Bacterial growth under confinement requires transcriptional adaptation to resist turgor pressure build-up

Abstract

Bacterial proliferation in confined spaces occurs in biofilms, intracellular compartments, or infection sites, generating mechanical constraints. We investigated how growth-induced mechanical pressure affects bacterial physiology using a microfluidic device ensuring nutrient access. We found that proliferating Escherichia coli cells generate forces in the hundreds of kPa range decoupling growth and division, producing shorter bacteria. Increased cytoplasmic protein concentrations and crowding induce division arrest and decreased protein synthesis. In contrast, growth-induced pressure keeps increasing during confinement. Theoretical modeling predicts this novel regime of steady pressure increase, termed overpressurization, driven by persistent metabolite synthesis. Under confinement, the Rcs pathway is activated, and rcs mutants display abnormal shapes only in the overpressurized state. Therefore, transcriptional adaptation is required to resist pressure build-up. A uropathogenic strain displayed the same confined growth phenotypes in vitro, highlighting the potential relevance of these pressurized regimes during infection.

Bacteria often grow in confined spaces within biofilms, host cells or infection sites. Here, Le Blanc et al. show that E. coli growth in confined spaces generates pressures of hundreds of kPa, and is associated with formation of shorter bacterial cells, cytoplasmic crowding, increased turgor pressure, and activation of stress responses.

Introduction

Mechanical forces shape the development and condition of all forms of life. Animal cells probe and apply forces on their environment, and their capacity to respond to mechanical signals, called mechanosensing, regulates diverse processes, such as growth, motility, state of differentiation, and behavior within tissues¹. Dysregulation of conserved mechanotransduction pathways is involved in the emergence of diseases impacting tissue development and homeostasis, such as cancer^2,3. Despite a major focus of the mechanobiology field on eukaryotic organisms, mechanical sensing and adaptation to forces are also essential for microbes, especially among bacteria colonizing diverse ecological niches⁴. Recent technological advances in live imaging and microfabrication have opened the path to the investigation of how single bacterial cells respond to mechanical cues⁵. Importantly, force sensing was proposed to contribute to infection by inducing specific mechanical morphotypes with enhanced tolerance and invasiveness^6,7.

Despite these recent advances, how forces experienced by bacteria in their natural environment influence their physiology, morphology, and growth remains poorly understood, especially in the context of disease progression during infection^6–9. Bacteria form dense multicellular communities in a wide range of conditions, such as during biofilm formation within a self-secreted polymeric matrix¹⁰. The combination of bacterial cell proliferation, cell-cell cohesion, and adhesion on a substrate induces the build-up of internal stress, suggesting that this might be a key general feature of biofilm growth⁹. The geometry of the microenvironment can also impose external constraints on growing bacterial cells^11–15. More generally, bacterial growth in a limited space leads to the formation of dense, confined colonies, leading to compressive forces. In eukaryotes, cells proliferating in a limited space generate pressures ranging from the kPa range for animal cells and up to the hundreds of kPa range in the case of yeast^16–18. These self-induced compressive forces affect the regulation of key cellular processes, from cell growth and division to differentiation and stress tolerance^19–21. However, the magnitude of growth-induced pressure for bacterial colonies and its impact on bacterial physiology at the single-cell scale remain elusive^11,22.

Historically, the bacterial cell wall was considered the sole element responsible for bearing the mechanical stress produced by turgor pressure and maintaining cellular shape, a perspective shaped largely by work in Gram-positive bacteria. It is now clear, however, that the architecture and mechanics of the bacterial cell envelope vary widely across bacterial species²³. In Gram-negative bacteria, the outer membrane is increasingly recognized as an important mechanical element²⁴, and additional surface layers such as the capsule can also play a role in maintaining envelope integrity²⁵. To this aim, bacteria rely on the activation of stress-response pathways to trigger transcriptional changes that strengthen surface structures, modulate cell wall synthesis, and maintain envelope integrity²⁶. Among them, the Cpx system and more recently the Rcs (Regulator of Capsule Synthesis) phosphorelay system have been shown to respond to mechanical perturbations of the cell envelope^5,26–28. Understanding which components are required in bacteria for mechanosensing and adaptation to confined growth might reveal novel mechanisms of adaptation to stress and specific molecular targets to fight bacterial infections.

Indeed, the potential impact of spatial confinement and the consequent generation of growth-induced mechanical pressure is particularly relevant for the infectious context, where bacteria grow within the host tissue and potentially experience increasing mechanical constraints as their number expands. For example, uropathogenic Escherichia coli (E. coli) invade cells of the urothelium, forming dense intracellular bacterial communities (IBCs)^29,30. These colonies can grow until the infected cell bursts and releases bacteria that can infect neighboring cells^31,32. Another example of pathogenic bacteria facing confinement during host invasion is Neisseria meningitidis growing and ultimately occluding the lumen of blood vessels⁸. Staphylococcus aureus can also form very tight colonies of deformed bacterial cells in canaliculi of cortical bone. This leads to chronic osteomyelitis cases, a major challenge in orthopedics³³. Bacteria growing in a physically limited environment thus occur in many infectious diseases, however further studies are required to dissect the functional impact of this process in disease pathogenesis.

Here, we combined in vitro experiments with theoretical modeling to unveil how bacteria respond and adapt to spatial confinement. We developed a microfluidic setup, the bacterial confiner, to confine E. coli in a controlled mechano-chemical environment while allowing live-cell microscopic observation at subcellular resolution. We found that mechanical confinement induces profound changes in bacterial physiology. Growth under confinement increases intracellular crowding and turgor pressure. These modifications in cell physical properties influence growth and division in a way that impacts bacterial morphology and are driven by changes in the bacterial transcriptional profile upon confinement. Specifically, the Rcs pathway leads to cell envelope remodeling essential to counterbalance high levels of turgor pressure and ensure bacterial shape maintenance. Importantly, uropathogenic E. coli displays similar shape changes in the microfluidic confining device, resembling those observed during IBC formation in urinary tract infection^31,32. Overall, our work integrates quantitative measurements of bacterial physiology and force dynamics with theoretical modeling to provide a biophysical and mechanistic elucidation of bacterial growth in confined spaces for commensal and pathogenic strains.

