Suspect Screening of Chemicals in Hospital Wastewaters Using Effect-Directed Analysis Approach as Prioritization Strategy

The increasing number of contaminants in the environment has pushed water monitoring programs to find out the most hazardous known and unknown chemicals in the environment. Sample treatment-simplification methods and non-target screening approaches can help researchers to not overlook potential chemicals present in complex aqueous samples. In this work, an effect-directed analysis (EDA) protocol using the sea urchin embryo test (SET) as a toxicological in vivo bioassay was used as simplified strategy to identify potential unknown chemicals present in a very complex aqueous matrix such as hospital effluent. The SET bioassay was used for the first time here to evaluate potential toxic fractions in hospital effluent, which were obtained after a two-step fractionation using C18 and aminopropyl chromatographic semi-preparative columns. The unknown compounds present in the toxic fractions were identified by means of liquid chromatography coupled to a Q Exactive Orbitrap high-resolution mass spectrometer (LC-HRMS) and using a suspect analysis approach. The results were complemented by gas chromatography-mass spectrometry analysis (GC-MS) in order to identify the widest range of chemical compounds present in the sample and the toxic fractions. Using EDA as sample treatment simplification method, the number of unknown chemicals (>446 features) detected in the raw sample was narrowed down to 94 potential toxic candidates identified in the significantly toxic fractions. Among them, the presence of 25 compounds was confirmed with available chemical standards including 14 pharmaceuticals, a personal care product, six pesticides and four industrial products. The observations found in this work emphasize the difficulties in identifying potential toxicity drivers in complex water samples, as in the case of hospital wastewater.


Introduction
The occurrence of contaminants of emerging concern (CECs), including pharmaceuticals and personal care products (PPCPs), pesticides or industrial chemicals, among other organic micropollutants, in the environment has increased during the last decades [1][2][3]. Although CECs are often found at trace levels, their daily use and continuous release into the aquatic environment are of particular concern and make the assessment of the risk that their occurrence poses to the environment and human health necessary [4].
The amount and variety of CECs is particularly large in industrial, urban and/or hospital wastewaters [4][5][6][7]. Wastewaters are often treated in wastewater treatment plants (WWTPs) but the elimination of CECs is not completely achieved and CECs as well as their transformation products can reach the aquatic environment [4,5,8]. Evidence of poor elimination of some CECs, including PPCPs, in WWTPs can be found in studies carried out in a WWTP located in the Basque Country (Galindo WWTP) [9], which receives the

Quality Assurance and Quality Control
In this work, a previously optimized and validated fractionation protocol for EDA was used [41] (see Table S1 in Supplementary Materials). Regarding the SET bioassays quality parameters, procedure blank samples (LV-B) and test control samples were prepared. On the one hand, solvent control samples (i.e., filtered seawater (CFSW) and DMSO (CDMSO)) were prepared and analyzed in parallel with the samples (see Section 3.3.2). Solvent control samples did not exhibit any effects on the two toxicological endpoints tested: sea urchin size increase (SI, %) and larvae morphology (IT) ( Figure S2). Three replicates of blank samples (LV-B) were fractionated to detect contamination during the fractionation step (LV-B-R). None of the tested blanks induced any effect on the SI (%) of the exposed larvae ( Figure S2) and neither carryover nor contamination was detected in the chemical analyses.

