The inducible chemical-genetic fluorescent marker FAST outperforms classical fluorescent proteins in the quantitative reporting of bacterial biofilm dynamics

To increase our understanding of bacterial biofilm complexity, real- time quantitative analyses of the living community functions are required. To reach this goal, accurate fluorescent reporters are needed. In this paper, we used the classical fluorescent genetic reporters of the GFP family and demonstrated their limits in the context of a living biofilm. We showed that fluorescence signal saturated after only a few hours of growth and related this saturation to the reduction of oxygen concentration induced by bacterial consumption. This behaviour prevents the use of GFP-like fluorescent proteins for quantitative measurement in living biofilms. To overcome this limitation, we propose the use of a recently introduced small protein tag, FAST, which is fluorescent in the presence of an exogenously applied fluorogenic dye, enabling to avoid the oxygen sensitivity issue. We compared the ability of FAST to report on biofilm growth with that of GFP and mCherry, and demonstrated the superiority of the FAST:fluorogen probes for investigating dynamics in the complex environment of a living biofilm.

Bacterial biofilms are confined, three-dimensional communities of adherent microbes. They display an outperforming specific lifestyle, which raises a number of fundamental and practical questions drawing interest from various disciplines [1][2][3][4][5][6] . Molecular biology advances have pointed out hundreds of genes that are up-or down-regulated in biofilm-dwelling cells compared to their planktonic counterparts 7,8 . Moreover, no biofilm-specific gene has been identified so far 9 , suggesting that the key properties of biofilms might emerge from the subtle interplay between the unique physico-chemical environment of the cells within the biofilm and the elicited biological responses. In addition, a central feature of biofilms is their structural, chemical and biological heterogeneity at the microscale 10,11 , a property reflexively interrelated with the multiple gradients of nutrients, metabolites, oxygen and others molecules or ions present in a biofilm. Therefore, biofilm description relying on data averaging on the whole population is blurred and the clarification of the mechanisms behind the development of thriving adherent bacterial communities requires spatially-resolved approaches. Thus, biofilm quantitative descriptors of both gene expression and physico-chemical parameters should be probed with a spatial resolution allowing to take into account the biofilm heterogeneity and its implications for the biofilm growth. The advances in genetically encoded fluorescent proteins 12 led to the development of fluorescence microscopy approaches for biofilm imaging 13 . Providing in principle molecular specificity, temporal and spatial resolution in

Spatio-temporal distribution of GFP fluorescence does not linearly report on gene expression.
To quantitatively evaluate the spatio-temporal distribution of the most classical genetic reporter-GFP-in the confined environment of an adherent bacterial community, we grew a biofilm of E. coli cells constitutively expressing the pilus F that promotes biofilm formation and the fluorescent protein GFP (GFPmut3) in a PDMS channel continuously supplied with growth medium at a flow rate of 1 mL/h. The growing biofilm was monitored by collecting bright field and fluorescence images at a frequency of 12 images per hour (1 image every 5 min). As previously shown, the bright field images can be converted in microscopic optical density (µOD) values, which linearly reported cell biomass 29 . GFP fluorescence intensity (I GFP ) and µOD were derived from the image stack reporting the first 24 hours of the biofilm growth. For each image (Fig. 1A), the mean values of intensity were averaged either over the whole field of view -R full -or over smaller regions of interest (ROIs) located on the edge or in the centre of the channel, so called R edge and R centre as shown in Fig. 1B. Our first striking observation was that I GFP and µOD kinetics, which initially overlapped, strongly diverged after 5 hours of growth of the biofilm. While I GFP curve saturated, µOD kept on increasing (Figs 1C and S1). Interestingly, the I GFP kinetics strongly depended, both in shape and intensity, on the localization of the ROI with respect to the channel edge (Figs 1C and S1). To better characterize this behavior, we plotted I GFP as a function of µOD ( Fig. 2) for two ROIs, one at the edge and one in the centre of the channel. We observed that I GFP deviation to linearity occurred for µOD values decreasing by four-fold from the edge to the centre of the channel. Our observations indicated that above a given biofilm development point, the amount of fluorescent proteins synthesized by cells no longer matched the number of cells reported by the µOD, contrarily to what is expected from a protein produced by a constitutively expressed gene. This discrepancy appears more strongly and quickly when the cells are further away from the edge of the channel while, under the hydrodynamical conditions used, no significant difference in biomass amount was measured between the edge and the centre of the channel. This suggested that an environmental parameter asymmetrically distributed in the channel affected the reporter fluorescence.
To better understand the temporal variation of the fluorescence, we measured the evolution of fluorescence in two concentric ROIs of different sizes in the middle of the channel as shown in Fig. 3. The smaller ROI was drawn around an initial cluster of a few cells and the larger one contained a higher number of cells distributed over several clusters within the ROI. Then, we considered the mean fluorescence from each ROI, i.e. the total fluorescence divided by the number of pixels. Interestingly, the fluorescence intensity of the two ROIs displayed very distinct regime changes after about 12 hours of biofilm growth. The signal from the small ROI exhibited a steep fluorescence decrease while the one from the larger ROI remained essentially constant. In the meantime, the µOD signal in the considered ROI increased exponentially, in agreement with the 3D growth of the biofilm. This behaviour indicated that, at the critical time τ c of the fluorescence regime change, the mean fluorescence of single cells decreased while the total fluorescence of the cell population remained constant. In other words, after the time τ c , a dilution phase starts where new fluorescent proteins are no longer produced, the daughter cells sharing then fluorescent proteins synthesized in the initial regime (the production phase). So, the fluorescence of the small ROI decreased due to the spread of the initial cluster lineage beyond the limit of the ROI and the dilution of the

GFP fluorescence in a growing biofilm is strongly impacted by the heterogeneous O2 concentration.
At that stage, we made the hypothesis that the GFP fluorescence in the biofilm was strictly controlled by a threshold level of molecular oxygen below which the final oxidation step of the GFP chromophore maturation did not occur. We hypothesized that, when the biofilm grows and the cell oxygen consumption increases, the level of oxygen in the channel decreases below the threshold required for effective GFP fluorescence maturation. To test this hypothesis, we measured the spatial distribution of oxygen in a mature biofilm after approx. 20 hours of growth using micelles functionalized with fluorescent Ruthenium complex displaying oxygen-dependent fluorescence lifetime.
Biofilms were incubated with Ruthenium micelles and imaged by fluorescence lifetime microscopy ( Fig. 4A) to derive oxygen spatial distribution (Fig. 4B) in the region extending from the edge to the centre of the channel. Control experiment performed in the absence of biofilm exhibited homogeneous O 2 distribution in the whole channel (SI, Fig. S2). Then, we compared the O 2 spatial distribution displayed in the presence of biofilm with that of I GFP and µOD measured under the same conditions ( Fig. 4C,D). We found that I GFP and O 2 spatial distributions displayed similar profiles with both high fluorescence and high oxygen levels close to the edge of channel (Fig. 4B,E) and low levels at the centre of the channel. In the meantime, cell biomass displayed no significant gradient from the edge to the centre of the channel (Fig. 4D,E). The O 2 gradient was consistent with the expected permeability of PDMS to oxygen. This was further confirmed by observing biofilms grown in glass channels with O 2 -impermeable edges, in which a uniform low level of fluorescence was observed in the channel (SI, Fig. S3).
Surprisingly, these results showed that after 20 hours of growth, colonization has expanded evenly over the channel, independently of the O 2 access suggesting that the biofilm growth was not affected by O 2 shortage. However, biomass transport and rearrangements during biofilm development could also account for this smooth topography.
