Our microfluidic chips were fabricated using two-layer soft lithography processes (Figure 1A) (Bentley et al., 2022). Even though the chips contained one, two or three rows of traps located between rows of pillars, all traps had the same size and geometry (Supplementary Figure S1). The pillars contained square cavities filled with air during chip operation and used for the autofocus function. The height of the bottom flow channel was 7.5 μm. The top layer had circular traps of diameter of 60 μm and was 7.5 μm high, such that the overall height of the droplets was 15 μm. Uncured polydimethylsiloxane (PDMS) consisting of a 10: 1 polymer to cross-linker mixture (Sylgard 184) was poured onto the master, degassed, and baked at 70°C for at least 4 h. Before this curing time at 70°C, a piece of 0.15 mm thick glass cover slip was immersed into the liquid PDMS and manually aligned above the trapping array. This method minimized droplet evaporation into the PDMS (Supplementary Figure S2). Following master mold fabrication, PDMS chips were plasma bonded (Diener Zepto) on thin cover slip glass (thickness 0.15 mm) to enable high-resolution imaging. Hydrophobic surface treatment was performed after bonding by flushing with 1% (v/v) trichloro (1H,1H,2H,2H-perfluorooctyl) silane (Merck) in HFE-7500 and, subsequently, placed in a 65°C oven for 30 min.

The microfluidic chip was placed on an epifluorescence microscope equipped with an automated stage. A self-digitization water-in-oil emulsion method was followed to make droplets (Figure 1B). First, the entire microfluidic chip was manually primed with a fluorinated oil (HFE-7500, Fluorochem Ltd.) solution supplemented with 1% w/w fluorosurfactant (008, RAN Biotechnologies) using a syringe. The main reason for using a surfactant in the oil solution is to reduce the interfacial tension between the sample and the oil phase. This facilitates the digitization process and prevents droplet coalescence (Cohen et al., 2010). The sample containing either bacteria or bacteria-phage mixture was then flushed through the chip using the opposite inlet to replace the HFE oil solution. The oil solution was subsequently flushed again through the chip forming droplets of volume 28 picoliters (assuming elliptical spheres).

Time-lapse imaging data were acquired using a temperature-incubated, μManager-controlled Olympus IX83 fluorescence microscope equipped with an LED light source (Lumencor SpectraX) and an automated stage (Marzhauser) (Edelstein et al., 2014; Chait et al., 2017). The incubation box (cellVivo) temperature was maintained at 30°C for all assays. Fluorescence and phase contrast images were acquired with an sCMOS camera (Hamamatsu Orca Flash4v3) every 3 min through a 40× objective (UPLFLN40XPH, Evident Olympus). Up to six positions were recorded in each time-lapse, capturing four to six droplets per field of view. Air cavities in pillars of the microfluidic chip aided software-based autofocus (on 640 nm reflection images) performed prior to acquiring each image set (Figure 1C), as described in Chait et al. (2017). We then recorded a phase contrast image to confirm that traps are populated with sample-containing droplets, and a GFP fluorescence image (to quantify fluorescence-based population size) at a focal plane mid-way through the droplets at each position (Figure 1D). We set exposure time to 20 ms for phase images, and 100 ms was the exposure time to image GFP (excitation 460–480 nm, emission 495–540 nm, dichroic mirror 490 nm).

Raw fluorescence images were divided by a shading corrector image to account for inhomogeneity of GFP excitation illumination across the field of view, as described in (Chait et al., 2017). To obtain the shading corrector, we took a median projection through a stack of GFP images from distinct locations on a glass cover slip overlaid with 50% (w/v) fluorescein (Model and Burkhardt, 2001). We then normalized this corrector image by dividing by its median pixel value. After shading correction, we performed background subtraction by using “mask” images. Masks were made by finding the thresholds in each frame for which the background fluorescence value, i.e., signal not coming from fluorescent cells within droplets, corresponded to 0. Then, each shading-corrected image was multiplied with the respective mask image for each recorded frame. This final image was used to set a circular region of interest (ROI) around each droplet. We extracted mean fluorescence intensity per ROI corresponding to mean droplet fluorescence intensity. For the analysis, we excluded frames in which the device was out of focus. All image analysis routines were done in the open-source image processing package Fiji (ImageJ) (Schindelin et al., 2012), and more details can be found in Supplementary Methods.

