This section includes the summarized results from the NT workflow of the 271 Luxembourgish surface water samples analysed that are of solely qualitative nature. The workflow started with the optimization of the ppm and peakwidth parameters to perform feature finding. An example of different DoEs visualized by perspective plots can be found in the supplementary Figure S1 for the samples of April 2020 (in negative mode). In addition, a visualization of the best parameters determined for positive and negative mode for the same month is shown in Figure S2. The full list of optimized feature finding parameters for ppm and peakwidth can be found in Table S2.
After optimizing the feature finding parameters, the actual NT analysis of the measured samples was performed. Figure 4 shows the applied patRoon workflow with data collected for the ten April 2020 samples. In total 75,263 positive and 43,697 negative features were found in the first step of the workflow, totalling to 118,960 features. After feature grouping and filtering the number was reduced to 24,005 features in 7,581 feature groups. The reduction of features in this NTA workflow helps simplify the analysis process, improve interpretability, enhance signal detection, prevent overfitting, and optimize resource utilization. After the generation and filtering of MS peak lists, 15,140 positive compounds and 12,546 negative compounds could be assigned to the feature groups (see Fig. 4). Applying the identification scheme explained in the Data Analysis section [38], 76 positive and 73 negative compounds could be identified, from which 93 were unique compounds (lower identification levels not covered here). 56 compounds were overlapping, which means that they were tentatively identified in both positive and negative mode.
Most of the rivers are interconnected in Luxembourg and therefore the same compounds appear in several measurements. There are catchment specific pollutants - monitored by AGE - appearing mainly in the regions indicated in Fig. 1[11]. Figure 5A shows overlapping features (using a Venn diagram) for the four rivers monitored regularly. The most feature groups were detected for the river Syr, which overlapped most with the surface water from Chiers and Alzette_E (975). However, all four rivers are located in different catchments with different, region-specific influences and therefore the overlap is not 100%. In Fig. 5B a Chord plot for all feature groups in all rivers in April 2020 is presented. All rivers showed several overlapping feature groups with clear overlaps of some rivers belonging to one catchment, e.g. Gander and Mosel. However, this is not always the case, looking e.g. at the two rivers in the Lower Sûre catchment or the large overlap between Alzette_E and Chiers.
The analysis steps presented in Fig. 4 were accordingly performed for all 34 months and the resulting tables can be found in the GitLab repository. In Tables S3 and S4 the number of 2, 3a and 3b identifications for positive and negative mode, their sum and the number of tentatively identified unique compounds per level can be found. There was a majority in level 2 identifications compared to the level 3 numbers, e.g. for the April 20 samples there were 58 level 2s, 22 3as and 17 3bs. The total number of positive, negative and unique identifications (without discriminating between levels) is demonstrated in Fig. 6 based on Table S4. The numbers of positive and negative (unique) identifications are presented in yellow and green and a black bar shows the total number of unique compounds. The count of identifications in positive mode is generally higher than the negative count and their overlap is shown in blue. Overall, a total of 2479 and 375 unique chemicals were identified with level 2, 3a and/or 3b. The chemicals identified per month and in total can be found in Table S5, including tentative identifications of pharmaceuticals like valsartan or metformin, agrochemicals like 4,6-dinitro-o-cresol (DNOC) or their TPs like Flufenacet ESA and industrial chemicals like benzotriazoles, methylbenzenesulfonamide or bisphenol S.
Classification
To get a better overview and group the tentative identifications, classification steps were performed. First, an ‘interannual’ (April results of all years) and an ‘intraannual’ (2021 results of all months) comparison was performed, looking at the number of identified compounds per classyFire class and parent class (superclass). The month April was the only one measured in all years and 2021 was the only year where samples were available for each month. In general, for the interannual and intraannual comparison, 12 main parent classes (superclasses) could be identified: organic oxygen compounds, organohalogen compounds, nucleosides, nucleotides, and analogues, organic nitrogen compounds, organosulfur compounds, lipids and lipid-like molecules, alkaloids and derivatives, benzenoids, phenylpropanoids and polyketides, organic acids and derivatives, nucleosides, nucleotides, and analogues and organoheterocyclic compounds. 50 unique sub-classes of those very general superclasses could be assigned (46 in 2021), giving a more detailed picture. The underlying data (total numbers and percentage of compounds found per class and superclass in the inter- and intraannual comparison) are included in Table S6. An overview of those compound classes can be seen in Fig. 7 using the summarized identification numbers of all analysed months in 2021.
