−1 dw). However, no clear pattern emerged as regards the influence of sampling site type on PFAS levels in sediments (Fig. S2). 3.3. Molecular patterns and spatial distribution In order to highlight similarities between locations, PFAS concentrations were analyzed with the help of unsupervised artificial neural networks, the most salient findings being described in this subsection. At national scale, PFOS was found to be the prevalent compound, accounting for 29% of ΣPFASs on average in the dissolved phase, and for 47% in the sediment. L-PFOS was the main isomer found in the sediment (88% of total PFOS), whereas L-PFOS and Br-PFOS were found in equal proportions in water (e.g., 48% for L-PFOS), which agrees well with the distribution of PFOS isomers in hydrosystems (e.g., Houde et al., 2008). On average, Br-PFOS and L-PFOS were the dominant congeners in the dissolved phase, followed by PFOA, PFHxA and PFHxS (Table S10). Short-chain congeners such as PFBA, PFPeA, or PFBS had a somewhat lower impact on the PFAS profile, each accounting for about 6–7% of ΣPFASs. Carboxylates and sulfonates were equally dominant: ΣPFCAs represented on average 50% and ΣPFSAs 47% of the PFAS pattern. However, considerable variations were observed between individual sampling sites, as summarized in the 12 × 15 Kohonen SOM (Fig. 3). While reference locations were usually grouped in the lower left-hand corner, the color gradient for each compound differed from one compound to the other, as shown in the corresponding component planes (Fig. S3). In addition, sites showing similar PFAS patterns were projected on neighboring cells of the hexagonal lattice. For instance, three peculiar locations (#PE03, #312 and #207) were associated in the same bright-green cluster, due to relatively high 6:2 FTSA levels. Thus, at the dam of SaintEtienne Cantalès (#PE03, ΣPFASs = 95 ng L− 1), 6:2 FTSA accounted for 72% of the PFAS profile, while site #312 on the River Seine (ΣPFASs = 44–87 ng L− 1) also exhibited high 6:2 FTSA relative abundance (32–52%) probably due to the impact of both the Paris conurbation and a perfluorochemical plant located in the River Oise basin. Interestingly, a very similar pattern to that of #312 was observed on a different watershed, in the cross-border town of Petite Rosselle near Germany (#207, ΣPFASs = 54–81 ng L− 1), where 6:2 FTSA accounted for 26–40%. Before flowing into the Rosselle, the Merle tributary crosses a

G. Munoz et al. / Science of the Total Environment 517 (2015) 48–56 Fig. 3. Similarities between surface water samples (dissolved phase, n = 333) evidenced by a Kohonen self-organizing map combined with k-means clustering. The first three numbers refer to location code (SI Table S1) and the last number to the sampling campaign. Corresponding component planes are available in the SI (Fig. S3).



