Derivation of water quality guidelines for priority pharmaceuticals

Abstract

 

Pharmaceuticals can enter freshwater and affect aquatic ecosystem health. Although toxicity tests have been carried out for the commonly used pharmaceuticals, evidence‐based water quality guidelines have not been derived. High‐reliability water quality guideline values have been derived for 4 pharmaceuticals—carbamazepine, diclofenac, fluoxetine, and propranolol—in freshwaters using a Burr type III distribution applied to species sensitivity distributions of chronic toxicity data. Data were quality‐assured and had to meet acceptability criteria for “chronic” no‐observed‐effect concentrations or concentrations affecting 10% of species, endpoints of population relevance (namely, effect endpoints based on development, growth, reproduction, and survival). Biomarker response data (e.g., biochemical, histological, or molecular responses) were excluded from the derivation because they are typically not directly relevant to wildlife population‐related impacts. The derived guideline values for 95% species protection were 9.2 μg/L, 770 μg/L, 1.6 μg/L, and 14 μg/L for carbamazepine, diclofenac, fluoxetine, and propranolol, respectively. These values are significantly higher than the unknown reliability values derived for the European Commission, Switzerland, or Germany that are based on the application of assessment factors to the most sensitive experimental endpoint (which may include biochemical, histological, or molecular biomarker responses) of a limited data set. The guideline values derived in the present study were not exceeded in recent data for Australian rivers and streams receiving pharmaceutical‐containing effluents from wastewater‐treatment plants. Environ Toxicol Chem 2016;35:1815–1824. © 2015 SETAC

INTRODUCTION

Our growing dependence on pharmaceuticals and their increased availability to consumers in many regions mean that a number of the commonly used active pharmaceutical ingredients, or their transformation products, are becoming detectable constituents of wastewaters [1], [2]. Depending on the effectiveness of the wastewater‐treatment process, there are real prospects for these active pharmaceutical ingredients or their transformation products to reach natural freshwater or coastal ecosystems, with the potential for effects on aquatic ecosystem health. Ecotoxicological investigations have been carried out for many of the popularly used pharmaceuticals; however, there have been limited attempts to derive evidence‐based water quality guidelines that enable regulatory agencies to determine whether measured environmental concentrations pose a significant concern.

The present study collates the available freshwater data for 4 pharmaceuticals—carbamazepine, diclofenac, fluoxetine, and propranolol—and derives high‐reliability guideline values for ecosystem protection of 99%, 95%, and 90% of species using species sensitivity distributions (SSDs) [3]. The latest revisions to the guideline derivation protocols [4] were applied. These involved the following. First, using effects endpoints for development, growth, reproduction, or survival in freshwater organisms and focusing on chronic 10% effect concentration (EC10) data, where available, rather than no‐observed‐effect concentration (NOEC) data and excluding biomarker responses (e.g., behavioral, biochemical, histological, or molecular responses). Second, ensuring that all toxicity data meet the required definitions of chronic tests (in particular, exposure duration should be ≥21 d for juvenile fish tests and ≥7 d for fish embryo tests. Third, high‐reliability guideline values defined as those derived from SSDs with 8 or more data points for chronic exposure (no conversions of acute data to chronic) representing at least 4 taxonomic groups and where the goodness of fit of the Burr type III distribution used in the SSD is acceptable (where the data set includes converted acute data and the fit is good, values are termed “moderate reliability” and where the fit is poor, “low reliability”). All data should be carefully evaluated to ensure they meet acceptability criteria [4].

EXPERIMENTAL PROCEDURES

A thorough review of the literature was undertaken for all freshwater ecotoxicity data relating to carbamazepine, diclofenac, fluoxetine, and propranolol and added to a new data set determined in our laboratories [5]. Because our priority is ecological protection based on population‐relevant endpoints, adverse effects on development, growth, reproduction, and survival were used to derive NOEC or EC10 values, per the recommendation by Hutchinson et al. [6]. This approach recognizes that biomarker responses based on biochemical, histological, and molecular endpoints may be highly useful for exposure monitoring [7], [8], providing mode‐of‐action information, and for developing adverse outcome pathways to help prioritize appropriate testing strategies for ecotoxicology research and risk assessment [9]. Data were sorted into acute and chronic freshwater tests, with the objective of obtaining at least 8 chronic NOEC or EC10 data points for species from 4 or more taxonomic groups. If this was achieved, acute data and chronic data having other endpoints (e.g., EC50 or lowest‐observed‐effect concentrations [LOECs]) were discarded; otherwise, lower‐reliability guidelines could be generated using a combination of converted acute data (using an acute‐to‐chronic ratio or a default value of 10) and chronic data. A quality check of the data as described by Hobbs et al. [10] was then undertaken, and only data of high or acceptable quality were retained as recommended for guideline derivation in Australia and New Zealand [4]. The basic data for each of the pharmaceuticals are summarized in Table 1.

