A behavioral model for mapping the genetic architecture of gut-microbiota networks

Abstract

The gut microbiota may play an important role in affecting human health. To explore the genetic mechanisms underlying microbiota-host relationships, many genome-wide association studies have begun to identify host genes that shape the microbial composition of the gut. It is becoming increasingly clear that the gut microbiota impacts host processes not only through the action of individual microbes but also their interaction networks. However, a systematic characterization of microbial interactions that occur in densely packed aggregates of the gut bacteria has proven to be extremely difficult. We develop a computational rule of thumb for addressing this issue by integrating ecological behavioral theory and genetic mapping theory. We introduce behavioral ecology theory to derive mathematical descriptors of how each microbe interacts with every other microbe through a web of cooperation and competition. We estimate the emergent properties of gut-microbiota networks reconstructed from these descriptors and map host-driven mutualism, antagonism, aggression, and altruism QTLs. We further integrate path analysis and mapping theory to detect and visualize how host genetic variants affect human diseases by perturbing the internal workings of the gut microbiota. As the proof of concept, we apply our model to analyze a published dataset of the gut microbiota, showing its usefulness and potential to gain new insight into how microbes are organized in human guts. The new model provides an analytical tool for revealing the "endophenotype" role of microbial networks in linking genotype to end-point phenotypes.

Keywords: The gut microbiota; competition; cooperation; network science; qtl; rule of thumb.

Figure 1.

Scatter plots of mathematical descriptors of mutualism (Z_mu), antagonism (Z_an), aggression (Z_ag), and altruism (Z_al) against the actual strength of mutualism (M_u) (a), antagonism (A_n) (b), aggression (A_g) (c), and altruism (A₁) (d) across 100 interspecific pairs of strains from E. coli and S. aureus at three distinct phases of microbial growth (lag, log, and stationary). Dots represent observations of different interspecific strain pairs at each time point. Note that the mutualism dots are those strains whose abundance is larger in co-culture than monoculture for both species, whereas the antagonism dots are those strains whose abundance is larger in monoculture than co-culture for both species. The relationship between two variables is roughly fitted by a curve, with correlation coefficient (r) given within each plot

Figure 2.

Microbial Z_mu-based mutualism (two-way arrowed line) networks (a), Z_an-based antagonism (two-way T-shaped line) networks (b), Z_ag-based aggression (one-way T-shaped line) networks (c), and Z_al-based altruism (one-way arrowed line) networks (d) at the genus level within the gut microbiota of the Hutterites in winter and summer. In each network, hub microbes are highlighted in dark circles. These hub microbes, expressed as leaders, antagonists, hawks, and beneficiaries in mutualism, antagonism, aggression, and altruism networks, respectively, are compared with other microbes from each network type, called followers, agonists, doves, and altruists, respectively, in bar graphs. The significance of the difference between each pair of these types was tested by a t-test statistic. The identity of each genus is labeled by a number (Table S1). The distribution of links owned by each genus within each network is given in the middle, separately for winter (w) and summer (s)

Figure 3.

Heatmaps of six indices (showing emergent network properties) constituting mutualism (a), antagonism (b), aggression (c), and altruism networks (d) among 101 genera for network properties for winter and summer

Figure 4.

Path analysis revealing how QTLs (outer) affect BMI as a final phenotype (inner) through the season-driven perturbation of microbial networks as an “endophenotype” (middle) (described by differences of six emergent property indices between winter and summer). Path coefficients are denoted by directed lines from QTLs to BMI (gray) and from networks to BMI (blue). Arrowed line and T-shaped line represent a positive and negative impact of path, respectively. Correlation coefficients between QTLs and networks are denoted by non-directed lines. In all cases, the magnitudes of path and correlation coefficients are proportional to the thickness of lines