Regulation by cyclic di-GMP attenuates dynamics and enhances robustness of bimodal curli gene activation in Escherichia coli

Curli amyloid fibers are a major constituent of the extracellular biofilm matrix formed by bacteria of the Enterobacteriaceae family. Within Escherichia coli biofilms, curli gene expression is limited to a subpopulation of bacteria, leading to heterogeneity of extracellular matrix synthesis. Here we show that bimodal activation of curli gene expression also occurs in well-mixed planktonic cultures of E. coli, resulting in all-or-none stochastic differentiation into distinct subpopulations of curli-positive and curli-negative cells at the entry into the stationary phase of growth. Stochastic curli activation in individual E. coli cells could further be observed during continuous growth in a conditioned medium in a microfluidic device, which further revealed that the curli-positive state is only metastable. In agreement with previous reports, regulation of curli gene expression by the second messenger c-di-GMP via two pairs of diguanylate cyclase and phosphodiesterase enzymes, DgcE/PdeH and DgcM/PdeR, modulates the fraction of curli-positive cells. Unexpectedly, removal of this regulatory network does not abolish the bimodality of curli gene expression, although it affects dynamics of activation and increases heterogeneity of expression levels among individual cells. Moreover, the fraction of curli-positive cells within an E. coli population shows stronger dependence on growth conditions in the absence of regulation by DgcE/PdeH and DgcM/PdeR pairs. We thus conclude that, while not required for the emergence of bimodal curli gene expression in E. coli, this c-di-GMP regulatory network attenuates the frequency and dynamics of gene activation and increases its robustness to cellular heterogeneity and environmental variation.


75
In this study we demonstrate that differentiation of E. coli csgBAC operon expression, leading 76 to formation of distinct subpopulations of curli-positive and -negative cells, occurs not only in 77 submerged and macrocolony biofilms but also stochastically in a well-stirred planktonic culture All strains and plasmids used in this study are listed in Table S1. A derivative of E. coli W3110

88
[26] that was engineered to encode a chromosomal transcriptional sfGFP reporter downstream    Macrocolony biofilms were grown as described previously [26]. Briefly, 5 µl of the overnight 123 liquid culture grown at 37°C in LB medium (10 g tryptone, 5 g NaCl, and 5 g yeast extract per 124 liter) was spotted on salt-free LB agar plates supplemented with Congo red (40 µg/ml). Plates

178
In order to investigate whether curli expression was uniform or heterogeneous within 179 planktonic E. coli populations, we next measured curli reporter activity in individual cells using 180 flow cytometry. The reporter was induced only in a fraction of cells, and this bimodality of curli 181 expression became increasingly more pronounced at later stages of culture growth, reaching 182 its maximum in the overnight culture ( Figure 1B). Thus, the bimodal induction of curli gene 183 expression is observed not only in biofilms but also in well-mixed planktonic cultures. Upon

184
. CC-BY 4.0 International license made available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprint this version posted October 2, 2022. ; https://doi.org/10.1101/2022.05.23.493020 doi: bioRxiv preprint inspection of the flask culture by microscopy, no cell aggregation could be observed,

185
suggesting that curli gene induction in a subpopulation of bacteria is not due to the formation 186 of suspended biofilm-like aggregates [39]. While curli activation was more pronounced in a cell 187 culture growing in an orbital shaker in a flask ( Figure 1B), bimodality was also observed for 188 cultures grown in a plate reader ( Figure S1B and Figure S2B). Notably, stimulation of curli 189 expression by SHX, or its suppression by additional nutrients, affected the fraction of positive 190 cells rather than their expression levels ( Figure S1B and Figure S2B).

212
We observed that level of curli gene expression in ∆pdeR ∆dgcM strain was lower than in 213 ∆pdeR strain, which confirms that DgcM can promote curli expression independently of PdeR

218
Most importantly, the bimodality of curli expression was still observed in the quadruple deletion 219 strain ( Figure 2F).

220
. CC-BY 4.0 International license made available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprint this version posted October 2, 2022. ; https://doi.org/10.1101/2022.05.23.493020 doi: bioRxiv preprint We thus conclude that, whereas the known c-di-GMP-dependent regulation of MlrA activity     Figure 2C). Another notable difference between flask and multi-well plate 253 cultures was that the low-fluorescence peak of the wild-type culture was not fully negative but 254 apparently contained a large fraction of cells with incompletely activated curli reporter, which 255 could also be seen in the ∆pdeH or ∆pdeR strains but not in the ∆pdeH ∆dgcE ∆pdeR ∆dgcM, 256 ∆pdeR ∆dgcM or ∆pdeH ∆dgcE strains. Similar results were obtained even upon prolonged 257 . CC-BY 4.0 International license made available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprint this version posted October 2, 2022. ; https://doi.org/10.1101/2022.05.23.493020 doi: bioRxiv preprint incubation in the plate reader ( Figure S4), confirming that the observed difference with the 258 overnight flask culture was not because of the different growth stage but rather due to 259 differences in growth conditions.

