Background

The ability of microbes to colonize different niches, i.e., their niche breadth, determines the composition and therefore networking of the microbial assembly when challenged by various environmental forces [1]. Generalist microbes are capable of occupying various niches, while specialists are those with a narrower habitat, i.e., found in fewer niches. Specialization can also be determined as unique if the microbe appears consistently only in a particular niche, condition or occasion [2, 3]. With the constant development of new tools for analyzing microbial communities, specific platforms for network analyses have also emerged. These tools allow the discrimination of association networks following their topologies as to density, clustering rate, diameter, length of shortest path, and other indices [4]. Accordingly, the organizational level of the community can be highly random, small-world, or scale-free [4, 5]. In random networks, the connectivity of microbe nodes is not easily disrupted since many microbial nodes can be reached by any other node through a short number of steps, and the associating nodes are more likely to be neighbors of each other [6]. A closer investigation into the different nodes in an assembly enables us to reveal the key hub microbes with the highest centrality and connectivity, as demonstrated by a fair number of edges (associations) and their proximity to other microbe nodes [7]. Likewise, driver microbes with the greatest contribution to the rewiring of a network under a certain weight (i.e., niche) can be identified by the changes in their associating members, type of association, and centrality [8].

Among numerous environmental forces, diet is considered to influence the composition of the gut-microbe assembly (GMA) of various aquatic invertebrates [9]. A particularly strong effect may be attained if animals are restricted to only one particular nutrient source. A prime example is sea urchins that can gain their energy from a single, usually preferred, algal species [10, 11]. In many cases such preference has been credited with positive impacts on performance, e.g., growth, gonad color and somatic index [12, 13].

Evidence that sea urchins that can be cultured when fed different macroalgae, each as a sole energy source, are favorable for study of the responses of GMA and microbial networking to changes in diet. Toward this goal we selected the sea urchin Tripneustes gratilla elatensis as a model algivorous species to examine whether and how diet restriction to a single species of alga, either Ulva fasciata (Chlorophyta) or Gracilaria conferta (Rhodophyta), will affect the gut microbial assembly compared to a plant-rich but algal-depleted pelleted diet. Not only changes in the composition and networking that could occur at the level of the entire assembly were considered, but also specific changes in the associations and centrality in the network of specific microbes in the assembly. It was further hypothesized that such changes may be more prominent in one of the three major regions in the gut (esophagus, stomach, or intestine) rather than over the entire organ. Hence, the experiment aimed at identification of the key microbes or microbe taxa in the gut of T. gratilla elatensis through their generalism or specialism in the different niches examined (diet types and gut regions) as per their occurrence in broader or narrower niches and also their content of genes for carbohydrates decomposition as potential contribution in feed metabolism.

Methods

Sea urchin cultivation with experimental diets

Fifty-four 6-month-old individuals of Tripneustes gratilla elatensis from a single hatch at the National Center for Mariculture (NCM, Eilat, Israel), were assigned for the feeding trial in an in-house experimental system which allows control of environmental conditions including the recommended ambient light for studies of this species [14]. This system comprises nine 90L rectangular tanks and allows maintenance of water quality, temperature, and oxygen saturation via a rapid flow of fresh filtered sea water, as well as continuous aeration of the tanks. Prior to restricting the diet to a given experimental feed treatment, all sea urchins were fed the algal diet of Ulva fasciata together with Gracilaria conferta recommended at the local integrated multi-trophic aquaculture system (IMTA). After two weeks of acclimation, homogeny in the physical parameters of body (‘test’) diameter (85.2 ± 2 mm) and weight (272.4 ± 17 g wet weight) between individuals was validated.

The experimental diets were comprised solely of either fresh Ulva fasciata (‘Ulva diet’) or fresh Gracilaria conferta (‘Gracilaria diet’) grown in the local IMTA system, or a pelleted feed made of plant-meals that lack algae or any marine footprint but contain whole wheat, rapeseed, soya pulp, corn starch and other essential minerals and vitamins (Additional file 1: Table S1). These diets differ greatly in their polysaccharide content. In Gracilaria, galactans are the primary polysaccharides, classified as agarans and carrageenans, and the latter also consist of sulfate ester groups in their repeating disaccharides in the poly-chain [15]. In Ulva, sulfated ulvans are the main cell-wall polysaccharides and their content in dry biomass may reach 38–54% [16]. In contrast to these marine polysaccharides in the algal diets, the primary polysaccharides in the pelleted feed used are plant cellulases and starch. The feeding trial was set in a completely randomized block design with triplicate tanks for each treatment containing nine individuals each, all of which were nourished ad libitum for a period of eight weeks. Feeding ended forty-eight hours prior to examination of the sea urchins in order to minimize remnant digesta in the gut. The sea urchins’ initial and final diameter, weight (dry and wet weight), and gonad weight, were measured in three randomly selected individuals in each tank. The specific growth rate and gonadosomatic index were then calculated as described elsewhere [12].

