Chronic toxicity of an environmentally relevant and equitoxic ratio of five metals to two Antarctic marine microalgae shows complex mixture interactivity

Metal contaminants are rarely present in the environment individually, yet environmental quality guidelines are derived from single-metal toxicity data. Few metal mixture studies have investigated more than binary mixtures and many are at unrealistically high effect concentrations to freshwater organisms. This study investigates the toxicity of five metals (Cd, Cu, Ni, Pb, and Zn) to the Antarctic marine microalgae Phaeocystis antarctica and Cryothecomonas armigera. Two mixtures were tested: (i) an equitoxic mixture of contaminants present at their single-metal EC10 concentrations, and (ii) an environmental mixture based on the ratio metal concentrations in a contaminated Antarctic marine bay. Observed toxicity, as chronic population growth rate inhibition, was compared to Independent Action (IA) and Concentration Addition (CA) predictions parameterised to use EC10 values. This allowed for the inclusion of metals with low toxicities. The biomarkers chlorophyll a fluorescence, cell size and complexity, and intracellular lipid concentrations were assessed to investigate possible mechanisms behind metal-mixture interactions. Both microalgae had similar responses to the equitoxic mixture: non-interactive by IA and antagonistic by CA. Toxicity from the environmental mixture was antagonistic by IA to P. antarctica; however, to C. armigera it was concentration-dependent with antagonism at low toxicities and synergism at high toxicities by both IA and CA. Differences in dissolved organic carbon production and detoxification mechanisms may be responsible for these responses and warrants further investigation. This study shows that mixture toxicity interactions can be ratio, species, and concentration dependent. The responses of the microalgae to different mixture ratios highlight the need to assess toxicity at environmentally realistic metal ratios. Parameterising IA and CA reference models to use EC10s allowed for the inclusion of metals at low effect concentrations, which may otherwise be ignored. Reference mixture models are generally suitable for predicting chronic toxicity of metals to these marine microalgae at environmentally realistic ratios and concentrations. (i) the toxicity of Cd, Pb, and Zn as chronic population growth rate inhibition to the Antarctic microalgae and armigera using an equitoxic and environmental mixture. Cellular biomarkers including chlorophyll a fluorescence intensity, changes to cell size and complexity, and relative intracellular lipid will also be investigated. (ii) the applicability of two reference models, and CA, predict chronic toxicity as microalgae population growth rate inhibition, parameterised to use EC10 values. (iii) metal-mixture interactivity to P. and C. armigera by IA and CA from the equitoxic and environmental


