Background

The importance of microbes for the evolution of higher organisms is starting to be realized [1, 2]. Metazoan evolution is not only driven by pathogenic microbes, as reflected by fast evolution of immune genes [3]. Rather, microbes often are facilitators of metabolic and environmental adaptations [2, 4, 5]. For instance, the gut microbial communities of wood-feeding roaches and termites facilitate thriving on a wood diet that is difficult to digest and poor in nitrogen. Nitrogen fixation and the digestion of wood depend on the termite gut microbiome [2, 6, 7]. In lower termites, lignocellulose degradation was initially mainly attributed to unicellular eukaryotes (protists) in the gut [8]. Recently, it has become evident that lignocellulose degradation is a synergistic effort of the termite, its associated protists, and bacteria [9,10,11]. In addition to their role in lignocellulose degradation, bacteria are also essential for the assimilation of nitrogen taken up from the environment. Nitrogen can be acquired from the environment either via fixation from the atmosphere [12, 13], or via nitrate reduction [14]. Also, nitrogen can be recycled from the metabolic waste product uric acid [15, 16]. By using genome sequencing and pathway reconstruction, these processes have been assigned to four major bacterial phyla in the termite gut: Proteobacteria (Desulfovibrio [17]), Spirochetes (Treponema [18, 19]), Bacteroidetes (Azobacteroides) [16], and Elusimicrobia (Endomicrobium [20, 21]).

Many bacteria in the termite gut live in tight association with protists, where they sit on the surface [22, 23], in invaginations of the cell membrane [17], or even inside the protist cells [24]. Such tight associations lead to frequent vertical transmission of bacteria between protist generations. In return, protists and bacteria are vertically transmitted between termite generations via proctodeal trophallaxis during colony foundation [25]. Vertical transmission has led to co-speciation between bacteria and their protist hosts, and sometimes even the termite hosts [26,27,28,29]. Evidence for horizontal transfer of protists between termite species, so called transfaunations, is limited to a few exceptions [30]. Hence, the termite host species association is rather strict, leading to strong phylogenetic imprints on protist community structure [31,32,33]. In comparison, the bacterial microbiome is more flexible, frequently transferred between termite host species [34], and affected by diet [33, 35,36,37,38,39,40,41].

There is evidence that the gut microbiome of termites has contributed to the adaptation of different termite species to their specific ecologies [33, 36, 42,43,44]. There are pronounced ecological differences between the so-called termite life types [45, 46]. Termite species of the wood-dwelling life type never leave their nest except for the mating flight. They feed on a relatively uniform bonanza resource, that is the piece of wood they built their nest in [47, 48]. On the other hand, foraging species leave their nest to forage for food and have access to additional nutrients [47, 49]. This likely imposes different selection pressures on the termite holobiont, in particular with regard to nutrient uptake. Because the microbiome is directly involved in nutrient uptake, it seems reasonable to hypothesize that it may also play a role in adaptation to life type related dietary differences. In this scenario, one would expect the life types to leave an imprint on microbiome structure and function. As such, searching for microbial imprints of a given life type can possibly provide us with a lead for microbiome-mediated adaptation.

One potential pitfall of such an endeavor is that microbiomes may bear imprints from transient microbes that were ingested from the environment. Transient microbes rarely form evolutionary relevant relationships with the host [50, 51]. Instead, they reflect short-term associations with microbes from the local environment the termites were collected from. For example, in previous studies [34, 36,37,38,39,40,41] microbiomes were profiled from termites that were collected from a variety of substrates and locations, sometimes including natural and laboratory environments in the same study. In such an experimental setup, microbiomes can be similar because termites ingested transient microbes from similar environments. For instance, microbiomes from different termite species may become similar because the termites ingested soil that harbors a discrete microbiome. In such a case it is challenging to disentangle effects stemming from ingested, transient soil microbes from those that resulted from potential adaptations of the holobiont to a soil diet. Critically, such similarities may then erroneously interpreted as involved in the adaptation to a soil diet. Similarly, rearing termites in different laboratories without acclimation to common conditions in the same laboratory could lead to artificial clustering of microbiomes. Strong laboratory effects have been shown for the microbiome of Drosophila [52]. Therefore, it is essential to reduce the impact of transient microbes in experimental setups to test evolutionary hypotheses.

