The strains and plasmids used in this study are listed in Supplementary Table 1. All strains were derived from E. coli W3110 (Serra et al., 2013b). Cells were grown in tryptone broth (TB) medium (10 g tryptone, 5 g NaCl per litre) supplemented with antibiotics, where necessary. Gene deletions were obtained via PCR-based inactivation of chromosomal genes (Datsenko and Wanner, 2000) or using P1 transduction (Miller, 1972). Km^r cassettes were eliminated via FLP recombination (Cherepanov and Wackernagel, 1995). For the construction of pOB2 and pVM42, mCherry and egfp genes were amplified by PCR using a forward primer containing the artificial strong ribosome binding site (RBS) ACAACTTAAGGAGGTATTC (Salis et al., 2009) and cloned into the pTrc99a vector with KpnI/HindIII.

For the construction of transcriptional genomic reporters derivatives of plasmid pKD13 (Datsenko and Wanner, 2000) were generated. For this purpose either Gibson assembly or circular polymerase extension cloning (Quan and Tian, 2009) were applied. All fragments including linearized vector with kanamycin cassette flanked by FLP recombination target sites, RBS, fluorescent protein and gene sites for insertion were first amplified using Phusion Polymerase and then assembled into one circular DNA fragment in a single reaction. In this way, constructs with an additional RBS (for transcriptional reporters), mCherry or superfolder GFP (sfGFP) genes and Km^r cassette flanked by the homologous regions for genomic insertion were created. The lambda Red recombination system (Datsenko and Wanner, 2000) was used to introduce fluorescent reporters into native chromosomal loci via homologous recombination. Positive clones were checked by PCR for the correct insertion using specific primers and subsequently sequenced. After Km^r cassettes were eliminated via FLP recombination (Cherepanov and Wackernagel, 1995), a second reporter was introduced in its native chromosomal locus.

For the analysis of E. coli biofilm formation, overnight cultures grown in TB in a rotary shaker at 30°C were diluted 1:100 into fresh TB medium and grown at 200 rpm to the mid-exponential phase (OD₆₀₀ = 0.6) at 30°C. The samples were further diluted to an OD₆₀₀ of 0.05 in TB medium and 400 μl of this suspension was seeded per well into the 8-well microscope slides with untreated surfaces (μ-Slide, 8-well untreated bottom, ibidi GmbH, Germany). TB was supplemented with 100 μg/ml ampicillin (Amp) and 10 μM isopropyl-β-D-thiogalactopyranosid (IPTG) when cells carried either plasmid pVM42 encoding egfp or pOB2 encoding mCherry for visualization of all cells in the biofilm. Biofilms were grown at 30°C without shaking for the indicated time. Gene expression and/or structure of undisrupted biofilms were visualized using confocal microscopy. In parallel, biofilms were separated into pellicle, supernatant and surface-attached cell fractions and subjected to flow cytometric single cell analysis.

In addition, biofilm formation on microtiter plates was quantified using crystal violet (CV) assays (O’Toole, 2011). For this purpose, overnight cultures were diluted 1:100 in fresh TB medium, regrown until OD₆₀₀ of 0.6, and adjusted to OD₆₀₀ 0.05. Three hundred microliter of this suspension was then loaded into a 96-well plate (Corning Costar, flat bottom, Sigma Aldrich, Germany). After 24 h of incubation at 30°C, the OD₆₀₀ of samples was measured, non-attached cells were removed from wells and wells were intensively washed with water. Cells bound to the polysterene plastic walls were subsequently stained with 1% CV solution, 300 μl per well for 20 min at room temperature. Then, CV was removed, wells were washed three times with water and then the plate was left to dry for 1 h. Remaining CV stain in attached cells was solubilized by adding 300 μl of 96% ethanol and measured at OD₅₉₅. Finally, all CV values were normalized by the respective OD₆₀₀.

Cells were collected from different biofilm fractions at the indicated time points of biofilm growth, washed two times and resuspended in phosphate-buffered saline (PBS) (8 g NaCl, 0.2 g KCl, 1.44 g Na₂HPO₄, 0.24 g KH₂PO₄) prior to flow cytometric analysis using BD LSRFortessa SORP cell analyzer (BD Biosciences, Germany). The pellicle fraction was collected by taking the thin film layer from the liquid/air interface using a pipette and resuspended in 1 ml PBS. After pellicle removal, 100 μl of culture medium was collected as the supernatant fraction. The rest of the supernatant was discarded and the remaining surface-attached cells were washed two times with PBS. Subsequently, surface-attached cells were scraped from the well surface using a 1 ml pipette tip and vigorously vortexed before the measurements to disrupt cell aggregates. The completeness of aggregate disruption was confirmed by flow cytometry, using fortward scatter (FSC) and side scatter (SSC) parameters. Data were analyzed using FlowJo software version 10.1r5 (FlowJo LLC, Ashland, OR, USA), applying a software-defined background fluorescence subtraction.

