Inhibitory interactions promote frequent bistability among competing bacteria

Abstract

It is largely unknown how the process of microbial community assembly is affected by the order of species arrival, initial species abundances and interactions between species. A minimal way of capturing competitive abilities in a frequency-dependent manner is with an invasibility network specifying whether a species at low abundance can increase in frequency in an environment dominated by another species. Here, using a panel of prolific small-molecule producers and a habitat with feast-and-famine cycles, we show that the most abundant strain can often exclude other strains--resulting in bistability between pairs of strains. Instead of a single winner, the empirically determined invasibility network is ruled by multiple strains that cannot invade each other, and does not contain loops of cyclic dominance. Antibiotic inhibition contributes to bistability by helping producers resist invasions while at high abundance and by reducing producers' ability to invade when at low abundance.

Figure 1. Scheme for measuring pairwise invasions and potential outcomes.

(a) Pairs of bacterial strains were added to test tubes at vastly different initial abundances and propagated for three cycles. Their relative abundance was quantified with high-throughput sequencing. (b) Each strain from a pair was competed twice, as either resident (high abundance) or invader (low abundance). The three potential outcomes are: bistability if neither resident was invaded, hierarchy if one resident was invaded and coexistence if both residents were invaded. (c) The invasion network for a panel of strains may be either completely hierarchical, where strains can be ranked by relative fitness in a way that explains all pairwise outcomes, partly hierarchical with a small fraction of non-hierarchical features, or essentially non-hierarchical.

Figure 2. Widespread bistability in pairwise invasions.

(a) Pairwise invasion matrix for a panel of 18 diverse Streptomyces strains. Strains are sorted by phylogeny constructed from partial rpoB gene sequences. Strain #1 is present in two replicas (labeled 1a,b). (b) Bistable pairings, in which two strains cannot invade each other, were a dominant feature of the invasion matrix. Coexistence was less frequent and mostly limited to strain #1, which was also the most phylogenetically distinct strain. (c) A few strains were involved in many bistable pairings. These ‘hubs of bistability' were more frequent than in randomized matrices with the same number of each type of pairwise link (P=1.7e−4).

Figure 3. The invasion network is partly hierarchical with multiple strains in the top level exhibiting bistable relationships with each other.

(a) Enrichment (green) or depletion (pink) of different triplet motifs relative to randomized networks preserving the number of each pairwise motif (Supplementary Fig. 3). Number of occurrences for each motif are given in the upper-left corner. Transitive invasions (leftmost motif) were highly enriched (***,P<1e−6), whereas the three intransitive motifs were highly depleted (***,P<1e−6). (b) The scoring scheme used in assigning hierarchy levels to strains rewards invasions pointing down the hierarchy and penalizes invasions directed against the hierarchy. (c) The invasion network overlaid on the hierarchy assignments that maximize the score in b. Strains were placed into seven hierarchy levels with six strains at the top level exhibiting bistable relationships with each other. Invasions going down the hierarchy are not shown, while others are shown in red. Bistability is denoted by a missing link between strains at the same level or by a dashed line for strains at different levels.

Figure 4. Antibiotic inhibition helps bistability.

(a) Inhibitions in the cross-streaking assay (Supplementary Fig. 6) overlaid on the invasion hierarchy. Almost all inhibitions were directed down the hierarchy (black inhibition arrows) and only a few were directed against the hierarchy (red inhibition arrows). (b) The number of pairs with invasions (blue) and non-invasions (black) are shown for cases in which the invader is inhibited (left pie-chart) or not inhibited (right pie-chart). The vastly different fraction of invasions in the two pie-charts indicates that inhibition greatly assists residents in resisting invasion. (c) The number of pairs with invasions (blue) and non-invasions (black) are shown for cases in which the resident is inhibited (left pie-chart) or not inhibited (right pie-chart). Only pairs in which the invader is at a higher hierarchy level (3⩾h_A−h_B>0) are considered to control for the tendency of inhibitions to point down the hierarchy. (d) The number of bistable (blue) and non-bistable (black) pairs is shown for pairs with inhibition (bottom pie-chart) and without inhibition (top pie-chart). Strains #14 and #6 at the bottom of the hierarchy were not included, as they were invaded by almost all other strains. A significant enrichment for bistability is evident among pairs with an inhibitory interaction.