Chemical prophage induction selectively removes Vibrio from a pelagic Sargassum-derived multispecies biofilm

Abstract

Background

Pelagic Sargassum has undergone significant range expansion and dramatic blooms in the Atlantic over the past 15 years. This alga’s microbiome provides symbiotic functions that are believed to contribute to its ecological success. Recent research shows that Sargassum-associated bacteria are enriched in integrated prophages compared to the surrounding seawater and that these prophages are inducible by chemical and ultraviolet treatment.

Results

Here, we investigated a Sargassum-derived in vitro multispecies biofilm encompassing the dominant heterotrophic microbial members associated with Sargassum to probe the impacts of prophage induction on the composition of Sargassum microbiomes. Induction was quantified by coverage-based virus-to-host ratios in chemically induced treatments with Mitomycin C and non-induced controls, and the community composition and metabolic profiles were analyzed after Mitomycin C treatment. Chemical induction led to a significant increase in abundance and virus-to-host ratio of viral genomes linked to Vibrio metagenome-assembled genomes. This was accompanied by altered biofilm community composition, with a reduction in Vibrio bacterial abundance that opened niche space for other biofilm members in the genera Pseudoalteromonas, Alteromonas, and Cobetia. The induced Vibrio-associated phages encoded genes involved in quorum sensing, biofilm formation, virulence, and host metabolism. Induction led to the depletion of 17 metabolic modules, including functions related to energy metabolism and nitrogen utilization.

Conclusion

Due to the high frequency of lysogeny in the Sargassum microbiome and the susceptibility of prophages to chemical and ultraviolet light induction, these results suggest that prophage integration and induction are mechanisms that contribute to structuring the Sargassum microbiome and its functional profiles, potentially aiding in microbiome flexibility in changing environmental contexts.

Supplementary Information

The online version contains supplementary material available at https://doi.org/10.1186/s40793-026-00925-4.

Introduction

The macroalgae genus Sargassum includes two species, S. natans and S. fluitans, that exhibit holopelagic life cycles, characterized by the absence of a benthic phase. Historically, pelagic Sargassum, herein referred to as Sargassum, was mainly restricted to the Sargasso Sea, Gulf of Mexico, Gulf Stream, Loop Current, and Caribbean Sea [1]. In oligotrophic areas of the North Atlantic, Sargassum rafts provide biodiverse hotspots for macrofauna [2–4] and fix and sequester carbon [5–7]. However, with the development of the Great Atlantic Sargassum Belt (GASB) across the Equatorial Atlantic since 2011, Sargassum biomass has increased drastically, reaching more than 20 million metric tons from the Gulf of Mexico to West Africa [8]. Sargassum super blooms have led to increased mass beaching events [9–11], with negative impacts on tourism and fishing, as well as potential health concerns and costly management [12–14]. These Sargassum blooms have also been linked to declines in seagrass beds and coral reefs due to reduced light and oxygen in nearshore water [15]. This unprecedented proliferation, along with the resulting ecosystem and socio-economic consequences, highlights the need to understand the mechanisms that have enabled Sargassum’s recent range expansion and growth.

Algal microbiomes include dense communities of epibiotic microorganisms that adhere to host surfaces by excreting extracellular polymeric substances, forming biofilms [16]. These biofilms have dynamic interactions with their algal host, often through nutritional symbiosis, in which both partners mutually benefit. Bacteria can utilize algal-derived oxygen, organic carbon [17, 18], and osmolytes such as dimethylsulfoniopropionate [19], putrescine, taurine [20], creatine, and sarcosine [21]. In return, the algal host receives vitamins and metabolites [22], such as vitamin B₁₂ [23], fixed nitrogen [24], and antibacterial compounds to exclude opportunistic pathogens [25, 26]. In many cases, bacteria are essential for algal growth, as axenic algae in the genus Ulva do not develop properly, whereas the addition of bacterial strains can induce a morphology similar to that observed in natural growth [27, 28]. Alternatively, biofilms can negatively impact their algal hosts by favoring primary and opportunistic pathogens [29, 30]. Building on this existing knowledge of biofilms in other macroalgae, recent studies have begun to investigate the potential role of the Sargassum microbiome in supporting algal growth by providing nutrient sources, thereby aiding its ability to thrive in a wide range of oceanic environments [31, 32]. While few studies are currently available, bacterial members of the Sargassum microbiome have been implicated in converting organic nitrogen [32, 33] and phosphorus [31] sources into bioavailable compounds, potentially supplying Sargassum with nutrient pools it would not otherwise be able to access. These studies indicate a central role of microbiomes in Sargassum physiology.

Across its range, Sargassum’s bacterial community is consistently dominated by the phyla Pseudomonadota, Bacteroidota, Cyanobacteriota, Actinomycetota, Planctomycetota, Chloroflexota, and Verrucomicrobiota [34]. Within Pseudomonadota, the most abundant phylum in the majority of Sargassum samples [34], the orders Rhodobacterales, Alteromonadales, and Vibrionales are frequently abundant [35–39]. However, microbiome composition varies with oceanographic context, with an increasing abundance of Vibrionales in the Caribbean and South Florida [38], in contrast to a dominance of Alphaproteobacteria, such as Rhodobacterales, in the Sargasso Sea and other regions where Vibrionales are less abundant [37]. These compositional differences are thought to reflect differences in microbiome function. Rhodobacterales are a metabolically diverse group with multiple trophic modes, contributing to vitamin production and to the synthesis and degradation of other metabolites [40]. Vibrio are heterotrophic bacteria with pathogenic strains capable of infecting humans and marine organisms; however, Vibrio species are found in a wide range of aquatic ecosystems and can be commensal and mutualistic members of microbiomes [41]. In Sargassum, Vibrio species were identified as the most abundant nitrogen-fixing bacteria [32] and have been shown to display pathogenicity-related phenotypes [42]. While these studies indicate an essential role of community composition in the Sargassum microbiome’s functions, the mechanisms that regulate community structure remain poorly understood.

In addition to bacteria, Sargassum biofilms host a diversity of bacteriophages (phages), viruses that infect prokaryotic organisms [35]. In seawater, phages can exert top-down control on bacteria by lysing their hosts and serve as a source of genetic diversity through horizontal gene transfer [43–45]. Temperate phages, which can integrate into their host’s genome as a prophage, encode genes that can impact their bacterial host’s growth, fitness, and virulence, in addition to protecting against superinfection by related phages [46–49]. Previous metagenomic analyses of S. natans in South Florida showed that the viral community of the algae is distinct from that of the surrounding seawater, and that Sargassum-associated bacteria harbor more prophages than seawater-associated bacteria [35]. Sargassum-associated phages encode genes for biofilm formation, and laboratory experiments showed that prophage induction (excision from the host genome and subsequent cell lysis) decreased biofilm production, suggesting that prophages have the capacity to modulate the growth of Sargassum biofilms [35]. However, the impact of induction on the microbial community composition and function is unknown.

Here, we hypothesized that prophage induction can cause changes in community composition and function by lysing dominant community members, opening niche space for other members of the microbiome. We tested this hypothesis in a Sargassum-derived multispecies biofilm representative of natural Sargassum microbiomes, using in vitro chemical induction and metagenomics.

Methods

Biofilm induction assay

The biofilm assay reported here was previously described in Stiffler et al. 2024, where it was initially conducted to test the inducibility of the integrated prophages, given that many prophages are domesticated and cannot be induced. Here, we quantify host and virus-specific induction levels and changes in community composition and metabolic profiles in these biofilms. Briefly, visually healthy Sargassum samples (S. fluitans and S. natans; Fig. 1a) were collected from close to shore at the University of Miami’s Rosenstiel School of Marine, Atmospheric, and Earth Science (RSMAES) campus on May 16th, 2023. About 150 g of Sargassum was collected from a single Sargassum patch in autoclaved containers, placed in a cooler, and transported back to the University of Miami’s main campus for immediate processing. The Sargassum was rinsed with DI water to remove loosely associated organisms, and the tissue was scraped with a cell scraper (Greiner Bio-One, Austria) to dislodge biofilms, which were collected into 200 mL of 100 kDa-filtered seawater. To further dislodge tightly associated microorganisms, algal tissue was placed in 250 mL tubes with 200 mL of 100 kDa-filtered seawater, vortexed for 20 min, and combined with the scraped biofilms. The biofilm suspension was filtered through a 10 µm filter mesh (Aquatic Experts, Greensboro, NC, USA) pre-treated with 10% bleach and rinsed thoroughly with DI water. The filtered Sargassum biofilm suspension was centrifuged at 8000 × g for 30 min to pellet bacterial cells, and the supernatant, containing extracellular viruses and debris, was removed. Pellets were washed with 100 mL of 100 kDa-filtered seawater, re-pelleted, and resuspended in 100 kDa-filtered seawater containing 25% glycerol before storage at −80 °C for future use. We refer to this as the Sargassum bacterial enrichment. Cell counts of the Sargassum bacterial enrichment were determined using epifluorescence microscopy [35].

