Dimethyl sulfide mediates microbial predator–prey interactions between zooplankton and algae in the ocean

Phytoplankton are key components of the oceanic carbon and sulfur cycles1. During bloom events, some species can emit large amounts of the organosulfur volatile dimethyl sulfide (DMS) into the ocean and consequently the atmosphere, where it can modulate aerosol formation and affect climate2,3. In aquatic environments, DMS plays an important role as a chemical signal mediating diverse trophic interactions. Yet, its role in microbial predator–prey interactions remains elusive with contradicting evidence for its role in either algal chemical defence or in the chemo-attraction of grazers to prey cells4,5. Here we investigated the signalling role of DMS during zooplankton–algae interactions by genetic and biochemical manipulation of the algal DMS-generating enzyme dimethylsulfoniopropionate lyase (DL) in the bloom-forming alga Emiliania huxleyi6. We inhibited DL activity in E. huxleyi cells in vivo using the selective DL-inhibitor 2-bromo-3-(dimethylsulfonio)-propionate7 and overexpressed the DL-encoding gene in the model diatom Thalassiosira pseudonana. We showed that algal DL activity did not serve as an anti-grazing chemical defence but paradoxically enhanced predation by the grazer Oxyrrhis marina and other microzooplankton and mesozooplankton, including ciliates and copepods. Consumption of algal prey with induced DL activity also promoted O. marina growth. Overall, our results demonstrate that DMS-mediated grazing may be ecologically important and prevalent during prey–predator dynamics in aquatic ecosystems. The role of algal DMS revealed here, acting as an eat-me signal for grazers, raises fundamental questions regarding the retention of its biosynthetic enzyme through the evolution of dominant bloom-forming phytoplankton in the ocean. Algal production of dimethyl sulfide plays a role in attracting predators and enhancing predation by zooplankton, thus mediating predator–prey relationships in the ocean.

fish 22 , sea birds 23 and penguins 24 , sense DMS and may use it to find areas of high prey biomass 25 . DMS is produced by algal DL during grazing 26,27 and viral infection 28 or by bacterial DL which uses algal DMSP as a substrate 29,30 . Nevertheless, the signalling role of the DL metabolic products, DMS and acrylate, during predator-prey interactions is still unresolved. Contradictory functions were proposed for DMS and DMSP, as anti-grazing defence 4,31 and pro-grazing chemoattractants 5 . Recently, the DL enzyme Alma1 was isolated from the bloom-forming coccolithophore Emiliania huxleyi (Prymnesiophyceae) 6 , providing a unique opportunity to unravel its ecophysiological role 6 . We study the role of DL during zooplankton-algae interactions by functional genomics and physiological approaches. We manipulated DL activity in prey cells to determine the influence of DMS on microzooplankton grazing dynamics. Microzooplankton  µm in size) are key marine herbivores, removing 49-77% of the photosynthetic biomass daily 32,33 . Their grazing on high DMSP producers leads to DMS production in algal prey cells 26,27 . We demonstrate that this DMS enhances grazing efficiency by the microzooplankton and mesozooplankton studied here and discuss the ecological and evolutionary implications of DMS-mediated herbivory in the ocean.
We first examined the effect of DMS emission on grazing efficiency during prey-predator dynamics between E. huxleyi high-DL strain CCMP373 (ref. 34 Fig. 1 and Supplementary Table 1); it is often used as a model microzooplankton to study the signalling role of DMS during grazing 4,31,35,36 . During short-term grazing experiments, DMS was emitted from grazed algal cells (~150 nM after 1 h) but not in prey-only control (Fig.  1b). To manipulate DMS production during grazing, we applied a recently developed inhibitor that selectively blocks the activity of the algal DL Alma1 but not the activity of known bacterial DLs 7 . This inhibitor, 2-bromo-3-(dimethylsulfonio)-propionate (Br-DMSP), blocks the DL activity of E. huxleyi when applied in vitro in 10 µM concentration 7 (Fig. 1a). To assess the possible indirect effect of Br-DMSP in live cells, we first tested the toxicity of the inhibitor in a dose-dependent manner and found that 0. decreased by ~50%, compared to the control grazed culture treated with methanol (Fig. 1b). The inhibitor caused partial reduction in DMS emission probably since suboptimal dose was applied to avoid non-specific effect on the grazer's physiology. In addition, we cannot exclude the potential production of DMS by associated bacteria in non-axenic E. huxleyi cultures ( Supplementary Fig. 1). Imaging of the grazer's food vacuole by fluorescence microscopy revealed that inhibition of algal DL activity led to >50% reduction in prey content as compared to ingestion of control cells after 0.5 and 3 h of predation (Fig. 1c,d). Furthermore, the ingestion rate of Br-DMSP treated cells declined by ~74% (~35 E. huxleyi per grazer per day) as compared to untreated control cells (~133 E. huxleyi per grazer per day), on the basis of quantification of prey clearance by flow cytometry ( Fig. 1e and Extended Data Fig. 3). Hence, despite apparently partial inhibition of DL activity by Br-DMSP, the grazing efficiency by O. marina was markedly reduced. To verify the specific effect of Br-DMSP on DMS production and grazing dynamics, we also used the alga Dunaliella tertiolecta as prey since it lacks the DL enzyme and does not produce DMS. Indeed, no change was detected in the consumption of D. tertiolecta by O. marina in the presence of Br-DMSP (Supplementary Table 2).
E. huxleyi cells produce minimal amounts of DMS during normal growth (Fig. 1b, Extended Data Fig. 1 and Supplementary  Table 1), probably because the DL and its substrate DMSP are segregated in different subcellular compartments 37 . However, algal senescence 38 and interactions with viruses 28 and grazers (Fig. 1b, Extended Data Fig. 1 and Supplementary Table 1) can trigger DL activity and DMS production as a result of damaged cellular membranes and mixing of DL and its substrate DMSP 26,34 . Here, reduction in the DL activity of the prey by ~50% significantly suppressed grazing efficiency by O. marina. We therefore suggest that increasing DL activity will enhance the grazing response by this grazer. As an attempt to mimic DL activity, we added exogenously DMS and/or acrylate, at a wide range of physiological relevant concentrations, to prey cells with low/no DL activity, namely E. huxleyi CCMP2090 (a high DMSP producer with low DL activity 34

