Artificially selecting microbial communities: If we can breed dogs, why not microbiomes?

Abstract

Natural microbial communities perform many functions that are crucial for human well-being. Yet we have very little control over them, and we do not know how to optimize their functioning. One idea is to breed microbial communities as we breed dogs: by comparing a set of microbiomes and allowing the best-performing ones to generate new communities, and so on. Although this idea seems simple, designing such a selection experiment brings with it many decisions with surprising outcomes. Xie and colleagues developed a computational model that reveals this complexity and shows how different experimental design decisions can impact the success of such an experiment.

Fig 1. Multiple microbial communities are created and incubated in parallel (colors represent different species or lineages, and white space represents cell-free growth medium) for a given amount of time as cells grow and divide.

Then, the trait value of each ecosystem is measured, communities are ranked, and the best communities are redistributed at a reduced initial population size to make a new selection round. Communities do not necessarily start with the same biomass or reach the same biomass after incubation. There are many ways to implement selection and redistribution. Figure adapted from [6].

Fig 2. Illustration of the problems associated with microbial community selection.

Each tube contains a community. Problems 1, 2, 3, and 5 show sequential rounds of selection, whereas problem 4 shows different communities within the same selection round. Letters represent species, and colors represent lineages (the two shades of green species A represent an ancestor and a mutant strain). Below each tube in a box is a trait value, which is used to select the best communities to be transmitted to the next selection round. Problems are described in Box 1.

Fig 3. Xie and colleagues optimize community selection.

(A) Their system is as follows: species 1 takes up resource R and makes by-product B, whereas species 2 takes up R and B to make the product of interest P. (B) For this system, the amount of product P a community produces is a function of biomass concentration: a community starting with greater biomass will make more product if selection rounds are short. Lengthening the selection round makes sure a community is evaluated by how much its cells produce individually rather than by how much biomass it has accumulated. Similarly, the initial fraction of species 2 affects community score if one selects early but not late. (C) Because of noise in initializing new communities, communities vary in their initial biomass and the fraction of species 2. The plots show the distribution of communities in a given round. Using flow cytometry reduces the noise compared with pipetting. Less noise means less bias in the resulting community scores because of these initial states. (D) Two selection methods were compared. In “top-dog” selection, the best-scoring community from the previous round is distributed among the fresh tubes of the next round until no cells are left, then the next best community is distributed, etc. The best community is highly favored. The “top-tier” selection method is less stringent because all top 10% of communities are equally distributed among the tubes of the next round. This second selection method worked better in Xie and colleagues’ model.