Coral-associated bacteria were isolated from four scleractinian coral species, Acropora millepora, Acropora tenuis, Pocillopora acuta, and Stylophora pistillata, all collected from Davies Reef (18°49’03.7”S, 147°38’39.6”E). Corals were maintained at the Australian Institute of Marine Science (AIMS) National Sea Simulator (SeaSim) and were healthy when the samples were collected, with no visual signs of bleaching or disease. Five coral fragments per colony were rinsed in sterile artificial sea water (ASW) prior to mucus collection using sterile 50 mL syringes fitted with 20-gauge hypodermic needles. Mucus samples were kept on ice and processed within an hour. In addition, two coral fragments per colony were placed in separate Whirl-Pak sterile sample bags (Nasco, United States) and immediately air-brushed with 5 mL of sterile ASW to remove coral tissue and their associated microorganisms from the coral skeleton. The tissue slurry was homogenised and transferred into sterile 50 mL centrifuge tubes, placed on ice, and processed within an hour.

To isolate coral-associated bacteria, each sample type (mucus and tissue slurry) was serially diluted in ASW (2-, 10-, 100-, and 1,000-fold; Figure 1A ). Aliquots (50 µL) of each dilution were then spread onto Difco Marine Agar 2216 (MA; Becton Dickinson, United States) or modified minimal basal medium (MBM) agar enriched with mixed carbon sources [300 mM; details in Table S1 ; (Curson et al., 2017)], methionine (C₅H₁₁NO₂S; 0.5 mM), and ammonium chloride, (NH₄Cl; 20 mM) as nitrogen source. All agar plates were incubated at 28°C in the dark for one week and inspected daily for growth and the formation of morphologically distinct individual colonies. Colonies were picked using sterile 20 µL pipette tips and resuspended in 5 mL of Difco Marine Broth 2216 (MB; Becton Dickinson, United States). The isolates were incubated at 28°C and 180 RPM until growth was visible. These liquid cultures were replated on MA and this procedure repeated until pure isolates were obtained. Isolates were then cultured in MB and aliquots of each isolate stored in 20% v/v glycerol at -80°C.

A 5 mL liquid culture of each isolate was grown overnight in MB at 28°C with agitation (orbital shaker at 180 RPM) before being centrifuged for 5 min at 10,000 g and the supernatant decanted. DNA extraction was performed using the DNeasy UltraClean Microbial Kit (Qiagen, Germany) according to the manufacturer’s instructions. Extracted DNA was resuspended in 20 µL of UltraPure DNase/RNase-Free Distilled Water (Invitrogen, United States) and quantified by spectrophotometry (NanoDrop ND-1000, ThermoFisher, United States). Aliquots of extracted DNA were diluted with sterile Milli-Q water to 10 ng µL^-1 and stored at -20°C until required.

Extracted DNA was used as template in PCR with the universal 16S rRNA genes primers 27F and 1492R (Lane, 1991), as well as the dsyB specific primers dsyB_deg1F and dsyB_deg2R that amplify a 246 bp region of the gene (Williams et al., 2019). Each PCR reaction mixture contained 1× reaction buffer, 2 mM of MgCl₂ solution, 1 mM of deoxyribonucleotide triphosphate (dNTP) mix, 0.4 µM of each primer, 0.5 µL of BIOTAQ DNA Polymerase (Bioline, United Kingdom), 1 ng µL^-1 of template DNA, and adjusted to a final volume of 25 µL with UltraPure DNase/RNase-free distilled water (Thermo Fisher Scientific, United States). PCR amplifications were as follows: (i) dsyB_deg1F/dsyB_deg2R: initial step at 95°C for 5 min; 30 cycles at 95°C for 30 s, 61°C for 1 min and 72°C for 15 s; and a final extension step at 72°C for 5 min; (ii) 27F/1492R: as described by Bourne and Munn (2005). PCR products were purified with the Wizard SV Gel and PCR Clean-Up System (Promega, United States) and visualised via electrophoresis on a 1% agarose gel stained with ethidium bromide.

PCR products were sequenced at Macrogen Inc. (Seoul, South Korea). The forward and reverse 16S rRNA and dsyB amplicon sequences were paired and the overlapping fragments were merged using Geneious Prime 2019.2.3 (Biomatters, New Zealand). Some dsyB amplicon sequences were too short (or of poor quality) to be merged and only one sequence (forward or reverse) was used.

