Mechanistic insight into DsyB/DSYB, key enzymes in marine dimethylsulfoniopropionate synthesis

Marine algae and bacteria produce eight billion tonnes of the organosulfur molecule dimethylsulfoniopropionate (DMSP) in Earth’s surface oceans every year. DMSP is an anti-stress compound and, once released into the environment, a major nutrient, signalling molecule and source of climate-active gases. The methionine transamination pathway for DMSP synthesis is used by most known DMSP-producing algae and bacteria. The S-directed S-adenosylmethionine-dependent methyltransferase (SAM-MT) 4-methylthio-2-hydroxybutyrate (MTHB) S-methyltransferase, encoded by the dsyB/DSYB gene, is the key enzyme of this pathway, generating S-adenosylhomocysteine (SAH) and 4-dimethylsulfonio-2-hydroxybutyrate (DMSHB). dsyB/DSYB, present in most DMSP-producing bacteria and haptophyte and dinoagellate algae with the highest known DMSP concentrations, is shown to be far more abundant and transcribed in marine environments than any other known DMSP synthesis pathway S-methyltransferase gene. Furthermore, we demonstrate in vitro activity of the bacterial DsyB enzyme from Nisaea denitricans, and provide its crystal structure in complex with SAM and SAH-MTHB, which together provide the rst mechanistic insights into a DMSP synthesis enzyme. Structural and mutational analyses imply that DsyB adopts a novel mechanism, distinct from any previously reported SAM-MT, in which the DsyB residue Tyr142 activates the sulfur atom of MTHB for nucleophilic attack on the SAM methyl group. Sequence analysis suggests that this mechanism is common to all bacterial DsyB enzymes and also, importantly, eukaryotic DSYB enzymes from e.g., algae that are the major DMSP producers in Earth’s surface oceans. and in marine algae containing DSYB. Our results provide novel insights into DMSP synthesis, shedding light on the global sulfur cycling. This study also offers a better understanding of the catalysis of SAM-MTs, especially the S-directed SAM-MTs.


Introduction
Approximately eight billion tonnes of the compatible solute dimethylsulfoniopropionate (DMSP) are produced annually in Earth's surface waters (1), constituting up to 10% of surface ocean organic carbon (2). Many marine algae, bacteria, corals and some plants produce DMSP (3) for its proposed functions as e.g., a compatible solute (4), grazing deterrent (5), antioxidant (6) and protectant against hydrostatic pressure (7). Furthermore, DMSP is a major nutrient for marine microorganisms, and a precursor for climate-active volatiles such as dimethyl sul de (DMS) (3,8,9). DMSP was thought to be mainly produced by marine algae in Earth's surface oceans, but recent studies suggest that bacteria, particularly in marine sediment, are also important DMSP producers (3,(10)(11)(12).
Three pathways for DMSP synthesis have been proposed based on the identi cation of intermediates and enzyme activities in various model DMSP producers: a methylation pathway in some plants and bacteria, a transamination pathway in algae and bacteria, and a decarboxylation pathway in one dino agellate (10,(12)(13)(14)(15) (Fig. 1). Of these ( Fig. 1), the transamination pathway is thought to be the most important in marine environments as it functions in the majority of DMSP producing algae (spanning dino agellates, haptophytes and diatoms) and bacteria (10,12,14). The committed enzyme of the transamination pathway ( Fig. 1) is an S-adenosylmethionine (SAM)-dependent 4-methylthio-2hydroxybutyrate (MTHB) S-methyltransferase that yields 4-dimethylsulfonio-2-hydroxybutyrate (DMSHB) (13,14,16). Recently, the key MTHB S-methyltransferase enzyme 'DsyB' (Fig. 1), was identi ed in many DMSP-producing marine Alphaproteobacteria (10). Functional DsyB homologues (~33% amino acid identity), termed DSYB, exist in most DMSP-producing dino agellates, haptophytes, corals and ~ 20% of diatoms (12). This is consistent with the detection of DMSHB and its oxidative decarboxylation to DMSP in some prymnesiophytes, diatoms and prasinophytes (14). dsyBDSYB genes are robust indicators of an organism's potential to produce DMSP (10,12). The centric diatom Thalassiosira pseudonana, which lacks DSYB, contains an isoform MTHB S-methyltransferase enzyme termed TpMMT, but this enzyme has not been studied in any other organism (17). Published and new analysis here (see below) shows that DsyB/DSYB are far more abundant in known DMSP-producing microorganisms (phytoplankton and bacteria) and in marine environmental metagenome and metatranscriptome datasets than other identi ed DMSP synthesis genes (10)(11)(12). Furthermore, acquisition of dsyB is su cient to enable some organisms to produce DMSP (10). Together these data suggest that transamination using DsyB/DSYB enzymes is the most important marine DMSP synthesis pathway.
DSYB and DsyB belong to the SAM-dependent methyltransferase (SAM-MT) family (10,12). SAM is the second most widely used enzyme substrate after ATP and is involved in many important biological processes (18). SAM-MTs are categorized based on the methyl accepting atom, usually O, N, C or S (19). The majority (54%) of known SAM-MTs are O-directed, whereas only 3% are S-directed, and up to now, all were in plants and mammals (19). SAM-MTs, which catalyse transmethylation via S N 2 nucleophilic substitution (20), have evolved three distinct mechanisms: proximity and desolvation (PD), general acid/base-mediated catalysis, and a metal-dependent mechanism (19). To our knowledge, DsyB was the rst reported S-directed SAM-MT identi ed in bacteria, and its mechanism is unknown.
Here, we investigate Nisaea denitri cans DR41_21, a marine alphaproteobacterium (21) predicted to produce DMSP, and characterize its DsyB enzyme. With no reported protein crystallographic or mechanistic studies of DsyB/DSYB or any DMSP synthesis enzyme, X-ray crystallography and mutational analyses were employed to establish the structure and mechanistic properties of a functional DsyB enzyme and its interaction with SAM and MTHB substrates. Our results provide the rst structural and mechanistic data for any DMSP synthesis enzyme. Furthermore, sequence alignment and structural analysis are used to elucidate mechanistic similarities between bacterial DsyB and algal DSYB enzymes.
We propose the mechanism for DsyB/DSYB, which is the rst for any bacterial/algal S-methyltransferase, and is distinct from all previously reported SAM-MT enzymes. We also probe marine microorganisms, metagenomes and metatranscriptomes for the presence of DsyB/DSYB and other key SAM-MT in DMSP synthesis pathways to investigate the importance of these proteins in the global oceans. Thus, this study identi es DsyB/DSYB as a new class of SAM-MT enzymes with a central role in driving marine microbial cycling of organosulfur.

Results And Discussion
Nisaea denitri cans DR41_21 is a DMSP-producing bacterium Isolated from coastal Mediterranean Sea surface waters, Nisaea denitri cans DR41_21 (DSM 18348) is a marine alphaproteobacterium of the Rhodospirillaceae family which was not previously known to produce DMSP (21). N. denitri cans contains a DsyB homologue, 337 amino acid residues in length with 59% identity to Labrenzia aggregata DsyB and is thus predicted to make DMSP (10). Indeed, cloned N. denitri cans dsyB conferred onto Rhizobium, a heterologous host that lacks DsyB and makes no DMSP, MTHB S-methyltransferase activity. Furthermore, N. denitri cans dsyB fully restored DMSP production (105 ± 3.4 pmol DMSP µg protein -1 ) to an L. aggregata dsyBdeletion mutant, which produces no DMSP without it (wild type L. aggregata produces 99.8 ± 1.2 pmol DMSP µg protein -1 (10)). N. denitri cans itself produced DMSP when grown in the absence of methylated sulfur compounds, and production and dsyB transcription was enhanced by increased salinity and by nitrogen starvation (Fig 2A-B). This work further con rms that the presence of dsyB and its transcription in a bacterium reports on the ability of the strain to produce DMSP and the levels it makes, respectively.
In vitro characterization of N. denitri cans DsyB As shown above and in (10), DsyB has MTHB S-methyltransferase activity when expressed in alphaproteobacterial hosts. However, L. aggregata DsyB and Chrysochromulina DSYB enzymes (12), had no detectable MTHB S-methyltransferase activity when expressed in E. coli. The same was generally true of the recombinant N. denitri cans DsyB enzyme puri ed from E. coli, although variable MTHB Smethyltransferase activity was observed in vitro by hydrophilic interaction chromatography mass spectrometry (HILIC-MS), see below. The reason for this lack of activity upon isolation is unknown; one possibility is that the enzyme requires an essential co-factor or modi cation that was provided by an algal or alphaproteobacterial host, but not by E. coli (12). This hypothesis was initially supported by the fact that addition of heat denatured cell lysate fractions (from a PD10 desalting column) liberated from the L. aggregata dsyB deletion mutant, which produces no DMSP, recovered N. denitri cans DsyB MTHB S-methyltransferase activity (Fig. S1). Similar complementation was shown with the addition of heatkilled Prymnesium parvum extract to DSYB in (12). The activated DsyB protein was shown to have K m and V max values of 0.14 mM and 365 nmol min -1 mg protein -1 , respectively, for MTHB ( Fig. 2C), which were similar to those previously established for P. parvum DSYB (0.09 mM and 294 nmol min -1 mg protein -1 ) in (12). The activated DsyB had a K m of 0.16 mM and V max 368.9 nmol min -1 mg protein -1 for the co-substrate SAM (Fig. 2D), which were also similar to those obtained with P. parvum DSYB (0.06 mM and 303 nmol min -1 mg protein -1 ) in (12).
Liquid chromatography with mass spectrometry (LC-MS) and/or native mass spectrometry was used in an attempt to identify the activation factor in the L. aggregata dsyBextract (Fig. S2). A prominent peak at 37,084 Da was observed in the LC-MS spectrum for both the as-isolated and activated samples, which corresponds to DsyB with its N-terminal Met residue cleaved (commonly observed for proteins overexpressed in E. coli) (22). A lower intensity peak at +131 Da, corresponding to the full-length protein (37,215 Da), was also observed in the as-isolated sample, indicating that the Met cleavage was not complete (Fig. S2A). There was an additional minor peak at +269 Da of unknown origin in the activated sample (Fig. S2B). Under non-denaturing conditions, both monomeric and dimeric forms of DsyB were detected in the as-isolated sample, a feature commonly observed in non-denaturing mass spectra of solution dimers (23)(24)(25). In the monomeric region, the main protein peaks (due to cleaved and non-cleaved proteins) were again observed, but, in addition, a number of adduct species were present in the spectrum. Two of these, at +36 and +98 Da, correspond to chloride and (most likely) phosphate adducts. An additional adduct at +63 Da was also observed, possibly due to metal ion binding. In general, the spectrum of the dimeric form of DsyB was less well-resolved, but the main protein peak (at 74,168 Da) along with chloride and possible metal ion adducts were all detected (Fig. S3). The non-denaturing mass spectrum of the activated monomeric DsyB (Fig. S2B) revealed a number of adducts, including those most likely due to chloride, phosphate and metal ion binding (all common within the as-isolated DsyB), along with an additional adduct at +122 Da (and at +244 Da), which is likely due to Tris buffer. Thus, we have no data to support there being a cofactor or modi cation of DsyB caused by the addition of the heat-killed L. aggregata dsyBextract to as-isolated DsyB, and further work is required to understand the variable nature of DsyB activity (see below).
The association of metal ions with DsyB was investigated further. Inductively coupled plasma mass spectrometry (ICP-MS) analysis revealed variable metal ion content with some preparations of as-isolated DsyB containing up to 0.85 Cu per protein, with other metals such as Ni (up to 0.5 per protein), Zn (0.4) and Fe (0.14) also detected. However, there was no correlation between metal ion content and activity of as-isolated samples.
Despite mostly lacking consistent in vitro MTHB S-methyltransferase activity (see below), native MS showed that the as-isolated N. denitri cans DsyB enzyme binds to SAM. The deconvoluted mass spectrum of a DsyB sample under non-denaturing conditions and containing 25 equivalents of SAM contained a peak in the DsyB dimer region at +870 Da (predicted mass of a (DsyB) 2 -(SAM) 2 is 74,966 Da), indicative of a (DsyB) 2 -(SAM-Cl) 2 adduct, that was not observed in the absence of SAM (Fig. S4A).
Evidence for SAM binding was also apparent in the monomer region, though the presence of chloride adducts spreads out the intensity in this region (Fig. S4B). Evidence for an MTHB-bound form of DsyB was also observed (although, again, the presence of chloride adducts spreads out intensity, Fig. S4B). Although care is needed in interpreting intensities of peaks in the non-denaturing mass spectrum, the low intensity of the SAM-and MTHB-bound forms of DsyB suggests relatively low a nities when these substrates are present individually.
Small molecule HILIC-MS analysis of reactions following addition of MTHB to DsyB-SAM resulted in the detection of substrates SAM and MTHB, and products DMSHB and SAH (Fig. 3). Non-denaturing mass spectrometry of similarly generated samples resulted in the loss of SAM-and MTHB-bound forms of DsyB (Fig. S4B). Together, the data are consistent with DsyB being a SAM-dependent MTHB Smethyltransferase.
Overall structure of DsyB To analyse the catalytic mechanism of DsyB, we solved the crystal structures of complexes of DsyB with SAM and with SAH-MTHB. The crystal structure of the DsyB-SAM complex was determined by the singlewavelength anomalous dispersion (SAD) method using a selenomethionine derivative (Se derivative) ( Table S1).
Crystals of the DsyB-SAM complex belonged to the P2 1 2 1 2 1 space group, with four molecules arranged as a tetramer in the asymmetric unit. Each DsyB molecule contains two domains, an N-terminal domain (N-domain, Met1-Ala125) and a C-terminal domain (C-domain, Thr126-Glu337), which can be seen binding to the SAM molecule (Fig. 4A). The DsyB C-domain contains seven β-strands surrounded by six αhelices, which together adopt the typical Rossmann-like α/β fold of Class I SAM-MTs (Fig. 4A). Structural analysis showed that two DsyB monomers are tightly intertwined, mainly through interactions of residues from the N-domains of two adjacent monomers (Fig. 4A). Analysis of DsyB using the PISA server (http://www.ebi.ac.uk/msd-srv/prot_int/pistart.html) predicted the DsyB dimer to be stable in solution. Indeed, gel ltration analysis indicated that DsyB is likely a dimer in solution (Fig. 4B), consistent with the non-denaturing mass spectrometry data above (Fig. S3, S4A). These results indicate that DsyB functions as a dimer in the same way as other SAM-MTs, whose N-domains are also responsible for dimerization (19,26,27). ICP-MS and LC-MS analyses showed that the as-isolated DsyB contained variable metals. However, in the crystal structure of DsyB-SAM complex, no explicit electron density associated with metals was observed, suggesting that the binding site of metals may not be speci c in DsyB.
The crystals of the DsyB-SAH-MTHB complex belong to the P2 1 space group (Table S1) and the resulting structure has a similar overall structure to that of the DsyB-SAM complex (Fig. 4C). Interestingly, in this case the MTHB co-substrate molecule is located between the C-domain and the N-domain of one DsyB monomer (Fig. 4C).

The conformational change of DsyB in binding SAM
During structural re nement of the DsyB-SAM complex, we found that three monomers (chain A, B and C) of the asymmetric unit contained SAM molecules. The structures of these three monomers are similar, with root mean square deviations (RMSDs) of ~0.5 Å between any two monomers. The chain D of the DsyB-SAM complex is not bound to a SAM molecule. The conformation of chain D is different to the other monomers bound to SAM, with a RMSD of ~2.4 Å between chain D and chain A. Moreover, residues Asp123 to Tyr143 in chain D exhibited weak electron density, suggesting that this region is highly exible. By superposing molecules of chain A and chain D, we observed that the N-domains of chain A and chain D are almost completely aligned, whereas the C-domain rotates ~10° as a rigid body (Fig. 4D). These structural differences indicate that DsyB possesses two conformations: an "open" form and a "closed" form. Although DsyB can bind SAM and MTHB individually (Fig. S4B), the binding of SAM triggers the conformational change of DsyB from the "open" form to the "closed" form, shrinking the cavity between the N-domain and the C-domain of DsyB and possibly promoting the subsequent binding of MTHB.
The structure of the DsyB-SAM complex is similar to that of the Streptosporangium sibiricum SibL protein (PDB code: 4U1Q), a C-directed Class I SAM-MT, with an RMSD of ~1.3 Å between these two structures. SAM binding also triggers the conformational change of SibL from an "open" form to a "closed" form to complete the formation of a binding site for its methyl acceptor 3-hydroxykynurenine (26). Similar conformational changes have also been observed in other C/O-directed SAM-MTs, despite their low sequence identities (19,27).

Binding sites of SAM and MTHB
The SAM molecule within the DsyB-SAM complex (Fig. 4A) is bound mainly by hydrogen bonds with residues in the DsyB C-terminal domain. DsyB residues Asp223 and Ala224 participate in binding the adenine ring of SAM; Asp196 forms hydrogen bonds with the ribose moiety of SAM; and Ser150, Gly173 and Ser239 interacts with the terminal amino acid moiety of SAM (Fig. 5A). A similar binding mode was observed between the same DsyB C-terminal residues for SAH in the DsyB-SAH-MTHB complex.
In the DsyB-SAH-MTHB complex, the electron density of the MTHB molecule is relatively poor and the distance between SAH and MTHB is more than 6 Å, which is too far to enable the methyl transfer reaction. We speculate that the position of MTHB observed in the structure is not the exact location of MTHB when the reaction occurs under physiological conditions, and that the observed structure represents a state where the MTHB molecule has not completely entered into the active site. Nevertheless, the location of MTHB clearly implies its initial binding site (Fig. 5B), and several residues likely involved in the binding of MTHB were identi ed, including Tyr97 and Gln101 from the N-domain and Tyr129, Tyr142, Gln146 and His291 from the C-domain of DsyB (Fig. 5C).
To determine the importance of the Tyr97, Gln101, Tyr129, Tyr142, Gln146 and His291 in DsyB binding to MTHB, we performed site-directed substitutions of these residues and assayed the in vivo MTHB Smethyltransferase activity of the resultant variant DsyB derivatives in R. leguminosarum. In vivo assays were performed in Rhizobium because of the sensitivity of in vitro assays, see above. Site-directed mutations of Tyr97, Gln101, Tyr129, Tyr142, Gln146 or His291 to alanine severely decreased the enzymatic activity of DsyB (Fig. 5D), indicating the potentially important roles of these residues in binding MTHB. In particular, activity of the Tyr97Ala and Tyr142Ala variants was completely abolished (Fig. 5D), suggesting that these two residues play key roles in the methyl transfer reaction.
Based on this mutational analysis, residues Tyr97 and Tyr142 of DsyB are the best candidates for catalytic residues. In the structure of the DsyB-SAH-MTHB complex, the distance between Tyr97 and MTHB is 3.9 Å, which is similar to the distance between Tyr142 and MTHB (3.7 Å). However, as discussed above, the DsyB-SAH-MTHB complex structure observed is likely an intermediate state, since the MTHB molecule needs to be closer to SAM to attack the SAM methyl group. Because the phenolic sidechain of Tyr97 is pointing in the opposite direction to SAM, and that of Tyr142 is located in between MTHB and SAM ( Fig. 5C), movement of MTHB from its initial binding site (observed in the structure) towards SAM would lengthen the distance between Tyr97 and MTHB, and shorten the distance between Tyr142 and MTHB. Therefore, Tyr142 is the most probable catalytic residue of DsyB. Interactions with Gln101, Tyr129 and Gln146 might be important in lowering the pK a of the phenolic proton of Tyr142, which may be essential for the activation of MTHB. The residue Tyr97 may play a vital role in MTHB binding based on mutational and structural analyses.
Three distinct catalytic mechanisms have been reported for SAM-MTs, including the PD mechanism, the general acid/base-mediated mechanism and the metal-dependent mechanism (19). Structural and biochemical analyses indicate that the activity of DsyB is not metal-dependent, but is catalytic residuedependent. Furthermore, MTHB, the substrate of DsyB, does not contain a thiol group, and so activation of MTHB for attack on SAM cannot occur via deprotonation. This implies that DsyB does not follow the typical acid/base mechanism where the substrate is activated by deprotonation. Thus, the catalytic mechanism of DsyB likely represents a novel catalytic strategy for SAM-MTs. Rather than acting as a catalytic base, the catalytic residue Tyr142 most likely activates MTHB by making the sulfur atom more nucleophilic, which is essential for the subsequent attack of MTHB on SAM.
The catalytic mechanism of DsyB Based on our structural and biochemical results, we propose DsyB rst binds a SAM molecule to generate a conformational change from "open" state to "closed" state, which promotes the binding of MTHB (Fig.  6A). When an MTHB molecule enters the active site, the deprotonated phenolic group of Tyr142 in DsyB activates the sulfur atom of MTHB for nucleophilic attack on the methyl group of SAM (Fig. 6B). Subsequently, the generated DMSHB and SAH are released, and DsyB can re-bind a SAM molecule from the intracellular environment in preparation for the next reaction.
Currently DsyB is the sole known bacterial SAM-dependent S-methyltransferase with a crystal structure, although there are S-directed SAM-MT structures available from mammal and plant enzymes (19). For example, thiopurine S-methyltransferase (TPMT) is a murine S-directed SAM-MT that methylates 6mercaptopurine (28). Unlike DsyB, TPMT does not contain an N-terminal domain likely involved in dimerization, as its N-terminus only constitutes 40 residues (28). Although Arg147 and Arg221 are possible participants in 6-mercaptopurine deprotonation, the modest decrease in the enzymatic activities of the corresponding mutants suggests that TPMT may possess the PD strategy for catalysis (28). Catharanthus roseus CrSMT1 is another S-directed SAM-MT that methylates a broad range of substrates including benzene thiol and furfuryl thiol (29). Homology modelling suggested that CrSMT1 contains an N-domain for dimerization (29), which is similar to DsyB. However, CrSMT1 is thought to use a histidine residue as a general base to deprotonate the thiol group of the substrate (29). Thus, although the Sdirected SAM-MTs only constitute a small portion of the reported SAM-MTs (19), their catalytic mechanisms appear diverse. Moreover, the catalytic mechanism of DsyB proposed here represents a novel catalytic strategy of SAM-MTs.

Universality of the catalytic mechanism of DsyB
The majority of bacteria containing DsyB are Rhodobacterales, which are abundant in marine environments, but this enzyme is also found in some Rhizobiales and Rhodospirillales (including N. dentri cans) (10,30). To investigate the ubiquity of the DsyB catalytic mechanism, we performed sequence alignment of DsyB proteins from different Rhodobacterales, Rhizobiales and Rhodospirillales bacteria (Fig. S5). Most residues involved in initial MTHB binding (Tyr97, Gln101, Tyr129, Gln146 and His291) and SAM binding (Ser150, Gly173, Asp196, Asp223, Ala224 and Ser239), and the catalytic residue (Tyr142) are highly conserved in DsyB proteins from these marine bacteria, indicating that mechanistic insight gained here for N. denitri cans DsyB has universal signi cance in bacteria containing DsyB.
Eukaryotic DSYB, which may originate from bacterial DsyB, is a key enzyme for DMSP synthesis in many phytoplankton, such as marine haptophytes, dino agellates and some diatoms (12). DSYB shares 33% sequence identity to DsyB, and we predicted the structure of DSYB from Chrysochromulina tobin CCMP291 by homologous modelling using SWISS-MODEL (https://swissmodel.expasy.org/) (31). Structural alignment of DSYB and DsyB indicated that residues involved in binding MTHB and the catalytic residue are perfectly superposed (Fig. S6). Moreover, sequence alignment of DsyB and DSYB from different marine algae showed that residues which play important roles in DsyB are highly conserved in different DSYB proteins (Fig. S7), suggesting that DSYB proteins adopt a similar catalytic mechanism to DsyB.
DsyB/DSYB are the most abundant and transcribed S-methyltransferase enzymes of known DMSP synthesis pathways in marine microorganisms and environments Having the identity of the key S-methyltransferases in diverse DMSP synthesis pathways (DsyB/DSYB and TpMMT in the bacterial and algal transamination pathway, and MmtN and BurB in bacterial methylation pathways, Fig. 1), we carefully analysed their presence in marine microorganisms and their abundance and transcript levels in published global 'omics datasets to quantify the potential environmental importance of these pathways.
This hierarchy of DMSP synthesis gene abundance in bacteria (DsyB > MmtN > BurB) was mirrored in marine environmental data. In the Tara Oceans prokaryotic database, both dsyB and mmtN were found throughout the water column (Fig. 7A), but no close homologues of BurB (e-value <1e-40) were detected. This is consistent with BurB-mediated DMSP production in Burkholderia spp., possibly for toxin production, not being an important process in marine systems. DsyB was signi cantly more abundant than mmtN in both the metagenomic (Kruskal-Wallis Χ 2 = 83.781, p<.001) and metatranscriptomic (Kruskal-Wallis Χ 2 = 33.64, p<.001) Tara Oceans datasets (Fig. 7A). Additionally, dsyB was found at 172 and 153 sampling sites (treating each depth as a separate site) in the metagenomes and metatranscriptomes, respectively, whereas mmtN was found at only 74 and 63 sites, respectively. Given this, our analysis of median abundance overestimates the contribution of mmtN to DMSP production in the global ocean. We therefore determined the relative abundance of dsyB:mmtN across depths at each sampling site in both metagenomes and metatranscriptomes (Fig. S8). DsyB was more abundant at almost all sites in the metagenomes, and was more highly expressed across most, though there were a number of locations in the South Atlantic and South Paci c where mmtN was predominant (Fig. S8). Taxonomic examination of both dsyB and mmtN sequences in the Tara database (Fig. 7B) showed that both genes were exclusively from Alphaproteobacteria, primarily within the Orders Rhodobacterales and Rhizobiales for dsyB, and the genus Thalassospira for mmtN (Fig. 7B). These data highlight DsyB as the most abundant, transcribed and, likely, important of the known bacterial DMSP synthesis enzymes in marine waters, which likely play a signi cant role in the global production of DMSP.
Moving to eukaryotic DMSP synthesis, we carefully analysed available transcriptome data from marine eukaryotes in the marine microbial eukaryote transcriptome sequencing project (MMETSP) (35). The TpMMT MTHB S-methyltransferase has only been characterised in the centric diatom Thalassiosira pseudonana CCMP1335, and close homologues (~70% protein identity) with the same singular domain structure only exist in 17/82 diatom transcriptomes (7 of which also contain DSYB), and no other phytoplankton (Table S3 and S4) (36-51). The next most homologous TpMMT-like proteins, present in e.g. Thalassiosira oceanica (EJK59074) and Fistulifera solaris (GAX25165) are more diverse (the methyltransferase domain being <50% identical to TpMMT), contain extra protein domains and, thus, are much larger proteins whose function is unknown. These TpMMT-like proteins cannot be considered as functional MTHB S-methyltransferase enzymes and were omitted from this study. In contrast, DSYB is found in the transcriptomes of 47/61 dino agellates and 24/30 haptophytes, organisms known to produce the highest levels of DMSP per cell (>50mM) (37,44). Furthermore, 15/82 diatom transcriptomes, typically known to produce lower cellular DMSP levels (generally < 50 mM) (37), and some Ochrophyta, Cnidaria and Cilophora transcriptomes also contained DSYB. These data show DSYB to be the most abundant and widespread DMSP synthesis enzyme known in eukaryotic DMSP-producing organisms.
Within the eukaryotic Marine Atlas of Tara Ocean Unigenes (MATOU) we found both DSYB and TpMMT within epipelagic (surface, SRF and deep chlorophyll maximum, DCM) waters. Initial examination showed DSYB to be more abundant in ≤ 3 µm fractions than in larger fractions (Fig S9). Data from these smaller ≤ 3 fractions that likely contain picoeukaryotes (CoP) were considered together. Likewise, fractions with a minimum lter size of ≥ 3 µm that likely exclude picoeukaryotes (ExP) were also considered together.

Conclusion
DMSP is an abundant and ecologically important organosulfur compound. DsyB/DSYB enzymes catalyse the committed S-methylation of MTHB to generate DMSHB, which is the key step of the transamination pathway for DMSP synthesis in most bacteria and algae (10)(11)(12). Furthermore, DsyB/DSYB enzymes are present in the most prodigious DMSP-producing haptophyte and dino agellate phytoplankton, and represent the most abundant and transcribed S-methylase genes of known DMSP synthesis pathways in marine waters. In this study, we solved the rst crystal structures of bacterial DsyB-SAM and DsyB-SAH-MTHB complexes and demonstrated conversion of SAM and MTHB into SAH and DMSHB. Based on structural and mutational analyses, the catalytic mechanism of DsyB is proposed, which represents a novel catalytic strategy of SAM-MTs, and has universal signi cance in bacteria containing DsyB, and in marine algae containing DSYB. Our results provide novel insights into DMSP synthesis, shedding light on the global sulfur cycling. This study also offers a better understanding of the catalysis of SAM-MTs, especially the S-directed SAM-MTs.

General in vivo and in vitro genetic manipulations
Plasmids (Table S5) were transferred to E. coli by transformation, and R. leguminosarum J391 or L. aggregata J571 by conjugation, in a triparental mating using the helper plasmid pRK2013 (55). Routine restriction digestions and ligations for cloning were performed essentially as in Carrion et al.,(56). The oligonucleotide primers used for molecular cloning were synthesized by Euro ns Genomics and are detailed in Table S6. Sequencing of plasmids and PCR products was performed by Euro ns Genomics.
The dsyB gene was ampli ed from N. denitri cans genomic DNA and cloned into the pET22b (Novagen, America) for the expression of DsyB with a C-terminal His-tag or into pLMB509, for expression in Rhizobium and Labrenzia (57). Amino acid substitution mutations in DsyB were generated using QuikChange® mutagenesis kit (Agilent Technologies) and the primers in Table S6. All site directed mutant (SDM) variant plasmids were veri ed by DNA sequencing.

Reverse transcription quantitative polymerase chain reaction (RT-qPCR)
RNA was isolated from 100 ml N. denitri cans cultures using RNeasy Mini Kit (Qiagen) according to the manufacturer's protocol with some modi cations. On-Column DNase digestion was performed with RNase-Free DNase Set (Qiagen). Reverse transcription of 1 µg of DNA-free RNA per sample was done using the QuantiTect Reverse Transcription Kit (Qiagen). PCR on RNA and cDNA samples con rmed that RNA samples were DNA-free.
Primers for RT-qPCR for N. dentri cans dsyB and housekeeping genes recA and gyrB, were designed using Primer5 (58) ( Table S6). Primer e ciencies were all 90-110% and within recommended limits. qPCR was performed on a StepOnePlus instrument (Applied Biosystems). Quanti cation was performed using the SYBR ® Green JumpStart™ Taq ReadyMix™ (Sigma-Aldrich) following the manufacturer's instructions. Reactions (20 µl) contained 2 µl cDNA, 0.8 µl of 10 µM primers and with an annealing temperature of 55 °C. For each condition and gene, the cycle threshold (Ct) values of triplicate technical and biological replicates were averaged. Relative expression levels were determined with e ciency correction (59). dsyB transcription was displayed as normalized fold change to the standard condition.

DsyB enzyme assays in vivo
To measure MTHB S-methyltransferase activity from pLMB509 clones (and SDM derivatives) in R. leguminosarum J391, cultures were grown (in triplicate) overnight in TY complete medium; 1 ml of culture was centrifuged at 20,000 g for 2 min, resuspended in the same volume of Y medium and then diluted 1:100 into 5 ml Y with 10 mM taurine (to induce expression, Sigma-Aldrich, T0625), 0.5 mM DL-MTHB (Sigma-Aldrich, 55875) and gentamicin, and incubated at 28 °C for 60 h before sampling for gas chromatography (GC) analysis to determine the amount of DMSP product (see below).
To measure MTHB S-methyltransferase activity from pLMB509 clones expressing the dsyB gene in the L. aggregata dsyBmutant strain J571, cultures were grown (in triplicate) overnight in YTSS complete medium. Following incubation, 1 ml of culture was then centrifuged at 20,000 g for 2 min, resuspended in the same volume of MBM medium and then diluted 1:50 into 5 ml MBM with 10 mM taurine (to induce expression; Sigma-Aldrich), rifampicin and gentamicin, and incubated at 30 °C for 24 h. DL-MTHB (0.5 mM) was added as substrate to the cultures and these were incubated for 4 h at 30 °C before sampling for GC and protein estimation by the Bradford assay.
To measure DMSHB/DMSP in Rhizobium or Labrenzia assay mixtures, 200 µl of culture was added to a 2 ml glass serum vial then 100 µl 10 M NaOH was added and vials were crimped with PTFE/rubber crimp caps immediately. Vials were incubated at 80°C for 10 minutes (to capture DMSHB as well as DMSP) and then for 24 h at room temperature in the dark before being monitored by GC assay. All GC assays involved measurement of headspace DMS using a ame photometric detector (Agilent 7890 A GC tted with a 7693 autosampler) and an HP-INNOWax 30 m × 0.320 mm capillary column (Agilent Technologies J&W Scienti c). Calibration curves were produced by alkaline lysis of DMSP standards in water. The detection limit for headspace DMS from DMSP was 0.015 nmol and from DMSHB was 0.3 nmol. DsyB activity is expressed as pmol DMSHB/DMSP mg protein −1 min −1 . Protein concentrations were determined using the Bradford method (BioRad). Control assays of Rhizobium or Labrenzia J571 containing pLMB509 were carried out, as above, and gave no detectable DsyB activity.
Protein expression and puri cation E. coli BL21 (DE3) containing pET22b::dsyB clones were cultured aerobically at 37°C in LB medium containing ampicillin. At mid-exponential growth (OD 600 0.5-0.7) 0.5 mM isopropyl β-D-1thiogalactopyranoside (IPTG) was added and the cells incubated at 20°C for 16 h. Cells were harvested by centrifugation (20 min, 7500 × g, 4°C), washed and resuspended in 25 mM Tris-HCl, pH 8.0, 150 mM NaCl. Cell lysis was performed by three passes through a French Press (16,000 psi), cell debris and unbroken cells were pelleted by centrifugation (30 min, 5,500 × g, 4°C) and the supernatant was recovered and subjected to centrifugation (60 min, 185,000 × g, 4°C) to pellet the membrane fraction. Soluble cell lysate was applied to a slurry of Ni-NTA resin (Qiagen) at a 3:1 ratio for 90 minutes with shaking at 4°C.
The lysate/slurry mix was loaded into Econo-Pac polypropylene columns, washed with 50 mM Tris-HCl, 250 mM NaCl, 20 mM imidazole, pH 8.0 and DsyB was eluted in 5 x 1 ml fractions using 50 mM Tris-HCl, 250 mM NaCl, 250 mM imidazole, pH 8.0. Fractions containing DsyB were concentrated and buffer exchanged into 2 ml of 50 mM Tris-HCl, 100 mM NaCl, pH 8.0 and applied to a Superdex 200 10/300GL preparative grade gel ltration column (Cytiva). Puri cation of DsyB and was monitored by electronic absorbance spectroscopy and SDS-PAGE; for the latter, proteins were visualized by Coomassie stain.
Aliquots of the puri ed protein were ash-frozen in liquid nitrogen and stored at −80°C until required.

DsyB enzyme assays in vitro
Labrenzia aggregata dsyB -J571 (10) was grown to late exponential phase in MBM. Cell lysates were prepared by centrifuging 100 ml of culture for 10 min at 2,500 g. The pellet was washed and resuspended with 20 ml 20 mM HEPES, 150 mM NaCl, pH 7.5 before cell lysis via French press (16,000 psi). The cell lysate was heat-treated at 80°C for 10 min to denature proteins, then applied to a 10 ml PD10 column, eluted over 10 ml, and collected in 1ml fractions. DsyB activity was monitored by performing in vitro enzyme assays with 50 µl of the separate heat-killed extract fractions, 1 mM SAM (Sigma-Aldrich), 1 mM DL-MTHB and 5 mM DsyB or no protein (control) in 400 µl total volume. Experiments were done as above with puri ed DsyB without addition of heat-killed extracts, but these gave no activity (data not shown).
Reactions were incubated at 28°C for 30 mins and then 800 µl of activated charcoal (38 mg ml −1 in 0.1 M acetic acid) was added to the samples and mixed. Samples were centrifuged for 15 mins at 14,000 g and the supernatant was retained. For GC analysis, 200 µl of the supernatant was added to a 2 ml vials with 100 µl 10 M NaOH and vials were immediately crimped. Crimped vials were then heated to 80 °C for 10 minutes (to release DMS) and incubated at 22 °C for 16 h in the dark. These samples were subsequently used for quanti cation of DMS by GC analysis (as above) and reported as nmol DMSHB mg protein −1 min −1 . No DMS was produced from the no DsyB protein controls.
For kinetics analysis of N. denitri cans DsyB, the as-isolated protein was activated by addition of 400 µl heat-killed fraction 5 (see Fig. S1). K m and V max values were determined by nonlinear analysis based on the initial rates using 1.97 µM DsyB and 0-2 mM SAM, or 0-2 mM DL-MTHB (Fig. 2). The reaction mixture was incubated at 28°C for 30 min before detection DMSHB, as above. The enzyme activities were linear with respect to incubation time and enzyme concentration. Origin version 8.5 was used to calculate K m .

Mass spectrometry analysis
Denaturing LC-MS was conducted using a Bruker microQTof-QIII electrospray ionisation time of ight mass spectrometer, operating in positive mode. The spectrometer was calibrated with ESI-L Low concentration tuning mix (Agilent technologies). Samples were prepared by ten-fold dilution of ~100 µM protein solution with 2% (v/v) acetonitrile and 0.1% (v/v) formic acid to 0.5 ml. Samples were chromatographically separated by an UltiMate 3000 HPLC system (Dionex) tted with a ProSwift reversed phase RP-1S column (4.6 × 50 mm; Dionex). Hystar (Bruker Daltonics) was used to coordinate mass spectrometer and HPLC operations. Bound proteins were eluted using an isocratic gradient (2 -100% B) at a ow rate of 0.2 ml min -1 using the following solvents: solvent A (water, 0.1% (v/v) formic acid); and, solvent B (acetonitrile, 0.1% (v/v) formic acid). The eluant was continuously infused into the source of the mass spectrometer operating with the following parameters: dry gas ow 8 L min -1 ; nebuliser gas pressure 1.8 bar; dry gas 240 ˚C; capillary voltage 4,500 V; offset 500 V; collision RF 650 Vpp.
Hydrophilic interaction chromatography (HILIC)-MS (60) experiments were performed using the same mass spectrometer and HPLC system tted with a Luna NH 2 column (2 x 100 mm) (Phenomenex). For HILIC chromatography the following solvents were freshly prepared: solvent A (95% (v/v) aqueous 5 mM ammonium formate pH 3.75, 5% (v/v) acetonitrile); solvent B (95% (v/v) acetonitrile, 5% (v/v) aqueous 100 mM ammonium formate pH 3.75). Standard compounds (SAM, DL-MTHB, SAH (Sigma-Aldrich), DMSHB (10)) were used to calibrate the elution pro le of the HILIC column. Samples were brought to 92% (v/v) acetonitrile and loaded onto a column pre-equilibrated with solvent B. An optimized HILIC gradient was applied and compounds eluted (0.6 ml min -1 ) using the HILIC gradient between solvent A and solvent B, as previously described (10). The eluant was continuously infused into the source of the mass spectrometer (optimized for 50 -600 m/z) with the following parameters: dry gas ow 8.5 L/min; dry gas 200 °C; nebulizer pressure 1.2 bar; capillary voltage 4500 V; offset 500 V; collision RF 400 Vpp. Each HILIC-MS run contained an internal sodium formate calibration segment at the end of the run.
For non-denaturing mass spectrometry, an aliquot of proteins was exchanged into 250 mM ammonium acetate pH 8.0 using Zeba spin (Thermo Scienti c) or PD mini trap (Cytiva) desalting columns and infused (0.3 ml/h) directly into the ESI source of the mass spectrometer with the following parameters: dry gas ow 4 L min -1 ; nebulizer gas pressure 0.8 bar; dry gas temperature 190 °C; capillary voltage 3000 V; capillary offset 500 V; ion energy 8eV; collisional RF 1500 Vpp; collision cell voltage 5 V; and, ion transmission range 1500 -5500 m/z. Processing, isotope pattern simulation and analysis of denaturing LC-MS, HILIC-MS and non-denaturing MS data were carried out using Compass Data Analysis version 4.1. For denaturing LC-MS and nondenaturing MS, neutral mass spectra were generated using ESI compass Maximum Entropy deconvolution algorithm version 1.3. Proteins masses are reported from peak centroids representing the isotope average neutral mass and compared to predicted masses (Expasy) (61).

DsyB activity detection using HILIC-MS
Heat-killed J571 fractions that restored MTHB S-methyltransferase were added to pure DsyB as above to yield activated samples for analysis here. Samples of as-isolated DsyB prepared in HEPES buffer or activated DsyB, were immediately desalted (PD10, Cytiva) into 25 mM Tris, 100 mM NaCl pH 8.0 prior to conducting HILIC or non-denaturing MS experiments. The methyltransferase activity of DsyB was measured using DL-MTHB and SAM as substrates, as previously described (17) after 1-week incubation. X-ray diffraction data were collected on the BL18U1&BL19U1 beamlines at the Shanghai Synchrotron Radiation Facility. The initial diffraction data sets were processed by the HKL3000 program (62).

Structure determination and re nement
The crystals of DsyB-SAM complex belong to the P2 1 2 1 2 1 space group, while the crystals of DsyB-SAH-MTHB complex belong to the P2 1 space group. The structure of DsyB-SAM complex Se derivative was determined by SAD phasing. The crystal structures of DsyB-SAM complex and DsyB-SAH-MTHB complex were determined by molecular replacement using the CCP4 program Phaser (63) with the structure of the Se derivative as the search model. The re nements of these structures were performed using Coot (64) and Phenix (65). All the structure gures were produced with the PyMOL (http://www.pymol.org/).

Analyses of DMSP synthesis genes in cultured microorganisms
The presence or absence of DMSP synthesis genes in 111 cultured DMSP-producing bacteria (published since the discovery of bacterial DMSP synthesis (7,10,11,32,33) was analysed (Table S2). This is based previously published work that analysed their sequenced genomes and/or used degenerate primers to detect the presence of burB, dsyB and/or mmtN. Percentage abundances were calculated for all three DMSP synthesis genes within these cultured organisms, as well as those containing both dsyB and mmtN.
These were manually curated to con rm identity (E value cutoff of 1e −30 for DSYB), although since TpMMT has only been shown to be functional in T. pseudonana, we did not assume any sequences below 70% identity to T. pseudonana TpMMT to be functional. Strains con rmed to contain DSYB and/or TpMMT are listed in Table S3 and summarised in Table S4, alongside literature reporting the presence of DMSP synthesis in that particular strain (if tested).
Veri ed sequences (7, 10-12, 67) were aligned using ClustalOmega (68), and pro le hidden Markov models (hmms) of dysB, DSYB, and mmtN were constructed using the hmmbuild function of hmmer 3.3 (69). Tara metagenome (OM-RGC_v2_metaG/MATOU_v1_metaG) (prokaryotic/eukaryotic, respectively), and metatranscriptome (OM-RGC_v2_metaT/ MATOU_v1_metaT) sequences together with their abundances and taxonomic assignations were downloaded from the Ocean Gene Atlas site (70) using an hmmsearch e-value threshold of 1e-70 (dsyB), 1e-80 (DSYB) or 1e-60 (mmtN). A blastp search (e-value threshold of 1e-80) was used for TpMMT, using the Thalassiosira pseudonana TpMMT sequence as query. Environmental dsyB/DSYB sequences were aligned with Nisaea denitri cans dsyB using ClustalOmega (68), and sequences that did not possess all six essential residues were excluded from further analysis. Environmental TpMMT sequences greater than 400 amino acids in length were also excluded from further analysis. Prokaryotic sequence abundances were normalized using the median abundance of 10 single copy marker genes/transcripts (71). This gave abundance as a percentage of single copy gene abundance (equivalent to the percentage of cells containing a copy) in the metagenome, and transcription as a percentage of single copy gene transcription in the metatranscriptome. These marker gene/transcript abundances were downloaded from the Ocean Gene Atlas using the hmm pro les developed by Milanese et al. (71) with an e-value threshold of 1e-80. The MATOU_v1_metaG (metagenomic) database featured few MIX and MES sampling sites (2 and 7, respectively), limiting the power of comparative analysis between sampling depths, thus, these sites were excluded from analysis. Statistical analysis was performed in R (version 4.02) using RStudio.

Accession numbers
The structures of DsyB-SAM complex and DsyB-SAH-MTHB complex have been deposited in the Protein Data Bank (PDB) under the accession codes 7DFD and 7DFF, respectively. Figure 1 Predicted DMSP biosynthesis pathways (10). Different pathways are shown in different colours.