Thermal plasticity in coral reef symbionts is mediated by oxidation of membrane lipids

The oxidation of polyunsaturated fatty acids (PUFA) is a common stress response across biomes with potential to trigger impairment of cell growth and reproduction. The oxidative stress theory of coral bleaching induced by global warming has been widely accepted to explain coral reef decline, but its underlying physiological mechanism remains under debate. Here we used lipidomic and population density data to examine cell cultures of three coral reef symbionts after a heat shock (sudden rise of 12 °C for 4 hours). Heat tolerance in S. microadriaticum and C. goreaui was characterized by preservation of thylakoid-derived glycolipids. Conversely, heat sensitivity in B. minutum was linked to elevated concentrations of oxidized PUFA esteried to glycolipids, suggesting that culture growth had ceased due to severe oxidative damage. Our ndings provide a basis to further understand the role played by oxidative stress in coral bleaching and reveal novel biomarkers for the monitoring of symbiont-coral health. indicate that the oxidation of PUFA in thylakoid membranes may explain cellular damage in Symbiodiniaceae under thermal stress and provide novel insights into the current knowledge of the oxidative stress theory of coral bleaching. We suggest that oxidation of PUFA in MLs may cause conformational alterations in energy transducing membranes, leading to energy limitation, impairment of cell growth and reproduction. This stress response is a common feature across biomes with the potential to trigger several morbidities ranging from neurodegenerative diseases in humans to worldwide coral reef decline. The biomarkers presented here have potential for the monitoring of coral health, and for improving the strategies for coral reef conservation. Lipidomics and metabolomics may be used in combination with other “omics” approaches (i.e., genomics, transcriptomics and proteomics) to better understand how the genetic code is interacting with cellular physiology and expression of phenotypic traits. An improved interaction between these areas will bring new insights for understanding symbiont-coral host cellular communication, as well as better predict coral bleaching events in warmer future oceans.


Introduction
Lipid composition of energy transducing membranes (e.g. cytoplasmic membranes of bacteria, thylakoid membranes of chloroplasts and mitochondrial inner membranes in eukaryotic cells) is pivotal for the survival of unicellular to complex organisms 1,2 . Energy balance in living cells is highly dependent on the e ciency of lipid membranes controlling the permeability of ions and optimizing electron transport at the membrane level, which creates proton gradients enabling mechanical energy output in the form of adenosine triphosphate (ATP) 3,4 . This process is mediated by altering the membrane uidity and viscosity that can be regulated by the length and saturation levels of fatty acid chains of different polar lipids 1,2,5 .The adjustment of membrane lipids composition in response to external abiotic stressors was suggested to be universally mediated by membrane homeoviscocity [5][6][7] , which assures optimal cellular functions and interlinks lipids with bioenergetics 1,2,4,8 .
Membranes enriched in polyunsaturated fatty acids (PUFA) are more likely to be oxidized by reactive oxygen species (ROS), especially free radicals 9,10 . If not contained by the cellular antioxidant machinery (e.g. enzymes and carotenoids), ROS can generate lipid radicals, which might propagate membrane lipid oxidation [10][11][12] , unbalancing proton gradients and thus disturbing the energy balance of the cell.
Oxidation of energy transducing membranes by ROS and the resulting lower concentrations of PUFA are suggested to impair distinct biological and ecological processes because of energy limitation, as shown in the inhibition of plant and algal growth 13,14 , coral bleaching [15][16][17][18] , and human neurodegenerative diseases 19-21 . Coral bleaching is characterized by physiological impairment of the symbiosis between cnidarian hosts and their algal symbionts (Symbiodiniaceae) and/or by loss of symbionts' photosynthetic pigments 22 , potentially triggered by excessive ROS generation in the symbionts and/or the host 18 . Higher temperatures lead to increased uidity of highly unsaturated thylakoidal membranes 23 and may cause leakage of high-energy electrons from the water splitting reaction at the photosystem II (PSII) in Symbiodiniaceae 24 . It was suggested 15,16,25 that this process leads to increased production of ROS and cell damage in the symbionts. Besides increasing temperatures, excessive light stress can also intensify the production of ROS via generation of singlet oxygen 26 . Symbiodiniaceae thylakoid membranes are enriched in glycolipids linked to PUFA 27 . The greater number of double bonds compared to mono (MUFA) and saturated (SFA) fatty acids in glycolipids increase their susceptibility to ROS 10 . Several studies have suggested that Symbiodiniaceae thermal tolerance and the bleaching susceptibility of their coral hosts were de ned by saturation levels of the symbiont's thylakoid membrane bond fatty acids 17,28-30 . However, this possible mechanism was inferred from the analysis of bulk fatty acid composition 17,28,30 or from a limited number of lipid classes 29,[31][32][33] . The oxidative stress theory provides a reasonable concept to explain coral bleaching leading to worldwide coral decline 34 , but up to this point, a lipid-based molecular explanation of thermal stress leading to thylakoid membrane oxidative damage is still missing.
