Biofilm development of an opportunistic model bacterium analysed at high spatiotemporal resolution in the framework of a precise flow cell

Life of bacteria is governed by the physical dimensions of life in microscales, which is dominated by fast diffusion and flow at low Reynolds numbers. Microbial biofilms are structurally and functionally heterogeneous and their development is suggested to be interactively related to their microenvironments. In this study, we were guided by the challenging requirements of precise tools and engineered procedures to achieve reproducible experiments at high spatial and temporal resolutions. Here, we developed a robust precise engineering approach allowing for the quantification of real-time, high-content imaging of biofilm behaviour under well-controlled flow conditions. Through the merging of engineering and microbial ecology, we present a rigorous methodology to quantify biofilm development at resolutions of single micrometre and single minute, using a newly developed flow cell. We designed and fabricated a high-precision flow cell to create defined and reproducible flow conditions. We applied high-content confocal laser scanning microscopy and developed image quantification using a model biofilm of a defined opportunistic strain, Pseudomonas putida OUS82. We observed complex patterns in the early events of biofilm formation, which were followed by total dispersal. These patterns were closely related to the flow conditions. These biofilm behavioural phenomena were found to be highly reproducible, despite the heterogeneous nature of biofilm.


INTRODUCTION
Biofilms are inherently heterogeneous communities. Biofilm activities are not the summation of their individual constituents in their planktonic state, but rather, the cells in biofilms exhibit distinctly different properties from their single freely-suspended homogeneous cultures. 1,2 The close physical proximity among microorganisms enables small molecules to diffuse effectively, allowing for quorum sensing/quenching signals 3,4 and may allow for the regulation of specific biofilm developmental patterns. [5][6][7][8] The initial phases of biofilm development are tightly correlated to their microenvironments. The importance of environmental gradients in selecting the most suitable microbial physiology to thrive in a given ecological niche was described already in the late 19th century by Sergey Winogradsky. 9 Quantification of these important observations has since been limited by the lack of tools for studying biofilm development from single cells to social communities. The heterogeneous nature of biofilms requires precise tools and procedures to allow for the quantification of their behaviour at the appropriate resolutions and at low background noise.
The study of microbial biofilms using traditional microbiology methods, like agar plates, microtiter trays, etc. are limited to non-flow environments that biofilms rarely encounter in nature. Ecological interactions are suggested to govern the development of microbial biofilms. 10 Therefore, biofilms should be studied under conditions that mimic their natural habitats such as biofilm flow cells. This concept led to the design of the presently most widely used growth chamber, with growth media pumped into straight channels by peristaltic pumps. Developed by Wolfaardt et al. 11 and later refined by Christensen et al., 12 these straight channel platforms provide unidirectional flow fields operated under well-established protocols. 13 Efforts to increase flow cell sophistication also resulted in a system for generating a two-dimensional flow pattern. 14 However, the ability to create specific well-controlled environments, including defined gradients, in these chambers remained limited.
The application of various lab-on-chip technologies into biofilm studies gave rise to various microfluidic flow cells with the capabilities of generating well-defined conditions such as chemical gradients, [15][16][17][18] hydrodynamic stresses [19][20][21][22] and temperature gradients. 23 However, these chambers are often custom-made for specific experiments. Moreover, most microfluidic devices are confined to shallow channels, and do not allow for accumulating sufficient biomass without the biofilm significantly altering the bulk environment even in the initial stages of biofilm development. An additional limitation of existing flow cells is that the channels are often sealed by permanent bonding, preventing the removal of intact surface-associated biofilm for further analyses, usually requiring the substrata to be destructively broken. 13 Recent developments in fluorescence microscopy, coupled with advances in fluidics and microfabrication facilitate dynamic biofilm studies by real-time live imaging at the spatial and temporal resolutions required to unravel the physics that shapes life at microscales. To achieve this, single micrometre spatial resolution and single minute temporal resolution are necessary. 24,25 Precision has to be incorporated into a flow cell system from its design, fabrication and the operation protocols. We present here a high spatiotemporal resolution approach for the real-time study of biofilm behaviour under well-controlled flow conditions. This flow cell was fabricated by micro-machining processes that were optimised for precision and reproducibility. In addition, the chamber has a removable substrate that allows for pre-treatments of the surface by surface modifications and for downstream analyses of the intact biofilm developed on that surface.