Fig. 1.

Mitomycin C treatment of a Sargassum-derived biofilm decreases biofilm formation. a Pictures of representative S. fluitans and S. natans samples collected for bacterial enrichment. b Cartoon representation of the Sargassum-derived biofilm induction experiment (Created in BioRender. Silveira, C. (2026) https://BioRender.com/m84ft3c). c Change in biofilm formation (quantified via OD540) of the Sargassum bacterial enrichment grown in 10% Sargassum-supplemented seawater 12 h after treatment for MC and control samples. (Control n = 5, MC n = 5; ANOVA, p value = 1.69e−05; Tukey’s HSD, p value = 2.495e−04). d Abundance of temperate and lytic viruses recovered from biofilms in naturally occurring Sargassum (n = 213, shown as the median fractional abundance across five replicates). e Abundance of the 17 bMAGs from biofilms that had greater than 0.5 genomes per million reads in in situ Sargassum samples (mean, n = 5) compared to their relative abundance in the surrounding seawater

To determine the impact of prophage induction on Sargassum-derived biofilms, the Sargassum bacterial enrichment was grown in a 10% Sargassum-supplemented seawater solution. To prepare the Sargassum-supplemented seawater, 100 g of mixed samples S. fluitans and S. natans were ground with a mortar and pestle, blended with 1 L of 0.45 µm filtered seawater, filtered through a 100 µm mesh and a 0.45 µm Sterivex filter (MilliporeSigma, Burlington, MA, USA), and autoclaved. For the biofilm assay, 5 mL of 10% Sargassum-supplemented seawater was inoculated with the Sargassum bacterial enrichment at a final concentration of 1.4 × 10⁶ cells/mL in eight individual 6-well plates (CytoOne, USA Scientific), with one well in each plate used as a blank. Four plates were used for optical density (OD) quantification at 540 _nm, and the remaining four were harvested for sequencing. The plates were divided into two groups: a control group with no treatment and an induction group that would be treated with 1 µg/mL of Mitomycin C (herein referred to as MC; Fig. 1b). All plates were incubated at room temperature (25 °C) for 12 h.

After 12 h, two replicate plates (each with five independent replicate wells and one technical control well) from each group were subject to OD₅₄₀ quantification or biofilm harvesting (pre-treatment time point). For OD₅₄₀ quantification, the media was removed, and the wells were rinsed twice with 1 mL of 0.02-µm filtered artificial seawater. The biofilms were stained with 2 mL of 0.01% crystal violet (Eastman Organic Chemical, Rochester, NY, USA) for 10 min and subsequently washed three times with 2 mL of 0.02 µm-filtered artificial seawater. For quantification, 2 mL of 100% ethanol was added to each well, and OD₅₄₀ was measured using a Synergy H1MD Hybrid Reader (BioTek, USA).

For sequencing, biofilms were harvested by removing the supernatant, adding 500 µL of 0.02 µm-filtered artificial seawater, and scraping the biofilms off with a cell scraper. Samples were stored at −80 °C until DNA extraction. At this time, the 10% Sargassum-supplemented seawater was changed for the remaining plates, MC was added to the induction group, and plates were left to grow for an additional 12-h recovery period before OD₅₄₀ quantification and harvesting of biofilms according to the same protocol.

DNA extractions and sequencing

DNA extractions of the biofilm samples were performed with a DNeasy Blood and Tissue Kit (QIAGEN, Germantown, USA). DNA concentrations were quantified using a Qubit 2.0 Fluorometer using a High-Sensitivity dsDNA kit (Thermo Fisher Scientific, Waltham, USA). DNA library preparation and sequencing were performed by AZENTA, Inc. in South Plainfield, NJ, USA, using the NEBNext Ultra DNA Library Preparation kit according to the manufacturer’s instructions (NEB, Ipswich, MA, USA). The genomic DNA underwent fragmentation using acoustic shearing via a Covaris S220 and was subjected to end-repair and subsequent adapter ligation following the adenylation of 3′ ends. Adapter-ligated DNA was indexed and enriched by limited-cycle PCR. The resulting libraries were verified using TapeStation (Agilent Technologies, Palo Alto, CA, USA) and quantified using Qubit 2.0 Fluorometer and real-time PCR (Applied Biosystems, Carlsbad, CA, USA). Sequencing was performed on an Illumina HiSeq 4000 (Illumina, San Diego, CA, USA) in a 2 × 150 paired-end configuration. The base calling was performed using the HiSeq 4000 or equivalent Control Software, and the bcl files were converted to fastq and demultiplexed using bcl2fastq v2.17.

Quality control and co-assemblies of metagenomic data

Shotgun sequencing yielded an average of 3.2 × 10⁷ raw reads per sample (Table S1 shows the number of reads before and after quality control). The reads were filtered and trimmed using BBduk v39.06 [50] with the following parameters: left and right trimming (qtrim = rl, trimq = 30), adapter trimming of both ends with a k-mer size of 23 (ktrim = rl, k = 23, mink = 11), a hamming distance of one, and tpe/tbo parameters. Read quality was quantified with FastQC v0.11.9 [51], and reads passing QC were co-assembled based on treatment (n = 5 per treatment) using MEGAHIT v1.2.9 [52], which generated 61,583, 48,143, 20,026, and 28,699 contigs for control pre-treatment, MC pre-treatment, control post-treatment, and MC post-treatment, respectively.

Viral genome identification and curation

Viral sequences were identified among the co-assembled contigs using geNomad v1.8.0 [53]. GeNomad employs a neural network model to identify nucleotide patterns, utilizing an encoder that contains a framework based on IGLOO architecture [54]. It then predicts open reading frames using a modified version of Prodigal [55] and queries predicted proteins against a custom set of plasmid and virus-specific protein profiles using MMseqs2 [56]. Viral contigs identified using geNomad were extended using COBRA v1.2.3 [57]. Briefly, COBRA assesses breakpoints in the assembly de Bruijn graph and uses coverage and paired-read spanning to determine whether contigs should be extended. Reads were mapped at 95% identity to co-assembled contigs within each treatment group using SMALT v0.7.6 [58] for COBRA analysis. SAM (Sequence Alignment/Map) files were sorted and indexed using SAMtools v1.10 [59]. Coverage depth files were generated from the mapping files using the Joint Genome Institute’s jgi_summarize_bam_contig_depths script, and the coverage file was formatted for COBRA using their coverage.transfer.py script. COBRA was run with these input files and maxk and mink parameters. Extended contigs were analyzed through geNomad for provirus identification and updated viral taxonomic classification. GeNomad viruses were dereplicated using Virathon (GitHub–Felipehcoutinho/Virathon: Genomic Analysis of Viruses of Archaea and Bacteria). Briefly, Virathon creates viral populations using Prodigal v2.6.3 [55] to identify open reading frames, count genes per sequence, and perform an all-vs-all BLASTn [60] comparison at the gene level to calculate the percentage of shared genes and average nucleotide identity (ANI). Viral genomes with 80% shared genes and 95% ANI are considered to be within the same population, and a greedy approach is used to retain the longest genome as the population representative. The quality of the dereplicated viral sequences was then assessed using CheckV v1.0.1 [61]. Viral sequences classified as medium/high quality, longer than 10 Kb, or as a prophage with a flanking region greater than 2000 bp by geNomad or CheckV were retained, yielding 334 viral genomic sequences for further analysis. The infection strategy, temperate or lytic, for these 334 viruses was predicted using DeepPL [62], which utilizes DNABERT [63], a pre-trained bidirectional encoder representation trained on a dataset with equal representation of nucleotide sequences encoding temperate and lytic genes. Viral sequences were annotated using MetaCerberus v1.3.1[64], which employed Prodigal and all available HMMs.

Identifying prophage induction using coverage-based virus-to-host-ratio

To determine if the identified prophages were induced in MC-treated samples, coverage data of prophage regions was compared to that of bacterial flanking sequences to generate a coverage-based virus-to-host-ratio (VHR) for each prophage in each sample using PropagAtE v1.1.0 [65]. First, reads were mapped at 99% identity to the original contigs in which the prophages were identified using SMALT v0.7.6 [58]. SAM (Sequence Alignment/Map) files were sorted using SAMtools v1.10 [59]. QC paired read files, contigs with prophage and host regions, SAM, sorted BAM, and a prophage coordinates file (obtained from geNomad and CheckV) were formatted for PropagAtE v1.1.0 input. We note that PropagAtE commonly uses a VHR of 1.5 as a threshold for identifying induction. However, in complex metagenomic samples, not all bacterial hosts are lysogenized, and not all prophages will simultaneously induce, which can potentially lower this ratio (here, 63% of VHRs were below 1). Here, we leveraged our study design, in which bacterial biofilms were harvested before and after treatment with a known inducing agent, to calculate VHR pre- and post-treatment and identify induced prophages as those with a statistically significant increase in VHR, even when values were below 1.5.

Identifying putative induction of viruses using viral fractional abundances

In addition to the induction of prophages identified with clear bacterial flanking regions, we investigated the putative induction of phages that could not be assembled with bacterial flanking regions, which could be caused by multiple integration sites [66, 67] or a small fraction of the host population being lysogenized [68]. The fractional abundance of all identified viral sequences was calculated by mapping the reads from each sample to the biofilm viral database at 95% identity using SMALT v0.7.6. SAM files were sorted and indexed with SAMtools v1.10. Fractional abundances (f_(i)) were calculated by taking into account contig length using the equation below [69].

1

where (r_(i)) is the number of reads mapped to each viral genome sequence obtained with SAMtools v1.10, L(mean) corresponds to the mean viral genome length (bp) in the database, L(i) denotes the length of the viral genome, and (T(j)) represents the number of viral mapped reads in each metagenome sample.

Binning and analysis of bacterial metagenome-assembled genomes

Bacterial metagenome-assembled genomes (bMAGs) were constructed by mapping QC reads to co-assembled contigs using Bowtie2 v2.3.5 [70]. Using SAMtools v1.10 [59], the SAM files were compressed, sorted, and indexed according to the binning algorithm specifications. Three binning programs were employed to build bMAGs: CONCOCT v1.0 [71], MetaBAT2 v2.12.1 [72], and MaxBin2 v2.2.6 [73]. All three binners combine contig coverage and sequence composition to cluster contigs into genomes. The resulting bins were refined, reassembled, and quantified using the bin_refinement, reassemble_bin, and quant_bins modules of MetaWRAP v1.3.2 [74]. One replicate sample from the control post-treatment group contained a bMAG that matched Salmonella, a common contaminant from human material. As this bMAG was assembled with sequences coming from a single replicate, and NMDS analysis indicated it was an outlier (Fig. S1), this bMAG was treated as a contaminant and removed from the study. The remaining bMAG relative abundances in the replicate sample where this bMAG originated were not distinguishable from the other replicates, so this metagenome was maintained for subsequent analyses after the removal of this bMAG (Fig. S1). Bin abundance was obtained through MetaWRAP in genomes per million reads using a modified version of SALMON v0.13.1 [75]. This abundance estimate considers a length-weighted bMAG coverage of individual contigs relative to the number of reads per sample. CheckM2 v1.0.2 [76] estimated bMAG completion and contamination. Taxonomy was assigned to bMAGs using GTDB-tk v2.1.1 [77, 78]. KEGG metabolism pathways were assigned to bMAGs using the anvi-run-kegg-kofams and anvi-estimate-metabolism modules of Anvi’o v8 with default settings [79]. Only modules that had a pathwise completeness of 70% or greater were retained for further metabolism analysis. bMAGs were clustered based on average nucleotide identity (ANI) using anvi-dereplicate-genomes module [79] with fastANI v1.32 using the ‘min-full-percent-identity’ and ‘similarity-threshold’ flags at 97% ANI [80]. The Newick tree file was visualized and annotated using iTOL v7 [81].

Host prediction

Prophages were linked to bMAGs by identifying contigs containing prophages within the bin. One bMAG (CT-PostInd-bin.5) had a viral sequence (CT-PostInd-BF1_k127_431539) binned within it, but with no flanking regions. This viral sequence was part of the same viral population as an identified prophage (MC-PostInd-BF1_k127_567814; prophage.8). We assume that the prophage identified within the MC post-treatment samples would not have been binned with its host due to induction-related coverage shifts. Therefore, BLASTn [60] was used to align the bacterial flanking region of the prophage to the non-prophage viral sequence within bMAG CT-PostInd-bin.5. The last 648 bp of the viral sequence matched the first 648 bp of the bacterial flank at 99.3% identity; thus, we deemed this a prophage-bMAG link (Table S2).

Bacterial flanks surrounding prophages (> 2000 bp) were extracted and compared to the GTDB GCF representative genomes (accessed September 2025) [78] using BLASTn [60]. Only matches that had greater than 85% identity and an e-value less than 1e-10 were considered (Data S7). For a genus to be assigned to a bacterial flank, 50% of the BLAST hits had to be assigned to a single genus. Additionally, to predict the viral host genus including bMAGs assembled from this study, we used iPHoP v1.3.3 [82] by adding the biofilm induction bMAGs to the ‘Aug_2023_pub’ host database using the GTDB-tk command gtdbtk de_novo_wf with the outgroup set to Actinobacteria. The results were then fed into the iPHoP command add_to_db and the new database was used with default ‘predict’ parameters; only matches with a confidence score of 90% were retained for analysis. All viral contigs identified by geNomad were removed from bMAGs before adding them to the ‘Aug_2023_pub’ host database. iPHoP combines six different methods to predict phage-host pairs. Using neural networks and random forest classifiers, it computes hit scores for potential hosts by combining the host-based prediction tools BLAST [60], CRISPR, VirHostMatcher [83], WIsH [84], and PHP [85]. It then couples this with the phage-based prediction tool RaFAH [86] to assign genus or genome-resolved phage-host pairs. These iPHoP host matches are refered to as iPHoP host links. If methods predicted different hosts, both matches were discarded from the results but still recorded in supplemental datasheets (Table S3 & Data S8).

Analysis of prophage and host-encoded AphA transcriptional regulator

One prophage (prophage.8) that was linked to a bMAG encoded a transcriptional regulator (phrog_13512) with homology to AphA (PadR transcriptional regulator family), a protein involved in biofilm formation and quorum sensing in several Vibrio species [87]. Interestingly, the AphA homolog was also identified as a potential biofilm gene encoded by a phage within the S. natans microbiome [35]. Therefore, we investigated the similarity of these phage and host-encoded proteins at the sequence and structural levels. MetaCerberus v1.3.1 [64] was used to annotate the bMAG host of the prophage.8 (CT-PostInd-bin.5; Vibrio sp.) and the viral sequence identified in S. natans metagenomes from our previous study [35]. CT-PostInd-bin.5 and one viral sequence (sarg_contig.k127_668789) encoded AphA homologs. A BLASTp [60] against the RefSeq non-redundant protein database [88] was used to identify homologous proteins to the Vibrio host and prophage.8 proteins. The 25 matches with the highest identity and coverage were selected for each, dereplicated at 99% identity using CD-HIT v4.8.1 [89], and compared with the two AphA homolog amino acid sequences from this study, the viral sequence from the S. natans metagenomes, and three of the viral encoded transcriptional regulators (phrog_13512) from the Prokaryotic Virus Remote Homologous Groups (PHROG) database for a total of 34 amino acid sequences. The proteins were aligned with MAFFT v7.511 [90], and the alignment file was then used to generate a maximum-likelihood tree with IQ-TREE v3.0 [91] with the following flags -st DNA -m TEST -bb 1000 -alrt 1000. The -m TEST flag employed standard model selection followed by tree inference, using a previously described model-finder analysis [92], and an ultrafast bootstrap method (UFBoot) [93]. The tree file was annotated in iTOL v7 [81]. To determine if the RefSeq AphA homologs were in putative prophage regions, the genomes associated with the proteins were downloaded and annotated with MetaCerberus v1.3.1 [64]; ORF start and end positions were confirmed, and the surrounding annotations were screened for viral genes (Data S14). The tridimensional structure of the AphA proteins encoded by prophage.8, CT-PostInd-bin.5, and their closest matches that have been isolated, V. scophthalmi (WP.3345556641), and Vibrio phage vB_VpaS_MAR10 (YP_007111862.1) [94] were predicted with AlphaFold 3[95]. Prophage.8’s AphA homolog was assessed as a homodimer using AlphaFold3 [95]. Predicted structures were aligned using the matchmaker command of ChimeraX v.1.10.1 [96].

In situ abundances of biofilm induction viruses and bMAGs

The presence of biofilm induction viruses and bMAGs in S. natans and seawater samples collected in South Florida was determined by mapping QC reads from S. natans and seawater to the viral sequences identified in this study at 95% identity using SMALT v0.7.6 [58]. SAM files were sorted and indexed using SAMtools v1.10 [59], and coverage was obtained using BBMap’s ‘pileup.sh’ script [50]. For a virus to be labeled as present in the S. natans and seawater samples, it had to have greater than 75% coverage. Viral fractional abundance was calculated using the previously described equation. Relative abundance of the 28 biofilm induction bMAGs in the same S. natans and seawater samples was quantified using the quant_bins module of MetaWRAP v1.3.2 [74]. bMAGs with fewer than 0.5 genomes per million reads in S. natans samples were considered absent.

Analysis of bacterial contigs

All co-assembled contigs were combined and dereplicated at 97% ANI using CD-HIT v4.8.1 [97]. Bacterial contigs were identified by using Kaiju v1.10.1 [98] with the proGenomes database v3 [99]. QC reads were mapped to the dereplicated bacterial contigs using Bowtie2 v2.3.5 [70]. Using SAMtools v1.10 [59], the SAM files were compressed, sorted, and the number of reads mapped to each contig extracted. Coverage information for the mapping was obtained using BBMap’s ‘pileup.sh’ script [50]. The fractional abundances (f_(i)) were calculated by accounting for contig length, as in Eq. 1. Alpha diversity (Shannon index), richness, and evenness of contigs across the different treatment groups were assessed using Kaiju v1.10.1 [98] to assign taxonomy to the contig at the genus level; when the genus of a contig was unknown, the lowest ranking taxonomy was used in its place. KEGG metabolism pathways were assigned to dereplicated bacterial contigs using the anvi-run-kegg-kofams and anvi-estimate-metabolism modules of Anvi’o v8 with default settings [79]. Only modules with a pathwise completeness of 70% or greater were retained for further metabolic analysis, and contigs were considered present in a sample only if they had 75% coverage.

Figures and statistics

All figures and statistical analyses were generated using R v4.3.2, and aesthetic edits were made in Adobe Illustrator 2025. Data spreadsheets were read using readxl v1.4.5 [100], and data sheets were re-formatted when needed with tidyr v1.3.1[101], dplyr v2.5.0 [102], tidyverse v2.0.0 [103], and stringr v1.5.1 [104]. The non-metric multidimensional scaling (NMDS), permutational multivariate analysis of variance (PERMANOVA), and similarity percentage (SIMPER) analyses, and Shannon diversity indices were conducted using vegan v.2.7-1 [105]. NMDS analyses were computed using Bray–Curtis dissimilarity matrices with 9999 permutations, and pairwise PERMANOVA analyses were performed using pairwiseAdonis v0.4.1 [106]. Bar plots, boxplots, pie charts, and NMDS plots were generated with ggplot2 v3.4.1 [107], ggExtra v0.10.0 [108], and ggpubr v0.6.0 [109]. Heatmaps were created with pheatmap v1.0.12 [110], viral genome maps with genoPlotR v0.8.1 [111], and bMAG percent nucleotide identity, and the protein phylogeny trees were visualized and edited with iTOL v7 [81].

Results

Mitomycin C (MC) treatment of the Sargassum-derived biofilm communities reduced biofilm growth, as measured by the change in OD₅₄₀ between pre- and post-treatment, where a significantly lower rate of biofilm formation was observed in the MC-treated group compared to the untreated control (ANOVA, p value = 1.69e−05; Tukey’s HSD, p value = 2.495e−04; Fig. 1c). Binning of contigs resulted in 28 bacterial metagenome-assembled genomes (bMAGs) that were > 50% complete (mean = 80.49%, SD = 14.82%) and < 10% contaminated (mean = 2.39%, SD = 2.12%; Data S1). We identified 1303 putative viral genomes/genome fragments categorized as medium or high quality, longer than 10 Kb, or labeled as a prophage with a bacterial flanking region longer than 2000 bp. These were dereplicated into 334 viral populations (herein referred to as viruses; Data S2).

Sargassum-derived biofilms are representative of naturally occurring Sargassum microbiomes

To assess environmental relevance, S. natans and seawater metagenomic reads collected off the coast of South Florida [35] were mapped to the viruses and bMAGs from this study. Of the 334 viruses derived from the biofilm assay, 64% were present in S. natans metagenomes (Fig. S2), but none were identified in surrounding seawater metagenomic reads. These overlapping viruses consisted of two prophages, 25 temperate phages, and 186 putatively lytic viruses (Fig. S2). The majority of these 213 viruses were unclassified (79.81%), with the remaining viruses belonging to Pokkesviricetes (10.33%), Revtraviricetes (5.63%), Caudoviricetes (3.29%), and Faserviricetes (0.94%) (Fig. S2). The 27 temperate viruses (including the two prophages) were more abundant in S. natans compared to the lytic viruses (Fig. 1d; Wilcoxon rank sum: p-value = 0.03). The three most abundant biofilm viruses in S. natans environmental samples included two temperate phages (MC-PostInd-k127_604294 and prophage.8) and one lytic virus (CT-PreInd-k127_631518) (Data S3). Within environmental S. natans samples, these three viruses accounted for 7.5%, 3.2%, and 4.2% of the biofilm assay viral community, respectively.

All bMAGs from the biofilm induction experiment together had, on average, 138.86 genomes per million reads in S. natans samples (Fig. 1e). An analysis of the relative abundance of these bMAGs in the S. natans samples demonstrated that 17 of the 28 bMAGs were present in environmental Sargassum samples. The most abundant bMAG, classified as Vibrio alginolyticus, had an average of 19.46 genomes per million reads, accounting for 10% of the total number of genomes (Fig. 1e; Data S4). Only nine of the 28 bMAGs were present in seawater, indicating that the in vitro biofilm was less representative of the seawater microbial community, as expected. Seven bMAGs were found in both S. natans and seawater samples: one Celeribacter halophilus, three Alteromonas sp., and three Vibrio alginolyticus. The ten bMAGs unique to S. natans included two Vibrio pelagius, four Vibrio sp., and four Pseudoalteromonas sp.