Here we show a time-dependent population density and a comprehensive untargeted lipidomic analysis (i.e., lipids, quinones and pigments) of a heat shock experiment (sudden rise of 12 ˚C for 4 hours) with three different species of coral reef symbionts: Symbiodinium microadriaticum, Breviolum minutum and Cladocopium. goreaui. These species are broadly associated with coral hosts 35 and were previously reported to display distinct thermal tolerances 36,37 Our detailed monitoring of lipid molecular species enabled us to identify biomarkers associated with physiological acclimation strategies regarding extreme heat stress and evaluate whether the cellular damage caused by heat stress was linked to membrane lipid oxidation.

Results
A comprehensive untargeted lipidomic assessment was performed under standard culture conditions (22 C; see methods section for details) to establish a baseline for the three Symbiodiniaceae species: S. microadriaticum (ITS2 type A1), B. minutum (B1) and C. goreaui (C1). A total of 276 compounds were identi ed and sorted into seven classes (i.e. pigments, membrane lipids as glycolipids, aminolipids, phospholipids and shingolipids; storage lipids and others) ( Table 1; Fig. S1), to differentiate their presence among distinct cell compartments. The most signi cant difference in abundance of lipid classes among the three species was observed in the concentration of membrane lipids (ML) (1 way -ANOVA F ssp. = 51.22; p < 0.01; Table S2) and their esteri ed fatty acids, which was up to 40% lower in B. minutum than in the other two species ( Fig. 1a and b). Most of the lipid compounds (75 out of 86) that signi cantly differed among the three species were ML components (Fig. S2). Of these 75 compounds, 43 of them were esteri ed to omega-3 fatty acids, either octadecatrienoic acid (18:4n-3), octadecapentaenoic acid (18:5n-3) or docosahexaenoic acid (DHA; 22:6n-3). Glycolipids were predominantly linked to 18:4 and 18:5, whereas phosphatidylcholine (PC) and 1,2-diacylglyceryl-3-(Ocarboxyhydroxymethylcholine (DGCC) were mostly linked to DHA (Table S1). The amounts of membrane bond oxidized PUFA (oxy-PUFA) did not differ signi cantly among the three species at baseline (Fig.1c).
To assess the in uence of heat shock on population density, cultures were sampled over time for both heat shock and control groups of each symbiont. Biomass sampling for lipidomics was performed immediately after the four-hour long heat shock (T 4 ), then 24 hours later (T 28 ) and nally at the end of the experiment at 240 hours (T 244 ). Even though the control of S. microadriaticum had low population density number at T 244 , no signi cant differences among the growing control cultures of the three species and the heat shock samples of C. goreaui were noticed (Table S3). On the other hand, heat shock samples of both S. microadriaticum and B. minutum showed an initial reduction to about half of their population, compared to control at T 4 , thus their growth curves were signi cantly affected (ANCOVA F HS [Spp] = 26.29; p < 0.01; Table S3) (Fig. 2). The only symbiont that fully ceased growth was B. minutum, indicated by the constant decline in its population density after heat shock (Fig. 2).
