A cooperative cobalamide biosynthesis guild in the endosphere of the edible aquatic plant Wolffia globosa Mankai

Abstract

Background

Cobalamin (vitamin B₁₂) is synthesized only by certain bacteria and archaea and is rarely found in plant-derived foods because plants neither synthesize nor require this cofactor. The edible duckweed Wolffia globosa Mankai is unusual in containing bioavailable cobalamin, suggesting a microbial origin. However, how cobalamin biosynthetic capacity is organized within angiosperm-associated microbiomes remains largely unresolved. Here, we investigated bacterial community structure and cobamide biosynthetic potential across the cultivation medium, plant surface, and internal tissues of Mankai to determine how cobalamin production is maintained in this aquatic plant microbiome.

Results

Bacterial communities differed significantly among compartments, with the endosphere forming a low-diversity, host-filtered microbiome enriched in specialized taxa. Genome-resolved metagenomics showed that only a minority of endophytic bacteria encoded near-complete cobamide biosynthesis pathways consistent with de novo synthesis. In contrast, many co-occurring taxa lacked multiple biosynthetic steps but were enriched in genes associated with cobamide precursor salvage and remodeling. Network analysis identified putative producer taxa as highly connected hubs linked to salvager populations, consistent with metabolite cross-feeding. Comparative genomic analysis demonstrated reduced cobamide biosynthetic gene complements in endophytic genomes relative to closely related free-living strains, supporting adaptive pathway reduction in the host-associated niche.

Conclusions

Cobalamin production in the Mankai endosphere appears to arise from a metabolically interdependent bacterial consortium rather than from single autonomous producers. These findings identify cooperative micronutrient biosynthesis as an organizing principle in plant-associated microbiomes and position Mankai as a tractable model for studying cobamide-mediated microbial cooperation in aquatic crops. Understanding these interactions may support microbiome-informed strategies to stabilize micronutrient production and functional resilience in controlled aquatic plant cultivation systems.

Supplementary Information

The online version contains supplementary material available at 10.1186/s40793-026-00917-4.

Background

Plant microbiomes play crucial roles in plant health, nutrient acquisition, and environmental adaptation, strongly influencing fitness and resilience to stress [1]. While the microbiomes of terrestrial plants have been extensively studied, aquatic plant microbiomes remain relatively underexplored [2]. Aquatic plants face distinct challenges, including nutrient limitation, rapid nutrient diffusion, and the need to maintain microbial communities that differ from the surrounding water column [3, 4]. They also contend with competition from algae and cyanobacteria [5], variable oxygen regimes at the plant-water interface [6], and fluctuating water chemistry and hydrodynamic forces [7] that can shape microbial colonization and plant-microbe interactions [8]. Among aquatic plants, the microbiomes of free-floating duckweeds (Lemnaceae) are of particular interest because of their potential in biotechnology, wastewater remediation, and sustainable agriculture [9, 10]. Duckweeds, especially Wolffia globosa Mankai (Fig. 1), have recently gained attention as a promising food source due to their rapid growth [11] and reported health benefits [12–14]. Of special interest is the reported presence of bioavailable cobalamin (vitamin B₁₂), a rare trait among plant foods [15–17].

Fig. 1.

Microscope image of a mother and daughter frond of W. globosa Mankai

Cobalamin, an essential water-soluble vitamin synthesized exclusively by certain bacteria and archaea and required for key processes in mammals, including DNA synthesis, red blood cell formation, and neurological function [18]. The lack of reliable plant-based sources of cobalamin poses a challenge for individuals following vegetarian and vegan diets. While members of the Rhodobacteraceae family have been identified as cobalamin producers in marine and freshwater algae-associated microbiomes [19, 20], the identity of cobalamin-producing bacteria in duckweed and the organization of their biosynthetic capacity remain largely unresolved [15, 16].

Recent studies have advanced our understanding of duckweed epiphytes [21–26]. Ishizawa et al. (2017) showed that Lemna minor harbors a stable bacterial community that can promote host growth and stress tolerance of the host. Toyama et al. (2009) identified bacterial communities involved in hydrocarbon degradation, suggesting a role for duckweed-associated microbiomes in phytoremediation. Using a synthetic bacterial community of six strains isolated from L. minor, a stable community was maintained for over 50 days, reflecting the family-level structure of natural duckweed microbiome’s [23]. While these studies highlight roles for duckweed-associated bacteria in nitrogen fixation [27], phosphorus solubilization [22], and pollutant degradation [28], the duckweeds endosphere has been largely overlooked. Unlike terrestrial phyllosphere communities, which can show substantial temporal variation driven by environmental exposure and plant developmental state [29], duckweed-associated microbiomes may provide a comparatively stable aquatic plant system, partly due to clonal reproduction and continuous interaction with the surrounding waterborne microbial pool [8]. The endosphere may be particularly important in rootless edible Wolffia, yet its diversity, community composition, and functional potential remain poorly characterized [30].

