Introduction

Microalgae (photosynthetic micro-eukaryotes) and bacteria are the most widespread and dominant planktonic organisms in aquatic ecosystems. As primary producers, microalgae form the foundations of the food web and directly influence global carbon and nutrient cycling, as well as the energy flow in aquatic ecosystems. A wide spectrum of associations between microalgae and bacteria have been reported, predominantly related to nutrient exchange, with important bottom-up effects on primary production. Associations include increased bioavailability of vitamins (e.g.B121,2,3), metals (e.g. iron4,5) and growth promoting hormones6 by bacteria in exchange for organic carbon from the algae. Apart from many beneficial interactions, some bacteria can have negative influences on algae by being algicidal and opportunistic pathogens7,8,9. A recent study on the bloom-forming Phaeocystis algae revealed that its microbiomes can take both symbiotic and opportunistic modes10. Thus, algal–bacterial interactions can range from mutualistic to parasitic, and a complex array of interactions can be hypothesised. For instance, previous work suggests that strong associations may exist between communities of microalgae and bacteria over a large geographical range in the open ocean11. However, the knowledge we have so far about specific microalgal-bacterial interactions represents only a small fraction of what potentially occurs in nature. Microalgae are highly diverse and innumerable, therefore gathering knowledge on their associations with equally ubiquitous and diverse bacteria is challenging.

High-throughput DNA sequencing such as 16S and 18S rRNA gene metabarcoding has been used to deliver insights into bacterial and eukaryotic community composition of diverse ecosystems. Microbial community abundance data can be used to identify associations between community members via microbial association networks12. Furthermore, combining community data with environmental data can reveal novel connections between microbial communities and their environment13. A network consists of nodes representing taxa and edges representing the associations between taxa. Microbial association networks targeting positive correlations are termed co-occurrence networks. A positive relationship is presumed when taxonomically relevant units co-occur or exhibit similarity in their compositions over multiple samples14 and biologically meaningful groups or communities of a network can be identified with network clustering15. Also, network metrics such as node degree, closeness centrality for example, can be used for quantitative description of communities and identify ecologically important taxa16,17. Various network analyses have reported co-occurring microbial taxa in different environments, identified clusters of microbes representing metabolic consortia, defined keystone species, documented recurrent microbial modules and helped elucidate microbial dark matter in microbial communities18,19,20,21. Ecological processes governing community structure, such as niche filtering and habitat preference, are thought to be reflected in co-occurrence network modules22. Overall, networks provide insights into the structure of a community and taxon co-occurrences, offering perspectives to use them to generate hypotheses about interactions that can be investigated with more focused studies. Thus, environmental sequencing data combined with network analysis can be a convenient addition to the toolkit for deriving potential interactions between organisms across myriads of environments.

Studies targeting 16S rRNA marker genes are generally focused on the bacterial communities and often disregard the chloroplast sequences that are also amplified. The shared origin of the chloroplasts of all oxygenic photosynthetic micro-eukaryotes and cyanobacteria23 enable the use of 16S rRNA marker gene to concurrently estimate relative abundances of microalgae and bacteria across samples, facilitating the construction of correlation networks to identify co-occurring taxa and speculate on their potential interactions. Robust co-occurrences can be detected through analysis of a sufficient number of samples, ideally covering temporal and spatial gradients, as these provide adequate variability in taxon abundances12. This can be coupled with stringent network building steps such as filtration of low prevalence organisms to reveal significant co-occurrences24. The ever-expanding sequence databases can be a great starting point to investigate such correlations. Analysis of the Tara Ocean dataset based on interaction networks has shown that abiotic factors are incomplete predictors of community structure22. Similar observations have been shown in phytoplankton blooms where environmental parameters were insignificant in influencing the community structure of plankton25. These results indicate that biological interactions are more influential in determining the community structure than environmental factors. Thus, taxon-taxon co-occurrence networks built on taxon compositions alone could capture potential interactions in nature. Our work is based on the premise that previously unknown associations between microalgae and bacteria may be discovered at large scales from existing 16S rRNA gene-based metabarcoding datasets using co-occurrence networks. In this study, we generated taxon-taxon co-occurrence networks using ten 16S rRNA gene metabarcoding datasets from the Earth Microbiome Project (EMP) that have sampled aquatic environments (both marine and freshwaters). These datasets were individually analysed to create local networks and network modules were used to identify significant co-occurrences of microalgae and bacteria.