Data were screened to ensure that the endpoints reported were acceptable as chronic tests according to agreed criteria [4], [10]. A freshwater SSD was then obtained from the data set using the BurrliOz, Ver 2, software to derive guideline values that were protective of 99%, 95%, and 90% of species with 50% confidence.

RESULTS AND DISCUSSION

Carbamazepine

A review of the literature found acute toxicity data reported for 6 species and chronic toxicity data reported for 17 species. Of these, acceptable chronic toxicity data were available for 11 freshwater species (2 cladocerans, 2 green algae, 1 blue‐green alga, 1 diatom, 1 midge, 1 rotifer, 1 cnidarian, and 2 fish) representing 8 taxonomic groups (Table 2). The cladoceran Ceriodaphnia dubia was the most sensitive, with an EC10 of 25 μg/L [11]. The data distribution using a Burr type III fit in the SSD was such that it had a long tail (Figure 1), which meant that a 99% protection guideline value could not be determined. The 95% protection guideline value was 9.2 μg/L (Table 3).

Carbamazepine enters the environment largely through discharges from wastewater‐treatment plants (WWTPs), in which it is not effectively removed [12], [13]. It has been detected in discharges from German plants at concentrations up to 6.3 μg/L [14]. Loos et al. [15] reported a mean concentration of 250 ng/L (maximum 12 μg/L) in studies of 122 European river waters. Indian rivers contained 6 ng/L to 128 ng/L [16], Spanish rivers contained 80 ng/L to 3090 ng/L [17], and the Pearl River in China contained 43 ng/L [18]. It has a relatively long half‐life of 38 d in natural waters in the presence of sunlight, with photolysis being the major degradation pathway [12]. Tixier et al. [19] reported a half‐life of 63 d in Lake Greifensee, Germany, indicating that it was relatively persistent.

In all cases, detected concentrations in receiving waters were below the derived guideline value. The guidelines recommended in Switzerland and Germany [20], [21] are considerably lower (Table 4). The Swiss environmental quality standard of 0.5 μg/L was derived by applying an assessment factor of 50 to the most sensitive reliable endpoint, that for reproduction of Ceriodaphnia dubia (25 μg/L) [20]. The available fish data were only for a 10‐d exposure and considered not acceptable for a chronic test, although in Australia and New Zealand the 7‐d test is acceptable for fish embryos and a 21‐d test is required for juvenile fish [4]. The chronic toxicity of chemicals to fish is routinely assessed by using fish early life stage test results. Previous studies have demonstrated that early life stage test results are highly predictive of results from longer‐term exposures including fish reproduction. Chronic exposure for an embryo is completely dependent on the length of time the species typically takes to develop, such as complete organogenesis, pectoral fin development, and jaw development, and is also temperature‐dependent within the same species. Fish full life cycle tests are generally required when there is a suspicion of potential endocrine‐disrupting properties.

In the Swiss study, the scope of the data analysis included both adverse effects data and biomarker responses in contrast to the present study's focus solely on population‐relevant effects (R. Kase, Ecotox Centre, Dübendorf, Switzerland, personal communication). Their guideline value is of unknown reliability given the smaller data set and the arbitrary assessment factor. Ferrari et al. [22], using a limited data set and a log‐normal distribution in a SSD, determined a 95% protection value (reported as a hazardous concentration to 5% of freshwater species) of 2.1 μg/L (Table 4), lower than our value of 9.2 μg/L with a large data set.

Diclofenac

Of 13 chronic data for diclofenac, 11 had EC10 or NOEC values suitable for guideline value derivation (Table 5). These comprised 2 cladocerans, 1 diatom, 2 green algae, 1 blue‐green alga, 1 rotifer, 1 angiosperm, 1 arthropod, and 2 fish, representing 8 taxonomic groups. The most sensitive species was the midge, Chironomus tepperi, with an EC10 of 760 μg/L [5].