In order to explore potential ecological imprints on the microbiome, we focused on an evolutionary switch between wood-dwelling and foraging life types in the Rhinotermitidae (Fig. 1). Reticulitermes species are of the foraging life type, while Prorhinotermes simplex is wood-dwelling. If the microbiome was affected by life type specific ecology, we would expect that the microbiome of Prorhinotermes simplex was similar to that of the other wood-dwelling species (Cryptotermes) although these are from a different family (Kalotermitidae). At the same time, the microbiome of the foraging Reticulitermes species should bear distinct features. Alternatively, if there was no ecological imprint, we would expect the microbiome to follow a phylogenetic pattern, with the Rhinotermitidae Prorhinotermes and Reticulitermes forming a cluster and the Cryptotermes species (Kalotermitidae) forming a second cluster. Using this experimental setup, we recently showed that protist community composition aligned with phylogeny, but bacterial communities aligned more strongly with wood-dwelling and foraging life types [33].

Fig. 1
figure 1

Schematic phylogeny of the five lower termite species used in this study from [33]. Branch length not drawn to scale. Colored boxes indicate the life type

Full size image

To further explore this, we investigated whether changes in microbiome composition are also reflected by changes in microbiome function, as would be expected if the microbiome played a role in adaptation. For instance, we would expect dietary adaptations to be reflected by changes to pathways involved in substrate breakdown and effective provisioning of limiting nutrients such as nitrogen. In order to test whether and which changes in the functional repertoire align with life type and could be involved in potential adaptation to different diets, we characterized the metagenome of two foraging species; Reticulitermes flavipes and Reticulitermes grassei. We compared their functional repertoire to that of three wood-dwelling species Prorhinotermes simplex, Cryptotermes secundus, and Cryptotermes domesticus. Because there can be substantial variation in microbial communities between colonies [53,54,55,56], we analyzed five C. domesticus, eight C. secundus, seven P. simplex, five R. flavipes, and four R. grassei replicate colonies. We focused on the persistent, long-term differences between microbiomes by controlling short-term effects caused by the influx of transient microbes. This was achieved by feeding a common diet of sterile Pinus wood for several weeks prior to sample collection.

Results

We analyzed a total of ~ 440 million metagenomic shotgun sequences. Between 974,176 and 8,949,734 sequences per sample were of microbial origin (Additional file 2: Table S1). Sequences were subsampled (rarefied) to 1,386,882 bacterial and 2781 protist annotated sequences per sample. For annotation, sequences were aligned to a reference database of clusters of orthologous groups of genes (COGs) with known function. These COGs represent the lowest level of the evolutionary genealogy of genes - non-supervised orthologous groups (eggNOG) hierarchical annotation. At the next higher level, the COGs are grouped into pathways (Additional file 1: Fig. S2, Fig. S3), and at the third and highest level, the pathways are grouped into three categories; “information storage and processing”, “cellular process and signaling”, and “metabolism”. We adhere to this definition of the eggNOG hierarchical terms throughout the study.

“information storage and processing” differentiates the protist metagenomes of wood-dwelling and foraging lower termite species

In our previous study [33] on the identical samples, the protist communities of the Rhinotermitidae Prorhinotermes and Reticulitermes clustered together, supporting a phylogenetic imprint on community composition. Here, we tested whether this pattern was also reflected by the functions encoded by the protist metagenome. Therefore, we annotated metagenome encoded functions in the shotgun sequences and compared the functional metagenome profiles across host species, using Bray-Curtis-Dissimilarity [57]. This index considers abundance of functional categories, thus avoiding arbitrary coverage cutoffs.

