Future water quality monitoring: improving the balance between exposure and toxicity assessments of real-world pollutant mixtures

Passive sampling

Using passive sampling methods to identify RBSPs during (chemical) monitoring offers several advantages for biological exposure characterization. Firstly, passive sampling can provide time-integrated information about specific aquatic pollutants over extended time periods (several weeks–months). This is a more realistic reflection of aquatic organism exposure in surface waters, compared to grab sample analysis (unless monitoring a discrete pollution event, where grab sampling may be more suitable). Secondly, passive samplers provide a measure of freely dissolved concentrations, rather than total concentrations. Freely dissolved concentrations are thought to be more comparable with single-chemical effect concentrations for aquatic species from laboratory studies (used in risk assessment for aquatic environments), due to their proportionality to the chemical potential and chemical activity [54]. Finally, passive sampler concentrations, after equilibration with sampled media, allow for a direct comparison of chemical levels in various compartments, thus helping to assess the compartmental distribution and to consider source and sink relationships, as well as to study accumulation and magnification of chemicals in aquatic biota. A guidance document on the use of passive sampling methods, resulting from the various efforts in SOLUTIONS, is provided [69].

A novel mobile dynamic passive sampling approach was introduced, which is applicable for characterizing chemical pollution along large rivers, lakes or sea transects, providing samples with chemical patterns integrated in time and space. This approach was applied in the Joint Danube Survey JDS3 [52, 68]. A case study on the combination of passive sampling, multi-residue analysis, and bioassays demonstrated the occurrence of 107 (out of 168) priority and emerging pollutants in the river Bosna (Bosnia and Herzegovina) [65]. The city of Sarajevo was identified as a major source of pollution, with downstream samples showing significant responses in all bioassays. While the estrogenic activity was largely explained by the specific estrogens measured, the drivers of the other observed effects remain largely unknown. Various compounds (diazinon, diclofenac, 17β-estradiol, estrone, benzo[k]fluoranthene, fluoranthene and benzo[k]fluoranthene) exhibited potential risks to aquatic biota, and indicated an inadequate water treatment infrastructure [65].

In situ large-volume solid-phase extraction

Combining chemical and bioanalytical techniques to characterize water samples shows great potential for identifying pollutants of potential concern more coherently, as component-based effect assessment can be compared with effect observations in the same sample [55]. Current limitations mostly relate to the amount of sample required for biological analysis, which typically involves multiple bioassays and enriched samples [1]. To overcome the logistical challenges associated with providing access to hundreds of liters of water samples, a novel, automated-solid phase extraction (LV-SPE) device was developed and tested for organic compound recoveries [60]. Good recoveries were observed for more than 200 compounds exhibiting a wide range of physico-chemical properties. Moreover, the generated extracts proved suitable for biotesting using various in vitro and in vivo bioassays [50], with effect recovery observations similar to those for chemical recovery for LV-SPE [51]. The device was used in various case studies comparing chemical and bioanalytical findings, to study how much of an observable effect in freshwater might be explained through chemical analysis using a bioanalytical equivalent concentration (BEQ) approach (e.g., [38, 47, 64].

Chemical analytical methods

The identification of RBSPs is desirable for chemical monitoring. Analytical methods to detect contaminants in low concentrations in water have improved progressively over time. A review of water contaminant detection performed at the onset of the SOLUTIONS project demonstrated that over 400 organic compounds were detected in European freshwaters, but that the overlap in the compounds analyzed was low due to the different multi-compound methods used [15]. Only 13 of the 426 compounds found were analyzed in all of the seven studies reviewed. Thus, it is currently difficult to tell whether reports on specific compound detections are site or method specific (or both). To improve transparency and allow for more rational choice and comparison of methods, we compiled standard operating procedures (SOPs) for more than 250 water contaminants. These SOPs were in two main forms. Firstly, ‘Master Methods’ contained detailed information on the multi-residue approaches developed to analyze different classes of priority and emerging pollutants (e.g., polycyclic aromatic hydrocarbons, polychlorinated biphenyls, organochlorine pesticides, polybrominated diphenyl ethers, novel brominated flame retardants, musks, perfluorinated compounds, currently used pesticides, pharmaceuticals and personal care products). Secondly, ‘Individual Compound Information Sheets (INCISE)’ contained essential and relevant information on the individual target compounds. These SOPs which were designed to simplify the identification and application of the developed methods by other research groups and laboratories, and to facilitate the effective monitoring and control of the targeted compounds, are compiled in a publicly available SOLUTIONS deliverable [71]. Kuzmanovic et al. [40] applied some of these multi-residue methods in the rivers of the Iberian Peninsula.