Schmitt‐Jansen et al. [23] exposed the green alga Scenedesmus vacuolatus to diclofenac in ultrapure water in sunlight and noted an increase in toxicity measured as growth inhibition, with time over 6 d, with the EC50 decreasing from 46.3 mg/L to 23 μg/L after 6 d. There was a rapid decrease in diclofenac concentrations caused by photodegradation, and the enhanced toxicity clearly resulted from the presence of degradation products. These data were not included because the tests were not conducted in natural waters and the pH was not recorded, nor were EC10 values calculated. It is unclear how the results relate to actual field conditions.

Concentrations in the range 310 ng/L to 930 ng/L have been detected in the effluents from a Swiss WWTP, with concentrations only marginally reduced during passage through the plant [24]. Diclofenac has been detected at <1 ng/L to 12 ng/L in Swiss lakes and at 11 ng/L to 310 ng/L in a nearby river [24] and at 110 ng/L to 220 ng/L in the Höje River in Sweden downstream of a WWTP [25]. Photolysis is the major degradation pathway, with half‐lives near 3 h at summer temperatures [23] (Buser et al. [24] reported 0.9 h) but up to 2 d in winter in some locations [26]. Diclofenac is ionized at the pH of most waters (dissociation constant [pK_a] = 4.2), so it is not readily volatilized, nor does it readily attach to particulates [24].

The measured concentrations are below the guideline value derived in the present study (Table 3) but would exceed the proposed environmental quality standard for the European Commission [27] (Table 4). A report on the European Union guidelines [28] indicated that these values are derived by applying an assessment factor of 10 to the lowest acceptable NOEC for fish. For rainbow trout, both Schwaiger et al. [29] and Triebskorn et al. [30] reported a LOEC of 1 μg/L for a histopathological effect, although the latter referred to a threshold of 5 μg/L for histopathological lesions. A NOEC of 0.5 μg/L was reported by Hoeger et al. [31] for monocyte infiltration/accumulation in livers of brown trout exposed to diclofenac for 21 d. They concluded that the adverse effects in various organs could “possibly compromise fish health.” The environmental quality standard of 0.05 μg/L proposed by the Swiss Ecotox Centre [20] was based on the application of an assessment factor of 10 to the above NOEC for brown trout (R. Kase, Ecotox Centre, Dübendorf, Switzerland, personal communication).

The current Australian and New Zealand approach to biomarker endpoints of this type is that they should not be used in the derivation of water quality guidelines, unless their ecological relevance can be demonstrated [4]. This approach is consistent with that of Hutchinson et al. [32], who advocated that biomarker responses or signals (such as vitellogenin, secondary sexual characteristics, gonadosomatic index, gonad histology, plasma steroids, enzyme induction, and gene expression) may provide valuable mode‐of‐action information to guide species selection and design of chronic testing for adverse effects. Gonadal histology is an indicator of reproductive effects. It has been demonstrated that oocyte atresia can estimate the fecundity and, hence, the reproductive potential of fish populations [32], [33].

A technical disadvantage of histopathology is its qualitative nature. Wolf and coworkers [34] recommended a pathology per review/pathology working group model for assessing the reliability of histopathology results for risk‐assessment process. This recommendation was based on their evaluation of previously published studies of diclofenac in rainbow trout (Oncorhynchus mykiss) [29], [30], [31]. Schwaiger and coworkers [29] evaluated sublethal effects in rainbow trout exposed to diclofenac concentrations ranging from 1 µg/L to 500 µg/L over a 28‐d period by histopathological methods. The histopathological examination of diclofenac‐exposed fish revealed alterations of the kidney such as hyaline droplet degeneration of the tubular epithelial cells and the occurrence of an interstitial nephritis. In the gills, the predominant finding consisted of a necrosis of pillar cells leading to damage of the capillary wall within the secondary lamellae. According to the pathology working group [34], even though hyaline inclusions were observed occasionally as an incidental finding in renal tubular epithelial cells of both control and diclofenac‐exposed trout, the severity of this alteration was never graded higher than minimal by the panel. The panel highlighted that varying husbandry conditions could also lead to numerous hyaline droplets in the proximal renal tubules of normal fish and urged caution in the interpretation of that particular finding. Based on the diagnostic inconsistencies among the 3 studies, the panel found limited evidence to support effects of diclofenac in trout and recommended the overall NOEC to be >320 μg/L [34].