Escherichia coli intact biofilms were visualized using an inverted Zeiss Axio Observer Laser Scanning Microscope (LSM) 880 equipped with a C-APOCHROMAT 40x/1.2 Water Corr-UV-VIS-IR objective. The following lasers were used: 458 nm, 488 nm, 514 nm Argon laser and 561 nm DPSS laser. Each experiment was performed at least three times on independent biological replicates. Obtained images and z-stack projections were visualized using Zeiss ZEN System imaging software (Zeiss). Thickness of pellicles was measured using an upright Leica TCS SP5 confocal laser scanning microscope equipped with HC PL FLUOTAR 10x/0.3 dry objective.

Quantitative image analysis was performed using Matlab (Mathworks) in several steps. First, image stacks showing constitutive fluorescence signal were up-sampled along the z-axis to obtain equal voxel side lengths. Second, noise was reduced by convolution with a small averaging filter kernel. Third, segmentation was performed by 3-level thresholding according to Otsu’s Method (Otsu, 1979) where class one was assigned to background, while class two and three where treated as foreground. Finally, the obtained binary image stack was dissected into cubes (side length: 5 px, approximately 1 μm). In cases where no constitutive fluorescence marker was available, the image stacks of each fluorescence channel were segmented individually and the volumetric data was merged before dissection (see Supplementary Figure 8). For each dissected volume the underlying unprocessed fluorescence intensities of GFP and mCherry were measured. In addition, the local cell density was determined by measuring the occupied volume inside a shell of 3 μm radius around the center of mass of each cube. Paraview was used for 3D-visualization (Ahrens et al., 2005; Flandin, 2015).

As a model to investigate changes in gene expression during formation of E. coli biofilms, we used the biofilm assay in microtiter plates under static conditions (i.e., without flow) at 30°C. Already after 6 h, and even more evident after 24 h, submerged biofilm structures of 50-60 μm thickness could be observed in this assay on the bottom of the well, and a pellicle of 10-15 μm thickness was formed at the liquid-air interface (Figure 1A). Taking into account the heterogeneity of such a static biofilm system, for our subsequent gene expression analysis using flow cytometry, we separated three physiologically different fractions: cells within the pellicle, cells attached to the surface of the well, and the individual cells and small aggregates that remain outside of these structures (termed “supernatant” in this article) (see Materials and Methods for details).

Confirming previous observations (Pratt and Kolter, 1998; Vidal et al., 1998), we observed that formation of biofilms under these conditions requires curli fibers and flagella, since deletion of major subunits of either curli (ΔcsgA) or flagella (ΔfliC) filaments led to large decrease of the biomass of submerged biofilms as quantified by CV staining (Figure 1B). Similarly, pellicle formation was completely abolished, as quantified by measuring cell density at the liquid-air interface using flow cytometry (Figure 1C).

Confocal laser scanning microscopy further revealed that both curli fibers and flagella are required for the 3D structure formation (Figure 1D). Quantitative analysis of the microscopy images (see Materials and Methods for details) confirmed that both ΔcsgA and ΔfliC strains do not form mature 3D structures, but display a uniform surface roughness profile in contrast to the wild type (Figures 1E,F). In order to further distinguish cells associated with structures and aggregates from single cells, we calculated the local cell density within the microscopy images (Figure 1G). The local density profile for the wild-type strain differed significantly from both mutant strains (Figure 1H). Whereas the wild type shows regions of high local density up to 100 μm above the surface, which correspond to cell aggregates and/or microcolonies that are visible in confocal images (Figure 1D; Supplementary Figure 1A), both ΔcsgA and ΔfliC strains show a high density only at the bottom of the well, which is likely to represent sedimented individual cells or cells being attached to the surface.