To identify the closest taxonomic-relative of each isolate, BLAST searches (Altschul et al., 1990) were conducted through the National Center for Biotechnology Information (NCBI). 16S rRNA gene sequences were then aligned using MAFFT (Multiple Alignment using Fast Fourier Transform) v7 (Katoh et al., 2002; Katoh and Standley, 2013) with default settings, then trimmed using trimAl v1.4 (Capella-Gutiérrez et al., 2009) to remove sites with more than 50% missing or degraded data. Following the Bayesian Information Criterion (Schwarz, 1978), the maximum likelihood phylogeny for each sample niche was calculated. Phylogenetic trees were constructed using IQ-TREE v1.6.12, with 1,000 ultrafast bootstrap replicates (Minh et al., 2013), and formatted using the ggtree package (Yu et al., 2017) in R (R Core Team, 2020).

The dsyB amplicons were translated into protein sequences and searches to find regions of local similarity were performed using BLASTP. These translated sequences were also aligned with DsyB from Labrenzia aggregata (AOR83342) (Curson et al., 2017) to assess their similarity using the Needleman-Wunsch algorithm (Needleman and Wunsch, 1970). A multiple sequence alignment (MSA) of the prokaryotic DsyB protein sequences was visualised to show conserved residues, conservative mutations, and divergence between the different homologues. The MSA was conducted using T-Coffee v11.00 (Notredame et al., 2000; Di Tommaso et al., 2011) with default settings and formatted using Boxshade v3.21. Conserved domains within the predicted DsyB sequence were detected using CD-Search (Marchler-Bauer and Bryant, 2004) against the Conserved Domain Database (CDD) v3.18 (Lu et al., 2019).

To confirm the DMSP biosynthesis capability of bacteria harbouring dsyB, each strain was cultured in 500 mL of either MB, yeast tryptone sea salts [YTSS; (González et al., 1996)], MBM broth or methionine-enriched MBM broth (final concentration 0.5 mM). The cultures were incubated at 28°C with agitation (orbital shaker at 180 RPM) for 24 hours before being harvested by centrifugation (3,000 g for 15 min at 4°C). The clarified medium was discarded and the remaining cell pellets snap-frozen with liquid nitrogen, lyophilised overnight (Dynavac freeze dryer, Massachusetts, United States; model FD12) and stored at -20°C until required.

The freeze-dried cell pellets were resuspended in 1 mL of deuterated methanol (CD₃OD; Cambridge Isotope Laboratories, Massachusetts, United States), vortexed at maximum speed for 5 min, and sonicated for 5 min at room temperature. A further 1 mL of CD₃OD and 666 µL of deuterium oxide (D₂O; Cambridge Isotope Laboratories, Massachusetts, United States) were added into the mixture (for a final CD₃OD to D₂O ratio of 3:1), which were then vortexed at maximum speed for 5 min and sonicated for 10 min at room temperature. Bacterial extracts were subsequently centrifuged at 3,000 g for 5 min. The particulate-free extracts (final volume of ~2.6 mL) were then used for subsequent analyses on the LC-MS and NMR.

LC-MS was used to assess the presence of intracellular DMSP in dsyB-positive bacterial isolates. Particulate-free bacterial extracts were analysed on an Agilent 1100 series high performance liquid chromatograph coupled to a Bruker Esquire 3000 quadrupole ion trap mass spectrometer (LC-MS; Bruker Daltonics, Massachusetts, United States) equipped with an electrospray ionisation interface (ESI). Extracts (5 µL) were separated on a reverse-phase Luna 3 μm HILIC column (Phenomenex, California, United States; 150 × 3 mm, with a particle size of 3 µm) maintained at 25°C. Separation was achieved using a programmed step gradient consisting of solvent A: 0.1% formic acid (HCOOH) in Milli-Q water and solvent B: methanol (CH₃OH, HPLC grade OmniSolv), at a flow rate of 0.5 mL min^-1. The column was pre-equilibrated at 60% B for 10 min prior to injection. The programmed step gradient was t = 0 min, 60% B; t = 12 min, 10% B; t = 14 min, 10% B; t = 15 min, 60% B; t = 20 min, 60% B; t = 22 min, 60% B. The ESI was operated in positive mode and the target mass of m/z 135, corresponding to the [M+H]⁺ of DMSP, monitored (established from a DMSP standard).

The presence of intracellular DMSP in dsyB-positive bacterial isolates was also assessed using NMR. A 700 µL aliquot of each bacterial particulate-free extract was transferred into a 5 mm Norell 509-UP NMR tube (North Carolina, United States) and analysed immediately using quantitative NMR (qNMR) via the ERETIC method (Electronic REference To access In vivo Concentrations) (Akoka and Trierweiler, 2002) to measure the concentration of DMSP, as described in Tapiolas et al. (2013).

NMR spectra of the bacterial extracts were recorded on a Bruker Avance 600 MHz NMR spectrometer (Bruker BioSpin, United States) with a triple resonance cryoprobe (TXI), referenced using CD₃OD (δ_H 3.31). ¹H NMR spectra were acquired as outlined in Tapiolas et al., (2013) using a standard Bruker solvent suppression pulse sequence. 2D NMR spectra were also acquired to confirm the assignment of DMSP. All spectra were referenced to residual ¹H and ¹³C resonances in CD₃OD. In addition, one extract was spiked with 14 µL of 50 mM DMSP to confirm the position of the methyl singlet, as NMR signals can shift as the sample matrix changes.

A phenol:chloroform extraction method, outlined in detail in Raina et al. (2016), was used to extract high-molecular weight DNA from a representative bacterial isolate producing DMSP (called AMM-P-2 hereafter). Extracted DNA was sent to the Ramaciotti Centre for Genomics (Sydney, Australia) for library preparation using the Nextera XT DNA Library Preparation Kit (Illumina, United States) and sequenced on the Illumina MiSeq system using V2 with 2×250 bp paired-end reads.

The MiSeq read set was trimmed, assembled, and error-corrected using Trimmomatic 0.38 (Bolger et al., 2014), SPAdes v3.13.0 (Bankevich et al., 2012), and Pilon v1.23 (Walker et al., 2014), respectively. All were implemented through Shovill v1.0.4 using default settings. Prediction of coding regions and annotation were performed with Prokka v1.14.6 (Seemann, 2014) using standard databases (i.e., ISfinder, NCBI Bacterial Antimicrobial Resistance Reference Gene Database, and UniProtKB (SwissProt)) and default settings.

To confirm the presence of dsyB within the genome of AMM-P-2, we used a reciprocal BLAST approach between the protein sequence reported by Curson et al. (2017) and the AMM-P-2 genome (E-value ≥ 1 e^-50). Similarly, to explore the genomic potential of the bacterium to utilise and metabolise other sulfur compounds (e.g., sulfate, cysteine), target genes were obtained from KEGG (Kyoto Encyclopedia of Genes and Genomes) (Kanehisa et al., 2004) and compared against the AMM-P-2 genome. The orthology of the highest scoring match (E-value ≥ 1 e^-50) was confirmed by conducting BLASTP analyses against the NCBI NR (non-redundant) database, followed by BLASTP analyses of retrieved best match against the AMM-P-2 predicted proteins.

Isolate AMM-P-2 was grown in MBM broth with methionine (final concentration 0.5 mM) (Curson et al., 2017), and the culture incubated at 27°C, 180 RPM, ambient lighting, and 35 practical salinity units (PSU). After three days, a 1:10 dilution of this starter culture was inoculated into 60 mL replicate cultures and incubated under different environmental conditions simulating stress experienced by tropical corals: (i) high temperature (32°C; T₃₂), (ii) low temperature (22°C; T₂₂), (iii) high UV (through a combination of Deluxlite Black Light Blue (18W) and Reptile One UVB 5.0 (18W) with an average total radiation in the incubator of 1.328 mW cm^-2 measured using a Solarmeter Model 5.0 UVA + UVB meter (Solar Technology, Pennsylvania, United States)), (iv) complete darkness, (v) high salinity (40 PSU), and (vi) low salinity (25 PSU). Three biological replicates were grown for each of the described conditions, including control conditions maintained at ambient lighting, and 35 PSU.

Culture density was monitored through time using spectrophotometry (2 µL was measured at 600 nm (OD₆₀₀) on a NanoDrop 1000 spectrophotometer; Thermo Fisher Scientific, United States). Cultures were sampled over four time points, corresponding to the mid-exponential (24 h), late exponential (28 h), early stationary (32 h), and late stationary (36 h) growth stages of the bacterium, as previously established by a standard growth curve. At each time point, one culture per treatment was removed and samples were taken for quantitative nuclear magnetic resonance (qNMR) analysis (50 mL; centrifuged for 5 min at 3,000 g, pellet snap frozen and stored at -20°C until analysis).

To convert the optical density data recorded with spectrophotometry into bacterial cell numbers, OD₆₀₀ and flow cytometry counts were carried out simultaneously on isolate AMM-P-2 grown under standard conditions in triplicate (27°C at 180 RPM, ambient lighting, 35 PSU, in modified MBM). After each OD₆₀₀ measurement, 100 µL of cells were fixed for 15 min in 2% glutaraldehyde for subsequent flow cytometry analysis. Samples were then stained with SYBR Green (1:10,000 final dilution; ThermoFisher, Massachusetts, United States), incubated for 15 min in the dark and analysed on a CytoFLEX S flow cytometer (Beckman Coulter, California, United States) with filtered MilliQ water as the sheath fluid. For each sample, forward scatter (FSC), side scatter (SSC), and green (SYBR) fluorescence were recorded. The samples were analysed at a flow rate of 25 µL min^-1. Microbial populations were characterized according to SSC and SYBR Green fluorescence (Marie et al., 1997) and cell abundances were calculated by running a standardized volume (50 µL) per sample.

Statistics were performed using IBM SPSS Statistics v27.0.1. qNMR signals associated with DMSP and acrylate were normalised to cell density and tested for significance using repeated-measures ANOVA, with Greenhouse-Geisser correction applied (a correction for sphericity). A simple main effect test, applied following significant interactions between treatments, was used to determine the difference between treatments at each time point for both DMSP and acrylate. A Pearson product-moment correlation was used to determine the relationship between DMSP and acrylate concentrations in AMM-P-2.

A total of 157 isolates were recovered from coral mucus (51%) and tissue (49%) ( Figure 1B and Table S2 ). The highest proportion of these were members of the Gammaproteobacteria (49%), followed by Alphaproteobacteria (32%) and Flavobacteriia (17%) ( Table S2 ). The taxonomic composition of the isolates in each compartment was different ( Figure 1 and Table S2 ), and few isolates were shared between coral species ( Figure 1B ).

From the 157 bacterial isolates screened, 14 harboured the dsyB gene (9% of total isolates). These strains all belong to the Alphaproteobacteria class, specifically the family Rhodobacteraceae, and include members of the Shimia (n=10), Roseivivax (n=3), and Pseudooceanicola (n=1) genera ( Figure 1C and Table S2 ). All of these dsyB-harbouring strains were isolated from Acropora millepora and Acropora tenuis, with 3 strains derived from the mucus and 11 from the tissue. More specifically, bacteria harbouring dsyB represented 19% (10 of 52) of the isolates from A. millepora tissue. The average sequence identity of the predicted (partial) DsyB amino acid sequences derived from the isolates to the protein from Labrenzia aggregata was ~65% ( Figure S1 ) (Curson et al., 2017; Curson et al., 2018). Interestingly, three different Labrenzia strains were isolated from corals, including two strains with more than 98% 16S rRNA gene sequence identity to L. aggregata, but dsyB could not be amplified from these isolates.

The 14 dsyB-harbouring strains were assessed for their ability to produce DMSP. The isolates were initially grown in MB and YTSS, but DMSP could not be detected in any of the cultures using either LC-MS ( Figure S2 ) or NMR ( Figure S3 ). However, LC-MS did detect a peak at m/z 135 and retention time of 5.9 min, consistent with DMSP, in all isolates grown in methionine-enriched MBM ( Figure S4 ). ¹H NMR analysis established the presence of a well-resolved diagnostic singlet at δ_H 2.95 ppm (2 × CH₃) (Tapiolas et al., 2013) ( Figure 2 ) and ¹H-¹H correlation spectroscopy (COSY) revealed two coupled methylene groups, S-CH₂- (δ_H 3.45, t) and -CH₂-CO₂H (δ_H 2.70, t) ( Figure S5 ). In addition, a ¹H-¹³C heteronuclear multiple bond correlation (HMBC) revealed long-range chemical shift correlations between: (i) the protons of the two methyl groups (S-CH₃) and the carbon of the S-methylene group (δ_H 2.94 - δ_C 43.21); (ii) the carboxyl-methylene protons of the carboxyl carbon (δ_H 2.69 - δ_C 172.03) and the S-methylene carbon (δ_H 2.69 - δ_C 43.21) ( Figure S6 ). Together, these COSY and HMBC correlations confirmed the structure of DMSP, thus verifying the presence of DMSP in the bacterial extracts.

To further characterise the genomic underpinnings of DMSP production in coral-associated bacteria, isolate AMM-P-2 was selected for whole-genome sequencing ( Table 1 ). This isolate belongs to the Shimia genus, which accounted for 71% (or 10 of 14) of dsyB-positive bacteria isolated. This strain exhibited >97% similarity with Shimia aestuarii based on its full 16S rRNA gene sequence.

Comparison of the predicted amino acid sequence of DsyB from S. aestuarii AMM-P-2 with previously characterised homologues representing the diversity of this protein family (n=14) revealed the presence of two conserved domains which are common to all dsyB orthologues ( Figure S7 ): (i) an S-adenosylmethionine-dependent methyltransferase (AdoMet-MTase) class I superfamily domain (pfam00891; E-value 4.02 e^-16) and (ii) a dimerization2 superfamily domain (pfam16864; E-value 4.13 e^-7). As in several other Rhodobacterales strains (Curson et al., 2017), isc [iron sulfur cluster] or suf [sulfur formation] gene clusters were present 5’ of dsyB in S. aestuarii AMM-P-2 ( Figure S8 ). However, in the region 3’ of dsyB, only limited synteny was observed between S. aestuarii AMM-P-2 and other Rhodobacterales.

To determine the source of sulfur used by S. aestuarii AMM-P-2 for DMSP biosynthesis, its sulfur metabolic potential was assessed. Distinct orthologues of all the enzymes in the sulfate reduction pathway were identified ( Figure 3 and Table S3 ), confirming S. aestuarii AMM-P-2 has the genetic machinery required to uptake and convert extracellular inorganic sulfate to sulfide ( Table S3 ). Following sulfide production, S. aestuarii AMM-P-2 can produce cysteine (through serine or homocysteine) and ultimately the DMSP precursor methionine (via aspartate; Figure 3 ).

Turning the attention to DMSP catabolism, S. aestuarii AMM-P-2 has the genetic potential to both demethylate and cleave this sulfur compound ( Figure 3 and Table S3 ). S. aestuarii AMM-P-2 has the entire dmdABCD suite of genes enabling conversion of DMSP to MeSH (Zhang et al., 2019). This bacterium also contains the DMSP lyase genes dddP and dddW, whose products liberate equimolar amounts of DMS and acrylate from DMSP (Zhang et al., 2019). It is also able to metabolise acrylate to propionyl-CoA through acuI, acuH and prpE genes (Wang et al., 2017). Significant matches to dddA and dddC (E-values ≥ 1.39 e^-93) which catabolise 3-hydroxypropionyl-CoA (3HP) to acetyl-CoA (Wang et al., 2017) ( Table S3 ) were also identified.

To assess the influence of environmental factors on DMSP production by S. aestuarii AMM-P-2, cultures were grown in methionine-enriched MBM under six different conditions and sampled four times over a 48-hour incubation period. The conditions tested are known to elicit stress responses in corals (high and low temperature, high and low salinity, high UV, and darkness), and were compared to controls. Intracellular DMSP levels per cell within each treatment were not statistically different through time (repeated-measure ANOVA, p = 0.078; see Table S4 ) and no significant interaction was identified between time and treatments (repeated-measure ANOVA, p = 0.204; Table S4 ).

Changes in salinity affected DMSP levels per cell ( Figure 4A ). Levels of DMSP in cultures grown under hyposaline conditions (25 PSU) were 63% lower than the controls (35 PSU) over all four time points. In contrast, the highest DMSP levels were recorded under hypersaline conditions (40 PSU), which were 80% higher than the controls. DMSP levels also increased when cells were grown under high UV (65% higher) or in complete darkness (43% higher) compared to control ambient light conditions. Finally, at higher temperature (32°C), DMSP levels were 42% lower than the controls, while the low temperature treatment (22°C) had no effect throughout the experiment.

Although it was not possible to quantify DMS with qNMR due to the volatility of this compound, DMSP lysis releases equimolar concentrations of acrylate, which was detected in every sample. Overall, the level of acrylate per cell was approximately one order of magnitude lower than that of DMSP ( Figure 4B ). These measurements confirm that a fraction of the synthesized DMSP is channelled through the cleavage pathway. A significant interaction between time and treatments was identified for acrylate levels per cell (ANOVA, p = 0.047; Table S5 ). However, the differences between control conditions and the different treatments were not as pronounced for acrylate compared to DMSP. The 32°C treatment exhibited the lowest levels of acrylate across all timepoints (with 68% less acrylate on average than the control), while the high salinity treatment gave rise to the highest concentrations (with 36% more acrylate).

The six treatments were divided into two categories: (i) those likely to elicit a stress response in bacteria (i.e., high UV, low and high salinity), and (ii) those unlikely to be stressors (i.e., darkness, 22 and 32°C), and the level of correlation of the intracellular levels of DMSP and acrylate between these assessed. No significant correlation was identified between DMSP and acrylate levels for the mild conditions (Pearson’s R = 0.23, p = 0.39; Figure 4D ). However, a strong positive correlation was identified for the stress treatments (Pearson’s R = 0.87, p < 0.001; Figure 4C ).

Of the 157 bacteria isolated from the mucus and tissues of four coral species, approximately 9% harboured the dsyB gene (14 out of 157). Although three genera (four species) of corals were screened, dysB-positive bacteria were only isolated from the two Acropora species, which correlates with the high DMSP production by members of this genus (Tapiolas et al., 2013). Note, it is possible that other DMSP-producing bacteria were present in the corals sampled, potentially containing mmtN or other unknown DMSP synthesis genes (Williams et al., 2019). All bacteria with dsyB belonged to the Rhodobacterales order (Curson et al., 2017), with Shimia species representing 71% of these isolates (10 out of 14). Shimia are metabolically versatile members of the Roseobacter clade (Choi and Cho, 2006), and are abundant in the water column, marine sediments, and commonly associated with eukaryotic hosts (Lenk et al., 2012; Luo and Moran, 2014), particularly phytoplankton (Ajani et al., 2018; Behringer et al., 2018) and reef-building corals (Chen et al., 2011; Zhang et al., 2021). The other dsyB-harbouring bacteria isolated in our screen belonged to the genera Pseudooceanicola – which are common DMSP producers in the water column (Zheng et al., 2020) – and Roseivivax – which have been isolated from corals (Chen et al., 2012) and can induce coral larval settlement (Sharp et al., 2015).

DMSP was unambiguously identified in extracts of dsyB-harbouring bacteria when the cells were grown in methionine-supplemented MBM. In MBM lacking methionine, DMSP signal intensity was close to or below the detection limits of the LC-MS and NMR. Addition of pathway intermediates, including methionine, have been shown to enhance DMSP production in bacteria, but most dsyB-harbouring strains can also produce methionine de novo (through sulfate assimilation) (Curson et al., 2017). This study therefore reveals the presence of bacteria capable of DMSP production in corals and shows that these microorganisms are likely to contribute to the DMSP production by coral holobionts.

Three strains of Labrenzia, the bacterial genus from which the dsyB gene was initially identified (Curson et al., 2017), were isolated from Pocillopora acuta and Stylophora pistillata. The 16S rRNA genes of one of the strains was 98.6% similar to Labrenzia alba, while of the other two were more than 98% similar to L. aggregata. However, the dsyB gene was not detected in any of the three strains. This is consistent with the previously-reported lack of de novo DMS production from a Symbiodiniaceae-associated Labrenzia strain (Lawson et al., 2020), suggesting that the ability to produce DMSP is not conserved across the Labrenzia genus.

