Toxicity of ten herbicides to the tropical marine microalgae Rhodomonas salina

Herbicide contamination of nearshore tropical marine ecosystems is widespread and persistent; however, risks posed by most ‘alternative’ herbicides to tropical marine microalgae remain poorly understood. Experimental exposures of the important but understudied microalgae Rhodomonas salina to seven individual Photosystem II (PSII) inhibitor herbicides (diuron, metribuzin, hexazinone, tebuthiuron, bromacil, simazine, propazine) led to inhibition of effective quantum yield (ΔF/Fm′) and subsequent reductions in specific growth rates (SGR). The concentrations which reduced ΔF/Fm′ by 50% (EC50) ranged from 1.71-59.2 µg L−1, while the EC50s for SGR were 4-times higher, ranging from 6.27-188 µg L−1. Inhibition of ΔF/Fm′ indicated reduced photosynthetic capacity, and this correlated linearly with reduced SGR (R2 = 0.89), supporting the application of ∆F/Fm’ inhibition as a robust and sensitive indicator of sub-lethal toxicity of PSII inhibitors for this microalga. The three non-PSII inhibitor herbicides (imazapic, haloxyfop and 2,4-Dichlorophenoxyacetic acid (2,4-D)) caused low or no toxic responses to the function of the PSII or growth at the highest concentrations tested suggesting these herbicides pose little risk to R. salina. This study highlights the suitability of including R. salina in future species sensitivity distributions (SSDs) to support water quality guideline development for the management of herbicide contamination in tropical marine ecosystems.

Microalgal toxicity tests for derivation of water quality guidelines. National WQGVs (referred to by ANZG 39 as default GVs) are derived in Australia to protect 99%, 95%, 90% and 80% (PC99, 95, 90, 80, respectively) of marine and freshwater communities by estimating community sensitivity from species sensitivity distributions (SSDs) 45 . The minimum data required for SSDs to meet WQGV criteria are toxicity thresholds for at least five species from at least four phyla that are characteristic of the receiving environment 45 . For a recent and detailed description of the methods and criteria in the Australian context see Warne et al. 45 . With rapid growth rates that allow for chronic exposure testing in a short period, marine microalgae represent a suitable taxon to contribute to future SSDs. Currently, SSDs are developed using toxicity data from chronic exposures that are ecologically relevant, and for microalgal toxicity testing inhibition of growth is the most common ecologically relevant endpoint 39,45 . However, strong correlations between effects on microalgae growth and reduced photosynthetic efficiency in estuarine microalgae as measured by Pulse Amplitude Modulation (PAM) fluorometry has been demonstrated for several PSII inhibitor herbicides 23 . The inhibition of effective quantum yield (ΔF/F m ′) by PSII inhibitor herbicides is proportional to the inhibition of photosynthetic efficiency at a given irradiance 46 and could be considered as a rapid, sensitive and non-invasive alternative for growth measurements in microalgae toxicity tests involving PSII inhibitor herbicides 23 . In previous studies, inhibition of ΔF/F m ′ has been extensively applied for assessing the toxicity of PSII inhibitor herbicides in microalgae 23,[46][47][48][49] and has also revealed herbicide-induced community tolerance in microalgae to PSII inhibitor herbicides over chronic exposures 50,51 . However, this sensitive photophysiological response may not be suitable as an ecologically relevant measure of whole organism stress for microalgae to non-PSII inhibitor herbicides where the mode of action does not involve PSII 34,52 . Further comparisons between the inhibition of growth and ΔF/F m ′ as endpoints for herbicide toxicity in marine microalgae are therefore warranted to demonstrate the relevance of using ΔF/F m ′ as an ecological relevant endpoint in future SSDs.
In order to improve WQGVs for herbicides and expand toxicity threshold data for tropical marine species to alternative herbicides, this study tested the effects of several herbicides on growth and ΔF/F m ′ to the marine microalgae Rhodomonas salina. This species was selected as a tropical representative of an understudied phylum, Cryptophyta, generally underrepresented in SSDs. In addition, this study aimed to derive no effect concentrations (NECs), which are the preferred toxicity estimates for inclusion in SSDs to derive WQGVs. Nine herbicides detected in the GBR and catchments 17,37 that indicated current toxicity data gaps (based on consultation with the Queensland Department of Environment and Science (DES)) were selected for testing, along with the reference herbicide diuron. The tested herbicides included the PSII inhibitor herbicides tebuthiuron, hexazinone, metribuzin, simazine, propazine, bromacil, and the non-PSII inhibitor herbicides, haloxyfop, 2,4-dichlorophenoxyacetic acid (2,4-D) and imazapic. The toxicity thresholds identified provide valuable toxicity www.nature.com/scientificreports www.nature.com/scientificreports/ data for alternative herbicides detected in GBR waters and will contribute to new and improved WQGVs for application in risk assessments.

Results
Assay performance. Rhodomonas salina displayed exponential growth in control treatments across all bioassays with SGR ranging between 1.07 ± 0.07 d −1 and 1.29 ± 0.02 d −1 (mean ± SD) ( Table 2). ΔF/F m ′ measurements of control treatments varied between 0.45 ± 0.02 and 0.53 ± 0.01 (mean ± SD). The carrier solvents (<0.01% v/v) had no significant influence on SGR compared with filtered seawater after 72 h (ANOVA, F ethanol (1,3) = 1.12; p = 0.37; F DMSO (1,3) = 0.15; p = 0.73). The reference toxicant diuron used in each growth test and fluorescence well plate assay inhibited SGR and ΔF/F m ′ between 30.1 ± 2.2% and 57.2 ± 2.8% and between 78.4 ± 2.0% and 97.7 ± 2.2% (mean ± SD), respectively (Table 2). This level of variability was expected between independent experiments conducted across 10 occasions and may have been due to minor differences in nutrients or the physiology of cells at the start of each test.
Physicochemical measurements indicated little variation within each treatment and across all tests over 72 h: pH 8.5 ± 0.4; salinity 34.2 ± 0.6 PSU, dissolved oxygen 8.0 ± 0.4 mg L −1 (± SD, n = 169 for each parameter), temperature 26.0 ± 0.6 °C ( ± SD, 10-min logging intervals). Herbicide concentrations were measured at 0 h and 72 h of each toxicity test to estimate the potential losses of herbicides due to degradation, volatilization or adsorption over the 72-h test duration. Chemical analyses showed that the time-averaged measured concentrations (between 0 h and 72 h samples) were within 20% of nominal concentrations for diuron, metribuzin, hexazinone, bromacil, tebuthiuron, and 2,4-D and between 30-50% of nominal concentrations for propazine, simazine and imazapic. No contaminant was detected in the control treatments and a summary of the nominal and measured concentrations can be found in Table S1.
Effects of PSII inhibitor herbicides on growth. Toxicity tests using R. salina were performed on seven PSII inhibitor herbicides, including the reference herbicide diuron ( Table 3). The growth of R. salina was inhibited by all PSII inhibitor herbicides, and diuron was the most toxic of all PSII inhibitor herbicides with an EC 50

Herbicide
Mode of action  39 (all of low reliability) based on freshwater species and proposed water quality guideline values (PGVs) 43,44,93 for 99%, 95%, 90% and 80% species protection (based on marine and freshwater species) against toxicity thresholds [no effect concentration (NEC); effect concentration inhibiting the specific growth rate by 10% (EC 10) )] values derived for Rhodomonas salina in this study (from Table 3). All concentrations in µg L −1 . NA signifies no available guideline values. Bold indicates herbicides tested in this study. *Level of protection unknown.
www.nature.com/scientificreports www.nature.com/scientificreports/ value of 6.27 µg L −1 ( Table 3). A summary of the slope and goodness of fit of each concentration-response curve (Sigmoidal, 4 parameter model) for SGR ( Fig. 1) is shown in Table S2. The comparison between relative potencies (ReP) based on EC 50 values to the reference herbicide diuron indicated the order of toxicity: diuron > hexazinone > metribuzin > bromacil > tebuthiuron > simazine > propazine ( Table 2). The EC 10 and modelled no effect concentrations (NECs) were also reported in Table 2 and showed similar orders of toxicity (Fig. 2).