Notably, both curli and flagella also provide structure to the macrocolony biofilms grown on the soft agar (Serra et al., 2013a,b), suggesting similarity between these two very different types of multicellular communities formed by E. coli. Among other tested potential matrix components or adhesins (Korea et al., 2010), only a lack of fimbriae showed more than 50% reduction in biofilm formation under our conditions, as quantified by CV staining (Supplementary Figure 1B). Interestingly, this defect was only observed for CV staining of the biomass but not in microscopy images (Supplementary Figure 1C), indicating that fimbriae are required for surface attachment but not for structure formation.

We next analyzed the expression pattern of curli and flagellar genes during transition towards the biofilm growth. For this purpose we constructed transcriptional reporters by integrating genes encoding superfolder GFP (sfGFP) or mCherry downstream of chromosomal csgA and fliC, respectively, and performed quantitative single cell analysis of gene expression in different biofilm fractions (Figures 2A–C) using flow cytometry. PcsgA-sfGFP expression could be detected 12 h after planktonic cells were seeded onto microtiter plates and increased with time; its mean levels were significantly higher in the pellicle and the surface-attached cells (P < 0.0001 according to unpaired t-test), compared to the supernatant fraction (Figures 2A,B). Notably, PcsgA-sfGFP expression was bimodal in all fractions (Figure 2B) and the increase in the mean fluorescence during biofilm development primarily occurred through an increase in the fraction of curli-ON cells (Figure 2C). Note that residual populations of curli-ON cells in the supernatant and of curli-OFF cells in the pellicle and surface-attached biofilm structures might be at least partly due to an imperfect separation of different fractions in our assay.

Confocal microscopy of intact 3D structures further confirmed bimodal expression of curli fibers in submerged biofilms, with high levels of PcsgA-sfGFP being mostly confined in cell aggregates/microcolonies and little or no expression in other cells within the biofilm (Figure 2D). Consistently, quantitative image analysis (see Figure 1G and Material and Methods for details) showed high curli expression in regions of higher local density (Figure 2E). As already observed in flow cytometry data for the supernatant fraction (Figure 2B), the majority of cells in the regions of low local density showed little or no GFP expression and thus most probably represent individual curli-OFF cells. Notably, the expression profile of PcsgA-sfGFP along the height of the submerged biofilm showed that the majority of cells expressing curli are found in a distinct region up to 40 μm from the bottom of the well (Figure 2F), which contains the majority of cell structures/aggregates (Supplementary Figure 1A).

An opposite expression pattern was observed for PfliC-mCherry. After 12 h of incubation, the mean expression of flagella was decreasing over time in all three culture fractions, with highest expression being measured in the supernatant fraction (Figures 3A,B). Overall, this decrease in flagellin expression was monomodal and gradual over time (Figures 3A,C). Confocal microscopy showed that PfliC-mCherry was primarily expressed in single cells in the supernatant or at the surface, but not associated with structures (Figure 3D). Quantitative analysis of these images further confirmed that flagellin expression was mostly confined to the regions with low local densities (Figure 3E). The expression profile along the height of the submerged biofilm also revealed that the majority of cells expressing flagellin was above 40 μm from the bottom of the well (Figure 3F), i.e., in regions where mostly single cells are found (Supplementary Figure 1A).

To directly investigate the interplay between expression of curli and flagellar genes at a single-cell level, we created a double reporter strain that expresses both PcsgA-sfGFP and PfliC-mCherry. Its analysis confirmed mutually exclusive expression of curli and flagella in all cell fractions and at all measured time points (Figures 4A–C and Supplementary Figure 2), with higher curli expression in the pellicle and in the surface-attached biofilm and with higher flagella expression in the supernatant fraction. Consistently, such mutually exclusive expression could also be seen (Figure 4D) for individual cells in intact submerged biofilms. This was confirmed by quantitative image analysis (Figure 4E) that clearly showed strictly exclusive expression of both reporters (Figure 4F).

