Global Mass Spectrometric Analysis Reveals Chemical Diversity of Secondary Metabolites and 44-Methylgambierone Production in Philippine Gambierdiscus Strains

Surveillance and characterization of emerging marine toxins and toxigenic dinoflagellates are warranted to evaluate their associated health risks. Here, we report the occurrence of the ciguatera poisoning-causative dinoflagellate Gambierdiscus balechii in the Philippines. Toxin production and chemical diversity of secondary metabolites in G. balechii GtoxSAM092414, G. balechii Gtox112513, and the recently reported Gambierdiscus carpenteri Gam1BOL080513 were assessed using targeted and untargeted UPLC-MS/MS analysis and radioligand receptor-binding assay (RBA). 44-methylgambierone was produced by all three strains, albeit with different levels based on RBA and UPLC-HRMS/MS analysis. The fatty acid composition was similar in all strains, while subtle differences in monosaccharide content were observed, related to the collection site rather than the species. Molecular networking using the GNPS database identified 45 clusters belonging to at least ten compound classes, with terpene glycosides, carbohydrate conjugates, polyketides, and macrolides as major convergence points. Species-specific peptides and polyhydroxylated compounds were identified in G. balechii GtoxSAM092414 and G. carpenteri Gam1BOL080513, respectively. These provide a glimpse of the uncharacterized biosynthetic potential of benthic dinoflagellates and highlight the intricate and prolific machinery for secondary metabolites production in these organisms.


INTRODUCTION
Ciguatera poisoning (CP) is one of the most common foodborne diseases associated with seafood consumption globally, having approximately 10,000-50,000 cases annually (Dickey and Plakas, 2010;Friedman et al., 2017;World Health Organization, 2020). Since 1988, 123 and 274 confirmed and suspected cases of CP, respectively, have been reported in the Philippines (Yñiguez et al., 2021).
A survey of reef fishes from the Visayan and Sibuyan Seas in the Philippines showed that 4.46% were positive for ciguatoxins based on a mouse bioassay (Montojo et al., 2020). The limited surveillance of CP-causative organisms and associated toxins may contribute to underreporting of CP cases in the Philippines.
Benthic dinoflagellates belonging to the genera Gambierdiscus, Coolia, and Fukuyoa are recognized as the toxin producers linked to CP (Berdalet and Tester, 2018;Murray et al., 2020). Grazing by invertebrates and herbivorous fishes allows the benthic dinoflagellates and toxins to enter the food chain (Munday et al., 2017;Holmes et al., 2021). While fishes are the main vectors of CP, mollusks, crustaceans, and echinoderms are also linked to CP (Munday et al., 2017;Holmes et al., 2021).
The associated symptoms of CP, such as gastrointestinal, neurological, and cardiovascular symptoms and paradoxical dysesthesia (temperature reversal) (Bagnis et al., 1979), can be traced to the cellular effects of ladder-shaped polyether toxins collectively called ciguatoxins (CTX). CTX are lipophilic molecules that biotransform and bioaccumulate in fish. To date, more than 50 CTX analogs have been identified from fishes and benthic dinoflagellates, grouped as Pacific (P-CTX), Caribbean (C-CTX), and Indian (I-CTX) (Soliño and Costa, 2018;Chinain et al., 2020). CTX are potent activators of voltage-gated sodium channels and show significant toxicity in mice when given orally (Holmes et al., 2021).
Another class of toxins in benthic dinoflagellates is the water-soluble maitotoxins (MTX) that accumulate in the fishes' digestive tract and liver. The presence of at least one sulfate moiety in the MTX backbone leads to significant hydrophilicity of these toxins and low bioaccumulation in fish flesh. MTX are the most potent marine toxins known to date, although MTX has lower toxicity compared to CTX via an oral route (Shmukler and Nikishin, 2017). The low oral bioavailability and bioaccumulation of MTX suggest that these compounds are not the main contributor to CP symptomatology (Holmes et al., 2021). To date, there are seven known MTX analogs, and their biological activity is varied (Estevez et al., 2020a. MTX-1, MTX-2, and MTX-4 cause massive Ca 2+ influx leading to cell death (Estevez et al., 2020a). MTX-3 causes a similar phenotype to CTX, although with lower potency (Boente-Juncal et al., 2019).
Among the MTX congeners, only MTX-1 have an assigned structure based on mass spectrometry (MS) and nuclear magnetic resonance (NMR) spectroscopy. MTX-1 is the largest nonpeptide toxin identified, with a molecular weight of 3425 Da, with 32 cyclic ether rings Sasaki et al., 1996;Shmukler and Nikishin, 2017). Structural analysis of MTX-3 showed that it is smaller than MTX-1, with a mass of 1039 Da . Several groups undertook a targeted purification of MTX-3 to determine the structure. The purified material was analyzed by NMR and MS and indicated that MTX-3 is the 44-methyl analog of gambierone (Boente-Juncal et al., 2019;Murray et al., 2019). The difference in bioactivity, approximately 1,000-fold lower potency of 44-methylgambierone than the original data for partially purified MTX-3, may indicate a potential divergence in structure between MTX-3 and 44methylgambierone (Holmes et al., 2021). Holmes and Lewis (1994) observed a potent toxic effect of partially purified MTX-3 when administered intraperitoneally in mice Lewis et al., 1994). Purified 44-methylgambierone had lower toxicity, with LD 50 of 20-38 mg/kg via intraperitoneal administration (Murray et al., 2020). Additional studies are necessary to ascertain the divergence in biological activity and structure between MTX-3 and 44-methylgambierone.
The complex structure of CTX and MTX and the limited biomass of dinoflagellate producers make the complete structural assignment and biological activity assessment of these toxins challenging. Several groups have implemented hyphenated liquid chromatography-mass spectrometry (LC-MS) to aid in toxin discovery. By using an LC-MS-based workflow, it is easier to screen extracts of benthic dinoflagellates for potential new congeners, and known CTX and MTX can be quantified and identified (Caillaud et al., 2010 and references cited therein;Chinain et al., 2010;Munday et al., 2017;Longo et al., 2019;Estevez et al., 2020aEstevez et al., , 2021Murray et al., 2020;Tibiriçá et al., 2020;Gago-Martínez et al., 2021 and references cited therein;Mudge et al., 2021). UPLC-HRMS profiling of 252 marine microalgae belonging to 32 genera showed that 44-methylgambierone production is ubiquitous in eight Gambierdiscus species (Murray et al., 2020). The benthic dinoflagellate Coolia and Fukuyoa were likewise producers of 44-methylgambierone (Murray et al., 2020), with a putative new analog likely to be present in Coolia (Tibiriçá et al., 2020). While 44-methylgambierone has low toxicity via the oral route, UPLC-HRMS quantitation showed significant amounts in the producer dinoflagellate and snappers, suggesting potential contributions to CP. The intracellular concentration of 44-methylgambierone in Gambierdiscus sp. is 5.8-74 pg MTX-1 eq cell −1 (Longo et al., 2019), while bioaccumulation in snapper liver and muscles can increase the concentration (Kohli et al., 2014).
Capitalizing on the improved methodology for mass spectrometry, untargeted metabolite analysis has become a mainstay tool to analyze chemical diversity in organisms. The Global Natural Products Social Molecular Networking Platform (GNPS) is based on spectral alignment to assess the similarities and relationships among molecules  and visualized using a molecular network. While GNPS is mainly used for biodiscovery of natural products (Teta et al., 2015;Naman et al., 2017;Ding et al., 2018;Via et al., 2018), it has been recently applied to dinoflagellate metabolites and demonstrated the unique metabolites from these organisms (Fiorini et al., 2020;Wu et al., 2020;Sibat et al., 2021). Molecular networking of five Dinophysis species identified the characteristic toxin profile for each strain and identified five new putative pectenotoxins . Differences in the metabolites of Pseudonitzchia during the reproductive stages were evident from the GNPS-based analysis (Fiorini et al., 2020). Compounds with unprecedented chemical scaffolds in Pseudo-nitzchia extracts, not represented in reference databases, were observed (Fiorini et al., 2020). Improved qualitative and quantitative screening of okadaic acid and dinophysistoxins was achieved through molecular networking of extracts from Prorocentrum lima (Wu et al., 2020). New esters of okadaic acid and dinophysistoxins were also identified (Wu et al., 2020). A scientometric analysis on dinoflagellates research recommended using metabolite databases such as Dictionary of Natural Products, AntiBase, MassBank, and GNPS to share metabolites information to advance biomolecule discovery (Oliveira et al., 2020).
In this study, we looked at the biosynthetic potential of Gambierdiscus carpenteri and two Gambierdiscus balechii strains from the Philippines to produce toxins and other classes of secondary metabolites using high resolution mass spectrometry (HRMS) and the GNPS molecular networking platform. We probed the chemistry of the three Gambierdiscus strains and obtained insights into the production of ladder-shaped polyether toxins and other classes of compounds.

Culture Condition and Morphological Observation
Three monoclonal cultures of Gambierdiscus spp. from The Marine Science Institute Red Tide Laboratory were used in this study ( Table 1). Cultures were routinely maintained in filtered natural seawater (30 psu) supplemented with full strength IMK medium (WAKO, Tokyo, Japan) at 25 ± 2 • C under 100 µmol photons m −2 s −1 of light using 40 W white fluorescent lamps, with a 12:12 h light:dark photoperiod. Large-scale 1-L cultures in Fernbach flasks were harvested after 21 days in the exponential phase by centrifugation at 4,000 × g for 15 min at 4 • C (Thermo Fisher Scientific).

Molecular Analyses
Genomic DNA was extracted from exponentially growing cultures using ISOLATE II Plant DNA Kit (Bioline, London, United Kingdom) following the manufacturer's procedure. The D8-D10 region of large subunit ribosomal DNA (LSU rDNA) gene was amplified using the primers D8F and D10R (Litaker et al., 2009). Amplification was conducted in 50 µL reaction mix containing 45 µL of PCR supermix (Invitrogen, California, United States), 2 µL of each primer and 1 µL of template DNA. The thermal condition of PCR was as follows: initial denaturation step at 95 • C for 2 min, followed by 35 cycles of 94 • C for 30 s, 60 • C for 2 min, and 72 • C for 60 min, and finally an elongation step of 72 • C for 5 min. Reaction was performed in a T100 Thermal cycler (Biorad, California, United States). Amplicons were purified using QIAquick Gel Purification Kit (Qiagen) and sent to 1st Base (Malaysia) for sequencing. DNA sequences were aligned using MAFFT v7.110 (Katoh and Standley, 2013) with taxa downloaded from GenBank. The multiple sequences were manually edited and/or trimmed using BioEdit Sequence Alignment Editor v7.2.5 (Hall, 1999) with 58 selected taxa of Gambierdiscus comprising 17 species. For outgroups, four Fukuyoa spp., two Alexandrium spp., Akashiwo sanguinea and Prorocentrum micans were used. Maximum likelihood (ML) analysis was performed using PhyML (Guindon et al., 2010) with 500 bootstrap replicates. The best fitting substitution model for the ML tree as selected by the Smart Model Selection (SMS) program (Lefort et al., 2017) was general time reversible (GTR) with gamma distribution (G = 0.301) plus proportion of invariable sites (I = 0.881). Bayesian inference (BI) was computed via MrBayes v3.1.2 (Ronquist and Huelsenbeck, 2003) using a Metropolis-coupled Markov chain Monte Carlo run for 10 million generations with sampling at every 100 iterations. The best-fit substitution model for the BI tree, selected by jModelTest 2.1.10 (Darriba et al., 2012), was TIM3 + G (0.899) + I (0.3000). GenBank accession numbers are provided in the phylogenetic trees.

Chemical Extraction
Gambierdiscus isolates were extracted according to the procedure of Murata et al. (1990). Lyophilized biomass from 1 L cultures were extracted using three volumes of acetone to yield the crude extracts. A portion of the crude extract (20.0 mg) was further fractionated on a Florisil R SPE-cartridge and eluted with 4:1 n-hexane:acetone (fraction A), 9:1 acetone:methanol (fraction B), and 1:1 acetone:methanol (fraction C). Fractions were dried and stored at -20 • C until further analysis.

Toxicity Assessment
Dried crude extracts were resuspended in methanol and subjected to radioligand receptor-binding assay (RBA) based on the IAEA-Tecdoc-1729 for the detection of ciguatera toxins (IAEA, 2013). The test kit for RBA consisting of tritium-labeled brevetoxin-3 ([ 3 H]PbTx-3), unlabeled PbTx-3 and porcine brain membrane was acquired from American Radiolabeled Chemicals. A 15 nM working solution of [ 3 H]PbTx-3 was prepared by diluting with the assay buffer. Solutions of unlabeled PbTx-3, with final assay concentrations of 0.01 ng/mL to 1.0 µg/mL in half-log dilutions, were prepared for the CTX calibration curve. The assay was performed in a 96-well microtiter filter plate with FB glass fiber filter (0.65 µm pore size) by adding the following solutions in order: 35 µL assay buffer, 35 µL PbTx-3 standard solutions or extracts, 35 µL [ 3 H]PbTx-3 working solution, and 105 µL porcine membrane homogenates. Plates were covered and incubated at 4 • C for 1 h, filtered using a vacuum manifold and washed with 200 µL ice-cold assay buffer (2 ×). A 50 µL aliquot of the scintillation cocktail (OptiPhase, PerkinElmer) was added to each well and the reaction plate was further incubated for 1 h at room temperature. Plates were counted using a microplate scintillation counter (MicroBeta R , PerkinElmer). Curve fitting of the PbTx-3 standards was performed using a four-parameter logistic fit (Sigmoidal, 4PL) with variable slope. Limit of detection for the assay was 2.0 ×10 −5 g PbTx-3 eq./g. Results are presented as pg PbTx-3 eq./cell by normalizing the acquired RBA values to the total cell counts of the cultures.