These results emphasize the complexity of interpreting GFP fluorescence in a growing biofilm, due to the strong dependence of the fluorescence intensity on oxygen concentration, itself depending, in the biofilm, on local cell concentration and on the distance to the main oxygen sources in the considered devices, typically close to the edge of the channel for PDMS devices but also close to the channel connectors for devices made of materials with lower permeability to O 2 such as for example glass or polystyrene 30 . Therefore, the fluorescence intensity is inconsistent with the biomass, unlike expected for a reporter under the control of a constitutive promoter. The O 2 requirement for dehydrogenating the α, ß bond of the protein Tyr-66 residue during GFP fluorescence maturation has been pointed out early by Tsien, who indicated that the protein could not be used in obligate anaerobes 27 . Here, we show that GFP fails to fluoresce when the oxygen level of the environment drops below a threshold which is quickly achieved in a growing biofilm. mCherry fluorescence also exhibits a severe non-linearity in a growing biofilm. To generalize our statement, we examined the fluorescence kinetics in a growing biofilm expressing the red fluorescent protein mCherry derived from Discosoma sp. Red (DsRed) fluorescent protein. As for GFP, mCherry expression was under the control of the promoter P R , an E. coli constitutive promoter, allowing to obtain expression that varies as cell biomass. In parallel channels, we initiated the growth of biofilms with cells expressing GFP and mCherry, respectively. We monitored fluorescence and µOD signal kinetics along the biofilm growth and plotted for both reporters the variation of the fluorescence amplitude as a function of the µOD derived from the same images (Fig. 5). We found that mCherry fluorescence also levelled off early in the biofilm development and that this saturation occurred for values of µOD as low as 0.03 while GFP fluorescence remained strictly co-linear up to µOD equal to 0.085. This result was in agreement with the larger sensitivity of mCherry to O 2 shortage due to the two oxidation steps involved in the maturation of mCherry chromophore.
We concluded that fluorescent proteins relying on a O 2 -dependent chromophore maturation could not be used as quantitative tracers in living bacterial biofilms, in which the level of oxygen is low and heterogeneous.
FAST outperforms GFP and mCherry in the complex environment of a living biofilm. To circumvent the oxygen dependence of classical fluorescent proteins, we tested FAST, a new fluorescent marker recently developed for labelling fusion proteins in living cells and organisms. FAST is a protein of 14 kDa engineered from the photoactive yellow protein (PYP) from Halorhodospira halophila to bind and activate the fluorescence of synthetic fluorogenic analogues of hydroxybenzylidene rhodanine (HBR). Because they fluoresce only when immobilized within FAST (because of conformational locking), HBR analogues do not exhibit nonspecific fluorescence in cells, enabling to detect FAST without the need for washing. FAST was shown to be functional in various expression systems including bacteria. As its fluorescence depends only on the interaction of the protein with its cognate fluorogen, FAST is fluorescent -provided that the fluorogen is present -whatever the oxygen level. In addition, an important feature of FAST is the ability to tune its emission colour from green-yellow to orange-red by changing the applied fluorogen, providing an experimental versatility not encountered with GFP-like fluorescent proteins. In a first set of experiments, we built a biofilm with MG1655-gfp-F cells transformed to constitutively express both GFP and FAST (see Materials and Methods). We supplied the fluorogen HBR-3,5-DOM at 2 µM in the biofilm growth medium to form an orange-red fluorescent FAST detected in the 'red' channel (593/40 nm) while GFP fluorescence was collected in the 'green' channel (536/25 nm), having first checked that the presence of the fluorogenic dye did not altered bacteria growth rate (SI, Fig. S4) and that no background fluorescence was induced in the biofilm by the fluorogens in the absence of FAST expression (SI, Fig. S4). The expression of FAST and GFP, monitored in the same biofilm, displayed quite distinct fluorescence kinetics (Fig. 6). GFP signal saturated after approx. 5 hours growth while FAST fluorescence increased exponentially up to 20 hours. In between, µOD signal levelled off after 13 hours of growth as expected due to the saturation of the detection for µOD values above 0.5. Interestingly, within the first 5 hours of the biofilm development, the three signals increased exponentially with the same kinetic constant, 0.39 ± 0.02 h −1 , showing parallel report of biomass and fluorescence over this initial time period (Fig. 6 insert).
We also examined the efficiency of the FAST protein in combination with another fluorogen, HBR-2,5-DM which emits in the green channel. We tested FAST:HBR-2,5-DM against mCherry, using MG1655-F cells transformed to constitutively express a mCherry-FAST fusion (see Materials and Methods). We grew biofilms in the presence of 2 µM HBR-2,5-DM in the medium flow and we collected time-lapse fluorescence and µOD images. mCherry and FAST fluorescence signals (Fig. 7A) diverged early during biofilm development as expected from the results displayed above regarding the dissimilarity between mCherry and GFP saturation stages (see   (Fig. 7B and movie S1) whereas FAST:fluorogen exhibits photophysical properties comparable to mCherry ( Supplementary Information, Table S1).
We conclude that the FAST:fluorogen complexes are efficient and accurate reporters in the heterogeneous and low-O 2 context of a growing biofilm.
As FAST enabled quantitative fluorescence measurements beyond the development limits imposed by µOD signal saturation, it also provided a longer time scale to determine biofilm cell biomass growth rate. We made this analysis for several characteristic edge and centre regions using exponential adjustment of the FAST fluorescence intensity versus time curves, = I Ie FAST kt 0 to derive k, the biofilm growth rate. Figure 8 shows the plot of k for the different ROIs together with the saturation value of GFP at the same location which reflected the amplitude of the lack of O 2 as shown above. Interestingly, we observed that O 2 shortage which caused GFP fluorescence failure did  not affect biofilm growth rate (Fig. 8). This means that the reduction of oxygen concentration, although blocking oxygen-sensitive reactions such as those controlling GFP or mCherry fluorescence, does not alter significantly biofilm growth.

Discussion and Conclusions
Bacterial biofilm research currently goes through a rapid-fire expansion due to the diversity and the complexity of the questions raised by such a living community and its tremendous impact on our human lives and natural environment. Here, we focus our attention on real time optical imaging of biofilms with the aim of observing the dynamics of the adherent community in situ with a high spatio-temporal resolution. In this context, quantitative microscopy is essential and very promising, but raises the issue of accurate reporting. This study interrogated the use of genetically encoded fluorescent reporters of the GFP family to image living biofilm. We show that GFP and mCherry exhibit high sensitivity to the local level of oxygen. This oxygen sensitivity, which prevents the use of GFP-like fluorescent proteins under anaerobic conditions as originally reported by Tsien and co-workers, is experienced for an O 2 level threshold rapidly reached under the concentrated and confined conditions of a growing biofilm, even under continuous flow. This leads to a spatial distribution of fluorescence intensities that depend on both the number of fluorescent cells and the local O 2 concentration but also on the history of cells involving the number of cell divisions or biomass transport processes.
We show here results obtained with GFP and mCherry, although we also observed similar behaviours using YFP, Venus or mKate2 (data not shown). This O 2 -dependence is critical given that O 2 distribution within a living bacterial biofilm depends a lot on the biofilm growth conditions such as the geometry of the setup, the O 2 permeability of the device material, the flow rate which controls the amount of advected O 2 but also all the conditions that will impact cell division rate and consequently cell consumption of O 2 such as growth medium or temperature. The topography of the biofilm development also plays a role. Thus, oxygen distribution is hardly predictable in such complex systems. Nevertheless, considering the bacterial O 2 consumption given in Riedel et al. 31 for an exponentially growing population of E. coli, i.e. 1.7 × 10 6 molec./s and an O 2 medium concentration equal to 3 × 10 22 molec./m 3 , we can calculate that the O 2 contained in a cubic micro-volume of 100 µm side length in equilibrium with an atmospheric pressure of O 2 or at the contact of a PDMS interface itself in equilibrium with atmosphere 32 , is fully consumed in less than 2 s by 10 4 bacterial cells, which corresponds to a 20 µm thick layer of cells at a 20% volume fraction on a 100 µm side square. Besides, using an O 2 diffusion coefficient in water of 3 × 10 −9 m 2 /s, it can also be evaluated that the characteristic time for an O 2 molecule to leave the 100 µm side cubic micro-volume is about 6 s. This emphasizes the high degree of O 2 consumption by the bacteria in the confined environment of a growing biofilm, in agreement with the severe gradient of O 2 observed in the channel for a mature biofilm. It is also worthwhile to consider the amount of O 2 advected by the flow working at 1 ml/h. Using the same O 2 concentration and consumption values as above, it can be shown that the 10 13 molec./s contributed by the flow in the whole channel are fully consumed by a mature biofilm, underlining the micro-aerobic environment that should be expected in a millifluidic channel impermeable to gas such as a glass channel but also the key role of the oxygen diffusing by the channel wall in a PDMS device. However, it should also be kept in mind that bacterial oxygen consumption is expected to fluctuate a lot depending on many environmental and physiological parameters.
Therefore, quantitative interpretation of the fluorescence intensity of GFP and its homologs is not possible in a living biofilm. We propose here an alternative approach based on FAST, a small inducible chemical-genetic fluorescent marker composed of a protein part interacting reversibly with an exogenously applied fluorogenic chromophore that is non-fluorescent until labelling occurs. Fluorescence arises from the conformational locking of the fluorogenic chromophore within the binding cavity of FAST 28,33,34 . This unique fluorescence activation mechanism does not require molecular oxygen, making FAST a very good candidate for imaging proteins and cells in low-O 2 environments. We demonstrate here that the FAST:fluorogen complexes efficiently labelled the biofilm, the small fluorogenic dyes diffusing freely and quickly in the biofilm. FAST allowed the reporting of exponential bacterial biofilm growth in flow at development stages where both GFP or mCherry fluorescence signal plateaued or decayed for hours. The ability to vary FAST colour changing the applied fluorogen opens exciting possibilities for various combinations with classical proteins. Indeed, beyond the measurement of a given gene expression per se, using the FAST system in combination with colour-complementing prototypical FP should allow real time monitoring of O 2 variations in ratio mode. Moreover, assuming effectively that below a given O 2 concentration threshold, no more fluorescent protein is produced and the cells are sharing the pre-existing pool could allow one to deduce the number of cell division number occurring locally by considering the FAST/FP fluorescence ratio to calculate the FP dilution factor.
Other families of non-prototypical genetically encoded proteins have emerged in the recent years 35 . Although O 2 -independent, due to their fluorogen-activated fluorescence, they show little irrigation of the microbe imaging field compared to the iconic GFP and none has been implemented in biofilm characterizations to the best of our knowledge. The LOV family proteins [36][37][38] which fluoresce upon flavin binding detain photophysical properties, interesting for metallic ions or oxygen singlet biosensing but potentially restrictive for gene reporting, which possibly explains its limited spread. UnaG, a recently discovered protein 39 from the Japanese eel, Anguilla japonica, fluoresces upon bilirubin binding in an O 2 -independent mechanism. Starting to be used in mammals as a hypoxia reporter and serum bilirubin sensor 40 , it might also offer an interesting tool for biofilm investigations.
The work we present here aims at stressing the absolute necessity to implement, in quantitative imaging studies dedicated to biofilms, fluorescent reporters displaying robust fluorescence whatever the local O 2 level in order to be able to use the fluorescence signal as an accurate reporter of the protein expression level. We show that this is not the case for the classical FPs such as GFP because of their oxygen-dependent fluorescence and we propose an alternative method based on the system FAST that we successfully implemented in an E. coli biofilm. Further developments of multicolour orthogonal tag:fluorogen pairs should enable the study of multispecies biofilms.

Materials and Methods
Plasmids, strains and fluorogens. Plasmids. pAG87pR and pAG101pR were engineered from pAG87 and pAG101 described in Plamont et al. 28 to replace T7 promoter with the lambda promoter pR for the expression of FAST and FAST-mCherry fusion in E. coli (details in SI). FAST is a variant of the photoactive yellow protein (PYP) containing the mutations C69G, Y94W, T95M, F96I, D97P, Y98T, Q99S, M100R, T101G 28 .
