A Critical Review of Organic Ultraviolet Filter Exposure, Hazard, and Risk to Corals

Abstract There has been a rapid increase in public, political, and scientific interest regarding the impact of organic ultraviolet (UV) filters to coral reefs. Such filters are found in sunscreens and other consumer products and enter the aquatic environment via direct (i.e., recreational activities, effluents) or indirect (i.e., land runoff) pathways. This review summarizes the current state of the science regarding the concentration of organic UV filters in seawater and sediment near coral reef ecosystems and in coral tissues, toxicological data from early and adult life stages of coral species, and preliminary environmental risk characterizations. Up to 14 different organic UV filters in seawater near coral reefs have been reported across 12 studies, with the majority of concentrations in the nanograms per liter range. Nine papers report toxicological findings from no response to a variety of biological effects occurring in the micrograms per liter to milligrams per liter range, in part given the wide variations in experimental design and coral species and/or life stage used. This review presents key findings; scientific data gaps; flaws in assumptions, practice, and inference; and a number of recommendations for future studies to assess the environmental risk of organic UV filters to coral reef ecosystems. Environ Toxicol Chem 2021;40:967–988. © 2021 The Authors. Environmental Toxicology and Chemistry published by Wiley Periodicals LLC on behalf of SETAC.


Text: Page Number
Text S1: Methods detailing the approach to the review. S4 Text S2: Data analysis methods including substitution for LODs and LOQs. S4-S5 Text S3: Toxicity endpoint distributions and methods for converting LOECs to NOECs. S5-S7 Text S4: Additional details on risk assessment methods. S7 Text S5: Further information on UV filter fate. S8 Text S6: Methods detailing approach to sediment data assessment for Table 2 S8-S9 Text S7: Methods detailing approach to coral data assessment for Table 3  S10 Text S8: Additional references S11 Tables: Table S1. List of organic UV filters measured in seawater near coral reefs and some of their details and physical and chemical properties. S12-S13 Table S2. Detailed summary of the twelve studies reporting concentrations of organic UV filters in seawater near or on coral reef locations. S14-S15 Table S3: Summary of the analytical methods and analyses, including quality control and assurance details used in the twelve studies reporting concentrations of organic UV filters in seawater. S16-S17 Table S4. Reported limits of detection (LOD), limits of quantitation (LOQ) and extraction recoveries for UV filters in seawater samples. S18-S19 Table S5: Detailed summary of the nine laboratory toxicity studies in intact hard and soft corals species exposed to organic UV filters. S20-S25  Table S7. S26-S27 Table S7: The risk quotients (RQs) used in Figure 7 in the main text. S28-S31

Supplementary Text:
Text S1: Methods detailing the approach to the review.
Two approaches were taken to obtain the data. First, combinations of the search terms "organic UV filter", "coral", "marine", "exposure", and "risk" were put into the Google Scholar search engine to identify relevant papers. Second, a search was conducted with the Web of Science database using the same terms and additional ones containing the specific UV filter chemical name (all variations e.g. "BP-3", "oxybenzone", "benzophenone" etc.) with results screened for any papers that contained chemical monitoring data in tropical regions and/or toxicity data in corals. Inorganic UV filters are beyond the scope of this review and subsequently in this text the term 'UV filter' exclusively refers to organic UV filters.