Introduction
Localised areas of Antarctica are contaminated as a result of historical and ongoing anthropogenic activities, particularly around research stations (Tin et al., 2009). Former waste disposal sites, abandoned work sites and stations, and fuel spills contribute to an estimated 1 million m 3 of uncontained contaminated pollution (Snape et al., 2001). Some of this pollution is mobilised from soils to the near-shore marine environment, generally by particle entrainment in summer meltwaters (Stark et al., 2006). This has resulted in elevated concentrations of a mixture of contaminants, including the metals Cd, Cu, Ni, Pb, and Zn in near-shore marine environments (Fryirs et al., 2015;Larner et al., 2006), causing toxicity to marine organisms (Cunningham et al., 2005).
To better understand how contaminants affect the Antarctic marine ecosystem, recent research has developed toxicity test protocols for a range of Antarctic and subantarctic organisms (Duquesne and Liess, 2003;Gissi et al., 2015;Hill et al., 2009;Holan et al., 2016;King and Riddle, 2001;Lewis et al., 2016;Marcus Zamora et al., 2015;Runcie and Riddle, 2007;Sfiligoj et al., 2015). These studies have only investigated the toxic response of organisms to single contaminants, and do not assess interactive effects of mixtures, as would be found in the Antarctic marine environment.
The toxicity of single metals may not reflect their toxicity in a mixture. Competition at cellular binding sites, changes to uptake pathways, and upregulation of detoxification mechanisms can influence the exposure and the response of an organism to metal mixtures (Duval et al., 2015). Recent efforts to understand and model the toxicity of metal mixtures has focused on freshwater organisms. We are only aware of two metal mixture studies conducted on a marine organisms -a marine mussel and a sea urchin (Deruytter et al., 2017;Manzo et al., 2010). There is therefore a deficiency in our knowledge about the application of metal-mixture models to marine and Antarctic organisms. This has left a significant gap in our understanding of metal mixture toxicity in marine systems.
Two reference models are widely accepted for the determination of metal mixture toxicity: Independent Action (IA) and Concentration Addition (CA) (Berenbaum, 1985;Jonker et al., 2005). Both models assume that the toxicity of the mixture components are non-interactive; but, have different approaches to combine individual contaminants' toxicities. IA calculates the toxicity of a mixture as the product of the individual contaminant's toxicity, making the assumption that they have dissimilar modes of action. The CA model expresses each contaminant of the mixture as a toxic unit (TU), calculated as the concentration of the contaminant in the mixture divided by a measure of its potency, such as a 10% effect concentration (EC10). CA assumes that contaminants within a mixture have a similar mode of action, and that contaminants are dilutions of each other. Both models rely on singlemetal toxicity data and assume no interactions between contaminants. However, deviations from model predictions are often observed. These interactions are described as: (i) synergism, where observed is greater than predicted toxicity, and; (ii) antagonism, where observed toxicity is less than predicted (Cedergreen, 2014). Mixtures are described as noninteractive where observed toxicity is predicted by the model.
The use of EC10 values as sensitivity estimates in ecotoxicology is growing because of its preferred use in the derivation of environmental quality guidelines Warne et al., 2014). This shift towards the use of EC10s may be because EC50 effects are too high to tolerate under a precautionary principle or EC50s are typically not environmentally realistic concentrations. Previous research has found that mixtures of contaminants at ≤EC10 may produce mixture interactions (Nys et al., 2017b). Yet, only a few studies have investigated environmentally relevant concentrations in chronic ecotoxicology studies (Nys et al., 2017b;Versieren et al., 2016).
Mixture models could help guide environmental assessment and decision making for sites impacted by contaminant mixtures in Antarctica. Under the Protocol on Environmental Protection to the Antarctic Treaty System, nations operating in Antarctica have a responsibility to ensure effective remediation of historical waste and to prevent further contamination (ATCM, 1991). Yet little progress in the remediation of historical waste has been made (Filler et al., 2015), with some exceptions (Errington et al., 2018;Snape et al., 2001;Stark et al., 2006), and routine monitoring of contamination and impacts is lacking (Braun et al., 2012;Hughes, 2010). A Clean-Up Manual has been developed to provide practical advice to national Antarctic programs about environmental management practices (CEP, 2013). However, this manual is yet to include environmental quality guidelines and lacks recommendations around the risk of contaminant mixtures. For temperate environments, CA is recommended as a good 'first tier' assessment of sites impacted by contaminant mixtures due to its generally conservative nature (Nagai and De Schamphelaere, 2016),which therefore tends to overestimate the toxicity of a mixture (Backhaus and Faust, 2012;Hochmuth et al., 2014;Versieren et al., 2016). Before existing contaminant mixture models can be applied to polar marine environments, however, research is needed to validate their suitability to polar marine organisms.
With only two metal mixture studies using marine species (using two invertebrates) reported to date, this study utilises organisms from an additional trophic level; marine microalgae. Microalgae are important components of the Antarctic marine ecosystem.
Phaeocystis antarctica (Phylum Haptophyta) is a common marine microalga associated with open marine waters and is known to be mucogenic (Rousseau et al., 2007). Cryothecomonas armigera (Phylum Cercoza) is a single-celled flagellated heterotrophic protist found in both Arctic and Antarctic marine waters, commonly associated with sea ice, and known to produce significant intracellular lipid stores (Kühn et al., 2000;Thaler and Lovejoy, 2012).
These species are both important components as primary producers in Antarctic marine food webs, and have different sensitivities to single metal contaminants (Gissi et al., 2015;Koppel et al., 2017).
This study aims to: (i) investigate the toxicity of Cd, Cu, Ni, Pb, and Zn mixtures as chronic population growth rate inhibition to the Antarctic microalgae P. antarctica and C. armigera using an equitoxic and environmental mixture. Cellular biomarkers including chlorophyll a fluorescence intensity, changes to cell size and complexity, and relative intracellular lipid concentrations will also be investigated. (ii) Assess the applicability of two reference models, IA and CA, to predict chronic toxicity as microalgae population growth rate inhibition, parameterised to use EC10 values. (iii) Investigate metal-mixture interactivity to P. antarctica and C. armigera by IA and CA from the equitoxic and environmental mixture.
Borosilicate 250 mL conical flasks used in tests were coated in a silanising solution (Coatasil, Ajax) to prevent metal adsorption. Plastic containers and consumables used were either new or acid-washed.
All chemicals were analytical grade or higher. Metal stock solutions were prepared in high purity water from metal salts, CuSO4, 3CdSO4.8H2O, PbCl2, NiSO4.6H2O, and ZnSO4 acidified to 0.1% v/v HCl.