Hospital Effluent Toxicity Evaluation and Identification of Toxic Fractions through EDA
The level of toxicity of the control samples and the raw sample of the hospital effluent was tested in triplicate at eight concentration levels in the range of REF 0.001-10. Based on a PROBIT-modelled log dose-response curve (i.e., size increase vs. REFs), the raw sample showed embryo growth inhibition and skeleton malformation activity within the concentration range tested, except the two lower ones (REF 0.001 and 0.01) where four-arm pluteus stage was reached and were not significantly different from the controls (ρ < 0.05, Kruskal-Wallis) (see Figure 1). Growth inhibition was observed at concentration levels equal to REF 0.1 while concentration values above REF 1 caused full inhibition. In this way, the EC 50 value was set at REF 0.07, and the calculated TU bio value was 14.1 at that EC 50 level. The IT values were also calculated for different REFs ( Figure S3A,B). Based on those values, a significant toxicity increase was observed in the range of 0.05-0.1 REF, and the maximum IT values (IT = 3) were reached at REFs ≥ 0.5.
The toxicity of the RS was tested at concentrations between REFs 0.25-1.5 to determine any total toxicity loss during the fractionation step. Based on the log-dose-SI response curve (see Figure 1), a bioactivity loss was observed for the recombined raw sample (RS) (EC 50 = 1.73). Since the quality parameters of the EDA fractionation are acceptable [41], the decrease in the bioactivity can be attributed to (1) the type of bioassay used, as it is a phenomenon often found when bioassays indicating adaptive stress responses or apical endpoints are used [42], or (2) the removal of the organic matter present in the raw sample due to fractionation [43].
A toxicity screening of all 18 primary fractions was performed at REF 0.07 (equivalent to EC 50 of the raw sample) and REF 0.3 (equivalent to EC 85 of the raw sample). Toxic effects measured as SI% and IT were only observed in three of the fractions tested (F6, F7 and F8) at REF 0.3 (ρ value < 0.05, Kruskal-Wallis) (see Figure 2A,B for REF 0.3 and Figure S4 for SI (%) at REF 0.07).
In order to reduce the sample complexity to identify chemicals, the three toxicologically active fractions (i.e., F6, F7 and F8) were combined (ΣF) (EC 50 ≈ REF 0.5) and fractionated again, obtaining 16 secondary fractions (see Section S2.2 in Supplementary Materials). Secondary fractions were tested at REF 0.5 considering the better isolation of the toxic compounds when fractionating, and therefore, the need for higher REFs for toxicity detection. However, none of the fractions obtained after the second fractionation stood out as toxic, suggesting a spread of the toxicity driver compounds over many fractions, below the detection limit (ρ value < 0.05, Kruskal-Wallis) ( Figure S5).
Comparisons of dose-response relationships between the raw sample, RS and ΣF revealed that the raw sample was the most toxicologically active, being significantly different (p-value < 0.05, 95% confidence level) ( Figure 3). Furthermore, the overlapping intervals of those may indicate that further isolation of the key toxicants and/or organic matter removal was achieved with the second fractionation, even if the toxicity was not recovered in a single fraction. Moreover, the EC 50 and TU bio values given in Table 1 sustain the significant differences between the raw sample and the toxicologically active fractions since about 16% of the hospital effluent bioactivity was recovered in ΣF (TU bio = 2.3), while the remaining unexplained activity was probably spread over the other fractions [30,42].  Figure 2A  In order to reduce the sample complexity to identify chemicals, the three toxicologically active fractions (i.e., F6, F7 and F8) were combined (ΣF) (EC50 ≈ REF 0.5) and fractionated again, obtaining 16 secondary fractions (see Section S2.2 in SI). Secondary fractions were tested at REF 0.5 considering the better isolation of the toxic compounds when fractionating, and therefore, the need for higher REFs for toxicity detection. However,

Identification of Unknown Compounds
Hospital effluent raw sample and all toxic fraction extracts were analyzed through LC-qOrbitrap and GC-MS to identify as many as possible unknown organic compounds. Concerning LC-qOrbitrap, in total, 446 features were among the compounds belonging to the relevant suspect lists in the raw sample, but only 94 features were present in the toxic fractions (see Table S2 in SI). Concretely, based on the Schymanski classification [44,45], the presence of 23 compounds was confirmed (level 1), 29 compounds were tentatively identified as probable structures (Level 2a or 2b) and 42 as tentative candidates (Levels 3 and 4) (see Table S2 in SI for the full list of identified features). In Figure S6 (in SI), the extracted ion chromatogram (A) and the mass spectra similarity between the experimental and the mzCloud library (B) of bisoprolol are included as an example of level 1 compound identification.
The compounds identified as level 1 were quantified in the raw hospital effluent sample (see Table 2). Among the compounds identified as level 1, 14 pharmaceuticals were detected at concentration levels between 3 ng/L (propranolol) and 37 ng/mL (furosemide). In addition to pharmaceuticals, seven pesticides (one herbicide, an insecticide and four fungicides), an industrial product, a preservative and a flame retardant were also quantified (see Table 2).

