Keitaro
Kasahara
ab,
Johannes
Seiffarth
ab,
Birgit
Stute
a,
Eric
von Lieres
ab,
Thomas
Drepper
c,
Katharina
Nöh
a and
Dietrich
Kohlheyer
*a
aIBG-1: Biotechnology, Institute of Bio- and Geosciences, Forschungszentrum Jülich GmbH, 52425 Jülich, Germany. E-mail: d.kohlheyer@fz-juelich.de
bComputational Systems Biotechnology (AVT.CSB), RWTH Aachen University, Aachen, Germany
cInstitute of Molecular Enzyme Technology, Heinrich Heine University Düsseldorf, Forschungszentrum Jülich GmbH, Jülich, Germany
First published on 31st March 2025
Microbial metabolism and growth are tightly linked to oxygen (O2). Microbes experience fluctuating O2 levels in natural environments; however, our understanding of how cells respond to fluctuating O2 over various time scales remains limited due to challenges in observing microbial growth at single-cell resolution under controlled O2 conditions and in linking individual cell growth with the specific O2 microenvironment. We performed time-resolved microbial growth analyses at single-cell resolution under a temporally controlled O2 supply. A multilayer microfluidic device was developed, featuring a gas supply above a cultivation layer, separated by a thin membrane enabling efficient gas transfer. This platform allows microbial cultivation under constant, dynamic, and oscillating O2 conditions. Automated time-lapse microscopy and deep-learning-based image analysis provide access to spatiotemporally resolved growth data at the single-cell level. O2 switching within tens of seconds, coupled with precise microenvironment monitoring, allows us to accurately correlate cellular growth with local O2 concentrations. Growing Escherichia coli microcolonies subjected to varying O2 oscillation periods show distinct growth dynamics characterized by response and recovery phases. The comprehensive growth data and insights gained from our unique platform are a crucial step forward to systematically study cell response and adaptation to fluctuating O2 environments at single-cell resolution.
Among the various environmental conditions, the availability of O2 is one of the most critical for microbial growth and physiology. O2 is intricately linked with a multitude of microbial processes, including iron homeostasis,16 oxidative stress,17 the development of pathogenic infections18 and biofilm growth.19 In addition to these microbial processes associated with O2 availability, O2 is also valuable as a primary electron acceptor for aerobic respiration in many microorganisms. In particular, facultative anaerobes, which are capable of growing under both aerobic and anaerobic conditions, adapt to changing O2 environments by switching their metabolic pathways between aerobic respiration and anaerobic respiration/fermentation.20,21 This ability of facultative anaerobes to adapt to different O2 environments has been extensively studied under single-shift O2 environments. Previous studies have primarily focused on examining intracellular adaptation, such as transcriptome,20 protein synthesis,22 metabolome,23 flux balance,24 and phenotypic adaptation like growth fitness.25 These studies have primarily been conducted in conventional cultivation setups, including microtiter plates, shaking flasks, and bioreactors. However, there is a lack of understanding regarding the cellular capability to adapt to rapidly fluctuating O2 environments. Recent studies indicate that microbial behavior in fluctuating environments, where conditions shift within seconds to minutes, can differ significantly from behaviors observed in single-shift experiments.26,27 Investigating the impact of O2 fluctuations on microbial growth would facilitate a more comprehensive understanding of adaptation processes that remain to be elucidated.
The study of microbial responses to O2 fluctuations has been hindered by several constraints. Conventional cultivation techniques do not facilitate rapid and precise changes in O2 concentrations on the timescale of seconds to minutes, nor do they allow for simultaneous, high-resolution data acquisition. Environmental control in conventional laboratory cultivation setups is typically slow, with limited precision in maintaining homogeneity, temporal consistency, and resulting O2 microenvironments. Additionally, these setups are often incompatible with fully resolving growth physiology under fluctuating O2 conditions, as repetitive sampling is impractical without disrupting the culture. Consequently, it has been challenging to analyze the cellular response in terms of microbial growth and physiology caused by rapid O2 fluctuations.
Today, microfluidic devices with precise environmental control and imaging at single-cell resolution are gaining attention as novel tools for creating oscillating environments on-chip and extracting microbial growth data with high temporal resolution. Previously reported microfluidic devices integrated O2 control based on gas diffusion through air-permeable polydimethylsiloxane (PDMS) membrane.28–32 Although some microfluidic systems have been developed to mimic oscillating conditions for pH, nutrients, and O2,27,33,34 detailed analysis linking microbial growth dynamics to fluctuating O2 environments remains limited. To better understand microbial adaptation processes under rapid O2 fluctuations, a platform is needed that allows for high-resolution, single-cell analysis of microbial growth, explicitly correlated with well-defined O2 fluctuations (Fig. 1A).
In this work, we investigated, for the first time, the growth dynamics of the facultative anaerobe Escherichia coli (E. coli) MG1655 under O2 oscillations occurring within minutes. To address the aforementioned limitations, we developed a double-layer microfluidic chip to facilitate rapid gas exchange within the cultivation chambers and frequent data acquisition accompanied by time-lapse microscopy to analyze cell division at the single-cell resolution (Fig. 1B). The PDMS microfluidic chip comprises two layers: an upper layer for gassing and a lower layer with multiple chambers for microbial cultivation. A thin intermediate PDMS membrane (65 μm) separates the two layers, facilitating rapid gas diffusion from the upper to the lower layer. The performance of the microfluidic chip was evaluated by spatially resolved O2 imaging in the fluid channel using fluorescence lifetime imaging (FLIM) microscopy and a fluorescent O2-sensitive dye. The microfluidic chip, automated time-lapse microscopy, and following deep-learning-based image analysis compose a versatile platform to analyze microbial growth and its correlation to applied O2 oscillations. The platform was employed to cultivate E. coli under well-defined O2 oscillating environments with varying oscillation periods, to examine cellular adaptation in a time-resolved manner. Here, we report periodically oscillating microbial growth dynamics composed of several adaptation phases and synchronized with applied O2 oscillations.
For gas control optimization, the on-chip gassing performance was simulated using computational fluid dynamics with experimentally determined gas-inflow concentration profiles resulting from interconnected mass flow controllers. Therefore, the O2 concentration was measured inside the supply tubing outlet under different gas-supply flow rates (100, 300, 600 mL min−1) when no chip was installed (Fig. 2C). This was necessary because likely dead volumes in the mass flow controller setup were affecting the resulting switching performance, mostly when O2 flow was fully switched off, and residual O2 remained inside the non-perfused tubing and connectors. The residual O2 was depleted relatively slowly by diffusion and delayed on-chip switching performance. When switching to higher O2 levels, this problem was not observed since all interconnections were continuously perfused, and no controller was switched off. As depicted in Fig. 2C, the simulated O2 level in the fluid channel exhibits a corresponding change from 21% to 0% when the O2 level in the inlet gas is changed from 21% O2 to 0% O2 at t = 0 min. The simulation results indicated that the gas-supply volume flow rate was the limiting factor in our design, mostly impacting the exchange time of O2 in the fluid channel rather than diffusion across the PDMS membrane. Based on the simulation results, the maximum total flow rate of N2 and O2 at 600 mL min−1 was applied to achieve rapid modulation of O2 within the fluid channel.
With the determined total flow rate, the O2 switching performance was experimentally validated by imaging the fluorescence lifetime of the O2-sensitive dye (tris(2,2′-bipyridyl)dichlororuthenium(II)hexahydrate, RTDP) inside the fluid channel with FLIM. Fig. 2D depicts O2 concentration measured in the fluid channel after the gas exchange from 21% to 0% and vice versa. The supply gas diffused into the fluid rapidly, achieving 99% of the aimed conditions (corresponding to a residual O2 concentration of 0.21% when switching from 21% to 0%) within 15 seconds in both switching directions. The O2 level in the gas supply was also switched between 21% and 0% at various oscillation half-periods T′ (T′ = 60, 30, 10, 5, 2, 1, 0.5 min), showing the robust experimental O2-level data when toggling between 21% and 0% at various T′, as shown in Fig. S1.† These device characterization results ensure a fast gas exchange in the order of seconds in the developed microfluidic device.