Seawater
Seawater was collected from Oak Park, Cronulla on the east coast of Australia. Seawater was filtered to 0.22 or 0.45 μm, depending on the test species (Table 1) and stored in the dark at 4 °C. Background metal concentrations in the seawater were measured by inductively coupled plasma -mass spectrometry (see Metal analysis) and were 0.013 ± 0.002 µg Cd.L -1 , 0.88 ± 0.03 µg Cu.L -1 , 0.47 ± 0.02 µg Ni.L -1 , 0.82 ± 0.02 µg Pb.L -1 , and 3.9 ± 0.1 µg Zn.L -1 . Dissolved organic carbon concentrations were analysed by total organic carbon analyser after 0.45 µm filtration and found to be 1.4 ± 0.3 mg C.L -1 . Salinity (salinity and conductivity meter, model 30/10 FT; YSI), pH (measured at the start and end of each test, model 420, probe ROSS 815600; Thermo Fischer Scientific), and dissolved oxygen saturation (Oximeter 330;WTW) were measured with instruments calibrated as per the manufacturer's instructions.

Algal culture
Phaeocystis antarctica (strain number AAD 133) and Cryothecomonas armigera (strain number AAD 139) were obtained from the Australian Antarctic Division, Kingston, Tasmania.

Toxicity test protocol
Filtered seawater was used as the control/diluent water in toxicity tests. Metal exposure treatments were prepared in 80 mL of seawater, supplemented with 0.15 mg.L -1 PO4 3and 1.5 mg.L -1 NO3 -(as KH2PO4 and NaNO3, respectively) to maintain exponential growth of the algae during tests.
Two metal mixtures were tested: an equitoxic mixture and an environmental mixture. The equitoxic mixture is a combination of Cd, Cu, Ni, Pb, and Zn at concentrations equivalent to the population growth rate inhibition EC10s for each alga determined from single metal tests (Gissi et al., 2015;Koppel et al., 2017), given in Table 2   . The environmental mixture was tested at increasing multiples to produce a concentration gradient of metals fixed at an environmentally realistic ratio.
Algal cells in exponential growth phase (8-12 days for P. antarctica and 25-30 days for C. armigera (Gissi et al., 2015;Koppel et al., 2017)), were washed to remove culture growth medium by centrifugation and resuspension in cold filtered seawater three times. The algal concentrate was then used to inoculate flasks at the start of tests at a density of 1-3 x10 3 cells.mL -1 (day 0).

Measures of toxicity
A BD-FACSVerse flow cytometer was used with a 488 nm excitation laser and multiple detectors to quantify cell density and the fluorescence and light scattering of individual cells in solution following methods described in Franklin et al. (2005), with specific modifications for P. antarctica and C. armigera outlined by Gissi et al. (2015) and Koppel et al. (2017), respectively.
Population growth rate was determined by a linear regression of cell densities over multiple time points (at least 4) during the exposure. The population growth rate (specific growth rate, μ) was calculated from the regression of a plot of log10 (cell density) versus time (hours) and was converted to equivalent population doublings per day).
Biomarkers of algal health were determined by fluorescence and light scattering of individual cells, including: cell size (forward-angle scattering <15°), cell complexity (side-angle scattering 90°), chlorophyll a fluorescence intensity (700 ± 27 nm band filter), and relative intracellular lipid concentrations (green fluorescence from the intracellular stain BODIPY 493/503 measured with a 527 ± 16 nm band filter) (Kerker, 1983). For each parameter, the fluorescence intensity of exposed cell populations was compared to control cell populations by establishing three regions on a one-dimensional plot of the parameter of interest (e.g. forward-angle scatter, red 700 nm fluorescence). Control cells were used to define a healthy region, R2, which contained >95% of the control populations. Nonoverlapping regions of increased (R3) and decreased (R1) fluorescence intensity were established to monitor shifts in fluorescence in metal-exposed cell populations. Results are presented as percent of the cell population in the R2 region (Koppel et al., 2017).