NaTurE MICrObIOLOgY
diatom Thalassiosira pseudonana (a low DMSP producer with no DL 39 ). In both cases, application of DMS and/or acrylate did not significantly alter the grazing response of O. marina (Extended Data Fig. 4). However, it is questionable whether a single dose addition of DMS/P can mimic the spatiotemporal gradients of these infochemicals produced by algal cells during microscale prey-predator interactions. To preserve the physiological context of DMS release, we genetically manipulated the DL enzyme in prey cells. As E. huxleyi and other major DMS-producing species are yet to be genetically amendable, we conducted heterologous expression of the E. huxleyi DL gene in the ecologically important diatom T. pseudonana [40][41][42] . T. pseudonana is similar in size to E. huxleyi (3-5 µm) and synthesizes DMSP (~10 mM cell −1 ) 39 but lacks a DL and does not produce DMS 6 . Thus, transgenic T. pseudonana with DL activity can serve as a tractable model system to study the direct signalling role of DMS during grazing interactions. We overexpressed in T. pseudonana the DL-encoding gene from E. huxleyi 373 (alma1) fused to green fluorescent protein (GFP). The GFP was conjugated at the C terminus to preserve a putative target sequence at the N terminus 43 (Fig. 2a,b and Extended Data Fig. 5). The morphology and growth of the transgenic diatom cells were not affected by the expression of Alma1-GFP (Extended Data Fig. 5a,b). The heterologous expression of Alma1-GFP in the transgenic T. pseudonana line, called Tp DL-GFP, was verified by western blot using an antibody raised against the Alma1 protein 6 (Extended Data Fig. 5d). The DL enzymatic activity in Tp DL-GFP cell lysate was comparable to E. huxleyi high-DL strains 6,34 (8.8 ± 2.9 fmol DMS cell −1 min −1 , Fig. 2c). To investigate the effect of DMS produced by Tp DL-GFP cells on grazing efficiency by O. marina (Fig. 2d), short (0-2 h) and long-term (1-6 d) feeding experiments were conducted. Although Tp DL-GFP cells produced DMS levels similar to those found in seawater samples (~10 nM) 27 , we did not detect a significant increase in DMS concentration in the media of grazed Tp DL-GFP (as in E. huxleyi, Fig. 2e). This is probably due to the smaller intracellular DMSP in T. pseudonana (10-20 mM) 39 as compared to E. huxleyi (~50-300 mM) 34,44 . Nevertheless, when O. marina was fed with Tp DL-GFP cells which produce ~10 nM DMS, the grazers accumulated significantly more prey in their food vacuole than when fed with wild-type cells, which do not produce any trace of DMS (P < 0.0001, Fig. 2f). In accordance, the immediate ingestion rate on Tp DL-GFP cells, measured during 30 min from prey addition, was around twofold to fourfold higher than on wild-type cells (24 ± 5 and 6 ± 4 T. pseudonana per grazer per day, respectively, P < 0.0001; Fig. 2g and Extended Data Fig. 6). Furthermore, the ingestion rate by O. marina was elevated when fed with T. pseudonana expressing only DL (Tp DL, Extended Data Fig. 7), while ingestion of T. pseudonana expressing only GFP (Tp GFP) was similar to wild type (Extended Data Fig. 6b). This eliminates the possibility that the GFP fluorescence of the prey modulated grazing activity and indicates that enhanced grazing on Tp DL-GFP cells was due to their DMS production.
Next, long-term feeding experiments were conducted in which O. marina was fed with Tp DL-GFP or wild-type diet for 6 d. Daily feeding with Tp DL-GFP cells significantly improved O. marina growth, as compared with wild-type cells (P = 0.0011, Fig. 2h). Thus, DL activity in prey cells expedited grazing dynamics, as well as enhanced the growth of the grazers on a longer timescale. This may be attributed to general higher consumption of algal biomass and also to specific assimilation of the DMS/P of the diatom as an essential nutritious organosulfur 36 .
We further examined if O. marina preferentially ingested DMS-producing cells or alternatively whether the diffusible DMS in the media may promote a general, non-selective grazing response. During double-prey competition trials, the ingestion rate of the total prey mixture containing Tp DL-GFP and wild-type cells (1:1) was over twofold higher as compared to ingestion of a mixture of wild-type cells with Tp GFP (Fig. 2i). This indicated that DMS formed by DL activity in grazed cells can increase the grazing likelihood of neighbouring cells with no DL activity. Therefore, DMS can potentially enhance grazing within natural algal communities where only a fraction of the cells secretes DMS. In contrast, we did not detect grazing preference by O. marina towards Tp DL-GFP cells as compared to wild-type or Tp GFP cells (Fig. 2i). Thus, O. marina grazing on Tp DL-GFP cells was non-selective and may be attributed to dissolved chemical signals such as DMS, which act as an appetizing infochemical enhancing general grazing activity.
To investigate the prevalence of DL-mediated herbivory, grazing assays were conducted with microzooplankton from different taxa, size and feeding modes (direct engulfment or filter feeding). The ability of the microzooplankton tested to produce and cleave DMSP has not been assayed but they lack homologues for known Alma1 enzymes 45 . The fold-change between the grazing rate on Tp DL-GFP and Tp GFP cells, which differ only by their DL activity, was calculated for each grazer species (if the fold-change was >1, Tp DL-GFP cells were consumed faster than Tp GFP cells  Table 3). In contrast, G. dominans grazed similarly on both prey types. Next, we assessed grazing by larger mesozooplankton (0.2-2 mm) on the transgenic diatoms. Cultured Artemia salina (brine shrimp) consumed Tp DL-GFP twice as fast as Tp GFP cells. Furthermore, wild calanoid copepods of the species Pleuromamma indica consumed Tp DL-GFP ~28 times more than they did the Tp GFP control (Fig.  3a). Notably, none of the tested predators exhibited a negative grazing response toward Tp DL-GFP prey (fold-change <1). While DL-mediated herbivory may be species-specific, it is clearly not specific to O. marina alone but shared by representatives of several important zooplankton taxa.
To further assess the ecological importance of DMS-mediated herbivory, we used mixed native microzooplankton assemblages from the Gulf of Aqaba in the northern Red Sea. Microscopy and 18S amplicon sequencing of these assemblages revealed a wide diversity of heterotrophic and mixotrophic species within the 5-200 µm size range, such as tintinnid ciliates, dinoflagellates, nauplii and small copepods (most cells <5 µm were excluded during net-tow harvesting; Extended Data Fig. 8 and Supplementary Figs. 2 and 3). The native grazers were subsequently fed with Tp DL-GFP or Tp GFP prey cells for 24 h. The uptake of the transgenic diatom cells by the natural grazers was confirmed by microscopy and quantified as grazing rate (g, (d −1 ); Fig. 3b,c, Extended Data Fig. 9 and Supplementary Fig. 4). The grazing rates measured in four experiments conducted during 2017-2020 were variable, perhaps due to changes in the taxonomic assortment of the grazer communities in each experimental flask. Nonetheless, in three out of four experiments, the native grazers removed the Tp DL-GFP prey cells significantly faster than the Tp GFP control (P < 0.016). The highest mean grazing rate was detected in February 2020, where Tp DL-GFP cells were removed on average around ten times faster than Tp GFP control cells (0.22 ± 0.12 d −1 , P = 3.3 × 10 −5 , Fig. 3b). Overall, most of the cultured and wild grazers tested here exhibited a varied degree of enhanced grazing on DMS-producing prey, indicating that DMS facilitates herbivory across a wide variety of zooplankton taxa representing different size and feeding strategies. Thus, the current study directly links DL activity in algal prey cells to enhanced grazing activity by zooplankton.
Since the historical discovery of DMSP and its catabolic product DMS from a seaweed about 85 years ago 47,48 , numerous phytoplankton were reported to produce these metabolites, including Letters NaTurE MICrObIOLOgY bloom-forming species such as E. huxleyi and Phaeocystis and coral symbionts such as Symbiodinium 6,10,49 . DMSP plays essential cellular roles as an osmolyte and antioxidant in several phytoplankton 13,14 and, when released to the water column due to cellular stress and lysis 16,50 , can serve as an important source of reduced sulfur and carbon for bacteria 30,51 . Here, we applied genetic and physiological approaches that enabled in vivo modulation of the DL enzymatic activity which cleaves DMSP to DMS and acrylate, and showed that DMS and/or acrylate enhanced predation. Hence, the evolutionary retention of the DL gene comes with a marked cellular cost of prompting grazing by diverse herbivores. This conundrum may indicate an essential physiological role for DL activity in algal cells

