Long-term monitoring of regulated organic chemicals, such as legacy persistent organic pollutants (POPs) and polycyclic aromatic hydrocarbons (PAHs), in ambient air provides valuable information about the compounds' environmental fate as well as temporal and spatial trends. This is the foundation to evaluate the effectiveness of national and international regulations for priority pollutants. Extracts of high-volume air samples, collected on glass fibre filters (GFF for particle phase) and polyurethane foam plugs (PUF for gaseous phase), for targeted analyses of legacy POPs are commonly cleaned by treatment with concentrated sulfuric acid, resulting in extracts clean from most interfering compounds and matrices that are suitable for multi-quantitative trace analysis. Such standardised methods, however, severely restrict the number of analytes for quantification and are not applicable when targeting new and emerging compounds as some may be less stable under acid treatment. Recently developed suspect and non-target screening analytical strategies (SUS and NTS, respectively) are shown to be effective evaluation tools aimed at identifying a high number of compounds of emerging concern. These strategies, combining highly sophisticated analytical technology with extensive data interpretation and statistics, are already widely accepted in environmental sciences for investigations of various environmental matrices, but their application to air samples is still very limited. In order to apply SUS and NTS for the identification of organic contaminants in air samples, an adapted and more wide-scope sample clean-up method is needed compared to the traditional method, which uses concentrated sulfuric acid. Analysis of raw air sample extracts without clean-up would generate extensive contamination of the analytical system, especially with PUF matrix-based compounds, and thus highly interfered mass spectra and detection limits which are unacceptable high for trace analysis in air samples.
In this study, a novel wide-scope sample clean-up method for high-volume air
samples has been developed and applied to real high-volume air samples,
which facilitates simultaneous target, suspect and non-target analyses. The
scope and efficiency of the method were quantitatively evaluated with organic
compounds covering a wide range of polarities (logP 2–11), including legacy
POPs, brominated flame retardants (BFRs), chlorinated pesticides and
currently used pesticides (CUPs). In addition, data reduction and selection
strategies for SUS and NTS were developed for comprehensive two-dimensional
gas chromatography separation with low-resolution time-of-flight mass
spectrometric detection (GC
Air monitoring programmes and case studies on the environmental fate of
anthropogenic pollutants including legacy persistent organic pollutants
(POPs) are important tools for environmental risk assessment. Furthermore,
data generated in monitoring programmes and case studies form the
foundations for integrated modern pollutant regulations and the
effectivity assessment of international agreements and conventions on POPs
(UNECE, 1998; UNEP, 2009a, b; EMEP, 2019). Air measurements of POPs are
commonly done using quantitative targeted analytical approaches in
combination with highly selective sample clean-up methods, often involving
destructive sample clean-up with concentrated sulfuric acid (H
The current demand for various chemicals in technical and day-to-day
consumer products is steadily expanding, leading to a constantly increasing
number of new compounds identified as potential environmental contaminants.
In the light of the continuously increasing numbers of chemicals in
commerce, the development of single-compound quantitative analytical methods
for each of these new compound groups is now considered ineffective,
time-consuming and expensive. Therefore, there is a strong demand to develop
targeted multi-compound analytical methods with the potential
supplementation with suspect screening and non-target screening strategies
(SUS and NTS). Many potential emerging contaminants are less persistent
and therefore rapidly degrade during destructive sample extraction clean-up
and processes (i.e. acid treatment, saponification, lyophilisation).
This limitation is a fundamental restriction for quantitative analyses of
such labile compounds and the identification of hitherto unknown
potential contaminants with similar physical–chemical properties. Hence,
there is an obvious incentive for the development of an alternative mild,
non-destructive sample clean-up procedure in order to retain the broadest
possible range of chemicals and as little as possible interfering matrix in
the clean extract. Today, the combination of unspecific sample extraction
and clean-up, together with high-resolution chromatographic and
detection methods, is considered a prerequisite for NTS and SUS strategies.
In particulate, the application of ultra-high-resolution chromatographic
methods (either liquid or gas chromatographic) in combination with high-resolution mass spectrometry (HRMS) enabled the identification and
characterisation of hitherto unknown environmental contaminants in different
matrices (López Zavala and Reynoso-Cuevas, 2015; Alygizakis et al.,
2016; Hernández et al., 2015; Masiá et al., 2014; Al-Qaim et al.,
2014; Hernández et al., 2007; Rostkowski et al., 2019; Schymanski et
al., 2015). Another advanced analytical tool for non-target-specific
analysis of environmental samples is comprehensive two-dimensional gas
chromatography (GC
GC separation compared to comprehensive GC
The overall aim of this study was the development of a wide-scope sample
clean-up method for high-volume air samples and the development of SUS and NTS
strategies optimised for GC
The samples of this study were based on the following: (i) the evaluation of the novel wide-scope clean-up method, which was based on a recovery test covering compounds within a wide range of polarities using spiked surrogate method evaluation samples and target analysis; and (ii) the application of the novel clean-up method to real high-volume air samples from the Birkenes Observatory in combination with the development of SUS and NTS strategies. For both (i) and (ii), glass fibre filters (GFF; 142 mm in diameter) and PUF plugs (7 cm in diameter, 4 cm in height), commonly used in high-volume air sampling (Kallenborn et al., 2013), were used.
For (i), spiked surrogate method evaluation samples (unexposed PUFs and
GFFs) were spiked with
Spiked standard mixtures for method evaluation samples.
For (ii), two dedicated real high-volume air samples were collected during
March–April 2015 at an EMEP background monitoring station, the Birkenes
Observatory in southern Norway (Aust-Agder; 58
The exposed real high-volume air samples (GFF and PUF) from Birkenes (ii)
were spiked with internal standard (ISTD) mixture (see Supplement Table S6 for
details) and GFFs and PUFs were Soxhlet-extracted separately for 8 h in
acetone
The samples from part (i) were quantitatively evaluated by target analysis using GC-HRMS. The detailed quantitative analytical methods applied here are described in Halse et al. (2011) and Kallenborn et al. (2013). A short description of these methods can also be found in the Supplement.
The real high-volume air samples (ii) were analysed on a comprehensive
high-resolution two-dimensional gas chromatograph coupled to a
low-resolution time-of-flight mass spectrometer with unit mass resolution
(GC
Further details on chromatographic specifications are given in the Supplement.
LECOs® ChromaTOF® (V 4.50.8) software (including its advanced features “Scripts” and “Statistical
Compare”), which also controls the GC
An in-house-developed post-acquisition workflow for GC
General strategy and levels of identification confidence for
GC
Data processing workflow and peak reduction during level classification.
Laboratory blank samples were included for both sample types (i) and (ii).
The blanks consisted of unexposed PUFs and GFFs and were treated as their
respective sample type (i) or (ii) regarding extraction, clean-up and
analyses. To ascertain whether a detected or reported compound has its origin in
sample (i) or (ii) and does not occur in the respective laboratory blank
samples for (i) or (ii), a compound needs to exceed a sample concentration
factor
There were no targeted compounds detected in blanks for part (i). ISTDs,
used in SUS and NTS of real high-volume air samples (part ii), were used for
quality assurance and sample normalisation but not for target
quantification. Visual comparisons of peak intensity and intensity ratios
from ISTDs were used to identify potential contamination and/or performance issues
of the GC
PUF plugs used for active air sampling would normally be reused after sample extraction and a complete cleaning procedure. Thus, PUF plugs for sampling and blank samples may be of different age, and thus the extractable PUF matrix will vary. Extracts from exposed, real high-volume air samples and laboratory blank samples can thus contain different peak distribution profiles. Blank filtration strategies are described in Sect. 3.2.1.
The application of the novel wide-scope sample clean-up method, with a
custom three-layer liquid chromatography method, was quantitatively
evaluated with targeted analyses using GC-HRMS of triplicates of unexposed
samples (PUFs and GFFs) spiked with a mixture of various compound classes
covering a wide range of polarity (logP 2–11). The results show that the
novel clean-up method provided extracts of similar cleanness and comparable
recoveries for acid-stable POPs as routine methods in monitoring programmes
for POPs. The recoveries of most of the targeted compounds were over 50 %
using the novel clean-up method (Table 2), which is in
accordance with the standard quality control (QC) requirements for this type of analysis. For
acid-labile compounds such as dieldrin, endrin, aldrin, isodrin,
heptachlor-
Summary of average recovery rates (%) for legacy POPs, BFRs, CUPs and CECs.