DNA extraction and sequencing

Three sea urchins from each tank (nine per diet type) were sacrificed for DNA sampling from the different gut regions. The tubular digestive tract was removed and separated into its three major regions [17] of the esophagus, which is closest to the mouth and where ingested food is covered by a mucous layer [18], and the anterior stomach and posterior intestine where secretion of digestive enzymes and nutrient absorption take place, respectively [19, 20]. Dissection was performed using sterile instruments, while residual digesta was removed, and the remaining tissue was rinsed gently with Ultra-Pure water (Additional file 1: Fig. S1a-d). Gut samples were homogenized in FastPrep-24™ 5G homogenizer (MP Biomedicals) and gDNA was extracted using the PureLink™ Microbiome DNA Purification Kit (Thermo Fisher Scientific, USA) following the manufacturer's protocol. Only DNA of high quality and quantity was amplified with the 515F (GTGYCAGCMGCCGCGGTAA) and 926R (CCGYCAATTYMTTTRAGTTT) primer set [21] by denaturation at 95 °C for 5 min, 28 cycles at 94 °C for 45 s, 50 °C for 60 s, 72 °C for 90 s, and a final elongation at 72 °C for 10 min. Bacterial 16S rRNA gene sequencing was performed at the Research Resources Center (University of Illinois, Chicago) by the Illumina MiSeq platform [22] using the reagent kit v3. Quality and quantity assurance included examination of negative and positive control samples (using fecal samples as positive control) on which all procedures both before and after sequencing were performed.

Data analysis

Analyses of GMA were performed on 15, 18, and 21 samples from urchins fed with diets of Ulva, algae-free pellets, and Gracilaria, respectively. Different gut regions were represented equally in each sample set. Bioinformatics analysis was performed using the Quantitative Insights into Microbial Ecology pipeline (QIIME; version1.9.1) [23]. After removing primers and linkers, pair-end reads were merged using PEAR [24], and sequences with a quality score below 30 or shorter than 300 bp were filtered out. Sequences were clustered into operational taxonomic units (OTUs) at 97% similarity cutoff using UCLUST [25] and the Silva database (ver. 132) [26], and chimera sequences were removed by ChimeraSlayer [27]. A subsequent alignment step by PyNAST [28] identified master sequences according to the most abundant criteria. Taxonomy assignment was established using the UCLUST classifier while discarding OTUs represented by less than five sequences or those identified as chloroplast or mitochondria. A Linear discriminant analysis Effect Size (LEfSe) [29] at false discovery rate (FDR) P < 0.01 was performed to identify bacterial taxa that can be considered as biomarkers of a certain diet or gut region.

Each of the microbes in the assembly was examined separately to determine whether it falls under the definition of core or unique microbe and secondly as an either generalist or specialist microbe. Core microbes were identified as those that appeared in at least 85% of samples of any specific habitat within the examined variable of diet, gut region, or both. On the other end, unique microbes were those that appeared in at least 85% of the samples of a specific gut region or diet but in no more than 25% of the samples of each of the other two diets or gut regions. Following this classification, niche breadth of each of the microbes was measured as recommended [30] by ranking OTU evenness of occurrence across various habitats (Shannon–Weaver diversity) [31], occupancy across habitats (Richness) [32], and the Levin's index [33]. Microbes with an overall rank of niche breadth within the top or bottom decile were defined as generalists or specialists, respectively, while all other microbes were defined as non-significant. Microbial association networks were analyzed using MetagenoNets [34] and the latent variable model Correlation interference for Compositional data through Lasso (CCLasso) [35] after log ratio transformation and measurement of edge weight (significant level of P < 0.005 and permutation of 200) for each node pair, excluding OTUs with low prevalence (< 0.001%) and occurrence (< 10%). Topology indices of density, diameter, cluster coefficient, average path length, and Jaccard edge index were analyzed for networks and participating nodes by Cytoscape [36]. The randomization level of association networks was evaluated following the common distribution model of nodes’ number of degrees [37]. Following the model, a random network will reveal Poisson distribution with many of the nodes having about the same number of degrees, while in a scale-free network a power-tail distribution is formed by a high number of nodes that have few degrees and vice versa. Hub nodes were identified using CentiServer [38] following their high degrees (edges), proximity centrality (short distance to any other node), between-ness (fraction of cases when node is in-between shortest track of all other pair nodes) and the Kleinberg's hub centrality [39] score as recommended [7, 40]. In addition to hub taxa, driver nodes are defined as those leading to rewiring of the association network under a specific treatment. Therefore, they present a low proportion of shared edges with other networks (Jaccard edge index; JEI), as well as exclusive enrichment of interacting partners (neighbor shift score; NESH) and a greater centrality (i.e., greater occurrence frequency in paths of other nodes) as compared to other nets (∆B). Such nodes were extracted from data using NetShift [8] employing a combined threshold of JEI < 0.3, NESH > 1.8, and a non-zero ∆B which was validated via pair-wised tests against each of the other two networks (diets or gut regions). Since identification was carried out in sets of paired networks, it should be noted that driver nodes identified here achieved the driver criteria in one specific niche network when compared in pair to each of the other two networks. Moreover, such microbes failed the criteria when compared with the other two network pairs.