Overall alterations in lipidome pro les after heat shock Thermal susceptibility was de ned by changes in the population densities of each symbiont and by the intensity of ML remodeling, where the largest differences, those with ANOSIM R (p < 0.01) values approaching 1, can be visualized using the multidimensional scaling plot (MDS) (Fig. 3a). Breviolum minutum drastically changed its membrane lipidome (R = 0.75) with no recovery at 24 (T 28 ) nor even 240 (T 244 ) hours after heat shock (Fig. 3a), which suggest irreversibly damaged to its cells and/or cell death at T 244 . Symbiodinium microadriaticum was an intermediate heat tolerant (R = 0.73) and C. goreaui was the most heat tolerant symbiont (R = 0.68). Differences among treatments of both S. microadriaticum and C. goreaui were within 30 Bray-Curtis similarity indexes, independent of heat shock treatment at all monitored times. The same similarity was shared only by the control of B. minutum (c.f. Fig. 3a). Glycolipids, plastoquinone and pigments, together with cardiolipin, PC and the total sum of oxy-PUFA esteri ed to glycolipids and DGCC were the ML classes that exhibited most variation among Symbiodiniaceae after heat shock according to principal component analysis (PCA; Fig. 3b). The greater distancing of heat shock treated samples of B. minutum indicated in both PCA and MDS (Fig. 3) also suggests that this heat sensitive symbiont suffered the greatest lipidome alterations. DGCC, cholesterol and other membrane components were less robust in explaining heat shock induced lipidome alterations in the Symbiodiniaceae species (cf. Fig. 3b, detailed on Table S1).
In S. microadriaticum, the concentrations of glycolipids ( Table S5) was only validated by a 2-way ANOVA species-speci c analysis representing a drop of 35% compared to the control samples. The concentration of oxy-PUFA esteri ed to glycolipids was uniquely lower than control at T 4 (F Ti = 27.61; p < 0.01; Table S5) ( Fig. 4). Cladocopium goreaui also had the lowest oxylipins content (F Spp. = 8.16; p < 0.01; Table S4) among all symbionts, independently of heat shock (Fig.4). Lysolipids (Fig. 5c), cholesterol and sphingolipids (Fig.S3) concentrations remained unchanged. Storage lipids concentration increased in C. goreaui at T 244 and the greatest change was observed for PUFA acyl chains esteri ed to TAG (F HS. [Spp.] = 6.62; p < 0.01; Table S5), which were three times greater than the control samples (Fig. S4b).

Discussion
Monitoring population densities and distinct lipid molecular species in coral reef symbionts after an acute heat shock enabled us to formulate different underlying cellular mechanisms regarding heat tolerance. Our ndings also highlight that cellular damage caused by heat stress was associated with oxidation of membrane lipids, particularly glycolipids that are a major structural component of thylakoid membranes 23 . The physiological strategies adopted to withstand heat shock exposure in each heat tolerant symbiont (i.e., S. microadriaticum and C. goreaui) were species-speci c and characterized by intensive lipidome remodeling. In contrast, the heat sensitivity in B. minutum was represented by ceased growth and a lipidome pro le indicative of thylakoid membrane damage, characterized by the reduction in glycolipids, chlorophyll-a as well as the PSII electron transporter, plastoquinone. The oxidation of membrane lipids in B. minutum was further evidenced by an increase in the concentration of oxy-PUFA esteri ed to glycolipids and total amounts of lysolipids and oxylipins after heat shock. Not surprisingly, the greatest concentration of these biomarkers appeared at the end of the experiment in B. minutum when its population density was the lowest.