Metal analysis
Subsamples (5 mL) of test solutions were taken from each flask at the start and end of each test (Table 1) Reported metal concentrations for each test treatment are the mean of initial and final dissolved metal concentrations.

Statistical analysis
Measures of toxicity were expressed as a relative effect (RE, Equation 1) as a percentage of control response, where y is the response of the exposure, and x̅ y control is the mean of the control response (Nys et al., 2016).

Equation 1
(% ) = y x̅ * 100 The analysis of the toxicity of metal mixtures uses previously published single-metal toxicity data for P. antarctica (Gissi et al., 2015) and C. armigera (Koppel et al., 2017). Data for P.
antarctica was reanalysed to determine EC10 values for population growth rate, cellular chlorophyll a fluorescence, size, and complexity using the statistical approach in the R environment outlined in Koppel et al. (2017).
Two reference models of mixture toxicity were tested to predict the toxicity of the metalmixture ratios:

(i) Independent Action
Independent Action was determined by Equations 2-5, where: yi is the response of the alga to metal i if it were exposed to a single metal at the same concentration found in the mixture. yi was determined by log-logistic model described by Equation 2, where xi is the concentration of metali, β is the slope parameter, and EC50 is the concentration of metali which causes a 50% reduction in the response variable yi. Independent Action response predictions were then determined as the product of the expected response of individual metals, shown by Equation 5.

(ii) Concentration Addition
Concentration Addition predictions (yCA) were determined by Equations 6 and 7 where xi is the concentration of metal i in the mixture, EC10i is the concentration of metali that causes a 10% inhibition to population growth rate, and ∑TUEC10 is the sum of the toxic unit contribution for each metal where each toxic unit is equivalent to a 10% reduction in population growth rate.

Equation 7
∑ ( 10 * 9 1 ) * ( The model parameters (i.e. EC10i and βi values for each metal) used in the reference mixture models were optimised prior to predicting metal-mixture toxicities as per the method given by Hochmuth et al. (2014). A random normal distribution of 40,000 sets of model parameters were generated about the means of each parameter. The mean and standard deviations of each EC10i and βi parameter were taken from the previously determined loglogistic model fits of single metal concentration-response curves for each microalga. Each parameter set was used to predict single-metal toxicity from existing single-metal toxicity datasets of P. antarctica and C. armigera. The parameter set that gave the lowest sum of squared error (SSE) was chosen to then predict the metal-mixture toxicities. The best parameters and SSE for each model are given in Supplementary information S1. The model predictions for single-metal concentration-response curves are given in Supplementary information S2 and S3 for P. antarctica and C. armigera, respectively.

Determination of significant mixture interactivity
To determine if there were significant synergistic or antagonistic deviations from the reference models, the IA and CA models were extended to include a deviation parameter 'a'. These are termed IASA and CASA and described by Equation 8 and 9, respectively (Hochmuth et al., 2014). To test if the deviation is significant, a nested F-test was conducted to determine whether the extended model provides for a significantly better fit than the reference model (Hochmuth et al., 2014;Jonker et al., 2005). (Jolley et al., 2018). The parameter pairs used in the reference models (IA and CA) and their extensions (IASA and CASA) are given in Supplementary information S1.