In this context, the microbiome structure and micronutrient availability in Mankai intensively managed production systems may jointly shape community stability and functional outputs, making Mankai a tractable model for phytobiome resilience in aquatic crops. We ask: (i) how the plant and its environment shape bacterial diversity in the cultivation medium and on/in the fronds (phyllosphere and endosphere fractions); (ii) how cobamide biosynthetic and salvage capacities are partitioned among taxa, and whether cobalamin production emerges from a minority of specialist producers supporting a broader guild of salvagers and auxotrophs; and (iii) whether comparisons of cobalamin producer genomes with closely related strains reveals evidence of adaptive metabolic streamlining in this host-associated niche. We addressed these questions by integrating 16 S rRNA gene amplicon sequencing, metagenomic analysis, comparative genomics of isolates, and cobalamin measurements in Mankai cultivated at industrial scale. We hypothesized that: (1) the Mankai endosphere hosts a low-diversity but strongly host-filtered bacterial community that is compositionally distinct from the phyllosphere and cultivation medium; (2) cobalamin biosynthesis in the endosphere is organized as a cooperative guild consistent with Black Queen type division of labour [31], in which a minority of endophytic taxa encode near-complete cob/cbi pathways while many others rely on cobamide precursor and corrinoid salvage; and (3) endophytes lineages exhibit adaptive pathway erosion of cobamide biosynthetic, reflecting metabolic interdependence in a stable host-associated environment. Together, these hypotheses provide a framework for understanding of cobamide-mediated cooperation and metabolic division of labor in plant-associated microbiomes.

Methods

Duckweed strain

We focused on W. globosa Mankai (Fig. 1), a free-floating aquatic plant and a domesticated cultivar (a proprietary strain of Hinoman Ltd, Be’eri, Israel), belonging to the Lemnaceae family. Each frond, typically 0.3–0.9 mm in diameter, is a simple thallus lacking roots, making it the smallest flowering plant [32]. The plant’s green, globular structure, as seen in the image, facilitates buoyancy and efficient nutrient absorption from the surrounding water [33].

Sampling and DNA extraction

Mankai was collected from Hinoman’s industrial cultivation site utilizing a fully automated, hands-free system with rotating raceway basins, all under closed-controlled green-house conditions at Kibbutz Be’eri. Four replicate samples were collected from three separate basins in April 2020. 12 samples of each matrix including the medium, phyllosphere, and endosphere microbial communities were analyzed. Water samples (n = 12) were filtered (0.2 μm), and plants samples (n = 24) were either surface-sterilized or washed with sterile nutrient solution before homogenization. Plants were surface sterilized using 5% hydrogen peroxide prior to tissue homogenization to minimize epiphytic carryover. The washed plant fraction is referred to here as the phyllosphere fraction; however, because whole fronds were homogenized after surface washing, this fraction is not strictly epiphytic and include tissue-associated bacteria. In this study, “endosphere” is used operationally to describe the sterilized-tissue-associated fraction recovered after surface sterilization and washing, with methodological limitations described in the Supplementary Information. DNA was extracted using the ExiPrep™ Plus Plant Genomic DNA Kit (Bioneer, Daejeon, S. Korea) and quantified with PicoGreen dsDNA assay (Thermo Fisher, Waltham, MA, USA). Full details of the sampling protocol, sterilization methods, and DNA extraction procedures are available in Supplementary Information.

Amplicon sequencing and analysis

Details of sequencing protocols and the full workflow of analysis, including statistical methods and thresholds, are provided in Supplementary Information. Briefly, amplicon sequencing targeted the V3–V5 region of the 16 S rRNA gene using primers 314 F and 926R, with plant chloroplast and mitochondrial sequences blocked using Peptide Nucleic Acids (PNA; PNA Bio, Thousand Oaks, CA, USA). Sequencing was performed on an 2 × 300 bp MiSeq platform (Illumina, San Diego, CA, USA). The raw data were deposited in BioProject: PRJNA1356720 (SRA Study: SRP645794).

Taxonomic classification of the amplicon reads was performed using QIIME2 [34] with SILVA 138 reference database [35]. ASVs were filtered to remove singletons, non-target taxa (e.g., chloroplast, mitochondria), and low-quality sequences. Diversity metrics, including alpha diversity (Shannon index, Faith’s PD) and beta diversity (Bray-Curtis, UniFrac), were analyzed. Differential abundance was tested using ANCOM.