The protection of energy transducing membranes against lipid oxidation can be mediated by the e ciency of antioxidant machineries (e.g. antioxidant enzymes 38 ). Breviolum minutum has been shown to have a less e cient enzymatic antioxidant machinery than S. microadriaticum and C. goreaui when exposed to heat stress 39 , which is consistent with the reduction in its population density after heat shock. Cladocopium goreaui has been reported to present the highest levels of superoxide dismutase and peroxidases (putative antioxidant enzymes that avoid propagation of superoxide and peroxide ions, respectively) among the three investigated species [39][40][41] . This is consistent with our ndings that revealed C. goreaui as the most tolerant symbiont (Fig. 2), the one with the lowest concentrations of both, oxy-PUFA esteri ed to MLs (Fig. 4) and oxylipins (Fig. 5d). Population density of C. goreaui did not change signi cantly after heat shock, but the concentrations of most of its MLs (except glycolipids), chlorophyll and plastoquinone were reduced when compared to control (Figs. 4 and S3). Lipidome data at T 244 indicated that some membrane fatty acids, especially PUFA, could have been reallocated to storage lipids ( Fig. S4), suggesting an energy saving strategy 21,42 and/or a mechanism of defense against oxidative stress 43 for preserving the thylakoid membranes speci c glycolipids in this symbiont. Symbiodinium microadriaticum has been reported to have lower basal antioxidant enzyme levels than C. goreaui, though it can upregulate the production of antioxidant enzymes 39 . This heat tolerant symbiont also increased the concentration of some classes of MLs after exposure to high temperatures (Figs. 5 and S3). Both strategies (synthesis of antioxidants and ML) are energetically costly and might result in relocation of available energy, jeopardizing other metabolic needs such as growth, which might explain why this symbiont was still able to recover but had its population density signi cant lower than C. goreaui. The initial drop in the population density of S. microadriaticum after heat shock (T 4 , Fig. 2) could have selected more thermal tolerant strains. However, it remains to be tested if the persistent population can keep the selected phenotypic traits after additional iteration of heat shock events, which would either suggest the selection of a certain heat resistant genotype or a phenotypic cell memory capacity. In the case of C. goreaui, where no signi cant drop in population density but an intense lipid remodeling was observed, a preserved phenotypic memory response is more likely and could be an underlying mechanism to cope with abrupt changes in the natural environment 44,45 . For instance, the apex of heat stress during reoccurring El Niño events over the past decades has the potential to either select for certain genotypes with higher thermal tolerance or select for strains with a higher phenotypic cell memory capacity. Thus, symbionts that represent such heat resistant traits are more likely to thrive in future warmer oceans 46 . The e ciency of these phenotypic traits for the metabolic exchanges within coral hosts in the case of an established symbioses remains to be explored and this might be a crucial point to predict the success of coral reefs in the future.
We considered the effects of heat stress on MLs and its link to coral bleaching using an untargeted lipidomics approach. The same subject has been investigated over the past two decades 17,[28][29][30][31][32][33] including studies that indirectly measured lipid oxidation and production of free radicals 47 . However, in the absence of analytical biomarkers of oxidative stress, we believe that some conclusions may be misleading. For example, Tchernov and colleagues (2004) 17 suggested that symbionts with higher concentrations of PUFA in thylakoid membranes were more susceptible to heat stress. However, in our study we demonstrate that heat tolerance/susceptibility in Symbiodiniaceae cannot be attributed to membrane saturation at baseline (see Fig. 1). Instead, our data revealed that thermal tolerance is linked to the capacity of maintaining membrane homeoviscocity stability by preserving the structural MLs (e.g. from oxidation; Fig. 3a and Fig. 4). In energy transducing membranes, higher unsaturation levels enable fast electron ow leading to high energy production 7 . However, such e ciency in energy transduction provided by highly unsaturated MLs is only possible with either an extremely constant environment or a concurrent evolution of powerful antioxidant machinery to preserve membrane PUFA against oxidation 1,2,9,10,38 . The latter was not the case for B. minutum. When ROS propagation cannot be contained, formation of lipid radicals leads to a chain reaction (Fig. 6). The addition of an oxygen-carbon bond in Oxy-PUFA may reorient the acyl chain, whereby it no longer remains in the membrane interior, but rather protrudes into the aqueous compartment 9 . Hydrolysis of the formed Oxy-PUFA leads to subsequent generation of lysolipids and oxylipins (Fig. 6). Higher concentration of lysolipids causes structural changes in membrane curvature and elasticity 48 , which unbalances the stability of ion channels 49 . Such alterations may also cause proton leakage 50 unbalancing nal energy output, which might inhibit cell growth and division due to a lack of energy (Fig. 6). HDoHEs were mediated enzymatically remains to be tested. We detected oxidation via singlet oxygen 55 (12-HODE, 10-HODE and 19-HDoHE), as previously reported for plants 56,57 . The great diversity of isomers showed in our study (see Figs. S5 and S6), lacking preferential generation of positional isomer type in the heat shock samples compared to controls, suggests lipid oxidation predominantly mediated by free radicals. Oxylipins and oxidized fatty acids esteri ed to MLs were present in all samples (including controls -Figs. 1c; S5 and S6), which is consistent with their pivotal role in biological systems as signaling molecules that accumulate as a function of abiotic and/or biotic stress 58,59 such as higher temperatures 14 . Oxidation-derived products in coral hosts (e.g., eicosanoids, prostaglandins and aldehydes) has shown to mediate symbiont-host communication 60 ; thus, symbiosis impairment might be a nal consequence of the accumulation of these signaling molecules in response to stress.