Using confocal laser scanning microscopy, we demonstrated live imaging at multiple locations in our flow cell over long-term experiments. Each specific position inside the chamber can be revisited at any time during the experiments with an accuracy of ± 2 μm limited only by the accuracy of the motorised stage of the microscope. We present a protocol to operate the flow cell including validation of flow pattern, biofilm experiments and quantitative biofilm growth analysis. Using the developed procedures, we demonstrate and quantify for the first time the complex yet highly reproducible dynamic formation and dispersal patterns in a model biofilm using a well-documented strain of Pseudomonas putida OUS82.

RESULTS
The aim of this study was to develop robust procedures to observe the behaviour patterns of the initial phases of biofilm development from attachment to biofilm build-up and dispersal. These observations were conducted under defined flow conditions. By applying a precise engineering approach for an in-depth understanding of life in microscales, we have observed and quantified, for the first time, fundamental developmental features during the initial phases of biofilm formation and dispersal.
We have designed, fabricated and validated a novel biofilm flow cell; assembled a system of precise flow control and mounted this system on an accurate motorised stage of an advanced confocal microscope; acquired high-resolution images and reproducible real-time imaging in three dimensions. The large number of confocal images were then quantified and analysed.
Hyperbolic biofilm flow cell The flow cell we developed has a channel with two inlets feeding into a hyperbolic expansion that then leads to one outlet ( Figure 1a and Supplementary Figure 1b). The hyperbolic expansion generates a linearly decreasing flow velocity (i.e., a negative linear velocity gradient) along the channel center-line (zone ii in Figure 1a). The flow fields for four flow rates used (Q = 0.1, 0.5, 1.5 and 4.0 ml h − 1 per inlet) were first simulated (Figures 1c and d) and then experimentally validated by particle image velocimetry, with good agreement between the simulated and measured flow velocities (Figures 1e and f). The flow velocity decreased linearly from x = − 1.49 mm to x = -− 8.99 mm resulting in a three times difference in magnitude at these two locations at all flow rates.
Real-time high-content three-dimensional imaging We observed the biofilm development of green fluorescent protein (GFP)-tagged P. putida OUS82 26 under defined flow fields using confocal laser scanning microscopy at real time, high content and high resolution. Each experiment, which was   Table 1).
The images were compiled into snapshots (Figures 2a and 3a) and four videos (Supplementary Videos 1-4), revealing the dynamic nature of biofilm formation and dispersal in this model organism. In all the experiments, we observed gradual biofilm formation up to a maximal biovolume followed by total dispersal. Dispersal was always observed to initiate at downstream (position 12) and propagate upstream (position 1). Significantly, different dynamics of biofilm formation and dispersal patterns    . Repeated experiments clearly demonstrated the reproducibility of these patterns. The same behaviours were observed to be repeatable within temporal resolutions of 4-17 min ( Table 1). The large number of collected microscopic images were analysed to quantify the dynamics of the biofilm behaviour (Figures 2b and 3b (Figures 5a-f).
Although the average number of initially attached clusters shown in bars varies significantly, it has little effect on the trend of initiation of dispersal shown in the line.  Figure 6a, the delay in the initiation of dispersal with increasing flow rates is apparent. Furthermore, the slope of the lines indicates a shift from gradual propagation of dispersal in the lowest flow rate into abrupt dispersal in all positions under higher flow rates. The shift from the gradually propagating dispersal to an abrupt process is further emphasised in Figure 6b. No correlation between flow rate and the resulting maximal biovolume at each position is illustrated in Figure 6c, except for the highest flow rate of 4.0 ml h − 1 , which has a significantly higher maximal biovolume in all positions. In addition, the average doubling time at this flow rate is significantly shorter, down to 47 min, in comparison with the longer doubling times of up to 71 min observed at the lower flow rates.