Metagenomic sequencing and analysis

Detailed description of sequencing protocols, functional profiling and genome binning approaches, see Supplementary Information. Briefly, nine Mankai samples collected from 3 separate basins were used for metagenomic sequencing conducted using NovaSeq 6000 after microbiome DNA enrichment. Read-based and assembly-based metagenomic techniques were used to characterize and annotate the metagenomes. The raw data were deposited in NCBI under BioProject PRJNA1356720 (SRA Study: SRP645794). Functional annotation of metagenomes was performed using HUMAnN2 [36] and MEGAN [37], mapping reads to UniRef90 and KEGG pathways [38]. Gene enrichment analysis for cobalamin biosynthesis pathways was conducted using clusterProfiler in R [39]. Metagenome-assembled genomes (MAGs) were reconstructed using MetaBat2 [40] and annotated with MetaErg [41]. MAG quality (completeness and contamination) was assessed with CheckM, and bins were retained according to standard genome-quality criteria. The resulting MAGs were deposited as MIMAG BioSamples under BioProject PRJNA1356720 (SRA Study: SRP645794).

Functional complementarity network

A full description of network construction and pathway curation is provided in the Supplementary Information. Briefly, to explore potential cobamide cross-feeding among endosphere taxa, we built a functional complementarity (metabolic dependency) network based on cobamide biosynthesis, salvage, and remodeling functions predicted from metagenome-assembled genomes. Nodes represent taxa, and edges represent the number of cobamide pathway intermediates that could be supplied by one taxon and utilized by another based on complementary gene content, thereby capturing putative exchange relationships inferred from pathway annotation rather than abundance correlations. Networks were constructed in R using igraph [42], and we required a minimum of three compatible intermediates to define an edge, to focus on higher-confidence metabolic dependencies. Cobamide gene presence and pathway completeness were assessed across metagenome-assembled genomes and isolate genomes to support interpretation of producer versus salvager roles. Importantly, edges in this network do not represent statistical co-associations inferred from compositional microbiome abundance data.

Statistical analyses

Details on normalization, statistical tests, and data visualization are available in Supplementary Information. Briefly, comparative analyses were performed in R (v4.0.3) using Kruskal-Wallis and PERMANOVA tests. P-values for differentially abundant taxa and pathways were adjusted to q-values using Benjamini-Hochberg false discovery rate (FDR) correction.

Cultivation and genome analysis of endophytes-equivalent isolates

MAGs reconstructed from the Mankai endosphere indicated near-complete genomes for several species-level bins. Based on this analysis [43], we selected four abundant taxa, Sphingopyxis terrae, Caulobacter vibrioides, Hyphomicrobium zavarzinii, and Hydrotalea flava as representatives for isolate-based validation. Because repeated isolation attempts from Mankai were unsuccessful, we obtained closely related species from culture collections (Table S1). Relatedness between MAG-level taxa and cultured representatives was assessed at the 16 S rRNA gene level, and the cultured strains should therefore be interpreted as species-level comparators rather than strain-matched counterparts of the endophytes.

Details on cultivation, DNA extraction, sequencing and analysis are provided in the supplementary information. Briefly, genomic DNA was extracted from pure cultures (Exgene Cell SV Mini Kit, GeneAll, Seoul, S. Korea), quantified by NanoDrop (Thermo Fisher), and sequenced on an Illumina platform to generate paired-end FASTQ files. Reads were quality-checked (FastQC) and processed with BBTools: deduplication/compression (clumpify), adapter/low-complexity filtering and quality trimming (bbduk), interleaving/merging of overlapping pairs (reformat, bbmerge), and k-mer–based error correction (tadpole). Low-quality or short reads were discarded. De novo assemblies were produced with SPAdes using multiple k-mer sizes [44, 45]; assembly quality was assessed with QUAST [46]. Processed reads were mapped back to assemblies with Bowtie2 [47]; alignments were converted, sorted, and indexed with SAMtools [48], and assemblies were polished with Pilon [49] to correct base errors and small indels. Open reading frames were predicted with Prodigal [50] and functionally annotated with Bakta [51]. Presence/absence of cobalamin (cob/cbi) biosynthesis genes was evaluated against KEGG map00860; genes in the aerobic corrin-ring and salvage pathways were recorded for each isolate and visualized as binary heatmaps in R (ggplot2) [52].

Cobalamin quantification in W. globosa Mankai and the isolates

A full description of cobalamin extraction and quantification by the bioassay is provided in the Supplementary Information. Briefly, cobalamin was quantified in Mankai and the isolates using VitaFast Vitamin B₁₂ (Cyanocobalamin) kit (R-Biopharm, Pfungstadt, Germany) according to the manufacturer instructions. Please note that in this manuscript we use cobamide as the broader family; cobalamin refers to the canonical vitamin B₁₂ form.