The protist functional repertoire clustered according to host family and genus (Fig. 2a), thus showing a dominant phylogenetic imprint. Family-wise clustering was supported by Redundancy Analysis (RDA): the model including host family explained more variance in the functional repertoire and produced lower Akaike’s Information Criteria (AICs) than the model based on life type (Table 1). For a more detailed view, we analyzed the three categories at the highest level in the eggNOG hierarchical annotation (Additional file 1: Fig. S2) separately. Cluster analysis of the categories “cellular process and signaling “and “metabolism” supported the notion that phylogenetic relatedness is an important factor for functional similarity (Additional file 1: Figs. S4B and D). In contrast, the portion of the metagenome assigned to “information storage and processing” (Fig. 2b) clustered primarily by life type. The stronger effect of life type than phylogeny on this functional category was also supported by higher explanatory power and lower AICs in RDA (Table 1).

Fig. 2
figure 2

Cluster dendrograms of the functional profiles of the protist and bacterial community. Community distances are based on Bray-Curtis Dissimilarities of a) all functions of the protist community (25,795 sequences), b) The category “information storage and processing” of the protist community (4527 sequences), c) all functions of the bacterial community (21,215,480 sequences), and d) the category “metabolism” of the bacterial community (10,586,058 sequences). Cd (red) = C. domesticus colonies; Cs (orange) = C. secundus colonies; Ps (green) = P. simplex colonies; Rf (blue) = R. flavipes colonies; Rg (lightblue) = R. grassei colonies. Green background = wood-dwelling life types; orange background = foraging life type. For protist functions involved in “information storage and processing” in the protist community, samples clustered according to life type. Similarly, the bacterial metabolic metagenomes clustered according to life type. Cluster dendrograms of all functional categories of protist and bacterial communities can be found in Additional file 1: Fig. S4 and S7

Full size image
Table 1 Models of the effects of life type and host family (phylogeny) on functional community profiles
Full size table

Identification of the functions that differentiate the protist metagenomes of the wood-dwelling and foraging species can hold clues with regard to the nature of potentially adaptive phenotypes in the protist metagenome. In order to do so, we performed a linear discriminant analysis (LEfSe: [58]). This analysis identified 22 over-represented COGs in foraging and 14 in wood-dwelling species (Fig. 3a, Additional file 3: Table S2, p < 0.05, q < 0.05, LDA score > 2, Additional file 1: Figure S6).

Fig. 3
figure 3

Differences in the functional content of the protist metagenomes of wood-dwelling and foraging species. a Circular dendrogram/hierarchy of all over-represented COGs in the category “information storage and processing” in wood-dwelling species (green) or foraging species (orange). Circle size at the edges scales with abundance of the COG. Colored branches indicate over-represented pathways. Over-representation was detected with LEfSe [58] (p < 0.05, q < 0.05, LDA > 2). A Venn diagram visualizing the total number, and the differentially abundant number of functions in each of the five pathways that constitute the category “information storage and processing” can be found in Additional file 1: Fig. S6. b Sequence coverage of wood-dwelling (green) and foraging (orange) species of examples of over-represented COGs mentioned in the text. Error bars represent 95% confidence intervals across replicate colonies

Full size image

The pathway “replication, recombination, and repair” was over-represented in the foraging species (Fig. 3a, Additional file 3: Table S2, p = 0.0001, q = 0.002). The over-represented COGs in this pathway included a DNA dependent DNA-polymerase (COG0470) and five helicases (COG0514, COG0553, COG1199, COG1204, ENOG410XNUT, see Fig. 3b for grouped analysis and Additional file 3: Table S2 for individual COG p- and q-values). In wood-dwelling species, the pathway “transcription” was over-represented (p = 0.0004, q = 0.003). The over-represented COGs in this pathway contained DNA binding domains and were supposedly involved in transcriptional regulation (COG5147, ENOG4111SAB).

The bacterial metabolic metagenome aligns with host ecology

In our previous study [33], the bacterial community composition of termite hosts clustered primarily by life type, which is consistent with ecology related differences between microbiomes. Following the rationale above, we tested whether this pattern was also reflected by the functions encoded by the metagenome.

Against the expectation from our previous study, the functional bacterial profiles showed no life type, but a phylogenetic imprint, which is in line with the protist functional profiles. Most samples clustered according to host family (Fig. 2c). Analyzing the three high-level eggNOG functional categories separately provided more detailed insight. The categories “cellular process and signaling “and “information storage and processing” supported the notion of strong phylogenetic effects on metagenome function (Additional file 1: Figs. S7B and C). In contrast, the metabolic metagenomes (Fig. 2d) clustered primarily according to host life type. Host life type was also a better predictor for metabolic functions than host family in RDA (Table 1).

Aside from these general patterns, several samples stood out. Samples Rg2 and Rg4 of R. grassei were on long branches in the dendrograms (Fig. 2 and Additional file 1: Figure S7), suggesting unusual functional profiles. Notably, these samples already stood out in our previous study [33] because of their unusual abundance of microbial taxa potentially due to infection with pathogens. This unusual composition was confirmed by taxonomic annotation in this study (see Additional file 1: Figure S8). Sample Cs7 (C. secundus) also clustered separately from the other samples. This was mainly driven by abundant transposases in this sample (53.1% of sequences) (for example COG1662, COG3385, or ENOG410XT1T, see Additional file 4: Table S3), accompanied by an increase in the frequency of Bacteroides (Additional file 1: Fig. S8) that are rich in conjugative transposons [59, 60]. We performed all analyses with and without these samples and found no qualitative differences (data not shown).

Bacterial metabolic functions that differentiated wood-dwelling from foraging species were identified using linear discriminant analysis (LDA). 105 metabolic COGs were over-represented in the wood-dwelling species, while 151 were over-represented in the foraging species (Additional file 5: Table S4, p < 0.05, q < 0.05, LDA score > 2, Additional file 1: Figure S11). All COGs described as over-represented or enriched in the following were subject to these p-value, q-value and LDA cutoffs. Because of their specialized diet, genes involved in nitrogen metabolism and lignocellulose break down like glycoside hydrolases (GH) are of particular interest, when focusing on metabolic differences among gut microbiomes in wood-feeding termites with different diets. In fact, among the genes involved in ‘carbohydrate transport and metabolism’ that were enriched in the microbiome of wood-dwelling termites, GHs were over-represented (43.3% of enriched genes versus 12% expected, exact binomial test, p = 2.12e-05, Additional file 5: Table S4, Additional file 6: S5). In the foraging termite species, only one gene with putative lignocellulolytic activity was over-represented (COG3858), suggesting that the wood-dwelling species have a higher potential for complex carbohydrate degradation. To further investigate differences in GH abundance between the microbiomes of wood-dwelling and foraging species, we performed a detailed pathway analysis using the carbohydrate-active enzymes (CAZy) database ([61], Fig. 4). All GHs acting in hemicellulose break-down were more abundant in the wood-dwelling species (Fig. 4b). Among the cellulolytic enzymes, ß-glucosidases were significantly more abundant in the wood-dwelling species. The other two enzymes involved (cellulase (endo-ß-1.4-glucanase), cellobiohydrolase) showed a trend into the same direction. All of the genes with cellulolytic or hemicelllulolytic activity were affiliated with Bacteroidetes (mostly members of the genus Bacteroides) or the genus Treponema. Additional support for the increased importance of hemicellulose utilization in the wood-dwelling species was provided by the over-representation of twelve COGs annotated as TonB-dependent receptors (ENOG410XNNV, ENOG410XNPQ or COG4206, see Additional file 5: Table S4). Apart from other substrates, these receptors are important for the uptake of plant-derived hemicellulose [62, 63]. All functions annotated as TonB-dependent receptors (or TonB-dependent associated receptor plugs) were affiliated with the genus Bacteroides (see Additional file 5: Table S4).

Fig. 4
figure 4

Differences in the functional content of the bacterial metagenomes of wood-dwelling and foraging species. a Circular dendrogram/hierarchy of all COGs in the category “metabolism” over-represented in wood-dwelling species (green) or foraging species (orange). Circle size at the leafs scales with the abundance of the COG. Over-representation was detected with LefSe [58] (p < 0.05, q < 0.05, LDA > 2). A Venn diagram visualizing the total number, and the differentially abundant number of functions in each of the five pathways that constitute the category “metbolism” can be found in Additional file 1: Fig. S11. b Pathway analysis of cellulose and hemicellulose degradation. Colored boxes of cellulolytic or hemicellulolytic genes indicate proportion of relative abundance of sequences affiliated with wood-dwelling (green) or foraging (orange) species. c Pathway analysis of nitrogen metabolism. Boxes for genes with functions in nitrogen metabolism indicate relative abundance in the two life types. d Pathway analysis of glutamate synthesis. Boxes in c) and d) show relative abundance in the two life types of genes with functions in nitrogen/glutamate metabolism. Pie charts show taxonomic association of the gene. All hemicellulolytic genes were over-represented in the wood-dwelling species. Also, a nitrogenase was enriched in the wood-dwelling species, while in the foraging species, genes involved in dissimilatory nitrate reduction were over-represented

Full size image

Because wood is poor in nitrogen, termites depend on an efficient system for conserving and upgrading nitrogen [6]. In the wood-dwelling species, a potential nitrogenase (nifH, COG1348) was over-represented (Fig. 4c, Additional file 5: Table S4). Nitrogenases are key enzymes in the fixation of atmospheric nitrogen and downstream ammonia synthesis. Nitrogenase activity was mainly affiliated with members of the genus Treponema (Fig. 4c). In contrast, in the foraging species, COGs involved in dissimilatory nitrate reduction (COG1251, COG5013, COG2181, COG0243, Fig. 4c, Additional file 5: Table S4) were over-represented. They were affiliated with a variety of different genera ranging from Desulfovibrio and Gordonibacter to Stenoxybacter, Enterobacter and Serratia. Serratia and Enterobacter are potential insect pathogens and contributed to the prevalence of one of the three nitrate reductases, narG (COG5013). Closer inspection of the source of these bacteria revealed that they mainly stemmed from the abnormal samples Rg2 and Rg4 that we suspected to carry a potential pathogenic infection. When we remove these samples from the analysis the increase of narG in foragers remains significant (p = 0.034).

For living on a nitrogen poor substrate, it can also be adaptive to effectively recycle nitrogen from the main waste product of the host’s amino acid metabolism, uric acid. Uric acid can be recycled through anaerobic ammonia production and downstream glutamate synthesis [6, 15, 20, 64]. In the wood-dwelling species a putative glutamate dehydrogenase (COG0334), involved in glutamate synthesis by ammonia assimilation, was over-represented. This glutamate dehydrogenase gene was mainly affiliated with members of the genera Bacteroides, Treponema and Desulfovibrio. In the foraging species, COGs with putative glutamine (COG0174) and glutamate synthase (COG0067, COG0069) function, were enriched (Fig. 4d). These COGs were affiliated with Desulfovibrio, Treponema, Pseudomonas and Acetobacterium.

Discussion

In this study, we assessed functional differences of termite metagenomes that underwent an evolutionary switch from wood-dwelling to foraging to identify putative contributions of the microbiome to ecological niche adaptation. To do this, we chose a set of five termite species (two foraging, three wood-dwelling species) and determined whether the functional profiles of the termite gut microbiome followed phylogeny of the host or aligned with host ecology. We hypothesized that alignment of microbiome function with termite life type is consistent with a contribution of the microbiome to termite holobiont adaptation to different diets. By comparing the functional content of microbiomes of different host species we focused on long-term evolutionary processes.