Results

Bacterial proliferation under confinement generates growth-induced pressure

To investigate how bacteria adapt to mechanical confinement during proliferation in a limited space, we developed a PDMS-based microfluidic chip in which bacteria grow inside chambers connected to 400 nm-wide nanochannels that prevent bacterial passage (Fig. 1A and Supplementary Fig. 1A, B). The bacterial confiner was designed based on previous work on yeast and mammalian cells and scaled down to dimensions adapted to trap 1 µm-wide cells^16–18. This setup allowed constant medium renewal while trapping bacteria inside multiple chambers connected in parallel (Supplementary Fig. 1C, D and Supplementary Video 1). In the bacterial confiner, bacteria proliferated, filled the 2D chamber floor (i.e., reached confluency) within ~5 h, and ultimately became densely packed, i.e., confined (Fig. 1B). Bacterial death during extended periods of confinement was rare (Supplementary Fig. 1E). We asked whether bacterial proliferation upon confinement led to the generation of compressive forces by measuring PDMS chamber deformation¹⁶. Semi-automatic tracking of the chamber contour showed that as soon as bacteria reached confluency, they pushed against the walls, leading to chamber deformation in the three dimensions (i.e., on the sides and top) and physical confinement (Fig. 1C, Supplementary Fig. 1F, G, and Supplementary Video 2). By contrast, in 5% of the cases, bacteria progressively exited the chambers when close to confluency without deforming the chambers (Fig. 1C). These cases were used as internal controls in all experiments to characterize how bacteria behaved in the absence of confinement. For quantification purposes, forces generated by bacteria were determined by calibrating PDMS deformability, and single curves were aligned by defining time 0 as the time frame that preceded the generation of measurable pressure against the walls, i.e., at confinement onset (Supplementary Fig. 1G–I). We found that bacteria generated growth-induced pressure (GIP) averaging 300 kPa after about 10 h of confined growth (Fig. 1D and Supplementary Fig. 1I). Pressure build-up was strictly dependent on efficient perfusion of the chamber despite minor changes in bacterial numbers (Supplementary Fig. 1J–L). 3D super-resolution imaging of E. coli K12 strain MG1655 with the inner membrane fluorescently labeled via a ZipA_TM-mCherry fusion³⁴ showed confinement did not induce any vertical cell orientation in these conditions (Supplementary Fig. 2A). Thus, 2D single-cell segmentation based on the inner membrane fluorescent marker was used to measure bacterial features and numbers in the 2D focal plane at the bottom of the chamber, the latter serving as a proxy for the number of bacteria in the chamber (Supplementary Fig. 2B–D). We found that the number of bacteria continued to increase exponentially for 1.5 h after confluency, which correlated with a progressive increase in growth-induced pressure (Fig. 1D). This was followed by a slowdown in bacterial proliferation, accompanied by a linear increase in growth-induced pressure until up to 10 h post-confinement. Thus, the bacterial confiner allowed dynamic subcellular observation of bacterial proliferation in a constrained space^11–13, while allowing efficient medium renewal and simultaneous pressure measurements. Using this device, we showed that bacterial proliferation upon confinement is sufficient to generate large compressive forces in the presence of nutrient perfusion. This further raised the question of how bacteria generate these forces and whether they adapt to mechanical confinement.

Fig. 1. Bacterial proliferation under confinement generates growth-induced pressure.

A Schematics of the bacterial confiner in top (xy) and side (xz) views. Growth medium is continuously renewed via flow through nanochannels (red arrows) (Supplementary Video 1). B Timelapse brightfield images of bacterial proliferation in the bacterial confiner in the absence of confinement (t = 0 h, left), at confluency (t = 5 h, middle), and upon confinement (t = 15 h, right). Time is indicated as hh:mm. C Schematics of chamber deformation during confinement, with an inset illustrating the measurement region. Below are kymographs of chamber edges for a control chamber (top), where confluent bacteria proliferated without deforming the walls, and a confined chamber (bottom), where bacterial proliferation induces wall deformation along the x-axis. Segmented chamber edges are represented in colors (Supplementary Video 2). Insets highlight the chamber deformation profile corresponding to the dashed white square regions. D Growth-induced pressure (GIP, in kPa) and number of bacteria in the 2D focal plane of observation as a function of time (n_chambers = 13, N = 3). The phases before confinement, at confluency, and upon confinement, shown in (B) are highlighted. Time 0 corresponds to the time that precedes GIP build-up. Data points correspond to mean values ± standard deviations. The number of bacteria follows an exponential growth curve until 1.5 h (red dashed line). Scale bars: 5 µm, scale bars insets: 1 µm.

Growth-induced pressure leads to rod cell shortening due to uncoupling between bacterial growth and division

To decipher how mechanical confinement affects bacterial physiology, we characterized its impact on bacterial cell shape using the inner membrane marker described above (Fig. 2A top and Supplementary Video 3). By simultaneously monitoring growth-induced pressure and cell size, we found that the single cell area decreased upon confinement (Fig. 2A bottom and Supplementary Video 3). Averaged bacterial area rapidly dropped at the onset of pressure build-up, reaching a stable minimal area of 0.5 µm² 2 h later, corresponding to a 4-fold decrease (Fig. 2B and Supplementary Fig. 3A, B). The size reduction was stronger than the starvation-induced shortening observed on agar pads and in mother machines¹⁴, suggesting that it differs from stationary-phase effects and is instead mediated by mechanical confinement and pressure build-up (Fig. 2C). We hypothesized that it might be due to an uncoupling between growth and division. To test this, we quantified bacterial length and width over time (Supplementary Fig. 2D). We found that this morphological transition was mostly due to a 75% decrease in bacterial length (Fig. 2D and Supplementary Fig. 3C), supporting our hypothesis. To go further, we imaged bacterial proliferation at higher temporal resolution to reconstruct single-cell lineages in the 2D plane of observation³⁵, quantified bacterial growth and division rates for each lineage, and averaged them across the entire cell population (Fig. 2E, F and Supplementary Figs. 2E, F, 3D, E). By using bacterial number as a readout for the onset of pressure build-up (Supplementary Fig. 2C), we found that while bacterial growth rate rapidly decreased at the onset of confinement, division rate persisted for a period of about 30 min (Fig. 2F). In other words, bacteria continued dividing for about 2 cell cycles while their growth was almost completely arrested, leading to cell shortening (Fig. 2E, F and Supplementary Fig. 3D). The division rate then decreased during the following 90 min, reaching a state where bacteria dramatically decreased their growth and division. Of note, while this trend was consistent across the chamber, spatial heterogeneities were observed between bacteria at the center and those at the edge facing the nanochannels (Supplementary Fig. 3F, G). In addition, these morphological changes were reversible upon pressure release, with bacteria recovering initial dimensions within 1–2 h (Supplementary Fig. 3H, I). Together, these data show that in E. coli, growth-induced pressure initially caused a diminution in bacterial cell size through the persistence of division, as previously reported for mammalian epithelial cells in a confluent monolayer^19,36. Our results identified 3 phases that followed growth-induced pressure build-up, defined as the onset of confinement (Fig. 2G): (1) bacterial growth declined rapidly while division persisted, causing a reduction in cell length; (2) division rate decreased, and finally (3) growth ceased, and division was nearly abolished while growth-induced pressure kept increasing, confirming accessibility of nutrients in the chambers at these late stages. This raised the question of bacterial division in the absence of volumetric growth.

