Bacterial community assembly based on functional genes rather than species

Abstract

The principles underlying the assembly and structure of complex microbial communities are an issue of long-standing concern to the field of microbial ecology. We previously analyzed the community membership of bacterial communities associated with the green macroalga Ulva australis, and proposed a competitive lottery model for colonization of the algal surface in an attempt to explain the surprising lack of similarity in species composition across different algal samples. Here we extend the previous study by investigating the link between community structure and function in these communities, using metagenomic sequence analysis. Despite the high phylogenetic variability in microbial species composition on different U. australis (only 15% similarity between samples), similarity in functional composition was high (70%), and a core of functional genes present across all algal-associated communities was identified that were consistent with the ecology of surface- and host-associated bacteria. These functions were distributed widely across a variety of taxa or phylogenetic groups. This observation of similarity in habitat (niche) use with respect to functional genes, but not species, together with the relative ease with which bacteria share genetic material, suggests that the key level at which to address the assembly and structure of bacterial communities may not be “species” (by means of rRNA taxonomy), but rather the more functional level of genes.

Metagenomic analysis of environmental microbial communities has revealed an enormous and previously unknown microbial diversity, and expanded our knowledge of their function in a variety of environments (1–5). Much still remains unknown, however, such as the principles underlying the assembly and structure of complex microbial communities, an issue of long-standing concern to the field of microbial ecology. To this aim, several recent studies have supported the “neutral hypothesis” (6–8), a largely stochastic model for community assembly, which assumes that species are ecologically equivalent and that community structure is determined by random processes (9, 10). However, there is also evidence that niche or deterministic processes play a role in community structure (11, 12); thus, both niche and neutral processes are likely to affect the assembly of complex microbial communities.

Support for these models is based on species abundance distributions, and critical functional aspects, such as the assumption of ecological equivalence, have for the most part not been tested. In this study, we examine the encoded functions of an algal-associated bacterial community and link patterns of function to patterns of community assembly. Following the results of an earlier study (13), we investigate these communities in the context of the lottery hypothesis, a model for community “assembly” derived from studies of eukaryotic communities, such as coral reef fish (14). This hypothesis incorporates both neutral and functional aspects and argues that ecological niches are colonized randomly from a pool of species with similar ecological function that can coexist in that niche (14, 15). Available space within that niche is colonized by whichever suitable species happens to arrive there first, meaning that colonization of space is random from within a functionally equivalent group of species. In the context of a bacterial community, this model implies that there are guilds of bacterial species, whose members are functionally equivalent with respect to their ability to colonize a particular niche (e.g., the surface of the seaweed Ulva australis), but that the composition of species in any particular community (e.g., a single U. australis individual) is determined stochastically by recruitment from within those guilds. Importantly, members of a guild can be phylogenetically related or unrelated. If this model is correct, different species from within these guilds should share functional traits, and a core suite of functional genes should be consistently present in all communities of a particular habitat, independent of the taxonomic or phylogenetic composition of its species.

In an earlier study (13) we characterized the bacterial phylogenetic diversity of seawater and U. australis, a member of a common green algal family often found in tidal rock pools or shelves around the world. We found that algal-associated communities were highly distinct from the surrounding seawater communities, but were also highly variable among individual algal samples, with only six operational taxonomic units (of a total of 528) at a 97% sequence identity cut-off occurring on all samples (13). This finding means that each U. australis sample hosts a unique assemblage of species (as defined by 97% 16S rRNA similarity). Given that the recruitment of new community members onto U. australis is most likely to come from the seawater, these results are somewhat contradictory with respect to dominant models of community assembly: the differences between seawater and algal communities imply selective mechanisms of assembly on the algal surface (niche partitioning), and the high variability between algal hosts is consistent with random colonization (e.g., neutral processes).