A further examination of the predicted metabolic and functional roles of the key bacterial taxa identified in T. gratilla elatensis was performed. To do this, the sequences of OTUs that were identified as key hub microbes and those categorized as core, core-generalists, generalists, specialists, or unique in a specific niche, were uploaded to the NCBI database for blasting with megablast [41]. Bacterial strains with the highest number of hits and an available representative genome were selected for further analysis of their metabolic repertoire. Considering the differences in the biochemical content of the experimental diets, the genomes of these bacteria were analyzed in the Carbohydrate-Active enZYmes database (CAZy) using the dbCAN2 meta server [42] to identify annotated genes that encode carbohydrate-active enzymes (CAZymes) of carbohydrate binding, carbohydrate esterases, glycoside hydrolases, and glycoside transferases. A heat-map of these results was produced to visualize the variation in the content and number of copies of the glycoside hydrolase (GH) and polysaccharide lyase (PL) genes (enzyme classes EC 3.2.1.-, and 4.2.2.-, respectively) between the selected microbes.

Statistical analyses

MicrobiomeAnalyst [43, 44] was used for examining each diet or gut region separately and also the combined effects of both these variables together. Rarefaction of data at a sequencing depth of 10,000 reads resulted in 51 samples for which differences in ecological indices of richness, alpha and beta diversity were measured by Kruskal–Wallis and post hoc Dunn’s multiple comparison test. Differences in microbial community composition were measured by Bray–Curtis dissimilarity and Permutational multivariate analysis of variance (PERMANOVA). Differences in relative abundance were verified by Kruskal–Wallis, at a false discovery rate (FDR) of P < 0.01. We selected the non-parametric Kruskal–Wallis test after refuting the null hypothesis and confirming that data in the sequencing results are not normally distributed following a Shapiro–Wilk test of distribution of the residuals (P < 0.0001).

Results and discussion

Diet shapes the gut microbe assembly of T. gratilla elatensis

No mortality or morbidity of sea urchins was observed during the entire period of the feeding trial. Sea urchin individuals fed the Ulva diet had a growth rate of 0.33% d−1 as compared to 0.27 or 0.1% d−1 on diets of Graciaria or pellets (P < 0.05), respectively (Additional file 1: Table S2). The faster growth rate under dietary Ulva as compared to Gracilaria or pellet diet is consistent with a previous study of T. gratila elatensis when fed similar diets of Ulva, Gracilaria and pellets over a much longer culture period of 400 days [12].

The fifty-one samples that reached sequencing depth of 10,000 reads resulted in 988,513 high-quality reads (18,305 ± 5208 per sample) that were clustered into 434 unique OTUs (117 ± 48 per sample; Additional file 1: Table S3). Neither diet nor gut region alone affected the richness of the GMA (H50,9 = 15.37; P > 0.05). The richest assembly was identified in the esophagus of the Gracilaria-fed sea urchins and the poorest assembly was found in the stomach of the pellet-fed sea urchins (Additional file 1: Fig. S2). The more diverse community in the anterior esophagus (H50,3 = 6.483; P < 0.05; Fig. 1a, b) agrees with results reported in previous studies on the sea urchin Lytechinus variegatus either prior to or after capture and rearing on artificial diet [45, 46]. This may be attributed to the rapid interactions with the external water environment and short retention time of ingested feed in the esophagus [47]. Lower pH and oxygen and high secretion of digestive enzymes in the stomach [19, 48] may have limited the assembly diversity in this niche. A similar characteristic was noted in those fed with artificial pellets (Fig. 1c), perhaps indicating disruption of the assembly due to the lack of algae or any other marine footprint in this diet. Diet type restriction induced dissimilarity between gut assemblies (F50,3 = 8.2, P < 0.01, R2 = 0.25, Stress = 0.143; Fig. 1d, e). The effect was intense enough to also differentiate assemblies of the different gut regions (F50,9 = 6.16, P < 0.01, R2 = 0.539; Fig. 1f; Additional file 1: Fig. S3a-c), with the strongest impact seen in the Gracilaria-fed sea urchins (P < 0.001 Additional file 1: Fig. S3a). This combined effect of the diet and gut region variables together (as compared to the effect of only diet or only gut region) resulted in a more diverse assembly in the anterior esophagus as compared to posterior regions in Gracilaria-fed urchins (P < 0.01), while an exact opposite pattern was noted in Ulva-fed sea urchins (P < 0.02) (Fig. 1c). An interesting finding was the relatively low variation in Shannon index of diversity within samples from different sea urchins when fed the Ulva diet, in each of the examined gut regions of these sea urchins (Fig. 1c). This can be attributed to the relatively consistent prevalence of the dominant taxa in the Ulva- fed sea urchins across samples of different biological replicates (sea urchin individuals; Additional file 1: Fig. S4b), which also resulted in a relatively low variation within individuals in terms of the community composition and diversity.