Identification of Unknown Compounds
Hospital effluent raw sample and all toxic fraction extracts were analyzed through LC-qOrbitrap and GC-MS to identify as many as possible unknown organic compounds. Concerning LC-qOrbitrap, in total, 446 features were among the compounds belonging to the relevant suspect lists in the raw sample, but only 94 features were present in the toxic fractions (see Table S2 in Supplementary Materials). Concretely, based on the Schymanski classification [44,45], the presence of 23 compounds was confirmed (level 1), 29 compounds were tentatively identified as probable structures (Level 2a or 2b) and 42 as tentative candidates (Levels 3 and 4) (see Table S2 in Supplementary Materials for the full list of identified features). In Figure S6, the extracted ion chromatogram (A) and the mass spectra similarity between the experimental and the mzCloud library (B) of bisoprolol are included as an example of level 1 compound identification.
The compounds identified as level 1 were quantified in the raw hospital effluent sample (see Table 2). Among the compounds identified as level 1, 14 pharmaceuticals were detected at concentration levels between 3 ng/L (propranolol) and 37 ng/mL (furosemide). In addition to pharmaceuticals, seven pesticides (one herbicide, an insecticide and four fungicides), an industrial product, a preservative and a flame retardant were also quantified (see Table 2). Concerning the analysis by GC-MS, the presence of two compounds (see Table S2), diethyl phthalate (DEP) and 2,4-di-tert-butylphenol (2,4-DBT), was confirmed and they were also quantified in the hospital effluent (see Table 2).
Several worldwide studies reported the presence of some of the detected compounds in hospital effluents. Santos et al. [7] reported the occurrence of the pharmaceuticals furosemide, ofloxacin, propranolol, sulfamethoxazole and verapamil in four different hospitals located in Coimbra (Portugal) at similar concentration levels to those determined in this work. Although at lower concentration levels, the compounds diphenhydramine, furosemide, methylparaben, omeprazole, primidone, propranolol, sulfamethoxazole and verapamil were also detected in hospital effluents monitored in New York [6]. The presence of fluconazole, furosemide and clozapine was also reported in a study carried out in Uppsala (Sweden) where hospital effluent samples were analyzed [46]. On a national scale, the occurrence of omeprazole, propranolol and sulfamethoxazole was observed in hospital effluent (Galdakao, Biscay) but at higher concentrations than the ones determined in this study. To the best of our knowledge, the occurrence of the remaining compounds detected in this study has not been reported in any other hospital effluent. Nevertheless, the occurrence of such compounds can be expected. For example, although the presence of the pesticides acetamiprid or spiroxamine was not reported in other hospital effluents, pesticides such as atrazine or metalaxyl were found in a hospital effluent located in Beijing, China [47]. The presence of OBT, the hydroxylated metabolite of benzothiazole, can be explained due to the extensive use of benzothiazole compounds (e.g., corrosion inhibitors, ultraviolet light stabilizers in textiles and plastics, food flavoring agents and pharmaceuticals) which can end up not only in environmental compartments [10,11,48] but also ingested by humans and then excreted as OBT via urine [49]. 2,4-DBT, DEP and TEP are widely used plasticizers (e.g., in PVC plastics) and have already been detected in several aqueous environmental samples (e.g., in sewage waters [4,50,51], surface waters [4,51,52] and drinking waters [50]); hence, they are likely to be found in hospital effluent.