However, based on current scientific evidence, biomarker responses per se should not be used to directly derive water quality guidelines. Moreover, it is recognized that interpretation of many biomarker responses in aquatic organisms is highly complex [32], [33], [34]. Important population‐relevant effect endpoints include survival, length, weight, development, fecundity, fertilization rate, hatching success, and sex ratios. The focus on population‐relevant endpoints for setting guideline values for pharmaceuticals is also proposed by Caldwell et al. [35], [36].

The use of an assessment factor results in a conservative, very low‐reliability guideline value. By contrast, the guideline value derived in the present study would be classified as high reliability based on the criteria being adopted for Australian and New Zealand water quality guideline derivation [4]. Using a limited data set, Ferrari et al. [22] applied a log normal distribution in a SSD to derive a hazardous concentration to 5% of species that protected 95% of species that was of the same order of magnitude as our value of 770 μg/L.

The European Commission's Scientific Committee on Health and Environmental Risks [28] raised a concern regarding the solubility of diclofenac being exceeded in some of the toxicity tests; however, data from Llinas et al. [37] suggest that this would only be an issue in mildly acidic solutions below the diclofenac pK_a. At the pH of natural waters, solubility limitations would not be an issue.

Fluoxetine

There is a large toxicity database for fluoxetine, comprising both acute and chronic freshwater tests as well as others based on behavioral and biomarker endpoints. Of these only 13 reported chronic NOEC, EC10, or 10% inhibitory concentration (IC10) endpoints, comprising 6 green algae, 1 arthropod, 1 angiosperm, 3 crustaceans, 1 gastropod, and 1 fish, representing 6 taxonomic groups (Table 6). Oakes et al. [38] found that the green alga Desmodesmus subspicatus was the most sensitive species to fluoxetine, with a NOEC of ≤0.6 μg/L. Given that NOECs are not a reliable endpoint, most jurisdictions, including Australia and New Zealand, recommend the use of EC10/IC10 values as a more defensible alternative [4]. In the supplementary information to Oakes et al. [38], the plotted dose–response curve showed an IC10 of 1 μg/L, so this was included in the database used in the present study. Along with this species, the New Zealand mud snail, Potamopygus antipodarum, was also very sensitive (Table 6) [39], [40].

The malformation endpoint for the African clawed frog, Xenopus laevis  [41] (Table 6), was deemed unacceptable for use in guideline value derivation because many noncontaminant factors can lead to the types of malformations reported. The 7‐d juvenile fish data for fathead minnow [42] were considered acute and not chronic according to the Australian and New Zealand data selection criteria [4], which require a 21‐d test, so this too was not included.

Fluoxetine is a racemate, a mixture of 2 stereoisomers with mirror‐image structures [4]. The (R)‐enantiomer is known as dextro‐propranolol. The (S)‐enantiomer is known as levo‐fluoxetine. The most common form is a racemic mixture (1:1) of the stereoisomers, supplied as the hydrochloride. To date only 1 study has examined the chronic toxicity of the stereoisomers and found that (S)‐fluoxetine was more toxic than (R)‐fluoxetine to fathead minnow, Pimephales promelas, whereas there was no significant difference in the responses of Daphnia magna  [4]. Fluoxetine photodegradation has a relatively long half‐life (160 d) [43], and its relatively high octanol–water partition coefficient means that it binds preferentially to particulate organic matter.

Measured concentrations of fluoxetine in natural waters are typically in the nanogram per liter range. Kolpin et al. [44] reported a median concentration of 12 ng/L for a range of US streams, and similar values have been reported for waters in Canada and the United Kingdom [38]. Wastewater‐treatment plant effluent concentrations are typically <500 ng/L [45], [46], [47].