G. Munoz et al. / Science of the Total Environment 517 (2015) 48–56

major industrial area at Carling (10 km southwest of #207), which may explain the distinctive PFAS profile. Another point worth mentioning was the pink cluster which exclusively comprised sites from the River Rhône basin (#600 series), corresponding to high levels of short to medium chain-length carboxylates. This is consistent with Loos et al. (2009) who observed relatively high values for PFOA (116 ng L− 1) in the River Rhône compared to other major European rivers (e.g., 25 ng L−1 for the Danube, Austria). For instance, downstream of Lyon (#614), ΣPFASs observed in the first campaign was 60 ng L−1, of which 50% was imputed to a single compound (PFNA), 17% to PFHxA and 8% to PFOA. In the next campaign, however, the PFHxA contribution increased dramatically to 77% (ΣPFASs = 67 ng L− 1), while PFNA dropped to 1%. L-PFOS, in contrast, remained steadily at around 5% in both sampling campaigns. While PFOS levels are likely generated by diffuse sources in this watershed, the high variability in PFHxA and PFNA levels might be partly attributed to the activities of a fluorochemical plant synthetizing polyvinylidene fluoride (PVDF), located about 15 km upstream of this location. Similarly, near the mouth of the River Rhône (#631, approximately 300 km downstream of #614), ΣPFASs fluctuated from 31 to 97 ng L− 1, PFHxA accounting respectively for 47–88% of ΣPFASs, thus corroborating the particular PFAS signature of the River Rhône (Dauchy et al., 2012). In sediments, L-PFOS dominated the mean profile, accounting for 41% of ΣPFASs (Table S11). This remained true when examining only the top 50% most contaminated sediments: L-PFOS still accounted for 40% of ΣPFASs, followed by PFDoA (11%), Br-PFOS (7%), several longchain carboxylates such as PFDA, PFUnA, PFTeDA and PFTrDA (range 5–7%), but also MeFOSAA (5%). The lower right-hand side of the SOM (dark green cluster) grouped sampling sites with high levels of PFOS precursors (Fig. S4–S5). In the river Risle (#305, ΣPFASs = 11 ng g−1 dw), PFOS precursors accounted for nearly 75% of the molecular pattern, with relatively elevated MeFOSAA and EtFOSAA levels (7.3 and 0.69 ng g−1 dw, respectively). The reasons for this peculiar profile are still speculative, the only facilities that could be identified near the sampling sites included a WWTP, a paper manufacturing plant and a plant processing zirconium alloys for nuclear applications (in which PFASs may be used for metal plating purposes). Sediments with high PFOS levels were found in the dark blue cluster. The most polluted sediment observed corresponded to an artificial lake near Bordeaux (#001) where ΣPFASs reached 25 ng g−1 dw, of which 77% was due to PFOS. EtFOSAA, FOSA and MeFOSAA also contributed to the PFAS profile, albeit to a lesser extent (3–9%). In contrast, sediments with the highest PFCA contributions were grouped in the pink cluster. For instance, in sediments downstream from Lyon (#614, ΣPFASs = 12 ng g−1 dw), carboxylates represented a remarkable 99% of ΣPFASs, PFTrDA being the essential congener (44%) followed by PFDA and PFDoA (20 and 10%, respectively). Thus, given the considerable size of our dataset, the Kohonen mapping proved to be a useful tool to highlight differences of PFAS levels and molecular patterns between locations. 3.4. Correlations and partitioning behavior L-PFOS and Br-PFOS were strongly correlated in the dissolved phase (cenken regression: slope = 1.12, τ = 0.86, p b 0.001, n = 333) (Fig. S6). This held true when excluding lakes from the regression, and high correlation coefficients were reported for each type of sampling site (τ = 0.85–0.89) except for reference locations where the correlation was weaker (τ = 0.39). Other relevant correlations included Br-PFOS/ PFHxS (τ = 0.82), L-PFOS/PFHxS (τ = 0.77), or L-PFOS/PFOA (τ = 0.71) (Table S12). In riverine sediments (n = 111) (Table S13), LPFOS best correlated with one of its precursors, FOSA (slope = 24.5, τ = 0.59, p b 0.001), whereas weaker correlations were found with C10–C14 carboxylates (τ = 0.37–0.46). Long-chain carboxylates were generally correlated with each other — e.g.: PFDA/PFUnA (τ = 0.45),