DISCUSSION
The application of the precise flow cell and robust procedures for high-resolution observation of biofilm development allows us to revisit the roles of microbial ecology in the development of biofilms. Even when using the presented robust procedures, the ability to reproducibly observe biofilm development in controlled environments is limited to its initial stages.
The mechanism of bacterial adhesion has been well established by K. Marshall. 27 The initial phases of bacterial attachment were later shown to be guided by niche selection. 28,29 In this study, we observed the transition from fluctuating numbers of adhered cells to stable, permanently adhered cells following 30 min of static inoculation and 1 h 30 min of flow. In each experiment, the permanently adhered cells then multiplied in a clonal mode at a constant doubling time (averaged over 12 positions) along the entire linear velocity gradient regardless of the significantly different flow along these positions (Supplementary Video 8b). The average doubling time ranged from 49 ± 2 min at the highest flow rate to 68 ± 2 min at the lowest flow rate.
Subsequently, deviation from the theoretical clonal cluster distribution was observed as many single bacteria detached from the concomitantly growing clusters. The newly attached cells and the established clusters continued to grow in size. During this period, there was an increase in the maximal observed biovolume per area from 24,067 μm 3 at the low flow rate of 0.5 ml h − 1 to 104,090 μm 3 under 4.0 ml h − 1 . This significant increase in the maximal biovolume could have been dependent on the overall flux of nutrient supply. Yet, we could not detect significant differences in biovolume (Figure 6c) and doubling time (Figure 6d)   along the x-axis in each of the experiments albeit having a linearly decreasing nutrient flux as generated by the velocity gradient. This phenomenon repeated itself in all the experiments regardless of the flow rate. Furthermore, there were no significant differences in the biovolumes and doubling times between the two control positions-2 and 12. These two positions have identical velocity in each experiment but are at opposite ends of the channelposition 2 near the medium inflow while position 12 near the medium outflow (Figures 6c and d). These clearly indicate that nutrient flux along the flow cell is not the limiting growth factor.
Only when maximal biovolume was reached, was the initiation of dispersal observed in all the experiments. Dispersal processes are clearly affected by flow and they are delayed with increasing flow rates. Dispersal always began at the downstream region close to the outlet of the flow cell and propagated upstream towards the inlet of the medium inflow. This phenomenon was most pronounced at the lowest flow rate. We postulate that it was affected by metabolites produced at the upstream region affecting the initiation of dispersal downstream. With increasing flow rate, the dilution of metabolites reduced this phenomena significantly. 30 In all the experiments conducted here, we have observed total dispersal. The possible re-establishment of the biofilm following a total dispersal event is unclear.
The experimental approach presented here is a result of true merging of microbial ecology criteria and engineering at the required precision. The combination of precise tools and robust experimental procedures provide the required reproducibility and high precision to address questions on the importance of microscale heterogeneity during the initial development of biofilm aggregates from a single attached cell to maximal biovolume and subsequent dispersal. We have provided here tools and protocols to study biofilms at the spatial and temporal scales required to quantify the physics of life at microscales. This may well be an effective experimental approach for the microbial ecology study of the inherently heterogeneous biofilm mode of life either in free biofilms or host-associated microbiomes. 31

MATERIALS AND METHODS
Set-up of flow cell system Our flow cell system includes (i) syringe pump(s) to precisely regulate flow of media into the flow cell; (ii) tubing to connect different components of the system; (iii) microvalves for flexible media control; (iv) a newly developed hyperbolic flow cell with a removable coverslip; and (v) effluent collectors.
We designed and fabricated a precise flow cell using in-house machining facilities to (a) create controllable well-defined environmental gradients inside the flow cell; (b) have a removable substratum (a microscopy coverslip) on which biofilms develop; this allows for various surface modification and facilitates post-analysis of the intact biofilm developed on the substratum; (c) enable long-term live imaging at multiple positions at high spatial and temporal resolutions. A detailed description of the flow cell structure and performance can be found in our patent. 32 For this investigation, the flow cell was composed of one disposable acrylic (poly(methyl methacrylate), DAMA Trading, Singapore) plate carrying a channel with a hyperbolic expansion, covered with a removable 22 mm × 22 mm × 0.17 mm microscopy glass coverslip (Marienfeld Superior, Lauda-Königshofen, Germany). The top view of the channel design and the structure of the flow cell are shown in Figures 1a  and b. The channel depth was 0.98 mm for all the experiments and can be easily modified for specific usage. Polymeric components of the flow cell, such as the acrylic plate carrying the channel profile, are designed to be disposable and can be mass produced by injection moulding; therefore lowering its cost, which will aid in the wide adoption of this system.
A two-syringe infusion pump (KDS200, KD Scientific, Holliston, MA, USA) was used to simultaneously control the media flow entering the two inlets of the flow cell. Three-way microvalves (VICI Micro Valve three-port 90°fl ow path, model JR-660310, Valco Instruments, Houston, TX, USA) were used to provide flexible control of flow during various steps of the experimental process, including removal of air bubbles arising from changing/replenishing media infused into the flow cell. Microbore PTFE Tubing (0.032″ ID × 0.056″ OD, Cole-Parmer Instrument Company, Vernon Hills, IL, USA) was used to allow for high temperature resistance, chemical inertness, low gas permeability and a low friction coefficient.
The compact set-up of the flow cell system on a confocal microscope Zeiss LSM 780 (Carl Zeiss, Jena, Germany) for live imaging of biofilm development is visualized in Supplementary Figure 1a. The assembly of the flow cell system is described in detail in Supplementary Note 1.

Simulation and validation of flow field
We used four flow rates in this study including Q = 0.1 ml h − 1 per inlet, Q = 0.5 ml h − 1 per inlet, Q = 1.5 ml h − 1 per inlet and Q = 4.0 ml h − 1 per inlet. We designed and constructed the channel geometry using SolidWorks (Dassault Systèmes SolidWorks, Waltham, MA, USA). Flow fields at all flow rates were simulated using COMSOL Multiphysics 4.2a-Laminar Flow module (COMSOL, Burlington, MA, USA). M9 minimal growth medium supplemented with cassamino acids 33 (Supplementary Table 2) was used for both the simulation of the flow field and biofilm development experiments. The density and viscosity of the M9 medium required for the simulation were measured to be 1,016 ± 2 kg/m 3 and 1.09 ± 0.01 mPa s, respectively. For the four flow field simulations, identical flow rates of 0.1 ml h − 1 , 0.5 ml h − 1 , 1.5 ml h − 1 and 4.0 ml h − 1 were set at the two inlets and the single outlet was set at atmospheric pressure. No-slip boundary conditions (velocity = 0) were imposed on the walls of the channel. The channel was meshed with physics-controlled mesh (calibrated for fluid dynamics) with maximum and minimum element sizes of 300 and 15 μm, respectively. The simulated mid-plane velocity fields (velocity field at half-depth of the channel) and centreline velocities were plotted and subsequently validated experimentally by particle image velocimetry (see Supplementary Note 2).
Microbial experimental procedure and confocal imaging A rigorous experimental procedure, including sterilisation and priming ahead of bacterial inoculation were applied to ensure a contaminant-free chamber (Supplementary Figure 4a). A scheme of the flow cell system setup including two modes of operation is given in Supplementary Figures 4b  and c. The flow mode was used as the default setting throughout the experimental procedure, except for a switch to the locked mode during media changing to prevent air being trapped, and during non-flow inoculation.