260
We further tested reporter activation under growth conditions that favor biofilm formation.       CC-BY 4.0 International license made available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprint this version posted October 2, 2022. ; https://doi.org/10.1101/2022.05.23.493020 doi: bioRxiv preprint

296
In order to investigate the dynamics of curli gene activation, and the impact of c-di-GMP 297 regulation, at the single-cell level, we utilized a "mother machine" device -a microfluidic chip 298 where growth of individual bacterial cell lineages could be followed in a highly parallelized 299 manner over multiple generations [35] ( Figure 5 and Supporting protocols). Since our design 300 of the mother machine allows rapid switching of the medium supplied to the cells ( Figure 5D),

301
we could mimic nutrient depletion in order to activate curli expression in continuously growing 302 single cells. For that, wild-type, individual DGC or PDE, or quadruple deletion strains were first 303 loaded into the mother machine from overnight cultures, allowed to grow in fresh TB medium 304 for several generations, and then shifted to the TB medium that was pre-conditioned by  (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprint this version posted October 2, 2022. ; https://doi.org/10.1101/2022.05.23.493020 doi: bioRxiv preprint expression by c-di-GMP reduces the rate but also the cell-to-cell variability of curli induction 332 within population.

Discussion 335
Expression of the curli biofilm matrix genes is known to be heterogeneous or even bistable in

637
. CC-BY 4.0 International license made available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprint this version posted October 2, 2022. ; https://doi.org/10.1101/2022.05.23.493020 doi: bioRxiv preprint

656
Depending on the pressure set at the bottom inlet, it is possible to select which one of the two media 657 flows into the central supply channel to the mother machine growth sites.

658
. CC-BY 4.0 International license made available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprint this version posted October 2, 2022. ; https://doi.org/10.1101/2022.05.23.493020 doi: bioRxiv preprint

680
. CC-BY 4.0 International license made available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprint this version posted October 2, 2022.

689
. CC-BY 4.0 International license made available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprint this version posted October 2, 2022.

695
. CC-BY 4.0 International license made available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprint this version posted October 2, 2022.

711
. CC-BY 4.0 International license made available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprint this version posted October 2, 2022.

721
. CC-BY 4.0 International license made available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprint this version posted October 2, 2022. ; https://doi.org/10.1101/2022.05.23.493020 doi: bioRxiv preprint

731
. CC-BY 4.0 International license made available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprint this version posted October 2, 2022. ; https://doi.org/10.1101/2022.05.23.493020 doi: bioRxiv preprint

743
. CC-BY 4.0 International license made available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprint this version posted October 2, 2022. ; https://doi.org/10.1101/2022.05.23.493020 doi: bioRxiv preprint

770
The device was produced using a two-layer soft lithography method as described previously 771 [1]. Based on our in-house made design of the channel layout, a 100 mm silicon wafer was 772 produced by e-beam lithography (ConScience, Sweden). The wafer contains the channel 773 layout as a positive relief. The mother machine traps, which are shown in blue in Figure 5A

793
. CC-BY 4.0 International license made available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is  CC-BY 4.0 International license made available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprint this version posted October 2, 2022. ; https://doi.org/10.1101/2022.05.23.493020 doi: bioRxiv preprint connected regions. After instance identification, two rounds of morphological dilation restored 831 each mask to its original size. Finally, a smooth outline for each cell was obtained as a two-832 dimensional spline defined by equidistant knots placed on the mask edge.

833
To track cells between time points, we applied a length conservation strategy for the cells along

840
Since cells may grow in length between time points, we also initialised a growth rate parameter 841 for each cell that biased the expected cell lengths for time point t+1 as a fold-increase in length.

842
To enable adaptation to the true growth rate, the growth rate parameter was updated by a 843 lagging average over 20 time points. To increase robustness to errors in segmentation, we 844 additionally allowed state transitions from one to many and many to one, and built a proposal 845 tree, which branched for all valid assignments lying within the length thresholds. We searched 846 for the proposal with the lowest average fold-change in matched lengths, but limited branching 847 by retaining only the 10 best proposals for subsequent nodes (cell outlines) in the tree. The 848 length thresholds were deliberately set loosely such that the (sum of) cell length(s) at t+1 could 849 decrease at most five-fold or increase at most two-fold relative to the (sum of) cell length(s) at 850 t. This increased the number of valid proposals, but was important in cases where the growth 851 rate estimate was poor. For transitions where one cell outline split into more than two, or 852 transitions where multiple outlines merged into one cell, a new label was generated for the 853 corresponding cells at t+1. We made one exception to this labelling strategy to account for 854 occasional ambiguity in segmentation near division events, where a cell segmented as two 855 sister cells could later be segmented as a single mother cell. Specifically, when two sister cells 856 -i.e., cells that were previously involved in a division event -merged into one, the label was 857 set back to that of the mother; at the next division event, the labels of the sister cells were also 858 retained. Finally, note that any outlines below a minimum size threshold of 50 pixels were 859 ignored. All errors in tracking were manually curated.

860
Cell length was estimated from cell regions as the 'major axis length' of the Matlab regionprops were identified as points for which the second derivative was less than -100 and either the first 872 derivative was zero, or its sign changed at the next time point. In cases where multiple 873 descendants shared a common peak event before branching, we counted that event only once.

874
The half-life of decay after a peak in fluorescence was determined by fitting an exponential 875 decay function with offset term to fluorescence data after the peak. Given the occasional 876 occurrence of a second peak in fluorescence, the decay function was only fit to time points   (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprint this version posted October 2, 2022. ; https://doi.org/10.1101/2022.05.23.493020 doi: bioRxiv preprint