One phage species with DNA genome, T7 (Escherichia virus T7) and two phage species with RNA genomes, MS2 (Emesvirus zinderi) and Qβ (Qubevirus durum) were used in this study, together with E. coli strains BW25113, W1485, and the derivatives of K-12 MG1655 harboring the constitutively expressed gfp reporter gene encoding for Superfolder GFP, with strain genotypes described in Supplementary Table S1. The F plasmid from strain W1485 was introduced by conjugation into strain TB193, now annotated as strain TB193 F+. Only strains that harbor the F plasmid can be infected by MS2 and Qβ, as RNA phage start infection by adsorbing to bacterial F-pili encoded on the F plasmid. Phage T7 can infect all listed E. coli strains. Bacterial cultures were grown in LB medium (1% tryptone, 0.5% yeast extract, 1% NaCl), which is a nutrient-rich medium. Frozen glycerol clones were streaked on LB agar plates to obtain single colonies. A single colony was used to inoculate overnight batch cultures shaking at 230 rpm. All overnight incubations and batch cultivations prior to phage infection were carried at 37°C. Bacterial cultures were supplemented with 0.01% glucose and 2 mM CaCl₂ prior to adding RNA phage for infection experiments [addition of divalent ions facilitates phage adsorption (Rappaport, 1965, Ács et al., 2020)]. LB medium was supplemented with 10 μg/mL chloramphenicol for selection after conjugation.

Overnight cultures of an E. coli strain were diluted 1 to 100 into 4 mL of fresh LB medium for 3 h at 37°C, to obtain exponentially growing cultures at ~5 × 10⁸ bacterial cells/mL. Bacterial cultures were then supplemented with 0.01% glucose and 2 mM CaCl₂. For phage infection experiments, aliquots of bacterial cultures were mixed with phage lysate of the known titer in a 1.5 mL-tube, yielding a specific multiplicity of infection (MOI, the ratio of phage particles to bacterial cells in a mixed culture), prior loading the sample into the microfluidic device. The exact time from preparing bacteria-phage mixture to starting the time-lapse microscopy experiment was noted (typically, 11–14 min). An additional aliquot of the bacterial culture was taken just before starting the microscopy experiment (or in the case of infection experiments, just before adding phage) to determine the initial bacterial population size. Serial dilutions of the culture aliquot were plated on LB agar plates and incubated at 37°C for 24 h. Bacterial population size was determined by measuring the density of colony forming units CFU, as: CFU/mL = n_colonies/(dilution factor × V_{diluted culture}), and these measurements are included in Supplementary Datasets. MOI values reported in figure legends were calculated based on the measured population size of each bacterial culture prior infection. As bacteria-phage mixtures were prepared prior droplet generation, the analyzed droplets had the same initial MOI in each experiment. We do expect some fluctuation in absolute initial phage numbers from Poisson statistics (Collins et al., 2015) but this arguably becomes less significant following the first lysis event after which the number of phages in droplets would increase dramatically, as the reported phage burst sizes are in the range: 179–182 phage T7 particles/cell (Nguyen and Kang, 2014; Xu et al., 2021), 2,000–2,100 phage MS2 particles/cell (Rappaport, 1965; Jenkins et al., 1974), 90–776 phage Qβ particles/cell (Woody and Cliver, 1995; Tsukada et al., 2009).

Overnight bacterial cultures of the E. coli host propagation strain (BW25113 for phage T7, W1485 for phages MS2 and Qβ) were diluted 1 to 100 into LB medium. After 4 h, exponentially growing cultures were inoculated with a single plaque in 4 mL of phage soft agar (1% tryptone, 0.1% yeast extract, 0.01% glucose, 0.8% NaCl, 2 mM CaCl₂, 0.7% agar; kept at 50°C), plated on phage plates (1% tryptone, 0.1% yeast extract, 0.01% glucose, 0.8% NaCl, 2 mM CaCl₂, 1% agar), and incubated overnight at 37°C for 20 h (Nikolic et al., 2023). The following day, soft agar containing plaques was removed into a 50 mL-Falcon tube with 12 mL of SM buffer (0.1 M NaCl, 8 mM MgSO₄·7H₂O, 50 mM Tris-HCl pH 7.5, 0.01% gelatin). Falcon tubes were centrifuged for 15 min at 4,000 g. The supernatant containing phage lysate was filtered twice through a 0.22 μm-filter into a new tube. Phage lysates were kept at 4°C, and their phage titer was determined by plaque spotting assays described below.

