Multi-region assessment of pharmaceutical exposures and predicted effects in USA wadeable urban-gradient streams

Human-use pharmaceuticals in urban streams link aquatic-ecosystem health to human health. Pharmaceutical mixtures have been widely reported in larger streams due to historical emphasis on wastewater-treatment plant (WWTP) sources, with limited investigation of pharmaceutical exposures and potential effects in smaller headwater streams. In 2014–2017, the United States Geological Survey measured 111 pharmaceutical compounds in 308 headwater streams (261 urban-gradient sites sampled 3–5 times, 47 putative low-impact sites sampled once) in 4 regions across the US. Simultaneous exposures to multiple pharmaceutical compounds (pharmaceutical mixtures) were observed in 91% of streams (248 urban-gradient, 32 low-impact), with 88 analytes detected across all sites and cumulative maximum concentrations up to 36,142 ng/L per site. Cumulative detections and concentrations correlated to urban land use and presence/absence of permitted WWTP discharges, but pharmaceutical mixtures also were common in the 75% of sampled streams without WWTP. Cumulative exposure-activity ratios (EAR) indicated widespread transient exposures with high probability of molecular effects to vertebrates. Considering the potential individual and interactive effects of the detected pharmaceuticals and the recognized analytical underestimation of the pharmaceutical-contaminant (unassessed parent compounds, metabolites, degradates) space, these results demonstrate a nation-wide environmental concern and the need for watershed-scale mitigation of in-stream pharmaceutical contamination.

Herein, we expand the assessment of stream-ecosystem pharmaceutical risk (exposure and hazard) [40-42] to four US regions (including SESQA) sampled during 2014-2017 to test the hypothesis that pharmaceutical contamination is common across the US in urban-gradient headwater streams, including in those with no NPDES-permitted WWTP discharges. The potential for cumulative-contaminant effects (hazard) to in-stream biota was assessed based on 1) occurrence and cumulative concentrations of pharmaceutical mixtures, and 2) cumulative Exposure Activity Ratios (∑ EAR ) [43-45] based on high-throughput screening data in Toxicity Forecaster [ToxCast™,46], as described in [44].

Site description and analytical method
Filtered water samples (10 mL) were collected by the USGS NAWQA Regional Stream Quality Assessments (RSQA) from perennial, wadeable (less than 10 m width and 1 m depth at baseflow) headwater stream sites in watersheds with varying degrees of urban and agricultural land use as part of four regional assessments ( (Fig 1, S1 Table). During each regional water-quality assessment period (spring to summer), 3-5 water samples were collected from sites representing region-specific gradients in urban and agricultural development and a limited number of single samples were collected from nominal low-impact (low-development) watershed sites. Site selection and sampling methodologies are as described [38,48,50]. Samples were syringe filtered (0.7 μm pore size glass-fiber) into baked (500 C) amber glass vials and shipped on ice for analysis at the USGS National Water Quality Laboratory (NWQL) in Denver CO. Direct aqueous injection (100 μL), isotope dilution, high performance liquid chromatography tandem mass spectrometry (HPLC-MS/MS) was used to quantify an environmentally-relevant, representative subset consisting of 111 human-use pharmaceutical and  ; among these, gabapentin, guanylurea, and hexamethylenetetramine were added to the method in 2017 and analyzed only in CaSQA samples. Typically interpreted as a pesticide environmental contaminant, piperonyl butoxide also is medically indicated for treatment of lice [53, 54] and, thus, retained herein. Two additional analytes (atrazine, herbicide; methyl-1H-benzotriazole, solvent/deicing agent) included in the NWQL pharmaceutical method are not recognized pharmaceutical agents and are not included herein. Analytes, with Chemical Abstracts Services numbers and laboratory reporting limits (RL), are listed in the supporting information (S2 Table).