Fig. 1
figure 1

Richness, diversity and similarity of T. gratilla elatensis GMA under different variables of diet and gut region. Shannon index considering a different diets, b gut regions, or c both variables (n = 51). Box plots represent a level of 95% confidence interval, bar plots represent the standard error, straight lines represent the median, and the black dot represents the mean. Dissimilarity between GMA is demonstrated by non-metric multidimensional scaling (NMDS) based on Bray–Curtis dissimilarities between GMA in different d diets e gut regions, or f both forces (n = 51)

Full size image

Predominance of bacterial taxa in the gut of T. gratilla elatensis

None of the 110 genera (Fig. 2a) from 17 bacterial phyla (Additional file 1: Fig. S4a, Table S4) predominated in a particular gut region, but the three taxa, Cyanobacteria, Firmicutes, and Fusobacteria, were predominant in the assembly in Ulva-fed sea urchins (Fig. 2b, P < 0.001) and exhibited (each in itself) a relatively similar prevalence in the different gut regions in all the examined diets. An opposite prevalence pattern of Spirochaetes vs. Tenericutes was evident under the Gracilaria diet. The Spirochaetes, primarily represented by the genus Spirochaeta, predominated in the esophagus and decreased in posterior regions, while the Tenericutes, which was primarily represented by a single OTU assigned as genus Candidatus hepatoplasma, predominated in the intestine and stomach and was lower in the esophagus (P < 0.001; Additional file 1: Fig. S4b). Fluctuations in prevalence allowed identification of potential diet biomarkers in the GMA (Fig. 2c) as the genera Fusibacter, uncultured Bacteroidetes bacterium, Roseimarinus, and Propionigenium for Ulva; Carboxylicivirga and Cohaesibacter for pelleted feed; and unassigned for Gracilaria; but none were specific to any gut region.

Fig. 2
figure 2

T. gratilla elatensis GMA composition under different variables of diet and gut region. a Illustration of sea urchin digestive tract (performed using magnetic resonance imaging [49] and image illustrator at https://www.nhm.ac.uk/our-science/data/echinoid-directory) followed by the relative abundance of T. gratilla elatensis gut microbes (Genera) in different gut regions and under each diet (n = 54). b Log-transformed count of bacterial phyla with greatest differences in prevalence under a particular diet, middle line represents median, and whiskers are drawn from the 10th to 90th percentiles (n = 54). c Gut bacterial genera identified as diet-biomarkers in T. gratilla elatensis via Linear discriminant analysis (n = 54)

Full size image

Niche specification in the gut microbial assembly of sea urchins

Generalist and core microbes colonize the majority of examined ecological niches, i.e., the different gut regions of T. gratilla elatensis, fed on the different diets. Therefore, these microbes may be highly important in aiding their host in changing environments [50, 51]. Adhering to the fixed rule of colonization of at least 85% of the examined samples (across any gut region and/or diet), the core microbes identified in the gut of T. gratilla elatensis included five OTUs belonging to taxa Candidatus Hepatoplasma, Ruegeria, Vibrio, and two uncultured members of the order Bacterioidia (uncultured bacterium and uncultured Bacteroidetes bacterium) (Table 1). Two additional taxa of uncultured Fusibacter and Roseimarinus were thusly determined as core microbes in all the examined gut regions and under diets of Ulva and Gracilaria. However, they marginally failed the selection threshold criteria in the pellet-fed urchins, appearing in only 83% of the samples under this diet. The cumulative abundance of each of the identified core microbes was above 10,000 reads (Fig. 3a). Moreover, the relative abundance of a core microbe obtained in any sample was greater than 0.01%. These findings on the occurrence and abundance of core microbes may aid in their confirmation as resident microbes in the gut of T. gratilla elatensis rather than transient microbes which may only pass through this organ.

Table 1 Bacterial characterization following niche breadth and occurrence in the different examined niches differentiates the assembly microbes into core, core generalists, generalists, specialists and unique microbes
Full size table
Fig. 3
figure 3

a Niche breadth of microbes in sea urchin GMA following the different measured indices of Levin's, Shannon–Weaver, and occurrence. Middle line in box plots represents mean value and whiskers are drawn from the 10th to 90th percentiles. Generalist or specialist microbes are shown at top or lowest decile, respectively (n = 434). b Occurrence and cumulative abundance of each of the microbes in GMA indicates differences of niche breadth and specialization in unique habitats. Each dot represents one microbe with a color indication indicating core, core-generalist, generalist, specialist, unique, or not significant (n = 434)

Full size image

Identified generalists included 18 microbes (8 of them assigned to phylum Proteobacteria or class Planctomycetacia) with relatively high occurrence (> 54%), diversity (> 2.95), and niche breadth (> 13.38; Fig. 3a). The 12 identified specialists were mostly annotated to Bacterioidetes (Table 1; Fig. 3a). Three core microbes of the respective genera of C. hepatoplasma, uncultured Bacteroidetes bacterium, and Fusibacter were also among generalists (i.e., core-generalists) and revealed the highest cumulative abundance (Fig. 3b). The highest occurrence and abundance of these core-generalists agrees with the predomination of these taxa in the GMA of other sea urchin species as well [52,53,54]. A recent model highlighted the important role of the generalist-specialist evolutionary cycle which allows the spreading of specialist species across diverse environments, thereby maintaining taxonomic diversity [2] while at the same time influencing the overall community and its differentiation [2, 3]. Following the model, the generalist-specialist cycle includes the expansion of generalist microbes across various ecosystems, where local environmental forces determine the specialization of some of their descendants in the specific niche, hence the latter’s determination as specialist microbes [2]. In another study, core microbes in fish gut revealed high strain variability [30].

Findings in the current study may suggest the fitness of phylum Bacteroidetes to this generalists-specialists cycle model with core and core-generalist OTUs (of genus of Roseimarinus and uncultured Bacteroidetes bacterium, respectively) as potential founders in the gut assembly of T. gratilla elatensis, and several other OTUs of these phyla (four of the genus Roseimarinus and one uncultured Bacteroidetes bacterium) identified as specialist microbes that occupied a narrower unique niche of the gut of Ulva-fed sea urchins. The fact that Bacteroidetes reveal high plasticity and genetic rearrangement, allowing their rapid adaptation to distinct ecological niches [55], makes them more adequately fit the suggested evolutionary model in the examined sea urchin. Despite the significant differences in niche breadth between generalists and unique or specialist microbes, the total abundance of microbes from these groups was relatively similar (Additional file 1: Fig. S5). This indicates that the prevalence of unique or specialist microbes surpassed that of generalists in each of the niches they occupied. This notwithstanding, the core microbes were present in the highest abundance in these niches.

Intriguingly, unique microbes were identified only in sea urchins fed with algae: Thirteen of these in Ulva-fed (annotated to phyla Bacterioidetes, Cyanobacteria, Firmicutes, Spirochetes, Tenericutes, and unassigned), two in Gracilaria-fed (Bacterioidetes and Planctomycetes), and one unassigned taxon that was shared between these diets (Table 1). One explanation for the Ulva- or Gracilaria-unique microbes identified here may be the previous identification of members of Bacteroidetes, Cyanobacteria, Firmicutes, Spirochaetes, and Plancomycetes taxa as epibionts on these algae [56, 57]. However, the relatively high Shannon diversity and Levin's indices of many of these OTUs that were measured in sea urchins fed these diets may also be explained by high specialization of these microbes in the narrow diet-mediated niches. This is particularly evident in the sea urchin-associated Bacteroidales and Clostridales fed on an Ulva diet and in sea urchin-associated uncultured bacterium (Planctomycetes) fed on the Gracilaria diet. Correspondingly, the contribution of these microbes to the decomposition of the complex polysaccharides in Ulva or Gracilaria should be considered, particularly since, despite their algivorous life style, sea urchin digestive systems contain few algal polysaccharide digestive enzymes [47, 58].

Assembly connectivity is determined via niche-specific hub and driver microbes

The overall association network of the GMA consisted of 27 nodes (each representing a single OTU/microbe) with 37 edges (i.e., of either co-occurrence or co-exclusion; also referred to as degrees). This network, however, was separated topologically and contained three sub-nets, each containing many fewer clusters of the participating microbes (Fig. 4a). This low connectivity of the general network highlights the importance of investigating microbial networks in the narrower niches of each different diet or functional region (Fig. 4b–g) to also allow the identification of niche-specific associations, clusters, hub microbes, or other changes. Examination of the variation in number of degrees between participating nodes in the different networks revealed a Poisson distribution of the degrees per node in networks of the different gut regions but a power-law tail distribution of this index in networks under different diets (Additional file 1: Fig. S6). These results suggest that the microbial networks in different gut regions or under different diets can be differentiated into random or scale-free networks, respectively [59]. Notably, many nodes in the network of the Ulva-fed sea urchins possessed a high number of degrees and formed a left tail distribution shape. This was exceptional compared to the other two diets, where only a few nodes revealed a high number of degrees, i.e., right tail distribution (Additional file 1: Fig. S6a-g). Considering the nature of associations, i.e., positive co-occurrence or negative co-exclusion, the number of positive associations in all examined niches surpassed that of the negative ones, with the greatest fraction of the total edges in the Ulva and intestine networks. However, the nature of many of the shared associations between networks was of competition/co-exclusion between Cyanobacteria and Bacteroidetes or Firmicutes (diets), or between nodes of Cyanobacteria, Spirochaetes, Firmicutes, Tenericutes, and Bacteroideteds (gut regions) (Additional file 1: Table S5).

Fig. 4
figure 4

Association network of T. gratilla elatensis GMA under different variables of diet and gut region. Association network presenting a the general network (regardless of diet or region), and in different examined niches of diets b Gracilaria, c Pellet, or d Ulva; or gut regions e esophagus, f stomach, or g intestine. Diamonds indicate hub nodes and triangles indicate driver nodes. Nodes are colored following annotation at Phyla level. Edge color indicates type of association as either co-occurrence (blue) or co-exclusion (red). h Venn diagrams reveal number of shared or unique associations in networks of different diets or gut regions (n = 54)

Full size image

Examination of the topology indices of the co-association networks (Additional file 1: Fig. S7a–g) revealed similarity in terms of the number of edges between participating nodes in networks of the different gut regions (14–16% similarity), but under different diets  the Ulva network revealed a greater dissimilarity from any other diet network with ca 2.5-fold more edges (Additional file 1: Fig. S7b). The significant high edges number in the Ulva network may indicate cross feeding relationships and richness in feed decomposition pathways [60]. The high number of edges under the Ulva diet resulted in many unique edges, 76% of total edges (Fig. 6h), but also in a significantly denser network and more sub-clusters of participating nodes than were seen in other diets (Additional file 1: Fig. S7d,e). Therefore, it is more likely that under the Ulva diet, any two microbes in the assembly will perform a direct association between themselves (most likely a positive one), as both the diameter and average path length were similar to those measured in other networks under different diets [61].

Our finding of a contradictory high-tail degree distribution under Ulva diet compared to other diets supports the advantage of the multi-parametric analysis for hub node identification as performed here. This analysis enabled the identification of an unassigned node as a hub microbe in the Ulva network as per its significant “high betweenness centrality” despite the somewhat lower “number of degrees”. The hub in the different networks included OTUs from six phyla of Bacteroidetes (3 OTUs of order Bacteroidales), Firmicutes (2 OTUs of order Clostridiales), Cyanobacteria (2 OTUs of order Gastranaerophilales), Proteobacteria (1 OTU of genus Photobacterium), Spirochaetes (1 OTU), Tenericutes (1 OTU of genus Candidatus hepatoplasma) and two more unassigned OTUs (Additional file 1: Table S6). Our identification of the hub nodes in the different networks as per the criteria for hub microbes [7] suggests Phylum Bacteroidetes as the keystone taxa, with a representative hub OTU in nearly all examined networks (Additional file 1: Table S6). Our additional findings of the relatively high strain variability of Bacteroidetes (richness; a total of 111 OTUs, 2nd highest) in the gut of T. gratilla elatensis sea urchin suggest further support for the importance of this keystone phylum. This strain variability revealed 3 core microbes, 7 specialist microbes (assigned to 5 different orders), and 4 Ulva-unique microbes (all assigned to genus Roseimarinus), all of this phylum.

Interestingly, the identified hub nodes in each network of the algal diets formed co-occurrence sub-clusters. Such a sub-cluster was also identified in the pellet-fed group but it included a fair number of co-exclusions between the hub nodes (Additional file 1: Fig. S8). Considering these findings, the synchronization between hub nodes in networks of the algal diets in terms of their co-occurrence may suggest a major role of these microbes in syntrophy/cross-feeding via complementary metabolisms [62]. Such syntrophy is greatly influenced by available resources (including feed) [63] and allows maintenance of functional redundancy in the community due to the presence of multiple taxa with shared metabolic function even when a particular node is eliminated from the network, [60, 64]. Under the algal diets tested here, this microbial syntrophy may be essential for the metabolism of the complex fibers that are the primary ingredients in these feeds [65, 66]. This could be compared, perhaps, with the syntrophic microbial metabolism of cellulases in cows’ rumens [67]. This is further supported by the fact that microbe connectivity in different regions of the gut was mainly through in-line/chain-form connections despite being spatially close in each of the examined regions.

Transfer of the sea urchins from the initial baseline algal diet of combined U. fasciata and G. conferta to the pellet diet treatment may have caused a more severe disturbance of their GMA than did the other two diet treatments in which the change was simply restriction to only one of these two algal species. This disturbance may have resulted in reduced diversity and a less dense network, consisting of low numbers of edges (also of exclusive ones), a fair number of co-exclusion associations, and fewer sub-clusters (Fig. 4b-d,h, Additional file 1: Fig. S7b,d,e). This perhaps indicates reduced stability and functionality of the bacterial community [6, 68] which may have resulted from switching the urchins to this new artificial diet made of ground-up land crops rather than natural marine feeds. A similar reduction was also measured in rodents when they were transferred to and from their natural feed as a result of periodic captivity [69], and in captive fish after rearing on an artificial pelleted feed [70].