The treemap in Fig. 7 shows that nine superclasses with several subclasses could be identified for the intraannual comparison of measurements in 2021. Most of the chemicals were categorized as benzenoids (43%) followed by organoheterocyclic compounds (26%) and organic acids and derivatives (8%). Comparing the intraannual results of 2021 with the interannual comparison of the month April between 2019 and 2022, additional chemical classes were observed. One purine nucleoside, one sulfoxide and one compound belonging to the pteridines and derivatives class were tentatively detected in 2019. Purine nucleosides are generally not considered to be harmful to the environment or human health, as they are essential components of normal cellular functioning. Some sulfoxides have been shown to have toxic effects (e.g. dimethyl sulfoxide, DMSO), particularly when they are not properly disposed of or when they enter the water supply [39]. However, looking at the measurement results of April 2019, the compound was sulforaphane (in positive mode) at the sampling points Alzette_E, Syr, Mess, Mamer, Attert and Alzette_M (Alzette sampling point Mersch-Berschbach), which is a naturally occurring compound that is safe for human consumption and is even used in cancer treatment. The same applies for pteridines and derivatives, some chemicals of this class have been shown to have toxic effects (e.g. atrazine), but the identified compound was in this case ribovlavin, also known as vitamin B2. Overall, these examples show (and it is important to remember) that the toxicity of a chemical is complex and context-dependent, and should be evaluated on a case-by-case basis. Generally, the toxicity assessment in terms of environmental and health hazards is difficult, as the toxicity of a chemical can depend on a variety of factors, including its chemical structure and specific chemical properties, concentration, mode and duration of exposure, and the susceptibility of the organism or sensitivity of the ecosystem exposed. Additionally, different chemical classes can have different toxicities for different organisms, and different endpoints (such as acute toxicity, chronic toxicity, carcinogenicity, mutagenicity, and reproductive toxicity) may also be relevant. It has to be considered that some compounds may have multiple classifications, and their potential impact on the environment and human health may vary depending on the specific application. The use of classyFire is examined further in the Discussion.
To identify possible sources and estimate the environmental impact of the exposome related chemicals, a classification of the compounds in the inter- and intraannual comparison was performed, using the PubChem metadata of each chemical. The categories agroChemInfo and drugMedicInfo were chosen to evaluate trends of agrochemical and pharmaceutical use in one year and over three years. Moreover, information about possible disorders and diseases related to a compound and known commercial uses were analysed using the disorderDisease and knownUse categories. The resulting total and percentage trends are visualized using four line charts in Fig. 8 and the raw numbers are summarized in Table S7. It has to be considered that the categories identified are not exhaustive, and there may be some overlap between them (multiple uses per chemical).
Looking at the intraannual comparison of all months in 2021 an overall increase of total numbers in all categories could be monitored, but the overall percentage (relative to total numbers) stayed roughly the same. A majority of chemicals were associated to disorders and diseases (between 67% and 79%), 53% (July) to 72% (March) of the chemicals were assigned to the class of pharmaceuticals and the percentage of agrochemicals was between 7% (March) and 32% (May). This corresponds to the usual usual ‘spraying rhythm’ of farmers who increase pesticide and herbicide spraying in May to lay a foundation for the harvest. Almost all identified chemicals (93–100%) had a documented use, with multiple matches per compound when looking at the individual case in PubChem. The interannual values showed a sharp decrease of total identifications in drugs, disorders and diseases and known use, either due to effects of the COVID pandemic or due to measurement variations (less likely as the agrochemical curve stayed more or less constant). The percentage values (% of total identifications) showed a constant trend between the years with nearly all identifications having a known use, 75–79% associated to disorders and diseases like neurodegenerative or cardiovascular diseases, ~ 70% being drugs and 12–22% agrochemicals.