Overnight cultures of the E. coli host strain were diluted 1 to 1,000 into LB medium. After 5.5 h, 200 μL of exponentially growing bacterial cultures were added to 4 mL of phage soft agar, and plated on phage plates (Nikolic et al., 2023). After 2–3 min, serial dilutions of the phage lysate were spotted on top of the soft agar containing bacterial host, and the plates were incubated overnight for 20 h, at 37°C for phages MS2 and Qβ, or at room temperature for phage T7. Phage titer was determined by measuring the density of plaque-forming units PFU, as: PFU/mL = n_plaques/(dilution factor × V_{diluted phage}).

Overnight E. coli cultures were diluted 1 to 100 into LB medium. After 3 h of cultivation, the cultures were supplemented with 0.01% glucose and 2 mM CaCl₂. Aliquots of the cultures were put into wells of a 96-well microplate, and each well was either infected with phage at the specific MOI, or the same volume of SM buffer was added to uninfected cultures. All samples were then diluted 1 to 10 into fresh LB medium, having final volume of 180 μL in each well. Four independent replicate cultures were analyzed in experiments with phage T7, and five independent replicates were analyzed in infection experiments with MS2 and Qβ. Growth of the cultures was recorded at 30°C, every 4 min for 800 min in total, as absorbance at 600 nm A₆₀₀, with shaking prior to each measurement (CLARIOstar Plate Reader, BMG Labtech).

For each growth curve, we employed the Microsoft Excel function slope from ln-transformed measurements over a sliding window of 40 min, i.e., over 11 time points in plate-reader assays or 14 time points in droplet assays. Bacterial growth rate μ was then defined as the maximum slope value for each growth curve, within the period t = 36–120 min for droplet experiments, and t = 20–120 min for plate-reader assays. In addition, growth rates μ of bacterial populations challenged with RNA phage were determined within the period t = 38–200 min for droplet experiments, and t = 24–200 min for plate-reader assays. Doubling times (plate-reader assays) or fluorescence-based doubling times (droplet assays) corresponded to ln2/μ. To mark the transition from initial bacterial growth to the onset of bacteriolysis, inflection points were evaluated for each growth curve from slopes calculated over a sliding window of 40 min, by identifying the time point where slope value changed from positive to negative. The middle time point of the sliding window was noted to be the inflection point. To evaluate differences between different conditions or different strains we used two-tailed two-sample heteroscedastic student’s t-tests.

We optimized the layout of a microfluidic device for easy self-digitization, enabling formation of stable droplet-environments and steady microscopy imaging routine. The microfluidic chip utilized in our experiments had two inlets, one for the oil phase and the other one for the aqueous samples, with a central large trapping chamber embedding dozens of traps and pillars (Figure 1A). A glass cover slip was immersed in the uncured PDMS and cured in place on top of the trapping area to minimize droplet evaporation and ensure the stability of droplets during the course of time-lapse experiments (Supplementary Figure S2). The chip was placed on an automated fluorescence microscopy imaging platform in an enclosed environmental box set to 30°C. We employed a three-step process to generate droplets on-chip following the self-digitization emulsion method, which means that emulsions were formed by consecutive flowing of aqueous and oil phases into the same chip (Figure 1B). In our study, the aqueous phase was the sample of either a sole bacterial culture, or a mixture of phage and susceptible bacteria cultivated in growth medium. The chips were first primed with the solution of oil containing surfactant, and the sample was then flushed through the chip using the opposite inlet so that the sample could fill all traps. The oil solution was then flushed again and droplets were passively formed within the traps. Interfacial tension and density difference between aqueous samples and oil generated anchored droplets that had the diameter of the designed traps (Supplementary Figure S1). In our experiments, droplets were formed in 96% of traps within the field of view, while the remaining 4% were either empty traps or traps populated with more than one droplet (Supplementary Datasets). Moreover, each microfluidic chip consisted of rows of traps followed by rows of pillars (Figure 1C). Air cavities were included at the center of pillars to identify them and provide fiducials at the glass interface to aid software-based autofocusing performed using 640 nm illumination prior to imaging at each time point and location (Chait et al., 2017) (Supplementary Figure S1). Stable autofocusing enabled steady image acquisition from several pre-selected positions on the chip per each time point, and ensured reproducible imaging despite microscope stage motion (Figure 1D). The small height of the chamber (15 μm) ensured accurate quantification of overall fluorescence produced by the cells. Finally, we utilized a rapid and simple image analysis routine from the widely-used open-source image analysis software platform Fiji (ImageJ) to quantify mean fluorescence intensity of each droplet (see Supplementary Methods). Altogether, we developed a device and workflow for the on-chip droplet generation and stable multidroplet microscopy image acquisition dedicated to the long-term monitoring of bacterial growth and lysis in droplets.

We first assessed how droplet-environments support the growth of Escherichia coli carrying chromosomally encoded, constitutively expressed superfolder gfp. An exponentially growing E. coli culture was loaded into a primed microfluidic device subsequently flushed with oil to form bacteria containing-droplets within traps. We monitored the increase in GFP fluorescence coming specifically from all cells during the time-lapse microscopy experiment, as the proxy for bacterial population growth in droplets. Despite known stochastic fluctuations in gfp expression in clonal bacterial populations, a broadly linear correlation between integrated droplet fluorescence and the number of fluorescent bacterial cells in droplets has been reported previously (Barizien et al., 2019; Taylor et al., 2022). For this and all subsequent droplet experiments, we aimed at loading 10 to 40 bacterial cells in each droplet, with the number of bacterial cells across droplets following a Poisson distribution, as previously established in Collins et al. (2015) and Taylor et al. (2022). We measured a mean 30-fold GFP fluorescence increase of the droplet populations over 400 min of the time-lapse experiment (Figure 2A, Supplementary Movie S1). Our analysis showed that the mean fluorescence-based doubling time of bacterial populations in droplets was 24 ± 8 min (mean ± standard deviation). The mean doubling time of the same strain growing in a plate-reader was 34 ± 2 min (Figure 2B). Overall, our results indicate that E. coli populations can successfully grow inside droplet-environments within designed microfluidic devices, with a growth pattern comparable to bulk batch cultures.

Next, we investigated whether we can utilize the same droplet-based setup to monitor the dynamics of bacterial populations challenged with phages. To this end, we mixed bacterial cultures of gfp-carrying E. coli with phage lysate prior to loading the sample into a microfluidic device. In all our experiments, we employed obligatory lytic phages, which are phages that promote bacteriolysis and kill infected bacterial cells. Subsequently, decline in bacterial population size due to lysis of bacterial cells, i.e., the biomass reduction within droplets, was detected as a decrease in overall GFP fluorescence. In the first set of experiments, we infected E. coli cultures with phage T7 at MOI ~1. T7 is a phage species with DNA genome, and is known for its short replication cycle (time from infection to the lysis of the host cell) of typically 15 min at 37°C, and 30 min at 30°C, and a high efficacy in killing E. coli cells (Jack et al., 2019; Xu et al., 2021). By measuring GFP fluorescence, we detected a severe reduction in the E. coli population size during T7 infection in droplets. Our analysis indicated that on average 99.4% of an E. coli droplet population was killed 218 min after adding T7 phage (Figure 2C), with 0–2 cells per droplet remaining at the end of the time-lapse experiment (Supplementary Movies S2, S3). We determined the time point when infected bacteria in droplets had a population dynamics shifting from dominant growth to dominant lysis, by finding the inflection point of GFP fluorescence curves. The mean time to the population decline, i.e., to the start of significant bacteriolysis of droplet populations, was 43 ± 7 min. Growth and lysis assays in a plate-reader showed that the mean time to population decline during T7 infection was 46 ± 2 min, with the eradication of bacterial cultures, i.e., >99.9% of the population killed, happening after 165 ± 92 min (Figure 2D). In addition, time to T7-induced bacteriolysis was not significantly dependent on the presence of the gfp reporter gene in the E. coli genome (Supplementary Figure S3). The presence of the F plasmid slightly increased the time to lysis, by 19% on average, without significantly affecting the time to whole population eradication (Supplementary Figure S3 and Supplementary Table S2). Overall, these experiments suggest that droplet-based technology can be utilized to analyze bacterial population growth dynamics in response to phage, and to quantify the efficacy of phages to kill bacterial populations.