Network rewiring was greatest due to dietary influence, particularly between the pelleted and either of the algal diets, and less significant regarding the gut regions (Additional file 1: Fig. S9a–f). This discrepancy was also supported by the relatively high number of driver microbes that rewired diet-type networks as compared to their numbers in gut-region networks (Additional file 1: Table S7). Six microbes were identified as clear driver nodes due to their altered associations and higher centrality in only one particular niche (Fig. 5). The networks of the pelleted diet and esophagus were rewired by differing driver nodes of Roseimarinus, while driver nodes of Sediminispirochaeta or C. hepatoplasma rewired the intestine or Ulva network, respectively. The lack of associations of Ulva- driver node C. hepatoplasma with hub nodes in this network was significant as compared to other drivers found in other networks. Therefore, despite the highest occurrence and abundance of this single OTU of phylum Tenericutes in all niches and examined sea urchin individuals, we assume that C. hepatoplasma acts as a host-dependent node in the assembly, as was also suggested for this bacterial species in the case of another sea urchin [46]. Under the Gracilaria diet, network rewiring was driven by two nodes, one of Spirochaetes and another of an unassigned microbe. These two drivers revealed a competitive association with each other under this diet. Moreover, while the Spichochaetes driver revealed negative interaction with all hub nodes of Gracilaria network, the other driver node (unassigned) co-occurred with them. As compared to other drivers, the unassigned driver in Gracilaria network also revealed greatest exclusive edges, i.e., interaction with an exclusive partner or with the same partner but via opposite association. Although both hub and driver nodes reveal great centrality in the network, the high number of degrees characterized the former but not necessarily the latter. However, in some cases where the node creates exclusive degrees that also increase its overall connectivity, the microbe is identified as both hub and driver node in this niche. This was the case with the unassigned and Roseimarinus nodes in Gracilaria and esophagus networks, respectively (Fig. 5, Additional file 1: Table S6, S7).

Fig. 5
figure 5

Driver nodes to rewire T. gratilla elatensis GMA association network under different variables of diet and gut region. Driver microbes were extracted from networks using NetShift following significant change in degrees and associated partners. Only nodes that presented significant change as compared to the other two diets or regions were identified as driver microbes. Associations of driver nodes in a different diets or b gut regions. Annotation (genus level) of driver microbes include (upper to lower order): denovo22514 (Spirochaeta2), and denovo28887 (Unassigned) as driver microbes in the Gracilaria-diet network; denovo36058 (Roseimarinus) in the pelleted-diet network; denovo31035 (Candidatus Hepatoplasma) in the Ulva-diet network; denovo6214 (Roseimarinus) in the esophagus network; and denovo14715 (uncultured bacterium of Spirochaetes) in the intestine network. Nodes are colored following annotation at Phyla level (n = 54)

Full size image

Generalism and specialism with regard to the potential metabolic functions of key microbes in sea urchin gut

A primary concern in the current examination of the niche breadth, specialization, and networking of the microbes in T. gratilla elatensis gut was the validation of these results in terms of metabolic diversity. This has been highlighted as a key milestone that will improve our current understanding of the customary ecological definitions of generalism and specialism [51, 71, 72]. To do this, analysis was made of the potential metabolic repertoire in genomes of the key microbes identified in the gut, including the key hub, core, core-generalist, generalists, specialists, and the unique microbes, particularly concerning their richness in CAZymes. Some specific differences between the selected microbes were observed (Fig. 6, Additional file 1: Table S8) which can be attributed to a specific gut region or diet type, and in some cases also to a specific stage in the metabolism of dietary polysaccharides. Altogether, the genomes of the selected microbes presented a total of 271 carbohydrate-related enzyme classes (including sub-families) in the CAZy database. 167 of them were classified as glycoside hydrolases (GHs), 27 as polysaccharide lyases (PLs), 36 as glycoside transferase (GTs), 23 as carbohydrate binding modules, 12 as carbohydrate esterases (CEs), and 6 related to auxiliary activities. Of these classes, only GHs and PLs are involved in polysaccharide decomposition [73] and were further analyzed. Three bacterial strains of Roseimarinus sediminis, Reichenbachiella agariperforans, and Sunxiuqinia elliptica (all members of the phylum Bacteroidetes and isolated from marine environments) were richest in terms of the number of different CAZyme classes. These species were also richest in their content of CAZymes for polysaccharide decomposition, with 81 (R. sediminis) 73 (R. agariperforans), and 50 (S. elliptica) GHs and PLs. Examination of the genome of R. sediminis revealed 4 exclusive PL classes (10,11, 29, and 38), while among these classes both the PL29 and PL38 (characterized as chondroitin-sulfate-ABC endolyase and glucuronan lyase, respectively) presented the highest number of copies in the genome. These findings appear adequate for the correlation of the four OTUs which were unique to Ulva-fed sea urchins in the current research (i.e., Ulva-unique) and annotated as R. sediminis, to potential activity in the metabolism of Ulva-polysaccharides in T. gratilla elatensis gut. The rich content of CAZymes in the genome of S. elliptica agrees with the known capability of this fermentative microbe to decompose various carbohydrates such as D-fructose, L-rhamnose, D-xylose, myo-inositol, sorbitol, trehalose, D-mannitol, ribose, raffinose, gluconate, malonate, and propionate, each as a sole carbon source [74]. Hence, this characteristic may be attributed to our current finding of R. sediminis as a key hub node in the stomach of T. gratilla elatensis, where rapid fermentation is required for the metabolism of refractile and sulphated carbohydrates [48, 75, 76] such as those derived from decomposition of the complex polysaccharides in Ulva and Gracilaria. We also suggest additional members of Bacteroidetes identified here as correlated with the initial depolymerization stage in the metabolism of Ulva or Gracilaria to be key taxa in the gut of sea urchins fed these diets. Among these is R. agariperforans which consists both of ulvan lyases (PL40 and PL37) and β-agarases (GH50 and GH86) that are correlated with degradation of Ulva-ulvan [77] and Gracilaria-agarose [78], respectively. The β-agarase GH86 is also found in another key species, Persicobacter. (referred to here as denovo796; also from the Bacteroidetes) [79], which was correlated in the current study with the Gracilaria-fed sea urchins (Gracilaria-unique). However, it should be noted that low CAZyme content in a genome is not necessarily singular evidence for the non-engagement of a microbe in polysaccharide metabolism. A potential example gleaned from this study may be the core-generalist Fusibacter tunisiensis (referred to here as denovo 3738) which consists only of 7 CAZymes (with a total of 11 copies) but presents a relatively unique characteristic in terms of energy gain from the reduction of thiosulfate and elemental sulfur into sulfide. This metabolism is part of the sulfur cycle in anaerobic salty environments such as marine sediment [80], while in the currently examined sea urchins it may be attributed to the metabolism of the algal polysaccharides. This was evident in previous studies on decomposition of Ulva and Gracilaria where sulfate-reducing bacteria were noted for their major role in the process [81, 82].