Fig. 3A and B show representative time series of phase contrast images of E. coli cultivated under aerobic (21% O2) and anaerobic (0% O2) conditions. Both cultivations started with a single cell at 00:00 h, with a resulting larger colony area at 21% O2 after 03:00 h cultivation time.
To further investigate whether various O2 concentrations also result in a corresponding decrease in cell growth in the microfluidic growth chambers, we cultured E. coli under constant O2 concentrations at 0%, 0.1%, 0.5%, 1%, 5%, 10%, and 21% O2 in separate experiments. As shown in Fig. 3C, the colony areas (Acolony), the sum of the individual cell areas, are quantified from the phase contrast time-lapse images. As evident from the plot, Acolony exhibits exponential growth, with the lowest rate being observed at 0% O2.
In Fig. 3D, the exponential growth rates μ were quantified based on Acolony in the exponential growth phase, showing comparable growth at O2 concentrations between 21% and 1%. The aerobic growth rate of around 2 fits the growth rate suggested in previous literature.35 Conversely, μ strongly decreases when the O2 level is below 0.5%. The relation of growth rate and O2 concentration was modeled by an adapted Monod kinetic, which resulted in KO2 of 0.07 ± 0.02%. As KO2 describes the O2 concentration in percentage at which the growth rate is reduced to half of the maximum growth under sufficient O2, the low KO2 indicates a strong decrease in growth rate at very low O2 levels. These results indicate that an O2 concentration of at least below 0.5% is required to observe a measurable change in the growth rate of E. coli within our device. Based on the gas exchange characterization shown in Fig. 2D, the minimum switching time, tmin, necessary to decrease O2 concentration below 0.5% and observe a detectable change between aerobic and anaerobic growth was approximately 15 seconds. The slower growth observed at low O2 concentration is in agreement with the Pasteur point (1% of the present atmospheric O2 level), below which is thought to inhibit heterotrophic aerobic respiration.36
Beyond colony growth, our data also provide insights at the single-cell level. In Fig. 3E, each gray plot represents the area of an individual cell (Asingle cell) measured at t = 2 h, with red plots representing the mean values. These mean values increase as oxygen concentration rises, which is consistent with the observation that cells with higher growth rates generally exhibit larger sizes.37–39 Interestingly, Asingle cell displayed considerable variation, ranging up to 14 μm2. This wide distribution suggests cell size heterogeneity within the population. Such heterogeneity in cell size might arise from a mixture of cells at different stages: smaller cells immediately post-division, larger cells just before division, and extensively sized cells with fewer division cycles. Such intra-population diversity can be effectively resolved using microfluidic cultivation combined with single-cell, time-lapse imaging.
At T′ = 30 min, μΔt shows a growth tendency similar to T′ = 60 min, characterized by the steep decrease right after the switch from aerobic to anaerobic conditions, and the following growth recovery till the end of the anaerobic gassing phase, as shown in Fig. 4B. At T′ = 10, 5, and 2 min, we observe μΔt hitting the lowest value, and the following gradual recovery phase, but never reaching μ0%, simply due to insufficient time for recovery and adaptation, as shown in Fig. 4C–E. In the case of T′ = 2 min, only a brief recovery phase is observed after the response phase. At T′ = 1 min, the steep decrease after the switch from aerobic to anaerobic conditions is observed without a recovery phase, followed by a fast increase right after the switch from anaerobic to aerobic conditions, as shown in Fig. 4F. As a result, the μΔt line plots at T′ = 2 min and 1 min represent simpler waveforms (monotonous up and down) compared to the other cases.
The single-cell area also exhibited a distinct increase under aerobic and anaerobic gassing phases. Fig. S2A–F† are plotted with Asingle cell obtained from individual cells growing in a representative chamber of each oscillation condition. Fig. S2† allows us to speculate how individual cells increase their cell size by following neighboring plots without needing cell tracking that requires more complicated analytical setups. As for overall tendencies, the plots show a faster area increase rate under aerobic than anaerobic gassing phases, similar to colony-area analysis. A rapid increase/decrease in Asingle cell was observed immediately after each gassing switch across all oscillation conditions. Notably, a clear recovery in Asingle cell was observed when the oscillation half-periods were sufficiently longer than tresponse (T′ = 60, 30, and 10 min).
The periodical comparison suggests that μΔt line plots from T′ = 2 and 1 min have simpler waveforms compared to the other T′ that are sufficiently longer than tresponse. To examine the waveform complexity of μΔt line plots at various T′, the frequency spectrum of μΔt line plots were analyzed using the fast Fourier transform (FFT) as shown in Fig. 5B. There are several frequency peaks at T′ = 10 and 5 min. These several peaks imply the complicated waveform of μΔt line plots due to the existence of response and recovery phases. In contrast, there is only one frequency peak at T′ = 2 and 1 min. The single peaks imply the simpler μΔt line plots, representing only the response phase. Notably, the highest peaks from FFT corresponded to applied O2 oscillation half-periods T′, showing that the periodic growth dynamics were synchronized with applied O2 oscillation periods (T′ = 10 min: 7.8 × 10−4 Hz, T′ = 5 min: 1.8 × 10−3 Hz, T′ = 2 min: 4.1 × 10−3 Hz, T′ = 1 min: 8.4 × 10−3 Hz).
In Fig. 6B, aerobic and
anaerobic for each T′ are summarized. At T′ = 60 min,
aerobic and
anaerobic are comparable to μ21% and μ0% respectively, indicating the sufficient recovery time and growth stabilization after the gassing phase shift. At T′ = 30 and 10 min,
aerobic is comparable to μ21%, whereas
anaerobic is below μ0%. This is due to insufficient recovery time under the anaerobic phases (tresponse < T′ < trecovery), resulting in an overall lower growth rate over anaerobic phases. This trend became more obvious at T′ = 5 and 2 min, with lower
anaerobic because of less time for growth recovery. Interestingly,
aerobic was higher than μ21% at T′ = 5 and 2 min. This high
aerobic is the result of the steep increase in growth rate right after the switch from anaerobic to aerobic gassing phases and insufficient time to adjust the growth rate to around μ21%, as shown in Fig. 4D and E. Lastly,
aerobic and
anaerobic at T′ = 1 min were close to each other, implying the growth adaptation attempt back and forth between aerobic and anaerobic phases, although insufficient time to adapt to either of gassing phases (T′ < tresponse). These results demonstrate a phase- and oscillation-period-dependent growth behavior that can be classified into several cases by growth characteristic values, tresponse and trecovery.
Furthermore, we investigated the difference in phase-averaged growth rate over periods to examine the growth robustness under repeated O2 oscillations. Growth data with more than 3 periods were analyzed (T′ = 10, 5, 2, and 1 min). As shown in Fig. 6C, aerobic and
anaerobic plotted over periods exhibit robust and steady trends, even with repetitive 60 periods at T′ = 1 min. This result indicates the versatility of the developed platform to stably create O2 oscillating conditions and analyze microbial growth under such conditions.
