Trusting the hand that feeds: microbes evolve to anticipate a serial transfer protocol as individuals or collectives

Abstract

Background: Experimental evolution of microbes often involves a serial transfer protocol, where microbes are repeatedly diluted by transfer to a fresh medium, starting a new growth cycle. This has revealed that evolution can be remarkably reproducible, where microbes show parallel adaptations both on the level of the phenotype as well as the genotype. However, these studies also reveal a strong potential for divergent evolution, leading to diversity both between and within replicate populations. We here study how in silico evolved Virtual Microbe "wild types" (WTs) adapt to a serial transfer protocol to investigate generic evolutionary adaptations, and how these adaptations can be manifested by a variety of different mechanisms.

Results: We show that all WTs evolve to anticipate the regularity of the serial transfer protocol by adopting a fine-tuned balance of growth and survival. This anticipation is done by evolving either a high yield mode, or a high growth rate mode. We find that both modes of anticipation can be achieved by individual lineages and by collectives of microbes. Moreover, these different outcomes can be achieved with or without regulation, although the individual-based anticipation without regulation is less well adapted in the high growth rate mode.

Conclusions: All our in silico WTs evolve to trust the hand that feeds by evolving to anticipate the periodicity of a serial transfer protocol, but can do so by evolving two distinct growth strategies. Furthermore, both these growth strategies can be accomplished by gene regulation, a variety of different polymorphisms, and combinations thereof. Our work reveals that, even under controlled conditions like those in the lab, it may not be possible to predict individual evolutionary trajectories, but repeated experiments may well result in only a limited number of possible outcomes.

Keywords: Digital microbes; Eco-evolutionary dynamics; Experimental evolution; In silico evolution; Predicting evolution; Resource cycle; Serial transfer protocol; Virtual microbes.

Fig. 1

Virtual Microbes model overview. a At the basis of the Virtual Microbe model is an artificial “metabolic universe”, describing all the possible reactions that can be catalysed. Resources (yellow and blue) are fluxed in, but building blocks (purple) and energy (red) must be synthesized to express proteins and transport metabolites across the membrane, respectively. b A Virtual Microbe only needs to express a subset of all possible reactions to be viable, and no metabolic strategy is necessarily the “right” one. c The individuals grow and reproduce on a spatial grid, and can only reproduce when there is an empty spot. Death happens stochastically or when a cell has accumulated toxicity by having excessively high concentrations of metabolites. Since only cells that have grown sufficiently are allowed to reproduce, we simulate evolution with no prior expectation

Fig. 2

Evolution of Virtual “wild types” under naturally unpredictable and fluctuating resource conditions. a Natural evolution is mimicked by (harsly) fluctuating resource conditions, resulting in a wide variety of resource conditions. The (actual) grid is 40x40, with four 20x20 subspaces where the rates of influx vary stochastically. These subspaces do not impede diffusion of metabolites or reproduction. The fluctuations of the A and C resource (blue and yellow respectively) are independent, resulting in a variety of different conditions. b We repeat the evolution in natural conditions 16 times starting from the same (minimally viable) initial clone (varying the mutations that happen) yielding 16 distinct WTs. These WTs are later transfered to a serial transfer protocol. c In the white labels we show how many of the evolved WTs adapted to use particular reactions. The thicker arrows represent the shared core genome which consists of two resource importers, a metabolic cycle, and a C-exporter (yellow). Transcription factors (diamonds) were always present across WTs, but only 11/16 WTs visibly display changes in gene expression correlated with changes in the environment

Fig. 3

Virtual Microbes adapt to anticipate the regularity of a serial transfer protocol. a Growth dynamics of early population (green) and evolved populations (blue) in terms of cell counts. (WT03#1 taken as an illustrative example). b-c Two WTs (green) and the population after prolonged evolution in the serial transfer protocol (blue) are shown as an illustration of the anticipation effects. Over the course of 3 cycles, the average cell volume is plotted against time for the ancestral WT (green) and for the evolved population (blue). The y-axis (cell volume) indicates the minimal viable volume and division volume (which are fixed for the model), and the evolved volume-at-transfer (as measured at the end of the third cycle). Daily and extended yield are measured as defined in the method section. After the third cycle, serial transfer is stopped (transparent area), showing decreased survival of the evolved populations with respect to their ancestor. d Stacked density distributions are plotted for the volume-at-transfer both early (transfer 0-40, green) and late (transfer 760-800, blue). e The evolved changes in yield both “daily” (within one cycle of the protocol) and “extended” (after prolonged starvation) for all 16 WTs

Fig. 4

Trajectories towards a growth versus yield trade-off end in either the high growth rate mode or the high yield mode. a Growth rate (average building block production rate) is plotted against daily yield (average population biomass within a single cycle), for all the 48 experiments after adaptation to 800 serial transfers. The black dotted line is a linear regression model (R² = 0.54). b Shows the initial points for all 16 WTs, which actually have a positive correlation between growth and yield (R² = 0.32) instead of the negative correlation (black dotted line). c-e These insets display how the repeated evolution of certain WTs produce very similar trajectories towards the trade-off (time points are day 0, 20, 40, 100, 200 and 800), ending in either high daily yield (c) or low daily yield (d). Other WTs diverge after reaching the trade-off, and thus show more diverse trajectories when repeated (e). The colours of the end point symbols depict different modes of adaptation as discussed in the next paragraph (grey = no coexistence, purple = (quasi-)stable coexistence, black cross = extinction due to over-exploiting the medium)

Fig. 5

Dynamics of neutral lineage markers reveal balanced polymorphisms based on the daily cycle. a-c Neutral lineage marker (random colours) frequencies are plotted along 800 serial transfers (left hand side) and along 3 cycles. Panel A shows an example with no coexistence which is found in 23 out of 44 replicates, and panel B and C show (quasi-)stable coexistence, found in the remaining 21 replicates. d shows, for all 3 replicates of all WTs whether or not coexistence of neutral lineage markers was observed (grey = no coexistence, purple = (quasi-)stable coexistence, black cross = extinction due to over-exploiting the medium). Also see Additional file 1: Figure S8

Fig. 6

Anticipation can entail polymorphism or a single lineage that switches between two metabolic modes. a Two lineages occupy different niches on the growth vs. yield trade-off WT02#01 diverges into a slow growing lineage (yellow lineage, average death rate ±0.015) and a faster growing lineage with elevated death rates (blue lineages, average death rate ±0.048), together anticipating the serial transfer protocol. b A single lineage anticipates the daily cycle by trimming and tuning the gene regulatory network. On the left the ancestral GRN, protein allocation dynamics, and resource concentrations are displayed over the course of 1 day. Next, after 400 days, all three independent simulations of WT07 are shown to have evolved to anticipate as a single lineage with two metabolic modes

Fig. 7

Individual and collective solutions have similar macro-level observables The daily yield for all the evolved populations is shown, for groups of individual / collective solutions with and without regulated gene expression. Colours and symbols are identical to previous figures (grey=no coexistence, purple=coexistence). Only the non-regulating, individual lineages perform significantly worse than any of the other groups (performing all 6 Wilcoxon rank-sum tests with α 0.05)