Effects of PSII inhibitor herbicides on effective quantum yield. Diuron, metribuzin, hexazinone, tebuthiuron, and bromacil, caused 100% steady-state inhibition of ΔF/F m ′ in R. salina after 24 h exposures ( Fig. 1). Propazine and simazine did not reach 100% steady-state inhibition, peaking at a maximum of 90% inhibition of ΔF/F m ′ at the highest concentration tested (Fig. 1). A summary of the slope and goodness of fit of each concentration-response curve (Sigmoidal, 4 parameter model) for ΔF/F m ′ ( Fig. 1) is shown in Table S2. The comparison of herbicide concentrations inhibiting ΔF/F m ′ by 50% (EC 50 ) revealed the order of toxicity: diuron > metribuzin > bromacil > hexazinone > tebuthiuron > propazine > simazine (Table 3). Comparable patterns were observed for the order of potencies with respect to ΔF/F m ′ EC 10 values ( Table 3).

Relationship between inhibition of effective quantum yield and growth. The relationship between
EC 50 values for SGR and ΔF/F m ′ obtained for each PSII herbicide was compared in two ways, with both demonstrating that inhibition of ΔF/F m ′ was more sensitive than inhibition of growth. Firstly, we compared the EC 50 ratios for SGR: ΔF/F m ′ which ranged from 1.5 -7.0 and averaged 4.3 (Table 3). Secondly, we plotted the linear relationship (R 2 = 0.87) of EC 50 values for each herbicide for SGR and ΔF/F m ′ (Fig. 4) which yielded a slope of 3.48.

Discussion
Toxicity effects of PSII inhibitor herbicides. Substantial reductions in both ΔF/F m ′ and SGR of R. salina were observed following exposure to all seven PSII inhibitor herbicides from four different chemical classes. Since Photosystem II is conserved across phototrophs 53 , the response of R. salina to the four classes of herbicides including phenylureas (diuron and tebuthiuron), triazines (simazine and propazine), triazinones (metribuzin and hexazinone) and uracils (bromacil) was expected. The toxicities of PSII inhibitor herbicides varied by over 20-fold with respect to inhibition of SGR and ΔF/F m ′, but a relationship between toxicity and chemical classes was not observed. For example, intra-class variations were wide within the phenylureas with diuron up to 18-times more toxic than tebuthiuron and similar disparities were evident within the triazinones (Table 3). Although all of these herbicides have the same mode of action, differences in PSII activity have been observed in a number of marine phototrophs 21,24,28,47 . Chesworth et al. 54 suggested that herbicides with a greater affinity and faster rate of binding  Table 2. Assay performance. Specific growth rate (SGR, d −1 ) and photosynthetic efficiency (ΔF/F m ′) measurements of control and reference (diuron, 4 µg L −1 ) treatments and diuron reference percent inhibition effect (Ref. inh (%)) (mean ± SD; n = 5 per treatment). *Note for the simazine toxicity bioassay, a reference treatment of diuron, 2 µg L −1 was used instead of 4 µg L −1 and therefore not included in calculations of the total mean.
www.nature.com/scientificreports www.nature.com/scientificreports/ to the Q B site accumulate more effectively leading to higher potencies of these herbicides. Furthermore, the binding of some herbicides to the Q B site lowers the redox potential of the plastoquinone Q A /Q A -redox couple within PSII, resulting in increased photooxidative stress and subsequently a higher toxicant PSII activity 55,56 , potentially explaining some of the differences observed. comparative species sensitivity. Several studies investigating toxicity of herbicides to tropical marine algae have applied standard test species, such as the diatom Phaeodactylum tricornutum 47,49,57 or symbionts of the family Symbiodiniaceae isolated from corals 21,58 . Toxicity values (EC 10 s and EC 50 s) from these studies and the present study are summarized in Table 4. While some EC x values are similar between species, others differ by up to an order of magnitude ( Table 4). Some of these differences will be due to inherent difference is in cell structure, biochemistry and physiology between different species. For example, Millie et al. 59 have shown that algae sensitivity to PSII inhibitor herbicides were related to differences in light-harvesting pigments under different  Table 3. Toxicity threshold summary. Derived effect concentrations (EC 10 and EC 50 from Fig. 1) and no effect concentrations (NECs from Fig. 2) with 95% confidence intervals for each herbicide, and relative equivalent potencies (ReP). NA indicates values could not be calculated. Concentrations are reported in µg L −1 .
www.nature.com/scientificreports www.nature.com/scientificreports/ light conditions. Guasch and Sabater 60 reported that the toxicity of PSII inhibitors was lower for diatom species that were already adapted to low light conditions. In addition, Tang et al. 61 observed higher herbicide sensitivity in chlorophytes compared to diatoms, suggesting some diatoms may apply an extra carbon fixation pathway, such as β-carboxylation that could compensate for the shutdown of PSII-based photosynthesis, and allow algal  www.nature.com/scientificreports www.nature.com/scientificreports/ metabolism to continue 62 . The sensitivity of algal species to PSII inhibitor herbicides has also been shown to be affected by cell size 63 . Although most growth tests are relatively standardized, care should be taken before directly comparing toxicity values between studies, as even subtle differences in experimental exposure and conditions are likely to affect responses. Prop. decline in SGR rel. cont.