Besides using classification workflows, data from other water studies can be used to determine possible sources of exposome related chemicals. Former studies looking at Luxembourgish surface waters provided evidence that there are more pharmaceuticals and agrochemicals entering the environment than those included in the target monitoring by AGE that could potentially cause harm [9, 10]. Regarding agrochemical compounds in Luxembourgish rivers, a study by Krier et al. [9] was conducted (same instrumental methods used here). The study identified 162 pesticides and 96 TPs in the water samples (several chemicals not allowed in Luxembourg), applying suspect screening and transformation product screening. 31 chemicals were confirmed at level 1 [9]. Comparing these results to this study an overlap of 36 agrochemicals was seen, listed in Table S9. As this study was not focussing solely on pesticides, several compounds identified at lower identification levels may have other candidates in the study by Krier et al..
Singh et al. [10] performed a suspect screening, identifying 93 pharmaceuticals, adding quantification steps later. The AGE monitoring however, included just five pharmaceuticals (list of AGE from 2019 and 2020 in Table S8): carbamazepine, diclofenac, ibuprofen, ketoprofen, and lidocaine. All five chemicals were identified in the work of Singh et al. as well as in this study. Moreover, 58 pharmaceutical compounds were tentatively identified in this study, that were already confirmed in the results of Singh et al. [10]. The compared lists and overlapping identifications are summarized in Table S10. Singh et al. also registered the trend of decreasing pharmaceutical load looking at the years 2019 and 2020, explaining it with the reduction of medical treatments due to the COVID pandemic and lower precipitation data [10].
The presence of pharmaceuticals in Luxembourgish rivers highlights the potential to enhance waste water treatment and disposal systems, thereby safeguarding public health. Luxembourg’s biological and mechanical WWTPs are not always designed to remove pharmaceuticals from waste water, and some compounds are not effectively degraded by current treatment processes. As a result, these compounds can end up in surface waters (as shown here), groundwater, and even drinking water sources. A combination of geographical information, information on flow paths and additional measurement data from the inlet of a WWTP was used to analyse possible sources of the chemicals found in the river Chiers. The river is located in the south-west area of Luxembourg, at the border to France, with exposure to a set of different sources of pollution (see Fig. 9). Its source is in Obercorn, it passes the WWTP in Petange (green) and 6 km later the sampling point of this study (blue), located at the border to France.
The measurements at the WWTP took place between May and June 2022, resulting in 409 tentatively identified chemicals. Comparing those findings to this study (all results from Chiers in 2019, 2020, 2021 and 2022), an overlap of 178 chemicals could be identified. Figure 10 shows the number of overlapping chemicals per month compared to the total identifications (AGE sampling point). Those chemicals were probably coming from the WWTP with sources before this sampling point and result from incomplete filtering or there was chemical input between the WWTP and the AGE sampling point. Other chemicals were effectively filtered by the WWTP system or could not be identified at the later sampling point. For the overlapping chemicals, the four PubChemLite categories analysed above were examined as well, resulting in 36 agrochemicals, 130 pharmaceuticals, 143 compounds associated with disorders and disease and 170 known uses. Consequently, the same trend with dominating identifications of pharmaceuticals (73%) could be observed here, even after the filtering of the WWTP. Even persistent synthetic chemicals, like the per- and polyfluorinated compounds (PFAS) perfluorooctanoic acid (PFOA) or perfluorobutanesulfonic acid (PFBS) were found before the WWTP and downstream of the Chiers. However, without having quantitative data on their concentration, little can be said about their environmental effects.