Mitomycin C induces Vibrio prophages and affects viral abundance

Of the 334 viral sequences in this study, 61% could not be taxonomically classified. The remaining viruses belong to the following classes: Caudoviricetes (25.1%), Pokkesviricetes (7.2%), Revtraviricetes (4.5%), Tectiliviricetes; Corticoviridae (0.9%), Faserviricetes; Inoviridae (0.6%), and Megaviricetes (0.3%) (Fig. 2a). The abundances of these viruses (Data S5) differed significantly between MC and control pre- and post-treatment (Pairwise PERMANOVA: p-values = 0.0067 and 0.0096 for MC pre vs. post treatment and control pre- vs. post-treatment, respectively) and post-treatment groups (Pairwise PERMANOVA: p-value = 0.0066 control post vs MC post-treatment). Pre-treatment groups did not differ (Pairwise PERMANOVA: p-value = 0.15 control pre- vs. MC pre-treatment) (Fig. 2b). SIMPER analysis identified 109 viruses driving the differences in viral abundance between MC and control samples; 10 of these viruses were identified as prophages. The change in fractional abundance in MC and control samples was calculated for all of the 109 viruses to identify potential viral induction candidates. Twenty-one viruses displayed a positive shift in abundance that was greater in MC samples than in controls (Fig. 2c). While none of these included the 10 viruses identified as integrated prophages, nine of these induction candidates were categorized as temperate phages encoding lysogeny-related genes (Fig. 2c): integrase, antitermination protein Ant, regulatory protein Rha, recombination protein Bet, recombination protein RecE, Cl repressor, antitermination protein Q, and replication protein O (Fig. 2d). This number is likely an underestimate, as many temperate viral genomes lack identifiable integration genes [68]. Eighteen of these viruses were classified as Caudoviricetes, one as Corticoviridae, one as Inoviridae, and the remaining virus was unclassified. Ten of the 21 induction candidates had a predicted host. Eight viruses were predicted to infect Vibrio, one Aliivibrio, and one Cobetia. The Inoviridae phage was predicted to infect Vibrio and encoded zonula occludens toxin (Zot) (Data S6). Viral genome k127_317508, which is a temperate phage predicted to infect Aliivibrio and experienced the second greatest change in abundance between groups, encoded a chitinase (Fig. 2d; Data S6).

Fig. 2.

Viral fractional abundance changes upon induction, and temperate phages display a significant increase. a Taxonomy of all viruses identified in the biofilm (n = 334 after quality thresholds described in the methods); unclassified (dark purple), Caudovircetes (blue), Pokkesviricetes (orange), Revtraviricetes (light green), Tectiliviricetes (light purple), Faserviricetes (light blue), and Megaviricetes (light orange). b Non-metric Multidimensional Scaling (NMDS) of the viral fractional abundance across sample types, CT post-treatment (n = 5, orange), CT pre-treatment (n = 5, light blue), MC post-treatment (n = 5, dark purple), and MC pre-treatment (n = 5, light green). Each dot represents a replicate. The ellipses indicate the 95% confidence interval around each sample type. c Median change in viral fractional abundance for the 21 viral induction candidates (SIMPER: p < 0.05). The colors represent the different treatments: control (orange) and MC (blue). The error bars indicate standard error. Temperate (white squares) and lytic (black squares) viruses are annotated under the infection (Inf) column. Taxonomy (Tax) is listed as unclassified (circle), Inoviridae (cross), Corticoviridae (diamond), Schitoviridae (triangle), and unclassified Caudoviricetes (star). iPHoP host matches (Host) are listed as no match (light purple), Vibrio (light blue), Aliivibrio (light green), and Cobetia (light orange). d Genome plots for the four viruses that had the most significant change in fractional abundance between treatment groups. Annotations for the plots are available in Data S6

Of the 334 viruses, 11 were identified as prophages, with an average length of 40,502 bp and an average flanking region length of 12,550 bp (Table S3). Eight prophages belonged to the class Caudoviricetes, one in the family Autographiviridae, one in Kyanoviridae, and one in Corticoviridae (Fig. 3a). Eight prophages were predicted to infect the genus Vibrio (Table S3; Data S7; Data S8). To directly assess induction of confirmed prophages, coverage-based virus-to-host ratios (VHRs) were calculated across treatment groups (Data S9). VHRs were similar between pre- and post-treatment controls (Pairwise PERMANOVA: p-value = 0.62), as expected, and differed between MC pre-treatment samples compared to post-treatment samples (Fig. 3b; Pairwise PERMANOVA: p-value = 0.015). SIMPER analysis identified four prophages (prophage.1, prophage.3, prophage.4, and prophage.8) that accounted for most of the difference in VHR between the MC pre- and post-treatment samples (Fig. 3b). Two of these prophages also differed between the control and MC post-treatment samples, although a comparison accounting for all prophages was not significant, indicating heterogeneity in prophage response to chemical induction (Pairwise PERMANOVA: p-value = 0.35). The four prophages identified in the MC samples were taxonomically classified as Caudoviricetes (two without family-level classification), one as Autographiviridae, and one as Corticoviridae. All were predicted to infect bacteria in the genus Vibrio (Fig. 3c). Two prophages, whose VHRs drove differences in both comparisons, encoded genes for host evasion, viral defense, and a transcriptional regulator belonging to the PadR family. Prophage.8 encoded hdf and ulx-like genes, which are a part of antirestriction systems to evade host defenses [112] and a transcriptional regulator with homology to aphA, a regulator of quorum sensing and biofilm formation in multiple Vibrio species in the PadR transcriptional regulator family [113, 114] (Fig. 3c; Data S6). Furthermore, prophage.10 encoded a gene with similarity to terB, which is involved in tellurite resistance and also plays roles in host antiviral defense systems [115].

Fig. 3.

Mitomycin C induces pelagic Sargassum biofilm-associated prophages. a Taxonomy of identified prophages (n = 11); unclassified Caudovircetes (blue), Autographiviridae (dark purple), Kyanoviridae (light green), and Corticoviridae (orange). b Virus-to-host ratio (VHR) for each prophage, normalized by prophage with a Z-score (median, n = 5 for each sample type). Prophages marked with a square (MC pre- vs. MC post-treatment), circle (CT post- vs. MC post-treatment), or star (both comparisons) were identified by SIMPER analysis as significantly contributing to the differences between samples (p < 0.05). c Genome plots of the four prophages contributing to differences in the MC pre- and post-treatment samples. Annotations for the plots are available in Data S6

Biofilm community composition shifts in response to Mitomycin C prophage induction

To investigate the impact of MC treatment and prophage induction on biofilm composition 12 h after treatment, we analyzed the abundances and genomic profiles of the 28 bMAGs identified in our dataset (Fig. 4). Note that because the changes in bacterial community composition reported in this section cannot be fully decoupled from the DNA damage toxicity of MC, changes are attributed to a combination of prophage induction and MC treatment. The bMAGs comprised 17 Vibrio, four Pseudoalteromonas, three Cobetia, three Alteromonas, and one Celeribacter. All of these are a part of common bacterial families found on Sargassum (Vibrionaceae, Halomonadaceae, Alteromonadaceae, and Rhodobacteraceae) [35–37]. Six bMAGs could be classified to the species level, with three Vibrio alginolyticus, two Vibrio pelagius, and one Celeribacter halophilus (Data S1). The relative abundance of the bMAGs, in genomes per million reads, differed significantly between sample types (Fig. 4a; PERMANOVA: p-value = 1e−04). The two pre-treatment groups did not differ (Pairwise PERMANOVA: p-value = 0.22), the post-treatment groups varied from their pre-treatment counterparts (Pairwise PERMANOVA: p-value = 0.0085 and 0.0078 for control and MC, respectively), and the variation between the two post-treatment groups was significant (Pairwise PERMANOVA: p-value = 0.025). SIMPER analysis identified 18 bMAGs that were differentially abundant when comparing the MC samples pre- and post-treatment, and between control and MC post-treatment (Fig. S3 depicts SIMPER p-values for all comparisons of the 18 significant bMAGs; Data S1). Among the 18 bMAGs driving differences between treatment groups, 13 Vibrio bMAGs decreased in abundance in the MC-treated samples (Fig. 4b). Pseudoalteromonas and Cobetia bMAGs became more abundant in the MC-treated samples. One Alteromonas bMAG did not display a change in growth in MC samples but decreased in the control (Fig. 4b and Fig. S3). An analysis including all bacterial contigs demonstrated that both post-treatment groups were more diverse than their pre-treatment groups (Wilcoxon rank-sum tests: p values = 0.01587 and 0.03175 for the control and MC-treated groups, respectively); the bacterial community diversity of the post-treatment groups did not differ (Wilcoxon rank-sum tests: p value = 0.4206; Fig. S8), demonstrating that even with changes in community composition, alpha diversity is not affected after 12 h. Among the bMAGs decreasing in abundance with induction, three had prophage links; two Vibrio sp. bMAGs (CT-PostInd-bin.5 and CT-PostInd-bin.8) and one Vibrio alginolyticus (MC-PreInd-bin.2) encoded prophage.8, prophage.3, and prophage.9, respectively. Five viral induction candidates, based on changes in relative abundance in the different treatment groups, were predicted to infect 92% of the 13 Vibrio bMAGs that decreased upon induction (Fig. 4b; Data S8). None of the 334 viruses were predicted to infect Alteromonas, Pseudoalteromonas, Cobetia, or Celeribacter assembled bMAGs.

Fig. 4.

Bacterial community composition shifts upon prophage induction. a NMDS analysis of bMAGs across Control (CT) post-treatment (n = 5, light orange), CT pre-treatment (n = 5, light blue), Mitomycin C (MC) post-treatment (n = 5, dark purple), and MC pre-treatment (n = 5, light green). Each dot represents a replicate. The genus of each bMAG is defined by the shapes: Alteromonas (square), Celeribacter (triangle), Cobetia (upside-down triangle), Pseudoalteromonas (diamond), and Vibrio (cross). The ellipses indicate the 95% confidence interval around each sample type. b Median change in abundance for the 18 bMAGs driving differences between the treatments (SIMPER: p < 0.05). The bar colors represent the different treatments: control (orange) and MC (blue), error bars indicate standard error. Prophages that were identified in bMAGs, are indicated by black circles. Viral links to bMAGs predicted by iPHoP are separated into those that are viral induction candidates (light orange; Fig. 2c) and those with non-induction candidates (light green). iPHoP methods and confidence scores are listed in Data S8

Prophage.8 encodes a gene with homology to the transcriptional regulator AphA encoded by the host. AphA belongs to the PadR superfamily, which is characterized by an N-terminal winged helix-turn-helix motif and a C-terminal dimerization domain. In many Vibrio species, it has central roles in quorum sensing, biofilm formation, and virulence [109]. CT-PostInd-bin.5’s AphA amino acid sequence was most similar to that of Vibrio scophthalmi (100% identity with 96% coverage; Fig. 5a; Data S10). The prophage amino acid sequence was most similar to an AphA homolog identified in V. alginolyticus (97.48% identity with 99% coverage; Fig. 5a; Data S10). This prophage-encoded protein was also observed in one viral sequence identified in the in situ microbiome of S. natans (sarg_contig.k127_668789), with 99.34% identity and 100% coverage (Fig. 5a; Table S4). An AphA protein phylogeny tree, consisting of AphA proteins from this analysis and their closest matches in the RefSeq database, had two distinct clades that separated from each other in 77.9% of the bootstrap iterations, where prophage.8 and its host were in separate clusters (Fig. 5a). Based on this observation, the surrounding genes in the genomes associated with AphA homolog reference sequences were analyzed for flanking viral gene annotations. This identified a group of putative prophage-encoded AphA homologs (Fig. 5a). However, one bacterial-encoded AphA homolog (EPP23146.1 Vibrio fluvialis) was not in a putative prophage region but separated from the other bacterial AphA proteins with the viral versions, suggesting that the viral versions originate from a previous Vibrio host, as the majority of putative prophages found did not originate from V. fluvialis. Both phage-encoded aphA (phrog_13512) genes from this study are flanked by viral genes (Ref-like RecA filament endonuclease, uncharacterized viral protein, and a probable phage tail protein) (Fig. S4; Data S6). The AphA homologs of prophage.8 and its host shared 34.78% identity (Table S4) at the amino acid level. However, the protein structures displayed high structural similarity between the core domains, but there was structural variation in other regions, with a similarity score of 302.3 and an RMSD of 1.216 Å over 82 pruned atoms (across all 156 pairs: 4.581) (Fig. 5b; Fig. S5). CT-PostInd-bin.5’s AphA was additionally aligned with the closest reference match originating from V. scophthalmi and the AphA from V. cholerae O395, which has been validated experimentally, and crystal structures are available [116, 117]. The protein sequence-alignment scores for CT-PostInd-bin.5 to V. scophthalmi and V. cholerae O395 were 946.2 and 853.6, respectively. The RMSD values between 179 pruned atom pairs were 0.244 Å (across all 179 pairs: 0.244), and 171 pruned atom pairs were 0.743 Å (across all 179 pairs: 0.950) for the V. scophthalmi and V. cholerae O395 comparison, respectively. Thus, indicating that these proteins have high structural similarity across the entire structure (Fig. S6a and b). Prophage.8’s AphA homolog was aligned to the closest version of the protein encoded by a temperate virus that has been isolated, Vibrio phage VB_VpaS_MAR10, which infects Vibrio parahaemolyticus [94]. The viral protein homologs had structural similarity between core domains but had the lowest alignment score across the alignments in this study, with a sequence alignment score of 290.5, and the RMSD between 77 pruned atom pairs was 1.388 Å; across all 152 pairs: 3.376 (Fig. S6c). Finally, the AphA homolog from prophage.8 was computationally tested for dimerization, as the PadR family of transcriptional regulators, including AphA, functions as a dimer. The C-terminal dimerization domain, consisting of three alpha helices, like other subfamily-1 PadR transcriptional regulators [118, 119], was predicted to homodimerize with an ipTM score of 0.81(Fig. S7)., however, when the whole protein was tested for homodimerization the ipTM score was significantly lower (0.51), likely because the N-terminal winged helix-turn-helix motif binds target DNA and does not dimerize (Fig. S7).

Fig. 5.

Prophage- and host-encoded AphA homologs are divergent but have structural similarity. a Maximum likelihood tree of 34 AphA homolog amino acid sequences, two from this study (indicated by stars), one encoded by a viral sequence from S. natans, RefSeq matches to both proteins dereplicated at 99% identity (n = 28), and three Vibrio phages. Purple and light green boxes represent AphA homologs encoded by bacteria and viruses, respectively. Light blue indicates that the AphA homolog is located within a putative prophage region. b Aligned AphA homolog structures for prophage.8 (blue) and CT-PostInd-bin.5 (orange). Protein structures are denoted by a star on the phylogenetic tree with the colors above. The sequence alignment score between the two proteins was 302.3, and the RMSD between 82 pruned atom pairs was 1.216 Å

Chemical prophage induction alters bacterial biofilm metabolic potential

Bacterial MAGs encoded 132 KEGG modules with a pathwise module completion of greater than 70% in 26 submodules (Fig. 6a; Data S11). Eight submodules were present in all bMAGs (Fig. 6a). To investigate if MC treatment and prophage induction shifted the microbiome metabolisms, the relative abundance of each bMAG that encoded a metabolism module was summed within each replicate as an indicator of module abundance. SIMPER analysis identified 17 modules that differed between the MC and control post-treatment groups (Fig. 6b) and 108 modules between the MC pre- and post-treatment. Here, we focus on the differences between the MC and control post-treatment samples, as they provide insight into which metabolic pathways were depleted by MC treatment. No modules were enriched and the 17 depleted modules were categorized into 10 different submodules: ATP synthesis, nitrogen metabolism, cofactor and vitamin metabolism, metabolic capacity, serine and threonine metabolism, polyamide metabolism, other aromatic metabolism, histidine metabolism, lipid metabolism, and other amino acid metabolism. The modules that decreased in MC treated compared to the control post induction groups were GABA biosynthesis, acetogen, ectoine degradation, cobalamin biosynthesis, thiamine biosynthesis, dissimilatory nitrate reduction, histidine degradation, fumarate reductase (prokaryotes), sulfate-sulfur assimilation, hydroxyproline degradation, phylloquinone biosynthesis, C1-unit interconversion (prokaryotes), cytochrome-c oxidase (cbb3-type), phosphatidylethanolamine biosynthesis, D-galacturonate degradation (bacteria), cytochrome-c oxidase (prokaryotes), and cytochrome bc1 complex respiratory subunit (Fig. 6b).

Fig. 6.

