Filamentation and restoration of normal growth in E.coli using a combined CRISPRi sgRNA/antisense RNA approach

CRISPR interference (CRISPRi) using dCas9/sgRNA is a powerful tool for the exploration and manipulation of gene functions. Here we quantify the reversible switching of crucial cellular processes by CRISPRi and an antisense RNA mechanism. Reversible induction of filamentous growth in E. coli has been recently demonstrated by controlling the expression levels of the bacterial cell division proteins FtsZ/FtsA via CRISPRi. If FtsZ falls below a critical level, cells cannot divide. However, the cells remain metabolically active and continue with DNA replication. We surmised that this makes them amenable to an inducible antisense RNA strategy to counteract FtsZ inhibition. We show that both static and inducible thresholds can adjust the characteristics of the switching process. Combining bulk data with single cell measurements, we clarify the role of bacterial heterogeneity and population dynamics for gene circuits affecting cell division. Filamentation is shown to strongly increase gene expression variability in the bacteria. Furthermore, we find reversible switching only in a small subpopulation of the bacteria, which takes over the population upon continued cell division. Successful restoration of division occurs faster in the presence of antisense sgRNAs than upon simple termination of CRISPRi induction.


Introduction
Bacterial growth rate and protein expression levels are tightly linked to the physiology of the cell (1). The control of physiological processes such as cell division or genome replication with artificial gene circuits therefore is of great interest for various applications in synthetic biology. For instance, growth control circuits could be used to switch between different growth modes to optimize metabolic processes. As a recent example, Wiktor et al. controlled the growth of E. coli bacteria by preventing replication initiation of the protein DnaA using the CRISPR/dCas9 interference mechanism (CRISPRi) (2). To this end, the authors blocked the DnaA boxes at the oriC using sgRNAs of the appropriate sequence and the non-cleaving mutant of the CRISPRassociated protein 9 (dCas9) (3).
A different approach is to alter the growth characteristics of E. coli bacteria by inhibiting the expression of the bacterial tubulin homolog FtsZ, which is the key player of the bacterial divisome (4). It has been shown that the ratio of FtsZ to FtsA is a crucial parameter for the initiation of fission ring formation (5). An imbalance of FtsZ and FtsA results in cell division arrest and bacterial filaments that contain multiple copies of the bacterial chromosome (6). Recently, Elhadi et al. (7) and others (8) (9) have shown that CRISPRi can be used to reduce and restore FtsZ and MreB levels and thus change the morphology of E. coli cells. Filamentation also occurs naturally under a wide variety of conditions and is surmised to have important biological functions (10,11). In this context, Sánchez-Gorostiaga et al. recently determined the physiology and recovery from FtsZ depletion with an IPTG inducible strain (12). In the present work, we reversibly induce filamentation in E. coli by targeting FtsZ using the CRISPR/dCas9 interference mechanism combined with an antisense RNA strategy. We first switch bacteria into the filamentous cell growth mode and subsequently reverse this process, such that the cells return into a normal growth phenotype. We quantify the switching process and dynamically control the system by the inducible expression of antisense RNAs that are complementary to the sgRNAs engaged in ftsZ knockdown in comparison to the removal of the CRISPRi inducers alone. We use single-cell fluorescence microscopy experiments from which we derive bacterial length distributions (13), growth and division rates, as well as reporter gene expression levels. This allows us to identify factors that affect the switching process. In particular, we find that a subpopulation with low dCas9/sgRNA expression levels can be actively switched back to normal division by the induction of antisense RNAs ('anti-sgRNAs'). The anti-sgRNA supported recovery process is found to occur considerably faster than switching in the absence of the antisense RNAs. In order to reversibly switch bacteria to filamentous growth, we first disturbed the FtsZ/FtsA ratio through CRISPR interference using appropriate sgRNAs (Fig 1A). We then restored cell division using different strategies, including the use of antisense sgRNAs ( Fig 1B). Three of at least seven known promoters for ftsZ (ftsZ2p, ftsZ3p, ftsZ4p) lie within the ftsA coding region (14). We targeted these three promoters for transcription blockage using sgRNAs with the corresponding sequences (cf. Fig 1 and S1 Table). The sgRNAs are induced by IPTG via T7 RNA polymerase and were encoded on a plasmid together with TetR-controlled dCas9 and mVenus reporter protein (the 'CRISPRi plasmid'). In order to inhibit ftsZ repression inducibly, we use an anti-sgRNA plasmid, from which sequences complementary to the three sgRNAs could be transcribed. The anti-sgRNAs were designed to absorb the sgRNAs via duplex formation in a similar manner as previously demonstrated by Lee et al. (15). In our approach, the anti-sgRNAs are induced by AHL (the acyl homoserine lactone (HSL) 3-oxo-C6-HSL) from pLux promoters (cf. S2 Table). Both plasmids contain sgRNA 'sponge' elements (vide infra), which are decoybinding sites for dCas9-sgRNA. RNA polymerase is inducible with IPTG. The 'anti-sgRNA plasmid' codes for anti-sgRNAs under the control of an AHL-inducible promoter. The sponge elements on the plasmids act as decoy binding sites for the corresponding dCas9-sgRNA complexes.

Results and Discussion
Decoy-binding sites to buffer CRISPRi. In contrast to tunable CRISPRi (tCRISPRi) with an inducible chromosome-integrated dCas9 (9), we here used plasmid-encoded dCas9 and decoy-binding sites ('sponges') to set the threshold for dCas9-sgRNA above which ftsZ is regulated down. Thresholding was necessary to buffer away dCas9 in the presence of leaky expression of sgRNA from the T7 promoter. Transformation of bacteria with the CRISPRi plasmid alone resulted in filamentous cells (cf. Fig 2A). We therefore further increased the threshold for CRISPRi knockdown of ftsZ by introducing additional sponge elements also on the anti-sgRNA plasmid, upon which bacteria containing both plasmids exhibited the intended normal growth morphology (cf. Fig 2B & S1 Movie). Hence, sponge elements were required at a high copy number (anti-sgRNA plasmid ∼500-700 per cell (16)), which is much higher than the copy number of the plasmid carrying the CRISPRi machinery (i.e., 20 ± 10 (17,18)). Expression of dCas9 in the absence of sgRNA or with sgRNAs of different sequence does not lead to filamentous growth under our experimental conditions. We can therefore rule out that dCas9 alone was responsible for the observed changes in cell morphology, as was reported previously for high level expression of the protein (19). Reversal of CRISPR interference using antisense sgRNA. In order to test the effect of anti-sgRNA on CRISPR interference, we initially performed a series of in vitro experiments in bacterial cell extract (20). Here we regulated the expression of mVenus using purified dCas9 together with sgRNA and anti-sgRNA. As shown in Fig 3B, mVenus expression is efficiently suppressed in the presence of dCas9 and sgRNA (S3 Table for the corresponding sequences), while the addition of anti-sgRNAs recovers the production of mVenus. We tested two different designs for the anti-sgRNAs, which were complementary to part of the sgRNAs. Our experiments confirmed, that sequestration of the spacer region and only part of the dCas9 handle by anti-sgRNAs is already sufficient to de-activate the CRISPRi mechanism (Fig 3A-C). In the in vivo experiments, both anti-sgRNA designs were used together. In a further control experiment, we checked the FtsZ levels by immunoblotting. CRISPRi efficiently decreases FtsZ expression in E. coli, while anti-sgRNA induction recovers the FtsZ concentration despite the presence of inducers (Fig 3D).