Besides the overlapping chemicals it is interesting to analyse chemicals identified only at the border to France and not at the WWTP inlet (in total 165 chemicals). Looking for example at the results from April 2022 (1 month before the Petange WWTP sampling), drugs like pregabalin (antiepileptic), tramadol (analgesic) and its TP n-desmethyltramadol (with high aquatic toxicity) were found besides other compounds like 1H-benzotriazole (anticorrosive). Again, most of the unique tentative identifications at the border to France were pharmaceuticals (56%) and related to disorders and diseases (65%). The detection of the antiepileptic pregabalin could indicate the medication being used in this area (see population distribution in Fig. 9). Industrial chemicals, like benzotriazoles, only identified after the WWTP, could result from activity in the equipment and accessories, electronics, engineering or metal industry located in the area between WWTP and border. Other PFAS identified solely at the later sampling point were perfluorononanoic acid (PFNA) and perfluoroheptanoic acid (PFHpA), known for their use as surfactants, in fire fighting foams, for the manufacturing of plastics and in the semiconductor industry. However, these substances are now being phased out in many applications due to their persistence in the environment and potential adverse health effects. The list of 409 chemicals compared to the Chiers results of this study can be found in the Tables S11 and S12.
Lastly, a comparison to the chemicals covered by the governmental target monitoring (AGE) was performed, using the screening lists from 2019 and 2020 and the published results from the water quality report in 2022 [11] (Table S8). A total of 40 identified chemicals were overlapping with the target monitoring of AGE, including eight (of 16) catchment specific pollutants: carbamazepin, metolachlor, terbuthylazine, chlorotoluron, tebuconazole, flufenacet, metolachlor ESA and metazachlor OXA. Among the eight not detected chemicals were e.g., metolachlor OXA and metazachlor ESA, both being TPs of metolachlor and metazachlor, just as metolachlor ESA and metazachlor OXA. 335 chemicals not covered by target monitoring remained and were ranked based on their frequency of occurrence (number of months out of 34) in the Luxembourgish rivers between 2019 and 2022. The top 54 chemicals identified, occurring in at least 13 months, were listed with their common use, the PubChem Chemical Identifier (CID), the number of occurrences and additional information like the CID of the parent compound (in case of TPs). Chemicals without an environmental or health effect according to PubChem data were excluded, e.g. natural products, food additives or ubiquitous compounds like caffeine, reducing the list to 41 entries (Table 1).
Table 1
List of chemicals with high occurrence in the 34 months analysed and not monitored by AGE.
Synonym
|
Use
|
Parent Name
|
PubChem CID
|
No of occurrences
|
Irbesartan
|
pharmaceutical
|
-
|
3749
|
32
|
1H-benzotriazole
|
industry
|
-
|
7220
|
32
|
4- or 5-methyl-1H-benzotriazole
(co-elution)
|
industry
|
-
|
8705 or 122499
|
31
|
Amisulpride
|
pharmaceutical
|
-
|
2159
|
31
|
Telmisartan
|
pharmaceutical
|
-
|
65999
|
26
|
Celiprolol
|
pharmaceutical
|
-
|
2663
|
25
|
Fluconazole
|
pharmaceutical
|
-
|
3365
|
25
|
Trimethoprim
|
pharmaceutical
|
-
|
5578
|
24
|
4-Acetamidoantipyrine
|
pharmaceutical
|
Metamizole
|
65743
|
23
|
Codeine
|
pharmaceutical
|
-
|
5284371
|
22
|
Desvenlafaxine
|
pharmaceutical
|
Venlafaxine
|
125017
|
22
|
Tramadol
|
pharmaceutical
|
-
|
33741
|
22
|
4-NP
|
industrial
|
-
|
980
|
22
|
Valsartan
|
pharmaceutical
|
-
|
60846
|
21
|
Fexofenadine
|
pharmaceutical
|
-
|
3348
|
21
|
Salicylic acid
|
industrial, pharmaceutical
|
-
|
338
|
20
|
Flecainide
|
pharmaceutical
|
-
|
3356
|
20
|
Tiapride
|
pharmaceutical
|
-
|
5467
|
20
|
TCEP
|
flame retardant
|
-
|
8295
|
19
|
TCPP
|
flame retardant
|
-
|
26176
|
19
|
Flufenamic acid
|
pharmaceutical
|
-
|
3371
|
19
|
Aspirin
|
pharmaceutical