Fig. 6
figure 6

A heatmap diagram of the variation in content and number of copies of glycoside hydrolase (EC 3.2.1.-) and polysaccharide lyase (EC 4.2.2.-) genes in the genome of the identified key microbes that fell into the determination of core (C), core-generalist (Cg), generalist (G), specialist (S), or unique (U) microbes, and the keystone hub microbes in the microbial networks (H)

Full size image

While the above results revealed some supportive evidence for the specialization of the diet-unique microbes in terms of their unique metabolic content as well, it is still questionable whether encompassing a broad or narrow niche in the gut of T. gratilla elatensis is indicative of microbes’ own metabolic arsenal (and vice versa). An interesting finding concerning this is the core microbe Candidatus hepatoplasma (a single OTU annotated as C. Hepatoplasma crinochetorum Av) identified here. This microbe, in spite of being fairly abundant in all examined niches of different diets and gut regions, reveals a relatively small genome of only 657,101 bp [83], no CAZymes, and also lacks a cell wall. This reinforces our previous assumption of this bacterium’s host-dependent lifestyle in the gut of sea urchins. Another common microbe in T. gratilla elatensis gut was the generalist Gloeobacter kilaueensis (Cyanobacteria) which was found in all the niches examined. Such generalism in terms of a wide niche breadth can be correlated in the current study with the photoautotrophic lifestyle of this microbe [84] essential to its success in the sea urchin’ light-penetrated gut, independent of the variation in available nutrients in the ingesta from different regions or diet types. The latter may be supported also by the relatively low content of CAZymes in this microbe’s genome. In contrast to these two microbes, nearly all the other microbes that occupied broad niches, i.e., core, generalists, and core-generalists, revealed a relatively rich carbohydrate metabolic capacity of between 14 and 81 CAZymes. Specialist microbes, with only two exceptions (R. agariperforans and U. croceus), presented a relatively poorer carbohydrate metabolism potential consisting of between 7 and 13 enzyme classes (including sub-families). This finding may emphasize the correlation between a narrow niche breadth in terms of occupancy and occurrence, as measured for these microbes in the current study, and specification in the niche in terms of carrying only relevant enzymes for carbohydrates metabolism which are vital for gaining energy from the here examined diets.

Conclusions

The mono-specific algal diet altered the gut microbial assembly of T. gratilla elatensis in terms of composition, networking, and potential metabolism. A dense and well-connected microbial network was measured in the Ulva-fed sea urchins and may suggest, together with the animal's rapid growth, that this alga was most nutritious among the experimental diets. Generalism and specialism of the microbes in the gut agreed with their richness in CAZymes, while specialization in narrower niches of diet types was evident in terms of both occurrence and the specific metabolic arsenal the microbe carries that may aid the host in nutrition. Finally, we conclude Bacteroidetes to be the keystone phylum in the gut of T. gratilla elatensis with many different strains inhabiting this organ. Among are key hub microbes in the niche-specific networks, core microbes that occupied nearly all niches, generalist microbes that were suited to broader niches, as they are rich in their CAZyme content for decomposition of various polysaccharides, and specialist microbes that were correlated with the initial depolymerization stage in the metabolism of Ulva or Gracilaria.