Emergent genetic oscillations in a synthetic microbial consortium

See allHide authors and affiliations

Science  28 Aug 2015:
Vol. 349, Issue 6251, pp. 986-989
DOI: 10.1126/science.aaa3794

Engineering cell population behavior

Attaining the full promise of synthetic biology will require designing population-level behaviors of multiple interacting cell types. As a start, Chen et al. engineered two strains of the bacterium Escherichia coli to produce signaling molecules that regulate transcription in the complementary strain (see the Perspective by Teague and Weiss). The signaling circuit was successfully designed to produce feedback loops that produce synchronous oscillations in transcription between the two strains. A mathematical model helped determine how to modulate the oscillations and control their robustness to perturbations.

Science, this issue p. 986; see also p. 924


A challenge of synthetic biology is the creation of cooperative microbial systems that exhibit population-level behaviors. Such systems use cellular signaling mechanisms to regulate gene expression across multiple cell types. We describe the construction of a synthetic microbial consortium consisting of two distinct cell types—an “activator” strain and a “repressor” strain. These strains produced two orthogonal cell-signaling molecules that regulate gene expression within a synthetic circuit spanning both strains. The two strains generated emergent, population-level oscillations only when cultured together. Certain network topologies of the two-strain circuit were better at maintaining robust oscillations than others. The ability to program population-level dynamics through the genetic engineering of multiple cooperative strains points the way toward engineering complex synthetic tissues and organs with multiple cell types.

Most synthetic gene circuits have been constructed to operate within single, isogenic cellular populations (14). However, synthetic microbial consortia could provide a means of engineering population-level phenotypes that are difficult to obtain with single strains (5). Indeed, several synthetic systems have been constructed to exhibit population-level phenotypes (69), including synthetic predator-prey systems (10), multicellular computers (11), and spatio-temporal pattern generators (12, 13). We constructed two genetically distinct populations of Escherichia coli to create a bacterial consortium that exhibits robust, synchronized transcriptional oscillations that are absent if either strain is grown in isolation. Specifically, we used two different bacterial quorum-sensing systems to construct an “activator” strain and a “repressor” strain that respectively increase and decrease gene expression in both strains. When cultured together in a microfluidic device, the two strains form coupled positive and negative feedback loops at the population level, akin to the circuit topology (i.e., how regulatory components within a circuit regulate each other) of a synthetic dual-feedback oscillator that operates within a single strain (14, 15). We used a combination of mathematical modeling and targeted genetic perturbations to better understand the roles of circuit topology and regulatory promoter strengths in generating and maintaining oscillations. The dual-feedback topology was robust to changes in promoter strengths and fluctuations in the population ratio of the two strains.

The two synthetic strains in our system were constructed to enzymatically produce and transcriptionally respond to intercellular signaling molecules (Fig. 1A). The activator strain produces C4–homoserine lactone (C4-HSL) (16), a signaling molecule that increases transcription of target genes within the synthetic circuits of both strains. The repressor strain produces 3-OHC14-HSL (17), which decreases transcription in both strains through a synthetic transcriptional inverter (18, 19) mediated by the repressor LacI. These two signaling mechanisms jointly create coupled positive and negative feedback loops at the population level when the two strains are grown together (Fig. 1B). Additionally, each strain, when active, produces the enzyme AiiA, which degrades both signaling molecules, resulting in another layer of negative feedback.

Fig. 1 The synthetic microbial consortium oscillator.

(A) Circuit diagrams of the activator (top) and repressor (bottom) strains. In the activator strain, transcription of rhlI and cfp are regulated by separate copies of the hybrid promoter, Prhl/lac, which is up-regulated by C4-HSL and down-regulated by LacI. In the repressor strain, cinI is driven by the hybrid promoter Prhl/lac and yfp is regulated by the hybrid promoter Pcin/lac, which is up-regulated by 3-OHC14-HSL and down-regulated by LacI. Both strains contain constitutively expressed copies of cinR and rhlR, which encode transcription factors that respond to the HSLs to regulate their respective promoters, and aiiA and lacI driven by 3-OHC14-HSL–responsive promoters. (B) Global topology of the dual-feedback consortium oscillator. The activator strain up-regulates genes in both strains. The repressor strain down-regulates genes in both strains. AiiA down-regulates signaling (dashed lines, omitted in Figs. 2 and 3 and figs. S2 and S7 for simplicity). (C) Representative time series of activator (blue) and repressor fluorescence (yellow), and activator population fraction (black, ratio of the area of activator cells to the area of the entire population of cells, as measured in pixels) for the consortium depicted in (A). Relative fluorescence values are the population average relative to the maximum after background subtraction. (D) Five images of the consortium from time points indicated by red dots in (C).

To observe the dynamics of the synthetic consortium, we used a custom-designed microfluidic device in conjunction with time-lapse fluorescence microscopy to observe the two strains as they grew together in a small chamber in which the diffusion time of the HSLs was small (see supplementary materials) (20). Each strain contained a gene encoding a spectrally distinct fluorescent reporter (cfp, cyan fluorescent protein, in the activator; yfp, yellow fluorescent protein, in the repressor), driven by promoters that respond to both positive and negative signals in the network (Fig. 1A). After an initial transient time, synchronous, in-phase oscillations emerged in the fluorescent reporters of both strains (Fig. 1, C and D). Neither strain oscillated when cultured in isolation (fig. S1). Oscillations had a period of ~2 hours and persisted throughout the experiments (usually more than 14 hours).

The circuit topology of our synthetic consortium consisted of linked positive and negative feedback loops, similar to the topologies of many naturally occurring biological oscillators (21, 22). However, a single negative feedback loop is sufficient to generate rhythms (14, 23). Therefore, we tested whether the additional feedback loops in our system were necessary for oscillations. We eliminated either the extra positive or extra negative feedback loop, or both simultaneously. To eliminate the extra positive feedback loop, we created a variant activator strain in which the hybrid promoter Prhl/lac-s was replaced with Plac. We eliminated the additional negative feedback loop by constructing a variant repressor strain without lacI. By combining one of the two variant activator strains with one of the two variant repressor strains, we generated consortia with four different topologies (fig. S2): P2N2 (containing both additional feedback loops), P2N1 (lacking the additional negative feedback loop), P1N2 (lacking the additional positive feedback loop), and P1N1 (lacking both additional feedback loops). Each topology still contained the negative feedback loop mediated by AiiA.

Each of the topologies described above generated rhythms if and only if both strains were present, confirming that additional feedback loops are not required for oscillations. Further, the oscillations were robust to fluctuations in the population ratio of the two strains in the microfluidic trap (Fig. 2, A to D). In contrast to the effect in a single-strain oscillator (14), removing the additional positive feedback loop lengthened the period in P1N2 and P1N1 (Fig. 2E). Furthermore, P1N2 and P1N1 generated double-peaked activator oscillations, in contrast to the single-peaked oscillations generated by P2N2 and P2N1.

Fig. 2 Dynamics of consortia with various topologies.

(A to D) P2N2 (A) and P2N1 (B) generate rhythms with shorter periods than P1N2 (C) and P1N1 (D). The CFP rhythms in the activator strain of P1N2 (C) and P1N1 (D) exhibited double peaks, in contrast to P2N2 (A) and P2N1 (B). Here, each line style is the same as in Fig. 1C. (E) Experimentally measured periods for each topology. The periods of P1N2 (P < 10−11) and P1N1 (P < 10−6) were not the same as that of P2N2 (t test with Bonferroni correction). Error bars are mean ± SD; from left to right, the number of measured periods was N = 19, 21, 15, 14. (F) The mean period (±SEM) of simulated oscillations with parameter sets leading to either single or double peaks in CFP oscillations among the 1506 parameter sets described in the main text. The periods in P1N2 and P1N1 were longer when CFP oscillations exhibited double peaks.

To understand the behaviors of the different topologies, we developed a mathematical model (24, 25) to simulate the intra- and extracellular dynamics of the key proteins and molecules (see supplementary materials and table S1). In our model, most of the parameters (32 of 40) were either obtained from the literature or by measurement of promoter activity under various conditions (fig. S3 and table S1). The unknown parameters were randomly and independently sampled from uniform distributions covering biologically realistic ranges (see supplementary materials). We obtained 10,000 parameter sets that led to oscillations in four different versions of our model, corresponding to the different network topologies. Of these 10,000 parameter sets, 1506 resulted in rhythms with approximately correct periods (100 to 250 min) for all topologies (fig. S4). More than 40% of these sets gave rise to double-peaked oscillations in the activator strain for both P1N2 and P1N1, but not P2N2 and P2N1, matching experimental observations (Fig. 2, C and D, and fig. S5, A to D). In simulations, the periods of P1N2 and P1N1 were longer when activator oscillations displayed double peaks (Fig. 2F). This indicates that the double peaks and period lengthening of P1N2 and P1N1 observed in experiments (Fig. 2, C to F) are related. Specifically, the model suggests that both the double peaks and period lengthening of P1N2 and P1N1 are caused by competition between RhlI and CFP for ClpXP-mediated proteolysis in the activator strain (26, 27).

We investigated how the additional feedback loops affect the robustness of oscillations, as theoretically additional feedback can increase the robustness of genetic oscillators (21, 22). We were interested in two types of perturbations: (i) variations in the population ratio of the two strains, and (ii) different promoter strengths within the circuit. Perturbations in the population ratio of the two strains arose naturally from variations in growth of the two strains within the microfluidic device. To perturb the promoter strengths within the circuits, we altered the Prhl/lac promoters used to drive cinI in the repressor strain and rhlI in the activator strain to have different expression strengths (see supplementary materials). The original activator strain contained a strong promoter, Prhl/lac-s, and the repressor strain a weaker promoter, Prhl/lac-w. Gene expression from the strong promoter was approximately 15 times that of the weak promoter. We also created a medium promoter, Prhl/lac-m, which was approximately 10 times as strong as the weak promoter (fig. S3).

To examine how changes to the promoter strengths and fluctuations in the population ratio affected oscillations in the mathematical model, we used the 1506 parameter sets described above and systematically altered the parameters governing promoter activities and the population ratio (fig. S3 and table S1). We then calculated the percentage of parameter sets that still led to oscillations in the four different topologies (fig. S6). The mathematical model predicted that P2N2 and P1N2 show similar robustness (i.e., the percentage of parameter sets leading to oscillations after perturbation was similar) and that both are more robust than P2N1 and P1N1 (Fig. 3, A and F, and fig. S6). Hence, the model predicts that the addition of a negative feedback loop, but not a positive feedback loop, has a pivotal role in generating robust rhythms because it tightly regulates repressor concentration (fig. S6) (22, 28). Thus, for different promoter-strength combinations, the P2N2 oscillator is expected to be more robust to differences in population ratio than P2N1 (Fig. 3, A and F). To test this prediction, we constructed P2N2 and P2N1 using the medium promoter, Prhl/lac-m, for both strains. P2N2 oscillated over a wide range of population ratios, whereas P2N1 occasionally stopped oscillating when the activator population fraction was low (Fig. 3).

Fig. 3 Additional negative feedback increases robustness.

(A) Percentage of the 1506 parameter sets (fig. S4) that led to oscillations in the model as a function of percentage of activator strain in the P2N2 consortium. Shown are the results for three different combinations of promoter strengths. S/W: Prhl/lac-s and Prhl/lac-w promoters in activator and repressor strains, respectively. S/M: Prhl/lac-s promoter in activator and Prhl/lac-m promoter in repressor. M/M: Prhl/lac-m promoters in both strains. (B to E) Example trajectories for the P2N2 topology with Prhl/lac-m promoters in both activator and repressor strains with different activator population fractions. (F) Percentage of the 1506 parameter sets that led to oscillations in the model as a function of percentage of activator strain in the P2N1 consortium. Symbols are the same as in (A). Results for P1N2 and P1N1 are given in fig. S6. (G to J) Example trajectories for the P2N1 topology with the M/M promoter configuration showing occasional loss of oscillations.

We also explored other configurations of promoter strengths within the four topologies. When the promoter driving cinI in the repressor strain was changed from Prhl/lac-w to Prhl/lac-m (and Prhl/lac-s was kept in the activator), all topologies still generated oscillations (fig. S7, A to D). However, P1N2 and P1N1 showed rhythms with a much shorter period than those of P2N2 and P2N1 (fig. S7, C and D). Furthermore, oscillations in the repressor strain were robust even though activator strain oscillations in P1N2 and P1N1 were low and unstable.

To understand why the P1N2 and P1N1 topologies exhibited strong and short-period repressor oscillations even in the absence of strong activator oscillations, we again turned to the mathematical model. Our simulations matched experimental data when Prhl/lac-m was used in the repressor strain in the P1N2 and P1N1 topologies (fig. S8, A and B). The model predicted that when Prhl/lac-m is used in P1N2 and P1N1, the mechanism responsible for generating oscillations is an intracellular negative feedback loop mediated by AiiA in the repressor strain and not the intercellular transcriptional negative feedback loop between the two strains (fig. S8). Essentially, the feedback loop mediated by AiiA in the repressor strain has a shorter delay time than the transcriptional loop between the strains, and hence the period becomes shorter (29).

Our results show that engineering dynamic population-level phenotypes in synthetic microbial consortia is possible with multiple intercellular signaling mechanisms. Because the population ratio within a consortium can fluctuate, it is important to engineer synthetic circuits that are robust to such perturbations. Overall, our synthetic microbial consortia provide a platform for testing the relation between population-level dynamics and genetic-level regulation.

Supplementary Materials

Materials and Methods

Figs. S1 to S10

Table S1

Movie S1

References (3052)

References and Notes

  1. Acknowledgments: This work was funded by the National Institutes of Health, through the joint NSF–National Institute of General Medical Sciences Mathematical Biology Program grant R01GM104974 (M.R.B. and K.J.), the Robert A. Welch Foundation grant C-1729 (M.R.B.), the Hamill Foundation (M.R.B.), NSF grant DMS-0931642 to the Mathematical Biosciences Institute (J.K.K.), and the China Scholarship Council (Y.C.). M.R.B., Y.C., K.J., and J.K.K. conceived and designed the study. Y.C. performed the experiments and analyzed the data. A.J.H. designed and fabricated the microfluidic devices. J.K.K. performed the computational modeling and analyzed simulations. M.R.B. supervised the project. All authors wrote the manuscript. The mathematical model and experimental data are archived in the BioModels Database at
View Abstract

Navigate This Article