Prop. decline in SGR rel. cont. www.nature.com/scientificreports www.nature.com/scientificreports/ Relationship between inhibition of effective quantum yield and growth. The inhibition of R.
salina growth was on average 4-times less sensitive to PSII herbicide exposures than the photoinhibition endpoint (ratios of EC 50 s SGR: ΔF/F m ′ for each herbicide can be found in Table 2). The correlation plot of EC 50 values for both endpoints had a slope of 3.5, also showing a greater sensitivity of ΔF/F m ′ to PSII inhibitor herbicides. Magnusson et al. 23 had demonstrated a relationship between SGR and ΔF/F m ′ inhibition by PSII inhibitor herbicides that was closer to 1:1 for two tropical benthic microalgae; Navicula sp. and Nephroselmis pyriformis. Both studies clearly demonstrated a link between inhibition in ΔF/F m ′ and decreasing growth rates; however, the direct link between the binding of PSII inhibitor herbicides to the D1 protein (reducing electron transport and causing damage to PSII) with growth is not necessarily expected to be 1:1 for all taxa and experimental conditions. Light intensity and light acclimation history have large influences on the relationships between photophysiology, primary production and growth 64 . Furthermore, it has been shown that the pigment structures in some microalgae, such as red algae (rhodophytes) can shift between PSI and PSII, potentially affecting the path of electron transport 65,66 and direct quantitative links between ΔF/F m ′, primary production, and SGR may be less certain in rhodophytes than for some other phototrophs. Cryptophytes also contain phycobiliproteins, the characteristic antennae pigments of the prokaryotic cyanobacteria and the eukaryotic rhodophytes 67 . The presence of phycobiliproteins in cryptophytes may allow shifting between PSI and PSII in R. salina although, phycobiliproteins are only present in the thylakoid lumen and as phycoerythrin 68 . While Magnusson et al. 23 measured effects on both growth and ΔF/F m ′ over 3 d, our comparison was between a chronic 3-d growth test and an acute 24-h ΔF/F m ′ test (as effects of PSII inhibitor herbicides on microalgae typically peak before 6 h and remain consistent over longer periods 69 ). While these differences in exposure durations make direct comparisons between the techniques (and against prior studies) more difficult, the exposure durations are optimal for each test type and the  www.nature.com/scientificreports www.nature.com/scientificreports/ linear relationship between effects of multiple PSII inhibitor herbicides on photosynthetic efficiency and growth remains strong. The consistency of these results for a variety of marine microalgae and reported in this study for the marine cryptophyte, in combination with the direct mechanistic link between ΔF/F m ′ and SGR for PSII inhibitor herbicides, suggests that ΔF/F m ′ provides a robust and sensitive endpoint for determining sub-lethal effect thresholds for these herbicides.   www.nature.com/scientificreports www.nature.com/scientificreports/ Toxicity effects of non-PSII inhibitor herbicides. R. salina was far more sensitive to the PSII inhibitor herbicides than the non-PSII inhibitor herbicides tested here. R. salina was insensitive to non-PSII inhibitor herbicides within the phenoxy family, haloxyfop and 2,4-D at the highest concentrations tested. Growth regulator herbicides, such as 2,4-D, inhibit the plant hormone auxin and are primarily used as selective herbicides for controlling broadleaves (dicots) 43 . This pathway is not present in cryptophytes, explaining the lack of toxicity of 2,4-D to R. salina. Previous studies reported similar observations of the low toxicity of 2,4-D on the growth rate of marine microalgae (Table 4), with the most sensitive species the diatom Chaetoceros calcitrans (21 d, EC 50 = 9.2 mg L −1 ). R. salina was also less responsive to haloxyfop, which targets the acetyl-CoA carboxylase enzyme involved in the synthesis of lipids and fatty acids in plants 44 . ACCase inhibitors target the homomeric (eukaryotic) form of the enzyme rather than the heteromeric (prokaryotic) form 44 and microalgae, such as rhodophytes and chlorophytes contain the heteromeric ACCase enzyme in their plastids 70 , likely explaining the insensitivity towards haloxyfop in cryptophytes. Imazapic was only toxic to R. salina at high concentrations. Imazapic inhibits the activity of the enzyme acetohydroxy acid synthase (AHAS or ALS), which is responsible for catalyzing the production of several branched-chain aliphatic amino acids across many aquatic phototrophs 71 . Other marine microalgae are similarly insensitive to imazapic, and the sensitivity to imazapic of enzyme variants in marine microalgae are unknown. No effect on the growth rates of the marine microalgae Navicula sp. and Nephroselmis pyriformis were observed after 10 d exposure at concentrations of up to 1.5 mg L −1 69 . Conversely, imazapic is far more toxic to freshwater phototrophs. For example, imazapic has an EC 50 of 6.1 µg L −1 for growth in the freshwater macrophyte Lemna gibba (duckweed) 72 . In macrophytes, imazapic is absorbed through the roots and shoots of plants, possibly explaining the lower toxicity of imazapic to microalgae 71 . Another factor to consider with respect to the sensitivity of marine species is whether the structure of imazapic may affect its exposure and bioavailability in seawater. Imazapic contains a carboxylic acid (COOH) which may result in complexation with Mg 2+ and Ca 2+ ions in the seawater 73 , or stabilize the herbicide at the seawater:air interface 74 . Both mechanisms could reduce the exposure and bioavailability of imazapic to marine species accounting for the low toxicities reported. toxicity thresholds for guideline development. Water quality guidelines are usually developed using SSDs 75 and ideally from NEC or EC 10 values for multiple diverse taxa 45 . However, most current herbicide WQGVs for marine communities are of low reliability (e.g. developed from toxicity data for as few as five species), or have not yet been developed due to the lack of data 39 . Currently, WQGVs exist only for the five priority herbicides (diuron, atrazine, ametryn, tebuthiuron, hexazinone) and four alternative herbicides (bromacil, MCPA, simazine, and 2,4-D) 39 (Table 1). There are no WQGVs for metribuzin, propazine, haloxyfop, and imazapic. A comparison of the existing ANZG WQGVs 39 and proposed guideline values (PGVs) 43,44 against herbicide toxicity thresholds (SGR: NEC and EC 10 values) for R. salina is presented in Table 1. The NEC and EC 10 values for hexazinone and bromacil were far lower than current PC99 WQGVs; however, the PC99 PGVs would all be protective of R. salina (Table 1). Nevertheless, most of the PGVs are of very low to moderate reliabiltiy 43,44 (Table 1) and could be improved by the incorporation of toxicity data from additional species, such as R. salina. Apart from diuron (the most toxic herbicide tested in this study), the NEC and EC 10 values were all greater than 2 µg L −1 and above concentrations that have been detected in tropical coastal waters 13,16,17,37 . However, the risks posed by these herbicides should not be assessed individually as they are usually detected in complex mixtures of multiple herbicides. Instead, their contribution to the total risk can be assessed using ms-PAF 42 , which accounts for all herbicides that have reliable SSDs (and WQGVs). The ms-PAF method has also been extended to include the additional influence of heatwave conditions on WQGVs for pesticides 76 . The exceedance of PC99 values for herbicide mixtures has recently been reported in water quality monitoring programs using ms-PAF, where individual herbicide did not exceed their own PC99 values 37 . The development of SSDs for alternative herbicides detected in the GBR using relevant toxicity data (such as the R. salina data presented here) will allow their contribution in predicting the cumulative risks of herbicide mixtures using ms-PAF.