Text S2: Data analysis methods including substitution for LODs and LOQs.
When multiple study sites were reported within a near-reef exposure study, all sites were included (e.g. beach and reef) unless sampling locations were located in regions where coral are not found (e.g. the Arctic; Tsui et al. 2014). If site replicates were reported, they were averaged (e.g. Tsui et al. 2017;Mitchelmore et al. 2019) and when temporal data was reported from the same site, it was considered an independent sample (e.g. Barger et al. 2015). Expanded datasets were obtained when possible from authors who only reported their monitoring data in a table or figure (i.e. Tsui et al. 2014;Bargar et al. 2015). Given the different approaches in reporting limits of detection (LOD) and limits of quantitation (LOQ) (including a lack of reporting) we applied a consistent approach across all studies for handling non-detects.
Left censoring of the environmental data was required when authors reported less than the limit of detection (<LOD). This was achieved by using the equation proposed by Antweiler (2015). For example, when a sample or replicate fell below the LOD data substitution according to Equation 1 was undertaken (Antweiler 2015).
!"#$%&%"%&'( = √" " * ,-. [1] Select studies did not report a compound-specific LOD and instead reported a range (Tashiro and Kameda, 2013;Kung et al. 2018). In the case of Tashiro and Kameda (2013) the lower value in the range was used to replace <LOD values for all UV filters in the dataset. This method was chosen because for each of the compounds requiring data replacement, a measured value was reported that was much less that the upper end of the reported LOD range and close to the lower end. In the case of Kung et al. 2018 a more conservative approach could be taken on inspection of the data. The highest LOD reported in the range was used in the left censoring equation. We also evaluated using the lowest LOD in the range and this only changed the data replacement value slightly (i.e. two places after the decimal), therefore it was determined that the use of either LOD would not significantly change the reported results and the highest was used.
If a paper only reported <LOD for a particular compound, no data was reported in Figure   3 and 0% detection frequency given. Data replacement was not used in these instances. reported by the study, the NOEC can be estimated by dividing the LOEC by 2, when the effect is between 10 and 20%. If the effect percentage is unknown a NOEC cannot be calculated. In the vast majority of cases, the treatment concentration below the LOEC was used as the NOEC. We recognize that this is not a perfect solution; however, a NOEC is preferable from a risk assessment perspective than a LOEC. It should also be noted that a dose-response relationship needs to be established to use any of these conversion methods. In certain cases this was not observed, for example several of the endpoints reported by He et al. (2019a,b) reported the LOEC at the highest concentration tested. Therefore, the NOECs reported in Figure 5 (main text) should be treated with caution and seen as preliminary.
Cumulative endpoint ecotoxicity distributions used the nominal exposure values, given that only a few studies attempted to conduct analytical verification and their monitoring was insufficient to understand the mean exposure concentration for each treatment.
The endpoint distributions presented in Figure 5 were derived from the toxicity endpoints summarized in Table S5  numbers (equivalent to the number of endpoints in the dataset) were normalized where the lowest number equals zero percent and the highest 100%. This data was then plotted with endpoint concentration on the X-axis and percentile on the Y-axis. This procedure was followed to create endpoint distribution, all analysis and graphs were created using Graphpad Prism software (Graphpad Software, 2017).
Text S4: Additional details on risk assessment methods.
Risk assessments were summarized in a single figure by plotting the risk quotient (RQ) reported for each compound assessed as described (e.g. Tsui et al. 2014). To make the Tsui et al.
(2017) risk assessment comparable with the others, risk quotients were calculated based on applying the assessment factor (AF) to the hazard data they selected (Danovaro et al. 2008;Downs et al. 2016) and dividing this by the water column data reported per site in each season.

Text S5: Further information on UV filter fate.
Even amongst freshwater solubility estimates, substantial variability is observed. For example the measured solubility for BP-3 reported in the ECHA database is 6 mg/L (Table 1), whereas values of 68 to 210 mg/L are reported in publications (Table S1). In addition to salinity, the fate of UV filters will also be dependent upon the water chemistry and physical environment (e.g. temperature). Many UV filters photodegrade and the extent is influenced by salinity and dissolved organic matter. For example Li et al. (2016)  Text S6: Methods detailing approach to sediment data assessment for Table 2.
The studies all reported data differently and therefore to represent it in Table 2  The Tsui et al. (2017) data was also reported per site for a total of 7 sites (4 in the wet season and 3 in the dry season). The median, minimum and maximum were calculated based on the means from the 7 sites. No data substitution was required to summarize this dataset. Text S7: Methods detailing approach to coral data assessment for Table 3.
The coral data was summarized similarly to the sediment data. Mitchelmore et al. (2019) provided data for each replicate and the average of the three replicates as the sample (each site = 3 replicates), for a total of 19 samples. The median, minimum and maximum is based the averages reported per site. For certain UV filters, there were limited detections so the data was not summarized in as an average per site (i.e. EHMC, 4-MBC, AVO, EDP and CN). CN, EHMC and EDP were detected in various replicates; however, the detections were not consistent (only 1 replicate of three) therefore data substitution technique applied in our analysis to calculate a sample average would not be appropriate because more that 60% of the data would need to be replaced (Antweiler 2015). In the case of 4-MBC and AVO, samples averages were calculated from the replicate data provided by the authors. When data substitution was required in order to calculate a sample mean (e.g. one replicate was a <LOD) the lowest LOD for coral reported by An average per species per site is provided. For comparability with the Mitchelmore et al. (2019) dataset, a site average was calculated from the species-specific data. This give a total of 7 samples (similar to their sediment dataset). The median, minimum, maximum detection frequency were calculated from this 7 sample set. Data substitution was required in order to calculate site averaged for BP-8, EDP and OC. Similarly to the other datasets, if more than one of the three replicates (e.g. >30%) of the data was <LOD, then the average was <LOD. The LODs used in the data substitution equation were 0.99, 0.22 and 0.12 for BP-8, EDP and OC, respectively. Table S1: Summary list of the organic UV filters measured in seawater near coral reefs (using the abbreviations suggested in Table 1 and in parentheses additional ones used in the literature), their chemical structures and some of the physical and chemical properties reported in the literature or predicted by EPISuite.        <LOD: limit of detection (the same as <MOD; method detection limit). <LOQ (the same as <MOQ) is the limit of quantitation. Note LODs/LOQs when reported vary for each UV filter and between each study. ^; combined total of both congeners (E/Z), $; potential sample contamination, *; sample depths are surface seawater samples (S), samples at coral depth or deep collections (D), or microlayer samples (ML). #; additional data on individual sample concentrations provided by the authors. c ; only 1 of 5 samples had a measured value, 4 of 5 were <LOQ. d ; the author (Dr. Tsui) provided a dataset of all individual sample numbers for all UV filters measured at sites near coral reefs which were 9 sites in total (site numbers 9-17) e ; of the n=60 sites, most are not marine samples or near coral reefs but marine coral reef data is indistinguishable using the data in the paper and supplementary file. f : this is the number of individual discrete data points used in this table (and Figure 2  LC50 is the lethal concentration that causes death (mortality) in 50% of the test population, EC50 is the effective concentration that causes 50% of the maximum response; NOEC is the no-observed effect concentration, LOEC is the lowest-observed effect concentration. N.R. = not reported. $ ; toxicity thresholds (i.e. LC/EC50, LOEC or NOECs) are those reported in the publications, those listed in italics are thresholds not implicitly reported but ones we have inferred based on the data presented in the main text or supplementary files, however, it should be noted that these were not reported by the authors or are derived statistically. When multiple toxicity thresholds are reported for the same biological endpoint the one most appropriate for use in a risk assessment is presented. If multiple toxicity thresholds are reported over the time course of an experiment the final timepoint toxicity threshold is reported. #; exposure was to a volume of sunscreen product/formulation containing multiple active/inactive ingredients ^; this study looked at a number of sublethal endpoints in larvae and adult nubbins (bleaching, settlement failure) many resulting in no effect at the highest concentrations reported, *; study used volume of active ingredient of unknown concentration. a ; this endpoint was also reported as 1.39 in the manuscripts Table 1 Figure 2B shows statistical difference between control and 2.28 µg/L concentration but text reports LOEC at 22.8 µg/L. e ; Figure 7A shows that the 22.8 µg/L concentration is statistically different from the control but text reports a NOEC at 22.8 µg/L. f ; this study also exposed corals to sunscreen products and are not reported in this table, these sunscreen product exposures used Acropora sp. and additional coral species (S. pistillata and Millepora complanate) and reported bleaching, zooxanthellae damage and viral load endpoints. g ; reported in Table 1 as 1000 but text and supplementary figures note a lack of significance at this concentration. h : mortality was not specifically measured but inferred as it is stated in the text, no information as to statistical significance is provided. ^^; this study placed coral nubbins in plastic bags with 2 L seawater, sealed the bags, and placed them on the coral reef. n.sig.=not significant from controls   Figure 7 in the main text. The RQs were either taken directly from the Supplementary Information of the corresponding publication or in the case of Tsui et al. (2017) re-calculated based -reflect MEC and PNEC rather than coral MECinternal and PNECinternal. The same assessment factor was applied to the endpoint and compared with the concentration measured in the water column, rather than internally.