Single-cell analysis of filamentation.
After induction of the CRISPRi mechanism with aTc and IPTG, bacterial cells rapidly stopped to divide and started the expression of mVenus (which also was under the control of a pLTetO promoter). We performed timelapse video microscopy studies (cf. Materials and Methods) to observe over time the growth and fluorescence of individual filamentous bacteria inside microfluidic trap chambers that are connected to fresh medium supply channels (S1 Fig). The cell length distribution of ftsZ-knockdown bacteria broadens and shifts towards greater lengths ( Fig  4A). The mean cell length increases from <L>=3 µm to <L>=21 µm in three hours with 500 µM IPTG and 107 nM aTc (which we defined as the 100 % induction level). As found by Li et al. (9), at low induction levels there is a co-existence of subpopulations of normal growing cells and filamentous ones. Upon 100 % induction of the CRISPRi mechanism, we were able to switch to filamentous cell growth (with a division rate approaching zero, cf. inset of Fig 4A). As shown in Fig

Efficiency of restoration of normal cell division.
We investigated three different strategies to revert the filamentous cells to normal growth: i) either passively by stopping the production of dCas9/sgRNAs via removal of the inducers aTc and IPTG, ii) by induction of the generation of anti-sgRNAs actively counteracting the CRISPRi mechanism (the 'antisense' strategy), or iii) by a combination of both methods (the 'active' approach) ( Fig 5A). Interestingly, the three strategies performed differently in different environments. In bulk experiments with bacterial cultures in flasks, all three strategies allowed us to switch the bacteria back to normal growth. As shown in Fig 5B for the passive and the antisense strategy, the cell length distribution shifted from about <L> = 10 µm to <L> = 4 µm within a few hours. A flow cytometer experiment performed for the active strategy showed that upon IPTG/aTc induction about 93 % of the cells switched to higher mVenus fluorescence intensity and higher forward scatter (FSC) signal (Fig 5C & S2 Fig). After supplementation of AHL (60nM for 3 hours) the non-filamentous population increased again from 6 % to 28 %. The used AHL concentration did not affect growth as shown in a control experiment (S3 Fig).
In contrast to the bulk experiments cells that were induced to grow filamentous inside microfluidic chambers usually did not re-initiate cell division under strategy ii), i.e., upon supplementation of AHL and in the presence of aTc and IPTG (S4 Fig) (the only exception, where strategy ii) was successful in microfluidics, is shown in S3 Movie). Only the removal of the CRISPRi inducers allowed a fraction of about 5 % of > 2500 analyzed cells to resume normal cell division (S4 Movie). Strategy ii) requires the sequestration of sgRNAs by anti-sgRNAs in the presence of dCas9 -thus dCas9 and anti-sgRNA will compete for binding to the sgRNAs present. The antisense strategy therefore will work better for lower dCas9 concentration and, accordingly, higher anti-sgRNA concentration. As these components are expressed from two different plasmids with different copy numbers, there may be large variations in the anti-sgRNA/dCas9-sgRNA ratio within a population (cf. also next section). The difference between the bulk and microfluidics experiments therefore can be interpreted as the result of simple statistics. In all three strategies, restoration of normal growth occurred within a few hours, which is however slower than in studies, where ftsZ was directly put under the control of an inducible promoter (12).

Transient phenotypic heterogeneity.
Our experiments in microfluidic cell traps revealed a considerable phenotypic heterogeneity upon induction of filamentation, which is reversed once normal growth is restored. In addition to their size, we characterized the cells with respect to their fluorescence intensity (Fig 6A & B). After 2 hours of continuous induction with IPTG/aTc, the coefficient of variation (CV=standard deviation/mean) of the cell size as well as of mVenus and mRFP expression increased considerably (Fig 6A). In contrast to the aTc-induced mVenus (CRISPRi plasmid), the constitutively expressed mRFP (anti-sgRNA plasmid) level dropped during filamentation (Fig S5). The decrease in mRFP fluorescence intensity could be caused by a dilution of the plasmid, a reduced protein expression rate or both. The same holds for the increasing CVs, which presumably are caused by greater variations in the plasmid copy numbers in the larger cells and/or in protein expression strength. Although both plasmid copy number controls are based on a negative feedback mechanism that measures the concentration of the plasmids in the cells (21), it has been shown that the copy number can change considerably depending on the bacterial growth rate (22). After removal of the inducers, a new population of normal growing cells restored similar CV values as the starting population.

Inducible restoration of normal cell growth in a subpopulation.
In contrast to the passive switching strategy described above (Fig 5D & S5), active switching according to strategy iii) switches back a different subpopulation of the filamentous bacteria (Fig 6B &  S6). This subpopulation is characterized by low mVenus and mRFP fluorescence levels (Fig 7A & B) and a large cell size (Fig 6B). The mean mVenus fluorescence of a fast responding bacteria is 0.2 times lower than the average of the main population. Low mVenus fluorescence should correlate with low dCas9 concentration (mVenus and dCas9 on the CRISPRi plasmid are both under TetR-control), which suggests that only in this case the expression of anti-sgRNAs can restore FtsZ levels enough to restart cell division. Re-initiation of AHL supported cell division on average started about 80 minutes after removal of IPTG/aTc, which is ≈ 180 minutes faster than in the absence of AHL (Fig 5D  & 7C). This means that in the confined space of a microfluidic growth chamber actively switched cells can take over the population before the other subpopulation (only responding to IPTG/aTC removal) switches back.

Conclusion
Using CRISPR interference and anti-sgRNA elements, we implemented a synthetic gene circuit that enabled us to controllably switch cell division off and on. Cells could thus be switched from normal cell division into a filamentous state, and also back to normal cell division. Due to the critical role of cell division in the bacterial life cycle, our circuit strongly interferes with the physiology of the bacterial host chassis. In order to be able to implement the circuit in viable bacteria at all, decoy-binding sites for the FtsZ-suppressing dCas9-sgRNA complexes had to be introduced as genetic buffer elements. Next to these static sgRNA sponges, we also utilized inducible antisense-sgRNA molecules that could be used to adjust the threshold for bacterial filamentation and thus modulated the switching process. An important aspect of a gene circuit that interferes with bacterial growth is that it very directly alters bacterial fitness. As a consequence, the circuit creates distinct dynamics both at the cellular and at the population level, which furthermore depends on the specific environment in which the bacteria grow. Using microfluidic single cell experiments, we observed that in this context only a fraction of the bacteria displays the desired switching behavior. However, these correctly switching bacteria can quickly take over the population, whereas others are expelled from the microfluidic chambers or die. Thus, on the population level the circuit appears to behave as designed. In this context, it is interesting to note that upon induction of filamentation, the heterogeneity of our bacterial population appeared to increase considerably, and, depending on the switching strategy, different subpopulations of this population were selected. One interpretation of this observation is that the displayed heterogeneity is part of a bacterial strategy to survive the conditions imposed upon them by the synthetic circuit (23,24). Even though the response of the bacteria to a gene circuit directly affecting bacterial division or growth thus appears to be complex and quite problematic from an engineering perspective, achieving control over physiological processes is still highly desirable. The conflict of interest between the engineer's agenda to produce a metabolite and the bacterial agenda to replicate (25), could be solved, in principle, by establishing molecular programs that switch between different cellular growth modes allowing an appropriate shift in resources (1,26,27). Such programs will inevitably have to take into account similar effects of bacterial heterogeneity and population dynamics as those described here.

Materials and Methods
Plasmids. We constructed plasmids by Gibson assembly of synthetic DNA fragments into the target vector (pSB1K3 and CRISPRi plasmid 44249 from addgene). The final plasmid sequences can be found in S2 Table. The sender strain was constructed in an earlier study and contains the gene for LuxI synthase (BioBrick part BBa_C0261). The sgRNA, anti-sgRNA and sponge element sequences are listed in S1 Table. Bacterial cell culture. Experiments with filamentous cells were performed in Escherichia coli BL21 (DE3) pLysS. Cells from glycerol stock were grown in 5 ml Luria-Bertani medium containing antibiotics selecting for both plasmids (CRISPRi and asgRNA plasmid) and incubated over night at 37 °C and 250 rpm. The following day, cells were diluted 1:100 and incubated for additional 2 h. Optical density (OD) values between 0.4-0.6 were obtained. From this batch, 1 ml of the culture was centrifuged and the pellet resuspended in 300 µl growth medium. The concentrated cells were immediately loaded on a microfluidic chamber until single or few bacteria were captured in the traps. In such a microfluidic chemostat with defined bacterial trap dimensions, we supplied the bacterial suspension constantly with fresh nutrients (LB medium, antibiotics and/or inducer chemicals or dyes) using a pressure flow controller (OB-1K, Elveflow).