Fig. 2. Growth-induced pressure leads to rod cell shortening due to uncoupling between bacterial growth and division.

A Timelapse confocal images of proliferating E. coli MG1655 bacteria fluorescently labeled at the inner membrane in the bacterial confiner (top). Corresponding colormaps of single-cell area (bottom). The phases before confinement (t = −2 h), at the time at which GIP builds up (t = 0 h), 1 h, and 6 h after GIP build-up are shown in a–d, respectively (Supplementary Video 3). Corresponding unconfined controls are shown in Supplementary Fig. 3A–D. B Quantification of single-cell area over time (n_bacteria > 47000, n_chambers = 4, N = 2). Letters refer to the phases depicted in (A). Colored regions correspond to the Before and Late confinement categories shown in (C). Black dashed line indicates t = 0 of confinement onset. C Quantification of single-cell area in the following conditions: exponential (n_bacteria = 482, N = 3) and stationary phases (n_bacteria = 410, N = 3) in agar pad, before (i.e., following bacterial loading in the chambers; n_bacteria = 65, n_chambers = 4, N = 2) and upon late confinement (n_bacteria = 1721, n_chambers = 4, N = 2) in the bacterial confiner. Statistical significance was assessed using an ordinary one-way ANOVA test. Experiments were repeated at least 3 times independently with consistent results; morphological analyses were performed on a representative dataset. D Quantification of bacterial length and width normalized by their average value before confinement (n_bacteria > 47000, n_chambers = 4, N = 2). Black dashed lines indicate t = 0 of confinement onset. E Zoom-in of timelapse confocal images acquired at higher temporal resolution during proliferation of E. coli in the bacterial confiner. Sequential divisions are indicated by white arrowheads. Colormaps represent single-cell area (Supplementary Video 3). F Quantification of single-cell growth and division rates over time. The gray shaded area indicates Phase 1, during which two complete division cycles occur in the absence of growth (n_bacteria = 26728, n_tracks > 800, n_chambers = 2, N = 1). G Schematics of the confinement-induced uncoupling between bacterial growth and division. Data points correspond to mean values ± standard deviations. Time is indicated as hh:mm. All scale bars: 5 µm.

Cytoplasmic crowding increases during Phases 1 and 2 and is sufficient to explain growth and division uncoupling

We next explored the impact of confinement on bacterial physiology at the subcellular scale, focusing on proteins and DNA. At the protein level, we investigated the impact of pressure build-up using two fluorescent reporters, one under the control of a constitutive promoter (P_R-meGFP) to have a global readout of protein transcription and another with the promoter of the ftsZ gene, a key component of the division machinery (P_ftsZ-FtsZ-mNeonGreen³⁷) (Fig. 3A, Supplementary Fig. 4A, B, and Supplementary Video 4). We found that in both cases, cytoplasmic mean fluorescence intensities, proportional to protein concentration, increased in the first two phases of confinement (Fig. 3B and Supplementary Fig. 4C–E). By computing the protein production rate, defined as the variation of the total fluorescent intensity at the scale of the chamber divided by the number of bacteria, we found that synthesis of both GFP and FtsZ declined at the onset of confinement, finally reaching a stalled state in Phase 3 (Fig. 3C and Supplementary Fig. 4F). However, while the volumetric growth rate decreased rapidly upon confinement onset, protein production was sustained for a longer time. This led to a non-specific increase in protein concentrations in the cytoplasm of confined bacteria during Phases 1 and 2, as demonstrated by the global rise in fluorescence intensities. Finally, protein production rate was largely halted, up to a level where degradation of an unstable protease-sensitive GFP became dominant over synthesis (Supplementary Fig. 4G–J and Supplementary Note 1).

Fig. 3. Cytoplasmic crowding increases during Phases 1 and 2 and is sufficient to explain growth and division uncoupling.

A Representative timelapse confocal images of proliferating E. coli MG1655 bacteria expressing cytoplasmic, constitutive GFP in the bacterial confiner (Supplementary Video 4). Corresponding unconfined control is shown in Supplementary Fig. 4B. B Temporal evolution of single-cell mean GFP intensity (n_bacteria = 50661, n_chambers = 4, N = 1) with respect to GIP build-up. C Quantification of total GFP production rate normalized by its value at time 0 of GIP build-up (dark green, n_chambers = 4, N = 1) and corresponding exponential fit (black, typical time τ = 33 min). Bacterial growth rate (gray) is computed on another dataset shown in Fig. 2D. D Representative images of E. coli bacteria fluorescently labeled at the inner membrane (gray) and DNA (blue) proliferating in the confiner (Supplementary Video 4). Insets zoom in on DNA occupancy in the cytoplasm. Corresponding unconfined control is shown in Supplementary Fig. 4K. E Schematics illustrating the hypothesis of an increase in protein concentration (shift from light to dark green) and DNA occupancy (blue) upon confinement and rod shortening (top). Temporal evolution of karyoplasmic ratio with respect to GIP build-up (bottom) (n_bacteria = 26894, n_chambers = 2, N = 1). F Representative images of E. coli bacteria expressing 40 nm-GEMs diffusive nanoparticles in their cytoplasm in the absence of confinement (left) and in Phase 2 (right) (Supplementary Video 4). 40 nm-GEMs trajectories are color-coded by trajectory duration (red: short; white: long ~2.5 s). Insets zoom into individual trajectories, with white arrowheads marking representative examples. Scale bars insets: 1 µm. G 40 nm-GEMs effective diffusion coefficient in different conditions: in agar pad (n_tracks = 1020, N = 2), before confinement (n_tracks = 1120, N = 3), Phase 1 (n_tracks = 444, N = 3), Phase 2 (n_tracks = 1431, N = 3), and Phase 3 (n_tracks = 2517, N = 3). The diffusion value of 0.015 µm²/s corresponds to the higher level of crowding reached (red). Datasets are represented as violin plots with median values and interquartile range indicated. Statistical significance was assessed using Kruskal-Wallis tests. H Schematics of a bacterium submitted to a hyperosmotic shock. I Schematics showing crowding increase after two successive hyperosmotic shocks (0.5 M then 1 M sorbitol, 5 min apart). J Representative timelapse images of E. coli fluorescently labeled at the inner membrane submitted to hyperosmotic shocks. White arrowheads indicate division sites. Time is indicated as mm:ss. K Fraction of dividing cells in response to hyperosmotic shocks (per condition: n_bacteria > 204, N = 4). L Fraction of dividing cells upon confinement (n_bacteria = 26728, n_tracks > 800, n_chambers = 2, N = 1). Note that the increase in dividing cells slightly precedes GIP build-up, likely because chamber deformation is measured at lower temporal resolution than bacterial division. Data points correspond to mean values ± standard deviations; Time is indicated as hh:mm; Scale bars: 5 µm, unless otherwise stated.

At the DNA level, we explored whether the initial persistence of bacterial division in the absence of growth could lead to higher DNA occupancy in the cytoplasm. For this, we used a HU-GFP fusion to measure the karyoplasmic or N:C ratio, defined as the ratio between DNA and cell area³⁸ (Supplementary Fig. 2D). We found that confined bacteria exhibited a 20% increase in karyoplasmic ratio during the first two phases of confinement compared to control, non-confined bacteria (Fig. 3D, E, Supplementary Fig. 4K, L, and Supplementary Video 4). This meant that pressure build-up induced the formation of small bacteria with a cytoplasm largely occupied by the nucleoid. Together, these observations on both proteins and DNA suggest that confined bacteria display higher levels of cytoplasmic crowding, as previously shown in eukaryotes¹⁷.

To directly evaluate cytoplasmic apparent viscosity, we expressed 40 nm-wide genetically encoded multimeric nanoparticles (40 nm-GEMs) that freely diffuse in the bacterial cytoplasm and tracked them³⁹ (Fig. 3F, Supplementary Fig. 4M, and Supplementary Video 4). In line with our previous results, we found that 40 nm-GEMs diffusion slightly decreased in the first phase of confinement and more steeply in the second phase (Fig. 3G), confirming confinement-induced crowding in bacteria. Collectively, these results demonstrate that, in the presence of nutrients, arrest of volumetric growth was associated with a highly crowded bacterial cytoplasm mediated by protein and DNA accumulation.

We next hypothesized that the progressive increase in cytoplasmic density upon confinement onset could first lead to division induction, because of an increase in protein concentration, and then to division arrest because of excessive crowding. Previous studies suggest that the accumulation of proteins of the divisome is sufficient to trigger bacterial division, both in steady-state conditions and upon hyperosmotic shock^40,41. By tracking 40 nm-GEMs particles in bacterial cells subjected to a range of sorbitol concentrations, we confirmed that hyperosmotic shock led to an increase in cytoplasmic crowding by the rapid increase in concentration of all proteins (Fig. 3H and Supplementary Fig. 4N). To mimic a progressive increase in crowding and inspired by a previous study⁴¹, we performed two successive hyperosmotic shocks at increasing sorbitol concentrations (0.5 M and 1 M) and monitored bacterial divisions (Fig. 3I). We observed that while bacteria often divided a few minutes after the first shock, they rarely divided after the second shock (Fig. 3J, K). Thus, a progressive increase in osmotically-induced crowding reproduced the confinement-induced change in bacterial division characterized by an increase in the fraction of dividing cells followed by a sharp decrease (Fig. 3K, L). These results suggested that, while an initial increase in protein concentration was sufficient to trigger bacterial division, consistent with the role of the concentration of proteins of the divisome to drive division, a further increase in protein density rather inhibited bacterial division. Interestingly, despite high karyoplasmic ratios (Fig. 3D, E and Supplementary Fig. 4O, P), the reduction in division rate in mechanically confined bacteria was not regulated by the nucleoid occlusion regulator SlmA⁴² (Supplementary Fig. 4Q–S). Therefore, these data support that cytoplasmic crowding regulates bacterial division upon confinement through physical means, by increasing protein concentrations, and then preventing protein diffusion and/or protein synthesis as proposed in yeast¹⁷, ultimately leading to the formation of non-growing and non-dividing, highly crowded bacteria. In this state, confined bacteria appear to have reached a quasi-frozen stalled state in terms of division and protein synthesis (Figs. 2F and 3C). Nevertheless, unexpectedly, we observed that the growth-induced pressure kept increasing steadily for hours after both protein synthesis and division rate had almost completely ceased (Fig. 1D). We propose to name this third phase of confinement the overpressurization regime.

Theoretical modeling points to a central role for cell anabolism in the overpressurization regime

To understand our surprising observation of steady pressure increase in confined bacterial colonies with reduced protein synthesis, we built a minimal theoretical model of bacterial growth that leverages recent knowledge in mechano-osmotic regulation^43–46 (Fig. 4A and Supplementary Note 2). We modeled confining chambers with a rectangular shape that deformed elastically in 3D due to the mechanical pressure exerted by bacterial growth (Supplementary Fig. 5A–D). Average bacterial cell shape was approximated by cylinders (Fig. 4B). Bacterial cell volume was determined by the balance between the osmotic pressure and the mechanical pressure difference throughout the bacterial envelope⁴⁷. This model recapitulates basic experimental observations. Initially, in the absence of confinement, growth-driven accumulation of osmolytes increases cell volume to ensure size homeostasis. Once bacteria completely fill the 3D space of the chamber, the intracellular osmotic pressure is balanced by the chamber walls (Supplementary Fig. 5A–D). Of note, 3D confluency (i.e., Filling time) is delayed by 30 min from 2D confluency (i.e., Confluency, as defined in Fig. 1B). According to this model, in the confined regime, and assuming homogeneous bacterial density and behavior (Supplementary Fig. 2A), growth-induced deformation of the PDMS chamber walls (3% of its initial size, Supplementary Note 2) provides a direct measurement of the quantity of trapped osmolytes inside bacteria. To note, because the chamber is sufficiently stiff, once confinement is established and growth-induced pressure can be measured, further pressure increases are accommodated by only minimal volumetric expansion, leading to an increased concentration of osmolytes rather than continued cell inflation.

Fig. 4. Theoretical modeling points to a central role for cell anabolism in the overpressurization regime.

A Schematics of the theoretical model for bacterial growth in a 3D rectangular space (L_x^c = 20 µm, L_y^c = 30 µm, L_z^c = 2.6 µm, projected area A_c) (left), which deforms elastically due to GIP (right). B Bacteria are modeled as cylinders of width w and length l(t) enclosed in a shell δ modeling the cell envelope (left). Osmotic pressure arises from the production of osmolytes, including proteins and small osmolytes (i.e., metabolites and counterions), whose synthesis depends on proteins (white and black curved arrows, resp.) (right). Protein production rate (g_P) depends on cytoplasmic crowding and defines bacterial growth (g_A) and division rates (g_D). C Bacterial growth upon confinement is modeled from time t₀, defined by theoretical 2D confluency (top). Theoretical t₀ (called Filling time) is delayed of 30 min compared to the experimental one (written Confluency) (Supplementary Fig. 5.A, B). This initial period is interpolated and shown as dotted lines. Experimental data (gray) and theoretical fits (blue) of GIP (top) and 2D number of bacteria (bottom) were used to set adjustable parameters. D Theoretical prediction (red) and experimental curve (gray) of bacterial area over time under confinement. E Theoretical prediction (red) and experimental curve (gray) of bacterial growth rate over time. F Theoretical prediction (red) and experimental curve (gray) of the bacterial division rate over time (Supplementary Note 2, Eq. 20). G Theoretical prediction of protein (g_P, dark green) and osmolyte normalized production rates (g_N, orange) over time. H Theoretical predictions of bacterial growth rate (magenta), division rate (light green), protein (dark green) and osmolyte normalized production rate (orange) over the first 4 h post confinement. I Schematics of Phases 1 and 3 illustrating the distinct effect of crowding on proteins and small osmolytes. In Phase 1, both are produced (white and black arrows, resp.) and freely diffuse in the cytoplasm (dotted green and orange arrows, resp.). In Phase 3, crowding impairs protein but not osmolyte synthesis, leading to overpressurization of the cytoplasm. J Schematics of the live cell enzymatic assay to assess metabolic activity (Methods). After entering the cell, CCF2/AM emits green fluorescence via FRET from the excited coumarin donor to the fluorescein acceptor (uncleaved state). Cells expressing β-lactamase convert CCF2/AM into CCF2, disrupting FRET and producing a blue fluorescence signal, indicating active enzymatic conversion (cleaved state). K Temporal evolution of the coumarin mean fluorescence intensity (MFI) over the entire chamber for bacteria expressing (blue) or not (gray) cutinase and β-lactamase, under constant perfusion of CCF2/AM (n_chambers = 8, N = 2). The increase of the blue signal at late stages of confinement indicates that cytoplasmic enzymes remain functionally active. Results were validated upon perfusion of CCF2/AM at the start of Phase 3 (Supplementary Fig. 5K). L Schematics of switch to medium supplemented with the translation-inhibitor antibiotic (Ab) spectinomycin at the entry of Phase 2. M Temporal evolution of GFP MFI per cell in confined control chambers (black), and upon spectinomycin treatment (green). N GIP in confined control chambers (black) and upon spectinomycin treatment (green) (n_chambers = 7, N = 1). Data points correspond to mean values ± standard deviations.

The model also recapitulates observations at the protein level: because proteins are initially produced at a normal rate while cell volume increase is constrained by the chamber walls, protein concentration increases. In the model, we implemented bacterial divisions for a doubled protein content, as suggested by prior work⁴⁰. This results in smaller cells, as observed experimentally in the first confinement phase. To account for the second phase, we incorporated the well-established notion that protein accumulation is accompanied by cytoplasmic crowding, leading to a decrease in protein production rate as observed experimentally in this work and reported by recent studies in yeast^16,17. Prior models relating pressure directly to the protein content would predict that once cytoplasmic crowding halts protein accumulation, growth-induced pressure should also stop increasing, which is not what we observed in the third phase of overpressurization.

Here, we introduced a key novel feature in the theoretical framework that decouples growth in volume (i.e., intracellular trapped osmolytes accumulation) and in mass (i.e., protein accumulation) based on recent work^44,48. This is made possible by the decoupling of two types of osmolytes (Fig. 4.B): (1) small osmolytes, i.e., metabolites and counterions, which dominate the intracellular trapped osmolytes due to their large numbers inside cells⁴⁹; and (2) proteins, whose contribution is mostly steric as they have a minor direct contribution to intracellular osmotic pressure due to their low numbers. The complete model is based on a simple force balance complemented by three ingredients: (1) cell division depends on the doubling of protein number⁴⁰; (2) protein accumulation depends on protein concentration with a saturation effect due to crowding¹⁷; (3) intracellular trapped osmolytes, responsible for the growth-induced pressure, are dominated by small biomolecules produced by proteins, with an accumulation rate that depends on the anabolic activity of cells (that we assume to be proportional to the total number of proteins in the cell for simplicity). This third ingredient is sufficient to predict the overpressurization regime, because instead of coupling the amount of osmolytes, and thus the pressure, to the protein content, it couples the rate of increase in the amount of osmolytes to the protein content. As a consequence, cells with a constant number of proteins would then increase their pressure linearly, as we observed experimentally in the overpressurization regime. This is reminiscent of the linear volume increase in cell cycle-arrested yeast cells⁴³, which was explained using the same model⁴⁴. In conclusion, with these simple features, the model accounts for the three phases of confinement.

Since our model relies on a small number of explicit parameters that can be measured experimentally, we were able to test it quantitatively. We first fitted three parameters using experimental growth-induced pressure measurements and the number of bacteria measured at the bottom of the chamber (Fig. 4C, Supplementary Note 2). We then produced predictions without any adjustable parameter on the other independent datasets at our disposal, namely bacterial area, growth and division rates, and diffusion of 40 nm-GEMs particles (Fig. 4D–F and Supplementary Fig. 5E). Notably, in Phase 3, the growth rate is non-zero but reaches very small values that fall within the error bars and are therefore below the detection limit (Figs. 4F and 2E). The quantitative agreement between theoretical predictions and experimental measurements validated our model, allowing us to define the different regimes of confined bacterial growth in physical terms (Fig. 4G–I and Supplementary Fig. 5F–J). Monitoring of metabolic activity with a live-cell fluorescence-based β-lactamase activity assay provides experimental evidence that, in Phase 3 of confinement, cytoplasmic enzymes remain functionally active, as predicted by the theoretical model (Fig. 4J, K and Supplementary Fig. 5K)^50,51. This is in agreement with the strict requirement of nutrient perfusion for pressure build-up in Phase 3. A key prediction of the model is that inhibition of protein synthesis between Phases 1 and 2 should not significantly affect the rise in growth-induced pressure (Supplementary Fig. 5F). To validate this, we perfused confined chambers with a sub-MIC concentration of the translation-inhibitor spectinomycin at the entry in Phase 2⁵² (Fig. 4L and Supplementary Fig. 5L, M). Confined bacteria exhibited a marked reduction in protein accumulation, confirming effective translation inhibition (Fig. 4M), while growth-induced pressure gradually rose over several hours to levels similar to untreated conditions (Fig. 4N). These experiments validate the model prediction that pressure increase is driven not by de novo protein synthesis, but by osmolyte production catalyzed by enzymes and pre-existing metabolically active proteins. Based on these findings, we reasoned that, by decoupling growth and essential biological processes such as protein and osmolyte production, mechanical confinement induces a unique bacterial state during which bacteria undergo an increase in their osmotic pressure in the absence of growth, a situation that can be referred to as overpressurization.

Rcs transcriptional response to mechanical confinement is required for shape maintenance in the overpressurization regime

By analogy to well-known mechanosensors in eukaryotes⁵³, we hypothesized that transcriptional adaptation to elevated turgor pressure may rely on the bacterial envelope. We focused on the Rcs and Cpx pathways as they sense defects at the outer and inner membrane/peptidoglycan layer, respectively^26,54,55. In addition, the mechanosensory ability of Rcs regulates the activation of the envelope stress response^27,28. To assess the activation of the Rcs pathway upon confinement, we used the P_rcsA-GFP transcriptional reporter⁵⁶. We found that the Rcs pathway was activated at the onset of confinement as soon as the pressure built up in the chambers (Fig. 5A, B, Supplementary Fig. 6A–C, and Supplementary Video 5). Of note, although Rcs was activated throughout the chamber, its expression was more pronounced at the chamber edges near the nanochannels. Similarly, the Cpx envelope stress response was transcriptionally activated upon confinement (Supplementary Fig. 6D–F and Supplementary Video 5), whereas the RecA pathway, induced upon DNA damage, was not (Supplementary Fig. 6G–I and Supplementary Video 5). To decipher the role of the Rcs pathway in bacterial adaptation to confinement, we used a bacterial strain deficient in the Rcs response regulator rcsB. Strikingly, we found that, while the rcsB mutant exhibited a normal rod shape in the absence of confinement (Supplementary Fig. 6J), it lost its shape in the overpressurization regime and produced inflated cells at the later stages of confinement (Fig. 5C, Supplementary Fig. 6K, and Supplementary Video 6), reaching unusually large cellular surface areas (Fig. 5D). The morphological changes in the rcsB mutant were most striking in the areas of the chamber where transcription of the Rcs regulon was the highest in wild-type cells (Fig. 5B). By contrast, a mutant deficient in the Cpx envelope stress response, cpxR, displayed only classical rod-shaped morphologies (Supplementary Fig. 6L and Supplementary Video 6). Synthesis of a colanic acid capsule played a partial role in shape maintenance as the mutant wcaJ displayed a subtle morphological phenotype upon pressure build-up in the microfluidic device (Supplementary Fig. 6M–P and Supplementary Video 6).

Fig. 5. Rcs transcriptional response to mechanical confinement is required for shape maintenance in the overpressurization regime.

A Representative timelapse images of proliferating E. coli labeled at the inner membrane (gray) and expressing the Rcs transcriptional reporter P_rcsA-GFP (red) in the confiner. The spatial pattern of reporter intensity is illustrated by a yellow dotted line (Supplementary Video 5). Corresponding unconfined control is shown in Supplementary Fig. 6B. B Kymograph along the x-axis of P_rcsA-GFP induction upon confinement using a Fire color scale (top). Percentage of bacteria which activate the Rcs stress response respect to GIP build-up (bottom). Mean values ± standard deviations are represented (n_bacteria = 35200, n_chambers = 3, N = 1). C Representative timelapse images of proliferating E. coli rcsB mutant labeled at the inner membrane (gray) in the confiner (Supplementary Video 6). Insets highlight aberrant bacterial morphologies marked with a white arrowhead. Corresponding unconfined control is shown in Supplementary Fig. 6J. D Bacterial area in Phase 3 for the WT and rcsB strains at the center (blue) and at the edges (red) of the chamber. Center and Edges regions refer to the spatial pattern depicted in (A). Data points and means are represented (per condition: n_bacteria ≥ 3928, n_chambers ≥ 2, N = 1). Statistical significance was assessed using ordinary one-way ANOVA tests. E Proposed model of E. coli growth upon confinement: (i) In the absence of spatial limitation, cell growth, division, protein, and osmolyte synthesis are tightly coupled to maintain size homeostasis. When bacteria face a lack of space: (ii) In Phase 1, growth rapidly declines (purple), increasing protein concentrations (green) and triggering accelerated cell divisions (black). This leads to more but smaller cells. (iii) In Phase 2, protein accumulation and DNA occupancy (blue) strongly increase cytoplasmic crowding (dark red), halting protein synthesis and cell division, while pressure increases steadily. (iv) In Phase 3, protein synthesis stops yet pressure continues to increase. Our model attributes this to continuous accumulation of trapped osmolytes driven by protein activity. In these conditions, Rcs-mediated envelope remodeling (red) is required to maintain cell shape during overpressurization. (v) At the global scale, this increase in turgor pressure generates large GIP onto the microenvironment (orange). Time is indicated as hh:mm. All scale bars: 5 µm.

These findings demonstrated that the Rcs pathway plays a key role in maintaining cell shape during prolonged confinement, potentially by remodeling the envelope to counteract elevated turgor pressure, as predicted by our theoretical model. We propose a simple mechanistic working model for confined bacterial growth, leading to the overpressurization regime (Fig. 5E): at this stage, activation of the Rcs pathway is involved in cell shape maintenance by strengthening the cell envelope in order to resist the increasing pressure. This suggests that cells can survive the overpressurization regime only by strengthening their envelope.

Validation of confinement-induced phenotypes in a uropathogenic strain

To explore the relevance of these results in the context of infection, we turned to a uropathogenic E. coli (UPEC) strain, UTI89, derived from a patient with a urinary tract infection (UTI)⁵⁷. These bacteria invade the urothelium, leading to the formation of tightly packed intracellular bacterial communities (IBCs)^29–31. Growth inside urothelial cells is accompanied by cell shape changes reminiscent of the observations with confined E. coli K12^31,32,58. Therefore, we hypothesized that UPEC might experience mechanical confinement during infection.

Experiments in the bacterial confiner showed that, similar to the K12 strain MG1655, UPEC built up large growth-induced pressure while undergoing major changes in cell size and shape (Fig. 6A–E). Cell shape changes were also reversible upon pressure release in the chambers (Supplementary Fig. 7A and Supplementary Video 7), echoing the morphological plasticity described during UTI upon cell exit³¹. By fitting the experimental parameters, we found that the key model prediction holds true for UPEC too, suggesting that the overpressurization regime could be a general mechanism of bacterial adaptation to confinement (Supplementary Fig. 7B–E). Strikingly, deletion of rcsB in UTI89 also resulted in large, amorphous cells in vitro (Fig. 6F, G). These results show that the phenotype of pressure build-up and transcriptional adaptation we characterized in the K12 model strain applies to the uropathogenic UTI89 strain.

Fig. 6. Validation of confinement-induced phenotypes in a uropathogenic strain.

A Representative timelapse images of proliferating uropathogenic E. coli UTI89 fluorescently labeled in the cytoplasm in the bacterial confiner (Supplementary Video 7). B GIP generated by the MG1655 (gray, data shown in Fig. 1D) (n_chambers = 13, N = 3) and UTI89 (purple) (n_chambers = 5, N = 2) strains over time. C Corresponding number of bacteria over time. In the case of UTI89, a representative dataset corresponding to (n_chambers = 3, N = 2) was analyzed. D Temporal evolution of single-cell area for the UTI89 strain (n_chambers = 3, N = 2). E Bacterial area upon late confinement for the MG1655 (n_bacteria = 695, n_chamber = 1, N = 1) and UTI89 strains (n_bacteria = 701, n_chambers = 1, N = 1). Statistical significance was assessed using an ordinary one-way ANOVA test. Experiments were repeated at least 3 times independently with consistent results; morphological analyses were performed on a representative dataset. F Representative timelapse images of proliferating UTI89 rcsB mutant fluorescently labeled in the cytoplasm in the bacterial confiner. Insets zoom in on aberrant morphologies marked with a white arrowhead. G Bacterial area in Phase 3 for the WT and rcsB UTI89 strains at the center (blue) and at the edges (red) of the chamber. Data points and means are represented (per condition: n_bacteria ≥ 2000, n_chambers ≥ 2, N = 1). Statistical significance was assessed using ordinary one-way ANOVA tests. Unless otherwise stated, data points correspond to mean values ± standard deviations. All scale bars: 5 µm.

Thus, UPEC exhibits morphological changes similar to those observed in the MG1655 strain upon confinement in our microfluidic chambers, consistent with the model’s predictions and matching the mean bacterial size reported in middle IBCs in vivo³¹. These results highlight the role of the overpressurization regime in adaptation to pressure build-up across bacterial strains. Finally, we found that the Rcs pathway is also involved in cell shape maintenance and bacterial fitness in the pathogenic strain, enabling resistance to increasing pressure, with potential implications as a therapeutic target for urinary tract infections.

Discussion

Using a novel microfluidic device adapted to the study of confined growth of bacterial colonies in the presence of nutrient perfusion, we identified a regime of quasi-complete growth arrest, stalled division, and repressed protein synthesis during which turgor pressure steadily increased, reaching hundreds of kPa, that we named the overpressurization regime. We found that the Rcs pathway is activated in this regime and that this activation is required for mechanical adaptation to turgor pressure build-up for commensal and uropathogenic strains.

During the different confinement phases, bacteria steadily build up a large compressive pressure, ranging from a few tens of kPa up to 300 kPa, in the range of values previously reported for turgor pressure in individual cells^59–61, in confined bacterial colonies^11,62, and in biofilm growth on adhesive substrates⁹. Confinement of eukaryotic cells also generates growth-induced pressure, with magnitudes scaling with osmotic pressures exerted on the cell wall in the case of yeast and on the actin cortex in mammalian cells^16,18,19. These results converge to support the idea that the main source of confinement-induced mechanical stress produced by a multicellular colony is the osmotic pressure generated by single cells. The range of growth-induced pressure values generated by a confined bacterial colony compares more to osmotic pressures generated by cells on ingested materials⁶³, than to forces encountered in mammalian tissues, which are typically in the 1–10 kPa range^64,65. We thus anticipate that it could have far-reaching consequences for intracellular bacterial infections of host tissues.

While in homeostatic conditions, bacterial growth and division were coupled to maintain a constant cell size, in the first phase of confined growth, the two were uncoupled, leading to major changes in cell volume. The limited space and stiff walls impose a sharp slowdown in bacterial growth. In the meantime, bacteria rapidly underwent multiple cycles of reductive divisions to finally arrest at a minimal cell size. This phase was highly reproducible across strains, despite some variability in the preceding phase, which sometimes showed a filamentation-like increase in cell length. Whether shear stress or quorum-sensing partially contributes to this response will be further addressed in future work. Rod shortening was reminiscent of stationary-phase cells; however, in this case, nutrients were constantly flowing, and bacteria generated increasing forces, unlike quiescent cells^14,66,67. While the impact of confinement on growth rate appears general to all living systems explored so far^16,68,69, its effects on the cell division cycle are very diverse. Budding yeast maintains a constant cell volume under mechanical pressure by arresting division when growth slows down¹⁶. Epithelial cells, similar to our observations in bacteria, display an uncoupling between growth and division rate in dense monolayers, which also induces a global reduction of cell volume, reaching a minimal value defined by genome size^19,36. While the increase in karyoplasmic ratio is detrimental in the epithelium, leading to DNA damage, we showed that E. coli had an extreme plasticity to changes in DNA occupancy and rapidly returned to their normal size and division cycle upon pressure release without DNA damage^36,70. These results point to the remarkable adaptation of bacteria to growth under pressure, in the same line as previous studies¹¹⁻¹³.

We found that one key feature of bacterial confinement at short timescales was an increase in intracellular molecular crowding, occurring primarily in Phases 1 and 2. Cytoplasmic crowding influences the activity of proteins and macromolecular complexes in all cells, including bacteria^71,72. We showed that at early stages of confinement, protein concentrations increased, and this process, reproduced by osmotic shocks, was sufficient to induce an uncoupling between division and growth. This was explained by a general increase in protein concentration, including those involved in the divisome machinery, such as FtsZ. Since osmotic shock and mechanical confinement have opposite effects on turgor pressure, this rules out a direct role of turgor pressure in division^40,41,59. Later, an increase of DNA occupancy mediated transition of the cytoplasm to a state similar to the previously described “colloidal glassy” state upon ATP or nutrient depletion⁷³, but in our case, triggered by a mechanical cue. Another layer of complexity arises from the distinct effects of osmotic shocks and mechanical confinement on nucleoid occupancy, raising the question of how cytoplasmic properties converge to a similar effective crowding.

An apparent paradox in our results concerned the increase in growth-induced pressure at late stages of confinement (Phase 3) while protein production was already arrested, raising the question of the origin of this sustained increase in pressure. We show that while proteins cease to increase in concentration, enzymes remain active (i.e., metabolism), leading to osmolyte accumulation and turgor increase, i.e., the overpressurization regime. This idea, supported by monitoring of enzymatic activity during bacterial confinement and the minimal theoretical framework developed in this study for two distinct bacterial strains, points to the mechano-osmotic role of metabolism. Unlike proteins whose role in growth has been the focus of several studies in the past few years, the osmotic contribution of small membrane-impermeable osmolytes such as metabolites (i.e., glutamate among others) to cell volume is a new concept in the field and an emerging mechanism for cell size control^43–45,48. Further studies combining multiplexed omics with next-generation microfluidic chips that enable confinement of large bacterial populations and cell retrieval at defined stages will make it possible to identify the small metabolites involved (i.e., glutamate and/or others), determine how their synthesis is regulated by cytoplasmic crowding, and assess their impact on growth-induced pressure under confinement^74,75. The late saturation in growth-induced pressure (>10 h) is not directly explained by our model and could have a variety of causes, such as passive or active shutdown of anabolic activity, which is out of the scope of this study. Nevertheless, theoretical modeling provided a convincing mechanistic explanation for our observations and highlighted the importance of cell metabolism in producing a persistent pressure increase in confined non-growing bacteria, consistent with recent models of turgor pressure regulation in exponentially-growing cells⁷⁴. Future work will incorporate the complex dynamic coupling between cell division and protein synthesis throughout confinement—when major shifts in bacterial metabolism and mechanics occur—to more accurately predict the overpressurization levels reached by bacteria at later stages.

A major prediction of the theoretical model is that confined bacteria, if they are constantly provided with nutrients and able to accumulate trapped osmolytes, will experience a massive internal turgor pressure increase while maintaining a proper morphology. We found that this activates stress responses, with a crucial role of the Rcs pathway in morphological robustness under mechanical confinement^{26–28,54,76}. Strikingly, in the absence of the rcsB-dependent transcriptional response, overpressurized (Phase 3) cells display large and heterogeneous shapes reminiscent of L-forms^77,78. The mild phenotype observed in the capsule-deficient mutant wcaJ is consistent with previous findings, suggesting a contribution of this structure to global cell mechanics²⁸ (Supplementary Fig. 6M–P). The transcriptional regulation of other components of the bacterial envelope by the Rcs pathway, acting as a mechanosensory system, such as peptidoglycan and outer membrane organization, are likely the main contributors to cell shape maintenance in the overpressurization regime^{24,54,75,76,79}. Interestingly, we found that the Cpx stress response was also transcriptionally activated during confined growth, but the corresponding mutant did not display any major shape defect^26,54. Activation of the Rcs pathway is therefore strongly correlated with bacterial adaptation to pressure increase during confinement. Interestingly, its activation is very fast and takes place at low pressure, suggesting that even if bacteria build up large pressures in the 100 kPa range, they can respond to stresses below 10 kPa that are relevant in terms of mammalian physiology. Future work will focus on developing novel imaging approaches to better characterize bacterial cell envelope structure and mechanics, including capsule dynamics and cell wall remodeling during confinement.

An interesting aspect of this global transcriptional reprogramming that we did not investigate in detail is that the spatial pattern of activation is not homogeneous throughout the chambers and is specific to each stress response pathway. This could be due to a mechanochemical coupling, with bacteria at the edges of the chamber having more access to nutrients compared to bacteria at the center. Other factors could contribute to this heterogeneity, including local differences in compressive forces exerted on bacteria and in shear stress. We indeed observed significant differences in division and growth rates for bacteria in these two locations, which finally resulted in a global growth arrest within the chamber (Supplementary Fig. 3F). These potential heterogeneities in nutrient access, metabolic activity, exposure to shear, and stress distribution might also explain the morphological defects of rcsB mutant strains at the chamber edges, leading to cell shape instabilities once the bacterial envelope is not rigid enough to counterbalance the increase in turgor pressure. However, we emphasize that mechanical confinement is strictly required to trigger Rcs activation and the associated abnormal morphologies in this mutant (Supplementary Fig. 6B, J). Neither nutrient limitation nor shear stress alone produced comparable phenotypes. Further studies with our confinement device, for instance, by using static confinement chambers or varying flow rates, as well as by tuning chamber dimensions and geometry, will allow a deeper understanding of the complex interplay between different stress responses and how integration of mechanical signals coordinates multicellular growth and cell fate specification in bacterial biofilms, as recently shown during organogenesis²⁰. It might also reveal other pathways required for bacterial survival during confined growth. In this work, we show that under conditions such as physical confinement, where cell volume expansion is constrained, and nucleoid occupancy is increased, turgor pressure can promote intracellular crowding and a glassy cytoplasmic state that slows diffusion, potentially impacting signaling dynamics (Fig. 3F, G)⁷³. It may also enhance efflux pump activity and strengthen envelope barriers (Fig. 5A, B), collectively reducing antibiotic susceptibility and improving survival under stress⁷⁵. