Multistable and dynamic CRISPRi-based synthetic circuits in E. coli

Gene expression control based on CRISPRi (clustered regularly interspaced short palindromic repeats interference) has emerged as a powerful tool for creating synthetic gene circuits, both in prokaryotes and in eukaryotes; yet, its lack of cooperativity has been pointed out as a potential obstacle for dynamic or multistable circuit construction, raising the question of whether CRISPRi is widely applicable for synthetic circuit design. Here we use CRISPRi to build prominent synthetic gene circuits that accurately govern temporal and spatial gene expression in Escherichia coli. We report the first-ever CRSPRi oscillator (“CRISPRlator”), bistable network and stripe pattern-forming incoherent feed-forward loop (IFFL). Our circuit designs, conceived to feature high predictability and orthogonality and low metabolic burden and context-dependency, allowed us to achieve robust circuit behaviors (e.g. synchronous oscillations) and to expand the IFFL into a twice as complex, two-stripe patterning system. Our work demonstrates the wide applicability of CRISPRi in synthetic circuits and paves the way for future efforts towards engineering more complex synthetic networks, boosted by the advantages of CRISPR technology.

2 design. Due to its RNA-guided nature, CRISPRi is highly programmable 4,5 , allows for easy design of sgRNAs that can be highly orthogonal 6 and whose behavior in different environments can be easily predicted in silico 7,8 . It also imposes low burden on host cells 2 and is encoded in shorter sequences than protein-based repressors, thereby facilitating circuit handling and delivery and reducing cost. A potential drawback of CRISPRi is the lack of non-linearity 9 of its dose-reponse functions. Cooperative protein transcription factors typically function non-linearly, a difference that might prevent the successful implementation of CRISPRi-based dynamic and multistable circuits [9][10][11] . Besides, unlike other RNA circuits that exhibit fast dynamics, low rates of dCas9:DNA dissociation in the absence of DNA replication 12,13 effectively slow down CRISPRi dynamics.
The last few years have seen a growing interest in developing CRISPRi-based synthetic circuits 6,9,[14][15][16][17][18][19][20][21][22] . However, despite the enormous potential of CRISPRi for synthetic circuit design, the use of CRISPRi circuits in prokaryotes has been largely focused on logic gates and to the best of our knowledge none of the flagship circuits in synthetic biology (namely, the bistable toggle-switch 23 and the repressilator 24 ) have been re-constructed using CRISPRi. Here, we fill this unaddressed gap by demonstrating that CRISPRi can be used for building some of the most notorious (synthetic) circuit topologies: the repressilator, a bistable circuit, and an incoherent feed-forward loop (IFFL, a.k.a. band-pass filter) that drives stripe pattern formation.
To maximize circuit predictability and orthogonality and minimize metabolic burden on the host, we employed a series of design features in all our synthetic networks. Circuit components were all expressed from a single vector (so-called "variable" vector) to avoid fluctuations in their stoichiometry. Circuit transcriptional units (TUs) were isolated from each other by strong transcriptional terminators and 200 bp spacer sequences, to avoid transcriptional readthrough and minimize compositional context 25 . To prevent mRNA context-dependency and provide transcriptional insulation within TUs, a 20 bp csy4 cleavage sequence 26 was used flanking sgRNAs and upstream of the ribosome binding sites (RBS) of the reporter genes. Orthogonal degradation tags 27 were used to avoid cross-talk between fluorescent reporters. Sequence repetition at the DNA level were minimized to prevent unwanted recombination events. The levels of dCas9 and Csy4 were kept constant by expressing them from constitutive promoters in a separate "constant" vector. All gene circuits were tested in E. coli (MK01 28 ) incubated in a rich defined media (EZ, Teknova) for maximizing cell fitness while reducing variability. In order to speed up the design-build-test cycle, we adopted a previously-described cloning strategy that allows for fast and modular assembly of synthetic networks 29 .
We first devised a simple circuit to assess the dose-response of CRISPRi repression. This 2-node network design displays Boolean NOT logic, in which node 1 (N1) is induced by Larabinose (Ara) and represses expression of node 2 (N2) (Supplementary Fig. 1). Levels of N1 and N2 in the NOT circuit were monitored via the fluorescent reporters mKO2 and sfGFP ( Supplementary Fig. S1b) carrying degradation tags (MarA and MarAn20) to accelerate reporter turnover 27 . We chose four single-guide RNAs (sgRNA-1 to -4) and their corresponding binding sites (bs-1 to -4), previously shown to be orthogonal to each other 6 . Maximal induction (0.2% Ara) of N1 produced sgRNA levels that repressed the expression of N2 in the range of 10 to 30fold ( Supplementary Fig. S1c), with sgRNA-2 being the strongest repressor. Truncation of the four 5' nucleotides ("t4") of sgRNA-1 and -4 reduced maximal repression by 20% and 45%, respectively, providing a means to tune-down repression ( Supplementary Fig. S1c). We also measured N2 fluorescence in the presence of different Ara concentrations ( Supplementary Fig.   S1d), which showed a modulable, dose-dependent CRISPRi repression, a prerequisite for building dose-dependent circuits.
We next sought to build a classical genetic circuit that has never been constructed before using CRISPRi: a bistable circuit (BC) 23 , where two nodes mutually repress each other resulting in two mutually-exclusive states. We designed a double-inducible BC in which each of the two nodes responds to a small molecule (Ara or N-(β-Ketocaproyl)-L-homoserine lactone, AHL) by producing a fluorescent reporter (mKO2-MarA or sfGFP-MarAn20) and a sgRNA (-1 or -4) that represses the other node ( Fig. 1a and b). When BC bacteria were subjected to perpendicular gradients of Ara and AHL following pre-incubation in either one or the other inducer, they displayed different reporter expression profiles depending on the previous state (Fig. 1c), a phenomenon known as hysteresis that characterizes bistable systems like the toggle switch 23   We next decided to build another flagship synthetic circuit, a stripe-forming IFFL ( Fig. 2) 30-32 . We designed a CRISPRi-based 3-node IFFL type 2 (I2), which relies exclusively on repression interactions 33 . To monitor the behavior of all three nodes of the circuit, we employed compatible fluorescent reporters fused to orthogonal degradation tags 27 : mKO2-MarA, sfGFP-avoiding fluctuations in their stoichiometry due to plasmid copy number variations. To corroborate that the CRISPRi IFFL was capable of true spatial patterning, we homogeneously spread bacteria on an agar plate and applied a central source of arabinose, which diffused forming a radial gradient. Consistent with the French flag model 34 , the initially identical, isogenic bacterial population carrying the synthetic network interpreted the Ara gradient into three discrete gene expression programs: "orange", "green" and "blue" (Fig. 2d).
We next wanted to explore the robustness of the CRISPRi stripe circuit with respect to variations in its molecular components. We assessed the performance of i) a circuit carrying a single reporter (sfGFP-MarAn20 in N3), ii) a circuit with a different set of sgRNA regulators (sgRNA-4, sgRNA-4t4 and sgRNA-3 instead of sgRNA-1, sgRNA-1t4 and sgRNA-2), and iii) a network with swapped N2 and N3 reporters ( Supplementary Fig. 5). All these circuits showed a stripe behavior ( Supplementary Fig. 5), demonstrating the robustness of our CRISPRi-based IFFL, which is enabled by our design aiming at high orthogonality and low context-dependency.
An important limitation that commonly affects synthetic gene circuits is the difficulty to expand their functionality by including existing circuits into more complex networks or by making them operate in a different environment, such as in co-existence with another circuit. To test whether a CRISPRi-based IFFL could operate in a similar manner in a different context, we decided to build two orthogonal CRISPRi IFFLs that would operate in parallel within the same bacterial cell, each producing a single (and independent) stripe in response to a common arabinose gradient. When bacteria containing both parallel circuits were submitted to different Ara concentrations, a double peak of gene expression could be observed at intermediate inducer levels corresponding to two overlapping peaks, each generated by one of the circuits, as designed (Fig. 2e). To our knowledge, this is the first study reporting a double synthetic stripe driven by a single (isogenic) cell population. The simultaneous yet independent operation of the double-stripe system was enabled by the intrinsic properties of CRISPRi (orthogonality, low burden, short encoding sequences) and the low context-dependency of the design. Next, we designed an IFFL in which both the control and the reporting are RNA-based.
To this aim, we modified our CRISPRi IFFL (in which the control is sgRNA-based) to remove the "orange" and "blue" reporters and substituted the original "green" reporter (sfGFP) with 8 copies of the dBroccoli (dimeric Broccoli) RNA aptamer 35 (Supplementary Fig. 6). When we incubated the engineered bacteria with the DFHBI-1T fluorophore we observed a low-background Broccoli stripe ( Supplementary Fig. 6b), showing that the band-pass behavior of an RNA-controlled IFFL can indeed be reported by an RNA molecule.
Finally, we wanted to explore whether CRISPRi-based circuits could be employed not only for spatial patterning, as demonstrated above, but also to create temporal patterns.
Oscillations are temporal patterns with key roles in many different biological phenomena. We sought to engineer an oscillatory pattern controlled by a dynamic CRISPRi circuit. To this aim, we designed a 3-node ring oscillator: the "CRISPRlator" (Fig. 3), in honor of the first synthetic oscillator (the repressilator) 24 . Bacteria carrying the CRISPRlator were grown in a continuous culture within a microfluidic device ( Supplementary Fig. 5) for 3 days (over 100 generations) and imaged every 10 min. The CRISPRlator generated robust oscillations between three states ("red", "green" and "blue") with a period of 10-12 h (14-17 generations, Fig. 3c and d).
Oscillations were synchronized at the population level (~110 cells per chamber) for hundreds of generations even in the absence of any engineered synchronization mechanism (Supplementary Movie 1) and can thus rival the optimized versions of the protein-based repressilator 36,37 . Together, our results demonstrate that the lack of cooperativity of CRISPRi is not an insurmountable obstacle for (complex) synthetic circuit construction and CRISPRi can effectively be used for generating dynamic and multistable behaviors. Future work will reveal the source of non-linearity in these circuits. The benefits inherent to CRISPRi circuits will likely prompt the construction of new/extended versions of the BC and the CRISPRlator, including designs hard to achieve with protein repressors, e.g. two independent clocks operating within the same cell. The universality of CRISPR should facilitate the transfer of our circuits into other species. We anticipate that our results will encourage the synthetic biology community to employ CRISPRi for gene circuit design and inspire future construction of more complex synthetic networks.

Circuit construction.
Genes encoding sfGFP, mKO2, mKate2, Csy4 and dCas9 were obtained as previously described 29 . mCherry and mCitrine were amplified from plasmids (pLacI(r)-mCherry(ASV) and pCI-Citrine-(ASV)) kindly provided by Sebastian Maerkl 36 . dBroccoli was amplified from plasmid pET28c-F30-2xdBroccoli, which was a gift from Samie Jaffrey (Addgene plasmid #66843). Cerulean 38 was purchased as an E. coli codon-optimized gBlock from IDT, and LuxR (BBa_C0062) was also codon-optimized and synthesized by GenScript. Primers were purchased desalted from Microsynth or Sigma-Aldrich. The circuits were constructed employing a previously described Gibson-based cloning framework that allows for the fast and modular cloning of synthetic gene networks 29 . Plasmids used in this study (Supplementary Table 1) are available through Addgene.

Microplate reader experiments.
Gene expression of fluorescent reporters was used to assess synthetic circuit performance; fluorescence was measured in microplate readers (except for the agar plate experiment in Fig.   2d and the microfluidic experiment in Fig. 3). MK01 28 electrocompetent cells were transformed with a "constant" plasmid encoding proteins required for circuit function (namely dCas9 and Csy4, plus LuxR when applicable) as well as with a "variable" vector bearing AraC (when needed) and the CRISPRi circuit (see Supplementary Table 1 for a detailed plasmid description). This was done for mKO2 and sfGFP separately, after which values for both reporters were added and divided by two to yield the total hysteresis in a scale from 0 to 1.

Flow cytometry.
Bacteria carrying the BC network were pre-cultured (~6 h) and cultured (16 h) as stated above, diluted 1:200 in phosphate-buffered saline (PBS) and analyzed with a BD LSRFortessa flow cytometer using a 488 nm laser in combination with FITC filter for sfGFP and a 561 nm laser with a PE-Texas Red filter for mKO2. 10'000 events were acquired with BD FACSDiva 6.2 software.
Bacterial cells were gated in the FSC-H vs SSC-H plot using FlowJo (FlowJo, LLC) and plotted using R 39 .
Agar plate assay. MK01 28 electrocompetent cells were transformed with the "constant" plasmid pJ1996_v2 29 , which carries dCas9 and Csy4, and a "variable" plasmid encoding the three-color IFFL (pJ2042.2). 2 ml of selective LB were inoculated with single cells and grown at 37° C for ~6 h; cells were pelleted at 4000 rcf and resuspended in selective EZ medium (Teknova) with 0.4% glycerol. 300 µl of 0.15 OD600 bacterial suspensions were added to pre-warmed (1 h 37° C) Petri dishes containing selective EZ medium (Teknova) with 0.4% glycerol and 0.9% agar, and suspensions were spread with sterile glass beads. After incubating 1 h at 37° C, a filter paper disk was placed in the center of the agar, and 15 µl of 5% Ara were delivered onto the disk. Plates were incubated at 37° C and imaged after 20 h with an Amersham Typhoon 5 Biomolecular Imager (GE Healthcare) using the following settings: mKO2: Ex. 532 nm, Em. 560-580 nm; sfGPF: Ex. 488 nm, Em. 515-535 nm; mKate2: Ex. 635 nm, Em. 655-685 nm. Grayscale images were converted to color images using Fiji 40 and overlaid.

Microfluidic experiments.
MK01 28 electrocompetent cells were transformed with the "constant" plasmid pJ1996_v2 29 and a "variable" plasmid encoding the CRISPRlator. Single colonies were used to inoculate 5 ml of selective LB, which were grown overnight at 37° C. Next morning, 3 ml of selective EZ containing 0.85 g/l Pluronic F-127 (Sigma) were inoculated with the overnight preculture in a 1:100 ratio and grown for 3-4 h at 37° C. Cells were centrifuged for 10 min at 4000 rcf and resuspended in ~10 µl of the supernatant to obtain a dense suspension, which was loaded into the PDMS microfluidics device. Cells were grown in a continuous culture inside microfluidic chambers (dimensions: 1.2 µm x 12 µm x 60 µm, h x w x l, purchased from Wunderlichips, see design in Supplementary Fig.   7) for 3 days with a constant 0.5 ml/h supply of fresh medium (selective EZ plus 0.85 g/l Pluronic F-127) and removal of waste and excess of bacteria, powered by an AL-300 pump (World Precision Instruments). Imaging was performed using a Leica DMi8 microscope and a Leica DFC9000 GT camera, with the following settings: Cerulean: Ex. 440 nm 10% 50 ms, Em. 457-483 nm; mCitrine: Ex. 510 nm 10% 50 ms, Em. 520-550 nm; mCherry: Ex. 550 nm 20% 200 ms, Em. 600-670 nm. Imaging started after 20 h to allow cells to fill the chamber and oscillations to stabilize, and images were collected every 10 min with LAS X software, and analyzed using Fiji 40 for both quantification and montage.

Author Contributions
JSM and YS designed research. JSM performed experiments, analyzed data, and prepared the figures. JSM and YS wrote the manuscript. All authors have given approval to the final version of the manuscript.

Conflict of Interest
The authors declare no conflict of interest.