Bacterial metabolism profile is altered post-treatment. a Average Nucleotide Identity (ANI) tree for all 28 identified bMAGs. bMAG completeness is represented by the purple to white color gradient, and taxonomy corresponds to the color of the tree node. Black circles indicate KEGG metabolism submodules that have at least one module with a pathwise completion greater than 70%. b KEGG metabolism modules that differed between the MC and control post-treatment samples, as determined by SIMPER (p < 0.05). The relative abundance of each bMAG encoding the module was summed within a replicate and normalized to the module with a Z-score

Acetogenesis, the process of producing acetate via the reduction of CO₂ anaerobically via the Wood–Ljungdahl pathway, is not a known functional pathway of bacteria identified in this study. However, many, including vibrios, produce acetate via fermentation using two key enzymes in the acetogen KEGG module: acetate kinase (AckA) and phosphotransacetylase (Pta) [120]. Cobalamin (vitamin B12) biosynthesis was another module that decreased post-treatment; however, Celeribacter halophilus (CT-PreInd-bin.7) was the only bMAG that encoded genes for additional cobalamin biosynthesis steps, and its relative abundance was not impacted by induction. Genes for six different pathogenicity modules with pathwise completion rates of 4–33% were present: ‘Vibrio cholerae pathogenicity signature, cholera toxins’ genes were identified in seven Vibrio bMAGs and ranged from 16 to 33% pathwise completion. Two Vibrio sp. and one Vibrio pelagius bMAGs encoded zonula occludens toxin (Zot), a prophage-encoded enterotoxin [121], and one Vibrio pelagius and two Vibrio alginolyticus bMAGs encoded thermolabile hemolysin (tlh) (Data S11). One Vibrio alginolyticus bMAG encoded both zot and tlh.

The bMAGs metabolism analysis represents only the dominant bacterial community for which genomes could be assembled. We complemented this analysis with KEGG module data for all bacterial contigs that represent the broader community (Data S12; Data S13). KEGG module profiles differed among treatment groups (PERMANOVA: p value = 1e−04) and specifically between the two post-treatment groups (Pairwise PERMANOVA: p = 0.0179), where 51 KEGG modules accounted for these changes (SIMPER, p < 0.05). Forty-six modules were more abundant in the MC post-treatment samples than in the controls post-treatment (Fig. S9). Within these 46 modules, 19.57% are involved in cofactor and vitamin metabolism.

Discussion

Biofilm formation is a common capacity of microbiomes, and the ability to grow as biofilms influences community composition [122–124]. Prophage induction can significantly impact biofilm development and interactions with an eukaryotic host [125]. Yet, the impacts of prophages on marine biofilms remain mostly unknown, despite their widespread presence and abundance [126–128]. In the experiments described here, over 60% of the bacterial and viral community members were also present and abundant in environmental samples of S. natans from South Florida (Fig. 1d and e), including several Vibrio spp. and Vibrio prophages. Like other eukaryotic hosts, Sargassum likely exerts some control over the bacterial community that grows epiphytically, although host control has been more extensively studied in animals [129]. We speculate that, as host control was lost in our in vitro system, where bacterial communities grew with Sargassum supplementation but not live Sargassum, within-microbiome interactions, such as competition and viral interactions, become relatively more relevant for overall community assembly. Despite loss of host control, the community retained abundant members, demonstrating that our experiments involve ecologically relevant taxa.

Prophages can have diverse effects on biofilm formation. The marine Sulfitobacter sp. CB2047, isolated from microalgae phycosphere, switches from predominantly planktonic to predominantly biofilm growth depending on the identity of its prophages [49]. Several studies in non-marine systems have demonstrated that cell lysis via spontaneous prophage induction enhances biofilm formation of single-species biofilms by releasing extracellular DNA that contributes to the biofilm polymer composition [130–133]. Conversely, prophage induction has been shown to cause biofilm decay in Geobacter sulfurreducens [134]. The data described here shows that the chemical prophage induction of a Sargassum-derived multispecies biofilm by MC resulted in a decrease in biofilm formation compared to the control 12 h after the induction (Fig. 1c) [35]. This decrease is likely a result of prophage-induced cell lysis, which was shown to occur predominantly in Vibrio (Fig. 4). Direct MC toxicity via DNA damage also likely contributed to the overall decrease in biofilm formation. A washout step would have helped minimize the effects of Mitomycin C toxicity. However, the centrifugation step necessary to separate cells from the Mitomycin C would have either mixed planktonic and biofilm cells if it had been performed without first removing the supernatant, or disrupted the biofilm and resulted in the loss of free viral particles embedded in the biofilm matrix if it included a scrape-and-wash step. Therefore, MC was maintained in the wells, and its toxicity presumably contributed to the observations discussed below. The extent of this effect is probably limited, given that there was overall positive biofilm formation in the MC-treated samples (i.e., cells were alive and dividing), only less than in controls (Fig. 1).

Chemical prophage induction modulates bacterial community structure

Among the 11 Sargassum-derived prophages identified in this dataset, seven were predicted to infect bacteria in the genus Vibrio, including four prophages that drove most of the differences in VHR between the control and MC post-treatment groups (Fig. 3b). These results suggest that in Vibrio-dominated Sargassum microbiomes, as is the case for Caribbean and Florida samples [35, 36, 38], phage integration and induction may have an impact on community composition and dynamics. Prophage induction with MC has been shown to introduce microbiome shifts in other systems, such as soil, resulting in an increase in bacterial diversity within 24 h [135]. While bacterial community diversity in our study increased after twelve hours, this was true for both control and MC-treated samples (Fig. S8) and could not be attributed to induction.

Vibrio prophages identified here were classified as Corticoviridae, a group of tailless double-stranded DNA phages present in many marine Vibrio genomes [136], and tailed phages in the Caudoviricetes group. One Inoviridae phage, a filamentous virus also common in vibrios [136], displayed an increase in abundance in the MC-treated group (Fig. 2c), but could not be confirmed as a prophage due to the lack of bacterial flanking regions or an episomal replication style [137]. The presence of these broadly distributed prophage groups in our inducible dataset from Sargassum further supports the idea that phage induction of the genus Vibrio may be a potential mechanism regulating Vibrio abundances in Sargassum. This aligns with our observation of decreased Vibrio abundance after MC treatment (Fig. 4b). A previous study showed that Sargassum prophages can be induced not only with MC but also by ultraviolet light [35], albeit to a lesser extent. As Sargassum is exposed to high ultraviolet radiation due to its pelagic lifestyle, Vibrio prophage induction is a plausible phenomenon in its natural setting.

The decrease in Vibrio bMAG abundance due to MC treatment was accompanied by an increase in the relative abundance of two Pseudoalteromonas sp. and one Cobetia sp. (and one Vibrio sp. to a lesser extent; Fig. 4b) and the persistence of one Alteromonas sp. that displayed a decrease in relative abundance in the control. This suggests that, in the short term, prophage induction, coupled with MC effects, decreases Vibrio spp. abundance while allowing other bacteria to persist or start growing. In our dataset, the only prophages linked to hosts through various methods were Vibrio-infecting prophages, and among non-prophage viral induction candidates, 8 were predicted to infect Vibrio versus 1 predicted to infect Cobetia, and none were predicted to infect Pseudoalteromonas or Alteromonas (Fig. 2c). This suggests that Vibrio bMAGs harbor a higher burden of inducible prophages, or that the expanding genera possess lower lysogeny frequencies or cryptic prophages resistant to MC. The selective Vibrio induction may contribute to Sargassum microbiome composition across its range: metataxonomic studies have reported that when Gammaproteobacteria comprise the majority of the Sargassum microbiome, two distinct profiles are observed: one with a greater proportion of Vibrio and another with a greater proportion of Alteromonas [37]. In this scenario, prophage induction could influence bacterial competition between these groups, as observed in other systems [138, 139]. Abiotic-driven mass prophage induction by ultraviolet light may also impact the alternating taxonomic dominance in Sargassum microbiomes across its geographical range. This dynamic could occur alongside community changes driven by algae senescence, which favor Vibrio [140]. Future studies analyzing prophage induction during senescence will be able to test the relationship between prophage induction and algae decay.