Assessment of Potential Toxicity of Identified Compounds
The potential toxicity of all the compounds identified as level 1 was assessed (Table  S3) at the concentration level quantified in the raw sample. At those conditions, none of the candidates showed any measurable effect at REF 50 (see Figure S7A,B in Supplementary Materials), concluding that, even if they were identified in the toxic fractions, none of them could be identified as responsible for the sample toxicity at tested conditions. Due to the lack of available ecotoxicological data for sea urchins, an additional toxicity assessment of the candidates was based on the acute toxicity values (EC 50 and LC 50 ) reported in the literature for fish and invertebrates (Table S3). The effects of the individual compounds observed in this study are in line with the results reported in other studies [53][54][55][56][57][58][59][60][61][62][63][64][65][66][67][68][69], online available databases (Pesticides Properties Database, PPDB, http://sitem.herts.ac.uk/aeru/ppdb/, accessed on 2 May 2022) and estimated values with Ecological Structure-Activity Relationship (ECOSAR) modelling (version 2.0, downloaded from https://www.epa.gov/tscascreening-tools/ecological-structure-activity-relationships-ecosar-predictive-model, accessed on 29 April 2022) since the EC 50 and LC 50 values are much higher than the calculated concentrations of the compounds. In any case, as can be observed in Table S3, EDDP (0.000032 mg/L) was ranked at the top of the list of toxicants for invertebrates, followed by TEP (0.02 mg/L) and 2,4-DBT (0.281 mg/L), fenpropidin (0.54 mg/L), isoproturon (0.58 mg/L), eprosartan (0.593 mg/L) and propranolol (0.8 mg/L) to a lesser extent. As for fish, the lowest values of LC 50 were compiled for eprosartan, 2,4-DBT and EDDP (0.011, 0.144 and 0.43 mg/L, respectively). Moreover, a previous EDA study to address the impact of WWTP effluents on coastal ecosystems through SET reported similar levels of fenpropidin (23 ng/L) and an experimental EC 50 value of 560,000 ng/L explaining <0.1% of the whole toxicity [30]. Prieto-Amador performed an ecotoxicological assessment of several phthalates in the sea urchin P. lividus, in which an experimental EC 50 value of 758.58 µg/L for DEP was reported [70], thus evidencing its low contribution to the toxicity detected in this study. On the other hand, Ribeiro et al. reported a significant increase in the percentage of total sea urchin abnormalities over 125 µg/L of propranolol [71].
The toxicity of the chemical mixture was also tested, but as occurred when the individual chemicals were tested, the mixture did not show toxicity up to a concentration 1000-fold above the original hospital effluent concentrations (see Figure S8).
Similar to the observations found in this work regarding unexplained toxicity, several studies integrating biological and chemical analysis showed unexplained effects for a variety of specific and non-specific bioassays [72][73][74][75][76]. In the case of non-specific bioassays, as is the case with SET, the elucidation of key toxic compounds can be hindered due to the alteration in many different pathways of the organism in a non-targeted way [72,76]. Moreover, the existence of several toxicity mechanisms/pathways leads to the consideration of possible interactions between the compounds present in the sample (e.g., synergism and/or antagonism) that could contribute considerably to the observed toxicity. For instance, Neale et al. [72] used the non-specific FET test to cover a wider range of endpoints to determine the effects on the human-impacted Danube River (Serbia), but the identified compounds could explain less than 0.2% of the toxicological effects on zebrafish. However, to a lesser extent, examples of unexplained toxicity can also be found in the literature, even when using specific bioassays. When using specific bioassays, the observed effects can be associated with specific modes of action (e.g., binding estrogen or androgen nuclear receptor or photosystem II), and the active compounds should share structural similarities. However, some examples in the literature pointed out the potential presence of nonidentified active chemicals in different water ecosystems (e.g., wastewater effluent or surface waters) [72][73][74], even if, for example, activation of the NFκB mediated inflammatory pathway was observed. In the same way, Kienle et al. [75] also reported a substantial difference between the biological effects and chemical results for the acetylcholinesterase inhibition bioassay on WWTP effluent samples, which could be related to the content of the dissolved organic carbon of the sample.

Reagents and Materials
All standard chemicals and materials used in this work are described in Lopez-Herguedas et al. [41].

Sampling
The sampling site and sampling method are thoroughly described elsewhere [41] and in the supplementary material (Section S2.1). Briefly, 45 L of a 24 h composite water sample was collected from the main discharging sewage drain of a hospital (holding > 800 beds and offering > 13 specialties) (Biscay, Basque Country, Spain) using an on-site large-volume solid phase extraction (LV-SPE (MAXX Mess-u. Probenahmetechnik GmbH, Rangendingen, Germany). For water sample analysis, SPE cartridges were eluted following the procedure proposed by Välitalio et al. [77] (see Supplementary Materials, Section S2.1). In parallel, and for quality control assurance, 5 L of UHPLC-MS grade water mineralized with analytical grade sodium chloride (0.1%, Merck, Darmstadt, Germany) was pumped 9 times through an LV-SPE system in order to have a blank sample (LV-B) The hospital effluent sample and blank sample were enriched 250 times (EF 250).

Effect-Directed Analysis (EDA)
EDA was used as a prioritization tool before chemical identification. To that aim, the preconcentrated hospital and blank samples were submitted to the EDA protocol described in a previous work of the research group [41] (see details in Figure 4 and Section S2 in Supplementary Materials). Briefly, the preconcentrated hospital effluent water sample (raw) and blank sample (LV-B) were fractionated by a 2-step procedure. The first fractionation of raw and LV-B was conducted using a C 18 semi-preparative column and, the toxic fractions were fractionated again using an aminopropyl semi-preparative column. A replicate of the raw sample was fractionated again and the fractions obtained in each fractionation were recombined (RS and ΣF-RS for the first and second fractionation, respectively) to assure that no analyte or bioactivity losses occurred during fractionation. The bioactivity of the raw sample (raw), recombined samples (RS and ΣF-RS), individual fractions as well as procedural (LV-B) and recombined blanks (LV-B-R) was tested using the SET bioassay (see Section 3.3.1.). In order to test the whole bioactivity in a unique fraction, the active fractions found in the first fractionation were combined (ΣF). Chemical analysis was limited to biologically active and neighboring fractions, the RS and ΣF-RS, and raw and blank samples (see Section 3.3.2).