NaTurE MICrObIOLOgY
which is yet to be discovered. DL activity in E. huxleyi strains, which span about four orders of magnitude range 34 , was initially described as a grazing-activated chemical defence mechanism 4,31 . Yet recent transcriptomes of the different E. huxleyi strains with a wide range of DL activity demonstrated that they differ also in the expression of hundreds of genes 6,52 . Hence, it is difficult to assign a role for a single trait as DL activity based on comparison between E. huxleyi strains as prey. Furthermore, the role of DL in defence against predators was shown to be species-specific (O. marina had only little or no feeding inhibition on high-DL prey) 31 . Past studies also suggested that bulk addition of DMSP, but not of DMS or acrylate, reduced grazing rates by some grazers 53 . In contrast, DMS and DMSP were later shown to act as chemo-attractant cues when introduced in minute plumes to diverse marine protists, including O. marina 5 . Here, we provide direct evidence for the signalling role of DMS leading to enhanced grazing by diverse grazers, which corroborates the suggested chemo-attraction role for DMS 5 . We therefore propose that DMS, and/or potentially acrylate, act as a pro-grazing signals and accelerate the removal of algal cells by their immediate protozoan predators. The response of marine protists to the algal-derived DMS or acrylate is analogous to the phagocytic activity of immune cells in mammalians, which are attracted to damaged and apoptotic cells via secreted 'eat-me' signals 54 . As mentioned above, the identity of the eat-me signal during predator-prey interactions is likely to be DMS but the role of acrylate, which is also produced by the DL enzyme, should be considered. Acrylate levels were not quantified here and further work is required to distinguish between the potential signalling roles of DMS and acrylate. Acrylate is a charged molecule and would require a transporter to enter the cell 55 . In bacteria, acrylate-utilization enzymes, AcuH and AcuI, catabolize acrylate to yield 3-hydroxypropionyl-CoA 30,56 but a similar mechanism in eukaryotes is still unknown. In O. marina, exposure to acrylate led to negative chemosensory response 35 and accumulation of acrylate in the food vacuole during grazing of E. huxleyi was suggested to inhibit ingestion 4 . Unlike acrylate, DMS is uncharged and can readily diffuse through biological membranes 57 . While bacteria oxidize DMS and also use it as carbon source via several enzymatic pathways 58 , a mechanism for DMS catabolism by eukaryotes remains to be discovered.
Our findings suggest that in the ocean, DMS-mediated herbivory may have a substantial impact on bloom dynamics of DMS-producing species like E. huxleyi. Since E. huxleyi strains are dramatically different in their DL activity levels 34 , we suggest that grazing of high-DL strains can release DMS that acts as an appetizing signal, leading to enhanced grazing pressure imposed on the entire blooming population (Fig. 4). Considering that blooms are prone to infection by pathogenic bacteria 16

NaTurE MICrObIOLOgY
At the ecological level, the response to DMS by zooplankton 19,21 , and specifically microzooplankton as shown here, is akin to what has been previously described for animals such as fish 22 , Procellariiformes birds 23 , penguins 24 and turtles 60 . These top-predators may use DMS as a foraging cue to find 'hot-spots' of high biomass and plankton concentrations 25,61 . Intriguingly, by facilitating a trophic cascade, DMS may promote indirect defence mechanisms by attracting predators of the herbivorous micrograzers. In the Southern Ocean, sea birds are highly sensitive to DMS emitted from Phaeocystis sp. blooms, track its source and feed mainly on krill, the main grazer of Phaeocystis sp. in this region 23 . In such a tritrophic mutualism scenario, sea birds may reduce grazing pressure on the phytoplankton. DMS was suggested to mediate pelagic tritrophic interactions in the plankton as well, where DMS-derived from phytoplankton-microzooplankton interaction triggers an indirect defence by attracting carnivorous mesozooplankton, which consume the microzooplankton and alleviate the grazing pressure imposed on the algal population 62,63 . This theory was supported by mathematical models 64,65 and evidence from coccolithophore and Phaeocystis blooms, where microzooplankton were found to be the main food source for mesozooplankton (copepods) rather than the phytoplankton directly 66,67 . Such copepods may be highly sensitive to DMS emitted during phytoplankton-microzooplankton grazing interaction [19][20][21] . In accord, we found that copepods and krill produced elevated levels of faecal pellets when fed with O. marina cells which were pretreated with DMS-producing prey (E. huxleyi or Tp DL-GFP cells). The mesozooplankton response was species-specific and may indicate an enhanced copepod predation in response to grazing-derived DMS (Extended Data Fig. 10). The interplay between algal DMS and grazing at likely different trophic levels adds a new dimension to the complex trophic interactions established in oceanic ecosystems and particularly during algal blooms. Overall, the ability of DMS to attract marine biota from diverse size ranges and taxonomic groups makes it a common foraging cue which may serve as a global eat-me signal across trophic levels. The prevalence of DMS and the ubiquity of DMS-mediated predator-prey interactions indicate that DMS plays fundamental roles in chemical communication among multiple trophic levels in the ocean.

Taxonomic composition of bacteria associated with E. huxleyi cultures.
Three strains of E. huxleyi (CCMP373, CCMP374 and CCMP2090) were grown without antibiotics and harvested during exponential growth phase. For each strain, 2 × 100 ml of culture (duplicates) were prefiltered on autoclaved 0.8-µm polycarbonate membrane filters (Merck Millipore, ATTP04700). The filtrate was collected in sterile bottles and concentrated on 0.2-µm polycarbonate membrane filters (Merck Millipore, GTTP04700). DNA was extracted from the filters using the DNeasy PowerWater kit (QIAGEN) according to the manufacturer's instructions. The V4-V5 region of the 16S ribosomal DNA region was amplified using the 515F-Y forward primer (5′-GTGYCAGCMGCCGCGGTAA-3′) 69 and the 806 R reverse primer (5′-GGACTACNVGGGTWTCTAAT-3′) 70 , combined with CS1 and CS2 Illumina adaptors. We used the following PCR mix: 12.5 µl of buffer MyTAQ HS 2× mix, 1 µl of each primer at 0.4 µM final concentration, 0.75 µl of DMSO 3%, 8.75 µl of ultrapure water and 1 µl of DNA template. We used the following PCR conditions: initial denaturation of 3 min at 94 °C followed by 28 cycles of 15 s at 94 °C, 10 s at 75 °C, 10 s at 55 °C, 30 s at 72 °C and final elongation of 10 min at 72 °C. ZymoBIOMICS Microbial Community DNA Standard (ZYMO RESEARCH) was used as a positive control and water was used as a negative control. All PCR products were prepared for Illumina sequencing on a MiSeq 2 × 250 base pairs (bp) by the IUC Sequencing Core. ASVs were called using DADA2 (ref. 71 ), annotated using the Silva reference database 72 and analysed using the phyloseq package 73 .

Genetic transformation of T. pseudonana.
To overexpress the Alma1-GFP fusion protein in T. pseudonana, wild-type cells were transformed using a biolistic particle delivery system fitted with 1,350-psi rupture discs (Bio-Rad). M10 tungsten particles (0.7 μm) were coated with 5 μg of plasmid DNA in the presence of CaCl 2 (2.5 M) and spermidine, according to the manufacturer's instructions. The plasmid encodes for nourseothricin resistance and Alma or Alma-GFP fusion protein. Chemo-attraction of zooplankton to leakage of DMS from phytoplankton such as E. huxleyi may facilitate initial grazing interaction (left). On dissociation of ingested cells during phagocytosis, the DL Alma1 degrades DMSP and releases more DMS into the water (middle). The microscale release of DMS may accumulate to high concentration (in yellow), which induces an increase in the grazing pressure (in purple) by diverse grazers on the entire phytoplankton population (right) and thus has a wide ecological implication in the marine environment.

NaTurE MICrObIOLOgY
Approximately 1 × 10 6 cells were spread on a plate containing solid medium (1.5% Bacto agar in FSW + F/2) a few hours before bombardment. Bombarded cells were recovered in F/2 media + silica and incubated at 18 °C under constant light overnight. The culture was then plated onto solid medium containing 100 μg ml -1 of nourseothricin. Plates were incubated for 10 d and resistant colonies were restreaked onto fresh solid medium containing nourseothricin. The colonies obtained were then screened for GFP fluorescence and DL activity.

Measurements of total DMSP, DMS and DL activity.
For determination of DMSP concentration, cultures (~3 ml) were acidified with 1.5% HCl and stored at 4 °C for >24 h (ref. 16 ). Samples were diluted (1:100) in double-distilled water and the total DMSP in the culture (particulate + dissolved) was converted to DMS by adding NaOH to a final concentration of 0.45 M. Glycine buffer (pH 3) was added to a final concentration of 0.8 M to neutralize the solution (pH 8-9). Samples were then measured for DMS concentration in sealed glass vials, using Eclipse 4660 Purge-and-Trap Sample Concentrator system equipped with Autosampler (OI Analytical). Separation and detection were done using gas chromatography-flame photometric detector (GC-FPD, HP 5890) equipped with RT-XL sulfur column (Restek) 7 . Although measuring total DMSP in the culture, we present our data as cellular DMSP (Fig. 2c) since DMSP was not detect in in the media of exponential T. pseudonana. As for E. huxleyi, dissolved DMSP in exponential cells was also negligible 16 . Therefore, we consider our total DMSP measurements to represent accurately the cellular DMSP pool, once the data were normalized per cell count and cell volume ( DMSP lyse activity (in vitro) was estimated as DMS release from cell lysates and measured as previously described 6 . Cells were harvested by centrifugation. Cell pellets were resuspended in lysis buffer (Tris 100 mM pH 8.0, NaCl 100 mM, DTT 1 mM, Triton 100 × 0.02%, Benzonase 250 units) and sonicated. Crude lysates were incubated with 10 mM DMSP while shaking for 10 min at 30 °C. Reactions were terminated by 1,000-fold dilution into sealed glass vials containing 30 ml of distilled water. The vials were kept on ice in the dark until the DMS measurement, conducted as described for the DMSP analysis. All measurements were calibrated against a standard curve of DMS, ranging from 5 to 300 nM (Sigma-Aldrich).

Western blot analysis.
For detection of the Alma1 protein expression, cells were harvested at exponential growth phase by centrifugation (10,000g, 15 min, 4 °C) and plunged into liquid nitrogen. Cell pellets were then resuspended in lysis buffer and bath-sonicated (5.5 s of sonication following by 10 s of rest, five cycles). Cell lysates and recombinant Alma1, used as a positive control 6 , were separated on an Any kD SDS-PAGE (Bio-Rad). For western blot analysis, we used a primary rabbit polyclonal antibody raised by immunization against the recombinant Alma1 protein 6 and a secondary horseradish peroxidase-conjugated antirabbit (Sigma-Aldrich). The antibodies were diluted in Tris-buffered saline containing 0.1% Tween-20 and 5% skimmed milk powder. The primary antibody was diluted 4,000-fold and the secondary antibody was diluted 10,000-fold. ECL-Prime reagent (GE Healthcare) was used for detection.

Scanning electron microscopy of T. pseudonana cells.
For morphological analysis of T. pseudonana cells, samples were collected during exponential growth phase. Samples (1 ml) were blotted at room temperature on Nuclepore track-etched polycarbonate membranes (Whatman), coated with 5 nm of iridium for improved conductance and imaged with a Sigma 500 SEM (Zeiss) using the InLens detector.
Grazing assays and determination of short-and long-term grazing and ingestion rates. Grazing assays were conducted to determine short-and long-term grazing and ingestion rates (g and IR, respectively). Before grazing assays, microzooplankton (O. marina, G. dominans and Strombidium sp.) were starved for 3 d to clear their digestive vacuole from their routine prey, D. tertiolecta. Prey cells (E. huxleyi or T. pseudonana) were added to the grazer culture to reach a final prey:grazer ratio of 3:1 or 10:1 (1-30 × 10 4 prey cells ml -1 ). For O. marina, grazer concentration was 3 × 10 3 cells ml -1 and assays were conducted for 0.5-24 h, as designated in each graph. In short-term grazing assays (up to 4 h), prey (0.1-0.2 ml) was added to the grazer culture in a 24-well plate (total volume in each well was 2.5 ml). Prey cells were also added to FSW as a growth control. Each group included four to six biological replicates. The plate was placed in the flow cytometer autosampler, where each well was sampled every 30 min to quantify the algal cells, for a total period of 2-4 h (Extended Data Fig. 3). Long-term grazing assays were conducted similarly to the short-term assays, except that the prey (~1 ml) were added to the grazer cultures in small flasks (50 ml each) which were mounted on a rotating plankton wheel and incubated in the growth condition indicated above. Samples were taken for cell counts by flow cytometry at t = 0 and 24 h. The flow cytometer, Eclipse iCyt (Sony Biotechnology), was equipped with 488-nm solid-state air-cooled laser, with 25 mW on the flow cell and with standard filter setup. Algal cells were identified by plotting the chlorophyll fluorescence (excitation 488 nm and emission 663-737 nm) against side scatter. At least 2,500 cells were collected in each sample. Grazer cells were counted manually. The algal specific growth rate (µ) was calculated for the control and grazing treatments. Values for g and IR were calculated on the basis of µ and normalized per grazer, as described by Frost 74 . For G. dominans and Strombidium sp., short-term grazing assays were conducted as described for O. marina, for a period of 3 h.
When using Br-DMSP 7 , the inhibitor was added to the E. huxleyi culture 2 h before grazing interaction. The inhibitor was dissolved with 10% MeOH and added in a final concentration of 0.2 µM, which was chosen on the basis of toxicity assays (Extended Data Fig. 2). The same volume and concentration of MeOH was used as control. The toxicity of the inhibitor for O. marina cells was tested after 2 and 3.5 h by using SYTOX Green (Invitrogen) staining to detect compromised cell membranes. Samples were stained with a final concentration of 1 μM SYTOX, incubated in the dark for 30 min and analysed by an Eclipse flow cytometer, as described below (excitation 488 nm and emission 500-550 nm). An unstained sample was used to eliminate the background signal.
O. marina vacuole content analysis. Manual quantification of ingested prey was adapted from previous studies 31 . Prey cells (E. huxleyi or T. pseudonana, 3 × 10 4 cells ml -1 ) were added to the O. marina culture (3 × 10 3 cells ml -1 ) at t = 0. Incubation of prey and predator were conducted in 50-ml culture flasks. Subsamples were fixed with 1% PFA at two time points (0.5, 3 h) for E. huxleyi or during a time course for T. pseudonana (8,17,30,45 and 60 min) prey. Fixed cells were gently transferred into an Utermoehl sedimentation system (Aquatic Research Instruments) and collected onto a microscope slide for 24 h. Ingested prey items were counted by using an IX71S1F-3-5 inverted Olympus microscope, equipped with an EXi Blue camera (Q Imaging). A total of 100 grazer cells were observed per sample. The average prey content was calculated for each sample.

O. marina growth curve.
To quantify the effect of DMS-producing prey on grazer growth, O. marina culture (3 × 10 3 cells ml -1 ) was divided to three dietary treatments: starvation, T. pseudonana wild-type and Tp DL-GFP. Each treatment contained four 50-ml small flasks, which represent biological replicates. Prey cells were supplied daily and added to a final concentration of 3 × 10 4 cells ml -1 . In the Tp DL-GFP treatment, prey cells were consumed almost entirely after 24 h (less than ~5,000 cells ml -1 remained), while in the Tp wild-type treatment, there was much less grazing and the cell concentration remained constant (~2.5-3 × 10 4 cells ml -1 ). Thus, fresh prey cells were added daily to a similar final concentration (3 × 10 4 cells ml -1 ) in each treatment, to avoid possible bias as a result of different prey availability in each treatment. The flasks were mounted onto a rotating plankton wheel and incubated with the growth condition indicated above. Daily samples for grazer cell count were fixed with 10% Lugol (Sigma-Aldrich) in a 24-well plate. O. marina cells were counted by using an IX71S1F-3-5 inverted Olympus microscope (described above). A total of 140-1,000 cells were counted per sample.
Grazing assays with natural microzooplankton assemblages. To assess grazing on transgenic diatoms by natural microzooplankton, experiments were conducted during 2017-2020 at The Inter-University Institute for Marine Sciences in Eilat, Israel. Surface water was collected from the pier area (29° 50′ N, 34° 91′ E) by using a concentrating plankton net (5 µm). Sampled water (4 l) was then gently sieved through a 200-µm mesh, to obtain the 5-200-µm fraction of microzooplankton community and used for the 'grazing' treatment. Seawater was also filtered through a 0.22-µm filter to remove all cells and used for the 'growth' treatment. Each water fraction (5-200 µm and <0.22 µm) was divided into 300-830-ml polycarbonate flasks. Diatom cultures (Tp GFP or Tp DL-GFP) were added to the flasks (3 × 10 4 cells ml -1 ) at t = 0, including five to six biological replicates ( Supplementary  Fig. 2). Incubation flasks were placed in a shaded water-table with ambient water temperature (22-23 °C) for 24 h. In 2020, the incubation flasks were mounted onto a rotating wheel (0.3 r.p.m.) and placed in the dark at 22 °C for 24 h. Samples for cell count were taken at t = 0 and after 24 h, using flow cytometry (FCM) as described above. Grazing rate was calculated as previously described 75 . First, the growth rate was calculated for each flask as: µ = (lnC 24 -lnC 0 )/(T 24 -T 0 ), where C 24 and C 0 are the concentrations of cells at times T 24 and T 0 , respectively. Then, the grazing coefficient g was calculated by: g = µ growth -µ grazing , using the average µ for the growth treatment.

Taxonomic composition of the natural grazer community in the Red Sea.
The identity of the natural grazer community was assessed by microscopy for experiments conducted in February 2017 and 2019 or 18S profiling for experiments conducted at January and February 2020. For microscopy, samples were collected from the grazing treatment, as described above, or from control bottles with no added diatom prey, and fixed with PFA (1%), as shown in Fig. 3c. For 18S profiling, samples were collected from the grazing treatment, as described Letters NaTurE MICrObIOLOgY above, and then gently filtered onto polycarbonate filters. The filters were plunged into liquid nitrogen and kept at -80 °C until DNA extraction. DNA was extracted from the filters using the DNeasy PowerWater kit (QIAGEN) according to the manufacturer's instructions, with a 10 min incubation at 65 °C after addition of PW1 buffer. The V4 region of the 18S rDNA region was amplified using the TAREuk454FWD1 (5′-CCAGCA(G/C)C(C/T)GCGGTAATTCC-3′) 76 and a modified V4Rev_Piredda (5′-ACTTTCGTTCTTGATYRATGA-3′) 77 , combined with CS1 and CS2 Illumina adaptors. We used the following PCR mix: 12.5 µl of buffer MyTAQ HS 2× mix, 1 µl of each primer at 0.4 µM final concentration, 0.75 µl of DMSO 3%, 8.75 µl of ultrapure water and 1 µl of DNA template. PCR conditions were as follows: initial denaturation of 2 min at 95 °C followed by ten cycles of 10 s at 95 °C, 30 s at 53 °C and 30 s at 72 °C then 15 cycles of 10 s at 95 °C, 30 s at 48 °C and 30 s at 72 °C, with final elongation of 10 min at 72 °C. DNA extract from an E. huxleyi culture (RCC1216) was used as a positive control and water was used as a negative control. All PCR products were prepared for Illumina sequencing on a MiSeq 2 × 250 base pairs (bp) by the IUC Sequencing Core. Amplicon sequencing variants were computed using the DADA2 pipeline 71 , annotated with the PR2 database 78 and analysed using the phyloseq package in R (ref. 73 ).
Grazing assays with A. salina and P. indica. For assessment of grazing rate by A. salina, experiments were conducted with 15-day-old females that were starved for 3 d. Four individuals were washed with FSW and carefully transferred using a pipette to 50-ml flasks containing FSW or transgenic T. pseudonana cultures (Tp DL-GFP or Tp GFP). Initial prey concentration was 1 × 10 5 cells ml -1 . Algal cells were counted by FCM at t = 0 and 1-3 h and g was calculated as described in Frost 74 . Grazing experiments with P. indica were conducted with wild animals collected at the northern Gulf of Aqaba, Red Sea (station A, 29° 28′ N, 34° 56′ E) by using 300-µm mesh nets. Populations of P. indica were carefully isolated from the heterogeneous assemblages. Groups of ten individuals were transferred to Petri dishes with FSW for 6-10 h to allow evacuation of their gut content. At t = 0, the animals were transferred to incubation flasks (300 ml, ten per flask), supplemented with FSW or transgenic T. pseudonana cultures (Tp DL-GFP or Tp GFP, 3 × 10 4 cells ml -1 ). Algal cells were counted by FCM at t = 0 and 48 h and g was calculated as in Frost 74 .

Mesozooplankton tritrophic grazing experiments.
To assess grazing and defecation by mesozooplankton in a tritrophic setup, samples were collected during 2019-2020 at the northern Gulf of Aqaba, Red Sea (station A, 29° 28′ N, 34° 56′ E) by using 300-µm mesh nets. Populations of Euphausia diomedeae, Rhincalanus nasutus and P. indica were carefully isolated from the heterogeneous assemblages and used in separate experiments, where their gut content and faecal pellet production were quantified in response to different treatments. Groups of 10-12 individuals were transferred to Petri dishes with FSW for 6-10 h to allow evacuation of their gut content. At t = 0, the animals were transferred to incubation flasks (300 ml), supplemented with FSW or O. marina and phytoplankton cultures. O. marina was fed with phytoplankton (3 × 10 3 cells ml -1 and 3 × 10 4 cells ml -1 , respectively) for a few hours before the addition of mesozooplankton, to allow the accumulation of grazing-derived chemical cues. Incubation flasks were mounted onto a rotating wheel (0.3 r.p.m.) and incubated in the dark at 22 °C for 48 h. After 48 h, the animals were collected by using forceps for DNA extraction. The copepods were washed twice with FSW to remove O. marina cells that were attached to their external body parts and then flash-frozen in liquid nitrogen (animals from the same flask were grouped into a single tube before freezing). The specific ingestion of O. marina cells was confirmed by detecting a specific O. marina DNA sequence within the copepod gut. DNA was extracted from whole copepods by classical phenol-chloroform extraction and then cleaned by AMPure XP beads (Beckman Coulter). Primers for quantitative PCR (qPCR) were designed by Primer3 and Amplify4 (W. Engels, University of Wisconsin 2015) to amplify the mitochondrial cob-cox3 gene, which is unique to O. marina 79,80 . Forward primer: 5′-TCATGCTTTTATCTTTCTATCCA-3′; reverse primer: 5′-AGCTAAGAATAAAGTAGAAGGAGA-3′. For qPCR, Platinum SYBR Green qPCR SuperMix-UDG with ROX was used as described by the manufacturer (Invitrogen). Reactions were performed on Step-One Plus real-time PCR System (Applied Biosystems) as follows: 50 °C for 2 min, 95 °C for 2 min, 40 cycles of 95 °C for 15 s, 60 °C for 30 s. For quantification of faecal pellets, samples from the incubation bottles were collected after 48 h and fixed with 10% Lugol (Sigma-Aldrich). Pellets were then sedimented over 24 h in the dark by using Utermoehl sedimentation system (Aquatic Research Instruments). The faecal pellets were imaged and counted under a light microscope. The width and length of each pellet was measured using the ImageJ software and their volumes were calculated assuming they are cylinders. The carbon content of the pellets was estimated assuming a carbon volume ratio of 55 (µg C mm -3 ) 81 .
Statistical analysis. Group comparisons were conducted with Student's t-test (for two groups) or ANOVA (for more than two groups) followed by Tukey's or Dunnett's posthoc tests. Experiments with repeated measures per vial were tested with mixed effects models, treating vial as a random factor. A generalized linear model with Poisson distribution was used when analysing discrete counts. All analyses were conducted in R v.4.0.2. Generalized linear mixed models were done using the package lmerTest v.3.1-2.
Illustrations were created with BioRender.com. Graphs were created with Excel GraphPad Prism.
Reporting Summary. Further information on research design is available in the Nature Research Reporting Summary linked to this article.

Data availability
The data supporting Figs. 1-3 and all Extended Data figures in this study are deposited in Dryad and are available at https://doi.org/10.5061/dryad.cjsxksn66. The amplicon sequencing data were deposited on NCBI; accession numbers are MZ645612-MZ645697 for the 16S analysis and MZ662955-MZ663668 for the 18S analysis. Source data are provided with this paper.

code availability
No new code was used in this study.   Fig. 3b.), seawater were collected by a concentrating plankton net (5 µm pore size) and then passed through a 200 µm mesh (as described in Supplementary Fig. 2). The 5-200 µm fraction contained various grazers that were fed with the transgenic diatoms. With the aim of characterizing the grazer population in the experiment, a sample of grazer-containing seawater was filtered prior to the addition of the diatom prey, and used for DNA extraction and 18 S sequencing. a, The relative abundance of taxonomic groups with grazer species. Two water samples are presented (technical replicates). Non-grazer taxa such as parasites, fungi, autotrophic phytoplankton and land plants were omitted from the graph. A detailed taxonomic analysis is shown for the main phyla represented in the analysis, namely arthropoda (b), ciliophora (c) and the superclass dinoflagellata (d).

NaTurE MICrObIOLOgY
Extended Data Fig. 9 | Uptake of T. pseudonana transgenic cells by wild tintinnid ciliates. Natural microzooplankton were collected from the Red Sea and incubated with Tp DL-GFP or Tp GFP cells as prey for 24 h. a, A variety of tintinnids observed during grazing experiments. Samples were fixed with Lugol. b, A tintinnid feeding on Tp GFP cells. Some ingested diatoms are observed inside the lorica. The ciliate itself is at the oral end of the lorica, collecting prey using its cilia. The diatoms' GFP cannot be clearly observed in this case, since this tintinnid has natural green fluorescence. c, A tintinnid fed with Tp DL-GFP cells. The ciliate cell itself is contracted inside the lorica. d, A magnified view of the inset in (c), showing the fluorescence of ingested Tp DL-GFP cells. The micrographs represent ~20 ciliate cells which were observed with ingested prey during different experiments as summarized in Fig. 3b in the main text. O. marina was pre-fed with different diets as indicated in the 'Prey' column. The crustaceans gut content was measured after 48 h by qPCR in order to estimate their direct feeding response on O. marina (Om). For technical reasons, gut content analysis based on prey-DNA content yielded very low values and could not be quantified accurately. Thus, only approximate evaluation of O. marina ingestion by the crustaceans is presented in the 'Om ingested' column. The crustaceans faecal pellet (FP) production was estimated after 48 h as a proxy for their ingestion. The pellets were counted manually under a light microscope and quantified for pellet count, volume and carbon content, assuming a carbon: volume ratio of 55 [µg C mm -3 ] 81 . Statistical significance (p-value) is related to FP biovolume and carbon, as was calculated by 1-way ANOVA with Dunnett's post-hoc test, for compering each diet to the no-prey treatment. ani = animal; d = day.