Genetically engineered control of phenotypic structure in microbial colonies

Abstract

Rapid advances in cellular engineering^1,2 have positioned synthetic biology to address therapeutic^3,4 and industrial⁵ problems, but a substantial obstacle is the myriad of unanticipated cellular responses in heterogeneous real-world environments such as the gut^6,7, solid tumours^8,9, bioreactors¹⁰ or soil¹¹. Complex interactions between the environment and cells often arise through non-uniform nutrient availability, which generates bidirectional coupling as cells both adjust to and modify their local environment through phenotypic differentiation^12,13. Although synthetic spatial gene expression patterns^14-17 have been explored under homogeneous conditions, the mutual interaction of gene circuits, growth phenotype and the environment remains a challenge. Here, we design gene circuits that sense and control phenotypic structure in microcolonies containing both growing and dormant bacteria. We implement structure modulation by coupling different downstream modules to a tunable sensor that leverages Escherichia coli's stress response and is activated on growth arrest. One is an actuator module that slows growth and thereby alters nutrient gradients. Environmental feedback in this circuit generates robust cycling between growth and dormancy in the interior of the colony, as predicted by a spatiotemporal computational model. We also use the sensor to drive an inducible gating module for selective gene expression in non-dividing cells, which allows us to radically alter population structure by eliminating the dormant phenotype with a 'stress-gated lysis circuit'. Our results establish a strategy to leverage and control microbial colony structure for synthetic biology applications in complex environments.

Extended Data Fig. 1 |. Establishment of phenotypic heterogeneity.

a, High-resolution image of full microfluidic trap after establishment of phenotypic heterogeneity. Blue color indicates IIDs (see Methods). Scale bar, 20μm. b, Initial growth in the microfluidic trap. Time-lapse images as in Fig. 1b with zoomed-in region around the eventual growth boundary (see also Supplementary Video 1). Blue color indicates IIDs (see Methods). Scale bars in top left and bottom left image, 15μm.

Extended Data Fig. 2 |. Growth patterns in different media compositions.

a, Kymographs indicating inter-image differences (IIDs, see Methods) as in Fig. 1c during the establishment of a steady state growth pattern in 170μm deep traps for growth media containing different amounts of glucose (the standard concentration is 0.2% w/v). For lower concentrations, growth extends further into the back of the trap with a smoother transition between regions of growth and no growth. Kymographs on the right of the black line show consistent behavior in the smaller (140μm deep) traps (the dashed line indicates the end of the trap). b, Distance from the mouth of the microfluidic trap at which the inter-image difference (IID) shown in a drops below detectable levels. For this purpose, the IID profile was averaged over 4 h of steady-state growth. The threshold for growth detection was chosen as the lowest possible level that safely detected the no-growth region in standard growth medium.

Extended Data Fig. 3 |. Transition to dormancy and protein expression in dormant cells.

a, Stress sensor activation at different background osmolarities. Kymographs showing growth (IIDs, blue) and GFP fluorescence (green) for lower NaCl concentrations. Compared to Fig. 1f, the fluorescence signal has been rescaled to visualize even weak activation of the pOsmy stress sensor. While the sensor is transiently activated upon growth arrest even in the absence of basal osmotic stress (0 mM NaCl), higher NaCl also elicits a measurable response in dividing cells close to the growth boundary. b, Position of the growth interface during sensor activation, analyzed as in Extended Data Fig. 2b, showing no detectable growth modulation by NaCl itself or the activation of the sensor module. See Extended Data Fig. 6c for additional controls. c, Expression of untagged RFP from the pLuxI promoter in growth-arrested cells. As in Fig. 1g, h, the plot shows average fluorescence measured in the growth-arrested back of the microfluidic trap. The blue shaded area marks the window of induction with AHL.

