Controlling the human microbiome

Abstract

We coexist with a vast number of microbes that live in and on our bodies. Those microbes and their genes are collectively known as the human microbiome, which plays important roles in human physiology and diseases. We have acquired extensive knowledge of the organismal compositions and metabolic functions of the human microbiome. However, the ultimate proof of our understanding of the human microbiome is reflected in our ability to manipulate it for health benefits. To facilitate the rational design of microbiome-based therapies, there are many fundamental questions to be addressed at the systems level. Indeed, we need a deep understanding of the ecological dynamics associated with such a complex ecosystem before we rationally design control strategies. In light of this, this review discusses progress from various fields, e.g., community ecology, network science, and control theory, that are helping us make progress toward the ultimate goal of controlling the human microbiome.

Keywords: community ecology; control theory; human microbiome; network science.

Figure 1:. The ecological network associated with a microbial community can have two different representations with different levels of complexity.

a, The first representation is a bipartite graph connecting two types of nodes: microbial species and chemical compounds (e.g., nutrients, metabolites, signaling molecules, toxins, etc.). Species can consume or produce consumable chemical compounds (e.g., metabolites); while reusable chemical compounds (e.g., signaling molecules and toxins) can stimulate or inhibit the growth of species. b, The second representation is a unipartite graph where nodes represent microbial species and edges represent pairwise inter-species interactions. One species can promote or inhibit the growth of another species. The unipartite graph can be considered as a projection of the bipartite graph onto the species nodes. Although the projection is not perfect, it does simplify the network reconstruction problem. Figure courtesy of Dr. Xu-Wen Wang.

Figure 2:. The human gut microbiome is highly personalized and very stable.

a, The taxonomic profile of the human gut microbiome varies a lot across different individuals. Here the stacked bar chart demonstrates the phylum-level gut microbial compositions of ~200 healthy adults in the HMP cohort. b, The taxonomic profile of the human gut microbiome is highly dynamic but very stable. In the absence of drastic interventions, the human gut microbiome can be considered as a dynamically stable ecosystem, continually buffeted by internal and external forces and recovering back toward a conserved steady-state. Here the stacked bar chart demonstrates the daily phylum-level gut microbial compositions of a healthy adult over ~200 days in the Moving Picture study. Figure courtesy of Dr. Xu-Wen Wang.

Fig. 3:. A control theoretical framework.

a, A toy community of N = 3 species (green, yellow, blue) with microbial interactions encoded in an ecological network 𝒢. The controlled ecological network 𝒢^c contains one control input driving species-3. b, Initial and desired abundance profiles shown in stacked bars. The control objective is to steer the community from the (undesired) initial state x₀ to the desired final state x_d, represented by two points in the state space of the system. c, In the continuous control scheme, the control inputs u(t) are continuous signals modifying the growth of the actuated species. d, In the impulsive control scheme, the control inputs u(t) are impulses applied at the intervention instants $\mathbb{T}=\left\{t_1,t_2,\cdots \right\}$, instantaneously changing the abundance of the actuated species. e, A minimum set of driver species can be identified from the ecological network 𝒢 by checking the graph-theoretical conditions of structural accessibility. Here, we show an ecological network involving the GnotoComplex microflora (a mixture of human commensal bacterial type strains) and C. difficile, inferred from mouse data (assuming the GLV model). Red (or blue) edges indicate the direct promotion (or inhibition), respectively. The five driver species are driven by five independent control inputs. f, Projection of the high-dimensional abundance profiles (states of the microbial communities) into their first three principal components (PCs). The calculated control strategies applied to the driver species succeed in driving the community to the desired state, using either continuous or impulsive control. Here, the controlled population dynamics is simulated using the controlled GLV equations. The intrinsic growth rates were adjusted such that the community has an initial “diseased” equilibrium state x₀ in which C. difficile is overabundant compared to the rest of species. We chose the desired state x_𝑑 as another equilibrium with a more balanced abundance profile. Figure adapted and modified from Ref..

Fig. 4. Personalized probiotic cocktails effectively decolonize C. difficile.

a, An ecological network involving the GnotoComplex microflora (a mixture of human commensal bacterial type strains) and C. difficile was inferred from mouse data. Red (or blue) edges indicate the direct promotion (or inhibition), respectively. b, A disrupted microbiota due to a hypothetic antibiotic administration. c, The restored microbiota due to the administration of a particular probiotic cocktail R_global. d, The trajectory of C. difficile abundance over three different time windows: (1) the initial healthy microbiota, (2) the disrupted microbiota, and (3) the microbiota post probiotic administration. In the third time window, we compare the performance of various probiotic cocktails in terms of their ability to decolonize C. difficile. Those cocktails were designed by considering the global ecological network (R_global), the ego-network of C. difficile (R_ego), and randomly chosen subsets of R_global (R₁, R₂ and R₃). R_{near–optimal} is obtained by excluding species-12 (i.e., K. oxytoca, which is an opportunistic pathogen) from R_global. e-h, We start from the same initial microbiota as shown in (a), but another hypothetic antibiotic administration leads to a different disrupted microbiota (f), which can be restored through another probiotic cocktail (g). Performance of different probiotic cocktails in decolonizing C. difficile vary (h). Note that since the disrupted microbiota (f) is different from that shown in (b), the optimal cocktail R_global in (h) is also different from that shown in (d). Figure adapted and modified from Ref..