|Home | About | Journals | Submit | Contact Us | Français|
Users may view, print, copy, download and text and data- mine the content in such documents, for the purposes of academic research, subject always to the full Conditions of use: http://www.nature.com/authors/editorial_policies/license.html#terms
The engineering of genetic circuits with predictive functionality in living cells represents a defining focus of the expanding field of synthetic biology. This focus was elegantly set in motion a decade ago with the design and construction of a genetic toggle switch and an oscillator, with subsequent highlights that have included circuits capable of pattern generation, noise shaping, edge detection, and event counting. Here, we describe an engineered gene network with global intercellular coupling that is capable of generating synchronized oscillations in a growing population of cells. Using microfluidic devices tailored for cellular populations at differing length scales, we investigate the collective synchronization properties along with spatiotemporal waves occurring on millimeter scales. We use computational modeling to quantitatively describe the observed dependence of the period and amplitude of the bulk oscillations on the flow rate. The synchronized genetic clock sets the stage for the use of microbes in the creation of a macroscopic biosensor with an oscillatory output. In addition, it provides a specific model system for the generation of a mechanistic description of emergent coordinated behavior at the colony level.
Synchronized clocks are of fundamental importance in the coordination of rhythmic behavior among individual elements in a community or a large complex system. In physics and engineering,the Huygens paradigm of coupled pendulum clocks 1–3 has permeated diverse areas from the development of arrays of lasers 4 and superconducting junctions 5 to GPS 6 and distributed sensor networks 7. In biology, a vast range of intercellular coupling mechanisms lead to synchronized oscillators which govern fundamental physiological processes such as somitogenesis, cardiac function, respiration, insulin secretion, and circadian rhythms 8–15. Typically, synchronization helps stabilize a desired behavior arising from a network of intrinsically noisy and unreliable elements. Sometimes, however, the synchronization of oscillations can lead to a severe malfunction of a biological system, as in epileptic seizures 16.
There is considerable interest in the use of synthetic biology to recreate complex cellular behavior from the underlying biochemical reactions that govern gene regulation and signaling. Synthetic biology can be broadly parsed into efforts aimed at the large-scale synthesis of DNA and the forward engineering of genetic circuits from known biological components. In the area of DNA synthesis, pathways have been perturbed and replaced 17 in an effort to understand the network motifs and transcriptional regulatory mechanisms that control cellular processes and elicit phenotypic responses 18. On a larger scale, progress has been made towards the creation of entire genomes, providing new insights into what constitutes the minimal set of genes required for microbial life 19.
The genetic circuits approach to synthetic biology involves the forward engineering of relatively small gene networks using computational modeling 20, 21. Here, the original toggle switch 22 and oscillator 23 have inspired the design and construction of circuits capable of controlling cellular population growth 24, generating specific patterns 25, triggering biofilm development 26, shaping intracellular noise 27, detecting edges in an image 28, and counting discrete cellular events 29. In the context of rhythmic behavior, there have been recent successes in the construction of intracellular oscillators that mimic naturally occurring clocks 30–33. In addition to their potential as biological sensors, these clock networks have led to insights regarding the functionality of circadian networks 34. A unifying theme for most of the genetic circuit studies is a particular focus on dynamical behavior. Thus the circuits are constructed and monitored in single cells, typically with fluorescent reporters, and new measurement technologies are often developed in parallel 35. In addition, since nonlinearities and stochasticity arise naturally from the underlying biochemistry, tools from the fields of nonlinear dynamics and statistical physics are extremely useful in both the generation of design specifications and for careful comparison between experiment and computational model.
The synchronized oscillator design (Fig. 1a) is based on elements of the quorum sensing machineries in Vibrio fisheri and Bacillus Thurigensis. We placed the luxI (from V. fischeri), aiiA (from B. Thurigensis) and yemGFP genes under the control of three identical copies of the luxI promoter. The LuxI synthase enzymatically produces an acyl-homoserine lactone (AHL), which is a small molecule that can diffuse across the cell membrane and mediates intercellular coupling. It binds intracellularly to the constitutively produced LuxR, and the LuxR-AHL complex is a transcriptional activator for the luxI promoter 36. AiiA negatively regulates the promoter by catalyzing the degradation of AHL37. This network architecture, whereby an activator activates its own protease or repressor, is similar to the motif used in other synthetic oscillator designs 30–32 and forms the core regulatory module for many circadian clock networks 13, 38, 39. In addition, theoretical work has shown how the introduction of an autoinducer in similar designs can potentially lead to synchronized oscillations over a population of cells 40, 41.
Most quorum sensing systems require a critical cell density for generation of coordinated behavior42. We modified the local cell density of the synchronized oscillator cells (denoted TDQS1) through the use of microfluidic devices 43 of differing geometries. The device used for monitoring the bulk oscillations consists of a main nutrient-delivery channel that feeds a rectangular trapping chamber (Fig. 1b). Once seeded, a monolayer of E. coli cells grow in the chamber and are eventually pushed into the channel where they then flow to the waste port. This device allows for a constant supply of nutrients or inducers and the maintenance of an exponentially growing colony of cells for more than four days. We found that chamber sizes of 100 × (80–100)µm2 were ideal for monitoring the intercellular oscillator, as they allowed for sufficient nutrient distribution and optimal cell and AHL densities. In the context of the design parameters, the flow rate can be modulated in order to change the local concentration of AHL. In addition, the device can be modified in order to permit the observation of spatial waves over longer length scales (see below).
After an initial transient period, the TDQS1 cells exhibit stable synchronized oscillations which are easily discernible at the colony level (Figs. 1c, 1d, and Supplementary Movies 1–2). The dynamics of the oscillations can be understood as follows. Since AHL is swept away by the fluid flow and is degraded by AiiA internally, a small colony of individual cells cannot produce enough inducer to activate expression from the luxI promoter. However, once the population reaches a critical density, there is a “burst” of transcription of the luxI promoters, resulting in increased levels of LuxI, AiiA, and GFP. As AiiA accumulates, it begins to degrade AHL, and after a sufficient time, the promoters return to their inactivated state. The production of AiiA is then attenuated, which permits another round of AHL accumulation and another burst of the promoters.
In order to determine how the effective AHL dissipation rate affects the period of the oscillations, we conducted a series of experiments at various channel flow rates. At high flow rate, the oscillations stabilize after an initial transient and exhibit a mean period of 90±6 minutes and mean amplitude of 54±6 GFP AU (Fig. 2a, Supplementary Movie 2). At low flow rate, we observed a period of 55±6 minutes and an amplitude of 30±9 GFP AU. Interestingly, the waveforms have differing shape, with the slower oscillator reaching a trough near zero after activation and the faster oscillator decaying to levels above the original baseline (Fig. 2b). We swept the flow rate from 180 to 296 µm/min and observed an increasing oscillatory period from 52–90 minutes (Fig. 2c). In addition, we found the amplitude to be proportional to the period of the oscillations (Fig. 2d), which is consistent with “degrade-and-fire” oscillations 44 observed in a previously reported intracellular oscillator 31.
In experiments conducted at low flow rate, we observed the spatial propagation of the fluorescence signal across the 100 µm chamber. In order to investigate these spatiotemporal dynamics in more detail, we redesigned the microfluidic chip with an extended 2 mm trapping chamber (Supplementary Information). Snapshots of a typical experimental run are presented in Fig. 3a (Supplementary Movies 3 and 4). A few isolated colonies begin to grow and subsequently merge into a large monolayer that fills the chamber (Fig. 3a: 66 minutes). At 100 minutes, there is a localized burst of fluorescence that propagates to the left and right in subsequent frames (Fig. 3a: 100–118 minutes). A second burst occurs near the original location and begins to propagate to the left and right as before.
To illustrate the spatiotemporal information contained in an entire 460-minute image sequence, we plot the fluorescence intensity as a function of time and distance along the chamber (Fig. 3b). Note the correspondence of this space-time plot to the images in Fig. 3a. During the first 100 minutes, there is no activity and the space-time plot is blue, indicating no fluorescence. Then at 100 minutes, there is an orange spot at around 1350 µm, corresponding to the burst in Fig. 3a. In the space-time plot, propagation of a wave to the left and right appears as an green-yellow concave line. The larger slope to the left of the burst-origin indicates that the leftward moving wave is traveling slower (~25µm/min) than the rightward wave (~35µm/min). Subsequent waves originating from a nearby location arise as additional orange-yellow intensity lines. These intensity lines indicate “annihilation events”, where a leftward moving and rightward moving waves collide and annihilate each other. While these events are striking in the movies (Supplementary Movies 3 and 4), they appear subtly in the space-time plot at locations where positive and negative slopes meet (300–400 µm in 2nd intensity line and on). As the traveling wave gets further from a burst location it breaks off into a packet (170 minutes) which travels leftward at 12.5 µm/min initially, and slows to 8.5 µm/min towards the end of the trap where the cell density is lower (between 118–200 minutes). The corresponding cell-density space-time plot shows that a higher density of cells is first reached at the center of the colony and is minimal towards the left-moving edge (Fig. 3c and Supplementary Movie 3). As a result, the critical cell and AHL densities for wave propagation are reached at different times and spatial locations.
We also investigated how the intercellular oscillator behaves in a three-dimensional colony growing in a 400×1000×4.0 µm3 microfluidic chamber (Figs. 3d, 3e, and Supplementary Movie 5). In this device, the colony grows radially over the course of 180 minutes without fluorescing until it reaches a size of approximately 100 µm. At this time, a large fluorescence burst originates from the center of the colony, with a bright band near the center (Fig. 3d: 228 minutes). During this first burst (273 minutes), the bright band shows that cells at an intermediate cell density have a larger amplitude and longer period than cells near the front or in the interior. As the colony expands an additional 50–100 µm in diameter, a second burst of fluorescence occurs at a similar intermediate cell density. Subsequent oscillations are seen as the cell growth front propagates, while weak oscillations arise and quickly die inside the colony.
In order to quantitatively describe the mechanisms driving bulk synchronization and wave propagation, we developed a computational model using delayed differential equations for protein and AHL concentrations (Supplementary Information). While conceptually the nature of oscillations is reminiscent of the degrade-and-fire oscillations observed in a dual delayed feedback circuit 31, 44, an important difference is the coupling among genetic clocks in different cells through extracellular AHL. The modeling of this coupling, and the related cell density dependence, allowed us to explain most of the non-trivial phenomenology of the spatiotemporal quorum clock dynamics.
A broad range of model parameters lead to oscillations (Figs. 4a–d), though there is a distinct absence of oscillations at small and large cell densities for low to medium flow values (Fig. 4c). The qualitative nature of the oscillations can be explained using Fig. 4a. Each period begins with the latent accumulation of both AiiA and LuxI, which after a delay burst rapidly to high values. That burst suppresses AHL and further production of AiiA and LuxI. Both of these proteins then decay enzymatically, after which the process repeats. As expected, the period of the oscillations is roughly proportional to the enzymatic protein decay time. The period grows with the external AHL flow rate (effective degradation) and the amplitude of the oscillations, in good agreement with the experiments (compare Fig. 4b with Figs. 3c and d).
We modeled the collective spatiotemporal dynamics of the clocks by generalizing the bulk model to include the coupling of individual oscillators through extracellular AHL. The model consists of a one-dimensional array of “cells”, each of which is described by the same set of delay-differential equations coupled to a common, spatially nonuniform field of extracellular AHL. The latter is described by a linear diffusion equation with sources and sinks due to AHL diffusion through the cell membrane and dilution. A small AHL perturbation in the middle of the array initiates waves of LuxI concentration (Fig. 4c), in excellent agreement with the experimental findings (compare Figs. 3b and 4c). The velocity of the front propagation depends on the external AHL diffusion coefficient D1 (Fig. 4d and Supplementary Information), and for experimentally relevant values of D1, the simulated front velocity is in good agreement with experimental data. In addition, cell density plays an important role in wave propagation. In order to model the evolution of the three-dimensional colony (Figs. 3c and 3d), we set the functional form of the cell density to be an expanding “Mexican hat”, as observed in the experiments. Oscillations are then suppressed by the high density of cells in the middle of the colony, and LuxI bursts only occur on the periphery of the growing colony of cells. This phenomenology is also in excellent agreement with our experimental findings (compare Figs. 4f and and3d3d).
On a fundamental level, the synchronized oscillations represent an emergent property of the colony that can be mechanistically explained in terms of the circuit design. Oscillations arise because the small molecule AHL plays a dual role, both enabling activation of the genes necessary for intracellular oscillations and mediating the coupling between cells. Since unbounded growth of the colony leads to an accumulation of AHL that ultimately quenches the bulk oscillations, we used open-flow microfluidic devices to allow for the flow of AHL away from the colony. At low cell densities, oscillations do not occur because intracellular gene activation is decreased as AHL diffuses across the cell membrane and out of the chamber. At intermediate cell densities (i.e. a full chamber), the increased production of AHL in each cell acts to mitigate the outward flow such that activation of the genes can occur in a rhythmic fashion, and colony-wide oscillations emerge in a seemingly spontaneous fashion.
A natural question arises regarding the behavior of individual cells in the absence of coupling. While experimentally we cannot turn off the coupling while maintaining intracellular gene activation, we addressed this question using simulations by artificially setting the AHL diffusion rate across the cell membrane to zero (with the other parameters fixed). We find that individual cells oscillate independently for any cell density since they are completely decoupled from the environment and each other. This result indicates that the coupling through AHL diffusion provides a means for the synchronization of individual oscillators at intermediate cellular concentrations.
In the mid seventeenth century, Chirstiaan Huygens serendipitously observed that two pendulum clocks oscillated in synchrony when mounted to a common support beam 1. While observations of synchronization in nature surely predate the age of enlightenment, Huygens is credited as the first to systematically characterize the synchronization of oscillators in terms of a known coupling mechanism (which, in the case of the pendula, he deduced as vibrations in the common support). We have shown how quorum sensing can be used to couple genetic clocks, leading to synchronized oscillations at the colony level. Given the single-cell variability and intrinsic stochasticity of most synthetic gene networks 27, 30, 31, 45, 46, the use of quorum sensing is a promising approach to increasing the sensitivity and robustness of the dynamic response to external signals. Along these lines, our results set the stage for the design of networks that can function as spatially distributed sensors or synthetic machinery for coupling complex dynamical processes across a multicellular population.
Three identical transcriptional cassettes for luxI, aiiA, and yemGFP were constructed by replacing a modular pZ plasmid’s promoter 47 (with yemGFP) with the lux operon from the native Vibrio Fischeri operon (luxR up to luxI stop codon) 48. LuxI and aiiA 49 genes were cloned in place of yemGFP and a degradation tag was added to the carboxy-terminal of each. A previously used MG1655 strain of Escherichia coli1 was transformed with plasmids pTD103luxI/GFP(colE1,Kan) and pTD103aiiA(p15A,Amp) to create strain TDQS1 (Suppl. Info).
Each experiment started with a 1:1000 dilution of overnight culture grown in 50mL LB (10g/L NaCl) with antibiotics 100µg/ml ampicillin(Amp) and 50µg/ml kanamycin(Kan) for approximately 2 hours. Cells reached an OD600 of 0.05–0.1 and were spun down and concentrated in 5mL of fresh media with surfactant concentration of 0.075 Tween20 [Sigma-Aldrich, St.Louis,MO] before loading in a device.
Images were acquired using an epifluorescent inverted microscope (TE2000-U, Nikon Instruments Inc., Tokyo, Japan), and chip temperatures were maintained at 37°C with a plexiglass incubation chamber encompassing the entire microscope. Phase contrast and fluorescent images were taken at 20× or 60× every 2–5 minutes and focus was maintained automatically using Nikon Elements software.
We thank Jesse Stricker for helpful discussions on plasmid construction, and Matt Bennett, Kurt Wiesenfeld, and Jim Collins for stimulating discussions during the preparation of the manuscript. This work was supported by the National Institutes of Health and General Medicine (GM69811), the DOE CSGF fellowship (TD), and CONACyT (Mexico, grant 184646, OMP).
Supplementary information, including methods, supplementary figures and tables, and timelapse microscopy movies, is linked to the online version of the paper at www.nature.com/nature.
The authors declare no competing financial interests.