SUMMARY:
While complex dynamic biological networks control gene expression and metabolism in all living organisms, engineering comparable synthetic networks remains challenging1,2. Conducting extensive, quantitative and rapid characterization during the design and implementation process of synthetic networks is currently severely limited due to cumbersome molecular cloning and the difficulties associated with measuring parts, components and systems in cellular hosts. Engineering gene networks in a cell-free environment promises to be an efficient and effective approach to rapidly develop novel biological systems and understand their operating regimes3-5. However, it remains questionable whether complex synthetic networks behave similarly in cells and a cell-free environment, which is critical for in vitro approaches to be of significance to biological engineering. Here we show that synthetic dynamic networks can be readily implemented, characterized, and engineered in a cell-free framework and consequently transferred to cellular hosts. We implemented and characterized the “repressilator”6, a three-node negative feedback oscillator in vitro. We then used our cell-free framework to engineer novel three-node, four-node, and five-node negative feedback architectures going from the characterization of circuit components to the rapid analysis of complete networks. We validated our cell-free approach by transferring these novel three-node and five-node oscillators to Escherichia coli, resulting in robust and synchronized oscillations reflecting the in vitro observation. We demonstrate that comprehensive circuit engineering can be performed in a cell-free system and that the in vitro results have direct applicability in vivo. Cell-free synthetic biology thus has the potential to drastically speed up design-build-test cycles in biological engineering and enable the quantitative characterization of synthetic and natural networks.
MAIN:
A central tenet of engineering involves characterizing and verifying prototypes by conducting rapid design-build-test cycles in a simplified environment. Electronic circuits are tested on a breadboard to verify circuit design and aircraft prototypes are tested in a wind tunnel to characterize their aerodynamics. A simplified environment does not exist for engineering biological systems, nor is accurate software based design possible, requiring design-build-test cycles to be conducted in vivo2. To fill the gap between theoretical design and laborious in vivo implementation for biological systems we devised a cell-free framework consisting of E. coli lysate (“TX-TL”)5,7 and a microfluidic device capable of emulating cellular growth and division3 (Fig. 1). Almost all prototyping can be done on linear DNA, which requires less than 8 hours to assemble and test. The cell-free framework provides a simplified and controlled environment that allows us to drastically reduce the design-build-test cycle8.
We first asked whether our cell-free framework could be used to run and characterize an existing synthetic in vivo circuit and chose to test the repressilator6 as a model circuit. We successfully implemented the original repressilator network in our cell-free framework and observed long-term sustained oscillations with periods matching the in vivo study (Fig 2, Supp. Video 1). We compared the original repressilator to a modified version containing a point mutation in one of the CI repressor binding sites in the promoter regulating LacI (Fig. 2a). This mutation increases the repressor concentration necessary for half-maximal repression (KM), and reduces cooperativity9. At long dilution times (td) both circuits oscillated, but with shifted absolute reporter protein concentrations (Fig. 2b). At decreasing dilution times amplitudes decreased and periods became faster with a linear dependence on td. Faster dilution times, however, did not support oscillations for the modified network (Fig. 2b-c). Experimentally, the range of dilution times supporting oscillations can serve as a measure for robust oscillator function, which generally diminishes with decreasing synthesis rates or when binding of one repressor to its promoter is weakened as in the OR2* mutant (Supp. Fig. 1). To give another example for an experimental characterization that would be challenging to perform in a cellular environment we analyzed the repressilator network in phase space showing limit cycle oscillations and invariance to initial conditions (Fig 2d).
The cell-free framework also allows rapid characterization of individual network components. We measured the transfer functions of repressor-promoter pairs in the repressilator network (Fig. 3a, Supp. Fig. 2a,b, Table 1) and found that the network is symmetric in terms of transfer functions. In the CI promoter OR2* mutant we observed the expected shift in KM and decreased steepness of the transfer function. We also characterized TetR repressor homologs as building blocks for novel negative feedback circuits (Fig 3a) and with the exception of QacR observed similar transfer functions as observed in vivo10 (Supp. Fig 2c).