Carbohydrate Extraction and Analysis
Gambierdiscus biomass was subjected to a two-stage sulfuric acid extraction according to the method of Templeton et al. (2012) to yield the total carbohydrates. The hydrolyzed carbohydrates were identified and quantified using the method of Schulze et al. (2017) with modifications. Monosaccharides were separated using an Acquity UPLC BEH Amide column, 1.7 µm, 2.1 × 50 mm (Waters) by a gradient program of acetonitrile/5 mM ammonium formate in water (both with 0.1% formic acid modifier): 90-75% acetonitrile for 8.5 min, and 75% acetonitrile for 4 min. Detection was done by multiple reaction monitoring (MRM) analysis (Shimadzu LCMS-8040). The optimized transitions for each standard are provided in Supplementary Table 1.
Data analysis was performed through manual peak integration using LabSolution (Shimadzu). Individual monosaccharide standards (Sigma) were prepared (0.156 µg/mL to 0.0098 µg/mL) in two-fold serial dilutions and injected three times for repeatability to generate the external calibration curves. Results are presented as % w/w (mg sugar/mg biomass), based on two biological replicates with three technical replicates each.

Lipid Extraction and Analysis
Solid-liquid extraction of the Gambierdiscus biomass was done according to the method of Bligh and Dyer (1959) to yield the chloroform soluble lipid extract. Fatty acid methyl esters (FAMEs) were prepared based on the AOAC official method 969.33 (AOAC, 2000b). Resulting FAMEs were separated and analyzed using gas chromatography with flame ionization detector (Shimadzu GC-2010) through the AOAC Official Method 963.22 (AOAC, 2000a). Chromatographic separation was done on a Supelco SP-2560 capillary column (0.25 µm, 100 m × 0.20 mm). Retention times of the eluted FAMEs were compared with known amounts of mixed reference standards composed of C6, C8, C10, C12, C14, C16, C18, C18:1, C18:2, C18:3, C20, C22, and C24. Peak areas were integrated using GCSolution (Shimadzu), and the response factors were calculated from the ratio between the peak area of the individual FAME and the internal standard (heneicosanoic acid methyl ester). Analysis was done using two biological replicates with three technical replicates each. Results are presented as % w/w (mg fatty acid/mg biomass).

Secondary Metabolites Analysis Using Global Natural Products Social Molecular Networking Platform
UPLC-MS/MS analysis of all Florisil R fractions (1 mg/mL in acetonitrile) was performed using a Waters Acquity UPLC R H-Class System with a Xevo R G2-XS Quadrupole Time-of-Flight (QToF) high-resolution mass spectrometer. A 1 µL aliquot of each sample was injected in a Phenomenex Kinetex 2.6 µm C18 100Å column (50 × 2.1 mm) and eluted at 0.35 mL/min, using a gradient program of acetonitrile/water (0.1% formic acid modifier): 40-100% acetonitrile for 5.5 min, and 100% acetonitrile for 2 min. The mass spectrometer was set to observe at m/z 100-2,000 in positive ESI mode with an automated data dependent acquisition (DDA) MS/MS scan. Three DDA scans were acquired for each sample with increasing ramp collision energies: 15-25, 30-45, and 50-70 eV. DDA scan for the Pacific ciguatoxin 3C (P-CTX-3C, WAKO Chemicals) standard was acquired using the 15-25 eV ramp collision energy.
Chromatograms were converted to mzxml format using freely available MSConvert software 1 . A molecular network was created using the online workflow 2 on the GNPS website 3 . MS/MS data of the Florisil R fractions from each isolate were grouped to form the molecular network. The precursor ion mass tolerance was set to 2 Da and an MS/MS fragment ion tolerance of 0.1 Da. A network was created, with edges set to cosine score >0.7 and more than six matched peaks. Further, edges between two nodes were kept in the network if and only if each of the nodes appeared in each other's respective top 10 most similar nodes. Finally, the maximum size of a molecular family was set to 100, and the lowest scoring edges were removed from the molecular families until the molecular family size was below this threshold. The spectra in the network were searched against the GNPS spectral libraries. All matches kept between network spectra, and library spectra were required to have a score >0.7 and at least six matched peaks.

Taxonomic Identification
Morphology Morphological characters (Figure 1) of the two Gambierdiscus strains (Gtox112513 and GtoxSAM092414) were identical to Gambierdiscus balechii (Fraga et al., 2016;Azanza et al., 2017). Cells were anterio-posteriorly compressed, measuring 53.6-63.5 µm (57.7 ± 2.7, n = 30) in depth (dorsoventral diameter) and 56.7-63.0 µm (60.1 ± 1.9, n = 30) in width. Cell shape was round to ellipsoid in apical view (Figures 1D,E). The epithecal and hypothecal surfaces were heavily areolated (Figures 1A,C,  2D). The presence of small and large accumulation bodies was observed ( Figure 1B). The nucleus was large and elongated ( Figure 1D). Cells had a rod-shaped, golden-brown chloroplast evenly distributed throughout the cell ( Figure 1F). Thecal plates of the two strains as observed with fluorescence microscopy, especially the taxonomically informative plates of the genus such as the shape of the second apical plate (2 ), third precingular plate (3 ), and second antapical plate (2 ), support their identification as G. balechii. Thecal plate tabulation of the Gambierdiscus strain (Gtox112513) is only shown (Figure 2). The epitheca was composed of apical pore (Po), three apical plates and six precingular plates. In apical view, the first apical plate (1 ) and sixth apical plate (6 ) were barely visible (Figure 2A) but can be seen at ventral view ( Figure 2C). Both plate 1 and 6 were the smallest among the apical and precingular plates, respectively. The second apical plate (2 ) was the largest of the apical series and shaped like a hatchet, i.e., having a suture ratio of 2 /1 and 2 /3 = 0.60 (n = 30). The third precingular plate (3 ) was asymmetrical (Figure 2A). The apical pore plate was oval with a fish-hook shaped slit (Figures 1A,D). The hypotheca was composed of two antapical plates (2 ), five postcingular plates (5 ), and a posterior sulcal plate (Sp). Among the five postcingular plates, the plate 4 was the largest. The second antapical plate (2 ) was narrow and pentagonal to trapezoidal ( Figure 2B). Marginal borders of the plates in the epitheca and hypotheca were overlapping (Figures 2A,B).

Carbohydrate Analysis
The total carbohydrate content of the three isolates was quantified using a colorimetric phenol-sulfuric acid assay (Masuko et al., 2005). G. carpenteri Gam1BOL080513 and G. balechii GtoxSAM092414 have comparable total carbohydrate content ( Figure 5A). G. balechii Gtox112513 showed the highest total carbohydrate content with 12.66 ± 1.92% w/w.
Carbohydrates were hydrolyzed according to the method of Schulze et al. (2017). Multiple reaction monitoring of the monosaccharides facilitated the identification and quantitation of the sugars. Galactose, mannose, and glucose were the major monosaccharides in the hydrolysates of the three Gambierdiscus isolates, with glucose being the principal monosaccharide ( Figure 5A). G. balechii GtoxSAM092414 and G. balechii Gtox112513 gave the highest glucose and galactose content, respectively ( Figure 5A). The mannose content significantly varied across the three isolates, with the highest amount in G. balechii Gtox112513 (Figure 5A).

Secondary Metabolites Analysis
Untargeted metabolite profiling of the extracts was performed using reversed phase UPLC-MS and MSMS analysis and subsequently annotated using the GNPS database to identify the compounds and toxins produced by the three Gambierdiscus isolates. Figure 6 shows the molecular network generated from the extracts of the three strains of Gambierdiscus and P-CTX-3C (in lavender) as a seed compound with a known identity. The entire molecular network consists of 541 nodes with 666 connections and 45 clusters with 271 single nodes. Cosine scores ranged from 0.70 to 0.96, suggesting that the putative hits from the GNPS database and the metabolites in the Gambierdiscus extracts may not be identical but would likely have the same chemical scaffold. Hence, clusters were annotated based on their common functional groups. Based on this, 45 clusters were generated belonging to at least ten compound classes (Figure 6).
of G. balechii GtoxSAM092414 and polyhydroxylated (22, 23) compounds from G. carpenteri Gam1BOL080513 (Figure 6). The seed compound P-CTX-3C appears in the molecular network as a single node (Figure 6 inset). Manual annotation of the single nodes in the molecular network revealed two precursor masses, 1021.65 and 1021.8 Da, in the three extracts (Figure 6 inset) that clustered together. Examination of the chromatograms showed that these ions have identical retention times ( Figure 7A) and MS 1 profiles ( Figure 7B) showed similar fragmentation patterns, therefore, corroborating with the observed pseudomolecular ion. 44-methylgambierone was detected in fraction B of the G. balechii strains, and in fraction C of G. carpenteri Gam1BOL080513 (Supplementary  Figures 3-5). The relative amount of 44-methylgambierone was semi-quantified by integrating the peak area corresponding to 44-methylgambierone (t R = 0.78-0.90 min) along with the solvent blank ( Figure 7C). G. balechii Gtox112513 showed the highest amount of 44-methylgambierone compared to G. balechii GtoxSAM092414 and G. carpenteri Gam1BOL080513.
With the analyses of phylogenetic position inferred from D8-D10 LSU rDNA sequences, the morphological identification of the two Gambierdiscus strains as Gambierdiscus balechii was further resolved by forming a well-supported clade that includes the sequence of the type material of G. balechii from Celebes Sea, Manado, Indonesia (KX268470). This clade is also composed of strains from Rawa Island, Malaysia (KY235255) and another strain originally identified as Gamberidiscus type 6 from Marakei, Kiribati (KJ125112, KJ125113) but now designated as Gambierdiscus balechii (Dai et al., 2017). The identification of Gambierdiscus balechii adds to the report of this species in the tropical Pacific (Fraga et al., 2016;Zhang et al., 2016;Dai et al., 2017;Tester et al., 2020).

Toxicity
The toxicity of G. carpenteri Gam1BOL080513 (12.36 ± 4.38 pg PbTx-3 eq./cell) was comparable to the highest toxin content reported by Vacarizas et al. (2018) at 7.48 ± 0.49 pg PbTx-3 eq./cell. The difference may be attributed to the extraction procedure performed in this study, using acetone instead of methanol. Compared to other G. carpenteri strains in literature Pisapia et al., 2017;Díaz-Asencio et al., 2019), the Philippine strain showed relatively higher toxin content. Several Gambierdiscus species associated with macrophytes identified in Cuba, including G. carpenteri, have toxin content below the RBA limits of quantitation (Díaz-Asencio et al., 2019). G. carpenteri strains isolated from Hawaii , the Caribbean , and Mexico  showed femtogram levels of CTX-3C eq./cell using cellbased neuro-2a assay (CBA-N2a).
The toxin content in the Philippine strains (0.13-0.79 pg PbTx-3 eq./cell) of G. balechii was also higher than the literature values. G. balechii strains from Marekei, Kiribati have toxin values ranging from 1.1 to 19.9 fg P-CTX-1 eq./cell (Dai et al., 2017), FIGURE 5 | Monosaccharide (A) and fatty acid (B) content of crude carbohydrate and lipid extracts of G. carpenteri Gam1BOL080513, G. balechii GtoxSAM092414, and G. balechii Gtox112513 biomass. The three strains showed significant difference in the monosaccharide content but not in the fatty acid composition. Results are presented as mean + SD from two biological replicates with three technical replicates each. Different letters indicate significant differences after one-way ANOVA analysis (P < 0.05).
while the Indonesian strain showed 3.4 ± 1.5 fg CTX-3C eq./cell  using CBA-N2a assay. However, a direct comparison cannot be established as the method for toxin determination, and the toxin standards used in these assays differed.

Chemical Analyses
An integrated targeted and untargeted metabolite profiling approach was done to comprehensively characterize the chemistry of the three Gambierdiscus strains from the Philippines. Monosaccharides and fatty acids were profiled and quantified directly using HPLC-MS/MS and GC-FID, respectively. Secondary metabolites were assessed using unbiased UPLC-MS and MS/MS techniques. Florisil R fractionation of the crude extracts was performed to improve the detection of compounds produced at low amounts and, therefore, allowed for higher metabolite coverage in the UPLC-MS analysis. We expand and report the first chemical analysis of Gambierdiscus spp. using the GNPS platform. We focused on molecular clusters with generated GNPS annotation and manually annotated toxin clusters.
Metabolites from G. balechii strains were concentrated in the more lipophilic Florisil R fractions A and B, while G. carpenteri Gam1BOL080513 metabolites showed higher richness in the hydrophilic Florisil R fraction C. The molecular network suggests that the Gambierdiscus strains produce metabolites with high chemical diversity, including carbohydrates, lipids, and polyketides. Exploration of the molecular network (Figure 6) identified strain-specific clusters. For example, a cluster of polyhydroxylated compounds (22, 23, 38-42) was only observed FIGURE 6 | Molecular network generated from HRMS/MS spectra from the Florisil R fractions of three Gambierdiscus strains, and P-CTX-3C standard using Global Natural Product Social Molecular Networking (GNPS) tool. Each node corresponds to a compound in the given strain (indicated by the color/s of the node). The nodes were color-coded according to the strains: G. carpenteri Gam1BOL080513 in blue, G. balechii GtoxSAM092414 in orange, G. balechii Gtox112513 in green, and the P-CTX-3C standard in lavender. Similarity between nodes is represented by the gray curves labeled by the cosine score. Annotation of compound class or family was performed through matching of fragmentation patterns of known compounds available in the GNPS server. Enlarged inset are manually annotated toxins, authentic standard (P-CTX-3C) and 44-methylgambierone with their chemical structures. The three Gambierdiscus strains contain 44-methylgambierone and chemically diverse secondary metabolites. A high-resolution image of the molecular network is provided in the Supplementary Material ( Supplementary  Figure 1). in G. carpenteri Gam1BOL080513 (Figure 6). The two G. balechii strains showed related chemistries, evident from the clustering of their secondary metabolites (Figure 6 highlighted in orange) and consequently, high convergence. These speciesand strain-specific clusters may be potential chemotaxonomic markers for identifying particular Gambierdiscus isolates. Further studies toward identifying these compounds and spatiotemporal variations on the production are needed to define the potential utility as chemotaxonomic markers.

Carbohydrates and Derivatives
The central cluster in the molecular network, which is a convergence of the three strains, contained annotated compounds belonging to terpene glycosides (2, 3) and carbohydrate conjugates (4-6). These are primarily carbohydrate-derived compounds connected to a lipophilic functional group. Based on the molecular network, these compounds are produced mainly by the two G. balechii strains.
Components of structural and storage polysaccharides were mainly detected through the targeted HPLC-MS-MRM approach ( Figure 5A). The monosaccharides in the three Gambierdiscus strains are typical for microalgae (Templeton et al., 2012;Ortiz-Tena et al., 2016;Schulze et al., 2017). Glucose comprises the majority of the biomass since the main polysaccharide in dinoflagellate theca is cellulose (Okuda, 2002), and the primary storage polysaccharide for photosynthetic dinoflagellates is starch  (Metting, 1996;Chen et al., 2013). Mannose and galactose were also reported as components of structural polysaccharides in dinoflagellates (Lewin et al., 1958;Brown, 1991). It is also the main organic component of mucilage found in the dinoflagellate Gonyaulax hyalina (MacKenzie et al., 2002) and a mixture of four phytoplanktonic taxa (Metaxatos et al., 2003). Benthic dinoflagellates, similar to Gambierdiscus, are known to produce mucilage as a means of attachment to their hosts (Heil et al., 1993;Rains and Parsons, 2015). In terms of the individual monosaccharide concentration, the glucose content was similar for the strains from Pangasinan, Philippines (Gam1BOL080513 and Gtox112513). A significant difference was observed for the strain from Eastern Samar, Philippines (GtoxSAM092414). In contrast, there was no evident relationship between the carbohydrate content and species suggesting spatial dependence. The carbohydrate content obtained in this study was significantly lower than other microalgae (Templeton et al., 2012;Ortiz-Tena et al., 2016;Schulze et al., 2017) and may be due to the partial hydrolysis of the complex carbohydrates using the sulfuric acid method.
In particular, the 4-methyl sterol, dinosterol, has been used as a biomarker in marine sediments to determine past dinoflagellate blooms (Boon et al., 1979). Marine-derived terpenes are also metabolites of particular interest because of their wide range of bioactivity and structural diversity (Gross and König, 2006).
Though most studies for these terpenes were in sponges and corals, symbiotic dinoflagellates particularly Symbiodinium sp. were identified as the actual producers of some of these compounds (Mydlarz et al., 2003). Three lipid clusters were observed in the extracts of G. carpenteri and G. balechii (Figure 6).  were putatively identified in the lipid and lipid-like molecules cluster. These compounds are common in all three strains. Other steroidal derivatives, mainly steroidal ketones (19,20), also clustered together and were specific to the two G. balechii strains. Steroidal ketones have been recognized as possible intermediates in the diagenesis of sterols in sedimenting particulate matter and marine sediments (Gagosian and Smith, 1979;Gagosian et al., 1982).
Free fatty acids clustered together for the three strains ( Figure 6) and corroborated with the GC-FAME analysis. The hydrolyzed fatty acids from lipids were similar across the three isolates ( Figure 5B). One of the primary fatty acids detected was C18:1, a precursor of docosahexaenoic acid (DHA) and eicosapentanoic acid (EPA). The C18:1 fatty acid was previously suggested as fatty acid biomarker for dinoflagellates and classified as essential fatty acids to grazers (Joseph, 1975;Parrish et al., 2000). PUFAs, characteristic of benthic dinoflagellates (Usup et al., 2008), were barely detected in the FAME analysis because of the lack of standards. GNPS was able to annotate PUFAs 20hydroxy-eicosatetraenoic acid (14, cosine score 0.84), 9-hydroxyoctadecatrienoic acid (15, cosine score 0.84), and DHA (16, cosine score 0.78) in the fatty acid cluster. Detection of C24:0 in all strains may also indicate the presence of very long chain PUFAs, which is hypothesized to be the precursor of C28 fatty acids octacosaheptaenoic [28:7(n-6)] and octacosaoctaenoic acid [28:8(n-3)] found in dinoflagellates (Leblond and Chapman, 2000). Very long chain PUFAs serve as a distinguishing feature of marine algae from higher plants (Harwood, 1998).

Polyketides
Since GNPS molecular networking is based on similarities among molecules, we used P-CTX-3C as a seed compound to cluster other structurally related molecules belonging to the CTX and MTX families of compounds. Surprisingly, P-CTX-3C and 44methylgambierone appeared as single nodes (Figure 6 inset).
The 44-methylgambierone cluster (Figure 6 inset) showed the pseudomolecular ion [M+H-H 2 O] + with a mass of 1021 Da. G. carpenteri Gam1BOL080513 and G. balechii Gtox112513 formed a single node for 44-methylgambierone with precursor mass of 1021.65 Da. G. balechii GtoxSAM092414 clustered with a separate node with precursor mass 1021.8 Da. The mass difference between 44-methylgambierone in the three Gambierdiscus species may be attributed to the mass error from the instrument rather than structural variation. Manual inspection of the MS and MS/MS spectra corresponding to these nodes showed similar m/z peaks (Figures 7A,B and Supplementary Figure 2) and retention time, indicating that the 44-methylgambierone in the three species are identical. Since 44-methylgambierone has an equivalent mass to 2,3-dihydroxyCTX-3C, the presence of a sulfate moiety is important to verify the maitotoxin backbone. The MS/MS fragmentation (Supplementary Figure 2) confirmed the presence of a sulfate moiety. 44-methylgambierone is the main MTX detected in this study presumably due to the high intracellular concentration of this compound in the three Gambierdiscus species.
However, the amount of 44-methylgambierone did not fully account for the observed toxicity in the RBA assay ( Figure 7C) for G. carpenteri Gam1BOL080513. Since 44-methylgambierone has a low potency, the high ciguatoxicity in the RBA assay for G. carpenteri Gam1BOL080513 may suggest the production of additional toxins. Accordingly, the low RBA activity observed for G. balechii Gtox112513 and G. balechii GtoxSAM092414 may imply that 44-methylgambierone is the principal marine biotoxin in these two strains.
Detection of other MTX and CTX in G. carpenteri Gam1BOL080513 was attempted using multiple reaction monitoring (Supplementary Methods and Supplementary  Table 3), as adapted from Estevez et al. (2021). Apart from 44-methylgambierone, other MTX and CTX analogs were not detected in G. carpenteri Gam1BOL080513 (Supplementary  Figure 6). The targeted monitoring also validated the results from the untargeted metabolite analysis.
Other MTX and CTX congeners are usually present in very low amounts, limiting the detection of the parent mass under positive ionization. Enrichment of G. carpenteri Gam1BOL080513 through column fractionation and subsequent targeted analysis for CTX and MTX may aid in the detection of other marine biotoxins produced by this species. More sensitive methods for mass spectrometry detection, such as multiple reaction monitoring, have proved to be successful in detecting minor CTX and MTX in benthic dinoflagellate extracts. Molecular networking in the negative ionization mode may improve the detection limit by capitalizing on the presence of a sulfate moiety in MTX.

Other Compound Classes
Pigments were also detected in the extracts of the three Gambierdiscus strains. Pheophorbide A (18) and pheophytin A (17), chlorophyll breakdown products, were putatively identified (cosine scores 0.80 and 0.87, respectively) in the chlorophyll derivatives cluster. Peridinin (10, cosine score 0.86) was also annotated in the polyketide cluster. Peridinin is an accessory pigment of several dinoflagellates, present as peridinin-chlorophyll-protein light harvesting complex used in photosynthesis (Song et al., 1976;Hofmann et al., 1996;Jiang et al., 2012). This pigment is also responsible for the characteristic orange to brown appearance of Gambierdiscus cells (Indelicato and Watson, 1986).
The targeted and untargeted chemical profiling and molecular networking of Philippine Gambierdiscus strains provided a glimpse of these organisms' biosynthetic potential and chemical diversity. At the same time, we gained insights into the similarities and differences in the metabolite production in the three Gambierdiscus strains. While the production of fatty acids, carbohydrates and 44-methylgambierone are common among the three strains, we observed speciesand strain-specific compounds. There is much more to marine toxins in the extracts of Gambierdiscus sp. The putative identification of polyhydroxylated and peptidebased compounds may suggest additional allelochemicals and potentially high value compounds from Gambierdiscus species. The putative identification from untargeted chemical profiling can guide targeted purification and chemical characterization of Gambierdiscus extracts.

DATA AVAILABILITY STATEMENT
The datasets presented in this study can be found in online repositories. The names of the repository/repositories and accession number(s) can be found below: https://www.ncbi.nlm. nih.gov/genbank/, OL437111; OL437112 https://massive.ucsd. edu/ProteoSAFe/static/massive.jsp, MSV000088378.

AUTHOR CONTRIBUTIONS
ZM: conceptualization, methodology, software, formal analysis, investigation, visualization, writing -original draft, review, and editing. GB: methodology, formal analysis, writingoriginal draft, review, and editing. JDC: formal analysis, resources, writing -review, and editing. JB: methodology, formal analysis, writing -review, and editing. MR and RA: resources, writing -review, and editing. LS-R: conceptualization, methodology, writing -original draft, review and editing, supervision, resources, project, and funding acquisition. All authors have read and agreed to the published version of the manuscript.