The analytical efforts mentioned above are restricted to target chemicals that are either known or suspected to occur and for which analytical standards are commercially available. To extend the universe of chemicals considered during water monitoring (for illustration see Fig. 1), we further explored novel routes of non-target chemical analysis using modern high-resolution mass spectrometry techniques.

Suspect screening and non-target analysis

While in target analysis, the analytes are known and standards are available, thus allowing for compound-specific method optimizations, in suspect analysis, standards are not necessarily available up front and in non-target analysis, even the structures of the compounds detected are not necessarily known [36]. Daily non-target monitoring of Rhine river water revealed its potential to support monitoring, revealing significant time series of unknown compounds at high intensities, for which the structures were subsequently elucidated and ultimately identified as industrial contaminants (Fig. 2 in [34]).

A major bottleneck hindering the widespread use of non-target analysis in RSPB identification results from the currently time-consuming and limited means of identifying compound structures out of the data-rich and complex mass spectral information. Within the SOLUTIONS project, we developed a workflow (Fig. 2) that ensured a systematic approach towards unknown compound identification, utilizing as much open data and software as possible. The workflow has been applied in several case studies, including the formation and elimination of transformation products through wastewater treatment including ozonation and several post-treatment steps [59]. Hierarchical cluster analysis across all treatment steps indicated that only a small portion of the non-target signals (9%) was formed during ozonation, while 54–83% of these signals were removed during post-treatment. The effectiveness of the advanced treatment and comparison of different post-treatment steps could already be demonstrated by bulk characterization parameters such as peak numbers or overall reduction in mass. The results of effect-based tools supported these conclusions regarding the effectiveness of treatment.

Beyond the identification of new compounds, NTS data may be used to characterize chemical contamination up to a continental scale by considering “chemical fingerprints” in surface water samples that may relate back to specific sources of chemicals or toxicity profiles [12]. The rapid progress in technology and distribution of powerful LC– and GC–HRMS technology for comprehensive NTS, open data repositories for digital freezing of samples together with the increasing availability of tools for big data evaluation and pattern analysis will pave the way for a more comprehensive assessment of chemical contamination in the near future.

Deconstructing mixtures into bioactive components

An approach is needed to identify (bioactive) candidate chemicals in complex mixtures that may be relevant on a larger spatial scale [24, 35]. The goal of a methodology that has been labeled virtual effect-directed analysis (vEDA) [24] is to assist explaining of biological effects by reducing the complexity of mixture components via multivariate statistics and pattern recognition methods on large sample numbers using a decomposition approach. This approach is able to handle peaks from non-target analysis, and thus is not restricted to previously known chemicals. Virtual EDA helps to identify peaks that co-vary with observed biological effects, suggesting these as candidate causative chemicals (Fig. 3). Obviously, this approach does not directly provide cause–effect relationships, but allows hypothesis generation, which must be confirmed using, e.g., literature and database review, calculation of toxic units using quantitative structure activity relationships (QSAR), or a full chemical and effect assessment with reference standards. Successful vEDA generally requires that:

1. 1.

   The observed effect is caused by a limited (small) number of toxicants among those present in the samples.

    
2. 2.

   Sufficient variance (larger than the data uncertainty) of the observed effect and chemical composition patterns occurs across the different samples.