A few of the spiked compounds had no recorded recovery (i.e.
chlorfenvinphos, chlorobenzilate, dichlorvos, endrine aldehyde and
etridiazole) or very low recovery (i.e. bromacil and chloroneb). The most
probable reason seems to be insufficient elution with the solvent used
(ACN
For compound characterisation, an already reported level classification
system for identification confidence by Schymanski et al. (2015) was
adopted and optimised for the GC
In comparison to target analysis, developed for the highest confidence level
of identification, SUS and NTS results have different confidence levels as
described above. In target analysis, isotope dilution analysis with ISTDs
is, beside others, a commonly applied technique (EFSA, 2010;
European Commission, 2017). The specific sample clean-up used here for
those selected compounds removes the bulk of disturbing matrix and other
potential deteriorating issues with potential effects on the chromatographic
separation. Hence, the results are reported as validated concentration
levels in table form for all target analytes (Fig. 2, Level 0), whereas for SUS and NTS a more general sample clean-up
procedure is necessary, which often does not remove all interfering matrix.
These SUS and NTS results are identified as extensive lists of relevant peaks
(often
The first step in reducing the originally long peak lists produced by
deconvolution of raw data is to identify and remove compound signals which
also occur in sample blanks. Since SUS and NTS at this stage
result in qualitative and semiquantitative rather than quantitative results,
the exact compound concentration in the collected air samples and blanks is
unknown. Therefore, blank compound filtration is based on comparison of
signal areas only. In order to compensate for response variation occurring
between real sample extracts and method blanks, a high threshold for
detection is applied that is considerably higher than utilised for traditional target
analysis. In our case, a compound in a real sample must exceed an area
factor
After automatic sample blank filtration for NTS and SUS analysis, the peak list of the air samples from Birkenes still covered a large number of compounds also confirmed in sample blanks. This poor efficiency of automated blank filtration can be explained by the differences in peak distribution profiles for the different blank samples and for the average of the blank samples compared to the real samples. Only 50 %–75 % of the identified blank contaminants were identical in the different blank samples. However, the automatic filtration procedure reduced approximately 10 % of the total peak number (reduction from about 26 000 to 24 000 peaks for PUF samples and 25 000 to 22 000 peaks for GFFs). Further strategies for peak filtration had to be applied to reduce the number of peaks. Such an effective filtration is necessary to provide a suitable platform for priority compound identification (Fig. 3, to reach A) and classification of the different confidence levels (Fig. 2, L1–L5).
During initial data processing, the ChromaTOF® software used here automatically finds all relevant signals or peaks, deconvolutes coeluting mass spectra, combines modulation slices and compares this spectral information against the set of chosen MS libraries. Hereby, it may happen that one signal in the chromatogram is associated with several peak markers, e.g. if the peak width is broader than the specifications used for automatic peak finding or peaks are tailing. Unfortunately, the automated deconvolution algorithm from ChromaTOF® can mark a single compound with several peak markers, which was shown in a study by Lu et al. (2008). Due to these limitations, the total number of originally detected compounds is usually lower than the number of peak markers. First, during comprehensive manual inspection (Fig. 3A), these additional false peak markers will be discovered and peak lists corrected for duplicate peak markers.
In this study, the data processing (DP) strategy was split into two parts, SUS (Fig. 3I) and NTS (Fig. 3II). After the initial automated peak identification, the peak lists from both DP approaches were merged to one L5 list for a manual check of identity (Fig. 3A) and further level of identity confidence classification.
During SUS DP (Fig. 3I), all MS of the
automatically detected peaks were searched against the MS library
reference information for SUS (in-house custom libraries of reference
standards and ISTDs, customised suspect library as described in Sect. 2.5,
and the SWGdrug mass spectral library; Oulton, 2019). Added ISTDs were
identified (L0), as were sample blank compounds. A second blank
filtration was performed and only compounds which exceed an area of
factor
For NTS DP (Fig. 3II), the LECOs statistical
compare® tool for the identification of all
compounds occurring in both PUF samples or both GFF samples was applied. With this
approach, it was possible to reduce the peak lists from approx. 30 000 to
3800 peaks for PUF and from approx. 25 000 to 5000 peaks for GFF samples.
After the initial automatic blank filtration (see Sect. 3.2.1), DP with the
NIST14 and suspect libraries as well as the application of NT scripts for the
identification of specific compounds of interest (i.e. halogenated) were
performed. The resulting peak list was further reduced to approx. 1000 peaks
per sample. These NT scripts, written in Visual Basic, were applied during
DP to identify brominated and chlorinated compounds based on their isotopic
clusters, as well as PAHs, phthalates and nitro compounds, with the help of
recognisable features in fragmentation patterns (Hilton et al., 2010).