|
-
|
2244
|
19
|
Sitagliptin
|
pharmaceutical
|
-
|
4369359
|
19
|
triethyl phosphate
|
industrial
|
-
|
6535
|
18
|
Adipic acid
|
industrial
|
-
|
196
|
18
|
Losartan
|
pharmaceutical
|
-
|
3961
|
17
|
4-Formylaminoantipyrine
|
pharmaceutical
|
Aminopyrine
|
72666
|
17
|
Sulfamethoxazole
|
pharmaceutical
|
-
|
5329
|
16
|
Carbamazepine-10,11-epoxide
|
pharmaceutical
|
Carbamazepine
|
2555
|
15
|
Bicalutamide
|
pharmaceutical
|
-
|
2375
|
14
|
Dibutyl phthalate
|
industrial
|
-
|
3026
|
14
|
PFOA
|
industrial
|
-
|
9554
|
14
|
2-Hydroxycarbamazepine
|
pharmaceutical
|
Carbameazepine
|
129274
|
14
|
Sulisobenzone
|
consumer products
|
-
|
19988
|
13
|
Sulpiride
|
pharmaceutical
|
-
|
5355
|
13
|
3-Hydroxypyridine
|
industrial
|
-
|
7971
|
13
|
1-methylbenzotriazole
(possible co-elution)
|
industrial
|
1H-Benzotriazole
|
25902
|
13
|
D617
|
pharmaceutical
|
Verapamil
|
93168
|
13
|
Ensulizole
|
consumer products
|
-
|
33919
|
13
|
Tributylamine
|
industrial
|
-
|
7622
|
13
|
Hydrochlorothiazide
|
pharmaceutical
|
-
|
3639
|
13
|
26 of 41 chemicals tentatively identified are classified as pharmaceuticals (20) or their TPs (six), predominating the analysis results. Two TPs of carbamezapine (parent included in AGE monitoring) were tentatively identified: carbamazepine-10,11-epoxide in 15 months and 2-hydroxycarbamazepine in 14 months, whereas the parent compound was only found in 11 months of this study. Desvenlafaxine, a TP of the antidepressant venlafaxine, was detected in 22 months; the parent compound was already detected in the study by Singh et al. [10] and in 12 months of this study, having a known impact on aquatic environments even at low levels [40]. Out of the 26 pharmaceuticals, seven parent and five TP chemicals were not covered by Singh et al. [10] as well as three parents of the five TPs: metamizole, aminopyrine and verapamil. They might be measured after the study by Singh et al. was conducted or have been missed due to variations in the identification approach. The example of pharmaceuticals proves that it might be worth adding (more) TPs of monitored chemicals to routine target monitoring, as the parent is sometimes not visible and their TPs cause risks as well. Furthermore, 16 chemicals were listed covering uses in industry, consumer products and as flame retardants. The omnipresent benzotriazole class was detected in nearly all measurement months (32) with four different chemicals or TPs identified in this study. However, due to technique limitations (LC-HRMS) like insufficient separation, not all isomeric species can be correctly distinguished with the chromatographic method used, resulting in multiple possible identifications. Besides, two organophosphate flame retardants (OPFRs), namely tris(2-chloroethyl) phosphate (TCEP) and tris(1-chloro-2-propyl) phosphate (TCPP) were detected in 19 months, both known for environmental and toxic effects [41–43]. OPFRs serve as a substitution for brominated flame-retardants (BFRs) such as polybrominated diphenyl ethers (PBDEs), which have been found to cause adverse health effects in many samples recently [41, 44]. The three industrial chemicals 4-nitrophenol or 4-NP (22 months), dibutyl phtalate (DBP) and PFOA (both 14 months) were frequently detected in Luxembourgish rivers as well. 4-NP, used in many industrial applications, is known to have severe environmental and human health effects [45], just as DBP, a plasticizer with high aquatic toxicity [46] and PFOA, already listed for elimination in the Stockholm convention [47]. The chemicals in question were not previously subjected to monitoring by governmental institutions in Luxembourg. Although PFOA, other PFAS or DBP are not on the official monitoring list, these are being measured at AGE using specialized target methods and we recommend these efforts continue. Upon confirming their presence and determining their concentrations, it is advisable to contemplate their inclusion in the monitoring list of AGE, due to their known environmental and health implications.