PFUnA/PFDoA (τ = 0.45), PFDoA/PFTeDA (τ = 0.50), pointing to similar sources or fate. The standard procedure to determine the sediment–water partitioning coefficient (Kd) is to calculate the ratio of the concentration in the sediment (ng kg − 1 dw) to the concentration in the dissolved phase (ng L−1), provided that the compound is detected in both compartments (“matching pairs” method). The sediment–water partitioning coefficient normalized to carbon content (KOC) is then derived from the following equation: KOC = Kd × 100 / fOC where fOC is the sediment organic carbon fraction (range: b 0.4%–22.9%). This procedure was applied to our dataset and, in addition, Kd and KOC were also estimated by a regression approach including non-detect handling, the calculation procedure being fully described in the SI (page S17). Average Log Kd and Log KOC calculated from the two approaches were in good agreement, the values returned by cenken generally falling within the confidence range of matching pairs (Table 2). The Log KOC of C7–C12 carboxylates fell in the range 2.9–3.9, consistent with findings by Kwadijk et al. (2010) in The Netherlands for C7–C9 carboxylates (range: 2.6–3.7). The cenken regression yielded a 2.9 Log KOC value for L-PFOS and PFOA, in line with the 3.0 and 2.8 average Log KOC values proposed by Zareitalabad et al. (2013). The influence of fOC on PFAS levels in the sediment was also investigated. The levels of PFDA, PFUnA and PFTrDA in the sediment positively correlated with fOC, indicating that fOC might be a major contributing factor of PFAS sorption onto sediments. For these carboxylates, the slope of the cenken regression ranged from 1.6 to 2.3, lower than that of sulfonates such as Br-PFOS (3.8) and L-PFOS (7.7) but higher than that of FOSA (0.5) (SI Table S14). However, Kendall's τ remained low (0.06–0.27), while the highest significance was for FOSA and L-PFOS (p b 0.0001), in excellent agreement with Ahrens et al. (2009). The same tendency was found when investigating the correlation between concentrations in the sediment and the fine fraction (b63 μm) of the sediment (Table S15), which suggests that, as observed for other organic pollutants (Karickhoff et al., 1979; Lee et al., 2006), grain size is another controlling factor of PFAS sorption. The combined influence of fOC (X1) and silt content (X2) on PFAS concentrations in the sediment (Y) was approached by multiple regression, using the following model: ln (1 + Y) = α + β*X1 + γ*X2 (see SI page S19). The model was significant for Br-PFOS, L-PFOS and FOSA. Only in the case of FOSA did we find significant coefficients for both explanatory variables (Table S16). Overall, these results suggest that fOC is a more important controlling factor than the fine fraction content as regards the sedimentary accumulation of PFASs. It has been previously suggested to plot Kd as a function of fOC for the purpose of normalizing PFAS levels in the sediment to those in the dissolved phase (e.g., Pan et al., 2014). In our study, this approach was not

Table 2 Average PFAS partitioning coefficients log Kd and log Koc obtained from 2 methods: i) calculation based on matching pairs only (the analyte level in the dissolved phase and in the sediment must be above the LOD); and ii) regression with the cenken function from NADA. Calculation method


Matching pairs

Cenken regression

Log Kd

Log Koc

Log Kd

Log Koc

2.2 ± 0.5 1.9 ± 0.6 2.5 ± 0.5 2.6 ± 0.6 3.0 ± 0.6 2.7 ± 0.6 4.1 1.8 ± 0.6 2.6 ± 0.3 1.9 ± 0.7 2.3 ± 0.8 2.4 ± 0.6

3.2 ± 0.4 3.0 ± 0.6 3.6 ± 0.5 3.7 ± 0.6 4.1 ± 0.6 3.8 ± 0.6 5.2 2.9 ± 0.6 3.7 ± 0.3 3.0 ± 0.7 3.4 ± 0.8 3.5 ± 0.5

2.2 1.8 2.2 2.2 2.8 2.6 NCa 1.7 3.0 1.8 1.9 2.2

3.3 2.9 3.3 3.3 3.9 3.6 NC 2.5 4.1 2.8 2.9 3.3

NC: regression not performed (n = 1).

