The EC (2021) EQS derivation for diclofenac examined the data discussed above (Additional Files) and proposed an EQS of 0.04 µg L-1, based on an estimated 10% effect level of 0.22 µg L-1 for stickleback (Gasterosteus aculeatus) from the mesocosm study (Joachim et al. 2021), divided by an Assessment Factor of 5.
This approach appears to diverge from EC (2018) guidance in two main areas:
- Despite the availability of an extensive and comprehensive dataset of reliable and relevant long-term ecotoxicity studies for diclofenac, a deterministic assessment has been applied.
There are more than sufficient reliable and relevant chronic aquatic toxicity data to allow a probabilistic (SSD) approach to EQS derivation for diclofenac. The application of a deterministic approach (especially one based on data derived from a mesocosm study) when such an extensive chronic toxicity dataset exists for diclofenac should only be attempted when a more statistically robust approach cannot produce a reliable result. This requires that all the available options for an SSD assessment have been thoroughly explored - including the evaluation of all potentially viable statistical models for the SSD curve, and the assessment of the effect of removing insensitive data from the upper portion of the distribution if bimodality is suspected.
- The single mesocosm study using diclofenac (Joachim et al. 2021) has been used to directly derive the EQS directly, despite apparently not meeting all the reliability criteria required by the guidance for employing such an approach.
Joachim et al. (2021) report on a 5-month mesocosm study with diclofenac that included exposure of caged freshwater mussels (Dreissena polymorpha) and free-living stickleback (Gasterosteus aculeatus) to nominal concentrations of 0.1, 1 and 10 μg L-1 in triplicate (with calculated ‘average effective concentrations’, based on measured values, of 0.041, 0.44 and 3.82 μg L-1). The mortality of female fish and mussels after 5 months of exposure to diclofenac are stated by EC (2021) to show a concentration-related response. However, there was both very high mortality in control replicates (up to 60% mortality for fish, and 41% for mussels) and significant variability between replicate mortalities across different mesocosms (both for controls and treatments) in both species. Except at the highest exposure concentration, the degree of variability in controls and treatments overlapped significantly, rendering reported differences between responses observed controls and the two lowest exposure groups to be highly questionable.
The EQS guidance (EC 2018) makes several references to the use of mesocosm data, including specific reliability criteria that must be satisfied if such data are to be used to directly derive an EQS:
- The experimental set-up of the mesocosm must be adequate for assessing the effects of the test substance, and reported in an unambiguous manner;
- The structure of the community of organisms exposed to the test substance in the mesocosm must be realistic, with respect to the type of community that might be exposed in the field;
- There must be an adequate description of the pattern of exposure to the test substance, especially in the compartment of interest (e.g. the water column);
- The statistical evaluation of the results from the study must be scientifically sound; and
- The endpoints investigation must be sensitive to the test substance, and in accordance with the mode of action of the chemical.
A number of these criteria do not appear to be sufficiently met to allow the Joachim et al. (2021) mesocosm study to be considered reliable for the direct derivation of an EQS.
Measured exposure concentrations; the full analytical results were not included in the main paper or Additional Files, and only time-weighted ‘Average Exposure Concentrations (AEC)’ were reported. An approach for calculating these time-weighted average concentrations (the van Wijngaarden et al (1996) AEC approach) was used by Joachim et al. (2021). This approach is specifically designed for use in a pond mesocosm that had been over-sprayed with a chemical on a single occasion once at the start of a study, mirroring the field use of plant protection products. It was not designed to be applied in a continuously dosed stream mesocosm. Nevertheless, Joachim et al. (2021) report AECs of 0.041 ± 0.016, 0.44 ± 0.05, and 3.82 ± 0.47 ug L-1 in the mesocosms treated with three different diclofenac concentrations. These can be compared with simple mean concentrations, which are more appropriate for a continuously dosed stream system, of 0.05-0.06, 0.43-0.49, and 3.86-4.17 ug L-1 diclofenac, respectively. However, both the AEC and simple mean concentrations mask such extremely large variations in exposure concentrations that they are both probably meaningless. For example, the measured exposure concentrations of diclofenac to which mobile organisms such as stickleback were exposed at the highest nominal concentration of 10 ug/L ranged from 0.14 to 7.235 ug L-1 diclofenac – a factor of over 50, over the course of the study.
In addition, the analytical results highlight that measured concentrations were <50% of nominal in all treatments at all measurement times, even at the inlet to the mesocosms. The analytical data also include a significant number of censored results (less than the limit of quantification (LoQ)) in the lowest two exposure concentrations at both 5 and 19 metres along the mesocosms. In addition, there are several sampling occasions across treatments when the diclofenac concentration apparently increased along the length of the mesocosm, which is a very unusual finding, especially given the >50% loss compared to nominal concentrations in solutions entering the mesocosms. This was particularly pronounced on the last sampling date when reported concentrations were higher in the lower mesocosm reaches in all mesocosms. Overall, this combination of issues with the analytical measurement of exposure concentrations would usually invalidate a study for use as key data in EQS derivation.
Secondly, the degree of control mortality for freshwater mussels and fish was high in the control mesocosms, with up to 41% mortality for (caged) mussels and up to 60% mortality for stickleback across control treatments by the end of the five-month study. This level of control response would invalidate these data if they were obtained from a laboratory study, yet they are used by EC (2021) for EQS derivation in the same way as laboratory data.
Finally, regarding the statistical analyses, Joachim et al. (2021) do not report an EC10 for any of the measured endpoints from the mesocosm in their paper, but the EC (2021) assessment estimates an EC10 estimate for female stickleback mortality of 0.22 µg L-1 diclofenac with a 95% confidence interval ranging over two orders of magnitude (0.0385 - 1.30 µg L-1). This is then used in a deterministic approach to deriving an EQS. The reasons for the high uncertainty in the female stickleback EC10 value are clear: the data are highly variable across a relatively small number of treatments (one control and three diclofenac concentrations). However, there is no evidence of a statistically significant effect on female stickleback mortality below the highest test concentration.
The mesocosm for diclofenac seems to be sufficiently reliable to be used as ‘supporting’ data, i.e. to qualitatively assess the uncertainties of the EQS derivation, and select an assessment factor, as detailed in the guidance (EC 2018). However, issues with data variability, high control effects, and uncertain exposure metrics mean that the outcomes of this study appear to be insufficiently reliable to be used as ‘critical’ or key data for direct derivation of an EQS.
In addition, the EC (2021) also derive further deterministic EQS for ‘laboratory’ data using the freshwater mussel mortality data from the mesocosm and use this as a ‘weight-of-evidence’ to support the mesocosm-based deterministic EQS. There is no precedent (nor mention in the guidance (EC 2018)) for including the results from a mesocosm in the derivation of a laboratory-data based EQS, using either deterministic or probabilistic derivation approaches. Indeed, a mortality endpoint with the degree of control response and within-treatment variability observed in the mussel data for diclofenac would be considered unreliable for use in EQS derivation if it came from a laboratory study.