Methods

Data acquisition

Publicly available EMP datasets26 were screened using the Qiita portal27 and 10 studies (4 marine and 6 Freshwater environments) targeting aquatic environments were chosen (Figure S1). Since the goal of this study was to individually analyse samples representing a particular environment to generate local networks and reveal co-occurrences, studies targeting any aquatic samples (water and sediment samples) were selected. To our knowledge, based on study and sample information provided on qiita portal for the selected studies, no size fractionation was carried out on the samples (water samples) provided to EMP. The data analysed in this study have amplified the same 16S rRNA gene V4 hypervariable region using 515F- ‘GTGCCAGCMGCCGCGGTAA’ and 806R-‘GGACTACHVGGGTWTCTAAT’ primer pair28,29 and sequenced on Illumina Hiseq2000. Detailed information on the selected EMP projects can be found in Supplementary material (Table S1) and can also be further explored using Qiita ids provided using the Qiita portal.

Sequencing data workflow

Raw demultiplexed reads were downloaded using the EBI accessions (provided in Qiita portal) from the European Nucleotide Archive and analysed individually employing a single bioinformatic pipeline built on Qiime2 version 2019.1030. Primer sequences attached to all reads were trimmed using cutadapt31. Sequence denoising, chimera checking and dereplication was performed in DADA232 to correct sequencing errors and remove low quality bases. Reads were truncated based on a median quality score of 30. The final outputs of DADA2, an abundance table of Amplicon Sequence Variants (ASVs) and fasta sequences of the ASVs were further processed as described hereafter. Taxonomy was assigned against Silva v132 [30] 16S rRNA gene sequences trained with a Naive Bayes classifier33. Chloroplast sequences were filtered using the Silva taxonomy file by identifying sequences assigned as “chloroplast”. These chloroplast sequences were then classified again using a Qiime2-compatible version of PhytoRef34 database accessed at35. All the algal taxonomic identities with PhytoRef had a confidence score of 0.7 or higher (Table S2 and S3). It is important to note that, each algal node in the networks (see below for details) represents a single ASV, likely representing one algal species or strain in the environment.

The DADA2 abundance table was filtered to create a bacteria-only abundance table (by removing mitochondrial, Archaeal and chloroplast ASVs) and a chloroplast-only abundance table (by retaining only the ASV ids assigned as chloroplast). The bacteria-only table was then collapsed at the genus level to reduce weak links in downstream network building36 and was merged with the chloroplast-only table to create the final abundance table. ASVs identified as microalgae were not collapsed since the 16S rRNA gene marker is highly conserved in algae and ASVs likely correspond to species or higher-level taxa34. This final abundance table was filtered to remove low prevalence organisms present in less than 10% of samples to prevent them from introducing artefacts in network inference.

Correlation analysis and network construction

We used a compositional data analysis tool, FastSpar37, which is a rapid and parallelizable implementation of the SparCC algorithm38 to compute correlations. FastSpar quantifies the correlation between all ASVs and assesses the statistical significance (p-values) of the inferred correlations using a bootstrap procedure. The correlation and p-value matrices created by FastSpar were used to create a network of significant correlations for each study separately. Co-occurrence networks were created using the igraph package39 in R studio version 1.4.1106. Undirected weighted networks were created using statistically (p < 0.05) significant correlations > 0.5 or higher (0.5 for freshwater and 0.6 for marine datasets). Coefficient cut-off of 0.5 was selected for freshwater datasets as a value above that resulted in fewer algal nodes in networks compared to 0.5. The idea behind the coefficient cut-off selection was to preserve as many links as possible with microalgae in resulting networks while choosing a considerably higher coefficient cut off to report stronger links. Networks were visualised in Cytoscape version 3.840.

Network analysis and module detection

The Network Analyzer plugin41 in Cytoscape was used to compute global network properties of each network to get an overview of node specific and edge-specific attributes. Networks were checked for the scale free nature by identifying the presence of highly connected nodes coexisting with nodes with fewer links42 using the node attributes. In order to identify communities, networks were clustered using the Cytoscape plugin clusterMaker43 using the MCL clustering algorithm44 with an inflation value of 2.0. Co-occurring microalgae and bacteria were identified using resulting modules (modules with nodes > or = 4). These co-occurring taxa (identified beyond the taxonomic rank Phylum) were recorded for each environment and summarised in heatmaps using Pheatmap v1.0.1245 in R.

Results and discussion

16S rRNA gene-based co-occurrence networks can recover microalgal-bacterial associations

We generated 10 co-occurrences networks representing aquatic environments using publicly available 16S rRNA gene datasets from the EMP. We reduced noise and false positives by including ASVs present in at least 10% of samples, filtering statistically insignificant and weaker correlations. We also used network modules that are indicative of ecological communities to identify potential interactions. In each co-occurrence network generated, we observed highly connected nodes coexisting with nodes with fewer links. This indicates the scale-free nature of the networks, a characteristic in real world networks46,47. The clustered networks (modules) comprised nodes representing microalgae or bacteria and the edges between them were instances of significant co-occurrences.

Analysis of the microalgal-bacterial modules identified 40 algal nodes co-occurring with at least one of 76 bacterial nodes in marine environments and 112 microalgal nodes with at least one of 311 bacterial nodes in freshwater environments (refer Table S2 and S3 for recovered co-occurrences with the correlation values). These significant co-occurrences inferred in marine and freshwater environments are summarised in Figs. 1 and 2, respectively. We identified algal nodes at different taxonomic levels although many could not be classified at lower taxonomic levels such as genus or species. To have consistency in algal node annotations, we have only used class-level classification in figures. Tables S2 and S3 provide full taxonomic affiliations of the nodes where possible. Even though taxonomic assignments were made at higher rank, our ASVs represent taxa at lower taxonomic levels. As the breadth of species included in reference databases such as PhytoRef increases, it is likely that more algal ASVs can be assigned to the species-level.

Figure 1
figure 1

Summarised interactions in marine environments. Correlation heatmap of co-occurring microalgae (columns) and bacteria (rows) in marine environments. “Taxonomy” represents the taxonomic affiliations of microalgal nodes at class level. “Taxonomy2” represents the taxonomic affiliations of bacteria (p_, c_, o_, f_ and g_ represent Phylum, Class, Order, Family and Genus, respectively). “Project” represents the accession numbers of EMP projects and indicate which project the microalgal and bacterial nodes were recovered from. “Sample_origin” indicates if each sample is planktonic or benthic. Heatmap colour gradient indicates correlation coefficients. Refer Table S2 for raw data used to generate the heatmap. Heatmap was generated using Pheatmap v1.0.12 (https://rdrr.io/cran/pheatmap/).

Full size image
Figure 2
figure 2

Summarised interactions in Freshwater environments. Correlation heatmap of co-occurring microalgae (columns) and bacteria (rows) in freshwater environments. “Taxonomy” represents the taxonomic affiliations of microalgal nodes at class level. “Taxonomy2” represents the taxonomic affiliations of bacteria (p_, c_, o_, f_ and g_ represent Phylum, Class, Order, Family and Genus, respectively). “Project” represents the accessions of EMP projects and indicate which project the microalgal and bacterial nodes were recovered from. “Sample_origin” indicates if each sample is planktonic or benthic. Heatmap colour gradient indicates correlation coefficients. Refer Table S3 for raw data used to generate the heatmap. Heatmap was generated using Pheatmap v1.0.12 (https://rdrr.io/cran/pheatmap/).

Full size image

Most modules exhibited high clustering coefficients (> 0.5) indicating that the neighbourhood of microalgal-bacterial communities are generally densely connected. High clustering coefficient values in networks were previously suggested as indicative of cross-feeding relationships and enriched degradation pathways48. We identified potential interactions of bacteria with diverse phyla such as Cryptophyta, Ochrophyta, Haptophyta, Chlorophyta, Streptophyta and Euglenophta representing 17 different taxonomic classes. Bacterial nodes in the marine modules were predominantly represented by Proteobacteria and Bacteroidetes followed by Planctomycetes and Verrucomicrobia. Similar to marine environments Proteobacteria and Bacteroidetes dominated the freshwater modules while Actinobacteria and Verrucomicrobia were the 2nd and 3rd most common bacterial taxa associated with microalgae (Figure S2). We also identified hub nodes with the highest node degree (number of edges connected to a node). Hub nodes represent highly connected nodes and are usually considered as keystone species49. Hub nodes (considering the top ten hubs) in both marine and freshwater modules were mostly represented by Alphaproteobacteria, Gammaproteobacteria (p_Proteobacteria) and Bacteroidia (p_Bacteroidetes). Other than these, members of Planctomycetes (mostly belonging to c_Planctomycetacia) were commonly found among the top hub nodes in freshwater modules. High prevalence and their characteristic associations with microalgae may explain the presence of Proteobacteria and Bacteroidetes as hub nodes in most modules50,51. The role of Planctomycetes in global nitrogen, carbon and sulphur cycles is gaining attention52,53. Thus, microalgal-Planctomycetes associations may be playing a crucial role in the global environmental cycles.

Based on the summarised significant co-occurrences, benthic diatoms exhibited different, and fewer, associations compared to planktonic diatoms (Figs. 1 and 2). For example, bacterial taxa such as unclassified Gammaproteobacteria, Caldilineaceae (Chloroflexi), Trichococcus (Firmicutes), Hoeflea (Proteobacteria), Pseudorhodobacter (Proteobacteria) and Cocleimonas (Proteobacteria) exhibited associations only with marine benthic diatoms and not with marine planktonic diatoms.

We investigated whether this network approach has recovered any known associations identified and confirmed earlier. We recognized known bacterial associates of microalgae such as Flavobacteriales54,55, Rhizobiales56,57, Cytophagales58,59 and Rhodobacterales60. Previously reported associations of microalgal genera with specific bacterial groups were also recovered in the analysis, including that between Bacteroidetes and the cosmopolitan Prymnesiophyceae genus Phaeocystis61,62,63 (Table S2) and that of diatoms (Bacillariophyta) with the genus Flavobacterium (Table S3)64,65,66. The ability to identify previously known associations suggests that the 16S rRNA gene-based network approach used in this study can yield biologically meaningful results.

We also compared co-occurrence networks built in our study with previously published networks of these data. Previous analysis of lake Mendota samples (EBI accession: ERP016591) by Kara et al67 described characteristics of bacterioplankton co-occurrence networks across three seasons. The network properties of these bacterioplankton networks such as clustering coefficient (0.167–0.256) and the characteristic path length (3.27–3.88) were comparatively lower than those of our microalgal-bacterial network (clustering coefficient > 0.5, characteristic path length = 3.965). Similar to the observation by Gilbert et al68, the co-occurrence network built on Western English Channel samples (EBI accession: ERP016541) featured many Alphaproteobacteria and Rhodobacterales nodes.

Both intrinsic and extrinsic factors may influence microalgal-bacterial associations

No clear taxonomic pattern was observed among the co-occurring microalgae and bacteria. In accordance with previous work on phytoplankton-bacteria co-occurrences67, taxonomically diverse bacteria co-occurred with different microalgal groups. Moreover, except for a few, there was no recurrence of bacterial genera (the lowest taxonomic identity of bacterial nodes in a module) across modules generated in each project. Since each project represents a specific environment, these observations indicate that each environment harbours its specific co-occurrence relationships. Project specificity in inferred co-occurrences can clearly be seen in the generated heatmaps (Figs. 1 and 2). Interestingly, a recent study68 has provided evidence for biogeographic differentiation of algal microbiomes, showing ecological boundaries driven by differences in environmental conditions altering the spatial scaling of the algal microbiomes. Another plausible explanation for such observations is the species specificity of algal microbiomes. Although some microalgal nodes could not be taxonomically identified at species level, each microalgal node in a module is an ASV which likely represents a single microalgal species or a strain. Species specificity of algal microbiomes has been predominantly shown using algal cultures50. In addition to supporting previous observations, our results show that species-specificity may be a global characteristic of microalgal microbiomes.