conclusion
Alternative herbicides may be practical substitutes for controlling weeds; however, their toxicity to non-target species such as R. salina could contribute to the combined risks posed by herbicide mixtures regularly detected in coastal waters in the tropics. In the present study, exposures of R. salina to increasing herbicide concentrations resulted in inhibition of ΔF/Fm′ within 24 h, indicating reduced photosynthetic efficiency which led to reduced growth rates over 72 h chronic exposures. Photoinhibition was a more sensitive endpoint over 24 h than inhibition of growth over 72 h; however, the relationship between inhibition of ΔF/Fm′ and SGR was linear and consistent. Importantly, the non-PSII inhibitor herbicides (imazapic, 2,4-D, haloxyfop) were substantially less toxic than the most toxic PSII inhibitor herbicides, indicating these herbicides pose little risk to this microalga in the marine environment. The toxicity thresholds (NECs and EC10s) derived here were higher than concentrations detected in tropical marine waters. However, the risk posed by these herbicides to marine species is better assessed by comparing measured values in the field against high-reliability WQGVs that are derived from SSDs. The current study contributes targeted data towards developing SSDs for alternative herbicides that are essential to improve predictions of the cumulative ecological risks posed by herbicide mixtures (using ms-PAF) detected in marine monitoring programs. While this study targeted some of the most frequently detected alternative herbicides in GBR waters, there remains a number of pesticides, including insecticides and fungicides with no current WQGVs and further testing is needed to address this. Methods test species and culture conditions. The cryptophyte Rhodomonas salina (Wislouch) 77 (CS 24/01) was purchased from the Australian National Algae Supply Service, Hobart (CSIRO). Cryptophytes are an important component of the primary producers in both freshwater and marine habitats, and changes in their abundance, composition and nutritional value may initiate an indirect bottom-up effect on higher trophic levels 78 . Many www.nature.com/scientificreports www.nature.com/scientificreports/ species are widespread and abundant in the sea within wide temperature ranges (5-29 °C), which make this phylum highly suitable for acute and chronic toxicity tests in a short period of time under both temperate and tropical conditions [79][80][81] . Cultures of R. salina were established three weeks prior to experimentation in Guillard's f 2 marine medium (0.5 mL of AlgaBoost F/2, AusAqua in 1 L sterile 0.5 µm-filtered seawater (FSW; pH 8.0, salinity 35.0 psu)) 82 . Cultures were maintained in sterile 500 mL Erlenmeyer flasks as batch cultures in exponential growth phase with twice-weekly transfers of 70 mL of a 3-to 4-day-old R. salina suspension to 300 mL f 2 medium under sterile conditions. Clean culture solutions were aerated and maintained at 26 ± 1 °C and under a 12:12 h light:dark cycle (90-100 μmol photons m -2 s -1 , Osram Lumilux Cool White 36 W).  Table S1. toxicity test protocol. Cultures of R. salina were exposed to a range of herbicide concentrations over a period of 72 h. Inoculum was taken from cultures in the exponential growth phase (4-day-old with cell density of approximately 1 × 10 6 cell mL −1 ). Prior to the inoculation of the test solutions, 15 mL of algae suspension (of the 4-day-old algal culture) was washed in 30 mL sterile 0.5 µm-FSW by centrifuge in 50 mL Falcon tubes at 1500 g for 5 minutes (Eppendorf Centrifuge 5810 R, Bio-strategy). The supernatant was decanted, and the cell pellet re-suspended in 30 mL of sterile 0.5 µm-FSW and homogenized by vortexing. This process was repeated three times to remove the nutrient-enriched f 2 culture medium, which might affect herbicide toxicity [83][84][85][86] . The cell pellet was finally re-suspended in about 15 mL of sterile 0.5 µm-FSW. The cell density of the concentrated algae suspension was measured from two 500 µL sub-samples by flow cytometry. The desired inoculum was calculated to a given starting cell density of 3 × 10 3 cells mL −1 in the following toxicity test. Individual R. salina working suspensions for each herbicide treatment were prepared in individual 100 mL Schott glass bottles by adding the required algae inoculum and sterile 0.5 µm-FSW. Each Schott glass bottle was finally dosed with a range of herbicide concentrations (Table S1). Bioassays for each herbicide were performed on different days with fresh algae, FSW and herbicide stocks. In each bioassay, a control (no herbicide) and reference (diuron, 4 µg L −1 ) treatment was added to ensure the response is reproducible. Diuron was chosen as a reference toxicant as it is a widespread contaminant and its toxic mode of action (PSII inhibition) and toxicity to a wide variety of microalgae are well understood (see Magnusson et al. 23 ).

Herbicide stock preparation.
Five replicated aliquots of 10 mL were transferred from the individual 100 mL Schott glass bottles into sterile 20 mL glass scintillation vials and incubated at 26.0 ± 0.6 °C under a 12:12 h light:dark cycle at 90-100 μmol photons m -2 s -1 (Osram Lumilux Cool White 36 W). Vials were randomized and swirled daily. Sub-samples of 500 µL were taken from each replicate to measure cell densities of algal populations at 0 h and 72 h using a flow cytometer (BD Accuri C6, BD Biosciences, CA, USA) equipped with red and blue lasers (14.7 mW 640 nm Diode Red Laser 20 mW 488 nm Solid State Blue Laser) and standard filter setup. The flow rate was set to 35 µL min −1 , 16-µm core size with a sample volume of 50 µL. Cell densities were obtained by plotting a two-dimensional cytogram. A fixed gating was used around the viable (chlorophyll fluorescing) cells, which allowed for differentiation of non-algal particles (debris) and dead cells from viable R. salina cells. Viable cells typically represented 75 -95% of particles counted (control treatment at 72 h). Each 500 µL sub-sample was analyzed by the flow cytometer two times and an average taken of the number of events that occur within the gated region. This process was then repeated for each replicate per treatment. Specific growth rates (SGR) were expressed as the logarithmic increase in cell density from day i (t i ) to day j (t j ) as per Eq. (1), where SGR i-j is the specific growth rate from time i to j; X j is the cell density at day j and X i is the cell density at day i 87 . chlorophyll fluorescence measurements. Acute effects of herbicides on the photophysiology of R. salina, measured by chlorophyll fluorescence as the effective quantum yield (ΔF/F m ′), were investigated in non-pyrogenic polystyrene 48 well-plates with lid (Nunclon Delta, Thermo Scientific) using imaging PAM fluorometry (I-PAM, Walz, Germany) 46,88 , following an exposure period of 24 h at an irradiance of 90-100 μmol photons m -2 s -1 . Inoculum was taken from mother cultures in the exponential growth phase (4-day-old with cell density of approximately 1 × 10 6 cell mL −1 ). Initial testing of varying cell densities indicated that consistent ΔF/ F m ′ measurement signals >0.45 46 were obtained at a starting cell density of 3.5 × 10 5 cells mL −1 (equivalent to cell density after ~3 d in the SGR inhibition test). Individual R. salina working suspensions for each herbicide treatment were prepared in individual 50 mL Schott glass bottles by adding algae inoculum and sterile f 2 (0.5 www.nature.com/scientificreports www.nature.com/scientificreports/ µm-FSW) marine medium. Each 50 mL Schott glass bottle was finally dosed with a range of herbicide concentrations (Table S1). Five replicated aliquots of 1 mL were transferred from the individual 50 mL Schott glass bottles across two 48-well plates (randomly) and incubated at 26.0 ± 0.6 °C under a 12:12 h light:dark cycle at 90-100 μmol photons m -2 s -1 (Osram Lumilux Cool White 36 W). Replicated seawater controls (SWC) (n = 5) or solvent controls (SC) and diuron references (4 µg L −1 ) were included randomly across each 48-well plate to ensure consistency in inhibition response between replicated algae cultures. Light adapted minimum fluorescence (F) and maximum fluorescence (F m ′) were determined and effective quantum yield was calculated for each treatment as per Eq. (2) 88 . The timing of plate preparation and measurements were staggered to ensure a consistent exposure duration of 24 h. Imaging PAM settings were set to actinic light = 1 (corresponding to photosynthetically active radiation (PAR) of 90-100 μmol photons m −2 s −1 ), measuring intensity = 11, gain = 3; damp = 2.
