Predicting bacterial interaction outcomes from monoculture growth and supernatant assays

Abstract

How to derive principles of community dynamics and stability is a central question in microbial ecology. Bottom-up experiments, in which a small number of bacterial species are mixed, have become popular to address it. However, experimental setups are typically limited because co-culture experiments are labor-intensive and species are difficult to distinguish. Here, we use a four-species bacterial community to show that information from monoculture growth and inhibitory effects induced by secreted compounds can be combined to predict the competitive rank order in the community. Specifically, integrative monoculture growth parameters allow building a preliminary competitive rank order, which is then adjusted using inhibitory effects from supernatant assays. While our procedure worked for two different media, we observed differences in species rank orders between media. We then parameterized computer simulations with our empirical data to show that higher order species interactions largely follow the dynamics predicted from pairwise interactions with one important exception. The impact of inhibitory compounds was reduced in higher order communities because their negative effects were spread across multiple target species. Altogether, we formulated three simple rules of how monoculture growth and supernatant assay data can be combined to establish a competitive species rank order in an experimental four-species community.

Keywords: agent-based model; bacterial interactions; community dynamics; competitiveness.

Figure 1

Growth differences of the four bacterial species in two different media: LB medium and GIM. (A) Growth curves over 36 h in shaken, liquid cultures. Shaded areas depict the standard deviation, and boxplots show the (B) maximum growth rate μ_max as OD₆₀₀ increase per hour (C) the inverse of the time to mid-exponential phase [1/T_mid], and the (D) integral (AUC). All growth parameters are derived from the growth curves in (A) using the Gompertz curve fit. Different letters above boxplots indicate significant growth differences (alpha = 0.05) between bacterial species using linear mixed models with experimental block as random factor. Boxplots show the median (line within the box) with the first and third quartiles, and the whiskers cover 1.5× of the interquartile range or extend from the lowest to the highest value if all values fall within the 1.5× interquartile range. Data are from 3 independent experiments, each featuring 3–4 replicates per condition, resulting in a total of 9–10 replicates per condition.

Figure 2

Boxplots show the relative growth of each species in the conditioned medium (30% supernatant + 70% fresh medium) of the other species both in (A) LB and (B) GIM medium, compared with a control treatment, depicted by the black dashed line (30% NaCl solution [0.8%] + 70% fresh medium) measured across 36 h of growth. For this plot, we used 1/T_mid for growth comparisons (see Fig. S3 for absolute readouts, see online supplementary material for a colour version of this figure), and we repeated the same analysis for μ_max (Figs S4, S5, see online supplementary material for a colour version of these figures) and AUC (Figs S6, S7, see online supplementary material for a colour version of these figures). Relative growth was calculated by dividing the absolute 1/T_mid, (estimates from curve fits) in the supernatant treatments by 1/T_mid in the control treatment. Asterisks depict significant differences (alpha = 0.05) of a species’ growth in the particular supernatant compared with its growth in the control medium using a linear mixed model with experimental block as random factor. Boxplots depict the median (line within the box) with the first and third quartiles; the whiskers cover 1.5× of the interquartile range or extend from the lowest to the highest value if all values fall within the 1.5× interquartile range. Data are from 3 independent experiments, each featuring 3–4 replicates per condition, resulting in a total of 9–10 replicates per condition. The underlying growth curves can be found in the supplementary information (Fig. S8, see online supplementary material for a colour version of this figure).

Figure 3

Boxplots show the fitness of the species indicated in the header relative to all other species in co-culture assays in (A) LB and (B) GIM medium. Asterisks depict significant differences (alpha = 0.05) of a species’ fitness relative to that of a competitor against the null hypothesis that none of the two species has an advantage (relative fitness = 0, black dashed line) using one-sample two-sided Wilcoxon rank tests. At relative fitness = 0, two species coexist at equal frequency. Boxplots show the median (line within the box) with the first and third quartiles, and the whiskers cover 1.5× of the interquartile range or extend from the lowest to the highest value if all values fall within the 1.5× interquartile range. Data are from six individual experiments with two replicates per condition, resulting in a total of 7–12 replicates per condition (note in a few cases sample size was <12 because obtaining countable colonies for both species in co-culture was very difficult). The corresponding figure with the absolute readouts can be found in the supplementary information (Fig. S9, see online supplementary material for a colour version of this figure).

Figure 4

(A) Species rank orders of competitive strength inferred from monoculture growth, the ability of a species to affect the growth of competitors in conditioned medium (supernatant), and co-culture experiments. All in vitro experiments were conducted in two different media: LB and GIM. For monoculture growth and supernatant assays, rank orders are shown for maximum growth rate (μ_max), the inverse of the time to mid-exponential phase (1/T_mid), and the growth integral (AUC). For the supernatant assays, we summed up all growth effects a species has on the other species and scaled them across species. To scale relative fitness values in co-culture assays, we implemented a stepwise process to avoid double-counting of the reciprocal fitness value. We started with the weakest species and summed up all its relative fitness values. Then, we moved to the second (third) weakest species and repeated the procedure leaving out any fitness effects that were already accounted for before. For all experiments, we scaled the values from weakest (left) to strongest (right) performer. A line demarks the transition from neutral/positive to negative effects on the other species. (B) Species rank order from in vivo competition experiments in the larvae of G. mellonella 12-h postinfection from a previous study [22]. The latter includes all raw data, while Fig. S10 (see online supplementary material for a colour version of this figure) depicts the relative fitness values underlying this figure. The calculation of the species rank order was done as for the co-culture assays. Both co-culture and in vivo rank order calculations are based on log10(CFU/mL) values. For more information regarding all calculations, please refer to Materials and methods section and Table S4 in the statistical analysis file.

Figure 5

Bacterial community dynamics simulated with an agent-based model and parameterized with experimental growth and inhibition data. Panels show the (1) species fractions over time, (2) mean species fraction over time, and (3) per cell toxin uptake for (A) the 4-species community and (B–E) all combinations of 3-species communities. All simulations were carried out under low diffusion (cell diffusion D = 0.0 μm² s⁻¹, toxin diffusion ∂ = 0.1 μm² s⁻¹, a structured environment) and high diffusion (cell diffusion D = 5.0 μm² s⁻¹, toxin diffusion ∂ = 10.0 μm² s⁻¹, an unstructured environment). In (1), lines show the mean and the standard deviation across 20 independent simulations. Boxplots show the median values (line within the box) across the 20 simulations with the first and third quartiles, and the whiskers cover 1.5× of the interquartile range or extend from the lowest to the highest value if all values fall within the 1.5× interquartile range. In (2), the shaded boxes depict the simulated mean species fraction in the absence of toxins.

Figure 6

Boxplots show the number of cells of each species in the four-species and all three-species communities of our bacterial consortium. Different letters above boxplots indicate significant differences (alpha = 0.05) between bacterial species using Wilcoxon paired signed rank tests for multiple pairwise comparisons with P-value adjustments using the FDR method. Boxplots show the median (line within the box) with the first and third quartiles; the whiskers cover 1.5× of the interquartile range or extend from the lowest to the highest value if all values fall within the 1.5× interquartile range. Data are from three individual experiments with three replicates per condition, resulting in a total of nine replicates per condition.