To investigate the potential of a representative DMSP-producing bacterium from corals to metabolise sulfur compounds, the genome of S. aestuarii AMM-P-2 was fully sequenced. S. aestuarii is capable of assimilating sulfate from seawater to produce sulfide and sulfur-containing amino acids. Although all the genes required for methionine synthesis were present, its growth on methionine-enriched media indicates that it can also use exogenous methylated sulfur compounds, enhancing the pool of reduced sulfur intermediates available for the synthesis of DMSP. The S. aestuarii AMM-P-2 genome also encodes both the demethylation and the cleavage pathways, allowing it to produce the sulfurous gases methanethiol and DMS. Given that metabolic interdependencies between different partners are common in symbiotic systems, it is important to note that S. aestuarii AMM-P-2 has all the required genes to produce DMSP de novo.

As in other Rhodobacterales (Curson et al., 2017), dsyB in S. aestuarii AMM-P-2 is located downstream of multiple genes (iscRS, sufBCDS) encoding iron-sulfur cluster (Isc) proteins involved in the formation of Fe-S clusters and cellular defence against oxidative stress. Specifically, the cysteine desulfurase IscS provides the sulfur that is then incorporated into Fe-S clusters and its deletion renders the cells hypersensitive to oxidative stress (Ayala-Castro et al., 2008; Rybniker et al., 2014). In addition, the IscR protein is a transcriptional regulator of the suf operon (Yeo et al., 2006) and both the suf and isc operons are highly induced by oxidative stress (Yeo et al., 2006). Therefore, the tight linkage of dsyB to the suf and isc operons suggests that DMSP transcription may be directly affected by oxidants (Sunda et al., 2002; Curson et al., 2017).

DMSP levels in S. aestuarii cells were affected by changes in salinity. Fluctuations in salinity are known to affect DMSP levels in corals (Gardner et al., 2016; Aguilar et al., 2017), while also impacting DMSP degradation in bacteria (Salgado et al., 2014; Liu et al., 2018). The nearly 5-fold increase in intracellular DMSP under hypersaline compared to hyposaline conditions reported here is consistent with previous reports in algae (Vairavamurthy et al., 1985; Kirst et al., 1991; Trossat et al., 1998), and bacteria (Curson et al., 2017), and supports the proposed function of this molecule as an osmolyte (Kirst et al., 1991).

UV exposure and complete darkness both caused significant increases in DMSP levels in S. aestuarii cells. The level of UV radiation applied here was approximately 50% higher than values previously measured at 1 m depth on the Great Barrier Reef (Nordborg et al., 2018), and is therefore likely to cause oxidative stress in the S. aestuarii cells. Indeed, because of their lack of pigmentation and low internal self-shading due to small cell volume, heterotrophic bacteria are amongst the most UV-sensitive organisms (Ruiz-Gonzalez et al., 2013). Similar increases in DMSP levels triggered by UV exposure have been reported in phytoplankton (Sunda et al., 2002; Slezak and Herndl, 2003) and have been attributed to the antioxidant capacity of DMSP (Sunda et al., 2002). More surprising was the increase in DMSP levels in cells grown in darkness. Recent studies have revealed that bacteria in aphotic environments, such as the deep ocean as well as coastal and deep sediments, produce substantial amounts of DMSP (Williams et al., 2019; Zheng et al., 2020). Although the effect of sunlight on bacterioplankton has been extensively studied at the community level (Ruiz-Gonzalez et al., 2013), the impact of prolonged darkness on the physiology of heterotrophic bacteria has received little attention to date. Our results suggest that the absence of light influences sulfur metabolism in S. aestuarii, and further investigation should aim to identify the mechanism driving the production of DMSP in aphotic conditions.

Intracellular acrylate levels in S. aestuarii AMM-P-2 were almost one order of magnitude lower than those of DMSP and decreased significantly over time in all growth conditions. Acrylate and DMSP levels were strongly and positively correlated under conditions likely to stress the bacteria (e.g., salinity, UV), but were decoupled under conditions that are stressful for the coral host but not necessarily for the bacteria (e.g., small temperature variations, darkness). This suggests that under conditions stressful for the bacteria, more DMSP is channelled towards the DMSP cleavage pathway, which generates equal amounts of DMS and acrylate, rather than the demethylation pathway (Gao et al., 2020). DMS and acrylate are efficient scavengers of hydroxyl radicals produced by the cells during stress (Sunda et al., 2002), which may explain why the cleavage pathway is preferred during stressful conditions.