Biofilms presumed to contain abundant cyanobacteria based on their green-to-brown color, were sampled in three mangroves on the Mayotte island of Grande-Terre: Dembeni (areas 1 and 2, 20 biofilms), Zidakani (4 biofilms), and Dapani (3 biofilms) between November 8^th and 11^th, 2018 ( Table S1 ; Figure 1 ). Biofilms, either benthic or attached to substrates including mangrove tree bark and debris were sampled using sterile tweezers, photographed, and transferred to sterile Falcon tubes in a cooler. Following Article 17, paragraph 2, of the Nagoya Protocol on Access and Benefit-sharing, a sampling permit was issued and published (https://absch.cbd.int/database/ABSCH-IRCC-FR-246969).

Back to the laboratory, biofilm samples were split. One fragment was stored in RNAlater™ (Ambion, UK) then frozen for DNA extraction. Another fragment was observed under a microscope. Cyanobacteria-rich fractions from 25 of the 27 biofilms were sampled using a Pasteur pipette under the microscope and inoculated on solid (5 g.L^-1 agar) Z8 culture medium, either as is, or containing 20 g.L^-1 of salt (Instant Ocean) (Rippka, 1988). Single filaments were isolated by repeated transfers on liquid media under an inverted microscope (Nikon ECLIPSE TS100), with or without cycloheximide to prevent fungi growth.

DNA was extracted from the 27 biofilms using the ZymoBIOMICS DNA mini kit (Zymo Research, CA) according to the manufacturer’s instructions, after a disruption step using glass beads (6 x 1min, 30 Hz, 1,800 oscillations per minute). A ~ 400 bp fragment of the rRNA-encoding gene corresponding to the V4-V5 variable region of Escherichia coli was amplified using primers 515F and 926R (Fadeev et al., 2021) and sequenced on an Illumina MiSeq platform (2x300 bp, paired end, Genoscreen, France). Company-provided mock communities of known composition were used as an internal control for the whole sequencing process. Raw reads were deposited into the GENBANK Sequence Read Archive (SRA, Bio project PRJNA947804).

Sequence reads were analyzed using QIIME2 (Hall and Beiko, 2018). Amplicon Sequence Variants (ASVs) (Fadeev et al., 2021) were identified using DEBLUR (Amir et al., 2017) using default parameters, i.e. a maximal probability for indels of 0.01 and mean read error rate of 0.5% for normalization. Chimeric sequences were identified using UCHIME (Mühr et al., 2020) and then removed (Edgar et al., 2011), and the taxonomic affiliations were obtained by using the sklearn-based classifier (GreenGenes 13-8-99 release). Sequences matching “Eukaryota”, “Chloroplast” and “Mitochondria” were discarded. Representative plots were obtained using GGPlot2.

Isolated strains and cultures were all monoclonal and non-axenic. A total of 43 strains were registered in the Muséum national d’Histoire naturelle (MNHN, Paris, France) culture collection of cyanobacteria and microalgae, where they are available upon request (https://www.mnhn.fr/fr/microalgues-et-cyanobacteries) (Hamlaoui et al., 2022), under labels PMC (Paris Museum Collection) 1096.18 to PMC 1143.19 ( Table 1 ). Biomass was generated for further characterization by culturing strains in increasing volumes of liquid Z8 medium (25°C; 15 µmol.m^-2.s^-1 white light; 16 h light: 8 h dark) without salt (PMC 1097.19, PMC 1099.19, PMC 1100.19, PMC 1101.19, PMC 1104.19 to PMC 1106.19, PMC 1111.19 to PMC 1115.19, PMC 1134.19, PMC 1136.19, PMC 1140.19) or with 20 g.L^-1 salt (other strains) for two months, depending on which conditions were leading to increased growth. The morphological characteristics (such as filament diameter, presence of heterocytes, akinetes, constrictions between cells) of strains were subsequently observed under the inverted microscope (Nikon ECLIPSE TS100, Japan).

DNA was extracted from cultures using the ZymoBIOMICS Fecal/Soil Kit (Zymo Research, Irvine, CA, USA), after a 3-minute disruption of cells with ceramic beads, using reagents and conditions as described in the manufacturer’s instructions (Ammar et al., 2014), and concentration was measured using Nanodrop and Qubit. Cyanobacteria 16S rRNA encoding genes were PCR amplified (35 cycles) using two targeted primer sets designed for this group. The annealing temperature of 58°C was used for primer set 8F (5’-AGAGTTTGATCCTGGCTCAG-3’) and 920R (5’-TTGTAAGGTTCTTCGCGTTG-3’), and annealing at 55°C was used for primer set 861F (5’-TAACGCGTTAAGTATCCC-3’) and 1380R (5’-TAACGACTTCGGGCGTGACC-3’) (Gugger et al., 2002; Ludwig, 2007). Both amplicons were sequenced and sequence chromatograms (Genoscreen, Lille, France) were analyzed, assembled by Geneious (https://www.geneious.com/) for each strain, and compared to the GENBANK database with BLAST (default parameters, standard databases, optimize for highly similar sequences program selected). Sequences were submitted to GENBANK under accession numbers OQ693634 to OQ693681.

A dataset was constructed consisting of sequences of the 43 isolated strains, their best BLAST hits, and representatives of major cyanobacterial lineages. Sequences were aligned by the secondary structure-aware Silva tool [Ribosomal Database Project website (Cole et al., 2009)]. After alignment, ambiguously aligned regions were deleted using Bioedit 7.2.5. Phylogenetic relationships were inferred using a Maximum Likelihood approach based on General Time Reversible Model (5 categories and invariants, see tree captions for details, selected based on the Akaike Information) (Posada and Buckley, 2004). A pairwise p-distance matrix ( Table S2 ) was generated to aid preliminary genus and species delimitation. Phylogenetic trees were constructed by using the software MEGA 7 (Newman et al., 2016).

The presence of four genes diagnostic to produce microcystin, anatoxin, cylindrospermopsin, and saxitoxin families, respectively, was tested using specific primer sets and PCR programs on all strains ( Table S3 ).

The 16S rRNA-encoding genes sequences obtained from all 43 strains were compared with ASVs obtained from the 27 biofilms to evaluate the relative abundance of isolated strains in natural biofilms. For this, cyanobacterial ASVs were extracted from biofilm datasets and aligned with strain sequences using Clustal W. Number of differences between most similar ASVs and strains were computed based on a 345 bp alignment. ASVs displaying 3 or fewer differences with strain sequences (~1%) were retained, and their respective abundance versus bacterial and cyanobacterial sequences in biofilm samples were checked. All analyses were conducted using MEGA-7 (Newman et al., 2016).

Chemical extracts were performed for the 23 strains for which enough biomass was available after two months. Each culture sample was stored in 50 mL Falcon tubes, containing 40 mL salt-free Z8 medium. After centrifugation at 4,000 g, three deep washes with unsalted Z8 medium were done, then the supernatant was discarded, and the cyanobacteria pellets were lyophilized under vacuum at -40°C for 12 h after freezing at -80°C for 30 minutes. Dry biomass was weighted, and 20 mg were re-suspended in a 2 mL solvent mixture consisting of MeOH/CH₃CN/H₂O (40:40:20). After three successive cycles of sonication (6 min cycle: 1 min ON; 30 s OFF) and centrifugations (14,000 g, 5 min), supernatants were evaporated and the dry crude extracts were weighed and suspended in MeOH (1 mg. mL^-1). Finally, extracts were filtered on a 0.2 µm membrane (4 mm in diameter Phenomenex®). An aliquot of 30µL was reserved for LC-MS/MS analysis. The rest of the crude extract samples was evaporated and diluted in DMSO (1 µg.µL^-1), after which 200 µL aliquots were used for activities assay.

UPLC-HRMS/MS analyses were achieved by coupling a LC system to a hybrid quadrupole-time-of-flight mass spectrometer Agilent 6546 equipped with an ESI source, operating in positive-ion mode. The UPLC (C18 BEH Acquity® Waters 100 × 2.1 mm, 1.7 µm column, 500 µL.min^-1 gradient elution (A: CH₃CN, B: H₂O + 0.1% formic acid) from 5% to 100% A, over 12 min) was used to perform chromatographic separation. Source parameters were set as followed: capillary temperature at 320°C, source voltage at 3500 V, and sheath gas flow rate at 11 L·min^-1. MS scans were operated in full-scan mode from m/z 100 to 1200 (0.1 s scan time) with a mass resolution of 60,000 at m/z 922. In the positive-ion mode, purine C₅H₄N₄ [M+H]⁺ ion (m/z 121.050873) and the hexakis (1H,1H,3H-tetrafluoropropoxy) phosphazene C₁₈H₁₈F₂₄N₃O₆P₃ [M+H]⁺ ion (m/z 922.009798) were used as internal lock masses. A permanent MS/MS exclusion list criterion was set to prevent oversampling of the internal calibrant. LC-UV and MS data acquisition and processing were performed using MassHunter^® Workstation software. Acquisition was performed in “Auto MS/MS” mode on the range m/z 100–1200 with an MS rate of 1 spectra.sec^-1 and an MS/MS scan rate of 3 spectra.sec^-1. Isolation MS/MS width was 1.3 m/z. Fixed collision energies of 45 eV was used. MS/MS events were performed on the four most intense precursor ions per cycle with a minimum intensity of 5,000 counts.

The MS data were converted from d (Agilent) standard data format to mzXML format by using the MSConvert software, part of the ProteoWizard package (Myers et al., 2017). Then, the converted mzXML files were processed with the MZmine software package v. 2.38 (Pluskal et al., 2010). The parameters were selected as follows: for mass detection, the centroid mass detector was utilized, with the noise level set to 2.5E3 for MS level 1 and 1.0E1 for MS level 2. The ADAP chromatogram (Du et al., 2020) builder was used, with a minimum group size of 4 scans, a 2.5E3 minimum group intensity threshold, a 1.0E3 minimum highest intensity, and a 10.0 ppm m/z tolerance. The wavelets approach was utilized for chromatogram deconvolution (ADAP). With a signal-to-noise ratio of 10, a minimum feature height of 1,000, a coefficient area threshold of 10, a peak duration range of 0.01 to 0.45 min, and an RT wavelet range of 0 to 0.1 min, the intensity window S/N was utilized as a S/N estimator. The isotope peaks grouper was used to detect isotopes with a m/z tolerance of 10.0 ppm, an RT tolerance of 0.3 min (absolute), a maximum charge of 1, and the most intense representative isotope employed. The join alignment method was used to align the peaks (m/z tolerance of 10 ppm, RT tolerance of 0.12 min). The weight of m/z was set as 50, the weight was set as 50, and the charge state must be the same. The same RT and m/z range gap filler was used to fill in the gaps in the peak list (m/z tolerance at 10 ppm, the retention time range 0.04 from 15.7 min). Finally, using the peak-list rows filter option, the aligned peak list was filtered to maintain only features linked with MS2 scans. LC-MS/MS data were deposited on MassIVE under the accession number: MSV000091466.

A feature-based molecular network was generated using Global Natural Products Social Molecular Networking (GNPS) (Pluskal et al., 2010; Wang et al., 2016; Nothias et al., 2020) (https://gnps.ucsd.edu/ProteoSAFe/static/gnps-splash.jsp). The edges of the network were filtered to have a cosine score greater than 0.6 and more than six matched peaks. Further edges between two nodes were preserved in the network if and only if each node was in the top 10 most similar nodes of the others. The network’s spectra were compared to GNPS’ spectral libraries. All network spectra and library spectra match had to have a score of at least 0.6 and at least six matched peaks. The molecular network was visualized using Cytoscape 3.9.1 software (Doncheva et al., 2019). The GNPS job parameters and resulting data are available at the following address (https://gnps.ucsd.edu/ProteoSAFe/status.jsp?task=5bd14c218bdf4a799d6fe98cb8880d20). The DEREPLICATOR annotation resulting data are available at the following address (https://gnps.ucsd.edu/ProteoSAFe/status.jsp?task=3b8d860a0b794769b25abe56d138ee34). To extend the annotations, two additional in silico tools, namely ISDB (Allard et al., 2016) and MolDiscovery (Cao et al., 2021) were used. Regarding ISDB tool, the annotation parameters were set as follows: spectral score (cosine > 0.2), parent mass tolerance (0.002 Da), and a maximum of five candidate structures were proposed. For the Moldiscovery processes, the parameters were: Precursor Ion Mass Tolerance 0.01 Da, Fragment Ion Mass Tolerance 0.01 Da, Max Charge at 1, and Min Significant Score at 10.0. Results of MolDiscovery annotations are available at the following address: (https://gnps.ucsd.edu/ProteoSAFe/status.jsp?task=5556da8361b74371b27ce1b260d34809). For more in-depth annotation, SIRIUS 5.6.2 (Dührkop et al., 2015) were used, the parameters of exporting MZmine data to SIRIUS were set as: weighted average m/z merge mode, mean intensity merge node, expected mass deviation 10 ppm, cosine threshold 0%, peak count threshold 0%, isolation window offset 0 m/z, isolation window width 0 m/z.

Crude extracts from 23 cyanobacteria strains were used to evaluate the cytotoxic activity against CHO (Chinese Hamster Ovary) cells. Cells were seeded in a 96-well microplate, 5,000 cells. well^-1 in a 200 µL culture medium (DMEM low glucose - PAN biotech, Germany) and maintained at 37°C and 5% CO₂/95% air for 24 hours. After the incubation period, each well was supplemented with 50 µL cyanobacterial extracts at a final concentration of 500.0, 250.0, 125.0, 62.5, 31.2, or 15.6 µg. mL^-1. For the fractions eluted from the LH-20 column chromatography, the final concentration in each well was 20.0 µg. mL^-1. Each concentration was tested in three replicates, and the experiment was repeated three times (Trang et al., 2021). Paclitaxel (Sigma-Aldrich, USA) was used as a positive control at concentrations ranging from 0.003 to 30µg.mL^-1. After 48 hours, cells were dyed with the MTT assay, and the optical density was measured using a Microplate reader SpectraMax Plus384 (GMI, USA) at two wavelengths of 570 and 690 nm (Bich-Loan et al., 2021). The IC₅₀ values (the concentration of extract at which 50% of cell growth was inhibited) were calculated by Excel and GraphPad Prism 8 software.

To investigate the activity against filarial nematodes, 200 blood-purified first stage larvae (microfilariae) of the rodent filarial nematode Litomosoides sigmodontis were cultured in vitro in 200µl RPMI 1640 medium (Sigma Aldrich) supplemented with HEPES 25mM, 10% FCS, 1% penicillin/streptomycin/amphotericin at 37°C, 5% CO₂ (Townson et al., 1987; Hawryluk et al., 2022). Each batch of microfilaria received 2µl (5 µg.µL^-1) of crude extract, or 2µl of DMSO solution as a control. Each condition was in triplicate. The motility of microfilariae was assessed by determining the mean worm motility scores from 0 to 3, with 3 corresponding to normal movement and uncoiled worm; 2 to slow movement and uncoiled; 1 to twitch movement and uncoiled; and 0 corresponding to no motility (dead worm). Scores were determined 2h post- addition.

The antibiofilm activity of the crude extracts was tested against biofilm-forming Pseudomonas aeruginosa MUC-N1 (Boukerb et al., 2020). The bacteria strains were cultured in 5 mL of LB medium (NaCl 10g/L, tryptone 10g/L and yeast extract 5g/L) at 37°C under orbital shaking at 125 rpm, overnight in triplicate. 200 μL of diluted culture adjusted to OD₆₀₀ of 0.05 in fresh test medium was inoculated in each well, except for the blank. Aliquots of cyanobacterial extract at c (concentration) = 50 µg. mL^-1 (1% DMSO) was added to the 200 μL P. aeruginosa suspensions.

The 96-well microtiter plate (Thermo Scientific™ Nunc™ MicroWell™ 96-Well) was incubated in a stationary phase for 24 hours. After measuring the optical density at 600 nm (OD₆₀₀) using a TECAN Infinite M1000, the coloration and quantification of the biofilm was realized according to the Coffey & Anderson protocol (Coffey and Anderson, 2014), after acetic acid addition, by measuring the OD₅₅₀. The evaluation of the biofilm formation is corrected according to the growth of the bacteria strains. The values are presented as the mean ± SD. Data were evaluated with GraphPad Prism Version 8.01 software.

The composition of prokaryotic communities was analyzed from 27 biofilms collected from three mangroves in Mayotte ( Table S1 ). Sampled biofilms occurred attached to sunken deadwood or debris (plastic, bamboo), covering the sediment surface, or adhering to mangrove tree roots and trunks ( Figure 1 ). Sequencing of the V4-V5 region yielded a total of 1,119,200 quality-filtered sequences (44,414 ± 23,522 per biofilm). Apart from sample S23, which yielded only 339 reads with 189 ASVs, the 26 other samples yielded 26,334 to 85,074 reads clustering into 992 to 47,941 ASVs (11,452 ± 9,052 ASVs).

Bacteria were abundant in biofilms ( Figure 5 ), with Alphaproteobacteria, Bacteroidota, Actinomycetota, and Cyanobacteria as the main taxa. The relative abundances of each taxon were highly variable among biofilms. Cyanobacteria represented between 1.35% (sample S26) and 61.07% (sample S30, Table S1 ) of reads. Interestingly, biofilms that displayed evident blue-green colors, such as samples S2, S21, and the three samples from Dapani (S27, 29, and 30) displayed particularly high abundances of Cyanobacteria compared to samples that did not have a greenish appearance (e.g., samples S13 and S26 illustrated in Figure 1 ). Over the whole dataset, cyanobacteria clustered into 157 ASVs, each recovered from 1 to 5 biofilms. Of these ASVs, 15 represented at least 5% of all reads in at least one biofilm sample, and could be considered as abundant members of the biofilm ( Tables 1 , S1 ).

A total of 43 cyanobacterial strains ( Table 1 ) were isolated from 10 of the 27 biofilms and successfully maintained in the PMC culture collection. During growth, they formed extensive fasciculate mats or small clusters in the medium, with colors ranging from brown-green, olive-green to dark green ( Figure 6 ). Analysis of the 16S rRNA-encoding gene sequences indicates that isolated strains belong to the cyanobacterial orders Oscillatoriales, Nostocales, Spirulinales, Coleofasciculales, Synechococcales, and Leptolyngbyales (22, 7, 5, 4, 4 and 1 strains, respectively), (Sturnecky et al., 2023). Strains were assigned to existing species and genera based on widely accepted 16S rRNA similarity cutoff values of 98.65% and 95% similarity for the 16S rRNA-encoding gene, respectively (Kim et al., 2014; Komárek et al., 2014), on monophyly, and on morphological features (filament diameter, presence of heterocytes, sheath, akinetes, constrictions between cells) overall consistent with members of these genera. When no published species or genus sequence fell within this similarity range and with the support of phylogenetic reconstruction, strains were assigned to hypothetical species (sp.) and genera (gen. nov.) (Caires et al., 2018a; Caires et al., 2018b; Berthold et al., 2021).

Based on sequence similarity criteria, strains were grouped into 22 species-level taxa according to selected threshold values ( Table 1 ; Table S2 ). Over half of the strains (26 out of 43 strains) belong to the orders Oscillatoriales and Coleofasciculales ( Figure 2 ). Among these, seven strains correspond to two previously described species, namely Capilliphycus salinus (PMC 1104.19 and PMC 1105.19) and Planktothrix mougeotii (PMC 1100.19, PMC 1112.19, PMC 1113.19, PMC 1114.19, PMC 1115.19). The rest of the Oscillatoriales strains can be assigned to undetermined species within genera Hydrocoleum (PMC 1102.19, 1103.19, and 1116.19), Capilliphycus (PMC 1119.19, 1121.19, 1127.19, and 1137.19), Neolyngbya (PMC 1131.19, 1132.19, and 1134.19), and Dapis (PMC 1108.19, Tables 1 and S2 ). Strain PMC 1135.19, showed a 6.1% dissimilarity to the closest identified Sodalinema, above the 5% dissimilarity threshold, and is not monophyletic with this genus ( Tables 1 , S2 ; Figure 2 ). For these reasons, it is assigned to a hypothetical new genus. Strains PMC 1129.19, 1130.19, and 1133.19 cluster together in the phylogenetic tree, with no named genus within the 5% similarity criterion, thus representing another new hypothetical genus. Coleofasciculales comprise strains PMC 1096.19, PMC 1098.19 (from Dembeni 1, sample S2), and PMC 1123.19 (Zidakani, sample S23) which clustered together in the phylogenetic tree, both groups representing distinct species-level taxa within genus Symploca, with which display 4.2% dissimilarity in the 16S rRNA sequence.). One strain, PMC 1122.19, belong to genus Coleofasciculus.

Seven strains belonging to the order Nostocales were isolated ( Table 1 ; Figure 3 ). Strains PMC 1097.19 and 1099.19 isolated from sample S2 were affiliated with Hapalosiphon welwitchii because of the 99% similarity in their 16S rRNA sequences, and their clustering within the tree of H. welwitchii (KJ 767019, Figure 3 ).

The four strains isolated from sample S29 (PMC 1136.19, 1138.19, 1139.19, and 1140.19) clustered into two distinct species-level clades in the phylogenetic tree, with several closely related sequences from species belonging to genera Wollea, Anabaena and Sphaerospermopsis ( Figure 3 ). Their morphological characteristics include tubular colonies with common sheath, straight trichomes, organized parallel and tightly in colonies, and widely oval, most similar to those observed in Wollea sp. (Kozlíková-Zapomělová et al., 2016) ( Figure 6 ), and these were thus assigned to two distinct hypothetical species within this genus. Finally, strain PMC 1101.19 isolated from sample S15 was assigned to a hypothetical species within the genus Brasilonema with which it is monophyletic.

Using the same criteria, other species belonging to orders Spirulinales, Leptolyngbyales, and Synechococcales were assigned to hypothetical species ( Figure 4 ). These belong to genera Jaaginema (PMC 1109.19), Halomicronema (PMC 1117.19), two distinct hypothetical species within genus Thainema (PMC 1125.19 and 1126.19 based on their 1.4% dissimilarity in the 16S rRNA-encoding gene), and two distinct species within genus Spirulina (PMC 1110.19 and 1111.19 in one species and PMC 1141.19, 1142.19 and 1143.19 in a second species), these latter two displaying the typical coiled morphology described in genus Spirulina ( Figure 6 ). Strain PMC 1124.19 was assigned to a hypothetical new genus.

PCR assays targeting genes involved in the production of microcystin, anatoxin, saxitoxin, and cylindrospermopsin yielded no amplicon for any of the 43 tested strains, despite that positive controls yielded amplicons ( Table S3 ), suggesting absence of the target genes in their genomes.

Out of the 22 species-level taxa identified among isolates, 13 were also identified as ASVs in 16S rRNA metabarcoding datasets obtained from 14 out of the 27 biofilms ( Table 1 ). For 10 of these taxa, the corresponding ASV was recovered from at least one of the samples from which the strains were isolated. According to biofilm community compositions, eight species isolated in culture happened to be abundant members of the cyanobacterial community (i.e., they represented over 5% of cyanobacterial reads) in at least one biofilm sample. In several cases, the species was even the main cyanobacterial member of this biofilm, with taxon Neolyngya sp. (PMC 1132.19, 1134.10, 1131.19) representing 78.0% of the cyanobacterial reads in sample S27, taxon Gen. Nov. 1 sp. (PMC 1124.19) representing 100.0% in sample S23, Symploca sp. 1 (PMC 1096.19, 1098.19) representing 60.7% in S2 and Planktothrix mougeoti (PMC 1100.19, 1112.19, 1113.19, 1114.19, 1115.19) representing 94.5% in sample S21. Finally, seven isolated species were also abundant members of the whole community in at least one biofilm sample (i.e. they represented above 5% of all reads, Table 1 ).

Cultures from a total of 23 out of the 43 identified strains, representing 11 species-level taxa, yielded enough biomass for chemical characterization by LC-MS/MS analyses. To analyze the chemical diversity of the 23 extracts, LC-MS/MS experiments were performed to enter the feature-based molecular networking workflow ( Figure 7 ). A total of 2,667 features and 131 clusters were visualized in Cytoscape 3.9.1. The distribution of cyanobacterial metabolites among the 23 extracts of the entire molecular network was mapped at the genus level using a representative color tag. Analysis of the network showed that some molecular clusters were restricted to specific genera, highlighting the distribution of closely related compounds ( Figure S1 ). The GNPS spectral library search yielded only eight hits including triphenylphosphate, triphenylphosphine, and pheophytin (https://gnps.ucsd.edu/ProteoSAFe/result.jsp?task=5bd14c218bdf4a799d6fe98cb8880d20&view=view_all_annotations_DB). As a way to enlarge the annotation coverage, DEREPLICATOR algorithm was used, which surprisingly failed to annotate any natural peptides. Nevertheless, two other in silico annotation tools ISDB and Moldiscovery allowed the annotation of 184 and 318 nodes respectively. They were mapped as red rectangles and shaped as triangles on the network, respectively ( Figure S1 ; Tables S4 , S5 ). Interestingly, 51 nodes were annotated by both tools, and thus appeared as red triangles on the network ( Figure S1 ; Table S5 ). Among these 51 annotations, seven candidate structures were identical ( Figure S2 ), and one of them at m/z 566.2516 ([M+H]⁺ C₃₂H₃₈O₉) was dereplicated by MarinLit as godavarin K,

None of the 23 cyanobacterial extracts tested against four pathogenic bacterial strains (S. aureus ATCC 25923, E. coli ATCC 25922, P. aeruginosa PAO1, and E. faecalis ATCC29212) displayed any significant antimicrobial activity, with minimum inhibitory concentration (MIC) values > 200 µg. mL^-1 ( Text S1 ). Of all 23 strains tested against the Chinese hamster ovary (CHO) cells, only PMC 1121.19, Capilliphycus sp. showed cytotoxic activity with a 50% inhibitory concentration (IC₅₀) value of 15.73 ± 1.64 µg. mL^-1. In addition, weak cytotoxic activity was also observed with both Hydrocoleum sp. Strains PMC 1103.19 and PMC 1116.19, with IC₅₀ values of 94.54 ± 3.98 µg. mL^-1 and 86.37 ± 4.16 µg. mL^-1, respectively ( Figure 8A ).

All 23 cyanobacteria extracts showed at least some degree of anthelminthic activity based on the reduced mobility of the Litomosoides sigmodontis microfilaria ( Figure 8B ). In particular, extracts from Capilliphycus sp. (PMC 1121.19) and Thainema sp. 1 (PMC 1125.19) completely inhibited their movements, with mobility scores down to 0 (no mobility). Extracts from Hydrocoleum sp. (PMC 1116.19), Wollea sp. (1138.19), Planktothrix mougeoti (1100.19), Thainema sp. 2 (1126.19), and Hapalosiphon welwitschii (1099.19) resulted in reducing microfilaria activity (< 0.5), meaning that microfilaria showed very limited mobility when exposed.

The antibiofilm activity of the 23 crude extracts was measured using the percentage of inhibition biofilm formation of the bacterial strain Pseudomonas aeruginosa MUC-N1. A majority of the cyanobacterial strains had low antibiofilm activity against this strain ranging from 2 to 30%. Three strains, namely PMC 1119.19, 1121.19 (Capilliphycus sp.), and PMC 1116.19 (Hydrocoleum sp.), showed moderate antibiofilm activity around 42% ( Figure 8C ). Three strains corresponding to two known species, namely PMC 1097.19, 1099.19 (Hapalosiphon welwitschii), and PMC 1100.19 (Planktothrix mougeotii) displayed the highest antibiofilm activities, ranging from 58.9 to 62.4%. In contrast, six other strains PMC 1101.19 (Brasilonema sp.), 1125.19, 1126.19 (Thainema sp. 1 and 2), 1127.19 (Capilliphycus sp.), as well as 1136.19, 1138,19 (Wollea sp. 1 and 2), showed negative inhibition, particularly strain 1125.19 (Thainema sp. 1, 153.7%) indicating that exposure increased biofilm growth compared to control. Other strains did not show clear-cut inhibition or promotion of biofilm formation.

Identifying cyanobacteria-containing biofilms in the field is not always straightforward. In this study, we sampled biofilms displaying different shades of green, as well as some brownish, slimy, or crustose biofilms. They mostly occurred on sediment surfaces and tree bark, and some of them, for example sample S21, covered very large areas. Despite a tendency for some of the greener ones to display higher relative abundances of cyanobacteria, the abundances were highly variable, pointing out the heterogeneity in biofilm composition. Furthermore, no single cyanobacterial ASVs occurred in more than five out of the 27 biofilm samples, confirming that different biofilms contain different cyanobacteria. Overall, this indicates that each type of biofilm is a potential source of novel strains of cyanobacteria and should be considered for a proper description of the benthic biofilm-forming cyanobacterial diversity.

The 43 strains isolated in this study represent 22 species-level taxa based on a 98.65% similarity threshold over the full-length 16S rRNA-encoding gene. Interestingly, 13 of the isolated species-level taxa were also identified during the characterization of biofilm composition using 16S rRNA metabarcoding. They represent only 8.9% of all identified cyanobacterial ASVs, suggesting that only a minor fraction of the actual diversity of cyanobacteria has been successfully isolated. This is consistent with the acknowledged fact that microbiologists tend to isolate only a small subset of the actual diversity present in Nature (Staley et al., 1985). However, if we consider only the most abundant cyanobacteria that occur in biofilms, the picture is slightly more encouraging. In fact, only 15 cyanobacterial ASVs represented more than 5% of the total 16S rRNA reads in at least one biofilm sample, while none of the other 142 ever represented more than 5% of the total reads in any biofilm sample. Focusing on filamentous morphotypes, we successfully isolated seven of these 15 most abundant cyanobacteria (47%), suggesting that isolating a significant fraction of the most abundant cyanobacterial members of benthic biofilms is tractable, and that continued isolation effort would likely yield additional taxa of ecological importance given their abundance.

Although various studies have successfully reported the isolation of cyanobacterial strains from mangroves (Ceccon et al., 2019; Wu et al., 2022; Zhang et al., 2022), these are usually not associated with a characterization of the communities from which they were isolated, and consequently, the authors are usually unable to address whether they have isolated any of the actual abundant ones. On the other hand, our results suggest that standard isolation procedures can already yield some of the abundant biofilm-associated cyanobacteria, as well as rarer members of the community. Examples of the latter likely include the nine species that we isolated as strains but never detected in biofilm 16S rRNA datasets, in line with the claim that lab-based isolation procedures also often yield rare bacteria (Ferris and Hirsch, 1991), as the culture media compositions or culture conditions are very different from that of natural environment. These rare bacteria may have been detected provided greater depth of sequencing. Isolating, cultivating, and maintaining the abundant members of biofilms in publicly available collections such as the PMC collection provides unique opportunities to unravel biofilm functioning and ecology (Liu et al., 2016; Tamminen et al., 2022), as well as their potential to produce new molecules, and preserve a snapshot of the ecologically important microbial diversity (Delgado-Baquerizo et al., 2016; Vitorino and Bessa, 2018; Gebbie et al., 2020). In situ abundance could even become a criterion to select which strains should be maintained in culture collections.

This study showed that only nine out of 43 strains can be assigned to a described species based on 16S rRNA comparative analysis, representing three species. The other 34 isolated cyanobacterial strains could not be assigned to a species, and belong to 19 species-level taxa for which no name could be proposed (Komárek et al., 2014). Based on current similarity criteria, five of the strains would even warrant classification in three new genera. High levels of novelty distributed over several orders of Cyanobacteria were recently reported in several mangrove habitats, for example in Guadeloupe island (Lesser Antilles), including strains within the Oscillatoriales, Synechococcales, and Nostocales (Duperron et al., 2020). Interestingly, some of the strains isolated in Mayotte have very closely related conspecifics in very distant locations. Strains corresponding to Thainema sp. 2 from this study (PMC 1126.19) were found abundant in the mangroves of Cananéia municipality (25·05^′02^″ S, 47·57^′42^″ W) and the city of Bertioga (23·53^′49^″ S, 46·12^′ 28^″ W), Brazil (Silva et al., 2014), despite being separated by the entire Atlantic Ocean and the African continent. This illustrates that some of the species could be widely distributed. Synechococcales diversity was limited in our study, while this order was for example found abundant in the Mooriganga Estuary, India (Bhadury and Singh, 2020). Differences in environmental factors including anthropogenic impacts as well as isolation bias (we focused on filamentous species) may have led to this result (Rigonato et al., 2012). Similarly, she strains in the Phormidium and Oscillatoria genera were present in 25% and 20% of mangrove forest samples in the Persian Gulf (Zaheri et al., 2021), while no species within these two genera was found in our study. These examples shown here illustrate that there also exist potential differences linked to geography.

Further features beyond simple 16S rRNA dissimilarity- and phylogeny-based criteria are required for determining the accurate status of strains as new genera or species. Polyphasic methods will be necessary to provide a proper taxonomic description, which are beyond the scope of this paper (Komárek et al., 2014; Cellamare et al., 2018). Nevertheless, the high level of taxonomic novelty found in the coastal mangrove biofilms in Mayotte is already self-evident based on our results. Very limited data are available on the taxonomic diversity of marine cyanobacteria in Mayotte, except for a recent paper describing Nostocales cyanobacterial strains, which are not closely related to the strains described herein, and belong to the genera Chrysosporum, Dolichospermum, Raphidiopsis (Duval et al., 2018). Overall, the novelty level shown here shows that the cyanobacterial diversity is mostly unexplored in Mayotte. The potential of tropical ecosystems, particularly mangroves, as reservoirs of cyanobacterial diversity (Aburto-Oropeza et al., 2008; Liao et al., 2020; Roberts et al., 2021), as well as the predominance of orders Oscillatoriales and Synechococcales among the recovered strains (Zubia et al., 2016; Zubia et al., 2019), have been highlighted by several authors.

The annotation of 2,667 nodes organized into 131 clusters at the genus identification level revealed that many were strain-specific or taxa-specific, suggesting important inter-strain heterogeneity in their produced molecules. The limited coverage of cyanobacteria in existing databases only allowed us reliable annotation of very few compounds.

Among the ISDB-DNP annotated nodes, 28 were from the marine field, 15 from cyanobacteria, and 6 compounds from bacteria. For a deeper analysis, the results generated from different in silico annotation tools, namely ISDB, MolDiscovery, and SIRIUS in conjunction with the marine natural product database MarinLit were combined. The three databases (MarinLit, ISDB, and MolDiscovery) consensually proposed godavarin K as the most probable annotated compound for [M+H]⁺ ion m/z 567.26008 ( Text S2 ). Interestingly, SIRIUS did not suggest godavarin K as a structure candidate. Further exploration of the database PubChem (Kim, 2021) revealed that the molecular formula C₂₉H₃₆O₉ (molecular weight 528.6 g.mol^-1) was erroneously attributed to godavarin K. Moreover, it turns out that this entry in PubChem was linked to the reference of godavarin A-J (Li et al., 2010), which does not include godavarin K.

In this study, the node annotated as godavarin K was previously isolated from seeds of mangrove tree Xylocarpus moluccensis, that were collected in the wetlands of Godavari estuary, Andhra Pradesh, India (Li et al., 2011). In addition to godavarin K, dereplication of the molecular formula C₃₂H₃₈O₉ against MarinLit resulted in four limonoids including granatumin C (Li et al., 2009), krishnagranatin B (Liu et al., 2018), thaigranatin R (Ren et al., 2021), and andraxylocarpin A (Li et al., 2012). All of them were obtained from seeds of mangrove trees X. moluccensis or X. granatum, or both. Furthermore, the in silico annotation tool SIRIUS proposed only granatumin C, as a candidate structure.

In addition, five isomers ( Figure 9 ) were shown after a filter of the formula C₃₂H₃₈O₉ in MarinLit was applied, and all five isomeric compounds were isolated from the seeds of the mangrove trees Xylocarpus moluccensis or Xylocarpus granatum or both, collected from wetlands in India or Thailand and published by the same team (Li, 2009; Li et al., 2011; Li et al., 2012; Liu et al., 2018; Ren et al., 2021). Our new results suggest that the five isomer compounds might have been produced by seed-associated cyanobacteria that may be associated with the mangrove trees. Whatsoever, organism-associated cyanobacteria must be considered as potential producers of some molecules that are usually attributed to their hosts (Habibi et al., 2017), as shown in a recent example in which cyanobacteria were the probable source of a molecule found associated with a marine sponge (Castaldi et al., 2022).

Interestingly, the node annotated as godavarin K was only found in strain Planktothrix mougeotii, PMC 1114.19, isolated from sample S21 collected in the Dembeni mangrove. Of note, four additional strains isolated from the same source were also identified as P. mougeotii but did not display the node annotated as godavarin K. This may due to chemical heterogeneity between individuals, possibly underpinned by genome variability between strains as previously highlighted in several studies on Microcystis and Aphanizomenon (Meyer et al., 2017a; Pérez-Carrascal et al., 2019).

No antimicrobial activity and limited cytotoxic activity on the cell lineages tested were detected for the 23 tested strains. On the other hand, some of these strains showed significant antihelminthic and antibiofilm activities and will be further explored to identify and characterize specific molecules responsible for these activities. Strains of Planktothrix mougeotii, for example, PMC 1100.19, displayed very high anti-biofilm activity in our lab-based test. In nature, antibiofilm activity could be relevant to the biofilm lifestyle by preventing overgrowth by competing microorganisms. Although actual in situ relevance of the activity measured in our test remains to be evaluated, P. mougeotii represents 94.5% of cyanobacteria and 44% of all bacteria in sample S21, suggesting it can outcompete other strains present. Similarly, our tests demonstrated anthelminthic activity in several strains. Nematode worms include various species that can graze on biofilms (Moon et al., 2021). Despite that our tests were targeted towards health-relevant nematodes, this type of strong anthelminthic activity could provide a level of protection against grazing. Certain species such as Hydrocoleum sp., which makes up 20% of cyanobacteria in a biofilm, displayed both types of activities. Indeed, one strain inhibits biofilm formation (PMC 1116.19) while another is efficient against nematodes (PMC 1103.19). Wollea sp. 2 (PMC 1138.10, 1140.19) represented 28% of cyanobacteria and 16% of all bacteria in one biofilm and also showed anthelminthic activity. However, opposite results were also observed with several strains displaying biofilm growth-promoting activity culminating in strain PMC 1125.19 (Thainema sp. 1). Biofilm growth inhibition is thus not systematic.

Activities observed in the different strains indicate that not all strains within a given cyanobacterial genus or species express the same activities, even when cultured under the same conditions. It suggests that specific genes or metabolic pathways may be present or expressed in some but not all strains of a given taxon, resulting in inter-strain heterogeneity that could be related to gene exchanges (e.g. lateral transfers). It also points towards the possibility that some of the activities are the result of interactions with members of the phycosphere, consisting of other microorganisms (mainly bacteria) that are isolated from the cyanobacteria and co-cultured with them (Zhu et al., 2016). This phycosphere interacts with and influences the physiology of the cyanobacterial strains, as has been seen in other organisms that coexist with microbiota, and may even be in an obligate symbiotic relationship. Investigating inter-strains genomic and phycosphere variability will be necessary to understand why different strains within a single species and cultivated under comparable conditions might demonstrate significantly distinct chemical compounds and activities.