The microfluidic cultivation chip was fabricated following a two-layer soft lithography. Briefly, a 100 mm silicon wafer was subsequently coated with two layers of the photoresist SU-8 (Microchemicals GmbH, Germany) and structured by photolithography. The masks for the photolithography were designed with the layout editor Clewin5 (WieWeb software, The Netherlands) and the generated GDS layout file is available as a Supplementary material.

The fabricated silicon wafer containing the SU-8 structure served as the mold for the following PDMS casting step. 50 g of a PDMS mixture (Sylgard 184 Silicone elastomer kit, Dow Chemical Company, USA) was prepared by thoroughly mixing the monomer base with the crosslinker in a ratio of 10:1, resulting in a homogeneous and opaque highly viscous mixture. This mixture was then degassed in a desiccator at 200 mbar underpressure for 1 h to remove any air bubbles. The resulting fully transparent PDMS mixture was poured over the silicon wafer mounted in a common petri dish. Thermal curing was then performed at 80°C for at least 1 h. The crosslinked PDMS slab was manually peeled from the silicon mold and cut into single chips. Then inlets and outlets for the fluid and the gas flow connections were manually punched using a punching tool (ϕ = 0.75 mm, World Precision Instruments, USA). Single PDMS chips were finally bonded to glass substrates (D263^®Bio, 39.5 mm × 34.5 mm × 0.175 mm, Schott AG, Germany) after an O₂ plasma treatment (Femto Plasma Cleaner, Diener Electronics, Germany) for 25 s. For full fabrication details, the reader is referred to Gruenberger et al. (2013) and Täuber et al. (2021).

Mass flow controllers calibrated for two different mass-flow ranges (0.5 - 10 mL/min and 20 - 600 mL/min, red-y, Vögtlin Instruments GmbH, Switzerland) and the gasses O₂ and N₂, were used to mix the gas at desired O₂ and N₂ ratios (gas supply pressure: p = 4 bar). A mini-incubator (50.0 × 82.0 × 18.0 mm) was designed (Solidworks, Dassault Systems, France) and fabricated by 3D printing (Original Prusa i3 MK3S, Prusa Research, Czech Republic) (Supplementary Figures S1A, B). The O₂ concentration inside the mini-incubator was continuously measured by an inserted laboratory-scale O₂ sensing probe (VisiFerm mA 120 H3, Hamilton, USA). The 3D CAD and printing files are available as Supplementary material. Fluid flow was controlled using a syringe pump (neMESYS, CETONI, Germany) for the continuous delivery of the sensor solution and the culture medium.

For O₂ sensing inside the microfluidic chip, 5 mg/mL of RTDP solution (Acros Organics, Belgium) dissolved in Milli-Q water (Merck Millipore, USA) was used. The prepared RTDP solution was protected from light and stored inside an anaerobic workbench (DG250, Don Whitley Scientific, UK) until use.

Microfluidic cultures and O₂ imaging were performed on an inverted and automated live-cell microscope (Nikon Eclipse Ti-E 2, Nikon, Japan) equipped with a 20 × objective (Plan Apo λ, Nikon, Japan), a 100 × objective (Plan Apo λ Oil, Nikon, Japan), and the perfect focus system (PFS) for focus drift compensation during time-lapse imaging. The setup was equipped with two digital cameras which can be used sequentially during time-lapse imaging. A CMOS camera (DS-Qi2, Nikon, Japan) was installed to the right microscope port to record phase-contrast and wide-field fluorescence images. The fluorescence excitation light was supplied by a SOLA LED (SOLA Light Engine, Lumencor, USA) and a fluorescence filter cube (GFP-3035D, Nikon, Japan).

The special FLIM camera (550 kHz frequency domain, pco.flim, PCO AG, Germany) connected to the modulated excitation laser (445 nm, 100 mW, pco.flim laser, Omicron-Laserage Laserprodukte GmbH, Germany) was mounted to the left microscope port. A customized filter cube with a emission filter = 440/40 (F47-440, AHF analysentechnik AG, Germany), a dichroic mirror = 495LP (F48-495, AHF analysentechnik AG, Germany), and a emission filter = 605/70 (F47-605, AHF analysentechnik AG, Germany) was used for FLIM. Additionally, the microscope setup was surrounded by a temperature incubator (Okolab, Italy), ensuring stable cultivation temperature. A customized chip holder fitting into the motorized X-Y-stage was used to mount the microfluidic chip and the 3D-printed mini-incubator (Supplementary Figures S1C, D). The chip holder CAD layout is available as Supplementary material.

E. coli MG 1655 carrying the vector pVWEx-1-gfpUV and a kanamycin resistance gene, was used for microbial growth experiments. GFP expression in the E. coli strain was induced by the addition of 0.1 mM isopropyl β-D-1-thiogalactopyranoside (IPTG, Carl Roth, Germany). Microbial cultures were performed according to the following procedure. The primary preculture was performed in a complex LB medium (Supplementary Table S1). For all subsequent culture steps, defined minimal M9 medium supplemented with 2% glucose (Carl Roth, Germany), 2% casamino acid (VWR, USA), and 50 μg/mL kanamycin (Carl Roth, Germany) was used (Supplementary Table S2). First, a shake flask containing 15 mL of LB medium was inoculated with a single colony from an agar plate and cultured at 37°C and 150 rpm for approximately 8 h. The second preculture containing M9 minimal medium was inoculated from the primary culture, with an initial optical density (OD₆₀₀) value of 0.05, and cultivated overnight at 30°C, 120 rpm. The final shake flask culture was inoculated from the second preculture with initial OD₆₀₀ = 0.3, and cultivated at 30°C and 120 rpm until the culture reached the logarithmic growth phase. Cells derived from the log phase culture were seeded into the growth chambers of the microfluidic chip.

An experimental system for the control of O₂ during microbial single-cell analysis has been developed (Figure 1). The system includes: (A) a microfluidic cultivation chip with a (de)oxygenator structure for on-chip O₂ control placed inside a 3D-printed mini-incubator, (B) online O₂ sensing by FLIM of the oxygen-sensitive dye RTDP, and (C) and automated image analysis procedure based on DL (Figure 1).

The 3D-printed mini-incubator was required to deliver a continuous gas flow of the desired O₂/N₂ composition over the PDMS chip. As the chip material PDMS is highly gas permeable (diffusion coefficient for O₂ ≈10⁻⁹ m² s⁻¹), it facilitates efficient gas exchange by diffusion across the entire chip surface (Vollmer et al., 2005; Markov et al., 2014). When we operated the chip without the incubator, anaerobic conditions were not possible due to uncontrolled O₂ influx from the surrounding laboratory atmosphere. Thus, the mini-incubator enables cultures in the absence of O₂ when 100% N₂ was supplied continuously and undesired influx was prevented.

Gas flow was controlled using four interconnected mass flow controllers (O₂ 0.5 - 10 mL/min, O₂ 20 - 600 mL/min, N₂ 0.5 -10 mL/min and N₂ 20 - 600 mL/min) to automatically mix O₂ and N₂ in desired ratios prior to the microfluidic chip and mini-incubator. The mini-incubator was operated at a total gas flow of 600 mL/min and a gas flow of 10 mL/min was delivered through the microfluidics. In the microfluidic chip, gas exchange was assisted with the (de)oxygenator structure which facilitates gas diffusion between the fluid and the adjacent gas channels across the PDMS membranes in between, as shown in Figure 1A. The syringe pump was used to deliver either the RTDP solution during on-chip oxygen sensing or the culture medium during single-cell analysis at a flow rate of 400 nL/min. The RTDP-based oxygen sensing was not performed simultaneously with the microbial single-cell analysis, but was used separately to characterize the performance of the device.

Dissolved oxygen in the fluid channel was determined by camera-based fluorescence lifetime imaging of the O₂-sensitive dye RTDP (Figure 1B). During microbial cultures, growth medium was provided continuously to maintain stable culture conditions. Growth chambers were arranged behind the (de)oxygenator structure to allow sufficient residence time and thus gas-exchange time of the fluid. This allowed the cells to be supplied with (de)oxygenated medium and cultured at the desired O₂ concentrations. Phase contrast and fluorescence intensity images were obtained using the conventional CMOS camera.

After microfluidic cultivation and continuous image acquisition from multiple chambers, the data can be quite large, making manual handling of the data time-consuming and difficult. Therefore, the acquired image sequences were processed with an in-house developed automated image analysis platform to extract interesting cellular properties such as cell number, area and fluorescence intensity to determine growth rates and cellular activities. Cellular properties can be extracted for each individual cell in the culture chambers, providing detailed information at the single-cell level (Figure 1C).

A new microfluidic chip with the required channel layout was designed to allow the on-chip control and simultaneous cultures (Figure 2A). The fluid channel (blue) is sandwiched between two adjacent gas channels (red). All supply channels are 100 μm wide and 10 μm deep. As shown in Figure 2Ai, ii, each channel comprises (i) a (de)oxygenator zone, followed by (ii) a cultivation zone with 80 cultivation chambers, allowing a statistically reliable number of microbial populations to be imaged during a single experiment.

Between the fluid channel and the adjacent gas channels, a 100 μm thick PDMS wall acts as a gas permeable membrane. The height of a cultivation chamber is 1 μm and the surface area = 50 μm × 30 μm, resulting in a volume of a single cultivation chamber of 1.5 pL. The height of the chamber corresponds to the diameter of a typical E. coli cell thereby restricting cell growth to a defined monolayer (Figure 2Aiii). Monolayer cell growth is highly advantageous for subsequent single-cell segmentation tasks on time-lapse images. Depending on the organism to be cultured, the resulting chamber height can be specified by adjusting the photoresist thickness of the spin-coated SU-8 layer during wafer processing. Cultivation chambers are arranged in series and each one is interconnected to two common fluid supply channels. During the cultivation phase, the volume flows in the parallel supply channels are equal. Therefore, no pressure difference and thus no convective flow is induced inside the growth chambers. As a result, mass transfer between the supplied medium and the growth chamber is governed by diffusion and the cells are not affected by shear forces (Westerwalbesloh et al., 2015).

As all gas channels are interconnected, only a single gas inlet and outlet port is required for on-chip gas flow, which results in uniform O₂ levels across the chip. In addition, all cultivation channels are equipped with adjacent gas channels on either side, creating a homogeneous O₂ distribution across the cultivation zone.

For illustration purposes, the microfluidic channels of the fabricated chip were visualized by the infusion of red and blue ink, as shown in Figure 2B. The chip contains three parallel culture channels, allowing three simultaneous cultures during a single experiment. The scanning electron microscope images of the fabricated chip show (i) the (de)oxygenator structure with the fluid channel in the center and the adjacent gas channels at the sides, and (ii) the cultivation chamber (i.e., growth chamber) with the rectangular structure and two connections to the fluid supply channels (Figure 2C).

Prior to the fluorescence lifetime imaging (FLIM), a calibration measurement must be performed with a reference slide containing a fluorescent dye with known and stable lifetime (lifetime = 3.75 ns, UMM-SFG, Starna Scientific, UK). This calibration is required to correctly determine the lifetime from unknown samples, in our case the lifetime of the O₂-sensitive dye RTDP. The calibration procedure must be performed every time the laser is shut down or operated at different frequency.

The performance of oxygen control was characterized using the RTDP sensor solution. The fluorescence of RTDP is quenched in the presence of O₂ according to the Stern-Volmer equation (Equation 1), where K_q is the Stern-Volmer constant (the quenching efficiency), τ is the experimentally measured lifetime and τ₀ is the lifetime in the absence of oxygen.

First, the RTDP solution must be calibrated at two known oxygen concentrations, at 0% O₂ giving τ₀ and at, for example, 100% O₂ to then derive K_q. In practice, however, the calibration procedure at 0% O₂ can be challenging because it requires an anoxic sensor solution. Several procedures have been demonstrated, for example, using a droplet of the sensor surrounded with 100% N₂ gas, or consuming O₂ by adding an O₂ scavenging chemical to the solution (Kalinina et al., 2016; Wu et al., 2019). However, none of these methods were compatible with our approach.

In the present work, the calibration was performed at 0% O₂ using a gas-tight glass syringe connected to a glass capillary filled with an anoxic RTDP solution to determine τ₀. The RTDP solution and the syringe were prepared entirely in the anaerobic workbench. Briefly, the solution was gassed with N₂ for at least 1 h in the anaerobic workbench, then filled into a gas-tight glass syringe (1.0 mL, ILS, Germany) to ensure an anoxic solution (Supplementary Figure S2A). The syringe was connected to a glass capillary (75 μm inner diameter x 375 μm outer diameter x 70 cm length, Molex, USA), which was then manually flushed with the anoxic RTDP solution. Prior to the assembly, 1 cm of the polymer coating was removed in the middle of the capillary using a flame and acetone to prepare an optical detection window for FLIM. The assembled syringe including the capillary was taken out of the anaerobic workbench, fixed above the objective on a suitable specimen holder, the capillary end was connected to a waste container and flushed continuously at 400 nL/min using the syringe pump. Then FLIM was performed to determine τ₀.

Furthermore, the RTDP solution was calibrated at 100% O₂ inside the microfluidic chip by using the on-chip microfluidic gas flow combined with the continuous gas flow through the mini-incubator at 100% O₂ to determine τ_100% (Supplementary Figure S2B). Briefly, the fluid channel was connected to a syringe (Omnifix^®F Solo 1mL, Braun, Germany) via PTFE tubing (0.5 mm inner diameter x 1.6 mm outer diameter, Chromatographie Service GmbH, Germany) and the RTDP solution was delivered at 400 nL/min. Gas flow was maintained at 100% O₂ for several hours until the oxygen diffusion reached equilibrium and a steady-state lifetime sensor output and τ_100% was determined. The obtained τ₀ and τ_100% were used to calculate the K_q value in Equation 1.

Calibration was performed at two different temperatures (T = 25°C and 30°C). As evident, the fluorescence lifetime of RTDP is also temperature dependent, thus calibration, measurement and microbial cultures should be performed at the same temperature.

The following lifetime values were determined during calibration. At T = 25°C, τ₀ = 573 ns and τ_100% = 160 ns (Supplementary Figure S2C), whereas at T = 30°C, τ₀ = 526 ns and τ_100% = 144 ns (Figure 3A). Using Equation 1, K_q was calculated as 2.58 at T = 25°C and 2.65 at T = 30°C. Using these K_q values and Eq. 1, the measured oxygen concentrations could be calculated from the measured fluorescence lifetimes.

A sequential step test was performed (at T = 25°C and 30°C) during O₂ control to characterize the dynamical performance of our setup. Therefore, a switching sequence was programmed using an in-house written LabView tool controlling the four massflow controllers. The O₂ / N₂ ratio of the gas mixture was then changed automatically to generate desired O₂ concentrations in the fluid flow. The RTDP solution was delivered continuously through the fluid channel, while the gas flow composition through the gas channels and the mini-incubator was set to 0%, 2%, 4%, 6%, 8%, 10%, 15%, 21%, and finally to 100% O₂ over a period of 30 h. The initial sensing phase at 0% O₂ was maintained for 4 h to ensure anoxic starting conditions. Then, the gas composition was altered every 3 h. During FLIM the RTDP solution was continuously delivered at 400 nL/min. The O₂ concentration inside the mini-incubator rapidly followed the desired values and changes were performed with negligible delay as evident from the oxygen sensor readings from the Hamilton probe installed inside the mini-incubator (data not shown).

Time-lapse images of the fluorescence lifetime were taken every 5 min at two successive channel positions, as shown in Figure 3B before and after the (de)oxygenator (ROI 1 and 2). The acquired image sequences from FLIM were processed with Fiji (Schindelin et al., 2012), a distribution of ImageJ (National Institutes of Health, USA) to determine the average lifetime for each image, as these images have a certain noisiness at the level of single pixels. From the representative false-color images determined for each O₂ concentration at T = 30°C (here ROI 1, Figure 3C), it is visible that FLIM delivered lifetime images, revealing homogeneous oxygen availability in the channel. The lifetime decreased with the increasing oxygen concentration, as expected. The time-lapse video of the RTDP solution during oxygen level switching is available as a Supplementary material. Oxygen values were then calculated using Equation 1 and plotted over time in Figure 3D (for T = 25°C see Supplementary Figure S2D).

The displayed O₂ concentration curve shows a response behavior to the concentration switches O₂ performed. Due to the diffusion-limited oxygen transport, the concentration follows an asymptotic progression until equilibrium and the final value is reached. The final oxygen values in the chip were determined in each equilibrium phase by averaging the measured concentration over the last 1h of each performed step, and plotted over the desired values in Figure 3E (for T = 25°C see Supplementary Figure S2E). The graph clearly shows that the final O₂ concentration reached very comparable values over the whole range from 0% to 100% O₂.

Finally, we studied the switching performance in terms of time. Therefore, the times required to reach 85%, 90% and 95% of the final O₂ concentration (t_85%, t_90%, t_95%) were determined and plotted in Figure 3F for T = 30°C (for T = 25°C see Supplementary Figure S2F). t_90% is indicated as central marker, whereas t_85% and t_95% are represented as the lower and upper bars. The plot show that t_90% was in the range of 80 to 90 min. t_90% seems independent of the applied O₂ step width, as the diffusion time t is independent of the concentration difference but depends on the diffusion distance d and the specific diffusion coefficient D (t≈d²/D).

The present system is limited in terms of fast oxygen changes, which can be performed below two hours. This is mainly due to the millimeter-thick PDMS chip. Obviously, the gas exchange can be accelerated by fabricating a thinner PDMS chip to reduce the time required for the gas exchange by diffusion (Polinkovsky et al., 2009; Markov et al., 2014). Our intention here was to develop a relatively simple setup enabling reliable switching and constant culture conditions.

A cell suspension was prepared using cells harvested from the logarithmic growth phase of the final shake flask culture and diluted to an OD₆₀₀ = 0.35. Cells were then seeded into the microfluidic cultivation chambers by delivering the cell suspension into the fluid channel by applying gentle pressure to the syringe while visually inspecting the seeding performance. After sufficient chambers were seeded with cells, M9 medium was used to rinse residual cells from the supply channels. M9 medium containing 0.1 mM IPTG was then continuously delivered at a flow rate of 400 nL/min. During the cultivation, time-lapse phase contrast and fluorescence images were taken every 5 min. Optimal illumination and camera exposure times have to be experimentally derived for each setup.

In the case of the anaerobic microfluidic cultivation, cell inoculation was performed entirely inside the anaerobic workbench to ensure complete isolation of the cells from O₂, but without the ability to follow the seeding process by microscopy. After seeding, the glass syringe filled with an anaerobic M9 medium containing 0.1 mM IPTG was connected to the fluid inlet. The entire cultivation setup, including the syringes, the chip and the mini-incubator, was assembled inside the anaerobic workbench. Finally, the prepared setup was removed from the anaerobic workbench and mounted on the microscope stage. The temperature inside the temperature incubator was maintained at 30°C.

The recorded phase contrast and fluorescence time-lapse image sequences from biological experiments were analyzed to extract the individual intracellular fluorescence intensities. As the image sequences contained more than 50k individual cells in total, the image processing was done partly automatically using DL segmentation in Jupyter Notebooks and Python (Seiffarth et al., 2022). First, the acquired time-lapse sequences were cropped into single cultivation chambers, aligned with Fiji, and stored in an OMERO instance (Allan et al., 2012). A pre-trained Omnipose (Cutler et al., 2022) segmentation model was applied to extract the contour of individual cells followed by a border filter eliminating segmentation artifacts closer than 0.5 μm to the image borders. The fluorescence signal for every individual cell is computed by averaging the fluorescence channel intensities inside every segmented cell contour.

After cell segmentation, cell tracking was performed using the activity-prioritized method (Ruzaeva et al., 2022) to explore the distributions of the fluorescence intensities inside every segmented cell contour over the cell generations. The cell lineages were computed using the default parameters of the cell tracker, utilizing the Omnipose segmentation model. Starting from the initially present cell, assigned with generation “zero”, the daughter cells gain one generation after every recorded division event. The analysis was performed for the fixed time point for each data set. The code used for cell segmentation and tracking is available at https://github.com/JuBiotech/Supplement-to-Kasahara-et-al.-2023a.