The high‐reliability guideline value for fluoxetine derived in the present study was 1.6 μg/L for 95% species protection. No reported environmental quality standard values could be found; however, a number of studies reported predicted no‐effect concentrations (PNECs) for fluoxetine in surface waters. These were all obtained by applying assessment factors to the most sensitive data (Table 4). Thus, Oakes et al. [38] obtained a PNEC of 0.012 μg/L by applying a factor of 50 to the Desmodesmus subspicatus data. Montforts [48] reported a PNEC of 0.031 μg/L using a factor of 1000 with algal toxicity data. Grung et al. [49] reported a PNEC of 0.004 μg/L, and Verlicchi et al. [2] reported a PNEC of 0.05 μg/L. All of these values are assumed to be conservative and of unknown reliability based on the use of assessment factors.

Sumpter et al. [50] discussed the fact that both vertebrates and invertebrates use serotonin as a neurotransmitter and, as such, fluoxetine as a serotonin reuptake inhibitor may have effects on fish (and invertebrate) behavior (e.g., swimming speed, schooling behavior). Such nonstandard endpoints have not been considered in our guideline value derivation.

Propranolol

Although there are published results for over 20 chronic toxicity tests, only 12 reported chronic NOEC or EC10 values, with the remainder only giving EC50 or LOEC values (Table 7). Although both an EC10 and an EC5 were available for the green alga Desmodesmus subspicatus, because of the greater errors around the EC5, the EC10 value was used for guideline derivation [51].

Data were obtained for 2 cladocerans, 1 diatom, 2 green algae, 1 blue‐green alga, 1 rotifer, 1 angiosperm, 1 arthropod, and 3 fish, representing 8 taxonomic groups. Of these, the fathead minnow, Pimephales promelas  [52], and the cladoceran, Ceriodaphnia dubia, were the most sensitive [53]. Like fluoxetine, propranolol is a racemate [54], with the most common form a racemic mixture (1:1) of the stereoisomers, supplied as the hydrochloride.

Propranolol has been detected in WWTP effluents in Germany at a median concentration of 170 ng/L (290 ng/L maximum) [14] and in Sweden near 30 ng/L [25]. Downstream river water concentrations were closer to 12 ng/L (590 ng/L maximum) and 10 ng/L, respectively. High concentrations are unlikely to persist because the laboratory‐determined half‐life for photolytic decomposition was 1.1 h [55]. For sunlight exposure, Liu et al. [56] extrapolating from laboratory studies, calculated a half‐life closer to 1 d in summer and 8 d in winter, with photodegradation being up to 19 times faster than biodegradation.

The present study yielded a high‐reliability freshwater guideline value for propranolol of 14 μg/L. This is almost 100‐fold higher than the value recommended for Switzerland [20]. Their environmental quality standard of 0.16 μg/L used an assessment factor of 50 applied to a NOEC of 8 μg/L for Ceriodaphnia dubia reproduction (R. Kase, Ecotox Centre, Dübendorf, Switzerland, personal communication) [53] (although the value reported in Ferrari et al. [53] was actually 9 μg/L).

CONCLUSIONS

High‐reliability guideline values have been derived for carbamazepine, diclofenac, fluoxetine, and propranolol in freshwaters applying a Burr type III distribution in SSDs of chronic toxicity data (NOECs or EC10s). Data were quality‐assured and had to meet acceptability criteria for “chronic” endpoints. Subchronic biomarker data were excluded from the derivation, and only data for ecologically relevant, population‐related effects were included. The derived guideline values for 95% species protection were 9.2 μg/L, 770 μg/L, 1.6 μg/L, and 14 μg/L for carbamazepine, diclofenac, fluoxetine, and propranolol, respectively. These population‐relevant values are significantly higher than the presumably conservative and unknown reliability values derived for the European Commission, Switzerland, or Germany that are simply based on the application of assessment factors to the most sensitive experimental endpoint (regardless of whether the endpoint is relevant to populations). The calculated freshwater guideline values are not exceeded in recent data for Australian rivers and streams receiving pharmaceuticals via WWTP effluents. Future chronic toxicity studies using a range of taxonomic groups will assist further in the robust ecological risk assessment of pharmaceuticals in the aquatic environment.

Acknowledgment

The present study was funded by the New South Wales Environmental Trust (Project 2009/RD/0008) and the Commonwealth Scientific and Industrial Research Organisation (CSIRO Project R‐705‐02‐ 029).

Conflict of interest

The authors declare no conflict of interest.

Data availability

Data and associated metadata are available from the corresponding author ([email protected]). BurrliOz Ver 2 software is available from https://research.csiro.au/software/burrlioz/

REFERENCES