These scripts are especially useful to detect compounds which would be
overlooked by a low MS library match or not listed in the MS libraries used.
In addition, a second blank filtration was performed, and only compounds
which exceed an area of factor
Similar to SUS, a manual check of the right identity of these NTS L5 compounds is needed in order to increase the level of identification confidence since all confirmations are only based on MS library comparisons or NT script filtrations. For manual inspection of each compound and further level classification, the lists from SUS and NTS were merged to one list for a more effective process (Fig. 3A).
Both DPs, SUS and NTS, used the forward match percentage to MS library
entries to reduce the number of peaks which require manual inspection. In
this step, the quality of an MS from a compound is of high importance to
match an MS library entry and thus be kept for further processing. The
quality of the MS of a compound is not only affected by interferences or S
(1) Isotope cluster of hexabromobenzene (HBB) in NIST14, (2) own
measured HBB on GC
In addition to the MS quality affected by the unit mass resolution of the ToF-MS detector, a lower library match could also be caused by different fragmentation patterns compared to the MS from the NIST14 library, which were obtained with a quadrupole mass filter in electron ionisation mode. Also here it was possible that compounds of interest could be rejected during a DP step due to a low match percentage to a NIST14 MS.
Further factors may limit the positive identification of a compound
including potential loss during sample clean-up. Our sample clean-up method
was optimised for the analysis of compounds covering a wide range of
polarity for GC
In the DP strategy chosen here, all confirmed compounds need to match all selection criteria used. However, the priority criteria need individual fine tuning for each dataset examined to avoid false positive and false negative listings as well as to minimise the occurrence of blank compounds. However, even after following this comprehensive data processing protocol, the possibility cannot be excluded that unconfirmed or excluded substances do not occur in air from Birkenes, southern Norway.
After comprehensive peak filtration from raw data to a reduced peak list for
manual inspection, all remaining compounds were initially classified as L5
(mass spectra of interest) (Fig. 3A), and
all compounds identified with ISTDs were classified as L0. The compounds
classified as L5 are further checked manually for their identity to reach a
higher level of identification confidence. For some compounds with a high
match percentage compared with the reference MS libraries and recognisable
For the high-volume air samples studied here from the Birkenes Observatory,
the merged L5 list from SUS and NTS available for manual inspection
(Fig. 3A) contains almost 1500 compound
suggestions: over 600 compounds from the GFF extracts (particulate phase)
and over 850 compounds from the PUF extracts (gaseous phase). More than 50 % of these compounds could be further identified and classified as L4, L3
and L2 during manual inspection of MS. This was possible for 350 compounds
from the GFF and for 655 compounds from the PUF. All L2 and L3 compounds
were manually checked against the blank sample before comparison to new and
in-house reference standards. For quality assurance, all reference standards
were analysed with the same GC
Overview of the L0–L4 compounds classified in air samples from Birkenes (southern Norway).
The L2 compounds include 11 potential PCBs. For those compounds the exact number of congeners might deviate since single reference standards for each PCB congener were not analysed. Polycyclic aromatic compounds (PACs) made up the largest subgroup of L3 compounds (see Fig. 6). Unknown halogenated compounds, which did not have any MS library match, were included in L4. An overview of the distribution of L0–L4 compounds in the GFF and PUF can be found in Table 3. The complete peak list of L0–L4 compounds is available in the Excel spreadsheet in the Supplement.
Overview of detected compounds confirmed with reference standards (L0 and L1) and probable structures (L2).
L3 compounds.
From 45 compounds classified as L1, 22 compounds are listed in one or more suspect lists, and from 80 compounds classified as L2, 28 compounds show similarity to one or more suspect lists (Table 3). As L2 compounds are not confirmed with reference standards, matches to suspect lists are slightly uncertain and compounds listed as L2 in Excel-SI may also represent different isomers.
