Predictability of microbiome dynamics

概要

Predictability of microbiome dynamics
Hiroaki Fujita

Background:
Microbial communities provide diverse ecosystem functions in a wide range of fields
such as medicine, industry, and agriculture. In stably managing those microbial functions,
forecasting of microbial community processes is of particular importance. However,
predicting microbiome dynamics is notoriously difficult because communities often show
abrupt structural changes, such as “dysbiosis” observed in human-gut and aquaculture
microbiomes. Elucidating the mechanisms and factors driving community structural
changes is one of the key challenges in microbial ecology. In this thesis, I focus on the
predictability and controllability of microbiomes by quantifying community-scale
stability as well as by exploring core species driving community dynamics.

Methods:
I examined the predictability of microbial community dynamics by estimating population
dynamics of constituent microbial species based on quantitative DNA metabarcoding
(Chapter 2). I monitored 48 experimental microbiomes for 110 days and observed that
various community-level events, including collapse and gradual compositional changes,
occurred according to a defined set of environmental conditions. I then applied analytical
frameworks of statistical physics and nonlinear mechanics to examine whether
community-stability indices could allow us to forecast major shifts in microbial
community structure.
I next investigated how community dynamics are driven by species interactions
within the microbial communities (Chapter 3). I compiled microbial community data
from two types of aquatic ecosystems, namely, fish aquaculture and experimental paddy
ecosystems. I then inferred dynamics of causality relationships between microbial species
based on a machine-learning framework. By examining changes in the structure of the
species interaction networks, I assessed potential roles of “phase-specific” species (or
indicator species) that show characteristic population growth in limited periods of time
through community dynamics.

Results:
I found that the abrupt community changes observed through the microbiome time-series
could be described as shifts between “alternative stable states” or dynamics around
complex attractors (Chapter 2). I then showed that collapses of microbiome structure were
successfully anticipated by means of the diagnostic threshold defined with the “energy
landscape” analysis of statistical physics or that of a stability index of nonlinear
mechanics. The results indicate that abrupt microbiome events in complex microbial
communities can be forecasted by extending classic ecological concepts to the scale of
species-rich microbial systems.
In the analysis of microbial interaction networks, I found that indicator species
occupied most upstream positions within the inferred interaction networks and that they
could impose causative impacts on many other species. Moreover, indicator species had
much more positive than negative impacts on the population dynamics of other species. I
further found that such positive interactions were condensed and strengthened within
limited periods of time, reinforcing abrupt alternations of dominant species sets. These
findings suggest that indicator species involved in time-varying webs of positive
interactions are keys to managing and controlling catastrophic shifts in microbial
communities and ecosystems.

Conclusion:
In this thesis, the predictability of microbial communities was empirically demonstrated
for the first time, to the best of my knowledge. I then explored the possibility of
anticipating and thereby preventing large community structural shifts based on
monitoring of community stability indices. Moreover, the reconstruction of ecological
network architecture provided insights into the potential mechanisms by which
alternations of dominant species sets could be promoted by dynamically changing species
interactions. Further empirical studies on experimental and field ecosystems are
necessary to deepen our understanding of the predictability and controllability of
biological community processes. ...