Results

Domain-level composition of W. globosa Mankai endophytes was assessed with MEGAN (Fig. S1). Bacteria dominated the microbiome, accounting for 97.8 ± 0.63% of the classified reads, whereas Archaea were not detected and unicellular eukaryotes were present at low relative abundance (0.3 ± 0.14%). Given the low abundance of Eukarya and the absence of detectable Archaea in our datasets, subsequent analyses focused on the endophytic bacterial community. However, because shotgun metagenomes were generated after microbiome DNA enrichment to reduce host DNA, the low relative abundance of eukaryotic reads should be interpreted cautiously, as this step may have reduced recovery of methylated eukaryotic DNA. In addition, MAG recovery and pathway reconstruction may be influenced by enrichment-related biases in taxon recovery.

Compartmental filtering creates a low diversity endosphere microbiome

Amplicon sequence analysis was performed based on the SSU rRNA gene region amplified from the DNA extracted from the medium, phyllosphere and endosphere of Mankai (Table S2). Based on the sequences’ analysis we estimated the diversity of the bacterial communities (Table S3) focusing on richness (observed ASVs) and diversity (Shannon entropy) indices (Fig. 2A). The bacterial community in the medium exhibited the highest species richness, with almost 150 observed ASVs. The phyllosphere and endosphere communities were host filtered and showed lower richness (Fig. 2A) consistent with strong physiological constraints at the plant-water and plant-tissue interfaces. Statistical analyses confirmed niche differentiation richness across these compartments (p < 0.05, Table S4). A similar trend was observed for diversity (Fig. 2B). The medium exhibited the highest Shannon diversity index (4.2), followed by the phyllosphere (3.3) and the endosphere communities (3.1). These differences in diversity between the three compartments were also statistically significant (p < 0.05, Table S4).

Fig. 2.

Bacterial diversity and composition across compartments of W. globosa Mankai culture. A Observed ASVs indicating bacterial richness in the medium, phyllosphere, and endosphere compartments. B Shannon entropy indicating alpha diversity of bacterial communities across the three compartments: medium (red; n = 12), phyllosphere (green; n = 12), and endosphere (blue; n = 12). C PCoA based on Bray-Curtis dissimilarity, showing distinct clustering of bacterial community structures. D Venn diagram illustrating the shared and unique ASVs among the across the three compartments. E Relative abundance of bacterial taxa at the genus level in each compartment, emphasizing differences in community composition. Genera labeled as “Rare” represent taxa with low relative abundance across all samples

To investigate differences in bacterial community composition, a PCoA based on Bray-Curtis dissimilarity was conducted. This analysis explained 80.8% of the total variation in bacterial communities. Distinct clustering patterns were observed, with the medium, phyllosphere, and endosphere communities forming distinct clusters (Fig. 2C) that were significantly different (p < 0.001, Table S5) Analysis of ASV distribution across the three compartments revealed overlapping and unique taxa. A total of 18 ASVs were shared among all three compartments, while six ASVs were shared exclusively between the medium and phyllosphere, 13 were shared between the phyllosphere and endosphere, and one was shared between the medium and the endosphere (Fig. 2D). These results show that the three compartments contain both shared and compartment-specific ASVs, with the endosphere retaining a distinct subset of the broader plant-associated and medium-associated bacterial pool. The amplicon sequencing results suggest that the endosphere bacterial community is distinct.

The bacterial community composition at the genus level was analyzed to identify dominant taxa in each compartment. The medium was dominated by uncultured members of the family Comamonadaceae (28%). The phyllosphere also showed a higher abundance of Comamonadaceae (44.3%), particularly the genera Rhizobacter (20%) and Methylophilus (14.3%). Additionally, the genus Acinetobacter from the order Pseudomonadales was detected at a high abundance (9.6%). The endophytic community was dominated by Comamonadaceae (43.1%), particularly the genera Mitsuaria (15.5%) and Methylophilus (10.6%). Rare genera constituted a notable fraction of the total community across all compartments, ranging from 28% in the medium to 10.9% in the phyllosphere and 6.7% in the endosphere (Fig. 2E).

Endophytic MAGs reveal a proteobacteria-dominated community with high-quality genomes

To complement the amplicon sequencing-based analysis, a metagenomic approach was employed to gain deeper insights into the taxonomic composition, phylogenetic relationships, and genomic characteristics of the endosphere communities of Mankai. Figure 3 demonstrates that most MAGs, including those of Sphingomonas, Caulobacter, and Ralstonia, exhibited high completeness (> 90%) and low contamination (< 5%), presenting the reliability of the genomic assemblies. This high-quality genomic data allowed for a more detailed exploration of the bacterial diversity within the endosphere.