To better understand the spatio-temporal pattern of curli expression, we studied the regulation of the stationary phase sigma factor, σ^S (Hengge-Aronis, 2002; Pesavento et al., 2008) during biofilm formation, using a translational fusion of mCherry to the σ^S-regulated gene osmY as a reporter. The expression pattern of the σ^S reporter resembled that of curli, with a gradual increase over time and significantly higher expression in the pellicle and surface-attached biofilm fractions (P < 0.0001 according to an unpaired t-test) after 48 h of biofilm growth (Figure 5A). The expression of OsmY-mCherry was also bimodal in all biofilm fractions at 24 h of biofilm growth or later (Figures 5B,C). Furthermore, in all culture fractions we observed a strong linear correlation between osmY and curli expression in a subpopulation of cells (Figures 5D–F; Supplementary Figure 3). Nevertheless, we also detected a subpopulation of cells that express OsmY-mCherry but not PcsgA-sfGFP. Most interestingly, a subpopulation of cells – particularly visible in the supernatant fraction at 24 h and in all fractions at 12 h – transiently expressed curli while having low σ^S activity, which indicates that curli expression switches to the ON state before full activation of σ^S. All of these demonstrate that σ^S activity is only partly correlated with curli expression. Similarly, in biofilm images we observed curli expression specifically in biofilm structures and aggregates, whereas OsmY-mCherry was detected both in structures and in single cells (Figures 5G,H, Supplementary Figure 4). Quantitative image analysis (Figure 5H) confirmed partial correlation between curli expression and σ^S activity, with a subpopulation of cells showing high σ^S activity but low curli expression (Figure 5I). In summary, this analysis indicates that at least three distinct populations arise in submerged biofilms after 24 h of incubation: cells producing flagella, stationary cells producing curli and stationary cells not producing curli.

Finally, we investigated the overall activity of the housekeeping sigma factor σ^D as well as the growth rate within E. coli biofilms in our open static culture system. A σ^D-dependent transcriptional reporter PrplL-mCherry (rplL encodes the 50S ribosomal subunit L12) showed steady monomodal expression, with no pronounced change over time or difference between the biofilm fractions (Figures 6A–C). This confirms that housekeeping genes are similarly expressed in all cells across the population. Consistent with this observation, we found little correlation between the expression of PrplL-mCherry and PcsgA-sfGFP (Figures 6D–F, Supplementary Figure 5). The same general pattern was observed within undisrupted submerged biofilms, with all cells expressing PrplL-mCherry irrespective of their association with the biofilm structures (Figures 6G–I). Thus there appears to be no significant correlation between curli expression and biosynthetic state of the cell.

A different pattern was observed for the single-cell growth rates. To monitor cell growth status in different biofilm fractions over time, we used a fluorescent growth rate reporter pOB44 (Claudi et al., 2014), which is based on a DsRed S197T variant called TIMER that spontaneously changes fluorescence from green to green/orange (Terskikh et al., 2000). Because green TIMER molecules have shorter maturation time than orange TIMER molecules, fast-growing cells that dilute both molecules with each cell division emerge green. In contrast, slowly growing cells accumulate both rapidly and slowly maturing TIMER molecules and appear green/orange, so that green/orange fluorescent ratios provide a readout for the growth rate. During biofilm development, we observed a stable green fluorescence and a gradual increase in red fluorescence in all fractions, with slightly higher values in the pellicle and in surface-attached cells (Figures 7A,B). A distinct subpopulation of slowly dividing cells apparently arises in the pellicle and surface-attached fractions already after 6 h of biofilm growth (Supplementary Figure 6) and later also in the supernatant fraction, gradually increasing with time (Figures 7C–E, Supplementary Figure 6). In biofilm structures, slowly proliferating cells were present almost exclusively in larger aggregates already at 24 h of growth and much more prominently at 72 h of growth (Figure 7F, Supplementary Figure 7). Importantly, rapidly dividing green cells were observed beneath and between aggregates of slowly dividing cells even at later time points. These observations were confirmed by quantification of the orange/green TIMER ratio, which showed both an overall increase throughout the growth of submerged biofilm and a particularly high level in a defined subpopulation of cells in the regions with highest local density (Figures 7G–I).

Although E. coli K-12 normally does not produce cellulose, expression of cellulose genes in this strain can be reactivated (Serra et al., 2013a). We observed that the corresponding cellulose-producing strain AR3110 formed submerged biofilms and pellicles with the biomass comparable to that of W3110 (Figure 8A), showing that presence of cellulose as a matrix component does not generally enhance the biofilm formation. Similarly, deletion of either curli (ΔcsgA) or flagella (ΔfliC) led to a decrease of biofilm and pellicle formation (Figures 8A,B) as well as of 3D structure (Figures 8C–F) in the AR3110 background, demonstrating that the presence of cellulose cannot compensate for the loss of curli matrix. The expression of PcsgA-sfGFP within biofilm structures of the AR3110 strain was again confined to dense cell clumps and aggregates (Figure 8G). Nevertheless, biofilms formed by the AR3110 strain were more compact and had larger thickness than those of the W3110 strain (compare Figure 8 with Figure 1), demonstrating that cellulose does contribute to the overall structure of the submerged static biofilms.