Extended Data Fig. 4 |. Actuator circuit in batch culture and microfluidics.

a, IPTG was added when the culture of E. coli cells equipped with the actuator circuit reached an OD600 of 0.13. Growth is halted through expression of GRP-ssrA from pLlacO-1 at a higher OD compared to Fig. 2a, consistent with the later addition of IPTG. b, Time series of growth (calculated as inter-image difference, IID, see Methods) and fluorescence in our microfluidic environment, averaged over the area in the trap indicated in c. Blue shaded regions mark the induction windows for all 8 IPTG concentrations tested. Cumulative effects can be observed when pulses of IPTG are given in short succession. c, Time-lapse images showing growth (IIDs, blue) and fluorescence (green) for the experiment shown in panel b during the last pulse of 300 μM IPTG (see Supplementary Video 3). The pulse begins at t = 17.6 h. d, Kymographs of growth (IIDs, blue) and fluorescence (green) for cell populations carrying the actuator circuit, which is induced continuously starting around t = 4 h. With increasing GRP-ssrA expression, the area of active growth extends further and further into the trap. e, Same data as in d, omitting fluorescence for better visualization of the growth pattern.

Extended Data Fig. 5 |. Numerical simulations of the sensor-actuator circuit (see Supplementary Discussion 2 for modeling details).

a, Activation of the stress sensor upon growth arrest in numerical simulations of a well-mixed batch culture of cells. The promoter strength of the sensor is modeled to increase with the NaCl concentration in the media (cf. Figure 1e). b, Actuator and population dynamics in batch culture upon IPTG induction. The arrow marks the start of induction. The production rate of the actuator protein is proportional to the IPTG induction level, halting growth earlier for higher IPTG (cf. Figure 2a). c, Spatiotemporal simulations of actuator protein induction and population dynamics in a microfluidic trap. 1 h and 5 h pulses of IPTG (amplitudes 0.5) lead to reversible growth resumption in the growth arrested region of the simulated microcolony (cf. Figure 2b and Extended Data Fig. 4). d, Numerical simulations of the full spatiotemporal model of the sensor-actuator circuit coupled with population and nutrient dynamics. Oscillations near the growth interface are observed for sufficiently high induction (NaCl) levels of the sensor promoter. e, In the oscillatory layer of the microcolony (compare d), induction of the actuator protein is followed by growth resumption, which is followed by growth arrest and induction of the actuator protein from the stress sensor, restarting the cycle (cf. Figure 2g).

Extended Data Fig. 6 |. Induction of oscillations in the sensor-actuator circuit.

a, Kymographs of growth (IIDs, blue) and fluorescence (green) for populations carrying the sensor-actuator circuit (Fig. 2c). Basal osmotic stress (NaCl) activates the sensor close to the interface (Fig. 1f) and initiates negative feedback leading to oscillations. b, Period and amplitude of fluorescence oscillations. Period is only shown for NaCl concentrations yielding finite oscillation amplitudes. Data points for each concentration correspond to two traps imaged at high resolution (see Methods). c, NaCl-dependent growth delay (measured as the time to reach OD 0.1 relative to 0 mM NaCl) in the sensor-actuator circuit (pOsmy-GRP-LAA) and multiple control strains: MG1655Z1 (no construct), pOsmy-GFP-LAA (the sensor component), pOsmy(LAA)-GFP-LAA (the sensor component with an additional ssrA-LAA degradation tag on the N-terminal Osmy part of pOsmy) and pOsmy-GRP (the sensor-actuator circuit without the ssrA-LAA degradation tag on GRP). The presence of growth delays exclusively in the GRP-containing constructs (exacerbated when the ssrA degradation tag is not present and GRP levels are increased) shows that the N-terminal portion of the Osmy protein (no matter whether expressed at high or low levels) has no growth-modulating effect for the media, NaCl concentrations and resulting expression levels relevant for this study (cf. Extended Data Fig. 3b). Note that these additional control constructs are not included in Supplementary Figure 4.

Extended Data Fig. 7 |. Unsuccessful elimination of growth-arrested cells with the ungated lysis circuit.

a, Time-lapse images of unsuccessful phenotype elimination with the ungated lysis circuit (LuxR-pLuxR-pLuxI-E-ssrA) for induction with 490 nM AHL (compare b,c). Lysis starts in the growing part of the population, causing growth-arrested cells to receive fresh nutrients and resume growth, while simultaneously also lysing. Cells regrowing despite circuit activation form disorganized colonies and display no visible phenotype pruning. Time stamps denote time after beginning of induction. Blue color indicates inter-image differences (IIDs, see Methods). See also Supplementary Video 5. Scale bar, 50μm. b, Growth and lysis dynamics (calculated via inter-image differences, IIDs, see Methods) in the region of the trap usually occupied by growth-arrested cells for different AHL induction levels of the ungated lysis circuit. After the initial lysis, no sharp lysis events are observed. Instead, uncoordinated phases of spatially inhomogeneous lysis, regrowth and growth arrest without lysis lead to fluctuating IID that settles at a low level (cf. end states in c). c, Images taken after 18 h of AHL induction with the indicated concentration, when traps have settled into a steady state. Blue color indicates inter-image differences (IIDs, see Methods). While cells in all traps initially lyse (b), no distinct phenotype elimination is observed and colony structure is disorganized (see Supplementary Discussion 1).

Extended Data Fig. 8 |. Stress-gated expression of rFP.

a, Example traces of OD and RFP fluorescence in plate reader experiments with the gating circuit (Fig. 3a) with RFP-ssrA-AAV as the target gene (left), compared to ungated expression from a standard bidirectional LuxR-pLuxR-pLuxI cassette with constitutive LuxR (right). Plotted induction levels were chosen to show similar levels of expression. b, Peak fluorescence in plate reader experiments of the same RFP-ssrA-AAV gating circuit (Fig. 3a). The stress-gated circuit exhibits increased sensitivity at low concentrations, which enabled us to determine gating fidelity (Fig. 3c) for much lower concentrations for this circuit compared to constitutive LuxR. Plot shows mean ± s.d., n = 4 wells. c, We tested the same circuits in microfluidics and measured the fluorescence in the growth-arrested region in the back of the trap in response to a 10 h induction with different AHL concentrations. t_20% is the time until 20% of peak fluorescence (as shown in Fig. 3d) are reached. Plot shows mean ± s.d., n = 7 microfluidic traps. d, Gating fidelity for stress-gated and ungated RFP-ssrA-AAV from plate reader experiments as in Fig. 3c, but measured in LB. e, Gating fidelity for stress-gated and ungated RFP-ssrA-AAV from plate reader experiments as in Fig. 3c, but measured in LB supplemented with 0.2% glucose. Symbols in d, e represent means of 2 wells each.

Extended Data Fig. 9 |. Numerical simulations of the gating circuit (see Supplementary Discussion 2 for modeling details).

a, Time traces for OD, nutrients and RFP as the target protein in the ungated circuit with constitutive production of LuxR. The arrow marks the beginning of AHL induction. b, Same time traces for the gated circuit with LuxR expressed from pOsmy, showing reduced expression during exponential phase. c, Gating fidelity (defined the same way as for Fig. 3c and Extended Data Fig. 8d,e) for the gated and ungated circuits, showing a ca. 10-fold stronger preference for stationary-phase expression in the gating circuit compared the ungated circuit. d, Time traces of OD and cellular lysis protein concentration in the ungated circuit for different AHL levels. Exponential growth rate is impacted at the same time as stationary-phase lysis becomes effective. e, Same time traces for the gated circuit for different AHL levels. The reduction of LuxR during exponential phase and its increase in stationary phase leads to efficient lysis without impacting exponential growth.

Extended Data Fig. 10 |. Behavior of the SGLc at low AHL concentrations near onset.

a, Images taken after 24 h of AHL induction with the indicated concentration, when traps have settled into a steady state. Blue color indicates inter-image differences (IIDs, see Methods). Lower concentrations than those shown here do not settle into steady states (see c) or show imperfect pruning. b, Growth and lysis dynamics (calculated via inter-image differences, IIDs, see Methods) in the region of the trap usually occupied by growth-arrested cells. No lysis occurred below AHL concentration of 2 nM. For higher concentrations and lysis onset times across parameters, see Fig. 4d–f. c, Time-lapse images showing oscillatory behavior of the SGLC close to onset of lysis (3 nM AHL, cf. panel b and Fig. 4d). Long delay between LuxR priming by the gating circuit and accumulation of sufficient lysis protein (Fig. 4e) causes sequential fill-up and lysis (see Supplementary Video 7). Time stamps denote time after beginning of induction.

Fig. 1 |. Microfluidic platform for observing E. coli growth patterns in spatially heterogeneous microcolonies.

a, Schematic illustrating a trap in our microfluidic device. The flow rate in the nutrient supply channel is ≈ 100 μm s⁻¹. b, Time-lapse microscopic images of a single microfluidic trap showing development of phenotypic heterogeneity shortly after cell loading (see Supplementary Video 1). Blue colour indicates regions of actively growing E. coli (based on IIDs; see Methods). Scale bars (top left and bottom right images), 15 μm. c, Kymograph of IIDs (blue) averaged along the horizontal dimension of the trap. d, The stress sensor consists of the σ^S-responsive pOsmy promoter driving ssrA-tagged GFP to visualize sensor activation (e,f). e, Time courses of optical density (OD, black) and fluorescence (fluo., green) in plate reader experiments display transient sensor activation that depends on the level of basal osmotic stress (NaCl). f, Time-lapse images (for 140 mM NaCl; see Supplementary Video 2) and kymographs of microfluidic experiments showing NaCl-dependent transient stress sensor activation (green) on initial growth arrest and later persistent activation in dividing cells close to the no-growth boundary. Blue colour indicates IIDs (see Methods). g, Protein expression in growth-arrested E. coli measured by induction of IPTG-inducible pLlacO-1 driving RFP. In the time-lapse images, red indicates RFP fluorescence and blue marks the regions of growth. Plotted time traces represent average RFP fluorescence in the growth-arrested region of the trap for different IPTG concentrations present during the induction window (blue shaded area). Expression in growth-arrested cells was also observed with pLuxI and AHL (see Extended Data Fig. 3c). a.u., arbitrary units. h, The same for pLuxI-driven RFP which is ssrA tagged and targeted for enzymatic degradation by ClpXP.

Fig. 2 |. Actuator growth modulation and spatiotemporal feedback.

a, Schematic of the actuator circuit expressing GRP-ssrA from the pLlacO-1 promoter. The circuit was induced with IPTG at an OD₆₀₀ of 0.03 (black arrow; see Extended Data Fig. 4a for induction at OD₆₀₀ 0.13). b, Kymographs showing growth (IIDs, blue) and GFP fluorescence (green) of E. coli cells with the actuator circuit in microfluidics. Pulses of IPTG shown by white marks. (See Extended Data Fig. 4 for time-lapse microscopic images and behaviour across several tested IPTG concentrations and for a continuous induction of the actuator circuit.) Note that even though GRP-ssrA is expressed from the same IPTG-inducible promoter as RFP (see Fig. 1g), its relative protein levels across the different phenotypes are not comparable to those of RFP, because the balance of protein synthesis, degradation and dilution is altered by the growth-modulating effect of GRP and the presence of the ssrA-degradation tag. c, Sensor-actuator circuit. The pOsmy stress sensor drives the GRP-ssrA actuator protein, giving rise to spatiotemporal negative feedback mediated by growth suppression that increases available nutrient concentration and relieves metabolic stress. d, Simulated kymograph from the spatiotemporal model of microcolony dynamics with the sensor-actuator circuit. After sensor and actuator components were separately tuned to match batch and microfluidic experiments with isolated circuits (a,b and Fig. 1e), the full model predicts periodic oscillations of actuator protein concentration (green) and growth (blue) near the no-growth boundary (see Extended Data Fig. 5 and Supplementary Discussion 2). e, Period and amplitude of actuator protein concentration oscillations in the spatiotemporal simulations. f, Time-lapse images of cells carrying the sensor-actuator circuit showing oscillations of GRP-ssrA fluorescence (green) at the growth boundary in medium supplemented with 71 mM NaCl needed for feedback activation (compare Extended Data Fig. 6 and see Supplementary Video 4). g, Local oscillations in GRP-ssrA fluorescence and growth (calculated via IIDs, see Methods) averaged over the area marked in f. h, Kymographs showing periodic oscillations of GRP-ssrA fluorescence (green) and growth (IIDs, blue) in a microfluidic colony of cells with the sensor-actuator circuit and expressing untagged RFP (red) from an AHL-inducible pLuxI promoter.

Fig. 3 |. Synthetic stress gate primes cells for phenotype-specific expression.

a, Stress-gated expression of a target gene. The pOsmy stress sensor drives expression of the regulator protein LuxR required to activate the pLuxI promoter on binding of external AHL. b, Constitutive LuxR expression (green) from the native pLuxR promoter in the standard pLuxI induction cassette (compare Supplementary Fig. 4) leads to sensitivity of both growing and growth-arrested phenotypes to pLuxI induction by external AHL. The gating module (purple) reduces LuxR levels during normal growth and primes cells with LuxR on transition into the non-growing phenotype, increasing pLuxI promoter sensitivity and thus target gene expression on induction with external AHL in a phenotype-specific manner. c, Stress-gated RFP circuit (target gene = RFP-ssrA-AAV) with constant AHL induction in the plate reader compared with an ungated circuit. Gating fidelity is calculated by dividing the peak RFP fluorescence (compare Extended Data Fig. 8b) by the RFP fluorescence at OD 0.7. Data shown are for peak fluorescence values >5% of peak fluorescence at maximum induction. d, The same circuits cultured in microfluidics and induced for 10 h at different AHL levels. Average RFP fluorescence in the growth-arrested region was determined as in Fig. 1h. Plot shows mean ± s.d. (n = 7 microfluidic traps). e, Batch growth of cells with the stress-gated lysis circuit (a with target gene = E-ssrA-AAV) and the ungated circuit at various inducer concentrations. f, Growth rates for the stress-gated and ungated lysis circuit, and corresponding controls expressing RFP-ssrA-AAV rather than E-ssrA-AAV. Plot shows mean ± s.d. (n = 4 wells). g, Lysis efficiency (percentage OD reduction within 1 h after reaching peak OD) for the two lysis circuits. Horizontal bars at the top of the plot indicate the inducer range where the growth rate reduction in exponential phase is <5%. h, Peak OD reached before lysis versus lysis efficiency for the gated and the ungated lysis circuit. The s.d. of peak ODs (vertical axis) is smaller than plot symbol sizes. Plots in c,g,h show mean ± s.d. (n = 4 wells).

Fig. 4 |. A colony-level phenotype filter.

a, Time-lapse images of a single microfluidic trap with cells carrying the SGLC (see Supplementary Video 6). Dormant cells are eliminated from the population whereas dividing cells are unaffected by the SGLC. Blue regions based on IIDs (see Methods) indicate regions of active growth and lysis. Time stamps indicate time after beginning of 17 nM AHL induction. Scale bar, 50 μm. b, Magnification of the interface region for the same time points as in a. Note the similarity of the growth pattern (including the growth boundary) after reaching steady state in a and b to the undisturbed pattern (see Fig. 1b and Extended Data Fig. 1). Dividing cells are found in the same region of the trap, whereas dormant cells have been eliminated by the SGLC. Scale bar, 15 μm. c, Magnification of a region in the rear of the trap, showing a small population of cells growing after the first lysis event, which eventually lyses when steady-state growth in the front part of the trap is established. Scale bar, 15 μm. d, Growth and lysis dynamics across AHL inducer levels for the SGLC. The curves were obtained by averaging IIDs (see Methods) in the region of the trap usually occupied by growth-arrested cells. Spikes correspond to sudden lysis events. Blue-shaded regions denote AHL induction windows. e, Time from induction to the onset of lysis was determined by detecting the first peak in the IID traces as in d. Experimental data were pooled from three different experiments for 0–3.5 nM, 0–20 nM and 0–140 nM AHL. Each data point corresponds to a single trap on the microfluidic chip. f, Effective stress gating by the SGLC enables specific elimination of dormant cells even for substantially higher AHL concentrations (compared with the minimum concentration necessary to induce killing; see Extended Data Fig. 10a), resulting in growth patterns very similar to those in a. Shown are time-lapse images of the steady state taken after 18 h of AHL induction with the indicated concentration. Growing cells are found in the same region as for the 0 nM control, whereas dormant cells are eliminated for all non-zero AHL concentrations (see Supplementary Video 8 for details).