Fig. 3.

Taxonomic composition, phylogenetic relationships, and genome quality of the W. globosa Mankai endosphere microbiome (n = 10). The left panel shows the relative abundance (%) of dominant species in the endosphere based on MEGAN taxonomic profiling of shotgun metagenomic reads. The middle panel presents phylogenetic relationships among these species, grouped by major bacterial phyla and classes. The right panel summarizes MAG quality metrics, including completeness (%) and contamination (%), with bars representing completeness (blue) and contamination (red). Thus, species abundance is based on read-level MEGAN analysis, whereas genome quality metrics are based on MAG reconstruction

The relative abundance and phylogenetic relationships among bacterial species derived from metagenomic sequencing suggest that the bacterial community was predominantly composed of members of Proteobacteria, particularly Alpha-Proteobacteria and Gamma-Proteobacteria, with contributions from Bacteroidota and Actinobacteria. Genera such as Sphingomonas, Hyphomicrobium, and Ralstonia were identified as key contributors to the community structure. The phylogenetic tree in Fig. 3 further provides an evolutionary framework for these taxa, illustrating their relationships and grouping them into distinct clades.

Fragmented cobamide pathways and widespread salvage genes indicate metabolic interdependence

Cobalamin was detected in Mankai at 2.8 ± 0.5 µg B12/100 g dry weight (DW) [17]. Recent work indicates that cobalamin producing epiphytes do not contribute to duckweed cobalamin [15], arguing against a surface-driven source. Moreover, cobalamin biosynthesis is restricted to bacteria and archaea, while bacteria dominate the endosphere (Figure S1). Because the phyllosphere fraction was obtained from water-rinsed plants, it likely includes both epiphytic and tissue-associated bacteria; therefore, overlap with the endosphere was expected. However, the lower diversity, distinct composition, and enrichment of potential cobalamin-producing taxa in the sterilized-tissue-associated fraction supported focusing downstream cobamide biosynthesis analyses on this compartment. Together, these observations support the endosphere as the plausible compartment linking bacterial biosynthesis to the plant’s cobalamin pool. We therefore confined downstream functional analyses to the endosphere associated bacterial communities.

The cobamide pathway begins with uroporphyrinogen III and proceeds through a series of cob/cbi-encoded steps to assemble the corrinoid macrocycle, followed by nucleotide-loop assembly and ligation to yield the active cofactor (Fig. 4). In the Mankai endosphere, the distribution of cobalamin-biosynthesis genes was uneven across taxa. Members of the Alphaproteobacteria, including H. zavarzinii, Hyphomicrobium sp., Devosia spp., Methylobacterium sp., and Labrys sp., carried near-complete gene sets (Fig. 4). By contrast, many Gammaproteobacteria and Bacteroidota lineages harbored incomplete pathways, implying reliance on exogenous cobamide intermediates. Importantly, pathway fragmentation and enrichment of salvage/remodeling functions were observed across multiple endosphere taxa rather than being driven by a single genome. Together with MAG quality assessment and the recurrent co-occurrence of incomplete de novo pathways with salvage/remodeling genes, this pattern supports a community-level interpretation, although missing genes in individual MAGs should be interpreted cautiously. It emphasizes the community capacity to recycle precursors supplied by other community members and underscores a functionally interdependent endosphere microbiome. This pattern is consistent with Black Queen dynamics, where costly functions are retained in a subset of community members and exploited by auxotrophs.

Fig. 4.

Presence/absence of cobalamin biosynthesis enzymes in the W. globosa Mankai endophytes community. The top heatmap shows the presence of aerobic genes (blue), the middle shows anaerobic gene presence (purple) and the bottom shows the final steps of cobalamin biosynthesis (green). The left panel illustrates cobalamin biosynthetic gene module

Metabolic dependency networks highlight cobamide-producer hubs and peripheral salvagers

The functional complementarity network identified a small set of taxa with near-complete cobamide biosynthetic capacity as highly connected hub nodes, linked to multiple taxa with incomplete pathways enriched in salvage and remodeling functions (Fig. 5). This topology is consistent with a producer-salvager structure in which corrinoid intermediates are putatively exchanged across the endosphere consortium. In this network, node size is proportional to relative abundance, and edges represent MAG derived metabolic compatibility based on ≥ 3 compatible cobamide intermediates inferred from complementary gene content rather than abundance correlations.

Fig. 5.