    

A case study on a time series of mutagenic wastewaters from a mixed industrial and municipal WWTP serves as an example. Varying levels of mutagenicity were detected at different time points over approx. 6 weeks, along with thousands of chemical signals of varying intensity from LC–MS non-target screening. Applying partial least squares analysis, the number of peaks of interest to explain the variability in mutagenicity was reduced to about 200 signals [35]. The overrepresentation (30 times larger) of nitrogen-containing compounds among the selected peaks, along with enhanced mutagenicity in a diagnostic Ames Salmonella stem (YG 1024) suggested aromatic amines as drivers of mutagenicity. After specific derivatization techniques were applied [44], several of these compounds could be identified. The intensity of two peaks, in fact two diaminophenazine isomers, were found to correlate with mutagenicity and were eventually confirmed as the drivers of the observed mutagenicity [45].

Ecological health status diagnosis

Monitoring the presence and abundance of different species forms the basis for an ecological status assessment of water bodies. Exposure to toxic chemicals can affect freshwater aquatic life and may lead to shifts in the composition of freshwater communities through the loss of sensitive species. Toxic pressure is, however, not the only possible reason for variations in the health of aquatic biota. Other non-chemical factors such as hydrological conditions or general water quality parameters can have a strong impact on aquatic ecology, possibly confounding diagnostic efforts.

Chemically induced shifts in community composition are expected to correlate with chemical exposure and can, thus, be used as detectors. Additionally, considering species traits, i.e., biological characteristics relevant for the elucidation of chemical effects, rather than using the occurrence of different species, can help to increase diagnostic power or ecological information [66]. Therefore, an approach statistically separating chemical from non-chemical impacts on taxonomy or traits-based community composition was developed and tested in SOLUTIONS [56]. It builds on the variance partitioning technique of Borcard et al. [7].

Prerequisites for the application of the variance partitioning method are chemical exposure data and information about other stressors such as general water quality parameters (pH, water hardness, nutrient levels, etc.) or hydro-morphological characteristics. A comprehensive statistical analysis of the relative importance of the different stressor groups for community composition can be obtained by a stepwise refinement of the explaining factors, increasing the resolution of the method by ‘zooming’ into single factor groups [56, 57]. In addition to the correlative analysis of the impact on variance, groups of chemicals (e.g., using class or modes-of-action) can be ranked and statistically tested concerning their impact on the composition of species or their properties across a number of sampling sites using Monte Carlo permutation testing. The methodology can be used to check whether measured chemicals show a statistically significant influence on existing community composition data.

This method was applied for a data-rich case study from the 3rd Joint Danube Survey, where information on the community compositions of aquatic macroinvertebrates, concentrations of about 300 organic pollutants and data on habitat characteristics, hydromorphology and general water quality parameters were available for 55 sampling sites along the whole Danube [56]. In this study (Fig. 4), variation of structure and trait composition of the invertebrate community was mainly explained by habitat and water quality parameters, whereas hydromorphological alterations were found to play a less important role. Physico-chemical water quality parameters explained a larger part of the variation in the invertebrate community, compared to metals or organic contaminants (Fig. 4). Nevertheless, 8.7 and 12.5% of the variation in the community compositions in this study were significantly correlated with organic pollutants, while the ‘flat’ exposure profile along the Danube sampling sites probably impaired the identification of a larger share [56]. This evaluation is dependent on the number of chemicals analyzed (here, 227 metals and organic compounds were considered) and the availability of toxicity data (75% of which had to be estimated using QSARs). Accounting for more contaminants and using more measured effect data would, therefore, also lead to an increased contribution. A second application on data from the Swiss EcoImpact study confirmed the usability of the variance partition approach. Up to 12.7% of variation in the community data was explained by chemical pollutants [14]. A related partition of variance exercise (redundancy analysis—RDA) was applied to Iberian Rivers [57]. The total explained variance on the biofilm and invertebrate communities was 86%, where 2.2% was directly attributed to organic micropollutants, 5.7% to land uses, and 10.6% of the environmental variables (nutrients, altered discharge, dissolved organic matter). The total shared variance of the three groups of stressor variables amounted to 41%.

The developed methodological approach allows additional information to be extracted by pooling different existing data. Moreover, it can be standardized and may be applied on a routine basis.

Effect detection using bioassay panels

Clearly, using ecological data to determine the link to specific chemical pollutants will remain elusive and of limited resolution for the foreseeable future due to the lack of diagnostic power of existing methods and/or the presence of multiple stressors, e.g., in urbanized areas. Bioassays are a complementary method to improve the detection of potential adverse effects from toxicologically relevant compounds and to help account for combination effects from mixture exposure [1, 13, 18, 70]. Moreover, if we strive to exit from the uncertainty bias (i.e., drive effect assessments by uncertainty factors to bridge the knowledge gaps), it is vital to add effect-based observations to the samples being assessed for chemical contamination.

Comprehensive analysis of uncharacterized but biologically active mixtures of organic contaminants requires additional effect-based approaches to be adopted for analyzing enriched water samples, as a complement to chemical analysis [25, 27]. We developed a systematic approach for capturing anticipated effects from such complex chemical mixtures. First, we compiled lists of water contaminants found in monitoring studies and translated these into mode-of-action (MoA) categories [15]. These MoAs were then used to devise panels of bioassays designed to comprehensively capture biological effects of chemicals expected in water [1, 23]. Subsequently, bioassay panels were applied to single water contaminants to test the MoA categorizations [48, 49], as well as mixture effect recovery in complex contaminated samples [2] and in various monitoring case studies [38, 47, 49, 52, 65].

The construction of a comprehensive diagnostic bioassay panel that captures all known MoAs using a specific bioassay is not yet realistic. For two-thirds of the 426 organic chemicals detected in monitoring studies of European freshwater, about 100 distinct biological molecular targets were identified and subsequently grouped into 30 categories of MoA. For the remaining chemicals, mostly transformation products, insufficient information was available to allocate a MoA [15]. There are few bioassays (typically, but not exclusively, in vitro assays) that are suitable for the proposed monitoring and capable of detecting specific biological effects beyond endocrine activity, photosystem II inhibition, mutagenicity and metabolic activation. Thus, for effect categories such as neurotoxicity or inhibition of mitosis and sterol biosynthesis, relevant water contaminants could escape attention [58].

At this stage, we suggest using modular panels of bioassays in effect-based water monitoring (Fig. 5). A module consists of different bioassays that allow similar interpretation in terms of specificity of exposure and effect observations. Different modules may be combined, depending on the monitoring scope, which may range from surveillance of individual contaminants or critical effects through to comprehensive status assessment. Modules are, thus, distinct from a tiered testing scheme as used in environmental risk assessment for single compounds. The first module comprises a set of short-term exposure organism assays, to capture an array of apical effects. While the diagnostic value for exposure will be restricted to the discrimination between phyla-specific effects, the assessment can be directly related to biological quality elements used in ecological status assessment. Thus, if the monitoring question relates to determining whether a compromised ecological status may be caused by chemical contamination, effect-based tools from the module using endpoints easily related to biological quality elements (fish, macroinvertebrates, phytoplankton, macrophytes) may be the first choice. The second module comprises of bioassays, cell- and organism based, that detect specific chemical–biosystem interactions that are indicative of chronic effects. Depending on our knowledge about the toxicodynamics related to a certain chemical–biosystem interaction, the assay responses may be used to either detect exposure for a defined group of compounds or indicate long-term effects, which may be overlooked when using short-term assays only. Here, a driving objective for the selection of bioassays would be the surveillance of drinking water abstraction and safeguarding against potential human health effects.

Indication of a long-term effect after exposure to a compound group may be performed through an established adverse outcome pathway, e.g., endocrine disturbance. The associated bioassays may be regarded as indicative for a defined long-term effect. For example, receptor-mediated endocrine disruptive activity of various types can be detected with high sensitivity using receptor-based assays that are either cell based or employ transgenic organisms. While a detailed discussion of AOPs is out of the scope of the current manuscript, it is fair to say that currently, we lack fully established AOPs ready for ecotoxicological application. It is, however, possible to integrate exposure detection across a larger number of compounds for stress response bioassays such as oxidative stress response or mutagenicity. This builds on the detection of signals downstream of a primary chemical–biosystem interaction and while they detect exposure, it remains context dependent whether the responses indicate compensation of or propagation towards adverse effects. Similarly, metabolic activation assays, such as AhR binding assays, which are typically promiscuous in their binding specificity, are known to elucidate pleiotropic responses and are, thus, easiest to use for mixture exposure detection.

Typical additional factors that determine the bioassay selection comprise rapidity, sensitivity, adequacy, statistical robustness, high reproducibility, accepted level of standardization, automated protocol, and demonstrated use for monitoring purpose, potential for inference, cost effectiveness and degree of representativeness as a biological proxy. Again the purpose of the study is most important, e.g., if assessment of contamination by lipophilic compounds is of concern, dioxin-like effects measurable with different types of bioassays, such as AhR receptor-binding assays, will be important. However, if more polar and water-soluble compounds are of concern, these assays are unlikely to detect much bioactivity. To foster the application of the bioassays applied in the case studies summarized above, one may refer to the standard operating procedures documented in the supporting information of Neale et al. [48, 49].

Bioassay findings cannot always be explained by monitored chemicals, as demonstrated in several case study investigations in surface [26, 38, 47, 49], and groundwater [39]. This is particularly true for assays indicative of more integrative effects. Therefore, it is useful to develop bioassay-specific trigger values that can serve for the assessment of observed effects at a given level of sample dilution/concentration. Whole effluent testing using bioanalytical methods, such as fish embryo, daphnia, luminescent bacteria or algae testing is performed already for wastewater surveillance in the German Waste Water Ordinance [32] under the Federal Water Act [31]. This regulation adopts the EU-WFD and provides a reference case for effect-based monitoring and assessment. Here the lowest inhibitory dilution in an apical bioassay deemed acceptable for a specific type of effluent is defined. In SOLUTIONS, the single chemical fingerprinting and mixtures studies were used to derive bioassay-specific effect-based trigger values (EBT), which would allow assessment of other effects beyond apical effect-based monitoring findings. The derivation of a more generalized approach for setting EBT values was based on the European environmental quality standards (AA-EQS), on data generated in SOLUTIONS, and on literature data [28]. The derived EBTs are still preliminary, because bioanalytical effect data are generally only available for chemicals with existing EQS. More single chemical data must be obtained for the different assays before specific EBT values can be considered ready for harmonization.

Effect-directed analysis

Despite the presence of mixtures of multiple compounds in environmental media, theoretical considerations and experimental findings suggest that the overall risk to individual organisms or populations of a species may be driven by only a few mixture components [3]. Thus, identification of the most significant chemicals contributing to observed effects will help to establish the corresponding cause–effect relationships and provide focus for potential management measures.

In some cases, well-known chemicals can explain observed biological responses (e.g., estrogenicity detected in surface water using in vitro assays often results from only a few natural and synthetic steroid estrogens such as estrone, estradiol, and ethinyl-estradiol) [33, 37, 38, 47]. However, in other cases routinely analyzed chemicals cannot explain observed biological responses (e.g., [26, 47, 49]. In these cases, bioassays provide a more comprehensive picture of the chemical burden in the aqueous environment. Efforts in the SOLUTIONS project, therefore, aimed to better understand which (organic) chemicals contribute to observable effects and what fraction of the effect is caused by unidentified chemicals, using a stepwise approach shown in Fig. 6.

When determining drivers of mixture toxicity using this approach, initially chemical target monitoring data should be evaluated using component-based mixture toxicity predictions. For chemicals with known effect concentrations (ECs), Toxic Units (TU) can be calculated as the ratio between environmental concentrations and ECs for a specific water quality element (e.g., phytoplankton or invertebrate fauna). This can be used to identify candidate drivers of toxicity. Assuming concentration addition as a model for mixture toxicity, the sum of toxic units (ΣTU) can be used as a default approach [5]. This step may already provide a basis for a tentative prioritization of sites and rivers of concern, as well as for the identification of sensitive biological quality elements. Existing chemical monitoring data may subsequently be compared with effect-based monitoring using a panel of biological endpoints (see above), that may be adapted for specific diagnostic purposes. More recent work has shown that it may be reasonable to expect a full mass balance for specific responses, for example receptor-mediated effects. An iceberg model [62] can be used to quantify the differences between expected (component-based) and observed (bioassay-based) mixture effects.

Subsequently, a full, unbiased EDA investigation can be used to investigate whether unidentified chemicals account for unexplained biological effects. The EDA methodology works without any previous information on the types and sources of pollution. The data evaluation and investigations suggested in the first step will help to decide if EDA should be applied to unravel any remaining unexplained toxicity. Basically, EDA reduces environmental sample extracts to less complex mixtures or individual compounds by fractionation and subsequent bioassay-directed selection of subsamples, so that relevant toxicants can be isolated and identified by chemical analysis [9]. Finally, identified toxicants need to be confirmed as the cause of the measured effect. This is carried out using analytical confirmation of identified structures as well as effect confirmation by testing neat standards and artificial mixtures. Furthermore, it may involve mixture toxicity modeling, and finally hazard confirmation, which should account for effects at higher levels of biological organization, such as populations and communities under realistic exposure conditions [10]. The power of this approach has been demonstrated, e.g., for anti-androgenic effects detected in vitro in a small river in Germany impacted by treated wastewater [43]. Parallel fractionation with different stationary phases together with testing and chemical screening of the resulting fractions was able to reduce the number of candidate peaks to very few peaks, eventually identifying and confirming the fluorescent dye coumarin 47 as the driver of the measured effect in vitro and in vivo in Medaka embryos. Site-specific compounds from local economic and social activities, such as in this example, typically go unrecognized in chemical analyses, as well as in ecological monitoring.

Effect-based monitoring combined with EDA and mixture effect experiments also support the identification and understanding of effects driven by the interaction of different compounds. This has been demonstrated by partial unraveling of mutagenic effects in the river Rhine, establishing synergistic effects of industrial aromatic amines with natural carboline alkaloids, which co-occur frequently in river water [46].

Ecology-directed analysis

This approach provides an adaptive and integrative method that systematically synthesizes the evidence from the different LOEs and provides optimum decision support for an ecologically oriented water management.

The overall status of each of the four individual LOEs is condensed and categorized into classes that indicate clear or moderate signals of pollution-driven impacts (Fig. 7). In an application example, data obtained for the single LOEs from the 3rd Joint Danube Survey were analyzed [6]. Results from in-depth chemical analyses of water samples (see above) were used for predictive mixture toxicity modeling (ΣTU). Results from a battery of in situ biomarkers in sentinel fish (Alburnus alburnus and Neogobius sp.) [17], including mainly markers for exposure (enzyme activities for biotransformation, oxidative stress and neuronal activation) and effects at cellular levels (genotoxicity), were analyzed and aggregated using an index for the average biomarker response. Finally, taxonomy- and trait-based analyses of fish and macroinvertebrate community data were performed to indicate ecological impacts [56]. A particular problem was the limited spatial overlap between sampling sites for the different LOEs. Results from a suite of in vitro bioassays, performed with extracts from LV-SPE and equilibrium passive sampling, were considered [47] but could not be fully included due to the spatial mismatch. An independent analysis regarding the evidence of genotoxic bioactivity was performed instead [61].

Despite these constraints, the toolbox application helped to identify a number of sites in the Danube where component-based mixture risk predictions, biomarker responses and community level responses, consistently indicated a chemical-driven impact. Many of the Danube sampling sites show clear indications of anthropogenic impacts, and in all cases the estimated toxic pressure suggests that pollutants are likely a contributing cause. However, the functional in situ responses for many sites indicate that the link from functional measures of toxic pressure to community effects is not as clear as anticipated when comparing chemical pressure and community effects. Here, biomarkers and average biomarker responses could provide additional information to support the overall evaluation of the chemical and ecological quality of water bodies.

The LOE-based approach transformed the multi-dimensional JDS3 data into a simplified matrix suitable for water managers and decision-makers, without losing crucial information. This matrix can serve as a basis for conclusive statements about the impairment of the ecological status at the various sites. It also pinpoints critical data gaps, which might stimulate and guide future chemical monitoring and ecological testing.