Strains. We used an E. coli MG1655-F strain carrying a non-conjugative F pilus plasmid to promote biofilm formation 41 and its fluorescent variant, MG1655-gfp-F carrying gfp-mut3 gene inserted on the chromosome under the control of the lambda-promoter pR. These strains were transformed with pAG101pR and pAG87pR, respectively, providing MG1655-F-FAST-mCherry which expressed the FAST-mCherry fusion protein and MG1655-gfp-F-FAST which expressed GFP and FAST proteins separately (details in SI).
Microfabrication and biofilm. Millifluidic device. Millifluidic polydimethylsiloxane (PDMS) channels of 30 mm in length, 1 mm in width and height were micro-fabricated and bound to glass coverslips using oxygen plasma activation of the surfaces. Stainless steel connectors (0.013″ ID and 0.025″ OD) and microbore Tygon tubing (0.020″ ID and 0.06″ OD) supplied by Phymep (France) were used to connect fluid flow as described into more details previously 29 . The medium was pushed into the channels at a controlled rate using syringe pumps. When specifically required, we also used square glass capillaries, 800-µm internal-side (Composite Metal Services, Shipley, UK).
Biofilm growth. All cultures were performed from LB agar plate colonies grown overnight in LB medium at 37 °C in an agitated Erlenmeyer flask in the presence of 7.5 µg/ml tetracycline, then diluted in M63B1 minimal medium supplemented with 0.4% glucose to provide an exponentially growing culture with an OD 600 equal to 0.2 after a few hours incubation at 37 °C. Biofilm growth was then initiated by injecting -approx. 3 × 10 6 cells -of exponentially growing cells with an optical density at 600 nm equal to 0.2 in the channel and allowing static settlement for 90 min before starting culture medium flow at 1 ml/h and imaging bacterial development in the channel. The whole experiment was thermostated at 37 °C.
Image acquisition. We used a Hamamatsu ORCA-R2 EMCCD camera for time-lapse acquisitions of 1344 × 1024 pixels images with 12 bits grey level depth (4096 grey levels) and captured an xy field of view of 330 µm × 430 µm. Bright field and fluorescence images were usually collected for 24 hours at the frequency of 12 frames per hour. Excitation LEDs were set at a 50% power level and exposure times were 50 ms or 500 ms for the green emissions and 800 ms for the red emissions.
Image analysis. Image intensity per pixel averaged on defined regions of interest (ROIs) was collected using the NIKON proprietary software NIS. The data sheets edited by NIS were next exported to Matlab for further analysis of the biofilm development kinetics and growth parameters determination. Microscopic optical density (µOD) was derived from transmitted light images according to µOD = ln (I 0 /I), where I 0 is the intensity (averaged gray level per pixel) recorded on a channel filled with medium only and I the intensity recorded on a channel containing a growing biofilm 29 , the presence of which attenuates incident light transmission. Fluorescence intensities were subtracted for their respective background using the contribution to the fluorescence intensity of a channel of medium in the absence of bacteria.
All curves are representative of three independent replicates. The statistical dispersion of the data is given in Fig. S1 where equivalent data coming from distinct positions and distinct experiments are shown. When numerical values are presented, they are given with a confidence interval which is the standard deviation of three equivalent measurements.
Oxygen measurement. FLIM imaging. Lifetime images were acquired using a wide field frequency domain fluorescence lifetime imaging microscopy (FD-FLIM) setup, and a single plane illumination (SPIM) obtained using a beam expander and a cylindrical lens 45 mm of focal length. The thickness of the illuminated plane was determined to be 7 ± 2 µm in the object plane of the microscope. The setup was equipped with a 445 nm laser diode module and an intensified CCD camera that can both be modulated synchronously from 0.1 to 200 MHz. Fluorescence lifetime was derived from the phase shift and relative modulation of the fluorescence sample when compared with a reference sample of known lifetime -here a hypericin solution, 3 ns lifetime.