Functional complementarity network for cobamide metabolism in the Mankai endosphere. Nodes represent taxa and are scaled by their relative abundance in the endosphere. Edges AG-derived metabolic complementarity based on compatible cobamide pathway intermediates inferred from annotated gene content; edge thickness reflects the number of intermediates supporting each putative exchange relationship. The network represents genome-inferred cobamide pathway potential and should not be interpreted as direct evidence of metabolite exchange or abundance-based co-occurrence

Within the network, H. zavarzinii emerged as a central hub, sharing the highest number of intermediates with M. ginsengisoli. Additional highly connected taxa, including Methylobacterium sp., M. purpurogeneiscleroticus, Devosia sp., and Ferrovibrionaceae, encoded most cobamide biosynthesis steps and showed extensive intermediate sharing with other endophytes, consistent with roles as primary producers of cobamides or key precursors.

In contrast, multiple taxa formed lower-connectivity peripheral nodes, including Sediminibacterium sp., Acidovorax sp., Ralstonia pickettii, Novosphingobium sp., UBA8621 sp., H. flava, Aquabacterium sp., Pelomonas sp., and C. vibrioides, which shared fewer compatible intermediates and were consistent with salvager-dominated strategies. The least connected taxa included S. terrae, Pseudoxanthomonas sp., Sphingomonas sp., Burkholderiaceae sp., Brevundimonas sp., UBA1930, and Chitinophagaceae sp. Finally, because this network is based on MAG derived metabolic complementarity, it does not encode negative (competitive) associations; accordingly, the network structure primarily reflects putative cooperative dependencies in cobamide metabolism.

To compare this pathway-derived network with abundance-based community structure, we performed an exploratory overlay with a co-occurrence network. This analysis showed that 25 of 33 significant co-occurrence edges (75.8%) were also present in the MAG-derived cobamide intermediate-sharing network. Although co-occurrence does not demonstrate direct metabolite exchange, this overlap supports the interpretation that taxa with complementary cobamide pathway potential tend to co-occur within the Mankai endosphere.

Comparisons to free-living relatives reveal pathway reduction in the endophytes

To characterize cobalamin biosynthetic capacity in the Mankai endosphere, we compared four species-level MAGs (Fig. 3) including S. terrae (3.96 ± 0.99% relative abundance), C. vibrioides (11.73 ± 2.85%), H. zavarzinii (15.83 ± 5.18%), and H. flava (4.98 ± 2.35%), with cultured representative strains obtained from culture collections (Table S1). Repeated attempts to isolate these endophytes directly from Mankai were unsuccessful (data not shown). The cultured representatives used in this study originate from environmental sources, including wetland soil (H. zavarzinii), freshwater systems (H. flava), and soil associated with wastewater environments (S. terrae), and therefore represent free-living lineages rather than plant-associated endophytic populations. Read mapping further suggested that these cultured strains are not identical to the Mankai-derived endophytic populations. Consequently, comparisons with Mankai-derived MAGs are interpreted at the species-comparator level rather than as strain-level matches. Accordingly, comparisons were used to evaluate species-level patterns in cobamide gene complements rather than strain-level differences. In this comparison, endosphere MAGs showed more fragmented cobamide gene sets than their cultured species representatives, a pattern consistent with reduced pathway completeness in host-associated genomes, although strain-level divergence limits evolutionary inference. Even the one isolated species C. vibrioides that was initially presumed to represent an endophyte differed from the endophytic MAG (Fig. 6) and is probably epiphytic rather than endophytic.

Fig. 6.

Presence/absence of cobamide biosynthesis enzymes in endosphere MAGs and cultured species representatives. Heatmaps summarize gene presence across the core cob/cbi set for genomes of cultured representatives obtained from collections (blue) and for MAGs (dark blue). Comparisons are intended to illustrate species-level differences in pathway completeness and do not imply strain identity between MAGs and cultured isolates (Table S1)

A binary heatmap (Fig. 6) summarizes gene presence/absence across the core cob/cbi set, contrasting MAGs (left; light brown = present, white = absent) with cultured counterparts (right; dark brown = present, white = absent). MAGs display a mosaic, fragmentary distribution of cobalamin pathway encoding genes, with multiple steps missing within individual endophytic genomes. By contrast, the cultured species show near-complete repertoires with only occasional gaps, consistent with de novo biosynthetic capacity at the isolate level. This was verified by testing for cobalamin production in the cultivated strains, where we identified that C. vibrioides and H. zavarzinii produced relatively high cobalamin levels, whereas S. terrae, and H. flava produced very low concentrations, which may reflect stored intracellular pools rather than substantial de novo production under the assay conditions (Table S6).

Discussion

Cobalamin biosynthesis is unusually complex, and in Mankai it is partitioned across endophytes, implying that costly functions are treated as shared metabolic public goods. Strong host and habitat filtering, together with resource constraints, likely favors division of labor and metabolic interdependence over fully autonomous production. Because the plant host is not cobamide-dependent, the system contrasts with classical algal mutualisms [20, 53], demonstrating microbial-microbial selection in a plant-associated niche.

Cobamide-mediated cooperation in Mankai endosphere

Duckweeds are unusual among land plants in that bioactive cobalamin has been detected in Mankai and shown to be bioavailable to humans [16, 17]. Yet classical algal and planktonic models, in which bacteria supply cobalamides to auxotrophic phototrophs in exchange for fixed carbon [19, 20, 54, 55], do not map directly onto this system. Angiosperms lack known cobalamin-dependent enzymes [56, 57], so any selective advantage for cobamide biosynthesis in the Mankai endosphere is most plausibly bacterial rather than host-driven.

The distribution of cobalamin biosynthesis genes among endophytes was uneven (Fig. 4). Several Alphaproteobacteria genera (e.g., Hyphomicrobium, Methylobacterium, Devosia and Labrys) carried gene sets consistent with de novo cobamide biosynthesis, whereas most MAGs lacked multiple steps but are enriched in salvage and remodeling genes (Fig. 4). Together with the pathway-derived interaction network topology (Fig. 5), this pattern supports a division of labor model in which a subset of producer taxa generates corrinoid intermediates and complete cobamides that are then salvaged by surrounding community members [58, 59]. The high centrality of Hyphomicrobium and Methylobacterium in the network (Fig. 5), and their links to multiple links to taxa with partial pathways, is consistent with this producer-salvager structure.

Unlike some macroalgal microbiomes where a single symbiont encodes the full pathway [60, 61], the Mankai endosphere appears to host a distributed cobamide network in which no single strain carries the full biosynthetic burden (Fig. 4). We therefore interpret cobamide production as a microbiome-centric trait that primarily supports the bacterial guild itself, and potentially the epiphytic community, rather than a direct nutritional requirements of the host [57]. This interpretation is consistent with bacterial dependence on cobamides for central metabolic reactions, whereas plants are generally considered not to require cobalamin for their own metabolism [18]. Accordingly, any plant-level benefit is likely indirect [62], through stabilization of cobalamin-dependent microbial functions that may support nutrient cycling, stress tolerance, pathogen suppression [63], and microbiome resilience at the plant-water interface. In an engineered cultivation system where trace metals such as cobalt may fluctuate, distributing production, salvage, and recycling across community members could buffer cobamide supply against local limitation and help stabilize cobalamin-dependent bacterial metabolism [64]. More broadly, vitamin-mediated interactions can contribute to stable community structure and sustained metabolic capacity [58], which may indirectly influence crop performance via nutrient cycling [65], colonization resistance [66], and robustness of microbiome functions under environmental fluctuation [1].

Compartmental filtering and endosphere community structure

The pronounced differences in richness and composition between the medium, phyllosphere and endosphere (Fig. 2) indicate strong environmental and host filtering along the plant-water interface. The cultivation medium harbored the highest richness and diversity, whereas the phyllosphere and especially the endosphere contained progressively lower-diversity assemblages with limited ASV overlap (Fig. 2). The ASVs shared across all three compartments likely represent taxa capable of persisting across the plant-water interface, whereas compartment-specific ASVs and the clear beta-diversity separation (Fig. 2) indicate selective enrichment within plant-associated niches. These patterns mirror observations in other aquatic plants, where waterborne source communities are filtered by plant surfaces and tissues [8, 15, 30].

Within Mankai, the endosphere was dominated by Proteobacteria (Fig. 3), including Hyphomicrobium, Sphingomonas, and other taxa frequently associated with nutrient cycling, stress tolerance, and adaptation to oligotrophic conditions [67, 68]. Reduced alpha diversity in the endosphere relative to the phyllosphere (Fig. 2) was consistent with stringent host-associated filters, potentially imposed by resource availability [69], or microhabitat structure [70]. Distinct taxonomic signatures across compartments suggest complementary ecological roles for epiphytic and endophytic communities, the former more directly exposed to the fluctuating aqueous environment, whereas the latter resides in a more stable, host-mediated niche [71]. It is within this host-filtered niche that the cobamide guild is assembled and maintained.