To understand if there is any factor driving general patterns of interactions, some insights may be gleaned from known functions of the co-occurring bacteria. Our observation was that despite their taxonomic differences, bacterial associates in modules often shared functional similarities. For instance, microalgae co-occurred with bacterial groups equipped with specific metabolic functional potentials such as algal polysaccharide degradation and provision of vitamins. A few examples of these are, co-occurrences with taxonomically diverse members of Bacteroidetes63 and Verrucomicrobia69 with polysaccharide degradation ability and Rhizobiales, Rhodobacterales70 and SAR11671,72,73 with Vitamin B12 synthesis. Therefore, our results stemming from observations across multiple datasets suggest that, irrespective of the environment, microalgae are associated with key functional types of bacteria. As most microalgal-bacterial functional interactions remain unknown, it is difficult to identify global trends in key functional types of bacteria associated with microalgal-bacterial communities. This urges the need for more functional studies to improve our understanding of microalgal-bacterial communities across the globe.

Emerging microalgal-bacterial associations that can guide functional studies

An advantage of using a network approach is the ability to analyse large scale datasets to unravel previously unknown associations. We believe that some of the inferred co-occurrences in this study may help guide focused research to shed light on the functional nature of interactions. Therefore, we further explored these co-occurrences which were prominent due to their frequent observations and importance in modules based on topological properties.

For instance, uncultured Deltaproteobacteria order NB1-j was identified as consistently co-occurring with Bacillariophyta in marine environments. In one environment where the NB1-j-Bacillariophyta link was observed (Surface water samples from Western English Channel, EBI accession: ERP016541), Bacillariophyta was the only microalgal node directly interacting with NB1-j. In another environment (Seawater metagenome samples from Catlin Arctic survey (2010 expedition), EBI accession: ERP020022), NB1-j was identified as the hub node with the highest node degree (43), in a module consisting of 77 nodes and 817 edges. This NB1-j node also had the highest closeness centrality (0.69). Closeness centrality measures how close a node is to another node and helps to identify centrally positioned taxa in the network24. Other than this NB1-j node, there were 3 more NB1-j nodes in this community. All together these 4 NB1-j nodes were directly interacting (edges) with 57 nodes (out of 73 other nodes) of which 25 were representing microalgal nodes (Fig. 3). Out of the 25 microalgal nodes, 13 represented the taxonomic class Bacillariophyta. Some of these Bacillariophyta were taxonomically identified at genus level as Chaetoceros and Fragilariopsis. Interestingly, except for one, all these Bacillariophyta nodes represented top hub nodes of the community with node degrees > 26. Most connected taxa are believed to have ecological relevance to the community as their removal causes the highest impact on many associations74.

Figure 3
figure 3

Interactions of the Deltaproteobacterial order NB1-j with microalgae in a marine environment. Figure shows the first neighbours (directly interacting taxa) including microalgal taxa co-occurring with Deltaproteobacterial order NB1-j. Nodes are labelled at the order level (If unclassified at the order level, higher taxonomic affiliations are provided). The edge width and node size are continuously mapped to edge weight (correlation strength) and node degree, respectively. Network image was generated using Cytoscape v3.8 (https://cytoscape.org/).

Full size image

Functional roles of NB1-j in marine ecosystems are largely unknown. It is believed to be involved in hydrocarbon degradation75 and has been reported mostly from marine sediments 70,76, mud volcanoes75, sponges77 and cyanobacterial mats78. To the best of our knowledge, this Deltaproteobacteria group is not commonly reported in algal microbiomes. However, a recent study found NB1-j ASVs associated with the coral skeleton algal symbiont Ostreobium79. A predictive metagenomic approach based on sponge samples originating from reef sites in West Java (Indonesia) suggest that NB1-j may be involved in nitrogen metabolism77. It was found that NB1-j was responsible for the elevated predictive gene count corresponding to N-cycling genes such as those encoding nitric oxide reductase (norB), nitrogenase (nifD) and hydroxylamine reductase. The available information indicates that the nitrogen cycling capacity of NB1-j may underly its association with algae, potentially facilitating the nitrogen needs of the algae while benefiting from algal organic carbon. However, a repertoire of interactions, as often expected from a keystone species, can be hypothesised between NB1-j and microalgae in a community.