Using three new repressors, BetI, PhlF and SrpR, we constructed a novel 3-node (3n) circuit (3n1) and observed high-amplitude oscillations over a broad range of dilution times with the same dependence of amplitude and period on td as for the repressilator (Fig 3b). In our characterization of the repressilator network and the 3n1 oscillator we found dilution rates to be critical for the existence, period and amplitude of oscillations. Protein degradation is similar to dilution in that it results in removal of repressor proteins. In order to study the effect of degradation we constructed a second 3n network (3n2) using TetR, PhlF and SrpR repressors on linear DNA. One version of the circuit used strong ssrA ClpXP degradation tags, while the second used untagged repressors. We observed oscillations for both circuits (Fig 3c). However, the circuit without ssrA-tag mediated protein degradation exhibited slower oscillations, which extended to lower dilution times, showing that protein degradation, just like dilution, affects oscillator function and period. Effects of ClpXP-mediated protein degradation, which have been shown to be important for existence and frequency of oscillations in vivo11,12, can thus be emulated in a cell-free environment.
Theory predicts that ring architectures built from an odd number of repressors oscillate, while even-numbered architectures have stable steady states13,14. We experimentally built and tested a 4-node circuit from LacI, TetR, PhlF and SrpR on linear DNA. Initial pulses of LacI inducer IPTG or TetR inducer aTc allowed us to switch expression into either one of the two stable steady states (Fig. 3d).
Encouraged by the robust oscillations observed in the 3n networks, we built two 5-node ring oscillators (5n) to test our prototyping environment on a novel synthetic network architecture (Fig. 3e). Despite their considerable complexity both circuits oscillated over a broad range of dilution times with the expected period lengthening, which could be as long as 19h. Comparing all ssrA-tagged 3n and 5n ring architectures, we show that the observed periods could be accurately predicted for all four networks by computational simulations (Fig. 3f). Our cell-free framework allows testing and characterization of complex networks including verifying networks cloned onto a single plasmid, which is the closest approximation to in vivo implementation (Supp. Fig. 3).
To validate our cell-free approach in vivo we cloned the 3n1 and 3n2 networks onto low-copy plasmids and co-transformed each with a medium-copy reporter plasmid into lacI-JS006 E. coli15. When tested on a microfluidic device (mother machine16), both 3n oscillators showed regular oscillations with periods of 6 ±1 hours for at least 30 hours (Fig. 4a, Supp. Video 2,3). Both oscillators were surprisingly robust as all cells undergoing healthy cellular division oscillated (n = 71) (Supp.. Fig. 4, Supp. Video 4).
We next tested our 5n oscillators in vivo. Due to loading effects17, 5n1 was not viable when co-transformed with a strong reporter. When tested with a low expression strength reporter both 5n oscillators showed robust oscillations that were maintained for at least 70 hours, and over 95% of all analyzed traps containing healthy cells oscillated (n = 104). In addition, both 5n networks oscillated with similar periods: 8 hours for 5n1, and 9 hours for 5n2 (Fig. 4b, Supp. Fig. 4, Supp. Video 5, 6, 7).
We also tested both 3n oscillators on a CellASIC system, which allows planar single-layer colony formation. In this system we observed a striking population level synchronization of daughter cells inheriting the oscillator state from their mother cells (Fig. 4c, Supp. Fig. 5a, Supp. Video 8,9). Synchronization was also apparent when using three different fluorescent reporters simultaneously (Supp. Fig. 5b, Supp. Video 10). We did not observe population level synchronization in the original repressilator, the OR2* mutant (Supp. Fig. 5c) nor the 5n networks. Synchronized oscillations were not reported with the original repressilator6, and have only been observed in oscillators using intercellular communication18,19. We hypothesize that the 3n1 and 3n2 synchronization is due to increased repressor concentrations as compared to the original repressilator network (Supp. Fig. 5d), which increases the inheritance of the period phenotype and minimizes the rapid de-synchronization expected from stochastic cellular protein fluctuations20. However, a quantitative characterization of the synchronization phenotype requires more in depth understanding of stochastic effects in vivo.
Because cells were synchronized, we were able to analyze the population as a whole to make general conclusions of oscillator behavior. We varied dilution time by using different media conditions and media flow rates, and found a direct relationship between division times and period, consistent with the in vitro data collected. Oscillation periods of the 5n oscillators were also consistent with our in vitro results and showed a similar dependence on doubling time (Fig. 4d).
Finally, we compared 3n1 and 5n2 with weak and strong reporters in vivo to analyze the effect of protein degradation on the oscillator period. We theorized that given a constant concentration of ClpXP, stronger reporters would result in more ClpXP loading, thereby slowing the period of oscillation. ClpXP is thought to influence oscillation dynamics in vivo in this manner11. We found that in the mother machine, both the period distributions of 3n1 and 5n2 showed this characteristic (Fig. 4e), which reflects our in vitro findings of differential –ssrA tag dependent period length (Fig. 3c).
We demonstrated the utility of our cell-free framework for biological systems engineering and component characterization. We observed some differences between the in vitro and cellular environment, particularly in the difficulty of predicting cellular toxicity and loading effects of the 5n oscillators in vivo. While more work is necessary describing and explaining differences between in vitro and in vivo environments8,21, the observed behavior of complex networks in our cell-free environment reflected network behavior in vivo well. The cell-free framework is thus a powerful emulator of the cellular environment allowing precise control over experimental conditions and enabling studies that are difficult or time consuming to perform in cells. With further developments in cell-free lysate systems and supporting technologies, the in vitro approach is posed to play an increasing role in biological systems engineering and provides a unique opportunity to design, build, and analyze biological systems.
METHODS:
TX-TL reactions
Preparation of TX-TL was conducted as described previously7, but using strain “JS006” co-transformed with Rosetta2 plasmid and performing a 1:2:1 extract:DNA:buffer ratio. This resulted in extract “eZS4” with: 8.7 mg/mL protein, 10.5 mM Mg-glutamate, 100 mM K-glutamate, 0.25 mM DTT, 0.75 mM each amino acid except leucine, 0.63 mM leucine, 50 mM HEPES, 1.5 mM ATP and GTP, 0.9 mM CTP and UTP, 0.2 mg/mL tRNA, 0.26 mM CoA, 0.33 mM NAD, 0.75 mM cAMP, 0.068 mM folinic acid, 1 mM spermidine, 30 mM 3-PGA, 2% PEG-8000. For experiments utilizing linear DNA GamS was added to a final concentration of 3.5 μM8.
Steady-state reactions
Experiments were performed in a microfluidic nano-reactor device as described previousely3,7 with some modifications to optimize the conditions for the lysate-based TX-TL mix. Reaction temperature was 33°C. Lysate was diluted to 2x of the final concentration in 5 mM HEPES 5 mM NaCl buffer (pH 7.2). The reaction buffer mix was combined with template DNA and brought to a final concentration of 2x. For a 24 h experiment 30 μl of these stocks were prepared. During the experiment, lysate and buffer/DNA solutions were kept in separate tubing feeding onto the chip, cooled to approximately 6°C, and combined on-chip. We ran experiments with dilution rates (μ) between approximately 2.8 and 0.5 h−1, which corresponds to dilution times, td = ln(2) μ−1, between 15 and 85 min. These were achieved with dilution steps exchanging between 7 and 25% of the reactor volume with time intervals of 7 to 10 min, which alternately added fresh lysate stock or fresh buffer/DNA solution into the reactors. Dilution rates were calibrated before each experiment. Initial conditions for the limit cycle analysis of the repressilator network were set by adding pre-synthesized repressor protein at the beginning of each experiment. For this, CI repressor (together with Citrine reporter) and TetR repressor (together with Cerulean reporter) were expressed for 2.5h in batch. On chip the initial reaction was mixed to be composed of 25% pre-synthesis reaction and 75% fresh TX-TL mix and repressilator template DNA. Then, the experiment was performed at a td of 19.2 ± 0.3 min. Initial conditions for the 4-node experiment were 2.5μM aTc or 250μM IPTG, and the experiment was performed at a td of 44.5 ± 0.9 min. DNA template concentrations used in steady-state reactions are listed in Supp. Table 4. Arbitrary fluorescence values were converted to absolute concentrations from a calibration using purified Citrine, Cerulean, and mCherry, which were prepared using previously published protocols utilizing a His6 purification method followed by size-exclusion chromatography and a Bradford assay to determine protein concentration8.
DNA and strain construction
DNA was constructed using either Golden Gate Assembly or Isothermal Assembly. The original repressilator plasmid, pZS16 was used as a template for initial characterization and for construction of the OR2* mutant. Transfer function plasmids were constructed by Transcriptic, Inc. For other plasmids, partial sequences were either obtained from Addgene10 or synthesized on gBlocks or ssDNA annealed oligonucleotides (Integrated DNA Technologies). For linear DNA, all DNA was constructed using previously published Rapid Assembly protocols on a “v1-1” vector8. Specific plasmids required secondary-structure free segments, which were designed by R2oDNA23. JS00615 was co-transformed with origin-of-replication compatible plasmids to create engineered strains. Specifically, negative-feedback oscillator units were cloned onto pSC101* low copy plasmids (ampR or kanR), while reporters were cloned onto colE1 medium copy plasmids (kanR or cmR). To modulate the reporter copy number, all experiments were conducted below 37°C24. Strain passage was minimized to avoid plasmid deletions due to the recA+ nature of JS006 and the high complexity of oscillator plasmids or triple-reporter plasmid. Based on the in vitro and in silico results, we used strong transcriptional and translational25 units to maximize gain.
Transfer function measurement
Transfer functions of the repressor – promoter pairs were determined in the nano-reactor device at a minimum of two different dilution times (Supp. Fig. 2). All tested promoters were cloned into a plasmid in front of a BCD7 ribosomal binding site and the Citrine open reading frame. A non-saturating concentration of 1nM plasmid was used in the experiment. The repressors were expressed from linear templates carrying the J23151 promoter and the BCD7 ribosomal binding site with time-varying concentrations, which were increased from 0 to 2.5nM and decreased back to 0 during the course of the experiment3. Simultaneously we expressed Cerulean as a reporter for the repressor concentration from a linear template at an identical concentration as the repressor template. From the concentration of the Citrine reporter we calculated the synthesis rate of the fluorescent protein over time using a model of steady state protein synthesis in the nano-reactor device3, where Pd and Pf are dark and fluorescent reporter concentration respectively, t is time, Δt is the time interval between dilution steps, dil is the volume fraction replaced per dilution step, which was determined during the calibration of the device, and mat is maturation rate of the fluorescent protein. Maturation times of Citrine and Cerulean were determined as described previously3 and were 15 ±4 min for Cerulean and 29 ±3 min for Citrine. Dark fluorescent protein was calculated from equation (2): and the synthesis rate was calculated from equation (1):
We used the sum of measured fluorescent Cerulean concentration and equation (3) for dark Cerulean as a measure of the total repressor protein present at any time during the experiment.
The synthesis rates were normalized to their respective maximal values (vmax) and plotted against the concentration of the repressor reporter using only repressor concentrations higher than 1nM. The transfer curves were then fit to a Hill function where y is the synthesis rate, ymin is the minimum synthesis rate, n is the Hill coefficient and KM is the Michaelis Menten constant for half maximal promoter activity. The fitting was performed in Igor Pro using orthogonal distance regression with ODRPACK95 assuming a 9% error in the measurements of Citrine and Cerulean fluorescence.
Vmax measurements
Relative promoter strengths (vmax values) were determined using the transfer function promoter plasmids. In vitro strengths were determined in 5 μl TX-TL reactions at a DNA template concentration of 1nM. Reactions were assembled in 384-well plates, overlaid with 35 μl Chill-Out Liquid wax (BioRad) and analyzed using a Biotek SynergyMx plate reader set to 33°C reaction temperature, and reading Citrine fluorescence with Exc:510±9nm and Em:540±9nm. For comparison, Citrine fluorescence at 6h was normalized to the value of pLacI.