Sea Urchin Embryo Test (SET)
Adult sea urchins (Paracentrotus lividus) were collected from an intertidal area of Armintza (43.43347 N, 2.89889 W, Basque Country) and maintained in an aquarium at the Research Centre for Experimental Marine Biology and Biotechnology, Plentzia Marine Station (PIE, Basque Country). Seawater tanks were maintained at 15 ± 1 • C with a natural photoperiod (12:12 h) and a continuous aeration and filtration water pumping system. Every two days, sea urchins were fed with macroalgae gathered in their sampling area and the dregs were siphoned. SET bioassays were performed following the procedure described by Saco-Álvarez et al. [78] detailed in the Supplementary Materials (see Section S2.3). Briefly, raw sample extracts and SPE eluted extracts were evaporated up to dryness under a gentle stream of N 2 at 40 • C on a turbovap system and re-dissolved using filtered seawater (FSW, 0.2 µm) containing 0.1% of DMSO (v/v). Three mL of the diluted samples (n = 3) were placed in glass vials (20 mL) with the fertilized eggs at a 40 eggs/mL concentration. Likewise, FSW (CFSW) and FSW with 0.1% of DMSO (v/v) (CDMSO) were also included in the test batch as solvent controls, and a copper (II) chloride solution (0.01-5 mg/L) was used as a positive quality control sample. Fertilized eggs (egg size at t = 0) and samples after the incubation stage (20 • C, 48 h in darkness) were fixed with one drop of 40% formalin.
Molecules 2023, 28, x FOR PEER REVIEW 10 of 17 fractionated again using an aminopropyl semi-preparative column. A replicate of the raw sample was fractionated again and the fractions obtained in each fractionation were recombined (RS and ΣF-RS for the first and second fractionation, respectively) to assure that no analyte or bioactivity losses occurred during fractionation. The bioactivity of the raw sample (raw), recombined samples (RS and ΣF-RS), individual fractions as well as procedural (LV-B) and recombined blanks (LV-B-R) was tested using the SET bioassay (see Section 3.3.1.). In order to test the whole bioactivity in a unique fraction, the active fractions found in the first fractionation were combined (ΣF). Chemical analysis was limited to biologically active and neighboring fractions, the RS and ΣF-RS, and raw and blank samples (see Section 3.3.2).

Sea Urchin Embryo Test (SET)
Adult sea urchins (Paracentrotus lividus) were collected from an intertidal area of Armintza (43.43347 N, 2.89889 W, Basque Country) and maintained in an aquarium at the Research Centre for Experimental Marine Biology and Biotechnology, Plentzia Marine Station (PIE, Basque Country). Seawater tanks were maintained at 15 ± 1 °C with a natural photoperiod (12:12 h) and a continuous aeration and filtration water pumping system. Every two days, sea urchins were fed with macroalgae gathered in their sampling area and the dregs were siphoned. SET bioassays were performed following the procedure described by Saco-Álvarez et al. [78] detailed in the SI (see Section S2.3). Briefly, raw sample extracts and SPE eluted extracts were evaporated up to dryness under a gentle stream of N2 at 40 °C on a turbovap system and re-dissolved using filtered seawater (FSW, 0.2 μm) containing 0.1% of DMSO (v/v). Three mL of the diluted samples (n = 3) were placed in glass vials (20 mL) with the fertilized eggs at a 40 eggs/mL concentration. Likewise, FSW (CFSW) and FSW with 0.1% of DMSO (v/v) (CDMSO) were also included in the test batch as solvent controls, and a copper (II) chloride solution (0.01-5 mg/L) was used as a positive quality control sample. Fertilized eggs (egg size at t = 0) and samples after the incubation stage (20 °C, 48 h in darkness) were fixed with one drop of 40% formalin.
Toxicity assessment was carried out by measuring two different sublethal points: (1) the index of toxicity (IT), where individual embryos were categorized for their malformation level according to Carballeira et al. [79], and (2) the size increase (SI) in the larvae recorded according to Saco-Álvarez et al. [78] (detailed in SI, Section S2.3, Figure S1). Toxicity assessment was carried out by measuring two different sublethal points: (1) the index of toxicity (IT), where individual embryos were categorized for their malformation level according to Carballeira et al. [79], and (2) the size increase (SI) in the larvae recorded according to Saco-Álvarez et al. [78] (detailed in Supplementary Materials, Section S2.3, Figure S1). Measurements were performed via an inverted microscope (Nikon Eclipse Ti-S) coupled to an electronic camera using NIS-Elements image analysis software v4.30 (Nikon Instruments BV, Europe). The hospital effluent (raw sample) was tested in dilution series (REF 0.001-0.01-0.05-0.1-0.5-1-5-10) to build the dose-response curve from which to derive the concentration value that causes effects in at least half of the population (effective concentration, EC 50 ).
Data treatment and statistical analyses were processed with the SPSS Statistics 23 package (v17, IBM SPSS), using data corrected by the control response. As the data set was not normally distributed (Shapiro-Wilk test), the non-parametric Kruskal-Wallis test was performed to compare treatments. The PROBIT model was used to determine the EC i values with 95% confidence limits.