The concentration of sphingolipids observed in B. minutum after heat shock was the highest at the end of the experiment when compared to the control samples (Figs. 2 and S3d). Accumulation of sphingolipids in response to heat stress has been reported across all life forms from unicellular bacteria and yeast to plants and mammals, as a strategy to adjust membrane uidity and permeability 61 . Furthermore, cell death through apoptosis, such as proposed for mitochondria under ROS damage, is known to be linked to upregulation in sphingolipids such as ceramides 62,63 . These alterations accompanied by dysregulation in cholesterol homeostasis (also noticed in B. minutum; Fig. S3c) are molecular signatures in aging and during evolution of neurodegenerative diseases in humans (i.e., Parkinson's disease 19 , Alzheimer 20 and amyotrophic lateral sclerosis 21 ). In the marine environment, apoptosis mediated by mitochondrial damage in cnidarian host cells was proposed to cause impairment of symbiont-host association and to promote bleaching 64 . Whenever bleaching is controlled by the symbionts or the host, thylakoid and mitochondria membranes' bioenergetics and lipidomes must be preserved to assure the success of this relationship.
Our ndings indicate that the oxidation of PUFA in thylakoid membranes may explain cellular damage in Symbiodiniaceae under thermal stress and provide novel insights into the current knowledge of the oxidative stress theory of coral bleaching. We suggest that oxidation of PUFA in MLs may cause conformational alterations in energy transducing membranes, leading to energy limitation, impairment of cell growth and reproduction. This stress response is a common feature across biomes with the potential to trigger several morbidities ranging from neurodegenerative diseases in humans to worldwide coral reef decline. The biomarkers presented here have potential for the monitoring of coral health, and for improving the strategies for coral reef conservation. Lipidomics and metabolomics may be used in combination with other "omics" approaches (i.e., genomics, transcriptomics and proteomics) to better understand how the genetic code is interacting with cellular physiology and expression of phenotypic traits. An improved interaction between these areas will bring new insights for understanding symbiontcoral host cellular communication, as well as better predict coral bleaching events in warmer future oceans.

Experimental setup
Symbiodiniaceae cultures (S. microadriaticum -ITS2 A1; B. minutum-B1 and C. goreaui -C1) were received from the University at Buffalo (NY-USA) and kept in the BMAK microalgae facility at the Instituto Oceanográ co from Universidade de São Paulo (IO-USP). The cultures of each symbiont were replicated equally into six 1L Erlenmeyer bottles (heat shock and control samples were performed in triplicates) containing sterile natural sea water with f/2 nutrient conditions 65 . Bottles were randomly distributed in a water bath maintained at 22 ºC. Sterilized temperature sensors (± 0.5 ºC) were added inside the cultures of each bottle to monitor temperature throughout the experiment. Cool uorescent lights were used, and light availability kept constant at 80 µE.m -2 s -1 in a 12:12 hours light: dark cycle. Initial cell number in each bottle was 500 cells.ml -1 and population densities were monitored over time (Fig. 2) using a Neubauer counting chamber. Heat shock was administered once all cultures were in exponential growth (240 hours of cultivation). A separate water bath was used and pre-warmed to 34 ºC using electrical aquarium heaters (± 0.5 ºC). Three bottles of each symbiont (nine in total) were transferred from 22 ºC to the pre-heated water bath container where a temperature of 34 ºC was reached after 20 min (T 0 ). Subsequently, the 34 °C heat shock period was applied for 4 hours. Samples were immediately taken at the end of heat shock period (T 4 ) to evaluate short-term lipidome alterations between the heat shock and control populations. Afterwards, culture bottles were transferred back to the 22 ºC water baths. Both treatment groups were sampled again after 240 hours (T 244 ) for analyses of long-term changes in lipidomes. Additional samples were collected from the heat shock populations one day after the event (T 28 ) to describe potential rapid recovery of lipidome pro les after a return to the initial temperature.
Bottles were slowly hand rotated for homogenization before sampling. No additional analysis was made to guarantee that only living cells were sampled in each culture bottles. All samplings occurred at the same time during the light phase. Samples for lipid and pigment analysis were taken with sterile pipets and subsequently ltered onto pre-combusted GF/F lters (10 min under 300 °C). Cell numbers exceed 1 million per lter and the true number of cells per lter was then calculated from the cell density in the remainder of the cultures after ltration on the sampling days.