In vivo strengths were determined using E. coli JS006 transformed with the same plasmids. Cells were grown at 29°C in MOPS medium supplemented with 0.4% glycerol and 0.2% casaminoacids. For each strain, three independent overnight cultures were diluted 1:50 and grown to mid-log phase. They were then diluted to a starting OD600 of 0.15 into 100 μl growth medium in a 96-well plate and grown in the plate reader at 29°C with periodic shaking measuring Citrine fluorescence. Fluorescence values were normalized to OD resulting in steady state values after 2 h. Average steady state values were normalized to pLacI for comparison with the in vitro measurement.
In vivo experiments
Mother machine16 experiments were conducted with custom-made microfluidic chips (mold courtesy of M. Delincé and J. McKinney, EPFL). E. coli cells were trapped in channels of 30 μm length, 2 μm width and 1.2 μm height. Before loading onto the device, cells were grown from a frozen stock to stationery phase. Cells were then concentrated 10-fold and loaded onto the chip. Experiments were performed using LB medium supplemented with 0.075% Tween-20 at a flow rate of 400 μl/h. Oscillation traces were collected from single mother machine traps using the background subtracted average fluorescence intensity of the entire trap.
CellASIC experiments were conducted using B04A plates (Merck Millipore, Darmstadt Germany). Flow rates were varied between 0.25 psi − 2 psi. Cells were grown from frozen stock in media at running temperature to stationery phase. Cells were then diluted 1:100 for 2 hours, and loaded on a equilibrated plate at 1:1000 or less to achieve single-cell loading efficiencies per chamber. To vary cellular doubling times, different growth media were used: LB (BD Biosciences), M9CA (Sigma Aldrich) with 0.2% glucose, 2xYT (MP Bio), MOPS EZ Rich (Teknova).
Cells were imaged in time series every 10-20 min using a 100x phase objective minimizing both lamp intensity (12% Xcite 120, Excelitas Inc. Waltam MA or 1-2% CoolLED pE-2, Custom interconnected Ltd., UK) and exposure times (<500ms) to limit photo-toxicity.
Analysis of in vivo data
Images were processed and stitched26, if necessary, using Fiji/ImageJ27. Fluorescence traces of cell populations with synchronized oscillations were extracted from CellASIC movies using background corrected mean fluorescence intensity from the entire field of view. For cells that were not synchronized over the complete field of view, we tracked regions of oscillating sister cells at the edge of the microcolony. We used ImageJ to define polygonal regions around those cells and manually shifted the polygonal region to track the front of growing cells. Periods were determined from fluorescence traces derived from mother machine and CellASIC movies by measuring the time from one oscillation peak to the next peak. Doubling times were estimated by averaging over the doubling times of at least ten individual cells.
Supplementary Information:
This manuscript contains Supplementary Information: Supplemental Figures 1 – 5, Supplemental Videos 1 – 10, Supplemental Tables 1 – 5, Supplemental Model Information.
Author Contributions:
All authors contributed extensively to the work presented in this paper.
Author Information:
The authors declare no competing financial interests. Correspondence and requests for materials should be addressed to S.J.M (sebastian.maerkl{at}epfl.ch).
Acknowledgements:
We thank Yin He, Transcriptic, Inc. and Holly Rees for cloning assistance, Jan Kostecki and Stephen Mayo for protein purification and size exclusion chromatography assistance, Rohit Sharma and Marcella Gomez for initial testing and modeling of oscillators in vitro, Kyle Martin for laboratory assistance, Adam Abate, Tanja Kortemme, and Charles Craik for laboratory space and equipment, Matthieu Delincé, Joachim De Jonghe, Marc Spaltenstein, John McKinney and Jin Park for mother machine material and assistance, Tim Chang and Benjamin Alderete for CellASIC assistance, and Michael Elowitz for insights and scientific support. This work was supported in part by EPFL and the Defense Advanced Research Projects Agency (DARPA/MTO) Living Foundries program, contract number HR0011-12-C-0065 (DARPA/CMO). Z.Z.S. is also supported by a UCLA/Caltech Medical Scientist Training Program fellowship, Z.Z.S and E.Y by a National Defense Science and Engineering Graduate fellowship, and Y.H. by a JSPS Fellowship for Research Abroad. The views and conclusions contained in this document are those of the authors and should not be interpreted as representing officially policies, either expressly or implied, of the Defense Advanced Research Projects Agency or the U.S. Government.