G. Munoz et al. / Science of the Total Environment 517 (2015) 48–56

implemented as such since it was not feasible to define a censoring threshold for individual Kd values. Hence, it was proposed instead to consider the PFAS levels in the dissolved phase as an additional explanatory variable. The combined influence of fOC (X1), silt content (X2) and PFAS concentrations in the dissolved phase (X3) on PFAS concentrations in the sediment (Y) was therefore approached by multiple regression. The model was significant for PFDA, Br-PFOS, L-PFOS and FOSA; for BrPFOS and FOSA, significant coefficients were reported for all explanatory variables while coefficients were either significant or near-significant for PFDA and L-PFOS (Table S17). Dissolved PFAS concentrations did not seem more preponderant than the other two investigated factors (Table S17). Since rivers are dynamic systems, non-equilibrium conditions between overlying water and bed sediment may reasonably be expected at numerous sampling sites; this may partly explain the latter observation. In addition, geochemical parameters not considered in this study may also have influenced PFAS sorption (e.g. pH or Ca2 +) (Higgins and Luthy, 2006). 4. Conclusions The present study examined the spatial distribution of PFASs in mainland France surface water bodies. Reference locations exhibited the lowest dissolved PFAS levels and the most polluted sites were generally found near large urban areas or industrial sites. A Kohonen mapping provided evidence for the existence of distinctive features, in some instances at watershed scale (e.g., predominance of carboxylates in the Rhône river), and in other cases between distant sites. The exceptionally large data set resulting from this national overarching campaign, combined with specific statistical treatments for non-detects, allowed for the generation of robust statistics when investigating correlations between compounds, partitioning coefficients, as well as the relative influence of sediment characteristics (i.e. organic carbon and silt content) on PFAS sorption. Our results also suggest that fOC is a more important controlling factor than the fine fraction content as regards the sedimentary accumulation of PFASs. Acknowledgments The authors thank the French Ministry of Ecology, the French National Agency for Water and Aquatic Environments (ONEMA) as well as River Basin agencies for their financial support, as part of the 2012 national prospective campaign on emerging contaminants (“Etude prospective sur les contaminants émergents”). This study has been carried out with financial support from the French National Research Agency (ANR) in the frame of the “Investments for the future” Program, within the Cluster of Excellence COTE (ANR-10-LABX-45); IdEx Bordeaux (ANR-10-IDEX-03-02) provided the PhD grant allocated to G. Munoz. The authors also acknowledge funding from the INTERREG ORQUE SUDOE project (SOE3/P2/F591), as well as the Aquitaine Regional Council and the European Union (CPER A2E project). Europe is moving in Aquitaine with the European Regional Development Fund (FEDER). Appendix A. Supplementary data Supplementary data to this article can be found online at http://dx. References Ahrens, L., Bundschuh, M., 2014. Fate and effects of poly- and perfluoroalkyl substances in the aquatic environment: a review. Environ. Toxicol. Chem. 33, 1921–1929. Ahrens, L., Yamashita, N., Yeung, L.W.Y., Taniyasu, S., Horii, Y., Lam, P.K.S., Ebinghaus, R., 2009. Partitioning behavior of per- and polyfluoroalkyl compounds between pore water and sediment in two sediment cores from Tokyo Bay, Japan. Environ. Sci. Technol. 43, 6969–6975.


