Background
Microbial communities residing in different ecological niches are known to play several key functions and often define the phenotypic characteristics of their environments. A typical microbial community consists of numerous bacterial/archaeal species belonging to diverse taxonomic lineages. Several recent studies have indicated that the functional behaviour of a bacterial species is not only dictated by its own genomic content, but is also influenced by the presence of other microbes that co-habit in that given environment [
1,
2]. In other words, the function of a given microbe in an environment is dependent on its interactions with other resident microbes present in that environment. Therefore, in order to obtain a holistic insight into the role of the microbial community in determining the phenotypic traits of an environment, it is necessary to understand the inter-microbial interaction patterns present within the environment under study.
A key focus of several concerted efforts by independent research groups as well as consortia like the human microbiome and the meta-HIT projects has been to profile as well as characterize the microbial communities residing in various body sites [
3‐
5]. Various studies have also attempted to identify differences in microbial communities inhabiting different body sites of individuals from certain geographies and age-groups [
6‐
14]. However, a comprehensive analysis across geographies as well as different age groups is still non-existent.
Cohabiting microbes in an environment can interact with each other in various ways. For example, they may have positive interactions like mutualism and commensalism, or negative interactions like parasitism, amensalism and competition. A few recent studies have attempted to infer such inter-microbial co-occurrence/exclusion networks across different environments [
15‐
17]. For example, a study on metagenomic datasets from 18 different human body sites, obtained from 239 individuals, has identified a global network of 3005 significant (positive and negative) interactions across 197 bacterial groups [
1].
Most of the reported studies have inferred the inter-microbial interactions based on the co-occurrence patterns of various microbes across samples [
1,
15‐
17]. In other words, a pair of bacterial/archaeal species was considered to ‘interact’ if their abundance profiles exhibited co-occurrence or mutual exclusion across multiple samples. Since the relationships between different microorganisms are predicted based on the similarity/dissimilarity in their abundance patterns in various samples, a correlation-based analysis is a key step in inferring the microbial association networks (in a given environment). While the edges in the co-occurrence network indicate positively correlated species, they depict negatively correlated species in the mutually exclusive network.
The co-occurrence as well as mutual exclusion networks can be utilized to investigate whether changes (or aberrations) in these networks can be associated with any disease or physiological disorder. The gut harbours one of the largest microbial communities in the human body. This microbial community is sensitive to environmental factors like diet, antibiotics as well as exposure to pathogens [
18‐
21]. A recent study indicated variations in the gut microbial co-occurrence networks of individuals with varying nutritional status [
22]. However, a comprehensive comparison of the gut microbial interaction networks from individuals belonging to diverse geographical locations is unavailable till date.
The motivation of obtaining and investigating a global (cross-geographic and cross-age-group) picture of the gut microbial communities, both in terms of their taxonomic composition as well as the inherent inter-microbial interaction networks therein, forms the basis of the current study. In this study, we have performed a comparative investigation of the gut microbial communities based on the available gut microbiomes from 399 individuals of various age groups and belonging to eight different nationalities. We have also performed an association analysis of the dietary consumption profiles of these nationalities with the composition of the gut microbiomes of the studied subjects, as well as the architecture of the inherent interaction networks.
Discussion
The current computational analysis was performed with the objective of profiling not only the microbial composition landscape of gut microbiomes of individuals belonging to different geographies and age-groups, but also their microbial interaction patterns. To the best of our knowledge, this is the first study of its kind performed on a large dataset (399 individuals from eight nationalities). The study indicated distinct geography as well as age-group specific trends, with respect to the composition, diversity and intra-group heterogeneity of gut microbial communities of individuals. However, unlike the signature trends in the community composition and diversity of gut microbes across geographies, the trends pertaining to gut microbial association/mutual-exclusion networks were found to be relatively cross-geographic, with distinct geography specific trends in the overall network properties of gut microbiomes. These specific patterns could be due to the inherent differences in the diet of individuals belonging to different nationalities. In order to evaluate whether the observed geography specific gut microbial networks are in accordance with the dietary habits of different nationalities, we attempted to investigate the relationships between the composition and network properties of the gut microbiome for different nationalities and the corresponding population level dietary statistics. Since the dietary intake patterns of the subjects whose gut microbiomes were analyzed in the current analysis were not profiled in the original studies, we considered nation-specific diet intake patterns from the Food and Agricultural Organization of the United Nations information repository (
http://faostat3.fao.org/download/FB/FBS/E). The analysis showed a correspondence between the per-capita intakes of the various food components across different nationalities (Additional file
17) and the respective gut microbiome structures.
Interestingly, many patterns observed in the diet analysis does corroborate with results from previous studies. For example, some genera belonging to Clostridia class, like Roseburia, Butyrivibrio, Eubacterium and Clostridium, were observed to be positively associated with the intake of meat (including fish), animal fat, milk, eggs and oils. This is in line with previous studies [
18,
36] which had shown that shifting to a high fat diet causes a shift in gut microbial composition having increased abundances of these specific genera. Faecalibacterium was observed to have a noticeably higher association with intake of vegetables and fruits as compared to other Firmicutes genera like Roseburia and Eubacterium (Additional file
15), which is in line with the observation reported in a previous study [
33]. The highest number of positive associations with different food items observed for Bacteroides could be a reflection of the high substrate versatility of species belonging to this genus. Notably, genome of
Bacteroides thetaiotamicroton has been shown to encode more than 200 families of carbohydrate active enzymes, indicating a higher variation in its substrate preferences [
37,
38]. In this regard, the most interesting observation made in the present study pertains to the differential association of dietary components with the driver genera for the two well known enterotypes [
7], namely Bacteroides and Prevotella. In contrast to the positive associations of Bacteroides with the wide range of dietary components (especially those rich in protein and animal diet), Prevotella was observed to be specifically associated with vegetarian contents like pulses, starchy roots, sugar crops and vegetables. This observation is in line with that obtained in a previous study which investigated the linkages between changes of gut microbial composition in individuals with long-term dietary patterns [
34].
In addition, association between the intakes of various food categories with the overall properties of the co-occurrence networks was more evident from the present study. It has been reported earlier that, based on the resources available in an environment, the functional interdependence or the interactions between co-occurring genera may be driven by metabolite exchanges between them [
2]. In the present study, the density of the co-occurrence networks, indicative of the functional interdependencies, were observed to decrease with consumption of food products like vegetables, eggs, tree nuts, fruits, milk, aquatic and sea food. On the other hand, the interdependencies amongst resident microbes in the gut were found to increase with the consumption of meat, animal fats, cereals, pulses and sugars/sweeteners. This is in accordance with the reported hypothesis of dysbiosis in the gut microbiome with consumption a high-fat-high-protein diet [
18,
33,
36]. Interestingly, increased functional interdependencies with higher nutritional status have also been reported earlier [
22]. Furthermore, the decrease in the functional interdependency amongst microbial community (network density) with food diversity could explain the apparent differences observed in the gut microbial co-occurrence networks for the Danish and the Spanish populations. Although belonging to the same continent, the Danish individuals were observed to have a much higher degree of functional interdependence (in terms of the average degree of nodes) in their gut co-occurrence networks as compared to the Spanish individuals. This trend is explained by the differences in the dietary patterns between the Spanish and Danish populations. The Spanish population have an evenly distributed diet compared to Danish population, indicated by a higher Berry Index as well as a higher HFD (Additional file
18). Previous studies have indicated that the microbes having similar nutrient preferences tend to co-occur together [
39]. Consequently, the gut of individuals having a homogenous diet (that is dominated by specific constituents) is likely to favour the growth of inter-dependent species having strong co-occurrence relationships among each other. On the other hand, gut of individuals having a highly variable or diverse diets (in terms of the different constituents) are likely to result in the growth of diverse bacterial groups with different nutrient preferences having lesser functional interdependence/competition among each other. The higher diversity in diet could therefore be a reason for the lesser interdependence among microbial genera observed for the Spanish individuals.
The most interesting observation of the current study is that the similarities in the genera level composition of interaction networks were not observed to be dictated by ethnicities or geographical proximity. A key example is the similarity between the microbial interaction networks of the Chinese and American populations (as compared to Japanese/Indian and European populations). The cross ethnicity similarities in the interaction patterns of gut microbiota is an interesting observation that requires further validation and profiling of the environment, hygiene as well as life-style habits (including dietary intake) of individuals belonging to different nationalities.
Distinct similarities in the gut microbial interaction networks, based on the age of the individuals were observed from the present study. Networks obtained for individuals below 30 years of age (G1–G3) were observed to be similar and distinct from those obtained for the age groups above 40 years. A key distinguishing factor was the increased functional interdependence between bacterial groups in the networks in the higher age groups.
In spite being the first of its kind study, the current study has distinct limitations primarily pertaining to the composition of cohorts constituting the age-groups and geographies. First, since some of the cohorts, especially those corresponding to French and Italian populations, have lower sample size, the reliability of the results obtained are specifically lower for these nationalities. Further, there is an inherent bias in the composition of individuals constituting the age-wise cohorts below 10 years of age, which primarily belonged to Indian nationality. Furthermore, sub-population specific biases could also occur in the diet-microbiota association analysis performed in the current study. This is because, while the dietary intake patterns used in the current study are population-level statistics obtained for entire nationalities, the gut microbiomes are only obtained from specific individuals in distinct neighbourhoods of a given country. Given the global nature of this study, as well as a more or less even representation of individuals in a majority of cohorts, a concordance was observed between the results obtained in the current study with previous reports, potential biases mentioned above. These inferences from the current study are likely to form the basis for future metagenomic investigations across much larger cohorts of individuals from specific regions.