The priority suspect lists chosen here were selected for the identification of the long-range atmospheric transport potential (LRATP) of CECs and hitherto unidentified CECs. However, the chosen suspects do cover the bulk of legacy POPs, CECs previously analysed at the Birkenes Observatory, and a large number of CUPs and non-regulated chemicals, especially own measured MS in the customised self-built libraries. The chosen suspect lists are considered relevant for Arctic air samples, and suspect prioritisation lists originate from different authors (Reppas-Chrysovitsinos et al., 2017; Brown and Wania, 2008; Coscollà et al., 2011; Hoferkamp et al., 2010; Howard and Muir, 2010; NORMAN network, 2016; Vorkamp and Rigét, 2014; Zhong et al., 2012) as well as self-built in-house suspect libraries (Table 3). A short summary of the data alignment of the suspect lists used and findings in our samples can be found in the Supplement.
The compounds and compound groups identified in the air samples from the Birkenes Observatory in this study are divided into three groups: (i) legacy POPs and PAHs, (ii) known CECs, and (iii) new potential CECs not previously reported in southern Norway (Birkenes; status October 2019). In addition to 36 already reported organic contaminants at Birkenes (including legacy POPs and known CECs), 92 new potential CECs with a match to reference standards (L1) or probable structures (L2) were identified (64 in PUF and 28 in GFF samples). It is interesting to note that 11 chemicals were common to the GFF and PUF sample. A total of 29 of the new potential CECs have an LRATP according to the Stockholm convention (UNEP, 2009a) half-life in air exceeding 2 d and may hence undergo long-range atmospheric transport.
Overall, 41 compounds identified as L0, L1 or L2 were also detected in high-volume air samples from the Zeppelin station (Ny-Ålesund) in Svalbard using the same analytical approach as in this study (Röhler et al., 2020).
A complete overview can be found in the Excel-SI spreadsheet, including information on the complementary findings in Arctic air samples, physical–chemical properties, additional information from a literature search, and further parameters on environmental properties (including persistence as well as the bioaccumulation and toxicity – PBT – classification by REACH and the Stockholm convention; European Parliament, 2018; UNEP, 2009a; see Table S7).
As summarised in Fig. 5, identified compounds were
grouped into different compound classes and arranged as previously detected or
previously not detected in air samples at the Birkenes Observatory (only
including L0, L1 and L2 compounds). For approximately
In total, 23 legacy POPs and PAHs were identified as L0, L1 or L2. The L0
and L1 were hexachlorocyclohexanes (
The presence of four known CECs (L0, L1 and L2), recently reported in
Birkenes air samples, was also confirmed by the approach applied here
(Nizzetto et al., 2019). These include the BFRs pentabromotoluene (PeBT, L2) and
hexabromobenzene (HBB, L0) as well as the organo-phosphorous flame retardants (OPFRs) triisobutyl phosphate (TBP, L1)
and
In addition to the identification of legacy POPs, PAHs and known CECs in air
samples from Birkenes, it was possible to detect 90 new potential CECs that
to our knowledge have not been previously reported in air samples from this
region. Most of these new potential CECs (
Of these 29 compounds, 14 were identified as L1 (
The four L1 compounds, which were identified in both the GFF and PUF samples,
were benzenesulfonamide (BSA),
The remaining five L1 compounds (only detected in PUF) were two intermediates, 1,4-benzenedicarbonitrile (terephthalonitrile) and 1-methyl-2-nitrobenzene (2-nitrotoluene), the biodegradation product tetrachloroveratrole, and the two combustion products 1-methoxy-2-nitrobenzene (2-nitroanisole) and 2-naphthalenecarbonitrile. Terephthalonitrile might be an intermediate for the production of the pesticide dacthal (Meng, 2012) and was detected together with two isomers of terephthalonitrile (probably positional isomers), which were classified as L2. 2-Nitrotoluene is used as an intermediate for the production of azo dyes and other dyes, rubber chemicals, agriculture chemicals, pharmaceuticals, and explosives (IARC, 2013; ECHA, 2008). The presence of 2-nitrotoluene may also indicate a degradation product of explosives like TNT (trinitrotoluene) (Mohsen et al., 2013). A possible local source could be a shooting range (6 km south-westerly) or military training area, which is approximately 30 km south-westerly from the Birkenes Observatory (NOU, 2004). The pesticide metabolite, or the bacterial biodegradation product tetrachloroveratrole, is formed during bleaching of wood pulp or chlorination of wastewaters in the pulp and paper industry (GovCanada, 2019; Su et al., 2008; Arinaitwe et al., 2016). Tetrachloroveratrole is a known priority pollutant found and monitored even in the Arctic (Su et al., 2008) but not previously reported in southern Norway background air. 2-Nitroanisole is mainly derived from combustion processes but can also be formed by atmospheric reactions (Stiborova, 2002). Large quantities of 2-nitroanisole were released into the atmosphere in the course of an accident at the Hoechst plant, Germany, in 1993 (Weyer et al., 2014). 2-Naphthalenecarbonitrile is related to plastic combustion, e.g. ABS (acrylonitrile–butadiene–styrene) plastic and polyester fabrics (Moltó et al., 2009; Watanabe et al., 2007; Wang et al., 2007; Moltó et al., 2006), but can also be used for the bluing of steel surfaces (Stefanye, 1972). The corresponding isomer 1-naphtalenecarbonitrile was classified as L2. Other compounds identified as L2 can be found in the Excel-SI spreadsheet.