We evaluated the ability of generally poorly characterized phages with RNA genomes to influence the growth of E. coli populations in droplets. MS2 (Emesvirus zinderi) and Qβ (Qubevirus durum) are the best known RNA phages, with longer phage replication cycle (typically 90 and 105 min at 37°C, respectively), different lysis machinery, and lesser efficacy in killing E. coli than DNA phage T7 (Rappaport, 1965; Jenkins et al., 1974; Woody and Cliver, 1995; Bernhardt et al., 2002; Tsukada et al., 2009; Nikolic et al., 2023). E. coli were mixed with RNA phage lysates at MOI ~10, to ensure a large excess of phage particles and therefore an increased likelihood of infection; specifically, MS2 lysate at MOI = 7, and Qβ lysate at MOI = 14. Droplets populated with bacteria-phage samples were then generated and trapped within a microfluidic device. Phage MS2 impeded bacterial growth in droplets, with fluorescence-based doubling time of droplet populations increasing more than 3-fold compared to uninfected populations, to 81 ± 16 min (Figure 3A, Supplementary Movie S4). Analysis of inflection points from GFP fluorescence curves indicated that droplet population size started to decline 374 ± 72 min after E. coli being challenged with MS2, in 63% of analyzed droplets. The remaining droplet populations either exhibited a reduced growth rate without population size decline, or resumed growth shortly after the initial population size decline (Supplementary Datasets). Overall, after 601 min of exposure to phage MS2, we measured a mean 2.8-fold GFP fluorescence increase of the droplet populations compared to the GFP fluorescence intensity at the beginning of the experiment. This increase in the fluorescence level suggests that a subpopulation of the infected droplet population continued growing in the presence of phage MS2. Recent single-cell studies have indeed indicated heterogeneity in the bacterial response to phage exposure: some cells within the infected bacterial population lysed, some cells entered a dormant, non-dividing state, and some cells continued growing and dividing with different elongation rates (Attrill et al., 2021; Nikolic et al., 2023). In addition, we measured the growth dynamics of the same E. coli strain challenged with phage MS2 at MOI = 10 in a plate-reader (Figure 3B). The population doubling time in a plate-reader was 46 ± 4 min during MS2 exposure, and the size of all plate-reader populations started to decrease 310 ± 45 min after adding phage MS2.

Furthermore, we detected a reduction in bacterial growth after challenging E. coli with phage Qβ, with the fluorescence-based doubling time of 204 ± 110 min for infected populations in droplets (Figure 3C, Supplementary Movie S5). The start of decline of the droplet population size was detected 233 ± 18 min after adding phage Qβ, in 94% of analyzed droplets. At the end of the time-lapse experiment involving 1,481 min of exposure to phage Qβ, we measured a mean 1.6-fold GFP fluorescence-decrease of the droplet populations. Similarly to the experiments with phage MS2, we measured growth dynamics of E. coli populations challenged with phage Qβ at MOI = 10 in a plate-reader (Figure 3D). The population doubling time during Qβ exposure was 43 ± 3 min, and decline in the size of all plate-reader populations was noticed 445 ± 17 min after E. coli being challenged with phage Qβ. Altogether, comparable phage-modulated bacterial growth dynamics between plate-reader and droplet experiments paves the way towards development of high-throughput multidroplet screening methods that can make bacteria-phage interaction analyses faster than standard microbiology methods.

In addition, the droplet experiment involving almost 25 h of phage exposure (Figure 3C) allowed us to test feasibility of the microscopy setup to monitor long-term bacteria-phage dynamics, which would be crucial for understanding underlying eco-evolutionary principles. Long-term exposure of bacteria to phages can generally lead to complex coevolutionary dynamics, with emerging bacterial antiphage resistance mechanisms as well as phage adaptation mechanisms facilitating more efficient infection of the host (van Houte et al., 2016; Koskella et al., 2022). In some of our experiments, we observed bacterial growth recovery during prolonged phage challenge, possibly due to mutations arising in bacterial genomes in the infected cultures, leading to antiphage resistance (Supplementary Figure S4). In conclusion, our findings indicate that utilizing droplet-based technology can provide quantitative temporal information on how phages modulate bacterial growth rate and alter population size of their bacterial host over longer periods of time.