Fluorescence time-lapse microscopy. The microfluidic PDMS (Sylgard 182, Dow
Corning) device was fabricated using standard soft lithography as previously described (28). The microfluidic device is a combination of a gradient mixer (29) and bacterial traps designed with dimensions of 200 µm x 10-50 µm x 1 µm as a H-shaped chemostat (S1 Fig). Time-lapse microscopy measurements were conducted on a Nikon Ti-Eclipse epifluorescence microscope controlled with NIS-Elements Imaging Software. The microscope was equipped with a sCMOS camera (Zyla, Andor), an automated x-y-stage (Prior Scientific, Cambridge, UK) and an incubator box (Okolab) to maintain an operation temperature of 37° C. All videos were recorded with 40x apochromatic magnification objectives. Every 5 to 20 min, images in phase contrast mode, YFP as well as RFP fluorescence mode (in combination with the appropriate filter sets) were taken for a total run time of up to 20 hours. The exposure times were automatically adjusted.

SUPPORTING INFORMATION
Additional movies, figures and tables based on descriptions in the text. Figures and tables are provided in a single "SI.pdf" file. S1 Movie. This video shows E. coli (with the CRISPRi and anti-sgRNA plasmids) in a microfluidic chamber without inducers of the CRISPRI mechanism. The images are an overlay of BF/phase contrast and fluorescence channels of mVenus and mRFP. Time is shown as hh:mm.
S2 Movie. This video shows filamentous growth of E. coli in microfluidic chambers upon induction with 500 µM IPTG and 107 nM aTc (100 % level). The images are an overlay of BF/phase contrast and fluorescence channels of mVenus and mRFP. Time is shown as hh:mm. S3 Movie. Active switching in microfluidic chambers. Filamentous growth is induced (215 µM IPTG and 46 nM aTc) for 2 hours. From there on the freshly supplied medium does not contain IPTG and aTc, but is supplemented with 50 nM AHL. The video starts after 1 hour of induction. One cell starts to re-divide about 50 minutes after the medium change. The images are an overlay of BF/phase contrast and fluorescence channels of mVenus and mRFP. The time is shown in hh:mm. S4 Movie. Passive switching in microfluidic chambers. Filamentous growth is induced (215 µM IPTG and 46 nM aTc) for 2 hours. After the time window, the freshly supplied medium is without inducers. The images are an overlay of BF/phase contrast and fluorescence channels of mVenus and mRFP. This video shows a bacterium that has a relatively low growth rate during induction and takes relatively long to start re-division. The time is shown in hh:mm. The mRFP (top panel) and the SSC signal (bottom panel) are plotted against the FSC signal. mRFP is constitutively expressed in initially normal cells but also after induction of filamentation. After switching actively back with AHL in the presence of filamentation inducers IPTG/aTc for 3h, a population with decreased mRFP signal arises, which finally represents the main population (after 7h of asgRNA induction). In the filamentous state, the SSC signal ramps up in proportion with the FSC signal (t=3h with aTc/IPTG). However, rescued cells quickly recover and the majority of the cells turn back to the initial scatter plot position as measured with normal growing and dividing bacteria (cf. not induced and after t=7h with aTc/IPTG/AHL). The images are an overlay of bright field (phase contrast) and the fluorescence channels of mVenus and mRFP. During the first 1 hour and 20 minutes the cells are exposed to 215 µM IPTG and 46 nM aTc (43%). After 1 hour, 60 nM of AHL is supplemented to the growth medium. One bacterium resumes cell division (marked by an arrow), which belongs to a persisting subpopulation. The time is shown in hours. Scale bar: 20 µm.  Table. Sequences of target sites, sponge elements, sgRNAs and anti-sgRNA. DNA sequences of the target sites of ftsA (Gene ID: 944778)) and the derived elements employed in this study.