Structural overview of L1 compounds classified as new potential CECs with LRATP.
Structural overview of L1 compounds classified as new potential CECs without LRATP.
Overall, 61 new potential CECs without LRATP were classified in Birkenes air
samples; 17 compounds were identified as L1 (
Four oxy-PAHs (1,2-BAQ, BPone, BAone, 9-Fone and one PAH: 3,6-DMPH) have
been previously detected in particle-related samples from three southern
European cities, with the highest concentrations during winter
(Alves et al., 2017), but to our knowledge they have not
been previously measured in southern Norwegian air samples. 3,6-DMPH and 9-Fone
were found in the PUF. BPone was found in the GFF, and 1,2-BAQ and BAone were found in the
GFF and PUF sample. The identified PAH and four oxy-PAHs were all previously
detected in wood combustion experiments (Czech et al., 2018),
and a local source cannot be excluded. A further group of compounds,
consisting of three terphenyl isomers (
Carbazole is mainly used in carbazole-containing polymers (PVK,
poly(-
A large number of L3 compounds (tentative candidates;
In the group of L4 compounds, 81 possible molecular formulas and unknown
halogenated compounds could be detected. Of these, 11 were classified as
potential unknown halogenated compounds (
A comprehensive sample clean-up method is one of the key factors for
successful SUS and NTS approaches. An ideal method removes interfering
matrix and at the same time keeps a maximum number of compounds of interest
in the extract. In this study, a novel sample clean-up method has been
developed and tested on spiked samples and real air samples. The results
demonstrate that this method is promising for target as well as SUS and NTS
analyses of regulated and emerging organic compounds in air samples. The
recoveries for legacy POPs and BFRs were comparable to those obtained with
the traditional acid clean-up method, but with the possibility to quantify
an extended range of compounds including the acid-labile POPs and BFRs. The
GC
In order to increase the effectiveness of future SUS and NTS studies in air,
expanding the suspect library with entries of relevant airborne contaminants
is considered essential. GFF- and PUF-based high-volume air sampling is a
widely used air sampling technique, but the polyurethane polymer used in the
foams generates a massive load of PUF-related matrix (often more than 20 000
compounds) which needs to be removed during sample clean-up or during post-acquisition data filtration. Reducing this load by developing cleaner PUFs
or replacing PUF with another adsorbent is an important next step in further
development of SUS and NTS methods for air samples. In future work, the
application of GC
All data are available in the Supplement. No further data available.
The supplement related to this article is available online at:
LR, MS, PBN and RK developed the idea behind this study.
LR performed chemical work and analysis, created the figures, and wrote the paper.
MS and PBN provided guidance and contributed to the paper preparation.
PR provided guidance for Z-Sep
RK provided financial support from internal NMBU funding, provided academic guidance and contributed to the paper preparation.
All authors read and approved the submitted paper.
The authors declare that they have no conflict of interest.
Special thanks to Anders Borgen at NILU for his help with target CUP A-C GC-HRMS analysis and quantification.
Compound structures were created using ChemOffice19 (PerkinElmerInformatics, 2019).
LogP and logD values were created using JChem for Excel (ChemAxon, 2019).
This research has been supported by the NMBU (grant no. 1205051013) and the NILU (grant nos. B111088, B116037). The study was funded by the following: NMBU, Norwegian University of Life Sciences, Ås, with an internal PhD grant; NILU, the Norwegian Institute for Air Research, Kjeller; and the Norwegian Ministry of Climate and Environment through two Strategic Institute Programs granted by the Norwegian Research Council (“Speciation and quantification of emerging pollutants” and “New measurement methods for emerging organic pollutants”).
This paper was edited by Ralf Ebinghaus and reviewed by two anonymous referees.