Oceanospirillales containing the DMSP lyase DddD are key utilisers of carbon from DMSP in coastal seawater

Ubiquitous and diverse marine microorganisms utilise the abundant organosulfur molecule dimethylsulfoniopropionate (DMSP), the main precursor of the climate-active gas dimethylsulfide (DMS), as a source of carbon, sulfur and/or signalling molecules. However, it is currently difficult to discern which microbes actively catabolise DMSP in the environment, why they do so and the pathways used. Here, a novel DNA-stable isotope probing (SIP) approach, where only the propionate and not the DMS moiety of DMSP was 13C-labelled, was strategically applied to identify key microorganisms actively using DMSP and also likely DMS as a carbon source, and their catabolic enzymes, in North Sea water. Metagenomic analysis of natural seawater suggested that Rhodobacterales (Roseobacter group) and SAR11 bacteria were the major microorganisms degrading DMSP via demethylation and, to a lesser extent, DddP-driven DMSP lysis pathways. However, neither Rhodobacterales and SAR11 bacteria nor their DMSP catabolic genes were prominently labelled in DNA-SIP experiments, suggesting they use DMSP as a sulfur source and/or in signalling pathways, and not primarily for carbon requirements. Instead, DNA-SIP identified gammaproteobacterial Oceanospirillales, e.g. Amphritea, and their DMSP lyase DddD as the dominant microorganisms/enzymes using DMSP as a carbon source. Supporting this, most gammaproteobacterial (with DddD) but few alphaproteobacterial seawater isolates grew on DMSP as sole carbon source and produced DMS. Furthermore, our DNA-SIP strategy also identified Methylophaga and other Piscirickettsiaceae as key bacteria likely using the DMS, generated from DMSP lysis, as a carbon source. This is the first study to use DNA-SIP with 13C-labelled DMSP and, in a novel way, it identifies the dominant microbes utilising DMSP and DMS as carbon sources. It highlights that whilst metagenomic analyses of marine environments can predict microorganisms/genes that degrade DMSP and DMS based on their abundance, it cannot disentangle those using these important organosulfur compounds for their carbon requirements. Note, the most abundant DMSP degraders, e.g. Rhodobacterales with DmdA, are not always the key microorganisms using DMSP for carbon and releasing DMS, which in this coastal system were Oceanospirillales containing DddD. 98tW24pEex88JAHxuSH5r1 Video abstract. Video abstract.

Many metagenomics and metatranscriptomics studies examine the abundance and taxonomy of DMSP catabolic genes and their transcripts to infer the activity of DMSP-degrading microorganisms in marine environments [3,19,36]. However, such studies cannot elucidate with certainty the microbes and pathways used to catabolise DMSP for carbon requirements because many microbes, particularly Rhodobacterales with dmdA and/or ddd genes, cannot grow on DMSP as a carbon source [37] and may degrade DMSP for signalling and/ or reduced sulfur requirements [12,38]. Thus, there is a clear need to establish which microbes in marine samples use DMSP as a carbon source and what DMSP catabolic pathways they contain.
Here we use 13 C-labelled DMSP (with only the propionate moiety 13 C-labelled; Fig. 2a) and DNA-stable isotope probing (DNA-SIP) [39,40] combined with metagenomics and 16S rRNA gene amplicon sequencing to identify organisms in coastal seawater that utilise DMSP (enriched in 13 C-heavy fractions) and DMS (enriched in 13 C-light fractions) as a carbon source (Fig. 2a). Culture-dependent methods were also used to examine bacteria using DMSP as a carbon source. Thus, this study uses DNA-SIP in a novel way to provide key insights into microorganisms utilising DMSP and DMS as carbon sources in natural seawater samples and to distinguish them from microbes that might catabolise these compounds for other uses.

Characterisation of DMSP cycling in coastal seawater
North Sea coastal water, sampled in Great Yarmouth, UK, in January 2018, contained 3.7 ± 0.4 nM DMSP and 0.9 ± 0.4 nM DMS. After addition of exogenous 100 μM DMSP, significant initial rates of DMSP removal (1.1 ± 0.1 μmol h −1 ) and DMS production (0.8 ± 0.02 μmol h −1 ) were detected over the first 48 h of incubation (Fig. 2b). Interestingly, no MeSH was detected in the incubation experiments, suggesting either that DMSP  13 C-DMSP and 12 C-DMSP (control). b DMSP and DMS levels in seawater samples incubated with 13 C-and 12 C-DMSP. Autoclaved seawater (AS) was used as abiotic control. Values show the average of three biological replicates. c DNA retrieved as function of refractive index of each fraction recovered after isopycnic ultracentrifugation. Samples in shaded backgrounds were used for downstream analysis. Triangles: seawater samples incubated with 12 C-DMSP (control). Circles: seawater samples incubated with 13 C-labelled DMSP lysis dominated over the demethylation pathway in these samples or that the MeSH consumption rates were equal to the synthesis rates.

Identification of candidate microbes cycling DMSP in natural (T0) coastal seawater
16S rRNA gene amplicon (16S) and metagenomic (MG) sequencing was used to identify microorganisms in the natural (T0) seawater samples with the potential to cycle DMSP. Analysis of both 16S and MG data revealed that the T0 microbial community was dominated by bacteria, especially Proteobacteria (relative abundance, RA 82.6 ± 7.3% of the total 16S reads), with archaea and eukaryotes present at very low RA (<1%; Fig. S1). Although algae, thought to be the major DMSP producers in photic seawater, were not abundant, potential DMSP-producing cryptophytes Teleaulax [41] and Guillardia, and the chlorophyte Bathycoccus [42] were identified in these T0 seawater samples (Fig. S2a). Furthermore, algal DMSP synthesis genes encoding proteins 100% and 86.6% identical to Prymnesium parvum DSYB and Thalassiosira pseudonana TpMMT [43,44] were not abundant in the T0 seawater metagenomes (Fig. S2b) and were present at similar levels to bacterial dsyB (Table S1). This is not surprising given the DMSP synthesis genes in chlorophyte or cryptophyte algae are not known. Furthermore, few algal DMSP lyase (Alma1) sequences were detected compared to bacterial DMSP lysis genes suggesting that bacteria are likely the major drivers of DMSP cycling in these seawater samples; see below (Table S1; Fig. S2b−c).
Regarding DMSP catabolism, Colwellia strains (RA of 33.2 ± 21.9%) from the Alteromonadales order together with the Rhodobacterales Amylibacter (13.4 ± 5.1%) and Planktomarina (5.3 ± 2.4%) were predicted to be the major DMSP degraders in these T0 seawater samples (Fig. 3, Table S2). Indeed, marine Colwellia spp. isolates have previously been suggested to be important DMSP catabolisers in polar samples [47], and like several Amylibacter spp., they contain dmdA [48]. Furthermore, many Amylibacter spp. and Planktomarina spp., abundant in other coastal samples [48], contain proteins homologous to ratified DddP DMSP lyases (Table  S3). However, it was surprising that Oceanospirillales and SAR11 clade bacteria, often abundant in surface seawater [49] and known DMSP degraders via lysis and/or demethylation pathways [17,27], only comprised 5.3 ± 3.0% and 1.9 ± 1.8% of the T0 microbial community, respectively, according to the 16S data ( Fig. 3; Table S4). MG analysis revealed that dmdA and dddP were the most abundant DMSP catabolic genes in the T0 samples, with 15.6 ± 4.1% and 5.8 ± 1% of bacteria predicted to contain them, respectively ( Fig. 4a; Table S1). The majority of dmdA sequences (48.3%) were from Rhodobacterales, mainly homologous to Amylibacter dmdA, and 21.7% were from Candidatus Pelagibacter of the SAR11 clade ( Fig. 4b and Fig. S4). Nearly all detected dddP sequences in T0 metagenomes (>98.8%) were also affiliated to Amylibacter ( Fig. 4b and Fig. S5). Bacteria containing dddD (1.6 ± 1.1%; mainly related to gammaproteobacterial Amphritea, Marinomonas and Colwellia strains), dddX (1.4 ± 0.3%) and dddQ (0.5 ± 0.5%) were also relatively abundant in the T0 seawater microbial community (Fig. 4a, Fig. S6 and Table S1). Other known DMSP lyases (dddL, dddW, dddK and dddY) were predicted to be in only <0.5% of T0 bacteria (Fig. 4a, Fig.  S6 and Table S1). These data suggest a typical coastal system where Rhodobacterales and SAR11 bacterial DMSP demethylation dominates over DMSP lysis through DddP and, to a lesser extent, DddD, DddX and DddQ [17,50,51]. However, the abundance and taxonomic profile of DMSP catabolic genes cannot predict their corresponding levels of transcription, enzyme abundance and/or activity. Furthermore, it is impossible for such an 'omics study alone to elucidate why these microbes catabolise DMSP, e.g. to provide carbon, sulfur or signalling needs. Therefore, we conducted a DNA-SIP experiment with Microbial community profile of coastal seawater samples at order (a) and genus (b) levels. Bacterial diversity of the natural (T0) and labelled (heavy; H) and unlabelled (light; L) fractions of 13 C-DMSP seawater incubations was analysed by 16S rRNA gene amplicon (16S) and metagenomics (MG) sequencing. "_1", "_2" and "_3" after the sample name represent biological replicates. Biological replicates from 13 C-heavy and 13 C-light fractions were respectively combined before MG sequencing due to their highly similar 16S rRNA gene community profile shown by DGGE (Fig. S7). Only classes and genera with RA >0.5% in at least one of the conditions are represented. Statistically enriched genera in the incubations with 13 C-labelled DMSP (13C_T) compared the natural (T0) samples based on 16S data are labelled with an asterisk. Classes and genera with RA <0.5% are grouped into "others". For 16S and MG data of samples incubated with 12   Relative abundance and taxonomic affiliation of DMSP cycling genes retrieved from coastal seawater metagenomes. a Relative abundance (RA) of DMSP demethylation (dmdA) and lysis (ddd) genes in seawater metagenomes. b Taxonomic affiliation of key genes involved in DMSP catabolism in seawater samples retrieved from metagenomic data. T0: metagenomes from natural samples; 13C_T: total microbial community from samples incubated with 13 C-DMSP; 13C_H: metagenomes from 13 C-heavy fractions; 13C_L: metagenomes from 13 C-light fractions. T0 values represent the average of three biological replicates. Biological replicates from 13 C-heavy and 13 C-light fractions were respectively combined prior to metagenomic sequencing (see the "Methods" section) 13 C-labelled DMSP to determine which microbes use DMSP as carbon source in these seawater samples.

Identification of microorganisms degrading DMSP for use as a carbon source by DNA-SIP
Known DMSP demethylation and lysis pathways in model organisms that use DMSP as a carbon source release MeSH and DMS, respectively, as gaseous products and utilise the propionate component for carbon ( Fig. 1). Thus, to identify microbes using DMSP as a carbon source, we synthesised DMSP from 13 C 3 -acrylic acid and 12 C-DMS (see the "Methods" section), the product hereafter being referred to as 13 C-DMSP, which was used in our DNA-SIP experiments with 12 C-DMSP as a control (Fig. 2a). This strategy was chosen to also enable identification of microorganisms utilising the DMS component of DMSP, i.e. those microbes enriched in incubations with DMSP but not necessarily being represented in the heavy fraction of 13 C-DMSP incubations.
The T0 seawater samples collected above were incubated with 13 C-DMSP or 12 C-DMSP (control) as substrate (100 μM) under 12-h light/dark cycling conditions to consider both phototrophic and heterotrophic DMSP catabolism. No MeSH production was detected in any of the incubations. DMSP degradation, DMS production and subsequent DMS removal processes were similar between the samples incubated with 13 C-DMSP and 12 C-DMSP (Fig. 2b). DNA was extracted after 96 h when 77 μmol DMSP L −1 (231 μmol C L −1 ) was assimilated and DMS levels were decreasing (by 0.7 ± 0.006 μmol h −1 during the last 24 h of incubation), suggesting that DMS degraders were also active at this timepoint. DNA was then separated into heavy ( 13 C-labelled) and light ( 12 C-labelled) fractions by isopycnic centrifugation (see the "Methods" section; Fig. 2c), which were analysed by denaturing gradient gel electrophoresis (DGGE). This technique showed clear differences in the 16S rRNA gene profiles from 13 C-heavy and 13 C-light fractions, whereas fractions from 12 C-DMSP (control) incubations had similar profiles (Fig. S7), indicating that DNA from microorganisms using 13 C-DMSP for carbon had been successfully labelled.
Subsequently, the heavy and light DNA fractions from 13 C-DMSP and 12 C-DMSP (control) incubations were subjected to MG and 16S sequencing to identify those microorganisms that used DMSP as carbon source. Sequence analysis revealed that Gammaproteobacteria dominated in both the heavy (13C_H) and light (13C_L) fractions from 13 C-DMSP incubations (Fig. 3a), but there were major differences at lower taxonomic levels (i.e. order and genus; Fig. 3b). Oceanospirillales, with a RA of 70.2% in MG and 81.2 ± 3.5% in 16S data, was the most abundant order in the 13 C-heavy fractions (13C_H), but only represented <10% of the bacterial community from 13 C-light fractions (13C_L; Fig. 3b, Table S4). Importantly, there was no such increase in Oceanospirillales RA between heavy (12C_H) and light fractions (12C_L) when the T0 samples were incubated with control 12 C-DMSP (Fig. S8a), indicating that these Gammaproteobacteria were the major microorganisms assimilating carbon from the propionate moiety of DMSP in the DNA-SIP experiments.
At the genus level, Amphritea dominated the heavy fractions from the 13 C-labelled microbial community (13C_H), comprising 55.9% and 52 ± 4.3% of the MG and 16S reads, respectively, followed by Marinomonas (9.1% RA in MG and 25.2 ± 0.5% in 16S data; Fig. 3b, Table S2). Other less abundant Oceanospirillales genera significantly enriched (P <0.05) in the 13 C-heavy fractions compared to the T0 samples, were Marinobacterium (to 1.3% RA in MG and 0.6 ± 0.1% in 16S data) and Oceanospirillum (to 0.5% RA in MG and 1.8 ± 0.1% in 16S; Table S2). These four Oceanospirillales genera were 6-to 60-fold more abundant (P <0.05) in 13 C-heavy (13C_H) than in the 13 C-light (13C_L) fractions and constituted 0.4−36% of the total microbial community of the samples incubated with 13 C-DMSP (13C_T, 16S data; Fig. 3b, Table S2). As expected, these genera were also highly abundant in both the 12 C-light and 12 C-heavy fractions (12C_L and 12C_H; Fig. S8b), further supporting them being key bacteria that used DMSP as a carbon source in the DMSP incubations.
In addition, two genera of the Bdellovibrionales order, Halobacteriovorax and an unclassified Bacteriovoracaceae, were also enriched during the 13 C-DMSP incubations, with their RA increasing to 0.7 ± 0.2% and 1 ± 0.1% of 16S reads in the 13C_T samples from <0.1% at T0 (Fig. 3b, Table S2). These Bdellovibrionales bacteria were ~10-fold more abundant in the heavy (13C_H) than in the light (13C_L) fractions from incubations with 13 C-DMSP (Table S2). No sequenced Halobacteriovorax strains contain known DMSP catabolic genes, so if they are true DMSP degraders they may utilise novel pathways and/or enzymes.
As only one DNA-SIP timepoint was analysed here, it is possible that some microorganisms enriched in 13 C-heavy fractions were labelled due to cross-feeding, i.e. they assimilated 13 C-DMSP catabolites released from primary degraders. This is unlikely for the Oceanospirillales given their dominance in the 13C_H fractions and that several previously studied Oceanospirillales strains, e.g. Marinomonas [22], Oceanimonas [52] and Halomonas [31], and bacterial isolates from this work, used DMSP as a sole carbon source and contained DddD (Table 1 and  see below). However, it is plausible that the predatory Bdellovibrionales may have consumed primary DMSP degraders and become labelled due to cross-feeding [53]. Interestingly, Gammaproteobacteria, including members of the orders Alteromonadales (Glaciecola), Vibrionales (Vibrio) and, most notably, Thiotrichales (unclassified Piscirickettsiaceae), were also enriched during the incubations with 13 C-labelled DMSP (RA 0.003−0.2% in T0 to 1.8−17.3% in 13C_T samples according to 16S data; Table S2). These Gammaproteobacteria were more abundant in the 13 C-light than in the 13 C-heavy fractions (RA >15-fold higher in 13C_L than in 13C_H fractions; Fig. 3b, Table S2), suggesting that they likely assimilated the 12 C-DMS generated from the lysis of 13 C-DMSP during the incubations with 13 C-DMSP (see below).
Strikingly, Rhodobacterales and SAR11 bacteria were >6-fold less abundant after the 13 C-DMSP incubations (13C_T) compared to the T0 samples (16S data, Fig. 3a, Table S4). Although some Rhodobacterales 16S rRNA genes, notably from Pseudophaeobacter and Planktomarina, were mainly present in the 13 C-heavy fractions consistent with them using DMSP as a carbon source, these bacteria comprised only 1% of the 13 C-labelled bacterial community (13C_H; Fig. 3b, Table S2). Furthermore, SAR11 bacteria represented 0.3% of the total microbial community from 13 C-DMSP incubations (13C_T) and were >10-fold less abundant in the heavy (13C_H) than light (13C_L) fractions (Fig. 3b, Table S4). These data imply that Rhodobacterales and SAR11 bacteria, which mostly possess dmdA [15,17] and at least one ddd gene [37,54], were not the major users of DMSP as a carbon source under these incubation conditions. It is possible that these bacteria, predicted to be key DMSP degraders in coastal waters [50,55], were using DMSP at the low T0 nM concentration for sulfur demands [56], and/or to generate DMS and/or acrylate as signalling molecules [12] and not primarily as a carbon source. Indeed, most Rhodobacterales with dmdA/ddd genes do not typically grow well on DMSP as sole carbon source under lab conditions [37], in comparison to Oceanospirillales bacteria with dddD (see below) [37] or Alcaligenes faecalis with dddY [26].

Abundance and taxonomy of DMSP catabolic genes in seawater samples incubated with 13 C-DMSP
After incubation with 13 C-DMSP, DMSP demethylation was no longer predicted to be the dominant DMSP catabolic pathway it was in the T0 seawater samples. Metagenomic analysis showed that the RA of dmdA in T0 seawater metagenomes (15.6 ± 4.1% of bacteria) decreased 10-fold to 1.8% after incubation with 13 C-DMSP (13C_T; Fig. 4a and Table S1). Of the dmdA sequences in these samples, 73% originated from the 13 C-heavy fractions and were mainly closely related to DmdA from Rhodobacterales, e.g. Lentibacter (RA 0.8%; Fig. 4b). These data indicate that some bacteria may have demethylated DMSP for carbon assimilation during the incubations with 13 C-DMSP, but that this process was far less significant than in the natural (T0) seawater samples. In support of this, the RA of the ancillary dmdCD genes (Fig. 1) S9). Conversely, dmdB sequences showed a higher RA in the 13C_T than in the natural (T0) metagenomes and were mainly affiliated to Amphritea bacteria (Fig. S9), which were highly enriched in the 13 C-heavy fractions (13C_H ; Fig 3b), have multiple copies of dmdB but not dmdA and are unlikely to demethylate DMSP (Table 1).
In contrast to dmdA, the RA of ddd genes increased to ~60% after the incubation with 13 C-DMSP (13C_T samples), most of which (97%) were retrieved from the 13 C-heavy fraction metagenomes (13C_H; Fig. 4a). Among the ddd genes, dddD was particularly enriched in the 13 C-heavy fraction, with 95.8% of bacteria predicted to contain it ( Fig. 4a and Table S1). Most of the dddD sequences were closely related to the Oceanispirillales genera Amphritea (84.5%) and Marinomonas (10.7%; Figs. 4b and 5a), supporting the 16S and MG data, which showed that these were the dominant genera in the 13 C-heavy fraction (13C_H; Fig. 3, Table  S2). dddD genes likely from Colwellia and Pseudovibrio were also retrieved from the 13 C-heavy metagenomes, although with lower RA (<0.5%; Figs. 4b and 5a). Therefore, the DNA-SIP experiment with 13 C-labelled DMSP clearly showed that Oceanospirillales were the key bacteria degrading DMSP for carbon via a DddD-mediated lysis pathway in these coastal seawater samples. This is not surprising since it has been previously reported that many Oceanospirillales bacteria with DddD, an enzyme with a high affinity for DMSP [57], can use DMSP as sole carbon source for growth [22,31,50].
A complete suite of ancillary genes involved in the downstream catabolism of DMSP (prpE, acuI, dddB, dddC and dddT) [31][32][33] was enriched in the samples incubated with 13 C-DMSP (13C_T) compared to the T0 samples ( Fig. S10 and Table S1). These genes mostly originated from the 13 C-heavy fractions (13C_H) and were affiliated to the Oceanospirillaceae genera Amphritea and Marinomonas (Fig. S10). Interestingly, dddA, whose product catalyses the same reaction as DddB in Halomonas [31], was present at a much lower RA in the 13 C-DMSP incubations (13C_T) compared to the natural (T0) seawater samples, and it was much less abundant in the 13 C-heavy fractions (13C_H) than dddB (4.6% vs 90.9%; Table S1). Thus, DddB likely represents the major route of malonate semialdehyde formation from DMSP in the seawater samples amended with 13 C-DMSP (Fig. 1). Similarly, MG analysis revealed that Oceanospirillaceae bacteria assimilating carbon from 13 C-DMSP in our incubations most likely metabolise acrylate via the PrpE-AcuI enzymes, rather than through AcuN-AcuK (Fig. 1). Although these four genes were mostly present in the 13 C-heavy fractions (13C_H), the RA of acuN and acuK in the total microbial community (13C_T) from the 13 C-DMSP incubations (1.8% and 21.9%, respectively) was lower than in the T0 samples (8.3 ± 1.7% and 34 ± 5.8%; Fig. S10 and Table S1). The opposite trend was observed with prpE and acuI, which had a 4-and 2-fold higher RA in the incubations with 13 C-DMSP (13C_T) compared to the natural (T0) samples ( Fig. S10 and Table S1).

Insights into DMSP cycling from metagenome-assembled genomes (MAGs)
To gain further insights into the microbial sulfur cycling pathways in this coastal environment, metagenomeassembled genomes (MAGs) were reconstructed and analysed. MAGs were screened for the presence of DMSP degradation genes and their RA in each sample calculated as described in Methods. dmdA, the dominant DMSP catabolic gene in T0 samples, was found in the most abundant MAG (5.1 ± 2.1% RA) from the T0 metagenomes (MAG 55, Alphaproteobacteria; Table S5) and also in several less abundant (RA <0.5% each) gammaproteobacterial (MAGs 11, 58 and 60) and Rhodobacterales MAGs (MAGs 3, 7, 18, 44, 68, 81; Fig. S4 and Table S5). MAG 44, taxonomically classified as Loktanella, also harboured a sequence encoding a DddW-like protein with 54% amino acid identity to R. pomeroyi DddW, which did not cluster with ratified DddW sequences (Fig. S6). Most of the dmdA-containing MAGs (except for MAGs 55, 58 and 68) also had dddP and were mainly classified as Rhodobacterales (Table S5). MAGs 55, 58 and 68 were only predicted to be 55.2 to 75.1% complete (Table S5), which might account for why no dddP homologues were detected. The Rhodobacterales MAGs whose taxonomy  from this study (b). a The tree shows DddD proteins from strains isolated in this study (in bold), previously ratified proteins (★), together with sequences retrieved from seawater metagenomes (MG), metagenome-assembled genomes (MAG), and reference strains (▲). Reference strains were selected as the most closely related strains to isolates from this study having publicly available genome sequences. Bootstrap values ≥50% (based on 100 replicates) are shown. Scale bar indicates 5% estimated phylogenetic divergence. b Gene function is indicated by a colour code detailed in legend. Bacteria in bold are those in which the gene clusters have been experimentally ratified could be resolved at lower levels, belonged to the genera Loktanella, Lentibacter and Amylibacter (Table S5), which is consistent with these bacteria often containing both DMSP demethylation and lyase genes (Table  S3). dddP was also found in a Planktomarina temperata MAG (MAG 73) that comprised 2.8 ± 1.9% of the T0 microbial community, and two gammaproteobacterial MAGs (<0.3% RA; Table S5 and Fig. S5). Most of the MAGs containing dmdA and/or dddP genes were not enriched after the 13 C-DMSP incubations (13C_T) compared to the T0 samples except for the Rhodobacterales genera Lentibacter (MAG 81) and Pseudophaeobacter (MAG 68) and gammaproteobacterial Neptuniibacter MAGs (MAG 36, also containing dddD), which were mainly present in the 13 C-heavy fraction (>17-fold higher RA than in 13 C-light fraction; Table S5). These data further support the hypothesis that most Rhodobacterales and Gammaproteobacteria with DmdA and/or DddP are important DMSP degraders in coastal waters but not primarily for carbon assimilation.
Consistent with the 16S data, the RA of an Amphritea MAG harbouring dddD (MAG 23) dramatically increased in the samples incubated with 13 C-DMSP (from 0.7 ± 0.5% in T0 to 35.3% in 13C_T samples) and was 12-fold more enriched in the 13 C-heavy (13C_H) than 13 C-light fractions (13C_L ; Table S5). Prior to this study, Amphritea had not been shown to catabolise DMSP, perhaps because many Amphritea spp. lack dddD or other known DMSP-lyase genes (Table S3).
Another two Oceanospirillales MAGs (MAGs 30 and 85) enriched in the 13 C-heavy fraction (13C_H) after incubation with 13 C-DMSP (26-fold and 3-fold more abundant than in the 13 C-light fraction 13C_L, respectively) were detected (Table S5). However, neither of these MAGs were assigned to Marinomonas, shown to be abundant in the 16S analysis described above, nor did they contain a known DMSP catabolic gene, again potentially due to their low completeness (61.4% for MAG 30 and 51.7% for MAG 85; Table S5).
As described above, Halobacteriovorax bacteria were enriched during incubation with 13 C-DMSP and one Halobacteriovorax MAG (MAG 43) was resolved from the metagenomic data. However, MAG 43 was more abundant in the 13 C-light (13C_L) than in the 13 C-heavy (13C_H) fractions after incubation and it lacked known DMSP catabolic genes (84.5% completeness; Table S5). Further studies are needed to evaluate if and how marine Halobacteriovorax catabolise DMSP.

Characterisation of bacterial strains isolated from seawater samples after incubation with DMSP
Sixty-six bacterial strains with distinct morphologies were isolated after incubations of T0 seawater with DMSP, 67% of which were able to catabolise DMSP yielding DMS or MeSH. All the DMSP-degrading isolates were either Alpha-(27%) or Gammaproteobacteria (73% ; Table S6). Twenty-one representative strains of the different colony types were selected (Table 1), and publicly available genomes of closely related strains (herein described as reference strains) were analysed for the presence of dmd/ddd genes. In addition, we sequenced the genome of six isolates with high DMSP-dependent DMS production rates, but where no closely related reference strains were available, or where these did not contain dmd/ddd homologues (Table S6).
All DMSP-degrading alphaproteobacterial isolates were Rhodobacteraceae, well known for their ability to catabolise DMSP [23,58]. The Sulfitobacter, Neptunicoccus and Pseudophaeobacter isolates, whose corresponding reference strains contained dddP, dddL and/or dddW DMSP lyase genes, released DMS from DMSP, with production rates ranging from 6 to 1411 nmol DMS·mg protein −1 ·h −1 ( Table 1 and Table S6). Most Rhodobacterales, including all reference strains analysed here except for Sulfitobacter, contained dmdA and were predicted to demethylate DMSP [17]. However, only one isolate, Litoreibacter sp. GY12, with no DMSP lyase activity, showed MeSH production from DMSP. In addition, of the twelve Rhodobacterales isolates, only two strains MB13-6 and MC12-6 grew on DMSP as sole carbon source (Table S6). Thus, these data further support the hypothesis that most Rhodobacterales do not catabolise DMSP for carbon assimilation but likely use it as a sulfur source, in oxidative stress protection [8] or for signalling processes [12].
Of the DMSP-degrading gammaproteobacterial isolates, 43% were Oceanospirillales from the genera Cobetia, Amphritea, Marinobacterium and Marinomonas, all of which had relatively high DMSP lyase activity (Table  S6). Indeed, Amphritea and Marinomonas were identified by the DNA-SIP experiments above, as being the two most dominant degraders of DMSP for their carbon demands ( Fig. 3b and Table S2). Furthermore, 63% of the Oceanospirillales isolates grew on DMSP as sole carbon source (Table S6), with their reference strains mostly containing dddD genes ( Table 1, Table S6 and Fig. 5). These data support the major finding from the DNA-SIP work that Oceanospirillales bacteria were the key DMSP degraders for carbon assimilation in these coastal waters.
Other gammaproteobacterial strains from the Alteromonadales (Alteromonas, Marinobacter and Pseudoalteromonas), Pseudomonadales (Pseudomonas) and Vibrionales (Vibrio) orders were also isolated from the seawater samples incubated with DMSP (Table S6). Alteromonas and Marinobacter strains had DMSP lyase activity and could use DMSP as sole carbon source, whereas Pseudoalteromonas sp. GY20 generated MeSH from DMSP but could not use DMSP as a carbon source. Although the genomes of the Marinobacter isolate and its reference strain harboured dddL, no DMSP catabolic genes were found in any genomes of the Alteromonas or Pseudoalteromonas reference strains ( Table 1). All six Pseudomonas spp. isolates had DMSP lyase activity and grew on DMSP as sole carbon source (consistent with previous work [59]), except for isolate MC12-18 (Table S6). In addition, all the genomes of the Pseudomonas reference strains and isolates analysed contained dddD (Table 1). Finally, Vibrio sp. GY15 produced MeSH from DMSP and could use the latter as sole carbon source (Table 1). To our knowledge, no Vibrio spp. have been shown to contain dmdA. Thus, it will be interesting to further investigate the DMSP catabolic mechanisms in the Gammaproteobacteria that cleave and/or demethylate DMSP but lack the known ddd and dmdA genes in their genomes. Given the low RA of Alteromonadales, Pseudomonadales and Vibrionales bacteria (<1.3% in 16S and <1.1% in MG data; Fig. 3, Table S4) in the 13 C-heavy fractions (13C_H) after incubation with 13 C-DMSP, they are unlikely to be as important catabolisers of DMSP for carbon assimilation as the Oceanospirillales in the coastal samples studied here, but they may be in other environments.
The dddD genes in the genomes and MAGs from Oceanospirillaceae bacteria (Amphritea, Marinobacterium, Marinomonas and Neptuniibacter) and Pseudomonadales (Pseudomonas) were always linked to ancillary genes involved in downstream DMSP catabolism (Figs. 5b and 6), showing similar gene synteny (dddBCDRT) to dddD from Marinomonas sp. MWYL1, which assimilates carbon from DMSP for growth [22]. This was also the case for Halomonadaceae bacteria with DddD, including Halomonas sp. HTNK1 and Cobetia sp. MC13-5 but with dddA replacing dddB [31]. Such linkage of DMSP lyase genes to their ancillary metabolic and transport genes, and their coordinated gene expression, is likely important in allowing bacteria harbouring them to utilise DMSP as carbon source.

The role of other bacterial groups in DMS and MeSH cycling
The DNA-SIP strategy used in this study potentially allowed the identification of microorganisms assimilating the 12 C-DMS component of 13 C-DMSP, which should be preferentially represented in the 13 C-light fractions (13C_L) from the incubations with 13 C-labelled DMSP.
Sequencing data revealed that the RA of Thiotrichales from the 13 C-DMSP incubations (13C_T) increased both in the 16S (from 0.5 ± 0.3% to 17.8 ± 8.6%) and MG (from 0.4 ± 0.1% to 15.6%) analyses compared to the T0 samples (Fig. 3a, Table S4). Furthermore, sequences from this order were >58-fold more abundant in the 13 C-light (13C_L) than in the 13 C-heavy (13C_H) fractions (Fig. 3a, Table S4). Such differences were not observed between light (12C_L) and heavy (12C_H) fractions of control samples incubated with 12 C-DMSP (Fig. S8a). Analysis of 16S data showed that these Thiotrichales bacteria comprised Methylophaga and other unclassified genera from the Piscirickettsiaceae family which, based on the higher phylogenetic resolution of the MG analysis, may also be Methylophaga spp. (Fig. 3b). Similarly, the RA of gammaproteobacterial Vibrio and Glaciecola strains were respectively 5-and 9-fold more abundant after the incubations with 13 C-DMSP (13C_T) than in the natural (T0) seawater samples and were mainly present in the 13 C-light fraction (137-and 15-fold higher RA than in the 13 C-heavy fraction; Fig. 3b, Table S2). These data imply that these Thiotrichales and Gammaproteobacteria can use the 12 C-DMS generated from the lysis of the 13 C-DMSP by Oceanospirillales bacteria with DddD for (See figure on next page.) Fig. 6 DMSP degradation pathways in microorganisms from coastal seawater samples. a Although Rhodobacterales (Roseobacter group) and SAR11 dominated the bacterial community of the natural (T0) seawater samples and their DMSP demethylation and cleavage genes (dmdA and dddP mainly) were relatively abundant in the T0 metagenomes (orange bars), the majority of the Roseobacter isolates were not able to grow on DMSP as sole carbon source. Thus, Rhodobacterales and SAR11 are predicted to use DMSP predominantly as a source of reduced sulfur and/or signalling in this coastal seawater. b Oceanospirillales were the major bacteria degrading DMSP for carbon requirements in the seawater incubations with 13 C-DMSP. dddD from Oceanospirillales was the most abundant DMSP lyase gene in heavy fractions from samples incubated with 13 C-DMSP (13C_H; red bars), although other DMSP lyase genes from Roseobacters, i.e. dddL and dddX were also present. Genes involved in the downstream catabolism of 3HP (dddBC in Oceanospirillaceae and dddAC in the Halomonadaceae) and acrylate (prpE and acuI) were also enriched in the 13C_H metagenomes compared to those from T0 samples. c DNA-SIP experiments showed that Methylophaga, a genus of the Piscirickettsiaceae family, and its gene encoding methanethiol oxidase (MTO) were highly abundant in the metagenomes from the 13 C-light fraction (13C_L; blue bars), indicating that these bacteria were the major degraders of the DMS generated from the lysis of DMSP by Oceanospirillales. Bar charts represent the relative abundance of key genes involved in DMSP catabolism in metagenomes from natural (T0) seawater samples, 13 C-DMSP incubations (13C_T) and heavy (labelled, 13C_H) and light (unlabelled, 13C_L) fractions from incubations with 13 C-DMSP. T0 data show the average of three biological replicates, whereas replicates from 13 C-heavy and 13 C-light fractions were pooled prior to metagenomics analysis (see the "Methods" section). MMPA, methylmercaptopropionate; MTA-CoA, methylthioacryloyl-CoA; 3HP, 3-hydroxypropionate; MalSA, malonate semi-aldehyde; MeSH, methanethiol; CCM, central carbon metabolism carbon assimilation. This is further supported by previous studies that have shown that several Methylophaga, Vibrio and Glaciecola strains catabolise DMS [60][61][62][63]. Indeed, the marine methylotroph M. thiooxydans, which was isolated from a DMS enrichment experiment and can use this compound as sole carbon and energy source for growth [64], is considered a model microorganism to study DMS degradation.
To investigate the underlying genetic mechanisms of DMS cycling, the MG data were interrogated with ratified genes involved in DMS metabolism (Fig. 1). Of these, tmm, whose product can generate dimethyl sulfoxide (DMSO) from DMS [65], was the most abundant DMS cycling gene in the T0 seawater samples (9.9 ± 1.7% RA) followed by DMSOR (5.3 ± 3.4% RA; Fig.  S11a and Table S1), which encodes a DMSO reductase enzyme that catalyses the reverse reaction ( [64], Fig. 1). Other genes involved in DMS metabolism such as mddA [66], ddhA [67] and dmoA [68] were present in <0.7% of the bacteria from the natural T0 samples (Fig. S11a and Table S1), suggesting that they have less important roles in this coastal seawater. mtoX, encoding MeSH oxidase [69], was the only known gene involved in DMS metabolism that was enriched in the incubations with 13 C-DMSP (13C_T; 3.1% RA) compared to T0 samples (0.003 ± 0.006% RA), with the majority of sequences being from the 13 C-light fractions (19-fold higher RA in 13C_L than 13C_H DNA; Table S1). As expected, all the mtoX sequences retrieved from the 13 C-light fractions were closely related to DMS-catabolising Methylophaga spp (Fig. S11b). In addition, a Piscirickettsiaceae MAG (MAG 21) containing an mtoX sequence (Table S5) was more abundant in the 13 C-DMSP samples (13C_T; RA 1.5%) compared to the natural (T0) seawater (RA 0.01 ± 0.004%) and was 34-fold more enriched in the 13 C-light (13C_L) than in the 13 C-heavy fraction (13C_H ; Table  S5), further supporting the notion that members of this family are key bacteria cycling DMS in this coastal seawater. These data are also consistent with previous work showing that mtoX, whose transcription and protein expression is upregulated by growth with DMS in Methylophaga thiooxydans, is the only known reporter gene for carbon assimilation from DMS in the Thiotrichales [70]. In M. thiooxydans and likely other microorganisms that use DMS as carbon source, DMS is proposed to be initially demethylated to MeSH [67]. However, the dmoA gene, whose product converts DMS into MeSH, was not enriched in the 13 C-DMSP incubations (13C_T) compared to T0 samples nor was it present in the Piscirickettsiaceae MAG or many of the Methylophaga available genomes (Table S1 and Table S5). Therefore, it is possible that the initial generation of MeSH from DMS in Methylophaga might be catalysed by a novel enzyme and that it rapidly enters central metabolism via a reaction mediated by MTO (MeSH oxidase) [68].

Conclusions
This study represents the first attempt to identify, in tandem, distinct microorganisms degrading DMSP and/ or DMS for carbon assimilation using DNA-SIP with 13 C-labelled DMSP.
Given that Rhodobacterales and to a lesser extent SAR11 with dmdA and dddP genes were highly abundant in the coastal seawater studied here, and that representative strains of these bacteria can grow on DMSP as sole carbon source [28,71], one would predict that they would play a major role in degrading DMSP for their carbon demands. However, DNA-SIP experiments showed that the less abundant Oceanospirillales bacteria in the natural (T0) samples were likely the key degraders of DMSP for carbon assimilation via DddD-mediated DMSP lysis. Oceanospirillales bacteria, e.g. the novel and dominant DMSP degraders in the DNA-SIP experiments, Amphritea and Marinomonas, were predicted to cleave DMSP into DMS and 3HP via their DddD enzyme, with the 3HP being assimilated and the DMS released for the benefit of a different suite of microorganisms (Fig. 6). Sequencing analysis also revealed that Rhodobacterales and SAR11 and their dmdA and dddP genes, though abundant in most surface waters, were scarcely present after incubation with 13 C-DMSP, suggesting that these important bacteria likely utilise DMSP demethylation and DddP-mediated DMSP lysis either for their reduced sulfur requirements [56], to protect against oxidative stress [8] or to generate signalling molecules [12,72]. These conclusions from SIP experiments with 13 C-labelled DMSP were supported by culture-dependent work since (i) most Oceanospirillales isolates, e.g. Amphritea, yielded DMS from DMSP; (ii) these bacteria could use DMSP as a carbon source and in nearly all cases possessed DddD; and (iii) most Rhodobacterales isolates contained dmdA and could cleave DMSP but were not able to use it as a carbon source. However, it should be noted that our DNA-SIP experiments used artificially high (100 μM) levels of 13 C-DMSP, which may have favoured the growth of, and carbon assimilation by, Oceanospirillales via DMSP cleavage rather than SAR11 and Rhodobacterales via DMSP lysis and/or demethylation. Indeed, marine bacteria are thought to favour DMSP cleavage over DMSP demethylation, upon exposure to high DMSP concentrations [21,73]. However, note that a recent study showed that DMSP demethylation required μM DMSP levels, far higher than those for DMSP lysis (>35 nM), to induce transcription of these competing dmd and ddd catabolic genes [74].
This study also provides a new strategy to identify key microorganisms degrading DMS as well as DMSP for carbon assimilation in environmental samples. The use of 13 C-DMSP where only the propionate was 13 C-labelled and not the DMS moiety efficiently identified members of the Piscirickettsiaceae family, especially Methylophaga (well known for their ability to grow on DMS [70]), as important bacteria likely degrading the DMS generated from DMSP for carbon assimilation in these coastal seawater samples. Thus, it will be interesting to use the SIP strategy applied in this work on more varied marine environments to elucidate the variability in those distinct microorganisms using DMSP and/or DMS primarily for carbon requirements.
Although wide-ranging conclusions have been drawn from the experiments conducted here on one coastal site, it should be noted that there (i) may be considerable variation in the types of microorganisms using DMSP and/or DMS in distinct environments and with different levels of available DMSP and (ii) was no gene/protein expression work complementing the DNA-SIP experiments, and thus however plausible the conclusions made here are, further work is required to establish that, e.g. the Amphritea dddD genes are expressed in the presence of DMSP. Nevertheless, this study has deepened our understanding of microbial DMSP degradation, and it provides a note of caution that analysis of environmental 'omics data for taxonomy and gene abundance alone can lead to significant misrepresentation of the importance and/or role of microbial groups in this process.

Sampling and quantification of DMSP and DMS in natural coastal seawater
North Sea surface seawater was collected from Great Yarmouth coast, UK, (52° 35′ 27.5928″ N; 1°44′ 17.9268″ E) on 27 th January 2018. To study the composition of the natural (T0) microbial community, 3 L of coastal seawater was filtered through 0.22-μm polycarbonate membrane filters (Millipore Corporation) using a vacuum pump. Filters were then stored at −80 °C for DNA extraction.
DMS concentrations in natural (T0) coastal seawater samples were quantified using a purge-and-trap gas chromatography (GC) system [75] using a flame photometric detector (Agilent 7890A GC fitted with a 7693 autosampler) and a HP-INNOWAX 30 m × 0.320 mm capillary column (Agilent Technologies J&W Scientific). DMSP content was quantified indirectly via alkaline lysis as previously described [76]. An eight-point calibration curve with DMS standards was made as in [76]. The detection limit for DMS in the headspace was 0.8 pmol.

DNA-stable isotope probing experiments
Synthesis of 13 C-DMSP 13 C-DMSP was synthesised in house from acrylic acid-13 C 3 and DMS (Sigma-Aldrich) as described in [31].

DNA-SIP experiments with 13 C-labelled DMSP
750 mL of coastal seawater were incubated in 2 L airtight bottles containing 100 μM of either 12 C-(control) or 13 C-labelled DMSP (Fig. 2a). To ensure the recovery of enough 13 C-labelled DNA for 16S and MG sequencing, six incubations with 12 C-and six with 13 C-DMSP were set up. All samples were incubated at 22 °C for 96 h (Fig. 2b) on a 12-h light (4000 lx) and 12-h dark cycle. After 96 h, when 77 μmol DMSP L −1 (231 μmol C L −1 ) were assimilated, the six biological replicates from 12 Cand 13 C-DMSP incubations were combined in pairs, respectively, and filtered through 0.22-μm polycarbonate membrane filters, resulting in triplicate samples used for subsequent DNA extraction.

Quantification of DMSP catabolism
DMSP, DMS and MeSH concentrations during SIP incubations with 12 C-and 13 C-DMSP were monitored by GC using the instrument and column cited above. To measure DMS and MeSH concentrations, 50 μl of headspace from 12 C-and 13 C-DMSP incubations were injected in the GC. To measure DMSP content, 1 mL seawater aliquots were taken from each replicate at selected timepoints and subsequently sparged with nitrogen to remove gaseous compounds. DMSP concentration was then measured via the addition of 10 M NaOH to 200 μL of the seawater samples in 2 mL gas-tight vials as described in [52]. Subsequent liberation of DMS was quantified by GC as above. An eight-point calibration curve of DMS and MeSH standards was used as in [52]. The detection limits for DMS and MeSH were 0.15 and 4 nmol, respectively.

Separation of labelled and unlabelled DNA
DNA from filters of natural (T0) seawater samples and incubations with 12 C-and 13 C-DMSP was extracted as in [77]. Four micrograms of DNA from 12 C-and 13 C-DMSP incubations was separated into heavy ( 13 C-labelled) and light ( 12 C-unlabelled) DNA by isopycnic ultracentrifugation as previously described [78]. DNA in each fraction was quantified using a Qubit dsDNA HS Assay kit (ThermoFisher Scientific) following the manufacturer's instructions. The density of each fraction was determined by refractometry using a Reichert AR200 refractometer (Reichert Analytical Instruments). Heavy and light DNA fractions from each sample were identified by plotting DNA abundance vs refractive index (as a proxy for density; Fig. 2c) and used for subsequent downstream analysis.

16S rRNA gene amplicon sequencing
To investigate the microbial diversity in samples from DNA-SIP experiments, we used DNA from three biological replicates from unenriched (T0) samples and labelled (heavy; H) and unlabelled (light; L) fractions from 12 Cand 13 C-DMSP incubations. 16S rRNA genes were amplified with primers 515F and 806R [79,80]. Triplicate PCR reactions for each sample were pooled before purification of PCR amplicons. Pyrosequencing was performed on Illumina MiSeq PE300 platform at Majorbio Bio-Pharm Technology Co. Ltd., (Shanghai, China), obtaining an average of 54936 quality-filtered reads per sample with an average length of 273 bp (Table S7). Sequences were analysed using Qiime [81], chimeras were excluded and OTUs assigned based on 97% similarity level. Taxonomic assignment was made using the SILVA database (Release 123 [82];) with 80% similarity threshold. Statistically significant differences in the RA of microbial groups from different samples were analysed using Student's t-test. The total microbial community from samples incubated with 12 C-(12C_T) and 13 C-DMSP (13C_T) prior to fractionation was reconstructed in silico, by adding the contributions from both light and heavy DNA, in proportion to the relative amount of DNA in the fractions.

Denaturing gel gradient electrophoresis (DGGE)
Bacterial 16S rRNA genes from DNA fractions from SIP incubations with 12 C-and 13 C-DMSP were amplified using primers 314F-GC and 518R [83]. Denaturing gel gradient electrophoresis (DGGE) was performed to visualise 16S rRNA gene profiles of the bacterial communities from 12 C-and 13 C-DMSP incubations following the protocol described by Green et al. [84].

Metagenomic analysis of DNA-SIP samples
Three biological replicates from natural (T0) seawater samples were subjected to metagenomic sequencing. As biological replicates from 13 C-heavy (13C_H) and 13 C-light (13C_L) fractions from incubations with 13 C-labelled DMSP showed highly similar 16S rRNA gene profiles in the DGGE analysis (Fig. S7), they were combined in equal proportions prior to metagenomic sequencing. Libraries from all samples were prepared by BGI (Shenzhen, China) for metagenomic sequencing without any amplification step. Shotgun sequencing was performed on an Illumina HiSeq X-Ten platform, with 2 × 150 bp paired-end reads. Metagenomic reads were quality-filtered and trimmed using SOAPnuke [85], obtaining a range of 13.2 to 14.1 Gb of high quality data per sample, with Q30 of each sample >90% (Table S8). Filtered reads were initially assembled by IDBA_UD [86] with different k-mer values for each sample. SOAPdenovo2 [87] was used to map reads from each sample to the assembled data, then combined with the N50 and mapping rates to choose the optimum k-mer and corresponding assembly results. Contigs shorter than 300 bp were excluded. Gene prediction was performed with MetaGeneMark [88]. Redundant sequences were removed using CD-Hit [89] at 95% identity and 90% coverage. Gene taxonomic annotation was determined by BLASTp (E ≤ 1e-5) against NCBI-nr databases using MEGAN software [90]. The RA of each gene was calculated as the percentage of its sequence coverage to the total sequence coverage (see Eq. 1). The sequence coverage was determined by the length of mapped reads relative to the reference sequence length.
where a i is the relative abundance of gene "i"; b i , the copy number of gene "i"; j, the total number of genes; X i , the total reads number aligned to gene "i"; and L i , the length of gene "i".
The sum of the RA of genes taxonomically assigned to a particular taxon was used to report the RA of that microbial group in metagenomic samples.
Profile  (Table S9) were used as training sequences to create the HMM profiles. HMM searches were performed against unique hits from seawater metagenomes using a cut-off value of E ≤ 1e−30 for DMSP cycling proteins and E ≤ 1e−5 for most DMS cycling proteins [3,91] (Table S1). Each potential sequence of interest retrieved from the analysis of seawater metagenomes was manually curated by BLASTp against the database of reference sequences (Table S9) and discarded if they had <40% amino acid identity to the corresponding ratified proteins. Given the large size of known DddD polypeptides [22,31], retrieved sequences <800 amino acids were excluded. Resultant sequences were used to construct a phylogenetic tree using MEGA v5.0 [92] with the Poisson model substitution model. Tree topology was checked using 100 bootstrap replicates. Only sequences whose top hits belonged to the same protein family and clustered with ratified sequences detailed in Table S9 were counted.
Finally, to determine the RA of bacteria and eukaryotes containing genes of interest, the number of unique hits of bacterial genes in seawater metagenomes was normalised to recA (a single copy marker gene possessed by the vast majority of bacteria), whereas unique hits of algal genes were normalised to ACTB (a phylogenetic marker gene for eukaryotes, encoding β-Actin protein) [3]. Hits of recA and ACTB were determined by HMM searches (E ≤ 1e−5) of sequences from a database obtained from RDP's FunGene [93] for recA and from Uniprotkb/ (1) swiss-prot for ATCB [94]. The RA of bacteria containing genes of interest in the total microbial community from incubations with 13 C-labelled DMSP prior fractionation (13C_T) was calculated as above.
Co-assembly of metagenomes from T0 samples and 13 C-heavy and 13 C-light fractions from incubations with 13 C-DMSP was performed to reconstruct metagenomeassembled genomes (MAGs) using MetaBAT2 [95]. Completeness and contamination of MAGs was assessed using CheckM [96], and only MAGs with <5% contamination were considered for further analysis. The phylogeny of MAGs was determined using PhyloPhlAn [97], followed by average nucleotide identity (ANI) calculations to further confirm taxonomy. The RA of each MAG in seawater metagenomes was calculated by the formula "MAG coverage / genome equivalent". The MAG coverage (copy number) of MAGs in each metagenome was evaluated using Anvi'o [98], by adding the coverage of each nucleotide in a MAG and dividing it by the MAG length. The genome equivalent of metagenomic samples (the total number of genomes per metagenome sample) was estimated by MicrobeCensus [99].

Isolation and characterisation of DMSP-degrading bacteria
Seawater from SIP incubations with 12 C-and 13 C-DMSP was serially diluted and plated onto marine broth agar, marine basal medium (MBM) supplemented with 10 mM mixed carbon source (glucose, succinate, sucrose, pyruvate and glycerol at 2 mM each) and MBM supplemented with 2 mM DMSP as sole carbon source [46]. After 48 h of incubation at 30 °C, sixty-six colonies with distinct morphologies were isolated and purified. To test the ability of the isolates to degrade DMSP, 300 μl of cultures grown on MBM with 10 mM mixed carbon source was transferred to 2-mL serum vials and supplemented with 0.5 mM DMSP. After 2-h incubation at 30 °C, DMS and MeSH liberated from DMSP was measured by GC. Media-only vials were set up as abiotic controls. Cellular protein content was estimated by Bradford assays (BioRad). Rates of DMS production are expressed as nmol·mg protein −1 ·h −1 .
To determine if isolates could use DMSP as sole carbon source for growth, they were grown in MBM with mixed carbon source, pelleted and washed twice with MBM medium without any carbon source. Cultures were then inoculated 1% (v/v) in triplicate into fresh MBM medium containing no carbon source, 10 mM mixed carbon source or 2 mM DMSP. Cultures were incubated at 30 °C for 10 days and growth was estimated by measuring cell density at OD 600 with a spectrophotometer. The statistical differences between the negative control (no carbon source) and cultures supplemented with DMSP or mixed carbon addition were determined by Student's t-test (P < 0.05 signifying the ability to grow on DMSP).
For identification, genomic DNA from DMSP-degrading strains was extracted using a Wizard genomic DNA purification kit (Promega) and 16S rRNA genes were amplified using primers 27F/1492R [100]. Purified PCR products were sequenced by Eurofins Genomics (Munich, Germany) and isolates were taxonomically identified using the Ezbiocloud website (http:// www. ezbio cloud. net/ ident ify). Representative strains from each genera/species were selected for further bioinformatics analysis to predict their molecular mechanisms to cleave DMSP. Publicly available genomes of their most closely related reference strains were screened for the presence of dmd/ddd homologous genes. Homologue sequences to ratified proteins involved in DMSP cycling (Table S9) were identified using local BLASTp, with thresholds set as E ≤ 1e−30, ≥50% amino acid identity and ≥70% coverage.
Quality of assembled genomes was analysed with CheckM [96], resulting in ≥99.1% completeness and <1.3% contamination. Genes were predicted using Glimmer 3.02 [102]. Identification of genes encoding homologous proteins to ratified DMSP cycling enzymes was performed using BLASTp, as described above.
Additional file 1: Figure S1. Microbial taxonomic profiles of seawater samples at the domain, phylum and class levels. Figure S2. Potential eukaryotic sources of DMSP in the natural (T0) coastal seawater. Figure  S3. Potential prokaryotic sources of DMSP in the natural (T0) coastal seawater. Figure S4. Maximum likelihood phylogenetic tree of DmdA proteins. Figure S5. Maximum likelihood phylogenetic tree of DddP proteins. Figure S6. Maximum likelihood phylogenetic trees of DddQ, DddL, DddK, DddY, DddW and DddX proteins. Figure S7. 16S rRNA gene profiles of seawater samples enriched with DMSP analysed by DGGE. Figure S8. Bacterial community profiles of coastal seawater samples analysed by 16S rRNA gene sequencing. Figure S9. Relative abundance and taxonomy of ancillary genes from the DMSP demethylation pathway in coastal seawater samples. Figure S10. Relative abundance and taxonomic affiliation of ancillary genes from the DMSP cleavage pathway in coastal seawater metagenomes. Figure S11. Relative abundance and taxonomic affiliation of DMS cycling genes in coastal seawater samples. Table S1. Relative abundance of genes encoding proteins involved in the cycling of DMSP, DMS and related compounds in metagenomes from seawater samples. Table S2. Relative abundance (RA) of main bacterial genera from seawater samples analysed by 16S rRNA gene amplicon (16S) and metagenomics (MG) sequencing. Table S3. Dominant bacterial genera in T0 seawater samples analysed by 16S rRNA gene amplicon sequencing. Table S4. Relative abundance (RA) of main bacterial orders from seawater samples analysed by 16S rRNA gene amplicon (16S) and metagenomics (MG) sequencing. Table S5. Metagenome-assembled genomes (MAGs) with homologous sequences to genes involved in DMSP cycling reconstructed from metagenomes from seawater samples. Table S6. Characteristics of bacterial strains with DMSP-degrading activity isolated from seawater incubations with DMSP. Table S7. 16S rRNA gene amplicon sequencing results for coastal seawater samples. Table S8. Statistics for metagenomic sequencing and assemblies. Table S9. Accession numbers of previously ratified enzymes involved in the cycling of DMSP and related compounds.