A potential pitfall of such an approach is that an alignment of the termite microbiome with life type-related ecology could also be caused by short-term differences between microbiomes that are merely transient. For example, microbes in the environment might differ between collection sites for the different host species. Further, ingestion of environmental microbes might lead to an association between microbiome and ecology. Similarly, differences in local food supply can lead to transient, short-term effects on the termite microbiome [56]. Consequently, such short-term differences reflect environmental differences at termite collection sites, rather than potentially adaptive, evolved differences between host-lineage-specific microbiomes.

For this reason, we chose to follow an approach where we control for environmental and dietary differences by acclimating all termites on the same (sterile) food source and to identical environmental conditions. We consider metagenomic patterns that persist under such highly controlled experimental conditions as robust and indicative of long-term, evolutionary acquired differences, rather than short-term imprints originating from differences in the environment or food source. It should be noted that the experimental setup poses a restriction to the number of sampled host species [33].

Increased potential for replication in the protists of foraging termite species

In the protist metagenome of foraging species, genes involved in replication were more abundant. High replication rates are expected to be more frequently under positive selection during recolonization of the gut with protists, when the gut environment has not yet reached carrying capacity [65]. Therefore, we would like to speculate that this difference is related to the fact that Reticulitermes guts have to be recolonized more frequently because they molt more frequently; the intermolt periods in Reticulitermes are about two weeks long [49], while they average almost two months in Cryptotermes [48]. During molting the protists are lost and the guts have to be recolonized through proctodeal trophallaxis from nest mates [66]. However, we are aware that differences in the relative abundance of housekeeping genes like those required for replication between protist microbiomes can not be clearly disentangled from differences in average protist genome size and therefore should be interpreted with caution.

Enrichment of genes for lignocellulose degradation in the microbiome of wood-dwelling termite species

While genes involved in replication differentiated the protist metagenomes of wood-dwelling and foraging species in our study, metabolic genes differentiated the bacterial metagenomes. Consistent with differences in their respective diets, the metagenomes of foraging and wood-dwelling species in our study differed by their potential for cellulose and hemicellulose utilization. Several GHs that have cellulolytic and hemicellulolytic function were over-represented in the metagenomes of wood-dwelling species (GH families 2, 3, 16, 43, mannosidases, xylosidases, glucanases, xylanases, Fig. 4b, Additional file 5: Table S4). A more detailed pathway analysis confirmed that hemicellulases are more abundant in the wood-dwelling species. This suggests a more pronounced role for lignocellulose degradation in the metabolism of the wood-dwelling species in our study. Accordingly, TonB dependent transporters were enriched in the microbiome of wood-dwellers. These transporters can shuttle hemicellulose and its building-blocks, in particular xylans and xylose through bacterial membranes [67, 68]. A large fraction of cellulases, hemicellulases, and putative TonB transporters were attributed to the genus Bacteroides. In Bacteroides, TonB dependent transporters are often co-localized and co-regulated with enzymes for polysaccharide degradation like hemicellulases [60, 69]. This suggests a partnership of enzymes and transporters in polysaccharide degradation. Bacteroides species from the human gut are also hemicellulose degraders [70], suggesting a distinctive role for the genus in hemicellulose degradation in termites as well.

The above-identified differences in functional potential between the wood-dwelling and foraging species in our study are suggestive of adaptations to utilize diets that differ in hemicellulose content. Hemicellulose content differs between wood species [71, 72]. The wood-dwelling Cryptotermes species in our study are mostly found in hardwood mangroves [73] where they can thrive on a bonanza food resource. The other wood-dwelling genus in our study, Prorhinotermes, lives in similar coastal habitats with a similar arboreal flora [74]. Hardwood is richer in hemicelluloses and the potential to use hemicelluloses is larger in the microbiome of species living on hardwood. On the other hand, Reticulitermes species originated in inland habitats [75], prefer soft woods like pine [76, 77] with lower hemicellulose levels, and accordingly, hemicellulolytic pathways are depleted.

Termites with different life-types rely on different forms of nitrogen uptake and recycling

Nitrogen is scarce in a wood-based diet. As a consequence, termites need to acquire additional nitrogen from the environment. The microbiome is essential for this process. In the microbiome of wood-dwelling species, which feed on a uniform lignocellulose diet, a potential nitrogenase gene was enriched (nifH, COG1348). Nitrogenases are the key enzymes in the fixation of atmospheric nitrogen and downstream ammonia synthesis. This nifH was mainly affiliated with treponemes that have been shown to play an important role in nitrogen fixation before [12, 18, 19]. In contrast, the microbiome of the foraging species in our study has a higher potential to provide nitrogen to the termite holobiont by dissimilatory reduction of nitrate (Fig. 4c). Nitrogen in the form of nitrate naturally occurs in soil. R. flavipes has been shown to acquire micro nutrients from soil [78] and to actively balance mineral uptake by food choice [79]. Therefore, it seems reasonable to assume that the microbiome of Reticulitermes relies on nitrogen from soil in the form of nitrate to balance the low nitrogen content of wood. The necessary nitrate reductases were found primarily in Desulfovibrio, Gordonibacter and Stenoxybacter that were found in association with Reticulitermes before and are shared between a wide range of termites [33, 80, 81].

Aside from obtaining nitrogen from the environment (atmosphere, soil), bacteria can also recycle uric acid nitrogen. All of these processes result in ammonia synthesis, the central metabolite of nitrogen metabolism. Ammonia is then further assimilated to glutamate. In the wood-dwelling species a glutamate dehydrogenase (COG0334) was over-represented. It was mainly affiliated with members of the Bacteroides, Desulfovibrio and treponemes. The foraging species seem to rely on another glutamate synthesis pathway, including glutamine (COG0174) and glutamate synthases (COG0067, COG0069). Accordingly, they were associated with a different set of bacteria including Pseudomonas, Acetobacterium, Desulfovibrio, and treponemes (Fig. 4d).

Phylogeny and ecology align with metagenome-encoded functions

Differences in the propensity for nitrogen uptake and recycling are likely to reflect differences in diet of the termite host species. Given differences in diet between the species that represent the different life types, it seems also reasonable to suggest that the changes in the repertoire of hemicellulases reflects adaptations of the microbiome to diets with different hemicellulose content. The finding that this manifested specifically in the metabolic functional repertoire, may suggest that potential selection acts in particular on metabolic functions.

Metabolic microbiome mediated adaptation to different diets can happen in two ways. First, acquisition of new microbes with adaptive functions could lead to adaptive changes of the microbiome. Second, genome evolution of microbes that are already associated with the host could lead to adaptation. Microbes that were already present before the onset of lineage specific adaptation are likely to be shared among host species. By contrast, newly acquired microbes are expected to be host lineage specific. We found that the bacterial groups that contributed most to the differentiation of metabolic functions are shared among all five host species (Treponema, Bacteroides, Desulfovibrio, Dysgomonas, Gordonibacter, Pseudomonas, Additional file 5: Table S4, Additional file 1: Figure S8). This supports that genome evolution of microbes that were already associated with the host contributed to potential adaptation in our model system.

Conclusion

We applied metagenomic sequencing of gut microbiomes from a controlled experimental setup to assess a putative contribution of the microbiome to host dietary adaptation that accompanies the evolutionary switch from wood-dwelling to foraging life types. We found that the overall pattern of microbiome variation reflected a phylogenetic signal. Interestingly, specific functions of the microbiome aligned with the underlying host ecology. The specific ecology related differences in microbiome function led us to hypothesize that the microbiome contributed to dietary adaptations, namely different hemicellulose and nitrogen contents. This hypothesis can now be tested, assessing host fitness under different dietary conditions. Such experiments will be crucial to disentangle adaptive functional changes from selectively neutral functional turnover or side effects from other adaptations.

Methods

Termite samples

All termites were collected from typical natural habitats (see [33]). They were kept under constant conditions (27 °C, 70% humidity) on autoclaved Pinus radiata wood from the same source for at least six weeks prior to the experiment. The feeding of Pinus represents a natural or near natural treatment; Pinus is a natural food source of P. simplex and Reticulitermes. Cryptotermes growth and behavior on Pinus recapitulates that on natural substrate [73]. The time of the acclimation period was chosen to lie well beyond the gut passage time of 24 h in lower termites [82, 83] and following Huang et al. [84], who showed that six weeks are sufficient for the microbiota to adjust to a new diet. That way, all excretable material like remaining food, transient microbes taken up from the environment that have no mechanisms to persist in the gut, and microbial DNA taken up before the experiment was made sure to be excreted. The samples were identical to those analyzed in our previous study, [33] where detailed information about animal collection, keeping, and cytochrome oxidase II based species identification and a phylogeny can be found.

DNA extraction and shotgun sequencing

DNA was extracted from a pool of three worker guts per colony using bead beating, chloroform extraction and isopropanol precipitation (see Additional file 1: methods section S13). Each of the 29 colony samples went through independent metagenomic shotgun library preparation and sequencing on an Illumina HiSeq platform (150 bp paired end reads).

Analysis

We employed a double filtering strategy to remove host DNA from our analysis. First, sequences were removed that mapped to available host genomes from C. secundus [85] and transcriptomes from P. simplex [86] and R. flavipes, provided by the 1KITE consortium (www.1kite.org, BioSample SAMN04005235) using BBMap [87] (for detailed workflow and more detailed information about used genomes and transcriptomes see Additional file 1: Fig. S14 and section S15). Of note, the sequences were not assembled, but individual reads were directly annotated. In a second step we used taxonomic and functional annotations with Megan6 [88] to retrieve only sequences that could be unambiguously assigned to either bacteria or protists. In order to compare the bacterial and protist data sets of all samples, they were rarefied to the number of sequences in the sample with lowest coverage, resulting in 1,386,882 and 2781 sequences per sample, respectively. Sample Cs4 was excluded from the analysis for insufficient sequence coverage (974,176 sequences), so was Cs5 from the protist data. Sample Ps5 did not pass the analysis pipeline and was also excluded.

Functional annotation with the eggNOG database resulted in the highest number of annotated sequences (21,215,480 annotated sequences in total) and was chosen for further functional analysis. Bray-Curtis Distances of functional abundances were clustered with the pvClust package in R [89]. Multivariate modeling was performed via RDA (Redundancy Analysis) and AICs as well as values for the proportion of variance explained were derived with the model selection tool ordistep and ordiR2step, as implemented in the R vegan package [90]. Models were compared to the null-model via ANOVA. To identify over-represented functions associated with the two termite life types, a Linear Discriminant Analysis (LDR) was performed using LEfSe [58] and visualized using GraPhlAn [91]. Pathway analysis of CAZy GHs was performed by blasting bacterial reads of all samples against the full CAZy protein database, using Diamond [92]. GH abundance was estimated by counting reads with matches on proteins with cellulolytic and hemicellulolytic functions [93]. Pathway analysis of the nitrogen metabolism was performed by searching COG IDs corresponding to the KEGG IDs among the over-represented COGs from the LEfSe analysis. A detailed workflow for full reproducibility, including scripts can be found in the supplementary pdf file in Additional file 1: Figure S14 and in sections S13 and S15 in the same file.