Here, we analyze the metagenomes of these communities to show that the algal-associated bacterial communities are functionally distinct from seawater communities, but contain a core of functional genes, which are represented across all algal samples. These functions are consistent with the ecology of surface- and host-associated bacteria, and importantly are distributed across a variety of taxa in individual communities, indicating functional redundancy across taxa. This mix of functional and random processes is consistent with the predictions of the competitive lottery model for the assembly of complex microbial communities. Moreover, this functional (niche) partitioning with respect to genes, but not phylogeny, in these assemblages highlights the potential difficulty in using bacterial species to test hypotheses derived from eukaryotic ecology, which focus on species as the critical unit. Given the relative ease with which bacteria share genetic material, the key level at which to address the assembly of these bacterial communities may not be species, but rather the more functional level of genes or gene clusters.

Results and Discussion

Algal-Associated Bacterial Communities Encode a Distinct Functional Profile.

We generated over 681 Mbp of metagenomic shotgun sequencing data from six algal (UA1–UA6) and eight seawater samples (SW3–SW10) (see Materials and Methods and Table S1). Environmental parameters were similar for each sample (see Materials and Methods and Table S2) and we found no evidence for increased viral or eukaryotic DNA between algal samples, indicating that these groups were a minor part of the microbial community at the time of sampling. Chao 1 estimates showed very similar levels of bacterial species diversity (ranging from 225 to 451) for the six algal communities (13). U. australis specimens were in the same developmental stage (i.e., fully matured) and the simple, two-cell layer alga is depauperate in bioactives found in many other algae, making it an ideal choice to minimize effects of host variability. Overall, these observations led us to the conclusion that bacterial communities on the U. australis samples collected existed under similar broad ecological conditions.

Sequencing data were used to create functional community profiles for each sample based on Clusters of Orthologous Groups (COG) (16) and SEED (17) annotations (details in Table S1). This process revealed that the algal-associated communities were functionally distinct from those in the surrounding seawater. Multidimensional scaling (MDS) plots show that the U. australis samples clustered together, and separately from seawater samples (Fig. 1A), and permutational multivariate analysis of variance (PERMANOVA) indicates that this difference is significant (P = 0.001). This distinctiveness was quantitative and multivariate, rather than qualitative, and there were few functions that were consistently present in one environment and consistently absent in the other. As such, the differences between the two environments lie predominantly in the relative abundance of particular functions as defined by COG and SEED annotations. Variation was also observed among seawater or U. australis samples, but this was primarily seen for those samples with the least (order-of-magnitude less) sequence data. When these samples were removed, the remaining algal samples still clustered together quite tightly (Fig. 1B). Bray-Curtis similarity shows that on average there was 15% similarity in species composition across samples (13), compared with 70% similarity in functional composition (COGs), indicating that despite large differences in species composition between hosts [as detected in our previous study (13)], many encoded functions are shared.

Fig. 1.

MDS plots based on Bray-Curtis similarity of functional gene profiles based on COG and SEED annotations of U. australis and planktonic seawater metagenomes. (A) MDS plots for all samples, including UA4, UA6, SW4, SW6, SW8, and SW10, which contained an order-of-magnitude less sequencing than the remaining samples. (B) MDS plots with low coverage samples removed. Samples from each of the respective environments cluster together based on their functional profile.

Functional Core in U. australis-Associated Bacterial Communities.

We observed a set of COG and SEED functions that contributed strongly to the difference between the two community types (Fig. S1 and SI Materials and Methods), and that were consistently abundant across U. australis samples. These were defined as the core functions of the U. australis associated bacterial communities (Fig. 2). These functions fit well with the current understanding of the ecology of an algal- or surface-associated bacterial community and could mostly be grouped into broader categories (Table S3) and are summarized here.

Fig. 2.

COG (A) and SEED subsystems (B), which comprise the characteristic functions of the U. australis community, by comparison with planktonic seawater, with SD across the six algal and eight seawater samples analyzed. COGs and SEED subsystems are presented in the order of their contribution to difference (highest to lowest, Top to Bottom) as assessed by SIMPER analysis.

Detection and movement toward the host surface.

Proteins associated with chemotaxis and flagellum-mediated motility were consistently abundant in the U. australis community, and are likely to be important for the detection and movement toward the algal host surface during colonization. Chemotaxis is essential for the development and maintenance of symbiotic, surface associations, as for example in symbiotic Rhizobium sp., which are chemotactically attracted to the flavenoids that induce the nodulation genes necessary for nitrogen fixation in the host plant (18). Flagella-mediated motility is also important for biofilm formation in a range of bacteria (19–22).