Bacillariophyta (Fragilariopsis, Chaetoceros and many other diatoms unidentified at lower taxonomic levels) also showed frequent co-occurrences with an uncultured clade of Planctomycetes, OM190 (SILVA taxonomy) in both freshwater and marine environments. In the Catlin Arctic survey samples (ERP02022), OM190 demonstrated frequent associations with Bacillariophyta. Three OM190 nodes were observed with high node degrees (29, 30 and 39) and these were directly connected to 59 nodes out of a total of 74 other nodes in the community. From the total of 35 microalgal nodes in this community, OM190 were directly connected to 27 nodes out of which 15 nodes were represented by Bacillariophyta (Figure S3A). Similarly in a freshwater environment (Lake Superior, Michigan, EBI accession: ERP016492), OM190 was identified as the 6th most connected node with a node degree of 89 and was directly connected to 32 microalgal nodes including Bacillariophyta (Figure S3B). As mentioned previously, highly connected nodes act as hub nodes in the community. Multiple OM190 acting as top hub nodes within their habitats indicate that they have high co-occurrences with both microalgae and bacteria.

OM190 shows deep branching within the Planctomycetes group and is usually found in different environments such as soil and seawater. Members of this clade are usually considered to be associated with macroalgae80, such as red81 and brown algae82. Since OM190 is yet uncultured, information on its metabolism is scarce. A metagenome-assembled-genome (MAG) for OM190 (likely OM190) with a rich diversity of secondary metabolite potential has been reported83. The production of secondary metabolites by OM190, including antimicrobial compounds may be one of the underlying reasons for their association with algae as they could protect the alga from undesired microbes. In a recent study84, it was shown that diatoms produce fucose-containing sulfated polysaccharides (FCSP) which can be hypothesised as an energy source for OM190. Many Planctomycetes have abundant sulfatases, but these were not confirmed in the existing OM190 MAG. In addition to their association with algae, OM190 and NB1-j seem to have a close relationship with one another. They were recently reported as abundant co-occurring taxa in Beaufort Sea surface sediments85. The reasons for their direct associations are not known and, along with their respective relationships with algae, this direct interaction is an interesting topic for future investigation.

More generally speaking, Planctomycetes often associate with macroalgae and favour a biofilm lifestyle82. Protein clusters that may be involved in Planctomycetes symbiosis or biofilm maintenance have been reported53, and production of sulfated polysaccharides by the alga that serve as the substrate for the abundant sulfatases produced by the Planctomycetes contributes to the reasons for successful associations between them53,80. Although successful associations are known between macroalgae and Planctomycetes, interactions with microalgae are largely unknown. Apart from their associations with diatoms, our network analysis identified taxa representing Planctomycetes (c_Phycisphaerae, c_Planctomycetacia) co-occurring with a range of microalgal groups. In a previous study, Planctomycetes closely related to Pirellula were identified as one of the dominant lineages associated with diatom blooms86. Results of our network analysis demonstrate that associations between Planctomycetes and microalgae may be as common as those they maintain with macroalgae.

Verrucomicrobia were also consistently associated with an array of microalgae in fresh and marine environments indicating that they may be common associates of microalgae. Here in our study, all the Verrucomicrobial members co-occurring with microalgae were represented by the taxonomic class Verrucomicrobiae. Among the genus-specific associations revealed in our network analysis is that of the Verrucomicrobial genus Lentimonas and microalgal genera Pyramimonas (c_Prasinophyceae), Phaeocystis, Tisochrysis, Haptolina, and Chrysochromulina (c_Prymnesiophyceae). A recent study showed hundreds of Lentimonas enzymes able to digest brown algal fucoidan87. Although fucose-containing sulphated polysaccharides were regarded as a macroalgal polysaccharide, microalgae such as diatoms have the potential to produce them84. We speculate that other microalgae might also produce this sulfated polysaccharide, or that members of Lentimonas may have a broader palate than just FCSP. Although widely distributed, the functional roles of Verrucomicrobia in aquatic ecosystems are not well understood due to the lack of cultured strains88. Some members of Verrucomicrobia have been shown to consume algal extracellular polymeric substances69,89,90. However, other factors contributing to their associations are not well understood.

Conclusion

In summary, our study illustrates the promise of using 16S rRNA gene-based co-occurrence networks as a hypothesis-generating framework to guide focused research and speculate on the functional nature of potential interactions. By studying multiple environments based on public datasets, we provided an overview of microalgal-bacterial communities in aquatic ecosystems from a network perspective. This identified a range of associations including previously unknown links that can set the stage for more focused research in the future.