The thorough growth analysis presented here demonstrates distinct growth dynamics induced by O2 oscillations, which are characterized by an immediate decrease in μΔt after the switch from aerobic to anaerobic gassing phases (response), followed by gradual increase (recovery), and later stabilized state. These distinguished cell behaviors occur depending on oscillation half-periods T′. This is reasonable, considering that the change from one metabolic pathway to another requires a series of biological events, such as signal transduction (in milliseconds), enzymatic reaction (in seconds), transcription (in minutes), and translation (in minutes), occurring at different time scales.26,40,41 For example, the O2 oscillation with T′ = 1 min was sufficient to rapidly and strongly decrease the E. coli growth rates in the respective anaerobic gassing phase (Fig. 4F). This observation could be explained by the rapid depletion of the ATP pool under O2 limitation, which occurs within the time scale of microbial responses to environmental fluctuations associated with enzymatic reactions and metabolite turnover under minute.26 A recurring increase in the E. coli growth rates were observed when T′ > tresponse (Fig. 4A–D). This adaptation to prolonged anaerobic phases is most likely the result of specific regulatory processes that alter gene expression patterns, leading to a gradual change in cell metabolism in minutes.25 Under the switch from anaerobic to aerobic gassing phases, the initial peak and the subsequent gradual decrease of μΔt was also observed. This temporal change in growth rate may be attributed to transient accumulation or excretion of metabolites as a result of maintaining homeostasis upon the gaseous transition in minutes.23,42 Lastly, the FFT and phase-averaged growth rate analyses revealed periodic and robust growth dynamics synchronized with the applied O2 oscillation periods. This result implies the cellular capability to respond and adapt to corresponding extracellular O2 environments and highlights the importance of O2 in determining cellular growth behavior. Regarding the metabolic switching under aerobic and anaerobic conditions, hybrid metabolism has been reported under microaerobic conditions, where both aerobic and anaerobic metabolisms are utilized.25,43 Therefore, metabolic switching is a continuous process that may not be clearly divided into aerobic and anaerobic states. Rather, the switching time of E. coli can be characterized and determined by various biological events, as mentioned earlier. For instance, the timescales of enzymatic reaction, transcription, and translation would fit our study. Follow-up studies would be valuable to further investigate the correlation between the biological timescales and O2 fluctuation timescales.
The demonstrated experiment and analysis platform can be strengthened with further improvements. In our platform, we measured the fluorescence lifetimes at 0% and 0.1% O2, which can be distinguished from each other. However, we have not yet made any further measurements in the range below 0.1%. Obviously, O2 sensing with the O2 indicator and FLIM has its limits in terms of sensitivity, which is a complex technical issue that depends on several parameters. For example, the measurement may be affected by the accuracy and resolution of the O2 control. A set of mass flow controllers connected to pure N2 and O2 gas supply was used in our setup, which also has limitations in resolution, especially at lower O2 concentrations below 0.1%, where the mass flow rate has to be set very low compared to an ideal operating range. A follow-up study should consider using a gas supply with a lower O2 concentration (for instance, 1%) instead of pure O2 gas so that the mass flow controller can operate in the recommended flow range when controlling O2 concentrations below 0.1%. FLIM and the O2 sensing dye can be characterized with such an improved setup for O2 control. Finally, sensitivity can be determined with finer resolution at low O2 concentrations. Another factor is the two-point calibration at known O2 availability. This calibration was done by flushing synthetic air containing either 0% or 21% O2. While controlling the O2 availability to 21% was credible, achieving a strict O2 control at 0% remained challenging due to potential disturbances from high air permeability and the possibility of residual air remaining within the PDMS. By improving the calibration method to ensure strict 0% O2 availability, such as by using chemical O2 scavengers44–46 compatible with the O2-sensitive chemical or by using a mini-incubator that allows flushing O2 depleted gas around the PDMS chip,47 more precise on-chip O2 control and sensing under anaerobic conditions may be possible to achieve.
The developed device and the finding regarding microbial behavior under O2 oscillation have the potential to be applied to a wide range of research fields. In terms of practical applications, the findings are useful in characterizing and improving industrial bioprocesses. The fluctuating environments resulting from heterogeneous conditions in large-scale bioreactors have been widely reported, which result in unexpected inefficiency and yield losses.13–15 Such fluctuations in industrial bioreactors, specifically O2 fluctuations, can occur under minutes.48 To address this issue, it is of the utmost importance to gain a further understanding of microbial behavior under rapid fluctuating environments. The developed device provides an on-chip environment that mimics rapid O2 fluctuations inside bioreactors. This enables the analysis of O2 fluctuation-specific microbial behavior, including the emerging phenotypic heterogeneity at single-cell resolution, which was previously not possible.
A comparable approach to recreate fluctuating O2 environments and live-cell imaging could also prove beneficial in fundamental biology and biomedicine. For instance, it is of interest to investigate pathogenic microorganisms (for example, Salmonella typhimurium) and their mechanism on virulence expression and host-cell infection. It has been recognized that pathogens use O2 as a signal to trigger their virulence, yet the underlying mechanism is elusive.18,49 Our analysis platform provides an optimal environment for such a study, where microbial behavior can be resolved at single-cell resolution under a well-defined O2 environment. Another example is to study the interaction of gastrointestinal host cells and microbial communities under fluctuating O2 environments. There has been growing evidence that O2 dynamics play a pivotal role in maintaining intestinal homeostasis.50,51 The intricate regulatory mechanisms at the interface of host cells and the microbiome and the role of O2 are of great interest since these interactions are linked to various diseases.51 Moreover, several reports imply the existence of O2 fluctuations in the intestine and the intestinal epithelial–microbiome interface that arise from periodic ingestion of nutrients or intermittent changes in O2 availability in the blood.52–54 Based on these previous reports, it is reasonable to assume that the timescale of O2 fluctuations for gut microbiome would be in the range of hours. In fact, previous research reported fluctuating bacterial abundance in wild meerkats throughout the day, with the relative abundance of aerobic (Cellulomonas) and anaerobic microbes (Clostridium) varying due to changes in O2 availability during that time.55 The presented device and analysis could be applied to study the interplay between host epithelial cells and microbiomes by emulating such an O2 fluctuating environment.
The O2 level in the chip was measured by fluorescence lifetime imaging (FLIM) and an O2-sensitive dye, tris(2,2′-bipyridyl)dichlororuthenium(II)hexahydrate (RTDP). The fluorescence of RTDP is quenched in the presence of O2, which can be quantified as a change in fluorescence lifetime (τ). The fluorescence quenching is described by the Stern–Volmer equation, as follows.
![]() | (1) |
The details of the image analysis from cultivation experiments are described in previously published papers.47,61 Briefly, the acquired image data in nd2 format was exported as TIFF files and pre-processed using Fiji, which included rotation, alignment (Correct 3D Drift62), and cropping. The pre-processed TIFF files were then uploaded to an OMERO server63 for subsequent analysis. For the automated image analysis, we developed Jupyter Notebooks and Python to perform deep-learning-based cell segmentation (Omnipose64) followed by filtering artifacts and extracting single-cell sizes. These Jupyter Notebooks are designed for a single time-lapse recording and provide video rendering to guarantee and document sufficient cell segmentation quality. We repeatedly apply the same Jupyter Notebook to all our timelapse images (scaling analysis), leading to fully automated image processing such that experiment results are obtained overnight. The codes for cell segmentation and analysis are openly available at https://github.com/JuBiotech/Supplement-to-Kasahara-et-al.-2025.
![]() | (2) |
The relation of growth rate and O2 concentration was modeled by a Monod kinetic65 including a growth offset for anaerobic growth, C (h−1) at 0% O2, as follows.
![]() | (3) |
Instantaneous growth rates, μΔt, the first derivative of Acolony, were calculated as follows.
![]() | (4) |
The fast Fourier transform (FFT) was performed using the Data Analysis Tools in Excel. The sample size was adjusted to 2n prior to FFT. For all the growth analysis, growth data between 0 h ≤ t < 1 h were omitted since growth data at the beginning of the cultivation was occasionally affected by high noise due to a low initial cell number. Datasets with Δt = 10 seconds were smoothed by a centered moving average (window size = 5) before calculating μΔt to reduce noise.
Footnote |
† Electronic supplementary information (ESI) available. See DOI: https://doi.org/10.1039/d5lc00065c |
This journal is © The Royal Society of Chemistry 2025 |