Attachment and biofilm formation.

An array of proteins functionally assigned to attachment and biofilm formation were overrepresented across U. australis samples. Functions include homologs of the OmpA protein (COG2197), which is required for adhesion to both mammalian and fish epithelial cells in a range of Proteobacteria (23, 24), Listeria internalin-like proteins, which enhance attachment and biofilm formation (25, 26), and the widespread colonization island, which is essential for biofilm formation, colonization, and pathogenesis in a range of bacteria (27). Proteins related to the production and excretion of galactoglycan, or exopolysaccharide II, were more abundant in the U. australis community, and apart from forming part of the biofilm matrix, is also essential for the establishment and maintenance of symbiosis in several Rhizobium strains (28–32). GGDEF and EAL domain proteins, which are involved in the production and degradation, respectively, of bis-(3′-5′)-cyclic dimeric GMP (cyclic-di-GMP), were also detected at a higher abundance (33, 34). Cyclic-di-GMP is an important secondary messenger, regulating the transition from a motile planktonic to a surface-associated biofilm lifestyle in a range of bacteria (34, 35), for example by up-regulating the production of adhesins and biofilm matrix components (36–39) or downregulating motility genes (34).

Genes encoding for Cbb3-type cytochrome c oxidases, which have a high affinity for oxygen and are associated with microaerobic metabolism in oxygen-limited environments (40), were overrepresented, as well as the function of nitrite and nitrate ammonification, considered the highest energy-yielding respiration systems after oxygen has been depleted (41). The higher abundance of proteins assigned these functions may be related to survival in biofilms, which are known to be spatially heterogeneous, containing pockets of low or no oxygen in some areas (42).

Response to the algal host environment.

Some of the overrepresented functions can be related directly to U. australis’ surface environment. For example, proteins associated with the metabolism of water-soluble polysaccharides produced by Ulva sp., such as rhamnose, xylose, glucose, mannose, and galactose (43), would enable bacteria to use these sugars as a source of carbon and energy and, hence, gain a competitive advantage in colonizing the host surface. Proteins associated with heat and osmotic stress (e.g., COG0668: small conductance mechanosensitive channel) (44–46) may be related to the desiccation of U. australis fronds, which occur in their intertidal habitats at low tide. Macroalgae also defend themselves against bacteria and herbivores by the release of reactive oxygen species, an oxidative burst (47), and the overrepresentation of proteins controlling oxidative stress might represent a protective mechanism for the surface community. Finally, Ulva sp. is known to take up and store heavy metals (48), and the overrepresentation of proteins associated with the export of heavy metals could thus be related to the presence of heavy metals in the algal host.

Regulation in response to environmental stimuli.

Some of the most overrepresented functions found in the microbial community of U. australis involve regulatory mechanisms in response to environmental stimuli. The high proportion of environmental signal transducers and transcriptional regulators could be indicative of the need to respond to changes in the host environment (e.g., osmotic; see above), to control the complex steps of colonization or biofilm formation and to mediate the interactions with other community members. For example, there are many homologs of COG0642 (histidine kinase), members of which are known to be involved in osmoregulation (49), multidrug export (50), sporulation (51), nitrate reduction (52), cell differentiation (53), and plant virulence (54). Also overrepresented is COG0583 (transcriptional regulator), which contains proteins that regulate transcription in response to plant exudates (55), virulence, motility, and quorum sensing (56). There are many examples of plants and their metabolites affecting gene regulation of associated bacteria (57–61), and it is likely that such regulators play an important part in mediating interactions between U. australis and the surface community.

Lateral gene transfer.

