Fig. 3. Migrant pool sampling gives a strong response to artificial ecosystem selection. This response is similar to that achieved with propagule sampling (Fig. 1). Mean F± 1 SE plotted. Here 49 runs were performed using migrant pool sampling with T_prop = 5,000 and P_mut = 0.01 for all runs, with arbitrary target vector (ā₁, ā₂, ā₃) = (0.2, 0.3, 0.5). Data are plotted for directed selection for either increase (high line) or decrease (low line) in distance of the abiotic environment from target state, F, as well as for a random selection control line that shows behavior in the absence of artificial ecosystem selection. Directed selection is stopped after iteration 30, at which point all selection is random.

Fig. 4. Response to artificial ecosystem selection (using propagule sampling method) is achieved with different target vectors for normalized abiotic environmental state. Mean F± 1 S.E. plotted for all runs. Here T_prop = 5,000 and P_mut = 0.01 for all runs, but different target vectors for abiotic environmental states are used. Sub-figure captions give target vectors and number of runs performed for each. Two data sets with different target vectors are shown, supplementing the example shown in the main body of the paper (Fig. 1). Directed selection is stopped after iteration 30, at which point all selection is random.

Fig. 5. Response to artificial ecosystem selection (using propagule sampling) with different mutation rates for individual reproduction. Here the target vector is (ā₁, ā₂, ā₃) = (0.2, 0.3, 0.5) and T_prop = 5,000 for all runs. Mean F± 1 SE plotted for all runs. P_mut takes values from the set {0, 0.01, 0.03, 0.05, 0.1} (number of runs for each value given in sub-figure captions). Fig. 5a: When the chance of mutation during individual reproduction is reduced to zero a strong response to selection is observed, and the adapted response is sustained even after directed selection pressure is removed at iteration 30. Fig. 5 b-e: As the probability P_mut of mutation at each genetic locus during individual reproduction is increased, the response to selection is weakened and relaxation to the nonselected state after iteration 30 is accelerated.

Fig. 6. Response to artificial ecosystem selection (using propagule sampling) when propagation time is varied. Here the target vector is (ā₁, ā₂, ā₃) = (0.2, 0.3, 0.5) and P_mut = 0.01 for all runs. Mean F± 1 SE plotted for all runs. T_prop takes values from the set {2,000, 5,000, 10,000, 20,000} (number of runs for each given in subfigure captions). As T_prop increases, the response to artificial ecosystem selection decreases, showing an inverse relationship between the time between selection events and the elicited response.

Fig. 7. Response to artificial ecosystem selection (using propagule sampling) varies inversely with mutation rate and propagation time. Plots show mean absolute difference in F between high and low lines (±1 SE) after 30 iterations of ecosystem selection, measured on multiple runs with varying mutation rate P_mut and propagation time T_prop. The target vector is (ā₁, ā₂, ā₃) = (0.2, 0.3, 0.5) for all runs.

Fig. 8. Ecosystems relax toward the non-selected condition when directed selection is removed. Relaxation depends on mutation rate, but not on propagation time. Difference in mean F between high and low selected lines (±1 SE) plotted against time for different mutation rates and propagation time. The target vector is (ā₁, ā₂, ā₃) = (0.2, 0.3, 0.5) for all runs.