MS/MS-Based Molecular Networking Approach for the Detection of Aplysiatoxin-Related Compounds in Environmental Marine Cyanobacteria

Certain strains of cyanobacteria produce a wide array of cyanotoxins, such as microcystins, lyngbyatoxins and aplysiatoxins, that are associated with public health issues. In this pilot study, an approach combining LC-MS/MS and molecular networking was employed as a rapid analytical method to detect aplysiatoxins present in four environmental marine cyanobacterial samples collected from intertidal areas in Singapore. Based on 16S-ITS rRNA gene sequences, these filamentous cyanobacterial samples collected from Pulau Hantu were determined as Trichodesmium erythraeum, Oscillatoria sp. PAB-2 and Okeania sp. PNG05-4. Organic extracts were prepared and analyzed on LC-HRMS/MS and Global Natural Product Social Molecular Networking (GNPS) for the presence of aplysiatoxin-related molecules. From the molecular networking, six known compounds, debromoaplysiatoxin (1), anhydrodebromoaplysiatoxin (2), 3-methoxydebromoaplysiatoxin (3), aplysiatoxin (4), oscillatoxin A (5) and 31-noroscillatoxin B (6), as well as potential new analogues, were detected in these samples. In addition, differences and similarities in molecular networking clusters related to the aplysiatoxin molecular family were observed in extracts of Trichodesmium erythraeum collected from two different locations and from different cyanobacterial species found at Pulau Hantu, respectively.


Introduction
Marine filamentous cyanobacteria are prolific sources of novel secondary metabolites, and a number of these compounds possess significant biological activity, having agricultural and pharmaceutical applications [1,2]. A notable example is a series of highly cytotoxic lead compounds belonging to the dolastatin class of molecules [3]. For instance, dolastatin 10 and synthetic derivatives, soblidotin (TZT-1027) and tasidotin, entered clinical trials in the 1990s as anticancer agents but dropped out in Phase II due to toxicity and/or lack of efficacy in trials [4]. A synthetic analogue, monomethyl auristatin E, was subsequently formulated as an antibody-drug conjugate, brentuximab vendotin Mar. Drugs 2018, 16, 505 2 of 15 (Adecetris ® ), for the treatment of Hodgkin's lymphoma and systemic anaplastic large cell lymphoma in 2012 [5].
In recent years, various studies revealed that certain marine cyanobacterial strains produce compounds that concurrently provide beneficial and detrimental effects to human health. One such compound is the polyketide-based cyanotoxin, aplysiatoxin. Cyanotoxins collectively cause a wide spectrum of health conditions, ranging from dermatitis to severe carcinoma [6][7][8]. Aplysiatoxin and its related compounds, which includes oscillatoxin A and nhatrangins, have been reported from Lyngbya majuscula, Oscillatoria nigro-viridis and Trichodesmium erythraeum [6]. The tumor promoting activity of aplysiatoxin and debromoaplysiatoxin is attributed to their specific molecular roles as protein kinase C activators. Interestingly, a simplified synthetic analogue of aplysiatoxin, aplog-1, was recently reported to possess anti-cancer properties with mode of action similar to bryostatin 1 [9][10][11][12][13]. In addition, studies revealed several aplysiatoxin-related compounds to display therapeutic activities, such as anti-Chikungunya and modulation of potassium ion channels [14,15]. A recent study by Richard et al. demonstrated that aplysiatoxin and its debrominated analogue induced expression of latent HIV-1 provirus in both cell line and primary cell models [16].
Due to the varied biological activities of the aplysiatoxins, they make good candidates for biomedical research as well as in their detection and monitoring of toxins during algal bloom occurrence. Given the correlation between global warming and occurrence of cyanobacterial blooms, the incidence of illness caused by contact with cyanotoxins can potentially be increased [8].
In contrast to research on hepatotoxins (e.g., microcystins and nodularin) as well as other neurotoxins (e.g., β-N-methylamino-l-alanine), very little work has been conducted on the development of detection methods of other cyanotoxins, such as the aplysiatoxins [17]. In this report, we employed MS/MS-based molecular networking approach for the rapid detection of aplysiatoxin class of molecules from environmental marine cyanobacterial samples obtained from four intertidal areas in Singapore. The detection method, based on either HPLC-HRMS/MS or UPLC-HRMS/MS combined with data analysis using Global Natural Product Social Molecular Networking (GNPS), led to the detection of both known and new analogues related to the aplysiatoxin class of cyanotoxins.

MS/MS-Based Molecular Networks of Marine Cyanobacterial Extracts
Organic crude extracts from each of the cyanobacterial samples were prepared and analyzed initially by UPLC-HRMS/MS Waters system (Massachusetts, USA). In addition, organic crude extract from TLT/PSK/001 were analyzed on both HPLC-HRMS/MS Thermo-Finnigan system (Massachusetts, USA) (TLT/PSK/001a) and UPLC-HRMS/MS Waters system (TLT/PSK/001b). Mass spectrometric data obtained from these analyses were then used to generate molecular networking clusters using the online platform Global Natural Products Social Molecular Networking (GNPS). Debromoaplysiatoxin (1) (Figure 2), isolated from the extract of TLT/PHC/001, was used as seed molecule in the analysis (HRESIMS of sodiated molecular ion m/z 615.3005, calculated for C32H48O10Na + 615.2987, difference = 2.9 ppm). Molecular networks of the extracts (an example is shown for TLT/PHB/001 in Figure 3), visualized using the software Cytoscape v3.5.1 (The Cytoscape Consortium, New York, USA), revealed the presence of clusters related to the molecular family of the aplysiatoxin class in all four environmental cyanobacterial samples (clusters A to E in Figure 4). In addition to the presence of the seeded debromoaplysiatoxin, five other related compounds ( Figure  2), namely, anhydrodebromoaplysiatoxin (2), 3-methoxydebromoaplysiatoxin (3), aplysiatoxin (4), oscillatoxin A (5) and 31-noroscillatoxin B (6), were detected in clusters A, B, C and E based on their HRMS (High-Resolution Mass Spectrometry) data (Table 1 and Figure 4). Furthermore, potential new aplysiatoxin analogues, based on their relatedness to 1, were detected from the clusters. These potential new compounds are indicated as 15 nodes from cluster A, eight nodes from both cluster B and C, seven nodes from cluster D and 17 nodes from cluster E (Figure 4). Some putative assignments for the new aplysiatoxin analogues were included, based on the HRMS m/z values.

MS/MS-Based Molecular Networks of Marine Cyanobacterial Extracts
Organic crude extracts from each of the cyanobacterial samples were prepared and analyzed initially by UPLC-HRMS/MS Waters system (Milford, MA, USA). In addition, organic crude extract from TLT/PSK/001 were analyzed on both HPLC-HRMS/MS Thermo-Finnigan system (Waltham, MA, USA) (TLT/PSK/001a) and UPLC-HRMS/MS Waters system (TLT/PSK/001b). Mass spectrometric data obtained from these analyses were then used to generate molecular networking clusters using the online platform Global Natural Products Social Molecular Networking (GNPS). Debromoaplysiatoxin (1) (Figure 2), isolated from the extract of TLT/PHC/001, was used as seed molecule in the analysis (HRESIMS of sodiated molecular ion m/z 615.3005, calculated for C 32 H 48 O 10 Na + 615.2987, difference = 2.9 ppm). Molecular networks of the extracts (an example is shown for TLT/PHB/001 in Figure 3), visualized using the software Cytoscape v3.5.1 (The Cytoscape Consortium, New York, NY, USA), revealed the presence of clusters related to the molecular family of the aplysiatoxin class in all four environmental cyanobacterial samples (clusters A to E in Figure 4). In addition to the presence of the seeded debromoaplysiatoxin, five other related compounds ( Figure 2), namely, anhydrodebromoaplysiatoxin (2), 3-methoxydebromoaplysiatoxin (3), aplysiatoxin (4), oscillatoxin A (5) and 31-noroscillatoxin B (6), were detected in clusters A, B, C and E based on their HRMS (High-Resolution Mass Spectrometry) data (Table 1 and Figure 4). Furthermore, potential new aplysiatoxin analogues, based on their relatedness to 1, were detected from the clusters. These potential new compounds are indicated as 15 nodes from cluster A, eight nodes from both cluster B and C, seven nodes from cluster D and 17 nodes from cluster E ( Figure 4). Some putative assignments for the new aplysiatoxin analogues were included, based on the HRMS m/z values.    Italicized labels indicate putative assignments; numbers in bold indicate compounds 1 to 6 (refer to Figure 2). Edge thickness corresponds to relative cosine similarity between nodes. Italicized labels indicate putative assignments; numbers in bold indicate compounds 1 to 6 (refer to Figure 2). Edge thickness corresponds to relative cosine similarity between nodes.

Discussion
Certain cyanobacteria strains are known to produce a wide array of toxic metabolites or cyanotoxins, such as aplysiatoxins, microcystins and nodularins [7]. Many of these cyanotoxins affect human health through their actions as tumor promoters, diarrhetic, and pro-inflammatory activities [18]. In addition, these metabolites may also limit various aquatic recreational activities, thereby affecting related economies. Such impacts are particularly acute in certain geographical regions, such as Singapore, where its location in the tropics provides a conducive environment for the occurrence, growth, and more importantly blooms of marine cyanobacteria [19]. For instance, regular occurrence of marine cyanobacterial blooms has been sighted at several offshore islands in Singapore, including Pulau Semakau, P. Seringat-Kias, P. Hantu, P. Ubin and P. Sakijang Bendera. The cyanobacterial samples used in this study were obtained from islands designated as tourist destinations with aquatic recreational activities, including SCUBA diving and swimming, as the main attraction. The persistent occurrence of marine cyanobacteria at these islands (personal observation by Lik Tong Tan) warrants setting up of a rapid and relatively straightforward analytical method for the detection of cyanotoxins from environmental samples. A rapid detection method is integral and equally imperative to prevent the general public's exposure to these toxic compounds.
In this study, molecular networking clusters of various environmental marine cyanobacterial extracts were analyzed using a MS/MS-based molecular networking approach [20]. Initial chemical investigation of the organic extract obtained from the cyanobacterial sample TLT/PHC/001 collected at Pulau Hantu resulted in the isolation of the known cyanotoxin debromoaplysiatoxin (1). This molecule was subsequently used as a seed molecule for the creation of the aplysiatoxin molecular family in the molecular network. Each node within the clusters represent an ionized compound and dereplication was performed by comparison with the m/z values of known aplysiatoxin-related analogues ( Table 2). In addition to debromoaplysiatoxin, a number of known analogues, anhydrodebromoaplysiatoxin (2), 3-methoxydebromoaplysiatoxin (3), aplysiatoxin (4), oscillatoxin A (5) and 31-noroscillatoxin B (6), were found to be present in the molecular network clusters A, B, C and E. The identity of these known compounds was confirmed by comparing the calculated m/z values with that obtained from HRMS/MS (differences of less than 5 ppm) as shown in Tables 1 and 2. Compounds 1-3 were present in clusters A, B, C and E. Additional compounds 4 and 5 were also present in cluster A while 6 was found in cluster E.
The phylogenetic results of the cyanobacterial samples, coupled with careful analysis of the molecular networking clusters of their extracts, revealed a number of interesting observations. Though both TLT/PHB/001 and TLT/PSK/001 cyanobacterial samples have been identified as Trichodesmium erythraeum, their clusters (A and E), related to the aplysiatoxin molecular family based on the seeded 1, were different ( Figure 4). Clusters A and E gave 15 and 17 nodes, respectively, both revealing the presence of all three compounds 1-3. The difference in the complexity of these two Trichodesmium erythraeum-derived molecular network clusters and the number of nodes present may be attributed to the versatility of the biosynthetic gene clusters present in the same cyanobacteria, and different environmental triggers present in the two locations where the cyanobacteria were collected. On the other hand, aplysiatoxin-related clusters B and C produced from two different cyanobacterial species, TLT/PHC/001 (identified as Okeania sp. PNG05-4) and TLT/PHC/002 (identified as Oscillatoria sp. PAB-21), respectively, with only slight difference in the m/z (within 5 ppm) observed between corresponding nodes. Interestingly, these two samples were collected from sites that were just 100 m apart at Pulau Hantu. In addition, when comparing the full molecular networks of these two different cyanobacterial species (Figure 5), minimal differences in the compounds, aside from the aplysiatoxin-related cluster, were observed. These observations suggest that similar compounds are largely produced as a result of the same environmental triggers, though by different cyanobacteria, and this can be validated with the help of results obtained through the investigations of related biosynthetic gene clusters of TLT/PHC/001 and TLT/PHC/002 in producing the aplysiatoxin-related compounds. by different cyanobacteria, and this can be validated with the help of results obtained through the investigations of related biosynthetic gene clusters of TLT/PHC/001 and TLT/PHC/002 in producing the aplysiatoxin-related compounds. It has been shown that the MS/MS-based molecular networking is useful for dereplication of known compounds as well as the discovery of potentially new natural products [21]. In molecular networking, MS/MS spectra are compared based on the similarity and differences in fragmentation patterns. Structurally related compounds can then be clustered as nodes due to certain degrees of similarity between the compounds expressed as edges in the molecular network. In addition, the availability of HRMS/MS spectral data of known cyanotoxins in GNPS database could facilitate the detection process [22]. Recently, Teta and co-workers used a similar MS-based molecular networking approach to detect new microcystin variants from the 2014 freshwater cyanobacterial harmful bloom at Green Lake, Seattle [23]. In another study by Briand et al., molecular networking was employed to track changes in secondary metabolic profiles, including microcystins and other peptides, of Microcystis aeruginosa strains due to intraspecific interactions [24]. In this present study, two nodes with m/z 629.219 and 629.220 observed in clusters B, C and E are likely to be compound 3. However, it is evident that only the node with observed value of 629.22 fits within the accepted ppm difference for compound 3 (Figure 4). Considering the presence of both nodes found within the same cluster, the node of m/z 629.219 could be a new analogue related to 2. A similar relationship was observed from the nodes of m/z 615.218 and 615.298 present in the same set of clusters ( Figure 4). Some putative assignments were provided to unknown compounds based on observed HRMS of the nodes within each cluster. An example is the compound with m/z value of 631.247 in clusters B and C that could correspond to an analogue of 3 having a reduced carbonyl moiety or cleavage to the oxide functionality of either tetrahydropyran rings.
The advantages of use and advancements in instrumentation provided a variety of MS/MS systems used by laboratories today. An interesting technical approach adopted in this study was to employ two different systems in the analysis, namely a low-resolution Thermo-Finnigan HPLC-MS/MS and a high-resolution Waters UPLC-MS/MS. The main reason was to make comparisons of the molecular networks created based on each system. The organic extract from TLT/PSK/001 was randomly chosen and molecular networks were then constructed individually using the MS/MS data acquired from both MS systems using similar parameters. Analysis of the molecular network clusters TLT/PSK/001a (D) and TLT/PSK/001b (E) using Thermo-Finnigan and Waters systems, respectively, revealed several differences. The first difference was the size of each molecular network due to a difference in quantity of nodes present, with 7 and 17 nodes in clusters D and E, respectively. Another difference was observed comparing compounds 1, 2 and 6 present in both clusters, where each corresponding node in both clusters gave different m/z values. A case in point is the node that corresponds to 2, with m/z of 597.267 and 597.261 observed in clusters D and E, respectively. The chemical identity of the corresponding nodes was achieved and confirmed to be 2 in both clusters, where a mass difference of within 5 ppm was determined based on the observed and calculated m/z obtained from its respective system. The last difference was the detection of different compounds from the same cyanobacterial strain in the two molecular networks. Compounds 4 and 5 were observed exclusively in cluster D obtained from the Thermo-Finnigan system; compound 3 observed in cluster E created using the Waters system. While the exact explanation could not be provided, these differences between the clusters may be attributed largely to the different MS instruments' operating parameters (i.e., resolution, type of ionization and detector), hence care should be taken while comparing molecular networks from different systems.
Current aquatic biomonitoring and ecotoxicology programs in Singapore are confined to the analysis of freshwater cyanobacterial strains for hepatoxins, such as microcystins and nodularin. Methods, such as mass spectrometry, capillary electrophoresis and qPCR, have been used for the detection of microcystins as well as for assessing cyanotoxin risk in freshwater cyanobacterial strains obtained from reservoirs in Singapore [25][26][27]. The unprecedented nature of this pilot study underscores the importance of the MS/MS-based molecular networking approach as a rapid analytical technique to detect non-peptidic cyanotoxins, such as the aplysiatoxins, from environmental samples. Moreover, recent reports showed that several aplysiatoxin analogues possess therapeutic activities, such as antiviral and potassium channel inhibition activities [14,15]. In light of these discoveries, the MS/MS-based molecular networking approach employed in this study could aid in the detection of new aplysiatoxin analogues that harbors varied biomedical potential, contributing to the growing list of marine natural products as therapeutic agents.

General Experimental Procedures
Proton NMR spectra were recorded in CDCl 3 on a 400 MHz Bruker (Billerica, MA, USA) NMR spectrometer using the residual solvent signals (δ H at 7.28 ppm) as internal standards. High-performance liquid chromatography isolation and purification of compound 1 was conducted on a Shimadzu (Nakagyo-ku, Kyoto, Japan) LC-8A preparative LC coupled to a Shimadzu SPD-M10A VP diode array detector. The UPLC-HRMS/MS analysis was conducted on a Waters (Milford, MA, USA) Xevo G2-XS qTOF with an ESI positive ion mode and data-dependent acquisition mode to obtain MS and MS/MS data. The HPLC-MS/MS analysis was conducted on a Thermo-Finnigan (San Jose, CA, USA) LCQ Advantage ion trap mass spectrometer with an ESI positive ion mode and data-dependent manner to obtain MS and MS/MS data. High-performance liquid chromatography preparative and analytical columns was obtained from Phenomenex (Torrance, CA, USA), and a UPLC analytical column was obtained from Waters. All solvents were HPLC-grade purchased from Thermo-Fisher Scientific (Waltham, MA, USA) and HPLC-grade water obtained by filtration using a Milli-Q Direct water purification system from Millipore (Billerica, MA, USA) unless otherwise stated. Deuterated NMR solvents were purchased from Cambridge Isotope Laboratories (Tewksbury, MA, USA).

Phylogenetic Analysis
Gene sequences were aligned using MUSCLE (European Bioinformatics Institute, EMBL-EBI, Cambridge, United Kingdom) [30]. The phylogenetic tree was plotted using maximum likelihood (ML) in MEGA, v6 (Pennsylvania State University, State College, PA, USA) [31]. The 16S rRNA gene sequence of each cyanobacterial sample was multiply aligned with 7 cyanobacterial strains obtained from the GenBank/EMBL/DDBJ databases. Bootstrap re-sampling was performed with 1000 replicates.

UPLC-HRMS/MS Analysis
Crude extracts were re-dissolved in 1 mL CH 3 CN with vortex mixing over 5 min before transferring into separate Eppendorf tubes. Tubes were then centrifuged at 10,000 rpm at 4 • C over 10 min and the supernatant was aliquoted and finally diluted with CH 3 CN to 10,000× dilution. Then, 1.5 µL of each 10,000× dilution sample was subjected to UPLC-HRMS/MS performed using the same high-resolution spectrometer and parameters stated above, together with a Waters ACQUITY BEH C 18 50 mm × 2.1 mm, 1.7 µm column and maintained at a column temperature of 40 • C and sample temperature at 4 • C using a step elution program of '1' based on the Waters system. Step elution program was as follows: Mobile phase of 98% CH 3 CN in 0.1% aq. HCOOH/0.1% aq. HCOOH, run time of 14 min and flow rate of 0.5 mL/min. All the mass spectra were recorded in the positive-ion mode. Data were collected in the data-dependent acquisition mode, in which the first ten most intense ions of a full-scan mass spectrum were subjected to MS/MS analysis. Tandem mass spectrometry scans were obtained for selected ions with a mass range of 100-2000 Da, MS scan time of 0.33 s over 12 min, MS/MS scan time of 0.10 s and a collision energy ramp of 10-50 V.

HPLC-MS/MS Analysis
TLT/PSK/001 was re-dissolved in MeOH and 20 µL of the solution was subjected (no-waste mode) to HPLC-MS/MS using a Thermos-Finnigan system with a Phenomenex Kinetex C 18 100 mm × 4.6 mm, 5 µm column. Elution at 0.5 mL/min using eluent A: 0.1% aq. HCOOH in H 2 O; eluent B: