Fig. 3: Subpopulation enrichment simulations incorporating growth-lysis linear correlation, variability, and random interactions.

a, b Illustration of an increasing variability among the different populations in their G/L correlations. Starting from a strict linear correlation of growth and lysis rates (G/L correlation) on the left, lysis rate variability was introduced by changing the slope and intercept of the G/L correlation to the right (10 and 20% of the normally distributed random numbers were added, see Results and Methods section for detailed procedures). The G/L correlations of subpopulations are depicted as solid lines, showing the lysis dynamics with respect to the growth rate modulation (e.g., due to approaching its carrying capacity or interactions with other populations). The marker representation follows that of Fig. 2. b shows the same lysis dynamics but is represented as the net rate (G–L) over maximum growth rates (G_m). c Fraction changes of subpopulations are shown over time when the simulations incorporate G/L variability and interactions. To the base case (no variability and interactions) shown in Fig. 2, G/L variability of corresponding degrees in Fig. 3a was introduced (column-wise). No interaction was introduced to the first row. The maximum interaction magnitude, max |I_ij|, was set to 15 and 30 for rows 2 and 3, respectively. One random interaction matrix (I) was generated and shared by scaling up the elements. Therefore, the fraction changes in the first phases (-A, growth only) are the same in each row. Different results of the fraction changes in the second phases are due to different lysis rates (column-wise), and interactions upon lysis (row-wise). d Antibiotic-mediated community restructuring is primarily driven by G/L correlations and further pushed by G/L variability and interactions among different subpopulations. EFs of each phase were calculated from c. The color and size of the maker follow the same indicators in a. The linearly correlated antibiotic-killing kinetics resulted in the linearly associated EFs of the community, close to the baseline enrichment dynamics. In increasing G/L variability, the EFs scattered from this tight linear association to all four quadrants, enable the diverse enrichment outcome. Increasing the magnitude of interaction terms (top to bottom rows) shuffled the order and broke the alignment of G_m and net rate rankings to EF₁ and EF₂, respectively, while the tight linear association of EFs remained. Combining two factors, EFs showed a shuffled, loose linear association spread from the baseline enrichment dynamics. The overlayed histogram shows the counts of slower (orange) and faster (blue) subpopulations along with the EF₁.