A screening process of plates containing algae suspension only was performed immediately prior to exposure with herbicides to ensure that ΔF/F m ′ > 0. 45. physicochemical analyses. Physico-chemical water quality parameters including pH and salinity (LAQUAact-PC110 Meter, HORIBA Scientific) and dissolved oxygen (HQ30D Portable Meter, HACH) were measured from individual 100 mL Schott glass bottles at 0 h and replicated 20 mL glass scintillation vials pooled for each concentration at 72 h. Temperature was logged in 10-min intervals over the total test duration (HOBO, Onset). Analytical samples were also taken from individual 100 mL Schott glass bottles at 0 h and replicated 20 mL glass scintillation vials pooled for each concentration at 72 h. Aliquots (1 mL) were transferred into 1.5 mL Liquid Chromatography amber glass vials and spiked with surrogate standards (i.e. diuron-D6, hexazinone-D6, metribuzin-D3, simazine-D10, propazine-D6, bromacil-D3, haloxyfop-D4, 2,4-D-13 C 6 , and imazapic-D7) at a final concentration of 10 ng mL −1 . Prior to analysis samples were stored at −20 °C, defrosted and centrifuged. Herbicide concentrations were determined by HPLC-MS/MS using a SCIEX Triple Quad 6500 QTRAP mass spectrometer (SCIEX, Concord, Ontario, Canada) equipped with a TurboIonSpray probe 10,89,90 . The mass spectrometer was coupled to a Shimadzu Nexera X2 uHPLC system (Shimadzu Corp., Kyoto, Japan) using a Phenomenex Kinetex Biphenyl column (2.6 μm 50 ×2.1 mm 100 Å) for analyte separation. Five μL of sample was injected on to the column followed by a linear gradient starting at 10% B for 0.5 min, ramped to 100% B in 4.7 min then held at 100% for 4.0 min followed by equilibration at 10% B for 3.0 min (A = 1% methanol in milli-Q water, B = 95% methanol in milli-Q water, both containing 0.1% acetic acid). The mass spectrometer was operated in both positive and negative ion mode using a scheduled multiple reaction-monitoring method (sMRM). Positive samples were confirmed by retention time and by comparing transition intensity ratios between the sample and an appropriate calibration standard from the same run. To provide estimates of 'measured' concentrations used for concentration-response modelling the geometric mean from measured start and end concentrations (time-weighted average) was assigned as the 'actual' concentration in that sample. The average loss from these measured concentrations was then applied to all nominal concentrations.

Statistical analyses.
All statistical analyses were based on measured herbicide concentrations. Mean percent inhibition in SGR and ΔF/F m ′ of each treatment relative to the control treatment was calculated as per Eq.
(3) 87 , where X control is the average SGR or ΔF/F m ′ of control and X treatment is the average SGR or ΔF/F m ′ of single treatments.
control t reatment control Nonlinear regression (Sigmoidal, 4-parameter) was used to produce concentration-response curves for each herbicide test (GraphPad Prism V 8.0.). Effective concentrations inhibiting ΔF/F m ′ and SGR by 10% and 50% with 95% confidence intervals (EC 10 /EC 50 ) relative to the control were interpolated from the equations of the curve fit. One-factor analysis of variance (ANOVA with replicates) was used to determine if there were significant differences (p < 0.05) in algal SGR rates and ΔF/F m ′ samples between various herbicide treatments. The relative potencies of the herbicides were determined using the relative equivalent potencies (ReP) compared to the reference herbicide diuron (EC 50 diuron/EC 50 herbicide) 23 . ReP values > 1 indicate potencies proportionally greater than diuron and ReP values < 1 indicate potencies less than diuron.
The estimation of no effect concentrations (NEC) was calculated in R (Version 3.6.1). The proportional decline in SGR (1-inhibition) was modelled as a function of log concentration of each herbicide using a Bayesian non-linear gaussian model using the R package jagsNEC 91 . This model has been specifically developed to derive no effect concentrations (NECs) but also allows the estimation of EC 10 and EC 50 values and is adapted from Fox 92 , and more generally defined by Eq. (4) 92 : i i i i i E[Y i |x i ] is the mathematical expectation of Y i (the response, e.g. in this case, the proportional decline in SGR) conditional on a given concentration x i . The model parameters for the generalized case are α (the response at zero or low concentrations, also called 'top'), −β (the rate of decay in the response after the NEC) and γ (the NEC value) 92 . For a gaussian Y, as used here, the model has the additional parameters Δ (an offset or intercept) and σ (the random error variance in Y). We used un-informative priors for the model parameters, including: α ~ dnorm(0, 0.1), β ~ dgamma(0.0001,0.0001), γ ~ dnorm(0, 0.01), Δ ~ dnorm(0, 0.1), and σ ~dunif(0, 29). Note www.nature.com/scientificreports www.nature.com/scientificreports/ that in jags dnorm is parameterized as a mean and precision (rather than mean and SD, as in R). Models were run with 10,000 Markov chain Monte Carlo (MCMC) iterations after an initial 'burn-in' period of 20000 iterations and for five separate chains. Trace plots were used to evaluate model fits and were found to have relatively good mixing in all cases.