Vibrio prophages have the genetic potential to impact quorum sensing and biofilm formation

Quorum sensing (QS) mediates individual and population behaviors and is controlled by extracellular signaling molecules known as autoinducers [141]. In Vibrio, two master regulators control QS: AphA, a part of the PadR protein family, which operates at low cell densities, and the LuxR protein family, which operates at high cell densities [87]. These two transcriptional regulators control the expression of one another and a vast array of other genes [142, 143]. Here, we identified a Vibrio prophage-encoded version of AphA, the low cell density master regulator of Vibrio QS. At low cell densities, bacterial AphA is upregulated, leading to the activation of biofilm formation and virulence genes, thereby enhancing colonization [87, 113, 114, 142–144]. The prophage-encoded aphA gene was flanked by viral genes and shared only 34% amino acid similarity with the host homolog, indicating that it was unlikely to result from host contamination. Despite this sequence divergence, the two proteins are structurally similar (Fig. 5b). The prophage-encoded AphA homolog is most closely related to another viral-encoded protein from a phage identified in S. natans metagenomes. QS has been shown to mediate prophage lysogeny decisions in Vibrio and to influence biofilm growth [145–147]. We speculate that the prophage-encoded AphA may similarly influence the lysogenic switch or bacterial quorum sensing via unknown mechanisms. This regulation could represent a way by which planktonic vibrios switch to biofilm growth in the presence of Sargassum as the algae moves to coastal waters and vibrios are enriched, or by which Sargassum-associated vibrios trigger planktonic dispersal during senescence. These hypotheses should be the focus of future studies.

The temperate phage that showed the second largest increase in abundance in the MC-treated group encoded a putative chitinase gene (Fig. 2c and d). Vibrio often associate with organisms that produce chitin, such as crustaceans, zooplankton, bivalves, and even some bryozoans [148, 149], and possess a conserved set of genes for chitin degradation, facilitating its utilization as a carbon and nitrogen source [150]. Although Sargassum itself does not produce chitin, holobiont members associated with Sargassum, like bryozoans, incorporate chitin and calcium carbonate into their exoskeletons [151, 152]. Vibrio spp. chitinases have been identified in several Vibrio infecting prophages [153], and a Pseudomonas sp. prophage encoding a chitinase enhanced the ability of its host, isolated from deep-sea sediments, to degrade chitin and form biofilms [48]. The identification of prophage chitinases in many Vibrio bacteria suggests lysogenic conversion of chitinases could be common in other marine environments.

Our study also identified one Zot-encoding inovirus whose increase in fractional abundance in the MC-treated group was partially responsible for the differences between treatments. Zot is a prophage-encoded toxin distributed across several Vibrio pathogens, with the best-studied example being the CTXΦ of Vibrio cholera [154, 155]. The C-terminal end of the Zot protein is involved in phage assembly, but the N-terminus is cleaved and secreted, where it breaks down tight junctions between eukaryotic cells [156]. Macroalgal cells do not form tight junctions; instead, they form plasmodesmata for intercellular communication and adhesion [157]. This indicates that some Vibrio retrieved in our study may use Sargassum as a dispersal mechanism, and may benefit from nutrients released by Sargassum, but may ultimately colonize animals hosts. A potential role for Sargassum as a vehicle for potential pathogenic Vibrio dispersal and growth has been proposed in previous studies [42, 158], especially in coastal waters where Vibrio become more abundant [38] and should be fully addressed in future studies.

Chemical prophage induction alters the metabolic potential of Sargassum-derived biofilms

Seventeen metabolic modules encoded by the bacterial community differed between post-treatment groups, all of them decreasing in the MC-treated group (Fig. 6b). Among the metabolisms depleted in bMAGs by MC treatment was dissimilatory nitrate reduction (Fig. 6b). Nitrogen metabolism is of interest concerning Sargassum because the algae’s tissue % N has increased by 35% since the 1980s, demonstrating that increased N availability is a factor contributing to Sargassum blooms [159]. Sargassum δ¹⁵N, an indicator of N source, varies across its range, indicating that the algae utilize multiple N pools [159, 160]. In certain regions, low δ¹⁵N and nitrogen fixation rates of Sargassum mats signal that N fixation by bacterial epiphytes on Sargassum is contributing to the N sources of these blooms [32, 33, 159, 160]. A study examining the relative abundance of genes involved in nitrogen cycling, using nifH as an indicator of nitrogen fixation, revealed that the majority of nitrogen-fixing bacteria on Sargassum were heterotrophic bacteria in the genera Vibrio and Filomicrobium [32]. In our study, we identified one Vibrio sp. bMAG that encoded the nifH enzyme; this bacterium also had genes for dissimilatory nitrate reduction and nitrate assimilation (Data S11). Several Vibrio spp. are documented as nitrogen fixers [161, 162]. ¹⁵N isotope tracing studies with sea urchins have demonstrated that microbially fixed N, via a Vibrio bacterium, can be incorporated into sea urchin tissues [163], suggesting that Vibrio bacteria could supply Sargassum with fixed N.

A functional analysis of all bacterial contigs differed from the bMAG analysis: 51 modules with differences between the control and MC post-treatment, 46 of which were enriched in the MC-treated (Fig. S9). This difference suggests that dominant community members for which bMAGs could be assembled encoded distinct metabolic functions that were depleted upon chemical induction, while lower abundance members (only represented among contigs) increased in relative abundance.

The ecological dynamics of prophage induction in Sargassum biofilms

In the context of the existing literature on prophage induction in biofilms, our data presents a duality regarding the ecological role of prophage induction: it may potentially facilitate biofilm formation and host colonization [164, 165], or act as a disruptor that lyses dominant Vibrio populations to alter community structure (Fig. 4). We speculate that both processes play a role in Sargassum biofilms: at low rates of spontaneous phage induction of a minor fraction of the population, induction may contribute to biofilm formation through the release of polymeric molecules, while mass induction events may abruptly shift community composition with the depletion of Vibrio. These mass induction events could be caused by intense exposure to ultraviolet radiation, as low wind speeds cause Sargassum mats to form large aggregates, bringing them closer to the surface, which has been coupled with reduced Vibrio abundances on Sargassum samples [37]. The environmental factors controlling prophage ecological dynamics in the Sargassum natural environment represent an exciting line of inquiry.

Conclusion

Our study shows that chemical prophage induction with Mitomycin C significantly alters the community composition and metabolic potential of multispecies biofilms representative of heterotrophic bacteria associated with pelagic Sargassum. Prophage induction reduced Vibrio abundances and opened niche space for less abundant Gammaproteobacteria in the genera Alteromonas, Pseudoalteromonas, and Cobetia. This reduction in Vibrio bacteria has the capacity to alter sulfur, nitrogen, and carbon metabolisms within the biofilm community. Induced viruses encoded genes involved in biofilm formation, virulence, and host metabolism. Overall, the data from this study suggest that viral infection strategy may be a biotic factor influencing Vibrio abundances associated with Sargassum, providing a potential mechanism that contributes to changes in community composition across Sargassum’s range.

Author contributions

CBS and AKS designed the study. AKS, NSV, and BAW analyzed the genomic data. AKS analyzed the results and wrote the manuscript. All authors contributed to the manuscript review.

Funding

AKS was funded by the University of Miami Dean’s Fellowship and an NSF GRFP (2023349872). BAW was funded by an NSF GRFP (2023353157). CBS was funded by NASA (80NSSC23K0676) and NSF (DMS2424579).

Data availability

Raw reads are available in the NCBI Sequence Read Archive (PRJNA1111456 for metagenomes from biofilm induction assay [Fig. 2, 3, 4, 5 and 6] and PRJNA1063307 for metagenomes of naturally occurring S. natans VIII [Fig. 1d & e]). Co-assembled contigs [166], bMAGs and the Newick file for the ANI tree [167], prophages with and without flanking regions [168], biofilm assay viral sequences [169], and AphA protein sequences and the Newick file for their protein phylogeny tree [170], are available on Figshare under the [Sargassum-derived biofilm induction project] project, [https://figshare.com/projects/Sargassum-derived_biofilm_induction_project/260945]. The code used for the bioinformatic pipeline and the R script for data analysis are available on the Silveira-Lab GitHub under the [Sargassum-derived-Biofilm-Induction-Bioinformatics-Pipeline] repository, [https://github.com/Silveira-Lab/Sargassum-derived-Biofilm-Induction-Bioinformatics-Pipeline].

Declarations

Competing interests

Cynthia B. Silveira is a member of the editorial board of Environmental Microbiome .