Type IV secretion system (T4SS), transposases, and intron-associated genes were overrepresented in U. australis samples. These functions are associated with lateral gene transfer, a source of dynamic genomic change that allows for rapid ecological adaptation (62), and which would provide a broad mechanism for facilitating the functional similarity of phylogenetically distinct bacteria on the surface of U. australis (see below). Biofilms are ideal environments for lateral gene transfer (63), and an abundance of transposases has been noted in other biofilm communities (64). A recent survey of the transposase families in different taxa suggested that transposases are most often transferred to other organisms within the same habitat and can be shared by distantly related taxa (65), and COG2801 (putative transposase) is a potential source of shared functional traits among different taxa in the U. australis bacterial community. Although T4SSs are also often associated with virulent host/bacterium interactions (66), this system can also mediate symbiotic interactions (67). For example, Mesorhizobium loti R7A encodes a “symbiosis island,” containing a T4SS homologous to the Vir pilus from Agrobacterium tumefaciens. It is believed that this T4SS transfers effector proteins into host plant cells (68). Although speculative, it is plausible that T4SS could be used by the bacterial community of U. australis to mediate symbiotic interactions, such as the transfer of compounds inducing correct morphology of the alga (69).

Defense.

Genes related to defense include multidrug transport, restriction modification systems, nonribosomal peptide synthase modules, and ABC transporters, which have reported links with virulence (70, 71), and are also abundant in the algal surface community. Algal-associated bacteria may protect the host by inhibiting the attachment of other bacteria and biofouling organisms through the production of secondary metabolites (71). Toxic and antibiotic compounds may be transported out of the cell via homologs of ABC transporters known to export multiple drugs (72). The presence of restriction modification systems indicates a need to minimize transduction or transformation, or to only allow for genetic exchange between bacteria that have similar restriction modification systems. All these functions can clearly contribute to the maintenance of the structural and genetic integrity and stability of the surface community, which is of particular importance in a system that is constantly exposed to a large number of potentially invading bacteria (e.g., planktonic secondary colonizers).

These functions are all consistent with the ecological role of the U. australis community and together provide support for the notion of a specific and stable core metagenome, which is functionally adapted to life on the alga’s surface.

Colonization of U. australis: Does Taxonomy Reflect Function?

In eukaryotic models of colonization via a competitive lottery, functional groupings (or “guilds”) often reflect taxonomic groupings (15, 73–76). Alternatively, there may be taxonomic redundancy, in which any given function is distributed broadly across a variety of taxa as opposed to being associated with any particular taxonomic group. To address this question of taxonomic redundancy, amino acid sequences of proteins assigned to six functions, which contributed most to the difference between algal and seawater samples, were analyzed using phylogenetic tree comparisons in Unifrac (77) and a taxonomic last common ancestor algorithm (MEGAN) (78) (see Materials and Methods). A large degree of phylogenetic dissimilarity of the core functions was exemplified by pairwise comparisons of samples, displaying between 65% and 97% dissimilarity (Table S4), indicating that the protein phylogenies for each core function were distinct in each sample. Unifrac P values indicated that for five core functions, the phylogenetic distribution of proteins between samples was not significantly different to that expected by chance (P values were > 0.05). The only exception was the pairwise dissimilarities of as little as 52% at a P value of 0.03 for COG2801 (putative transposase). Transposases are part of mobile genetic elements and therefore are likely transferred between members of the different communities. This finding would make those communities more similar and the differences appear less random with respect to the function of COG2801.

Taxonomic assignment with MEGAN further showed that the core functions were generally widely distributed across major bacterial groups present on U. australis, namely the α- and γ-proteobacteria, Bacteroidetes, and Planctomycetes, suggesting that a broad range of bacteria possess the ability to carry out these functions (Figs. S2–S7). At the level of species or genus, proteins were distributed across a variety of taxa, which differed from sample to sample, and often a function was assigned to a species that was present in one sample only (although it should be noted that proteins are not necessarily from the specific species assigned, but have high similarity to the homologous protein from that species). This finding means that protein functions are derived from distinct lineages that can only be crudely assigned to very high-level taxa (e.g., phyla) or, when a low-level assignment (e.g., species/genus) is possible, there is little overlap between samples. Together the result of the phylogenetic and taxonomic analysis of proteins support the assertion that the core functions are not restricted to a particular taxonomic group. This result also means that different taxa provide the core functions to the community and that members from different taxa could form a functional guild.

Structure of Bacterial Communities: Assembly of Functional Genes or Assembly of Species?

Although our samples are taken over a limited time scale, and thus do not fully accommodate potential successional or historic changes in these communities, the evidence presented here and in our previous study (13) is most consistent with a competitive lottery model for community assembly on the surface of U. australis. Originally proposed for coral reef fish (15, 73), and subsequently applied to plant (74, 75) and parasite communities (76), the lottery model combines functional (niche- or guild-based) and random components as drivers of community structure. Specifically, species with similar trophic or other ecological properties are able to occupy the same niche within an ecosystem, and the particular species that occupies a particular space is then determined by stochastic recruitment. This means that within a group of species with similar ecologies, the “lottery” for space is won by whoever gets there first (15, 79).

In our system, as long as a bacterium has the necessary functional characters (defined here as particular gene functions) to colonize or grow on U. australis, the specific assemblage of species present on any given algal surface is stochastic, determined by which members of the guild happen to be available to colonize from the water column when space becomes available. Although the lottery model was originally proposed for relatively phylogenetically narrow groups of organisms (e.g., families of coral reef fish), we argue here that it also explains community assembly when the species pool spans multiple bacterial phyla.

More broadly and independently of any specific theory of community assembly, our results imply that genes or gene clusters are as or more important than species for understanding community assembly in bacterial systems. Dawkins (80) has famously argued in the context of evolutionary theory that individuals, and by inference collections of individuals such as species, are essentially containers for collections of genes. This approach contrasts with ecological models for community assembly, in which species (or higher taxonomic levels) are the fundamental metric. Such models are largely drawn from studies of eukaryotes, and so this species-focused approach is not surprising, given the general assumption of substantial genetic coherence within eukaryotic species.

However, the utility of the species concept for bacteria has been challenged in a number of ways (81–83), ranging from the level of genetic similarity necessary to define a species, to the extent (or lack thereof) of the genomic coherence of “species,” because of the occurrence of substantial genetic exchange among taxonomically distinct bacteria. Indeed, in our system this frequent genetic exchange was supported by the abundance of functions for horizontal gene transfer (e.g., type IV secretion and transposase). We also found that analysis of functional gene systems revealed a considerable biological pattern that was not evident by focusing only on patterns of species diversity (13). Similar observations have been made for the human microbiome project (84, 85). This finding has at least two implications for studies of microbial community ecology. First, tests of community assembly theory [e.g., neutral theory (8)] using species as the key parameter may be misleading, because, if function does not map onto taxonomy, then accumulation or assembly of species will always appear random. Second, almost all of the theory currently used to understand patterns of microbial diversity is derived from eukaryotic ecology. However, our results raise the general possibility that genes and their functions may be more useful in testing models of community ecology for bacterial communities. This may be one important way in which the ecology of bacteria differs from that of eukaryotes.

Conclusion

This study provides insight into the link between community structure and function for a complex, algal-associated bacterial community. This community contains a consistent functional profile, with features related to a host-associated lifestyle, and functional similarities exist within phylogenetically distinct members from different host individuals. Although the community members on U. australis contain functional similarities, we do not yet know whether they form guilds that are specific to U. australis, or whether they are more generally adapted to other living or inanimate surfaces. Nevertheless, our evidence is consistent with community assembly via a competitive lottery mechanism. This model encompasses both selective and neutral processes, and could apply to other complex host-associated microbial communities, such as the human microbiome, where a consistent core of functional genes is detected across hosts (85), but species-level community composition is highly variable (86). The lack of correspondence between function (as determined by functional gene systems) and phylogeny in this system suggests that genes, rather than species, may be the appropriate parameter for understanding patterns of diversity in many microbial communities.

Materials and Methods

Sampling.

