of coral bleaching and exposure to diseases (6)(7)(8)(9). Diseases may result in coral colony mortality and can drive the local extinction of corals (10). One of the biggest threats to Caribbean reefs is the continued spread of stony coral tissue loss disease (SCTLD). This virulent disease spread rapidly across Florida's Coral Reef over the course of several years, affecting ~22 scleractinian coral species since its first observation in 2014 (11), and has since spread throughout the Caribbean (12). Disease susceptibility, resilience, and lethality vary significantly among affected species (11,(13)(14)(15)(16), but underlying mecha nisms behind resilience remain unknown.

The Scleractinian (stony coral) coral holobiont is diverse, consisting of the coral animal (host), the endosymbiotic dinoflagellate algae, archaea, bacteria, viruses, and fungi (17)(18)(19)(20). Many coral species are sessile organisms that rely heavily on symbiosis with photosynthetic microalgae (Symbiodiniaceae) for their daily energy requirement and through associations with microbial communities for nutrition and defense (17,(21)(22)(23)(24)(25). Within the holobiont, multiple interactions occur among all members. The coral animal provides a physical habitat for the endosymbiont and a diversity of microorganisms within the holobiont. Nutrient exchange between the coral host and Symbiodiniaceae includes the exchange of phosphorus and nitrogen (from the coral animal), oxygen, and carbon (from the Symbiodiniaceae, generated through photosynthesis). Studies have shown that Symbiodiniaceae can modulate intra-and interspecies susceptibility to bleaching (26,27) and disease (28). Metabolite exchange between endosymbiotic zooxanthellae and bacteria influences the fitness of the endosymbiont (29). Bacteria selected by the coral holobiont fulfill several key roles in holobiont metabolite cycling and pathogen defense (30). Endozoicomonas, for example, are proposed to provide vitamins to both endosymbiotic dinoflagellates and the coral host (24). The bacterial communities are uniquely structured between the coral skeleton, tissue, and mucus (25). While the microbial community composition within the skeleton and tissue is associated with the coral host species, environmental factors may have a greater impact on the community structure within coral mucus (25). Beneficial bacteria (symbionts) also produce natural products that can provide competitive advantages to the coral holobiont and aid in responding to pathogens (31). Studies that profiled heat-sensitive and heat-tolerant corals have established that symbiosis as well as heterotrophy play a key role in determining which species thrive in the face of increasing ocean tem peratures and that endosymbiont identity is an important factor in survival (32)(33)(34)(35)(36)(37). Although the cause of SCTLD remains unknown, dysbiosis in the coral holobiont occurs with breakdown in the host-endosymbiont relationship in the gastrodermis resulting in necrosis and opportunistic infections by bacterial pathogens (38). There is also evidence of in situ symbiophagy (symbiont degradation) by the coral host in SCTLD-exposed corals (39,40).

'Omics techniques such as transcriptomics, proteomics, and metabolomics hold promise for delivering insights into biochemical pathways that may drive differences in disease response (41). Pairing multiple 'omics techniques enabled key insights into how Endozoicomonas can provide key immune response-related metabolites (such as vitamins) to the coral host and symbiotic zooxanthellae within the coral holobiont (24). Additionally, by comparing transcriptomes of stony corals of species Acropora hyacinthus either resilient or sensitive to bleaching stress, the expression of genes important for survival even under non-stress conditions in a phenomenon termed frontloading was observed (42). The differential expression of orthologs related to vesicular trafficking and signal transduction was positively correlated to species-specific susceptibility to SCTLD (39). With advancements in data annotation strategies, untargeted metabolomics is also being increasingly employed to generate, refine, and validate hypotheses to untangle interactions between different members of the holobiont (43,44). Metabolomics has been largely applied to profile different genotypes of corals (45), locations (46,47), delineate biochemical pathways important in heat tolerance (48-50), and the effect of environmental factors such as use of sunscreen (51). We compare the findings of these studies to the observations in this work throughout our paper.

We hypothesized that metabolomes of coral species with different reported SCTLD susceptibility would vary in their metabolomes, and such variations could guide future work aimed at understanding of the pathways implicated in disease resilience. The sampling site in this study, Dry Tortugas, Florida, was being monitored regularly in anticipation of SCTLD beginning in September 2020, and the disease was first observed on 29 May 2021. Thus, the species in this study were not affected by SCTLD at the time of sampling (January 2020) but are representative of Florida species that are known to be susceptible to SCTLD. There are only a few investigations that compare metabolic or lipid profiles of field-collected corals that are SCTLD susceptible (41). With the unabated spread of SCTLD along the Florida reef, opportunities to profile inter and intraspecies variation in metabolomes ahead of disease and post disease were envisioned to generate testable hypotheses to delineate biochemical pathways underlying disease susceptibility. To test our hypothesis and compare metabolomes of coral species with different SCTLD susceptibilities, we collected healthy coral fragments of four coral species, Orbicella faveolata, Montastraea cavernosa, Meandrina meandrites, and Colpophyllia natans, ahead of the SCTLD front in the Dry Tortugas. M. meandrites and C. natans are highly susceptible to SCTLD, while O. faveolata and M. cavernosa are defined as moderately susceptible (13,15,16). SCTLD susceptibility is defined by the length of time between the disease's arrival to a reef and observation of lesions on a particular species, rates of lesion progression, and prevalence among species (15,16). M. meandrites was one of the first reported species affected in the Dry Tortugas (52), while C. natans recruits spawned from parents in the Dry Tortugas showed ex situ lesion progression rates of 24.9-31.1%/day, which is in range for highly SCTLD susceptible corals (53). The Dry Tortugas is a unique habitat for corals along the Florida coral reef system where species such as Acropora palmata, Siderastrea siderea, and Porites astreoides have exhibited faster growth rates and enhanced reproduction relative to conspecifics in the Florida Keys (54). This may be attributed to oceanographic conditions that drive periodic upwelling, which is favorable for heterotrophy, and cooler temperatures and greater distance from urbanization and sources of pollution (54). Exogenous untargeted metabolome profiles of Dry Tortugas' seawater samples were previously reported to be distinct from the profiles of seven other zones within Florida's Coral Reef prior to SCTLD arriving at the Dry Tortugas (46). Thus, the visually healthy corals in this study from the Dry Tortugas provided the unique opportunity to examine and compare the metabo lomes of several coral species growing under optimal growth conditions in Florida (54)(55)(56) before this region was affected by SCTLD.

In this work, we apply an untargeted high-performance liquid chromatography-mass spectrometry (LC-MS)-based approach to profile the metabolomes of a small sample set of four coral species (M. meandrites, C. natans, O. faveolata, M. cavernosa) utilizing recently developed advanced compound annotation methods to identify metabolites underlying the interspecies differences observed. While metabolomics analysis has been performed on Caribbean Scleractinian corals (43,45,46,(57)(58)(59)(60)(61), our understanding of differences between the metabolomes of visually healthy Caribbean stony coral species is limited. Thus, we seek to address this gap in knowledge by describing chemical classes that are variably detected between four visually healthy coral species from the Dry Tortugas National Park sampled in January 2020 (52). In this study, we identify an endosymbiont-derived vitamin E pathway and a host-derived acylcarnitine pathway that were significantly variable among species. We describe additional chemical diversity by partitioning the crude extract of whole coral. Lastly, we report differences in the bioactivity of partitioned extracts of whole corals against bacterial pathogens.

Metabolome extracts from four stony coral species (Orbicella faveolata, Monstastraea cavernosa, Meandrina meandrites, and Colpophyllia natans) were subjected to LC-MS analysis (Fig. 1; Table S1). The resulting data were analyzed using a variety of data visualization and metabolite annotation tools (Fig. 1C). Unsupervised principal com ponent analysis (PCA) revealed that metabolome profiles of M. meandrites had the lowest intraspecies variation compared to the other coral species (Fig. 2A). The largest interspecific separation captured on the first principal component (PC) was observed between M. meandrites and M. cavernosa. Four PCs captured interspecies variation between M. meandrites and the other species, and intraspecies variation for M. caver nosa (Fig. S1A through E). The largest intraspecies distribution on PC1 was observed for C. natans, followed by O. faveolata and M. cavernosa. While the metabolomes of M. cavernosa fragments were spread across PC2, tighter clustering was observed for the other species along this component. PCs 3 and 4 captured metabolome varia tion between individual extracts, revealing that additional factors beyond species are captured within the metadata analysis (discussed further below) (Fig. S1A through E). We calculated alpha and beta diversity for each coral (using Shannon entropy and Brays-Cur tis similarity metrics, respectively, Fig. S2) to quantify metabolome similarity. A principal coordinates analysis on the Bray-Cutis similarity matrix constructed on the metabolome data revealed tighter clustering for M. meandrites, while the other species were spread along the first principal coordinate (Fig. S2A). The within species beta diversity was significantly larger than beta diversity between species (Fig. S2B, Mann Whitney U Test, P = 0.00082), further supporting that the metabolome analysis captures variation driven by factors beyond coral species. There was no significant difference between the alpha diversity for each species found using a Kruskal Wallis Test (Fig. S2C, P= 0.119).

Using permutational multivariate analysis of variance (PERMANOVA) (62), we found metabolome variation differed significantly across coral species (Pseudo-F 3,13 = 2.192, P = 0.007), with no significant effect of site (as a random effect, P = 0.670). We queried whether SCTLD susceptibility categorization (OFAV, MCAV = moderate; CNAT, MMEA = high) affected metabolome variation. We could not examine interactive effects due to a lack of Susceptibility × Species replication. We performed PERMANOVA with suscept ibility as a single fixed factor and found that metabolome variation was significantly different between the two groups (Pseudo-F = 3.191, P = 0.001). Additionally, a model with susceptibility as a fixed factor and species as a random effect found that Spe cies(Susceptibility) was slightly significantly different (Pseudo-F 1, 2 = 1.715, P = 0.021). Therefore, both coral species and SCTLD susceptibility affect metabolome variation, but it is unclear how the two interact in affecting metabolome differences. We cannot fully disentangle the effects of species and susceptibility with the current sampling regime in this study; therefore, such efforts are an important avenue for future inquiry. Sampling species with moderate resilience (i.e., some individuals are susceptible, others never develop lesions) where both affected and unaffected colonies were sampled would be a potential way to disentangle this effect. A non-metric multi-dimensional scaling (nMDS) plot by species was constructed to visualize metabolome variation using a method appropriate for smaller sample sizes. The nMDS plot divided the samples into four distinct clusters. Consistent with the PCA, all M. meandrites samples clustered together in one group, and a larger spread was observed for other species with metabolomes of M. cavernosa displaying the largest intraspecies variation (Fig. S3). These observations aligned with previous findings where apparently healthy M. cavernosa from a SCTLD endemic site in Broward County showed similar intraspecies variation (43). The proximity between M. cavernosa colonies on the reef in this previous study explained the variation cavernosa). (C) Untargeted metabolomics data were acquired, processed, and analyzed with a variety of methods. Metabolomics data available through public data sets (mined using MASST) and acquired on cultured algae was used to assign the biosynthetic source of annotated metabolite features. (D) Schematics of representative interactions between the coral holobiont members are shown. The host genotype, the microbiome composition, and the endosymbiont Symbiodiniaceae species as well as the complex interplay of interactions between them can confer resilience to the increased frequency and impact of coral diseases.

in some instances but did not completely explain the metabolome variation observed for M. cavernosa in Broward County (43). The beta diversity analysis conducted in this study revealed three pairs of corals from the site B or D as having the greatest Bray-Curtis similarity score to each other. These include OFAV2B/CNAT12B, MCAV6B/OFAV12B, and M. meandrites (MMEA, purple), and O. faveolata (OFAV, blue). The SCTLD susceptibility categorization is included in the key ("Highly, " "Moderately"). Axes are labeled with the corresponding variance explained by each principal component. A, B, C, D refer to the site from which the coral was sampled (Table S1).

(B) Hierarchical clustering analysis reveals a separate cluster for all M. meandrites samples, while the other species are distributed across clades. Colored branches correspond to species as outlined in (A). The letter at the end of the sample name corresponds to the sampling site. The x-axis represents the distance between the samples/clades. (C) UpSet Plot showing the distribution of detected metabolite features. The number above each bar represents the number of features in that intersection. "Set Size" denotes the total number of features detected in each coral species. The inset table includes a number of features detected within each species, as well as the number of annotated features reported in this paper. MMEA2D/MMEA24D. Thus, we see further evidence of reef site driving metabolome similarity in some instances although for the MMEA pair we cannot disentangle the effect of site and species on the metabolome similarity.

An unsupervised hierarchical clustering analysis (HCA) revealed M. meandrites was the only species that clustered within a single clade (Fig. 2B, purple branches), further indicating the relatively low intraspecies variation. The HCA also reveals that the corals within the aforementioned pairs identified through the Bray-Curtis similarity analysis have the greatest metabolome similarity to each other (Fig. 2B). There was significant variation in metabolomic variation (multivariate dispersion) among the coral species (F 1,3 = 7.944, P = 0.045). Overall, MCAV was the most variable (average distance to group centroid = 33.3) with the variation being significantly higher than the variation observed for MMEA (average distance = 19.2, P = 0.036). OFAV was the second most variable (average distance = 29.3) and the variation was significantly higher than the variation of MMEA (P = 0.030). The relative metabolomic variation did not differ between pairwise comparisons performed for other coral species. This phenomenon has been observed in deep sea corals where interspecies differences rather than site-dictated clustering of metabolite profiles (63). However, intraspecies differences may be attributable partially to site, as Haydon and colleagues found metabolite differences in Pocillopora acuta based on reef site, even after acclimation of the corals in aquaria as is the case in this study (48). Thus, when comparing multiple species of corals from different reef sites, replication of species collected from each site should be conducted where possible. The intraspecies metabolomic variation observed in this study may further be partially explained by different genotypes (45), and both intra-and interspecies variation may be influenced by the endosymbiotic profile, microbial community, bleaching history, and stimuli/stressors unique to the sampling site (21,22,64,65). Acquiring data on seawater (exometabolomics), cataloging abiotic factors at reef sites, and profiling the endosymbiont and microbial community will aid in disentangling what additional factors drive metabolome variation. In a metabolomics study of cultured Symbiodinium species, Klueter et al. noted that metabolite profiles varied by species and the degree of metabolome variation was not ubiquitous across species given the different classification error rates of each symbiont species (66). The distinct metabolome profiles of M. meandrites compared to the other corals species in this study could be influenced by the symbiont types of the coral species. The M. meandrites sampled for this study may host symbionts with highly similar metabolomes or interaction networks with the associated microbiome, while the other coral species may host symbionts and/or microbiome with more diverse metabolomes. Incorporating microbiome analysis and symbiont typing into metabolome studies would be beneficial toward delineating how the degree of observed metabolome variation may correspond to the symbiont and the microbiome species present. Another factor that may contribute to the intraspecies metabolome variation captured in this study is cryptic lineages observed within the studied species (67). Cryptic coral species lineages may share phenotypic traits but have distinct underlying genomic differences (67). M. cavernosa (68,69) and O. faveolata (70,71) are reported to have cryptic lineages. The genomic differences between the coral hosts imply a strong possibility for metabolite differences (since the metabolome reflects the functional biochemical state of the system, in this case the coral holobiont). Proven association with different symbiont genera [reported for O. faveolata (70)] and potential association with different microbial assemblages (yet to be studied for the species in this report) among cryptic lineages could further diverge metabolome profiles, as all the members of the coral holobiont influence and contribute to the metabolome. Such phenomena (67) may well explain the metabolome variation of M. cavernosa and O. faveolata in this study, and although cryptic lineages have not yet been identified for Colpophyllia (67), the observed metabolome variation in C. natans may be partially explained by this as well.

The UpSet Plot generated to show the distribution of metabolite features revealed the greatest number of unique features were detected in M. meandrites extracts, followed by M. cavernosa, C. natans, and O. faveolata (Fig. 2C). Since the statistical analyses revealed a distinct metabolome profile for M. meandrites and a greater intraspecies metabolite variation captured for the other coral species, unique features present in M. meandrites and features that were variably detected among M. meandrites and other coral species were prioritized for annotation and are described below.

A metabolite feature m/z_RT (m/z: mass to charge, RT: retention time in min) 449.398_21.3 min, uniquely detected in M. meandrites extracts, was proposed as α-tocopherolhydroquinone by the in silico annotation tool MolDiscovery. SIRIUS with CSI:FingerID also proposed the annotation for this feature as α-tocopherolhydroquinone. We searched for α-tocopherolhydroquinone spectra in the literature and used MS 2 spectral matching with a published spectrum (72) of silicated α-tocopherolhydroqui none to further support the annotation (Fig. S4A). This feature clustered in Feature Based Molecular Networking (FBMN) analysis with another feature at 447.383_21.3 min representing one unsaturation from α-tocopherolhydroquinone (Δm/z = 2.015). We annotated this feature as α-tocopherolquinone, the oxidation product of α-tocopherol hydroquinone (73,74) and confirmed this annotation with an analytical standard of α-tocopherolquinone (Fig. S4B and C). To identify additional metabolites, we applied unsupervised substructure discovery using MS2LDA (75) (Fig. 3A; Table S2).

It is interesting to note that α-tocopherolquinone, an oxidation product of α-toco pherol (76,77), and α-tocopherolhydroquinone were exclusively detected in highly SCTLD susceptible M. meandrites and C. natans but not in O. faveolata and M. cavernosa with moderate SCTLD susceptibility (15,16). These features were also not detected in the extracts of cultured Symbiodiniaceae included in this study (Fig. 3B). The Durusdiniumassociated mangrove coral Pocillopora acuta was previously reported by Haydon and colleagues to accumulate α-tocopherol in summer (48). This observation was suggested as a possible mechanism of "frontloading" associated with the resilience of Durusdiniumassociated corals (48). Tocopherols are the most prominent antioxidants that counteract lipid peroxidation. Using transcriptomics, the gene for arachidonate 5-lipoxygenase (ALOX5) was found to be significantly differentially expressed with its highest expres sion in the most susceptible corals, including C natans (39). ALOX5 is a key enzyme in mediating lipid peroxidation (78) which can lead to cell death such as apoptosis, ferroptosis, and pyroptosis (78). Thus, we searched the literature to identify studies that might link α-tocopherol(hydro)quinones with lipid peroxidation and cell death. While linking specific metabolites with processes is outside the scope of this study, we can speculate on possible functions of metabolites and determine future avenues of inquiry based on literature precedence. Indeed, a recent report updated the mechanism of action for iron-dependent anti-apoptotic activity (ferroptosis) of α-tocopherol (79). Tocopherol was suggested to be the pro-vitamin E form, while the (hydro)quinone forms produced from the oxidation of α-tocopherol were shown to be the activated forms responsible for the prevention of cell death (79). Thus, it is possible that our detection of the α-tocopherol(hydro)quinones indicates the coral cells are frontloading the activated form of vitamin E, which is counteracting lipid peroxidation resulting in the detection of α-tocopherol(hydro)quinones.

When corals were previously challenged with bacterial pathogen-associated lipopolysaccharides, susceptible corals demonstrated a transcriptome response related to apoptosis, while resistant corals transcribed genes related to autophagy, a more modulated response to stressors (80). The damage threshold hypothesis proposes that coral disease susceptibility is inversely related to the upper limit of damage a coral can withstand before harmful effects are observed (81). Corals with a low damage threshold (high susceptibility) may be unable to modulate immune responses; either mounting too high of a response, leading to auto-immune challenges, or too low of a response before cellular death is imminent. Thus, the varied detection of tocopherol(hydro)quinones in this study should be further investigated to determine if these metabolites serve as a biomarker of corals particularly susceptible to disease, represent stressor history, and if they vary temporally with disease progression.

M. meandrites and O. faveolata had the highest relative abundance of α-tocomo noenol (Fig. 3A). α-Tocomonoenol was previously detected at higher abundance in apparently healthy M. cavernosa compared with diseased corals (43). The analog α-tocotrienol (m/z 424.333) with three degrees of unsaturation was exclusively detec ted in SCTLD-affected M. cavernosa, while other unsaturated analogs were likewise detected at higher abundance in the diseased corals (43). In this study, where we have analyzed healthy corals ahead of the SCTLD front, α-tocotrienol was not detected. Based on these results, we hypothesize that tocotrienols may serve as biomarkers for coral disease, wherein accumulation coincides with disease progression. A time course study that tracks how tocotrienol analogs and tocopherolquinones accumulate in response to disease exposure is required to validate this hypothesis. Recent work highlighting differential detection of tocopherol upon heat stress among resilient and susceptible species (48, 82) and our work reporting the detection of different tocopherol analogs among healthy and SCTLD-affected coral colonies suggest that this endosymbiont pathway likely plays an important role in coral health and resilience (48, 83); warranting studies that monitor tocopherol-related metabolite production over time after disease exposure.

Feature 476.373_14.4 min was proposed by SIRIUS with CSI:FingerID as docosatetrae noyl carnitine (C22:4) (Fig. 4A and B). To confirm this annotation prediction and to determine if other acylcarnitines were present in our data, the output of the MS2LDA analysis was consulted. This feature shares MS2LDA substructure motif 185 with feature 276.180_2.8 min, which was annotated as hydroxyhexanoyl carnitine (C6:0-OH) based on the MS 2 fragment peak at m/z 217.107 associated with the neutral loss of trimethylamine (Δm/z = 59.07, Fig. 4C; S4A). A variety of acylcarnitines were further annotated with the aid of the GNPS spectral library, MassQL, substructure motif 185, and SIRIUS with CSI:FingerID (Fig. 4; Fig. S4B through O; Table S2). The putative annotation and m/z_RT are included for each feature (Table S2; Fig. S5). (K) The hierarchical clustering analysis based on the log-transformed abundances of the acylcarnitines. The SCTLD susceptibility categorization is included in the key ("Highly, " "Moderately").

Acylcarnitines are typically host-derived metabolites, and these metabolites were not detected in the cultured Symbiodiniaceae extracts in this study. Acylcarnitines have been detected at higher abundances in the daytime exometabolomes of Porites and Pocillopora compared to algae (turfing microalgae, macroalgae, and crustose coralline algae), where they are hypothesized to play a role in nitrogen and phosphorous cycling (84). Acylcarnitines play an integral role in metabolism of fatty acids in mitochondria (85) and maintenance of available pools of free coenzyme A (86). In the diatom Phaeodacty lum tricornutum, propanoyl-carnitine and butanoyl-carnitine accumulate under nitrogenstarvation (87). Accumulation of acylcarnitine concentrations has been linked with cell toxicity (88), mitochondrial dysfunction (89)(90)(91), and dysfunction in cellular bioenerget ics in humans (88). Acylcarnitines have been found to be upregulated in corals upon exposure to octocrylene, an ingredient used in sunscreens (51), which is the only study reporting conditional dysregulation of acylcarnitine levels in corals found in our literature search. In this study, the interspecies variation of all features annotated as acylcarnitines were analyzed with a Kruskal Wallis test with Dunn's post-test (adjusted P < 0.05). The acylcarnitines with fatty acyl tails with C13-C20 are classified as long chain, and tails > C21 as very long chain (92). Features that were differentially detected showed two interspecies patterns based on the acyl chain length (Fig. 4B, C and F through J; Fig. S4D through O). The hydroxyhexanoyl acylcarnitine and very long chain acylcarnitines were detected at a higher intensity in M. meandrites (Fig. 4B, C and F through I). The accumulation of long-chain acylcarnitines is associated with several metabolic diseases in humans (92). Differences in acylcarnitine profiles have also been reported as indicators of frailty in humans (93). Given that certain acylcarnitine analogs are detected at higher intensity in M. meandrites (Fig. 4; Fig. S4), a highly SCTLD-susceptible species, it is possible that acylcarnitine profiling could represent disease history and/or higher susceptibility to disease. Interestingly, when an HCA was performed on only the annotated acylcarnitine features, a clear separation of M. meandrites from other coral species was observed (Fig. 4K). Thus, host-derived acylcarnitines display a species-specific profile. Since several acylcarnitines were variably detected in these apparently healthy corals, the understud ied role of carnitines in disease resilience and susceptibility in corals should be further investigated.

Several unknown acylcarnitines were distributed differentially across the coral species (Fig. 4J; Fig. S4G through O). Upon manual inspection of fragmentation spectra, we propose the annotation of these features as acylcarnitines containing fatty acid esters of hydroxy fatty acids (known as FAHFAs) (Fig. S5A; Table S2). The fragments at m/z 85.028 and 144.102, the presence of a fragment corresponding to hydroxylated fatty acid of acylcarnitine (CAR 14:1-OH), and the presence of an additional fatty acid tail fragment (C18:0) supported the annotation of an FAHFA-containing acylcarnitines [Fig. S5A, bottom spectrum; CAR 14:1-(O-18:0)]. FAHFAs are a conserved class of lipids that are widely reported, including in dietary plants (94,95), as defense molecules in caterpillars named as mayolenes (96,97), as anti-inflammatory metabolites in humans (98), and in the corallivore Crown-of-Thorns Starfish (99). Oxidative and environmental stress increase the synthesis of FAHFAs and ornithine-conjugated FAHFAs (100,101). Acylcarnitines containing FAHFAs have not been previously reported and warrant further investigation for structural characterization and their role in coral biology. We searched for these acylcarnitines features in the publicly available data sets on the MassIVE server using MASST (102). These features were found in several marine organism-derived data sets including data sets from several coral species (Table S3) but were not observed in human-derived data sets. These observations further strengthen the role and application of modern methods in data analysis in untargeted metabolomics to discover biologically relevant metabolic pathways and generate testable hypotheses. Here, access to public data sets on these pristine endangered coral species is advantageous.

Several differentiating features were identified as diacylglyceryl-carboxyhydroxymethyl choline (DGCC) betaine lipids. Feature 774.584_19.6 min was a GNPS library match to DGCC(36:5) (Fig. 5A; Fig. S6A). The fragment peaks at m/z 490.373 and 472.363 in the MS 2 spectra, which are characteristic of the chemical substructure containing a 16:0 fatty acyl tail, enable further annotation of this feature as DGCC(16:0_20:5) (Fig. S7). This feature was variably detected among coral species, present at the highest abundance in M. cavernosa. As expected, the feature 490.373_13.1 min, annotated as lyso-DGCC(16:0) known to be a constituent of healthy corals (43,58,103), was detected in all species (Fig. 5B; Fig. S6B). We used MassQL to search for additional DGCC analogs containing a 16:0 fatty acyl tail (Fig. S6C). This approach permitted the annotation of additional metabolite features, detected at highest abundances in M. meandrites, as lyso-DGCC(16:0) analogs (Fig. 5; Fig. S7; Table S2). Diacylated and unsaturated DGCC betaine lipids have been previously proposed as biomarkers of coral bleaching (58,103). The increase in lipid unsaturation is suggestive of increased cell death when the antioxidative capacity of cells is lowered (104). DGCC betaine lipids are biosynthesized by Symbiodiniaceae (50). We searched the metabolite data acquired on cultured Symbiodiniaceae for the presence of the annotated DGCC analogs. While monoacylated lyso-DGCC(16:0) was detected in all cultured Symbiodiniaceae genera, the diacylated analogs were notably absent in Durusdinium extracts, the genera known to be most thermotolerant (105-108) (Fig. 5H). Roach et al. noted a higher abundance of lyso-DGCCs in historically non-bleached corals, while unsaturated and DGCCs were abundant in historically bleached corals (58). Rosset et al. observed significantly higher abundance of lyso-DGCC and unsaturated DGCCs in thermotolerant D. trenchii as compared to Cladocopium C3 in both control and heat-stressed conditions (49,50). Symbiodiniaceae genera show differential responses to thermal and irradiance stress, which affects the entire holobiont response to stressors (64,(109)(110)(111). The variable detection of the DGCC(16:0) analogs in the coral extracts may indicate variable bleaching history or the presence or absence of certain Symbiodi niaceae species in the coral colonies sampled. Previous reports suggest that the DGCC lipid profile is influenced by the host (112). Given that algae transform their membranes in response to a variety of stimuli and stressors (113)(114)(115), it is also possible that the variable detection of the diacylated DGCC(16:0) analogs is reflective of host-dependent shifts in betaine lipid profiles of Symbiodiniaceae.

Carotenoids are important antioxidants in photosynthetic organisms. Symbiodiniaceae produce several carotenoids such as peridinin, fucoxanthin, astaxanthin, diatoxanthin, diadinoxanthin, and neoxanthin, with peridinin being the most prevalent and abundant (116,117). We examined whether endosymbiont-derived pigment profiles contributed to variation among the coral colonies analyzed in this study. Several pigments were annotated using mass spectral search and literature search (Fig. 6; Fig. S8A through F; Table S2). The features annotated as pigments were also analyzed by HCA (Fig. 6B). The pigment profile did not display clear interspecies variation but did display intraspecies variation. Peridinin was detected in all coral extracts, while fucoxanthin was detected in only a few coral extracts (Fig. 6A). Among cultured Symbiodiniaceae in this study, peridinin was detected in all genera, whereas fucoxanthin was only detected in thermotolerant Durusdinium cultures (Fig. S8G). Since Wakahama et al. reported a negative correlation between the presence of fucoxanthin and peridinin in a variety of symbiotic and free living Symbiodinium strains (118), we confirmed the detection of fucoxanthin in peridinin-containing coral extracts using an analytical standard of fucoxanthin (Fig. S8F). The detection of both pigments may represent the presence of multiple strains of Symbiodiniaceae within the cultures.

The crude extracts from the whole coral samples were further partitioned into ethyl acetate (EtOAc) and butanol (BuOH) solvents, and the bioactivity of these fractions was tested against the potential SCTLD-associated pathogens Vibrio coralliilyticus OfT6-21 and OfT7-21, Leisingera sp. McT4-56, and Alteromonas sp. 120) using an agar disk-diffusion assay (Fig. 7A; Fig. S9A). Partitions only exhibited activity against V. coralliilyticus strains. The largest zones of inhibition were observed for BuOH partitions of M. meandrites against both pathogens (Fig. 7A; Fig. S9A). Thus, untargeted metabolomics data were acquired on BuOH partitions of all species. The metabolite data were analyzed following the scheme outlined in Fig. 1. Within the BuOH partitions, additional 560 metabolite features were detected (Fig. 7B). The UpSet Plot analysis showed the greatest number of unique features was detected in C. natans extracts, followed by M. cavernosa, M. meandrites, and O. faveolata (Fig. 7B). We used CANOPUS to predict the chemical classes of these features (Table S4). For the features uniquely detected in the BuOH partitions, none of the metabolites in the CANOPUS-predicted natural product pathways were significantly enriched in M. meandrites compared to the other species (Fig. 7C). The UpSet Plot and CANOPUS output were used to guide compound annotations (Fig. 7D).

A feature, detected exclusively in BuOH partitions at 280.157_7.9 min was annotated as Tau-C10:0 based on MS 2 spectral matching (Fig. S10A). We also observed the presence of the N-acyl taurines in several publicly available data sets acquired on diatoms, dinoflagellates, and seawater by searching the MS 2 spectrum of this metabolite in MASST (Table S3). N-acyl taurines have been implicated as important signaling molecules in several human processes including postprandial glucose regulation (121), but these molecules have not been previously described in corals. Thus, partitioning crude extracts into organic solvents can enable detection and characterization of low-abundance metabolites, which are otherwise below the limit of detection. Feature 267.960_5.3 min with an isotopic pattern of a brominated compound was uniquely detected in the The SCTLD susceptibility categorization is included in the key ("Highly, " "Moderately"). class as hydropyrimidine carboxylic acids and derivatives. Caelestine A, a brominated quinoline carboxylic acid, has been reported as a possible indicator of a response to heat stress in the invasive bryozoan Bugula neritina (122). A feature at 615.346_5.9 min was proposed as a tunicyclin G analog by DEREPLICATOR, which compares experimental MS 2 spectra against predicted in silico MS 2 spectra of peptides (123) and CANOPUS predic ted the chemical class of this feature as oligopeptides. We putatively annotated the feature 615.346_5.9 min as a polypeptide with partial sequence NGAI/LA (Fig. S9C). This polypeptide was detected in M. cavernosa and O. faveolata BuOH partitions. Although we could not link the enhanced antibacterial activity with specific molecules or chemical classes due to a small sample size, these annotations show that partitioning of whole coral metabolomic extracts increases the breadth of detected metabolites and enables annotation of low-abundance natural products. Annotation of individual metabolites is still a tedious and manual task. As spectral libraries are populated by the community, and in silico compound annotation methods advance, these data sets will be a valuable resource to untangle mechanisms of symbiosis between members of coral holobionts. Research Article mSystems December 2024 Volume 9 Issue 12 10.1128/msystems.00856-24 15 Downloaded from https://journals.asm.org/journal/msystems on 30 January 2025 by 128.61.217.165.

The visually healthy corals collected from the Dry Tortugas, a region of Florida with oceanographic conditions that support coral productivity, revealed interspecies metabolome differences. Tight clustering of the M. meandrites metabolome indicated similar metabolite profiles, while higher metabolome variation was found for C. natans, M. cavernosa, and O. faveolata. Metabolites driving the variation between species included tocopherol(hydro)quinones, diacylated betaine lipids, and acylcarnitines. This is the first report describing differences in acylcarnitine profiles between coral species and the discovery of potentially novel analogs containing an additional fatty acid group.

Given the specificity of these acylcarnitine-FAHFA to only marine organisms based on the MASST search, the biochemical function of these molecules is of particular interest.

The role of acylcarnitines in cellular energetics is well established; the varied detection of acylcarnitines in corals may indicate variability among species in their ability to readily utilize these pathways. Future work will focus on the structural description of these carnitines. How the profiles of metabolites reported in this paper change over time should also be characterized to determine their viability as biomarkers of health, disease, and lesion progression. The juxtaposition of M. meandrites SCTLD susceptibility and the observed highest bioactivity of the BuOH partitions of the extracts of this species against putative secondary SCTLD pathogens generates interesting avenues for future study, including how molecular dynamics of pathogen response and disease susceptibility might explain discrepancies between disease susceptibilities in the field while metabo lite extracts show high antibacterial activity when challenged with in lab-assays to a specific pathogen. Additional studies can also incorporate knowledge of environmental factors like heat stress to determine how biochemical disease dynamics and susceptibil ity may shift in the field.

As SCTLD appears to affect Symbiodiniaceae and disrupt their relationship with the host (13,38,40,43), it is imperative to understand the differences in chemical cross-talk between the corals and endosymbionts. Symbiodiniaceae dynamics within the host (e.g., relative abundance, density, species) will likely have an effect on the metabo lomic profiles. In this study, several Symbiodiniaceae metabolites driving interspecies differences and endosymbiont-derived carotenoid pigments displayed both inter-and intraspecies variation suggesting the presence of different endosymbiont genotypes in these samples and/or different endosymbiont-host dynamics among species. Thus, metabolomic studies on Symbiodiniaceae directly isolated from field corals will enable source tracking to tease apart host-derived, diet-derived, and endosymbiont-derived compounds. Given that the endosymbiont fraction can be isolated from corals using mechanical methods such as the air brush technique (124), we propose future studies also include comprehensive metabolic profiling of these endosymbionts prior to and upon exposure to disease. Such differences should be interrogated across corals with different disease susceptibility and with different endosymbiont profiles. Furthermore, studies of metabolomic profiles taken at discrete time-points after disease exposure will provide insights into the transitory response of corals to disease stressors. With the increased application of untargeted metabolomics methods to study coral physiol ogy, availability of annotated data sets, and our ability to mine public data sets using methods such as MASST, discoveries pertaining to chemical interactions between the coral host, the endosymbiont, environment, and the microbiome that define health status are a real possibility. By advancing our knowledge of the biochemical pathways involved in coral health and susceptibility to disease, we can disentangle the sources of metabolites, and how they change with time and increasing anthropogenic and climate-related stressors.

Whole coral fragments of a maximum size of 200 cm 2 were collected on SCUBA from four visually healthy Scleractinia coral species, O. faveolata (n = 4), M. cavernosa (n = 4), M. meandrites (n = 4), and C. natans (n = 3). These were collected in January 2020 from four sites outside of the Dry Tortugas National Park (Fig. 1; Table S1). This collection occurred ahead of the SCTLD front, which was first observed at the Dry Tortugas in May 2021 (125). Collection also occurred during a time of year when temperature stress and any associated paling or bleaching of the corals should not have been occurring, and none was observed at time of collection. Corals were chiseled at the base until they released from the substrate and then were transported back to the diving vessel in 18.9 L plastic bags filled with ocean water. Collected corals larger than 25 cm 2 were cut down to this size on the diving vessel using an AquaSaw (Gryphon C-40 CR). The cut portions and whole colonies were stored in a 1,000 L covered insulated container (Bonar Plastics, PB2145) filled halfway with ocean water. Air stones within the container allowed for aeration and water movement. A complete water change was performed on the container four times daily. Collections were conducted over 2 days before corals were transported the morning of the third day. The cruise was sponsored by the Florida Department of Environmental Protection and sample collection was covered by permit FKNMS-2019-160 to Valerie Paul. All corals collected from the field were transported to the Smithsonian Marine Station in Fort Pierce, FL. For transport, individual colonies were wrapped in plastic bubble wrap moistened with ocean water and then placed in a cooler.

Upon arrival, corals were rinsed with filtered seawater (FSW) and stored in a large indoor recirculating system holding approximately 570 L of FSW at 25.5°C ± 0.3°C. The FSW was collected from an intake pipe extending 1,600 m offshore South Hutchingson Island, Port Saint Lucie, FL and was filtered progressively through 20, 1.0, 0.5, and 0.35 µm pore filters. While stored prior to use in the recirculating system, the FSW constantly circulated through a 20-µm pore filter, a filter canister with ROX 0.8 aquarium carbon (Bulk Reef Supply), and a 36-watt Turbo-twist 12× UV sterilizer (Coralife) in series. The recirculating system contained a UV sterilizer (same model as described), two circulating pumps (AquaTop MaxFlow MCP-5) to create water movement, and a row of 6 blue-white 30 cm 2 LED panels (HQPR) to create 150-250 µmol photons m -2 s -1 for the captive corals. Corals were stored in the recirculating system for 5 days prior to sampling to allow them time to recover after transport, with a partial water change on the fourth day. All corals were held together in a single table.

After the 5th day, the corals were cut into smaller segments with a rock saw, and the blade was constantly sprayed down with UV/filter-sterilized seawater to cool the blade and wash off any debris, thus reducing cross-contamination between corals. The coral fragments ranged from 1 to 13 cm 2 in surface area (6.2 + 3.7 cm 2 , mean + SD) were immediately frozen (-80°C) and lyophilized the next day. Coral fragments were lyophilized overnight and then extracted twice using a 2:2:1 mixture of ethyl acetate (EtOAc), methanol (MeOH) ,and water (H 2 O) at room temperature in 20 mL scintillation vials. For the extraction, coral fragments were covered with an excess of solvent mixture, sonicated for 5-10 min and left to sit for 3 h on the initial extraction and overnight for the second extraction. The liquid extract was then transferred into a round-bottom flask using filter paper to prevent the transfer of coral fragments. The coral extract was then dried via rotary-evaporation (Buchi Rotovapor R-210) in a 35°C water bath (Buchi Heating Bath B-491) and weighed to determine the amount of crude extract. The extracts were dried in vacuo and 0.5 mg of the extract was transferred to Eppendorf tubes for metabolomics data analysis. The extracts were stored at -20°C until UPLC-MS data were acquired.

We previously reported on profiles of Symbiodiniaceae isolates provided by Mary Alice Coffroth from the University of Buffalo Undersea Reef Research (BURR) collection (43). Given the challenges involved in isolating and culturing Symbiodiniaceae, we used this publicly available data (43) (MassIVE identifier MSV000087471) in this current study to aid in determining the biosynthetic producer of detected metabolites. Briefly, the endosymbionts were isolated by Mary Alice Coffroth from Orbicella faveolata corals sampled between 2002 and 2005 from the Florida Keys. Isolate extracts were sent by Richard Karp and Andrew Baker (University of Miami). Culture conditions included incubation at 27°C in F/2 media, with 20 μE of light on a 14:10 diurnal cycle. Extracts of the culture were performed as previously described (43), using 2:2:1 EtOAc:MeOH:H 2 O to extract pelletized cellular cultures. Solvents were removed, and the dried samples were transferred in 3:1 MeOH:H 2 O into a 1.5-mL microcentrifuge tube. After centrifugation, the supernatant was transferred to a microcentrifuge tube and removed via SpeedVac for 3 h. The extract was frozen, lyophilized, and analyzed using UPLC-MS/MS. Please see Deutsch et al. 2021 for detailed methodology (43).

The dried extracts were resuspended in 100% MeOH containing 1 µM sulfadimethoxine as an internal standard. The samples were analyzed with an Agilent 1290 Infinity II UHPLC system (Agilent Technologies) using a Kinetex 1.7 µm C18 reversed phase UHPLC column (50 × 2.1 mm) for chromatographic separation, coupled to an ImpactII ultra high resolution Qq-TOF mass spectrometer (Bruker Daltonics, GmbH, Bremen, Germany) equipped with an ESI source for MS/MS analysis. MS/MS spectra were acquired in positive mode as previously described (43). Metabolomics data on the cultured Symbiodiniaceae from the Burr Collection were previously acquired (43). The strains utilized are reported in Table S5.

The raw data were converted to mzXML format using vendor software. MZmine 2.53 was used to extract metabolite features with steps for mass detection, chromatogram building, chromatogram deconvolution, isotopic grouping, retention time alignment, duplicate removal, and missing peak filling (126). These processed data were submitted to the feature-based molecular networking workflow on the Global Natural Product Social Molecular Networking (GNPS) platform (127). The output of MZmine includes information about LC-MS features across all samples containing the m/z value, retention time, the area under the peak for the corresponding chromatogram, and a unique identifier for each feature. The quantification table and the linked MS² spectra were exported using the GNPS export module (126,128) and the SIRIUS 4.0 export module (126,129). Feature Based Molecular Networking was performed using the MS² spectra (.mgf file) and the quantification table (.csv file). The molecular network was generated as previously described (43). The molecular network and the generation parameters are available here. The molecular network was visualized using Cytopscape v3.7.2 (130). The MS2LDA analysis was performed as previously described with default parameters (131,132). The MassQL Sandbox Dashboard (133) (v 0.3) on the GNPS platform was used to construct the spectral pattern queries for the MassQL search. Feature annota tion was performed using SIRIUS with CSI:FingerID, MolDiscovery (134), GNPS spectral library matching, MassQL, MASST, and literature searches. The metabolite annotations presented herein follow the "level 2" annotation standard based upon spectral similarity with public spectral libraries, spectra published in the literature, and through spectral comparison with the analytical standards as proposed by the Metabolomics Society Standard Initiative (135). All mzXML files included in this study can be accessed publicly on the repository Mass Spectrometry Interactive Virtual Environment (MassIVE) with ID MSV000089633. The commercial analytical standard for α-tocopherolquinone (catalog number 35365) was purchased from Cayman Chemical Company and the commercial analytical standard for fucoxanthin (catalog number 16337) was purchased from Sigma Aldrich.

Prior to statistical analysis, blank subtraction was performed as previously descri bed (43) to out features detected in the solvent and media controls. Unsuper vised multivariate statistical analyses including principal component analysis (136) and hierarchical clustering analysis (137) were performed using MetaboAnalyst 5.0 (138), and pareto scaling was employed prior to the analyses. The Plotter Dashboard (v.0.5) on the GNPS platform was used to construct boxplots for metabolite features of interest. A nonparametric Kruskal Wallis test with Dunn's posttest was performed in R studio. The UpSet Plots were made using the Intervene app (139).

To test for an effect of coral species on metabolomic variation, we used a permuta tional multivariate analysis of variance (PERMANOVA) (62). Coral species was treated as a fixed effect (four levels), with site included as a random nested effect (four levels). The PERMANOVA was based on a Bray-Curtis similarity matrix (140), type III (partial) sums of squares, and 999 random permutations of square-root transformed data (to down-weight heavily dominant variables) under a reduced model. Both PERMANOVA and non-metric multidimensional scaling (141) plot were constructed from a Bray-Curtis similarity matrix of square root transformed data, which was performed using PRIMER v7 (142). To quantify metabolomic variation within and between coral species, we calculated their multivariate dispersion using the PERMDISP routine (143). PERMDISP calculates the distance of each observation (in this case, each coral sample) to its group centroid (in this case, each coral species) and then compares the average of these distances among groups. It is a multivariate extension of Levene's test, with the P-values obtained using permutations of the raw data. This allowed us to make inferences about the relative size of the clouds in multivariate space within and between coral species. The tests were based on the same transformed data and Bray-Curtis similarity matrix as our PERMANOVA tests. Shannon entropy (144) was calculated for the alpha diversity metric using a Jupyter Notebook. A Bray-Curtis similarity matrix of log-transformed data was constructed using Primer v7 for the beta diversity metric. A principal coordinates analysis (145) was constructed on the Bray-Curtis similarity matrix.

Several in silico tools were used to aid in metabolite annotations. MolDiscovery compares in silico generated MS 2 spectra of small molecules to user-uploaded exper imental MS 2 spectra (134). SIRIUS computes putative chemical formulas based on user-uploaded MS 1 isotopic peaks and MS 2 fragmentation patterns (129). CSI:FingerID transforms MS 2 spectra into predicted structural fingerprints that enable matching to fingerprints generated from structure databases (146). CANOPUS, which predicts the chemical class of metabolites by utilizing CSI:FingerID's predicted structural fingerprints, proposed the chemical class of 449.398_21.3 min as Vitamin E compounds (147). Unsupervised substructure discovery performed through MS2LDA (148) enabled annotations of several classes. Substructure motif 108 consisted of fragmentation peaks characteristic of tocopherol substructure (Fig. S3D). Motif 185 containing characteristic carnitine headgroup fragment peaks (91,149) at m/z 85.028 and 144.102 (Fig. 4D) aided acylcarnitine annotations. Supervised substructure discovery was performed using MassQL, an MS query language platform that outputs metabolite features based on sets of user-defined fragment peaks and neutral losses (133).

Crude coral extracts were partitioned to remove salts and separate compounds based on polarity. First, 3 mL of EtOAc was added to 20 mL scintillation vials containing dry crude extracts. Vials were sonicated to resuspend the extracts for 30-60 s. Three milliliters of H 2 O and another 1 mL of EtOAc was then added and the vials swirled to mix. Vials were then left to separate into distinct layers. The EtOAc layer was transferred via glass pipette into a clean and pre-weighed 20 mL scintillation vial. An additional 2 mL of EtOAC was then added to the crude mixture with water, swirled to mix, and left to separate again. The EtOAc layer was again transferred into the vial containing the EtOAc partition. The EtOAc partitions were then dried via a SpeedVac vacuum concentrator (Thermo Scientific Savant SPD121P) at 35°C. The remaining aqueous extract was then partitioned using n-butanol (BuOH). Approximately 2 mL of BuOH was added to the aqueous extract, to mix, and then left to sit until distinct layers formed. The BuOH partition was then transferred into a clean and pre-weighed 20 mL scintillation vial. Another round of BuOH partitioning was performed by adding an additional 1 mL of BuOH to the aqueous extract, mixing and allowing time for a final separation. The BuOH layer was transferred to the vial containing the initial BuOH partition, which was then dried via rotary-evaporation and analyzed using UPLC-HRMS. The BuOH partitions were resuspended, and LC-MS data were acquired and analyzed following the procedure outlined in "Mass spectrometry data acquisition and analysis, " above.

Coral extracts were tested for antibacterial activity using disk diffusion growth inhibition assays against putative coral pathogens, Vibrio coralliilyticus OfT6-21 and V. coralliilyti cus OfT7-21, Leisingera sp. McT4-56 and Alteromonas sp. McT4-15. To make pathogen lawns, overnight liquid cultures of pathogens were grown by inoculating 2-3 mL of seawater broth (4 g/L tryptone and 2 g/L yeast extract in 0.22 mm filtered seawater) with individual colonies of each strain and shaking culture tubes at 150 RPM and 28°C (Benchmark Incu-shaker 10LR). To coat seawater agar (seawater broth with 15 g/L agar) plates with a pathogen lawn, a 200 mL aliquot of liquid culture (OD600 = 0.5) was added to each plate (150 mm × 15 mm) and spread using sterile glass beads.

Coral partitions were tested by solubilizing partitions in MeOH to a concentration of 6.25 mg/mL and applying 4 µL aliquots to sterile paper disks (Whatman Grade 1) in triplicate (final amount 25 µg/disk). A filter disc with 4 µL of nalidixic acid at 15.62 mg/mL (62.5 µg) was used as a positive control. Negative controls were disks treated with MeOH only. Disks were given time to dry completely and then carefully transferred with sterile forceps to seawater agar plates containing freshly coated pathogen lawns. Disk diffusion plates were then incubated at 28°C overnight. After incubation, zones of inhibition (ZOI) were measured using digital calipers from the edge of the paper disk to the edge of the zone of bacterial growth inhibition.