Metabolic specialization and adaptive pathway reduction in endophytes

Although distributed cobamide biosynthesis is common across microbial ecosystems [3], our data show that this organization also occurs within the Mankai endosphere. Because MAG reconstruction was performed only for the endosphere fraction, we cannot determine whether this pattern is unique to the endosphere or reflects the broader cultivation microbiome. The contribution of epiphytes to duckweed cobalamin accumulation remains unresolved, but our data support the endosphere-associated community as a key compartment for cobamide metabolism. Comparison of endosphere MAGs with cultured relatives (Fig. 6) suggests metabolic specialization within the Mankai endosphere. Repeated attempts to culture the dominant endophytic taxa were unsuccessful, consistent with reports that many duckweed-associated bacteria are poorly culturable or require plant-derived factors for growth [30]. Endophytic genomes showed fragmented cobamide pathways with multiple missing steps, whereas closely related free-living strains retained near-complete cob/cbi repertoires and produced cobalamin in culture (Fig. 6). Together, the pathway fragmentation (Fig. 4), network structure (Fig. 5), and comparison with free-living relatives (Fig. 6) supports cobamide interdependence and suggest that community context may reduce selection for maintaining the full pathway in any single genome.

This pattern aligns with Black Queen dynamics, where costly functions are retained by a subset of “helpers” while many lineages evolve auxotrophies and depend on cross-feeding of essential metabolites [31, 72, 73]. Cobalamin biosynthesis is metabolically expensive and constrained by cobalt availability, making it a plausible target for partial pathway loss in taxa that can access shared metabolite pools. Accordingly, the loss of complete pathways in some endophytes (Fig. 6), coupled with widespread salvage genes (Fig. 4), is consistent with adaptive pathway erosion that lowers metabolic burden while preserving access to vitamin B₁₂ within the consortium [74, 75]. Because the closest free-living comparators were obtained from culture collections (Table S1) rather than isolated directly from Mankai, we cannot fully separate evolutionary history from current ecological interactions. Even so, the consistent differences between endophytic MAGs and their close relatives (Fig. 6), together with the network structure (Fig. 5), support host-associated metabolic streamlining rather than random gene loss.

From an application perspective, duckweed is increasingly developed as a sustainable aquatic crop for food and biotechnological production [76, 77]. Our results suggest that maintaining a cobamide-interdependent endosphere consortium could help stabilize microbiome function and vitamin-related traits under managed cultivation. Future work integrating transcriptomics, metabolite flux measurements, and controlled manipulation of key producer and salvager taxa will be essential to move from genomic potential and inferred interactions to directly measured cobamide exchange, and to develop microbiome informed management or inoculation strategies that improve the robustness of duckweed-based production systems.

Conclusion

Cobalamin biosynthesis in W. globosa Mankai appears to arise from a metabolically interdependent endophytic guild rather than a single producer taxon. The low-diversity, Proteobacteria-dominated endosphere contains a small set of Alphaproteobacterial members with near-complete cob/cbi repertoires alongside multiple taxa enriched in cobamide precursor utilization, remodeling, and corrinoid salvage functions, consistent with division of labor and pathway reduction. This distributed cobamide network propose microbial-microbial interactions as key drivers of cobamides metabolism in a host-associated niche and provides a tractable framework for studying cobamide-mediated cooperation and phytobiome functional resilience.

Supplementary Information

Below is the link to the electronic supplementary material.

Abbreviations

16S rRNA

16 S Ribosomal RNA

ASV

Amplicon sequence variant

BMC

BioMed central

DNA

Deoxyribonucleic acid

FDR

False discovery rate

KEGG

Kyoto encyclopedia of genes and genomes

MAG(s)

Metagenome-assembled genome(s)

MEGAN

MEtaGenome ANalyzer

NCBI

National center for biotechnology information

PCoA

Principal coordinates analysis

PERMANOVA

Permutational multivariate analysis of variance

QIIME2

Quantitative insights into microbial ecology versions 2

rRNA

Ribosomal RNA

SRA

Sequence read archive

SSU

Small subunit rRNA

Vitamin B12

Cobalamin

Author contributions

OG, ML and VCC conceived the study. AFY, OO, LG and RD developed the methodology. AFY, OO and RD performed the investigations. OO and RD conducted the formal analyses and curated the data. OG, ML and VCC provided resources. RD, OO and OG wrote the first draft of the manuscript. RD, OO and OG prepared the figures and visualizations. All authors contributed to review and editing of the manuscript. OG, ML and VCC supervised the project. OG administered the project. OG, ML and VCC acquired funding. All authors read and approved the final manuscript.

Funding

This research was funded by the Israel Ministry of Agriculture and Rural Development Grants No 16-38-0038 for OG, VC & ML, the Jewish Charitable Association (ICA) and the Goldinger Trust for OG. The authors acknowledge the Kreitman scholarship for their generous support to AFY, and AKIS for its generous support to RD.

Data availability

The datasets generated and/or analysed during the current study are available in BioProject PRJNA1356720 (SRA Study: SRP645794).

Declarations

Ethics approval and consent to participate

Not applicable.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.