Chemical Analysis
Samples and fractions exhibiting significant toxicity in the SET bioassay were analyzed using LC-HRMS and GC-MS analyses and the methodologies described elsewhere [4,80] (see Supplementary Materials, Section S2.4). Blank samples (LV-B, LV-B-R) were also analyzed simultaneously, to avoid false positive results.
Concretely, the LC-HRMS measurements were run in a Thermo Scientific Dionex UltiMate 3000 UHPLC coupled to a Thermo Scientific™ Q Exactive™ Focus quadrupole-Orbitrap mass spectrometer (UHPLC-q-Orbitrap) equipped with a heated ESI source (HESI, Thermo-Fisher Scientific, CA, USA). The LC-qOrbitrap measurements were acquired in negative and positive ionization in the Full scan-data-dependent MS2 (Full MS-ddMS2) discovery acquisition mode in the m/z 70-1050 Da, which allowed us to identify compounds through suspect screening analysis as well as multi-target analysis. Suspect analysis was performed employing Compound Discoverer 3.1 software (CD, Thermo-Fischer Scientific) following the workflow described by González-Gaya et al. [41], with some modifications (detailed in Supplementary Materials, Section S2.4.1). Regarding the multi-targeted analysis of the 187 target compounds available at the laboratory (see details in Table S1 in Supplementary Materials), the compounds were identified and quantified using TraceFinder 4.1 as detailed in the Supplementary Materials (see Section S2.4). Instrumental characteristics including ionization mode, retention time (R t ), and instrumental limits of quantification (LOQ instrumental ) and identification (LOI instrumental ) of the 187 target compounds are detailed in Table S1.
Regarding GC-MS analyses, an Agilent 6890 gas chromatograph coupled to an Agilent 5975 mass spectrometer system (Agilent Technologies, Palo Alto, CA, USA) (see Supplementary Materials, Section S2.4.2) was used. The measurements in SCAN mode allowed the identification of unknown non-polar compounds from the comparison of experimental mass spectra with the ones available in the NIST library, and the identified compounds were afterwards confirmed using standards.

Potential Toxicity Assessment of Detected Compounds
The potential toxicity of compounds identified in the toxic fractions of the hospital effluents was evaluated through the estimation of toxic units (TU). TU values were calculated as described in a previously published EDA study [30]. Concretely, chemically based TU values (TU chem ) were calculated from the sum of the concentration of compounds identified in the toxic fractions (C i ) normalized to individual 50% effect concentrations (EC 50(i) ) (see Equation (1)), for comparison with biologically derived TU (TU bio ) (Equation (2)). Additionally, an artificial mixture containing the compounds identified in the toxic fractions was prepared by fortifying filtered seawater at the concentration levels at which contaminants were quantified in the hospital effluent (concentration range of 10 −3 -10 8 ng/L). Artificial TU (TU artificial mixture ) was determined using Equation (3).
For the TU determination, SETApp [40] was used to automatically measure the two endpoints of the SET (i.e., IT and SI). Briefly, larvae were transferred to a 24-well microplate at 48 h post-fertilization, and an image set of each well was obtained using the automatic image reader Cytation 5 (BIOTEK). Images were loaded in the SETApp to automatically detect, isolate and measure all larvae, and determine the malformation levels according to the work of Carballeira et al. [79].

Conclusions
The EDA approach enabled the identification of potentially toxic chemicals among all the hundreds of chemicals that are present in a complex water sample, as is the case with hospital effluent. Concretely, the fractionation of the sample and toxicity testing using SET bioassays enabled the identification of three potential toxic fractions among the 18 fractions into which the raw sample was firstly divided. In this way, EDA was used in this work as a prioritization strategy in order to identify as many as possible chemicals that may be responsible for the raw sample toxicity. Suspect screening of the toxic fractions using LC-qOrbitrap and GC-MS allowed the identification of 25 potentially toxic unknown compounds including different pharmaceuticals (bisoprolol, clozapine, cyclophosphamide, diphenhydramine, eprosartan, fluconazole, furosemide, ofloxacin, omeprazole, pentoxifylline, propranolol, primidone, sulfamethoxazole, verapamil), pesticides (acetamiprid, EDDP, fenpropidin, imazalil, isoproturon, spiroxamine), industrial compounds (2,4-DBT, OBT, DEP, TEP) and a preservative (methylparaben), among other tentatively identified substances. However, sample toxicity was neither attributed to any particular compound as a predominant driver nor confirmed as the result of a combined effect of the cocktail of chemicals present in the hospital effluent.
The discrepancies observed between analytical and bioanalytical contamination suggest that even though efforts have been made to identify unknown toxic compounds, there are more unidentified compounds at concentration levels below the limits of detection and/or not detectable by the chemical analyses used in this work. In this way, these results support the need to improve the detection limits of analytical techniques to facilitate the identification of undetected/unknown contaminants. On the other hand, the lack of ecotoxicological data on detected contaminants for sea urchins should also be highlighted. As a model organism for assessing the ecological status of coastal ecosystems, it would be necessary to broaden the knowledge of the effects in sea urchins of these pollutants, considering their ubiquitous reported presence in the environment. Furthermore, the observations found in this work highlight the importance of further investigating the different toxicity pathways as it could help us understand how contaminants interfere in different non-specific bioassays as well as lead to the presence of mixture effects.
Supplementary Materials: The following supporting information can be downloaded at: https:// www.mdpi.com/article/10.3390/molecules28031212/s1, Figure S1: Examples of how to measure the maximum dimension according to Saco-Álvarez [43] in P. lividus at fertilized eggs (A) and completely developed larvae at 4 arm plateaus stage (B); Figure S2: Toxicity of the tested procedural blanks (LV-Blank and LV-R-Blank) and the difference with the Raw and RS samples considering only the SI (%) endpoint. A comparison with solvent controls (CFSW and CDMSO) is also included. Green coloured bars refer to non-toxic samples/concentrations, while red bars refer to toxic samples/concentrations. Error bars correspond to the standard deviation (SD) of the three replicates; Figure S3 Figure S8. Results of sea urchin P. lividus larvae exposed to different concentrations of the artificial mixture containing toxicity driver candidates using the SETApp  Table S1: The electrospray ionization (ESI) mode, m/z, retention times (tR) and instrumental limits of identification (LOI instrumental, ng/mL) and quantification (LOQ instrumental, ng/mL) for the target compounds classified according its application, CAS number, molecular formula, InChIKey and the mass analyser used in their analysis; Table S2: Formula, candidates, application, molecular weight (MW) and parameters to determine levels of identification of the compounds classified according to Schymanski criteria [49] with a mzCloud best match >70% in both ionization modes. Features detected in the active fractions (F6, F7 and F8) are also indicated; Table S3: Acute toxicity data of the potential toxic candidates for fish and invertebrates. Compounds structure is also included. Reference [81] was cited in the Supplementary Materials. Funding: Authors acknowledge financial support from the Agencia Estatal de Investigación (AEI) of Spain and the European Regional Development Fund through CTM2017-84763-C3-1-R and CTM2020-11686RB-C31 projects and the Basque Government through the financial support as a consolidated group of the Basque Research System (IT1446-22). The authors are grateful to the Consorcio de Aguas de Bilbao and especially to Iñigo González. Naroa Lopez-Herguedas is grateful to the Spanish Ministry of Economy, Industry and Competitivity for her predoctoral scholarship FPI 2018 (PRE2018-086493).

Informed Consent Statement: Not applicable.
Data Availability Statement: Data will be made available on request.