Quality assurance quality control (QAQC)
HPLC-MS/MS pharmaceutical analysis included addition of 19 surrogate standards (400 ng/L nominal final concentration) to field-filtered samples to evaluate whole-method recovery (median = 102%, interquartile range [IQR] = 95-110%, range 2-320%). A total of 44 field blanks were prepared across all regions by processing pesticide-grade blank water through all field collection equipment and analyzing as above [52] for environmental samples. Only 9 of the 88 pharmaceutical analytes detected at least once in environmental samples in this study were also detected in field blanks. Among these, 6 (metformin, methadone, metoprolol, nevirapine, norethindrone, and omeprazole/esomeprazole) were only detected in a single field blank; corresponding environmental data were not blank adjusted, but interpretation of results below blank detection levels warrants caution. Lidocaine, nicotine, and caffeine were detected in 5 (11%), 5 (11%), and 2 (5%) field blanks, respectively; environmental concentrations less than corresponding 90 th percentile field blank concentrations were blank corrected to nondetect (nd).

Data handling, statistics, and ∑ EAR analysis
The reporting limits for pharmaceutical analytes were determined using DQCALC 2 software, a spreadsheet-based tool for graphically modeling relative standard deviation versus concentration and assigning a data precision statement in water analytical methods [56](RLDQC; S2 Table). Laboratory-estimated water concentrations below the RLDQC (positive detections with reduced quantitative certainty) were used as is (S3A-S5 Tables). Pharmaceutical results were aggregated to estimate maximum-and median-concentration exposures relative to the 108 pharmaceutical analytes (111 for CaSQA). S3A and S3B Table include the sample and detection counts, respectively, by compound and site for all pharmaceuticals detected at least once in the study. S4 Table includes the maximum concentrations of each pharmaceutical detected in this study by compound and site. S5 Table contains median concentrations (all samples) by compound and site and only includes those pharmaceuticals that were detected in at least half of the samples at one or more sites. S6 Table contains land-use/land-cover (LULC), specific conductivity, and major ion (maximum and median concentrations of Ca, Mg, Na, Cl, SO 4 , K) data.
No permits were required for this work. All data are also available from the USGS National Water Information System (NWIS) [74] and from USGS ScienceBase [75].

Pharmaceutical mixtures were common in all regions
Among the 308 wadeable streams sampled across all regions during the 2014-2017 RSQA studies, multiple pharmaceuticals (pharmaceutical mixtures) were detected at least once (maximum exposure dataset) in 95% (248) of the 261 multiple-sample, urban-gradient sites and in 68% (32) of the 47 single-sample, non-urban, presumptive low-impact, sites (Fig 1, S3B and S4 Tables). Importantly, pharmaceutical mixtures were detected in at least half of the samples (median exposure dataset) from 81% (212) of the multiple-sample, urban-gradient sites (S1 Fig, S5 Table). Cumulative (sum of detected pharmaceuticals) maximum and median detections ranged 0-60 per site (median: 4; IQR: 2-8) and 0-43 per site (median: 2; IQR: 1-4), respectively, with the centroid (mean) of cumulative pharmaceutical detections per site estimated (PERMA-NOVA) to be greater (permutation N = 9999 probability of being the same = 0.0001) in eastern (NESQA, SESQA) than in western (PNSQA, CaSQA) study streams (Fig 2A and 2B, S4 and S5 Tables). Site-specific cumulative maximum and median concentrations ranged nd-36,142 ng/L and numbers of pharmaceuticals detected in water samples from wadeable streams in each region (total sites in parentheses) as part of the USGS Regional Stream Quality Assessment (RSQA). Circles indicate maximum (left) or median (right) data for sites sampled multiple times (n = 3-5). Open triangles indicate data from single-sample sites and do not differ between maximum and median plots. Boxes, centerlines, and whiskers indicate interquartile range, median, and 5 th and 95 th percentiles, respectively, for multiple-sample sites (circles), only. For each plot, p is the permuted probability that the centroids and dispersions across all groups are the same and different letters indicate groups with pairwise probabilities that are less than 0.05 (PERMANOVA) for multi-sample sites only. Numbers above X-axes (bottom plots) indicate numbers of sites in each region with no detections (nd) under maximum and median exposure conditions. https://doi.org/10.1371/journal.pone.0228214.g002 Pharmaceuticals in USA urban-gradient streams (median: 88 ng/L; IQR: 29-306) and nd-8,756 ng/L (median: 19 ng/L; IQR: 3-75 ng/L), respectively ( Fig 2C and 2D, S4 and S5 Tables). The cumulative maximum concentrations were not clearly different (PERMANOVA; probability of being the same, p = 0.0812) between regions. There is some evidence to suggest the centroid of cumulative median concentrations was greater (permutation N = 9999 probability of being the same = 0.0154) in the eastern region (NESQA, SESQA) streams than in PNSQA streams, with CaSQA streams intermediate, but extensive overlap of the data distributions was apparent in all cases. The apparently greater detections and, possibly, median concentrations of pharmaceuticals between eastern and western regions are consistent with reported differences in cumulative 2016 prescription drug sales (87 vs 50 million US dollars [76]) and populations (79 vs 50 million people [77]) between corresponding eastern and western region states, respectively.
Likewise, the spatial detection frequencies (the percentage of study sites at which analytes were detected) of several individual pharmaceuticals and pharmaceutical groups differed markedly between regions (Fig 3 and S2 Fig). For example, the antidepressants, venlafaxine and desvenlafaxine, were detected at a substantially higher (approximately four times higher) percentage of eastern region sites than western region sites. The percentages of eastern region study sites, at which pharmaceuticals associated with seasonal and perennial allergies (fexofenadine, pseudoephedrine/ephedrine, diphenhydramine, loratadine) were detected, were generally double those observed in the two western study regions (Fig 3), despite similar reported percentages of hay fever and respiratory allergies for children [78] and adults [79] in the eastern and western US in 2018. Correlations between co-detected analytes are presented in S8 Table. Land-use/land-cover and major ion predictors of pharmaceuticals Frequent occurrence, multiple detections per site (median: 4 compounds per site across all sites including in pre-selected low-impact watersheds), and elevated cumulative concentrations (up to >36 μg/L per site) emphasize the need for identification, monitoring, and mitigation of pharmaceutical sources in wadeable-stream ecosystems. ANOSIM analysis of sitespecific pharmaceutical and LULC data matrices indicated marginal (Global R range: 0.043-0.239) differences between RSQA study unit, urban-center, and ecoregion groupings (S7 Table). Likewise, weak correlations (RELATE; ρ range: 0.029-0.116) were found between median pharmaceutical similarity matrices (detection/concentrations under median conditions) and watershed LULC similarity matrices, with multiple individual urban development metrics (e.g., percentage of developed industrial/military [Dev_IndusMilitary2012] and semi-  Table).
The pharmaceutical contaminants observed at the 75% of RSQA sites without NPDES-permitted discharges, however, confirm previous conclusions that WWTP outfalls are not the only important pathways of pharmaceutical contaminants to urban/suburban streams [1,7,88,89]. Other potential urban-gradient sources of pharmaceuticals to streams include aging sewer infrastructure [90,91], combined (sanitary/stormwater) sewer overflows [92][93][94][95][96], private septic and on-site waste-handling systems [97][98][99], gray-water systems [100][101][102], green space and golf course wastewater reuse [103], and animal waste runoff [104][105][106]. Notably, a recent national reconnaissance demonstrated that untreated stormwater can be an important episodic source of mixed pharmaceuticals to surface waters, at levels comparable to and often exceeding those in treated WWTP effluent [107]. Thus, these results reiterate the need for whole-watershed, contaminant-mitigation approaches, including improved pharmaceutical disposal practices, wastewater treatment and transfer systems, and stormwater controls.