Standards
Internal standards for sphingolipids, phospholipids and storage lipids were obtained from Avanti Polar Lipids, Inc. (Alabaster, AL. USA). They were added to each sample allowing for further identi cation and quantitative corrections. Description and concentration of each component are described in Table S7. Aminolipids and glycolipids as well as pigments had external calibration curves and standards were obtained from Avanti Polar Lipids, Inc. (Alabaster, AL. USA) and DHI Labs (Denmark), respectively.

Lipid extraction
Lipid extraction was performed according to an adaptation of the Bligh & Dyer (1959) 66 method. Each GF/F sample lter was macerated and homogenized in 1 mL of 10 mM phosphate buffer (pH 7.4) containing deferoxamine mesylate 100 μM. Then 800 μL of methanol and 200 μL of internal standard mix (10 μg.ml -1 ) were added. Next, 4 mL of chloroform/ethyl acetate (4:1) were added to each mixture and thoroughly vortexed for 1 minute. All procedures were performed on top of crushed ice to minimize evaporation of the solvents. Samples were sonicated for 20 min to enhance breakage of glycomembranes. After centrifugation (2000g for 6 min at 4 °C) the lower phase containing the total lipid extract (TLE) was transferred to a new tube and dried under N 2 gas. Dried TLE was again dissolved in 100 μL of isopropanol for analysis and the injection volume was set at 1 µL.
Lipidomic analysis and data processing TLE was analyzed using an ESI-TOFMS (Triple TOF 6600, Sciex, Concord, US) interfaced with a highperformance LC system (UHPLC Nexera, Shimadzu, Kyoto, Japan), as described previously in Chaves- MS/MS data was analyzed with PeakView® and lipid molecular species were identi ed by an in-house manufactured Excel-based macro. Spectrums, showing fragment breaks including exact masses and retention times used for identi cation, are exempli ed in the additional information. Pigments, plastoquinone, cholesterol, DAG, TAG and CE were fully analyzed in the positive mode. Aminolipids and glycolipis lipids were identi ed in the positive mode but quanti ed in the negative mode. All other membrane lipids and FFA were fully analyzed in the negative mode. Area of each lipid species were obtained by MS data from MultiQuant®. For quanti cation, the peak area of each lipid species was divided by the peak area of their corresponding internal standard (shown in detail in Table S7). Pigments, cholesterol and FFA had external calibration curves relative to LysoPC (17:0) internal standard (IS); DAG had external calibration curve relative to TAG (17:0/17:0/17:0) IS. PI, aminolipids and glycolipids had external calibration curves which were relative to PC (17:0/17:0) IS. They were injected separately following dilution curves with eleven points each based on concentration ranges described in Tables S8  and S9. Throughout the dilution curves, each point had half the concentration of the previous point and lower limits were de ned based on MS inferior limit of detection. Speci c correction factors were calculated as slopes from graph curves of each external standard divided by their respective abovementioned internal standards. For these speci c groups, nal concentrations were obtained from the peak area ratio divided by their respective internal standard area ratio and multiplied by their respective correction factor (for details see Table S8 and S9). No external calibration was performed for plastoquinone. Its peak area was divided by the peak area of Lyso PC (17:0) because of its similarities in retention time. Therefore, plastoquinone concentration values still not precise as the other lipid compounds and they are better interpreted by comparing within samples.
Oxylipidomic analysis TLE was spiked with the internal standard 5-HETE-d8 (100 ng) and 9-HODE-d4 (100 ng) and analyzed using an ESI-TOFMS (Triple TOF 6600, Sciex, Concord, US) interfaced with a high-performance LC system (UHPLC Nexera, Shimadzu, Kyoto, Japan). The HODE and HDoHE standards were synthesized and puri ed as previously described in Derogis et al. (2013) 67 . Samples were loaded into a BEH column (UPLC ® C18 column, 1.7 µm, 2.1 mm i.d. x 100 mm) with a ow rate of 0.5 mL/min and oven temperature at 35 °C. For RPLC, The mobile phase A consisted of acetic acid/water/acetonitrile (0.02:50:50), while mobile phase B composed of acetic acid/acetonitrile/isopropanol (0.02:50:50) for the lipid analyses performed in negative ionization mode 68 . The linear gradient during RPLC was as follows: from 0.1 to 55% B over the rst 4 min, 55 to 99% B from 4-4.5 min, hold at 99% B from 4.5-6.5 min, decreased from 99 to 0.1% B from 6.5-7 min, and hold at 0.1% B from 7-10 min. The MS was operated in negative ionization mode, and the scan range set at a mass-to-charge ratio of 200-1000 Da. Data for lipid molecular species identi cation and quanti cation was obtained with the targeted product ion acquisition method. Data acquisition was performed with a period cycle time of 0.56 s with 100 ms acquisition time for the MS1 scan and 10 ms for the MS/MS scan. Data acquisition was performed using Analyst® 1.7.1 with an ion spray voltage of -4.5 kV and the cone voltage at -80 V. The curtain gas was set at 25 psi, nebulizer and heater gases at 50 psi and interface heater at 500 °C. Speci c fragments used of each oxidized lipid were manually identi ed using PeakView ® and ChemDraw ® softwares (Table S10), as previously described 67 . Identi cation and quanti cation were performed by monitoring the speci c fragments of each analyte using Multiquant ® software. The area of analytes was obtained by using 5 mDa as the maximum acceptable mass error. The area ratio obtained for each lipid molecular specie was calculated by dividing the peak area of the lipid by the corresponding internal standard (Table S10). The concentration of lipid species was calculated by applying the area ratio in a calibrate curve constructed for each analyte. Data are presented as relative percentages. This analysis was performed separately from lipidomics, so that data generated here are not comparable to relative percentages, neither to abundances of lipid compounds monitored through lipidomics.

Statistical analysis
All lipid compounds described in our work (276 lipid molecules) were sorted into the main lipid classes (Table S1). Differences among the lipidomes of symbionts growing at 22 ˚C (controls) were stablished before applying the heat shock considering the total sums of each lipid class. For that we used a one-way ANOVA analysis followed by Tukey's HSD (p < 0.01) corrected by a false discovery rate (FDR) routine using MetaboAnalyst 4.0 software 69 . A two-way analysis of covariance (ANCOVA) was performed to test for signi cant differences in cell density development over time between the control and heat shock groups of each species. ANCOVA used time (hours of experiment) as a covariable and Heat shock ((T) /Control (C)) nested within each species. Pairwise comparisons between heat shock and control within the same time; and heat shock over time (T 4 , T 28 and T 244 ), for each specie were tested with Tukey's HSD post-hoc test (p < 0.01) (JMP v8.0 statistical software).
Further analyses on lipidomic focused on 133 membrane lipids and pigments organized in their classes and performed separated from other characterized compounds (i.e. storage lipids, DAG and FFA). First, multidimensional scaling (MDS) and analysis of similarity (ANOSIM) (both ran with Primer 6.0 70 ) were used to visualize and test the differences among heat shock and control in each species and among species. Positive pairwise R values from ANOSIM were used as indicators of the intensity of membrane lipid remodeling caused by heat shock in each Symbiodinaceae. Principal component analysis (PCA) was used to identify the groups of membrane lipids responsible for the main differences caused by heat shock (classes detailed in Table S1). After this screening, each one of these lipid classes, together with storage lipids and oxylipins were tested for signi cant differences among the experimental units with a 3way mixed ANOVA to evaluate the differences between heat shock and control among species using heat shock (HS; two levels: control and treatment) orthogonal to time (Ti; two levels: T 4 and T 244 ) and nested within each Symbiodiniaceae species (Spp.; three levels: A1, B1 and C1). A further 2-way crossed ANOVA was performed to evaluate the differences between heat shock and control individually in each species using heat shock (HS; two levels: control and treatment) orthogonal to time (Ti; two levels: T 4 and T 244 ).
Pairwise comparisons were tested with Tukey's HSD test, signi cance threshold was set at p < 0.01. The same matrix of 133 membrane lipids was used for similarity percentages (SIMPER) analyses (using Primer 6.0) comparing heat shock and control of each symbiont for identi cation of individual lipid compounds as heat stress biomarkers. All data of parametric tests were tested for homoscedasticity and log-transformed when necessary. Data was log-transformed (except in SIMPER analyses), so that the variations in less abundant lipids are comparable to more abundant compounds.