Metal mixtures
Equitoxic mixture The equitoxic mixture tested 5 metals at their population growth inhibition EC10 concentrations, to P. antarctica and C. armigera, respectively (Tables 2 and 3). Nickel was not toxic to P. antarctica in single-metal exposures but was included in this mixture to observe competition effects. The EC10 is increasingly used in place of the EC50 as a basis for environmental quality guidelines for single contaminants . For some contaminants the reliability of higher effect measures (such an EC50s) may be confounded by the solubility limit of metals in marine waters. For example, at a concentration of 1,800 µg Pb.L -1 , which is approximately the Pb solubility limit in seawater, the population growth rate of C. armigera was 75% of the control (Koppel et al., 2017). In such cases, these contaminants may contribute to toxicity or interactivity when present in a mixture and using their EC10 value then allows them to be included in reference models. This has been demonstrated as necessary, as mixtures with concentrations ≤EC10s have demonstrated mixture interactivity, such as synergism and antagonism (Nys et al., 2017b;Versieren et al., 2016).

Environmental mixture
The to any clean-up or remediation efforts (Snape et al., 2001).
The environmental mixture has a high Cu concentration relative to Cd, Ni, Pb, and Zn when compared against individual toxicity, i.e. Cu is the most toxic metal to both microalgae by 1 to 2 orders of magnitude with EC10s of 2.8 and 22 µg.L -1 to P. antarctica and C. armigera, respectively (Gissi et al., 2015;Koppel et al., 2017). Therefore, it was expected that Cu would be the main contributor to observed toxicity. Zinc was present at the highest concentration of all the metals in the ratio. However, it generally had a low toxicity to both algae in single-metal exposures, with EC10s of 217 and 366 µg.L -1 to P. antarctica and C.

Growth rate inhibition
The equitoxic mixture reduced population growth rate to 67 ± 8%, relative to the control treatment ( Fig. 1, Table 2). Nickel was not toxic in single-metal exposures but was included in the equitoxic mixture at 420 ± 70 µg.L -1 to investigate any contribution to mixture interactivity. There were no significant differences in the population growth rate, cellular chlorophyll fluorescence, or cellular size or complexity when compared to the equitoxic mixture treatments that excluded Ni (Supplementary information S4). Therefore, equitoxic mixture exposures with and without Ni were pooled for the analysis. The environmental mixture tested multiples of the ratio from 5 to 60 (Table 2). This mixture was toxic to population growth rate in a concentration-dependent manner, decreasing population growth rate to 10% at a multiple of 40.

Cellular chlorophyll a fluorescence
The equitoxic ratio resulted in an approximately equal proportion of the cell population with an increased and decreased fluorescence intensity; however, 81% of the cell population remained unaffected. The environmental mixture increased fluorescence intensity in a concentration-dependent manner (Fig. 1 B). In single-metal exposures to P. antarctica, Cu and Ni increased chlorophyll a fluorescence intensity, Pb decreased fluorescence intensity, and Cd and Zn caused greater variability (i.e. approximately equal increases and decreases).
The different fluorescence shifts suggest different modes of action of individual metals.
However, in single-metal exposures these were generally low responses; <30% difference to the response in control populations at all concentrations tested. The increased variability observed in response to the equitoxic mixture could suggest non-interactive toxicity, but the relative contributions of each metal and underlying mechanism is unknown. The environmental mixture reflects a Cu-only response with a greater magnitude of shift e.g.
70% population shift at a multiple of 20 with a Cu concentration of 14 µg.L -1 which was the chlorophyll a fluorescence EC50 in single-metal exposures (Supplementary information S5).

Cell size and complexity
The equitoxic mixture increased the size of cells in 17% of the cell population and decreased complexity in 11% of the population. The environmental mixture increased the variability of cell size at multiples ≥5, which was coupled with a concentration-dependent increase to complexity in 100% of the cell population (Fig 1. C and D).
In single metal exposures, the complexity and size of P. antarctica cells were affected by Cd and Cu, but not by Ni, Pb, or Zn. Exposure to Cu ≥6 µg.L -1 increased the range of cell sizes observed and increased complexity in a maximum of 50% of the cell population (Supplementary information S5). The single-metal Cu response is consistent with the response to the environmental mixture; however, the magnitude of the response is much greater in the environmental mixture at comparable single Cu exposure concentrations.

Growth rate inhibition
The equitoxic mixture reduced population growth rate to 54 ± 3%. For the environmental mixture, the population growth rate was unaffected at multiples below 30; above this there was a concentration-dependent decrease of the growth rate to 4 ± 5% at a multiple of 100 (Table 3, Fig. 2 A).

Cellular chlorophyll a fluorescence
The equitoxic mixture showed approximately equal proportions of the cell population with increased and decreased fluorescence intensity, with a combined 26 ± 4% of the population affected. The environmental mixture increased chlorophyll a fluorescence in up to 30% of the population until a multiple of 50. At multiples greater than this, 62% of the cell population had decreased florescence at a multiple of 100 (Table 3, Fig. 2, B).
In single-metal exposures, Cd, Cu, Ni, and Pb increased fluorescence intensity, while Zn

Cell size and complexity
The equitoxic mixture had a small effect on cell size and complexity, with 10% of the cell population affected (Table 3, Fig. 2 C and D). In exposures to the environmental mixture, size increased in 20% of the population and complexity decreased in 30% of the population ( Fig. 2 C and D).
Single-metal treatments (except Cd) affected the size of C. armigera, with Ni, Pb, and Zn reducing size in a maximum of 10% of the population, while Cu increased cell size in 70% of the population (Koppel et al., 2017). However, only Cd reduced cell complexity, with an EC10 of 269 µg.L -1 (Supplementary information S5). The equitoxic mixture had concentrations of Cd, Cu, and Ni that were expected to reduce cell size, and a concentration of Cd expected to reduce cell complexity based on single-metal exposures. However, this was not observed in the equitoxic mixture. Exposure to the environmental mixture increased size in a population of cells roughly equivalent to the population with decreased complexity. These changes could be related, as an increased size in the absence of intracellular changes could lead to a perceived decreased complexity. The mechanism behind the change in size is unknown but could be due to differences in osmotic pressure resulting from changes to membrane permeability (Jamers et al., 2009).

Intracellular lipid concentrations
The equitoxic mixture resulted in a reduction of intracellular lipid concentrations, with 27 ± 11% of the exposed population having decreased green fluorescence compared to control populations ( Fig. 2 E). Despite having lower metal concentrations, the environmental mixture resulted in a greater reduction of lipid concentrations. However, this was not concentration dependent with 40-60% of the cell population having reduced lipids at multiples >10.
In single-metal exposures, Pb > Cu > Cd elicited the greatest effect on lipids, with EC10s of 11, 33, and 89 µg.L -1 , respectively. Changes to lipid concentrations were a more sensitive measure of toxicity than changes in growth rate for Pb and Cd. The response to the equitoxic ratio was less than expected based on the concentrations of Cd, Cu, and Pb.
Curiously, there were no differences to lipid concentrations in exposures of equitoxic mixtures where Cd was excluded or included at 50 µg.L -1 or 500 µg.L -1 (Supplementary information S6), despite Cd showing high lipid-reducing toxicities at these concentrations in single-metal exposures (Koppel et al., 2017).
The environmental mixture elicited unexpected results. At a multiple of 50, only Cu and Zn concentrations are equivalent to their single-metal intracellular lipid EC10 (i.e., no toxicity to lipid concentrations was expected at lower multiples, and less toxicity was expected in the 50x multiple compared to the equitoxic mixture). This may be explained by cellular detoxification and division. Where cell division is not greatly inhibited, lipids may be diluted as cells divide, especially when coupled with energetically expensive detoxification mechanisms (Farese and Walther, 2009;Lavoie et al., 2016). That is, at low toxicities (normal growth rates) lipid concentrations may be most affected. However, this wasn't observed, with variable but roughly equivalent reductions at all multiples >10 (Fig. 2 E).

Equitoxic mixture
The equitoxic mixture had metals at concentrations that summed to ∑TUEC10 of 4.5 ± 0.8 for P. antarctica and 6.63 ± 0.04 for C. armigera, which resulted in population growth rates of 67 ± 8% for P. antarctica and 54 ± 3 for C. armigera, Fig. 3. Both microalgae had similar responses to the equitoxic mixture; observed toxicity was well predicted by IA while CA overestimated toxicity, i.e. was antagonistic. The antagonism as measured by CA was significant for both algae, but the power of the Tukey's HSD test for C. armigera was low due to the low number of replicates, n=3. IA calculates mixture toxicity from individual contaminants under the assumption of independent modes of action, whereas CA considers contaminants to have the same mode of action. The equitoxic mixture specifically investigated how metals may interact at concentrations where they are known to exert equivalent toxicity. That IA better predicted toxicity than CA suggests that the metals have a dissimilar mode of action, rather than joint modes of toxicity (Berenbaum, 1985).