Seawater was collected from Botany Bay (SW3 and SW4: 33°59′S, 151°14′E) on the 3rd of January 2005 and on the 19th of January 2005 from Bare Island (SW5 and SW6: 33°59′S, 151°13′E) and 200 m away in Botany Bay (SW7 and SW8). Seawater was collected again from Bare Island (SW9 and SW10) at the 18th of October 2006 to coincide with the sampling of U. australis at this site. Two-hundred liters of seawater were collected for each sample (exception is 100 L each for SW9 and SW10) from a depth of 2 m, and immediately serially filtered through 20-, 3-, 0.8-, and 0.1-μm filters. U. australis thalli were collected (wet weight: 20 g per sample) from two different rock pools at Bare Island in October 2006 (UA1 and UA2: 33°59′S, 151°13′E) and again on the 7th of February 2007 (UA3 and UA4). Thalli were also collected from two different rock pools ∼9 km away at Shark Point, Clovelly, on the 7th of February 2007 (UA5 and UA6: 33°91′S, 151°26′E). Sampling was performed between 10:00 AM and 12:00 PM after outgoing tides to ensure rock pools were well flushed with the surrounding seawater, and only algae of the same approximate size were collected to ensure that they were in the same developmental stage of their life cycle. Water parameters were also measured (Table S2) and showed similar values across the years. The sampling design and environmental measurements suggest that each sample of U. australis was subjected to similar environmental conditions.

DNA Sequencing, Assembly, and Annotation.

DNA was extracted from the six algal (UA1–UA6) and eight seawater samples (SW3–SW10), which correspond to samples from ref. 13. Bacterial DNA was extracted from the surface of the algal fronds as described previously (87), which leaves the algal host intact and extracts total DNA from the entire surface community. For the filtered seawater samples, DNA was extracted from the 0.1-μm filter, as previously described (88).

Large-scale shotgun sequencing was performed and sequences were quality-filtered, assembled, and annotated (details in SI Materials and Methods). Each ORF was searched against the COG database (16) and a matrix of raw counts per COG and per sample was generated. Data were also submitted to the MG-RAST server (89) in the form of unassembled individual reads. Matches to the SEED database (17) with an e-value of less than 10⁻²⁰ and a minimum alignment length of 100 amino acids were used in a matrix of counts per SEED subsystem per sample.

Statistical Analysis of Metagenomic Datasets and Core Functions.

Matrices of raw counts of COG and SEED functional annotations per U. australis or seawater sample were standardized to account for the unequal sequence coverage between samples. Bray-Curtis similarity matrices were calculated and used to generate MDS plots. PERMANOVA (90) were carried out to compare samples from each environment. Similarity percentage analyses (SIMPER) (91) were carried out to determine the contribution of each COG or SEED subsystem to similarities within, and difference between, environments. The PRIMER-6 package was used for all multivariate statistical analysis (92).

The top 50 COG and SEED subsystems which contributed most to the differences between the two environments (as assessed by SIMPER), were selected for further analysis. After removal of those, which were not consistently more abundant in the algal dataset, the remaining 36 COG and 38 SEED subsystems were defined as comprising the core functions of the U. australis bacterial community. These functions were grouped into broader categories with respect to a surface-associated lifestyle with a living algal host (see SI Materials and Methods for details).

Assessment of Phylogenetic Similarity and Taxonomic Orgin of Core Functions.

Protein sequences from each U. australis sample for the six COGs, which contributed most to the difference from seawater libraries, were aligned in ClustalW 1.83 (93) and maximum likelihood trees were built in RAxML 7.0.4 (94) using 100 bootstraps. Tree files were analyzed with the Unifrac Webserver (http://bmf2.colorado.edu/unifrac/) (77) to calculate the proportion of branch length, which is unique or shared in each environment. The unweighted algorithm was used as each protein sequence was unique. Sequences were also compared against the National Center for Biotechnology Information nonredundant protein database using BLASTP (95) and analyzed with MEGAN 3.7 (78), which uses a last common ancestor algorithm to assign a likely taxonomic origin to each sequence. The predicted origins of proteins were compared to determine if particular functions were associated with particular taxa. Only sequences from samples UA1, UA2, UA3, and UA5 were used in MEGAN analysis; samples UA4 and UA6 contained fewer than 10 sequences for each function because of the lower depth of sequence coverage obtained.