Antignac, J.P., Veyrand, B., Kadar, H., Marchand, P., Oleko, A., Le Bizec, B., Vandentorren, S., 2013. Occurrence of perfluorinated alkylated substances in breast milk of french women and relation with socio-demographical and clinical parameters: results of the ELFE pilot study. Chemosphere 91, 802–808. Boiteux, V., Dauchy, X., Rosin, C., Munoz, J.F., 2012. National screening study on 10 perfluorinated compounds in raw and treated tap water in France. Arch. Environ. Contam. Toxicol. 63, 1–12. Buhrke, T., Kibellus, A., Lampen, A., 2013. In vitro toxicological characterization of perfluorinated carboxylic acids with different carbon chain lengths. Toxicol. Lett. 218, 97–104. Clara, M., Gans, O., Weiss, S., Sanz-Escribano, D., Scharf, S., Scheffknecht, C., 2009. Perfluorinated alkylated substances in the aquatic environment: an Austrian case study. Water Res. 43, 4760–4768. Dauchy, X., Boiteux, V., Rosin, C., Munoz, J.F., 2012. Relationship between industrial discharges and contamination of raw water resources by perfluorinated compounds. Part I: case study of a fluoropolymer manufacturing plant. Bull. Environ. Contam. Toxicol. 89, 525–530. De Solla, S.R., De Silva, O., Letcher, R.J., 2012. Highly elevated levels of perfluorooctane sulfonate and other perfluorinated acids found in biota and surface water downstream of an international airport, Hamilton, Ontario, Canada. Environ. Int. 39, 19–26. Giraudel, J.L., Lek, S., 2001. A comparison of self-organizing map algorithm and some conventional statistical methods for ecological community ordination. Ecol. Model. 146, 329–339. Helsel, D., 2006. Fabricating data: how substituting values for nondetects can ruin results, and what can be done about it. Chemosphere 65, 2434–2439. Helsel, D., 2012. Statistics for Censored Environmental Data Using Minitab and R. 2nd ed. Wiley. Higgins, C.P., Luthy, R.G., 2006. Sorption of perfluorinated surfactants on sediments. Environ. Sci. Technol. 40, 7251–7256. Hloušková, V., Lanková, D., Kalachová, K., Hrádková, P., Poustka, J., Hajšlová, J., Pulkrabová, J., 2014. Brominated flame retardants and perfluoroalkyl substances in sediments from the Czech aquatic ecosystem. Sci. Total Environ. 470–471, 407–416. Houde, M., Czub, G., Small, J.M., Backus, S., Wang, X., Alaee, M., Muir, D.C.G., 2008. Fractionation and bioaccumulation of perfluorooctane sulfonate (PFOS) Isomers in a Lake Ontario food web. Environ. Sci. Technol. 42, 9397–9403. Houde, M., De Silva, A.O., Muir, D.C.G., Letcher, R.J., 2011. Monitoring of perfluorinated compounds in aquatic biota: an updated review. Environ. Sci. Technol. 45, 7962–7973. Huston, C., Juarez-Colunga, E., 2009. Guidelines for Computing Summary Statistics for Data-sets Containing Non-detects. Jaspers, V.L.B., Herzke, D., Eulaers, I., Gillespie, B.W., Eens, M., 2013. Perfluoroalkyl substances in soft tissues and tail feathers of Belgian barn owls (Tyto alba) using statistical methods for left-censored data to handle non-detects. Environ. Int. 52, 9–16. Kannan, K., 2011. Perfluoroalkyl and polyfluoroalkyl substances: current and future perspectives. Environ. Chem. 8, 333–338. Karickhoff, S.W., Brown, D.S., Scott, T.A., 1979. Sorption of hydrophobic pollutants on natural sediments. Water Res. 13, 241–248. Kwadijk, C.J.A.F., Korytár, P., Koelmans, A.A., 2010. Distribution of perfluorinated compounds in aquatic systems in The Netherlands. Environ. Sci. Technol. 44, 3746–3751. Labadie, P., Chevreuil, M., 2011a. Partitioning behaviour of perfluorinated alkyl contaminants between water, sediment and fish in the Orge River (nearby Paris, France). Environ. Pollut. 159, 391–397. Labadie, P., Chevreuil, M., 2011b. Biogeochemical dynamics of perfluorinated alkyl acids and sulfonates in the River Seine (Paris, France) under contrasting hydrological conditions. Environ. Pollut. 159, 3634–3639. Lam, N.H., Cho, C.R., Lee, J.S., Soh, H.Y., Lee, B.C., Lee, J.A., Tatarozako, N., Sasaki, K., Saitoe, N., Iwabuchi, K., Kannan, K., Cho, H.S., 2014. Perfluorinated alkyl substances in water, sediment, plankton and fish from Korean rivers and lakes: a nationwide survey. Sci. Total Environ. 491–492, 154–162. Lau, C., Anitole, K., Hodes, C., Lai, D., Pfahles-Hutchens, A., Seed, J., 2007. Perfluoroalkyl acids: a review of monitoring and toxicological findings. Toxicol. Sci. 99, 366–394. Lee, L., 2013. NADA: Nondetects and Data Analysis for environmental data. R Package Version 1.5-5. Lee, S.J., Kim, J.H., Chang, Y.S., Moon, M.H., 2006. Characterization of polychlorinated dibenzo-p-dioxins and dibenzofurans in different particle size fractions of marine sediments. Environ. Pollut. 144, 554–561. Loos, R., Gawlik, B.M., Locoro, G., Rimaviciute, E., Contini, S., Bidoglio, G., 2009. EU-wide survey of polar organic persistent pollutants in European river waters. Environ. Pollut. 157, 561–568. McLachlan, M.S., Holmstrom, K.E., Reth, M., Berger, U., 2007. Riverine discharge of perfluorinated carboxylates from the European continent. Environ. Sci. Technol. 41, 7260–7265. Munschy, C., Marchand, P., Venisseau, A., Veyrand, B., Zendong, Z., 2013. Levels and trends of the emerging contaminants HBCDs (hexabromocyclododecanes) and PFCs (perfluorinated compounds) in marine shellfish along French coasts. Chemosphere 91, 233–240. Olsen, G.W., Burris, J.M., Ehresman, D.J., Froehlich, J.W., Seacat, A.M., Butenhoff, J.L., Zobel, L.R., 2007. Half-life of serum elimination of perfluorooctanesulfonate, perfluorohexanesulfonate, and perfluorooctanoate in retired fluorochemical production workers. Environ. Health Perspect. 115, 1298–1305. Pan, G., Zhou, Q., Luan, X., Fu, Q.S., 2014. Distribution of perfluorinated compounds in Lake Taihu (China): impact to human health and water standards. Sci. Total Environ. 487, 778–784. Pistocchi, A., Loos, R., 2009. A Map of European Emissions and Concentrations of PFOS and PFOA. Environ. Sci. Technol. 43, 9237–9244.


G. Munoz et al. / Science of the Total Environment 517 (2015) 48–56

Prevedouros, K., Cousins, I.T., Buck, R.C., Korzeniowski, S.H., 2006. Sources, fate and transport of perfluorocarboxylates. Environ. Sci. Technol. 40, 32–44. R Core Team, 2013. R: A Language and Environment for Statistical Computing. R Foundation for Statistical Computing, Vienna, Austria. (ISBN 3-900051-07-0, URL bhttp:// www.R-project.orgN). Rigét, F., Bossi, R., Sonne, C., Vorkamp, K., Dietz, R., 2013. Trends of perfluorochemicals in Greenland ringed seals and polar bears: indications of shifts to decreasing trends. Chemosphere 93, 1607–1614.

Wolf, C.J., Schmid, J.E., Lau, C., Abbott, B.D., 2012. Activation of Mouse and human peroxisome proliferator-activated receptor-alpha (PPARα) by perfluoroalkyl acids (PFAAs): further investigation of C4–C12 compounds. Reprod. Toxicol. 33, 546–551. Zareitalabad, P., Siemens, J., Hamer, M., Amelung, W., 2013. Perfluorooctanoic acid (PFOA) and perfluorooctane sulfonic acid (PFOS) in surface waters, sediments, soils and wastewater — a review on concentrations and distribution coefficients. Chemosphere 91, 725–732.

Spatial distribution and partitioning behavior of selected poly- and perfluoroalkyl substances in freshwater ecosystems: a French nationwide survey.

The spatial distribution and partitioning of 22 poly- and perfluoroalkyl substances (PFASs) in 133 selected rivers and lakes were investigated at a na...
1MB Sizes 0 Downloads 13 Views