(a) (b)
Observed CA IA Observed CA IA

Environmental mixture
Observed and predicted toxicities from the environmental mixture to the population growth rates of P. antarctica and C. armigera was concentration-dependent and well described by a log-logistic model (Fig. 4). Both models gave similar predictions to both algae; however, each alga had a different response. Antagonism to P. antarctica and concentrationdependent mixture interactivity to C. armigera, where there was antagonism at low and synergism at high effect concentrations (Fig. 4 A and B, respectively).
The only significant mixture interaction (p < 0.05) for the environment metal mixture ratio was antagonism measured by IA to P. antarctica (p = 3.37 x10 -5 , Supplementary information S1). No other interaction was significant for either model or algae. Antagonism by CA to P.
antarctica was not significant despite its predictions being very similar to IA.
The response to C. armigera was not significant, possibly because the mixture interactivity was concentration-dependent. That is, significant mixture interactivity at high or low concentrations is masked when the response across the whole concentration range is assessed (Fig. 4 B). In such cases, a full factorial experimental design (rather than individual ratios) or a different statistical approach assessing concentration-dependent mixture interactivity may be needed (Deruytter et al., 2017;Nys et al., 2015). While the synergism by IA and CA to C. armigera at high toxicities was not determined to be significant, similar responses are well documented with freshwater organisms (Cedergreen, 2014;Nys et al., 2017bNys et al., , 2015. However, this is not always the case, with Ni recently found to induce synergism at low concentrations and antagonism at high concentrations when in a binary mixture with Cu (Deruytter et al., 2017).
It was expected that Cu would cause the most toxicity, given it is the most toxic single metal to both microalgae and present in the ratio at the second highest concentration. Zinc was less toxic to both algae but was present at 4.5 times the concentration (in µg.L -1 ) of Cu. To both algae, exposures ≥20 of the environmental mixture had Zn concentration within the 95% confidence interval of the single-metal Zn EC10. To P. antarctica this caused antagonism, but to C. armigera this led to concentration-dependent interactions with antagonism up to a 50% effect, beyond which synergism was observed.
Cellular biomarkers give clues to the mechanisms behind these results. The changes to chlorophyll a fluorescence in C. armigera were coincident with the mixture interactivity shift from antagonism to synergism in population growth rate. The fluorescence shifts suggest toxicity from Cu at multiples <50 and Zn at multiples >50, thus Zn may be antagonist at low concentrations but synergistic at high concentrations. However, changes to cell size and complexity suggest Cd toxicity, possibly affecting membrane permeability. So, although Cu is the most toxic and Zn the most abundant, the influence of Cd, Ni, and Pb cannot be discounted. Especially considering in the equitoxic mixture they were non-interactive by IA and antagonistic by CA.
The antagonism observed to P. antarctica may be due to the production of dissolved organic carbon or detoxification mechanisms as inferred from cellular biomarkers. P. antarctica is known to produce dissolved organic carbon to increase Fe and Zn uptake in the Southern Ocean (Alderkamp et al., 2012;Saito and Goepfert, 2008). Dissolved organic carbon is known to bind strongly to Cu, effectively reducing its toxicity (Wood et al., 2011). This may explain the observed antagonism in the environmental mixture. In contrast, less is known about the dissolved organic carbon production of C. armigera, but it is not known to be mucogenic.
A large proportion of P. antarctica had increased cellular complexity at low concentrations of the environmental mixture, compared to C. armigera ( Fig. 1 D compared to Fig. 2 D). This could indicate detoxification of metals by intracellular sequestration in P. antarctica.
Previous research has shown Cu can be sequestered by polyphosphate bodies or phytochelatins to intracellular bodies, which would be observed as an increase to cell complexity (Adams et al., 2016;Levy et al., 2008). Such a response wasn't observed in C.
Different chlorophyll a fluorescence changes in the microalgae suggest different metals are causing toxicity. P. antarctica had a greater proportion of cells affected than C. armigera, and the fluorescence shift was different. P. antarctica had a consistent increase, while C.
armigera had an increase up to multiples of 50 followed by a decrease in fluorescence intensity. The increase in fluorescence intensity suggests P. antarctica and C. armigera had an impaired electron transport chain in the photosynthesis pathway (Guo and Tan, 2015), possibly a result of Cu toxicity. The decrease in fluorescence intensity in C. armigera suggests Zn toxicity, possibly impairing the light harvesting apparatus of chlorophyll moieties (Küpper and Andresen, 2016). These changes in biomarkers indicate different physiological responses to metal mixtures and give insight into mechanisms behind these differences.

Metal mixture modelling for environmental management
Studies of the toxicity of metal mixtures are predominately freshwater based, with few studies investigating mixture toxicity to marine organisms, such as the mussel Mytilus edulis (Deruytter et al., 2017) or sea urchin Paracentrotus lividus (Manzo et al., 2010). The response of freshwater organisms may not be comparable to those in seawater; however, the metal-organism interactions that cause toxicity may be similar. These include complex specific-and non-specific uptake processes, metal-specific modes of toxicity, specific or general detoxification mechanisms, and rapid homeostatic regulation of cellular physiology (Cedergreen, 2014;Versieren et al., 2017). The mixture interactivity found in this study is largely within the range observed in freshwater microalgae.
Most metal-mixture studies only investigate binary metal mixtures. For example, in freshwater studies to green microalga Zn has shown to protect against Cd to Chlamydomonas reinhardtii (Lavoie et al., 2014) and Pb to Micrasterias denticulate (Volland et al., 2014). While Cd has been shown to be synergistic in the presence of Cu to Chlorella sp. (Franklin et al., 2002). This study found that the equitoxic mixture was non-interactive by IA and antagonistic by CA and the environmental mixture was found to elicit more complicated interactions which were microalgae dependent. More research is needed to investigate more complicated metal mixtures, which may better represent environmental contamination.
Previous studies consistently find that CA tends to overestimate toxicity (Nys et al., 2018).
This has been observed both in metal mixture studies (Nagai and De Schamphelaere, 2016;Nys et al., 2017aNys et al., , 2015 and in studies with organic contaminants (Cedergreen et al., 2008).
As a result, CA is recommended as a conservative first-tier screening model for the environmental management of contaminant mixtures (Cedergreen, 2014;Nys et al., 2018).
This trend was observed in the equitoxic mixture, but not in the environmental mixture, where both models gave largely equivalent predictions to the environmental mixture (see SSEs in Supplementary information S1). At low concentrations (<1 TU), the environmental mixture has a more realistic ratio and metal concentrations than the equitoxic mixture.
Thus, either IA or CA should be suitable to predict toxicity to P. antarctica or C. armigera in environmentally realistic conditions.

Conclusion
This study demonstrated that mixture interactivity of five metals could be ratio, concentration, and microalgae specific. Adapting IA and CA mixture reference models to use EC10 values as parameters allowed for the inclusion of metals which individually only exhibited low toxicities. This will be beneficial for future studies as the use of low-effect toxicity values such as EC10s, particularly in environmental quality guideline derivation, grows.
The equitoxic mixture was non-interactive by IA and antagonistic by CA to both P. antarctica and C. armigera. This implies a greater likelihood that the contaminants have dissimilar modes of action. The environmental mixture was antagonistic by IA and CA to P. antarctica, with IA being a significant interaction. To C. armigera, the environmental mixture was concentration-dependent with antagonism at low effect concentrations and synergism at high effect concentrations by both models. Mixture interactions in both P. antarctica and C. armigera were observed concurrent to changes in biomarkers, including chlorophyll a fluorescence, cell complexity, and lipid concentrations. These biomarkers give insights into toxicity and detoxification mechanisms.
Frameworks designed to manage the environmental risk of metal mixtures typically recommend a 'first-tier' screening using a CA model (Backhaus and Faust, 2012;Farley et al., 2015). This study showed either models are suitable at environmentally realistic ratios and concentrations of metal mixtures, where there were equivalent predictions from both models to both microalgae. However, a better understanding of mixture interactivity to other Antarctic marine organisms is still needed guide environmental management.