At the start of each experiment, the flow cell was first sterilised with 70% v/v ethanol (Merck, Singapore) in DDW for 15 min at a flow rate of 1.0 ml h − 1 per inlet. This was followed by priming the chamber with M9 medium for 15 min at the same flow rate. This step ensured that all ethanol was flushed out. In the next step, the flow cell was inoculated with a suspension of defined strain of P. putida OUS82::GFP 26 that expresses GFP constitutively. This strain was grown overnight in M9 medium and diluted to an optical density of 0.005 at 600 nm and the cell density was measured using colony-forming unit counts (3.43 × 10 6 ± 5.51 × 10 5 per ml). The inoculum was delivered using an initial flow rate of 4.0 ml h − 1 per inlet for 1 min to completely fill the tubing from the valves to the two inlets. The flow rate was then reduced to 1 ml h − 1 per inlet for 4 min. After this stage, the valves were switched to the locked mode for static inoculation. This event was defined as the reference zero time point t = 0 in all the experiments. Two syringes filled up with M9 media were precisely positioned onto the syringe pump. Any air trapped in the tubing was removed into the two effluent collectors (Supplementary Figure 4c Quantification of three-dimensional biofilm confocal images For the quantification of biofilm development, we used the threedimensional stacked images and computed biofilm biovolume by using Imaris (Bitplane, Zurich, Switzerland). We calculated biofilm clustering at 36 positions using the surface segmentation algorithm of Imaris. The biovolume of individual biofilm clusters i at position p at imaging cycle n is defined as V pni . The segmentation parameters used for the above computation were defined as: (a) absolute intensity threshold of 10; and (b) minimum object size of three voxels (each voxel is a cuboid of 0.42 μm × 0.42 μm × 0.78 μm). The computed biovolume is defined to be the volume of the bacteria cells excluding the additional volume of the extracellular polymeric matrix.
V pn -total biovolume per imaging window at position p and imaging cycle n, was calculated by summing the biovolume of all the clusters in the imaging window of 212.55 μm × 212.55 μm determined by the image acquisition parameters and the specific objectives used.
where p is the position (p = 1 to 36), n the imaging cycle number (n = 1 to 47 for flow rate 0.1 ml h − 1 , n = 1 to 62 for flow rate 0.5 ml h − 1 , n = 1 to 60 for flow rate 1.5 ml h − 1 and n = 1 to 73 for flow rate 4.0 ml h − 1 (see Supplementary Table 3 for the conversion between actual experiment time t and imaging cycle n)). N pn is the total number of clusters in the imaging window at position p and imaging cycle n. N pn is varied at different positions and different imaging cycles. To allow for comparison between different positions at each flow rate, V pn was normalised against the total biovolume V pna À Á at the respective position using a reference time point at 2 h (i.e., imaging cycle n a ) in all experiments. Time point 2 h was chosen because the bacteria are permanently attached to the surface by that time. The number of clusters at this time was taken as the number of initially attached cluster, N ap . The normalised total biovolume at position p and imaging cycle n (Vnorm pn ) is calculated as follows: The apparent growth rate at position p was assumed to follow an exponential equation described as: where V pn (t) is the total biovolume at position p at time t (i.e., imaging cycle n), V pn0 is the initial total biovolume at position p (i.e., at imaging cycle n 0 ), g p is the apparent growth rate at position p and has a positive value, t is the time. The average growth rate at position p ðg p Þ was calculated by fitting V pn during the period from the start of experiment (n 0 = 1) to the time of maximal observed growth at that position; at imaging cycle n pmax is the cycle at which total biovolume reached its maximal value. The doubling time at position p, t dp , was calculated by: The distribution of cluster sizes at position p, at imaging cycle n, was plotted by sorting V pni in ascending order against the total number of clusters at that position N pn (with p being the position, p = 1 to 36). The bubble plot represents the cluster size and its spatial distribution at each defined time point. Each bubble represents an individual cluster i at position p taken in imaging cycle n. Its diameter, D pni , was computed from V pni by assuming the cluster as a sphere as follows: The centre of the bubble was the centroid of the respective clusters.
A summary of the experimental parameters and specifications of image acquisition is presented in Supplementary Table 1.