More, the results argue for research and implementation of new high-frequency or continuous sensor technologies for direct or indirect monitoring of pharmaceutical contaminants in next generation water observing systems [e.g., 108,109] in urban settings. Surface-water ions (and related conductivity measures) have been suggested as potentially useful surrogates for indirect monitoring of pharmaceutical contamination in streams [e.g., 110,111], because physiological ions (electrolytes) and pharmaceuticals are both primarily excreted in urine [112][113][114][115] and are frequently reported together in wastewater-impacted streams at locally-elevated concentrations [e.g., 92,93]. The most useful monitoring approaches are expected to be fixed-station or single watershed applications [111], due to the potential confounding effects and 95 th percentiles, respectively. Numbers to the right of each plot indicate the percentage of sites within the region at which the compound was detected at least once. Gabapentin, guanylurea, and hexamethylenetetramine were analyzed only in CaSQA samples. of site-to-site variability in non-wastewater sources of ions in urban-gradient streams [116,117], including geologic minerals [118], fertilizer runoff [104], road salt [119,120], and concrete infrastructure [121][122][123]. Nevertheless, the current multi-region dataset provides a unique opportunity to test the broad geospatial validity of the approach by assessing the probability (p) of no correlation (H 0 : rho (ρ) = 0; permutation N = 9999; Spearman Rank-Order Correlation) between in-stream pharmaceutical and ion concentrations (S9 Table).
The results support the potential utility of surface-water ions as surrogates for wastewaterassociated contaminants like pharmaceuticals [110,111]. Although no relation (p � 0.2528) was observed between specific conductivity and cumulative pharmaceutical metrics under median exposure conditions, correlations were observed between pharmaceuticals and sodium or chloride (p � 0.0182), with the strongest correlations observed for potassium (p � 0.0001). Comparison of the Spearman correlation coefficients provided additional insight into the importance of the yellow-water/wastewater pathway relative to potential confounders (i.e., non-wastewater sources [104,[116][117][118][119][120][121][122][123]). Correlations between median pharmaceutical metrics and median concentrations of sodium and chloride were weak (ρ � 0.202), consistent with numerous confounding non-wastewater sources of these ions in urban settings [116,119,120]. However, promising correlations (ρ range: 0.314-0.328) were observed between median concentrations of potassium and cumulative median detections and concentrations of pharmaceuticals, indicating the potential for potassium as an indicator of in-stream pharmaceutical contamination in fixed-place or single watershed applications and consistent with the strong correlation (R 2 > 0.89) reported between instream concentrations of potassium and pharmaceuticals in the Leine River watershed in Germany [111]. The stronger broad regional correlation in this study between potassium and pharmaceuticals may reflect comparatively less variability in non-wastewater potassium sources as well as the usage of potassium salts (e.g., ferrate, ferrocyanate) as floculants/coagulants in wastewater treatment [124]. Emerging sensor technologies that hold promise for next generation monitoring of potassium (and other ions) include recently described nanorod-based potassium ion sensors [125] and multi-parameter potentiometric microanalyzers (lab-on-a-chip platforms) developed for space travel [126][127][128] and environmental water quality monitoring [129,130].

Potential for mixed-pharmaceutical biological effects
The 111-pharmaceutical analytical space assessed in this study is a fractional indicator of the presumptive pharmaceutical-contaminant universe, with more than 4000 active ingredients (parent compounds) [2,131] in current use and an unknown chemical-space [132] of metabolites and environmental degradates [13]. Given the breadth of species, life stages, biomasses, and concomitant vulnerabilities present in urban-gradient aquatic food webs [133][134][135][136] and the designed bioactivity of commercial pharmaceuticals [1-6, 14-17, 19-24, 137], their detection in RSQA headwater streams is prima facie evidence of the potential for molecular toxicity and sub-lethal effects in non-target, organisms in urban-gradient headwater streams across the US [138][139][140][141].
The in vitro ToxCast-based EAR approach provides an additional line of evidence for sublethal effects at a reported concentration [43,73], supports estimation of cumulative effects (∑ EAR ) of mixed-contaminant exposures using the CA-model methodology [63][64][65][66][67][68], and predicts probable effects consistent with traditional in vivo water-quality benchmark-based toxicity quotient (TQ) approaches (EAR = 0.001 comparable to commonly-employed TQ = 0.1 effects threshold) [45]. ToxCast [46] includes exposure-response metrics for 9000+ organic chemicals and approximately 1000 standardized, predominantly-vertebrate, molecular bioassay endpoints (e.g., nuclear receptor, DNA binding) [70, 142,143]. ToxCast EAR results for estimated maximum and median pharmaceutical exposure conditions are summarized in Fig  4 and S11-S14 Tables. Given the diversity of organisms and concomitant range of vulnerabilities in surface-water food webs [133][134][135][136], we employed the recently suggested effects-screening threshold of 0.001 [45], as described [44]. Of the 88 pharmaceuticals detected at least once in this study, only 43% (38) had acceptable ToxCast data at the time of access. Under maximum exposure conditions, 63% (194) of study sites had one or more compounds with individual EAR greater than the 0.001 effects-screening threshold, 65% (201) had cumulative EAR (SEAR max ) � 0.001, and 3 sites had SEAR max � 1 (Fig 4). These results indicate transient exposures with a probability of vertebrate molecular effects were common in urban-gradient headwater streams across the US. Of the 62 pharmaceuticals in the estimated median exposure dataset, 61% (38) had exposure-effects data in ToxCast. Approximately 25% of the study sites had individual or cumulative EAR med � 0.001 under the estimated median exposure conditions (Fig 4), indicating that sites with persistent exposures with a probability of molecular effects were common in urban headwater streams across the US. Zebra fish (ZF; Danio rario) embryo metrics in ToxCast inform organism-level as well as vulnerable, early-life-cycle effects in fish [144,145]. Thus, the results also indicate the potential for pharmaceutical effects to fish at the organism level in at least some headwater streams, because 7 and 5 sites, respectively, had SEAR max ZF and SEAR med ZF � 0.001 across all ZF endpoints.

Implications for stream ecosystem health and remediation
The results indicate substantial pharmaceutical-contaminant concerns in wadeable, urbangradient, headwater streams not only in SESQA [7], but in other regions across the US [1, 4, 6, 19], irrespective of WWTP discharge. Crucially, the pharmaceutical-analyte space [52] assessed herein is an order(s) of magnitude underestimate of the presumptive pharmaceuticalcontaminant universe, with 4000+ parent compounds in current use [2] and unquantified numbers of environmental metabolites/degradates [13,132]. Considering only those pharmaceuticals assessed in this study, individual concentrations up to μg/L levels and multiple detections per site (median = 4 across all sites including pre-selected, low-impact watersheds) at cumulative concentrations ranging more than 36 μg/L are notable concerns, given documented adverse impacts of individual pharmaceuticals at low ng/L concentrations [20] and the widespread cumulative exceedances of in vitro molecular effects thresholds (∑ EAR ) observed in these urban-impacted headwater streams.
Metformin was the second most frequently detected pharmaceutical across all sites in this study and the fourth most prescribed pharmaceutical in the US, with an estimated 81 million  [153,154]. US and global metformin usage is expected to increase, as a first-line diabetes therapy [155][156][157] and treatment candidate for polycystic ovarian syndrome [158] and cancers [159]. Metformin is excreted essentially unchanged in human urine [160], poorly removed by wastewater treatment technologies [161], considered environmentally recalcitrant [161,162], and increasingly reported in environmental samples [7,9,163]. Environmentally-relevant [161,164,165] metformin exposures in the μg/L range have recently been shown to induce biological responses in fish [166][167][168][169], including up-regulation of vitellogenin mRNA [170,171] and other gene targets [169,171,172], male intersex in fathead minnow (Pimephales) [170], and behavioral modifications in Siamese fighting fish (Betta splendens) [167]. Guanylurea, metformin's only currently recognized persistent environmental degradate, is often observed in surface waters at higher concentrations than metformin [155,[163][164][165] and, importantly, has been recently reported to cause growth effects in Japanese medaka (Oryzias latipes) similar to metformin but at low (<10) ng/L concentrations [166].
Fish and fish-embryos are widely-used animal models in the pharmaceutical development pipeline [173,174], including for anti-diabetics [175,176]; from this perspective, fish are arguably pharmaceutical target organisms with unintended environmental exposures. Consistent with this use, individual and simple mixtures of pharmaceuticals have been shown to cause unintended effects to the health and behavior of laboratory and wild fish at environmentallyrelevant concentrations [24, [177][178][179] and the potential biological impacts of characteristically complex environmental pharmaceutical cocktails are global concerns [64, 68,180,181]. In light of the documented potential for pharmaceutical bioconcentration [182][183][184] and trophic transfer [182] within aquatic food webs and for trophic transfer of pharmaceuticals from aquatic to riparian food webs [185], measured water concentrations may substantially underestimate the ecological exposures and effects from in-stream pharmaceutical contaminants [182,185]. Thus, considering potential individual and interactive effects of the 88 pharmaceuticals detected in headwater streams herein and the recognized orders-of-magnitude analytical underestimation of the presumptive pharmaceutical-contaminant (parent compounds, metabolites, degradates) space, the results of the present study demonstrate a nation-wide need for watershed-scale pharmaceutical-contaminant mitigation approaches that extend the current emphasis on WWTP-effluent sources to include more broadly distributed inputs such as septic systems, leaking wastewater transfer systems, and urban stormwater runoff.  Table. Watershed-specific GIS metrics (see S15 Table for GIS data dictionary) Table. Summary statistics for multivariate relations between pharmaceutical cumulative detection/concentration (log-transformed and normalized) data matrices (Euclidean distance) and site-specific LULC or major ion data matrices (Euclidean distance) assessed by non-metric multi-dimensional scaling (NMDS), one-way analysis of similarity (ANO-SIM), and permutation-based (permutations = 999) cophenetic correlation (RELATE) routines.

S9 Table. Spearman rho (r) rank-order correlation coefficients (r; lower triangle) and 2-tail probability (permutation N = 9999) that no correlation exists (upper triangle) for site-specific cumulative median pharmaceutical detections or concentrations (nanograms per liter, ng/L), major ions, and select GIS metrics identified by multi-variate RELATE and BEST analyses.
(XLSX) S10 Table. Compound:Endpoint combinations excluded from ToxCast evaluation due to unreliable concentration-response relationship and resulting lack of confidence in activity concentration at cutoff (ACC). (XLSX)

S11 Table. Site-specific Exposure Activity Ratios (EAR) under Maximum exposure conditions for those compounds with exact Chemical Abstract Service (CAS) number matches and with reliable concentration-response relationship and ACC data in ToxCast.
(XLSX) S12 Table. Site-specific Exposure Activity Ratios (EAR) under Maximum exposure conditions for all bioassay endpoints within each class shown. Data are for those compounds with exact Chemical Abstract Service (CAS) number matches and with reliable concentrationresponse relationship and ACC data in ToxCast. (XLSX) S13 Table. Site-specific Exposure Activity Ratios (EAR) under Median exposure conditions for those compounds with exact Chemical Abstract Service (CAS) number matches and with reliable concentration-response relationship and ACC data in ToxCast. (XLSX) S14 Table. Site-specific Exposure Activity Ratios (EAR) under Median exposure conditions for all bioassay endpoints within each class shown. Data are for those compounds with exact Chemical Abstract Service (CAS) number matches and with reliable concentration-response relationship and ACC data in ToxCast. (XLSX) S15 Table. Data dictionary describing metrics in S9 Table. Additional citation details provided in S16 Table. (XLSX) S16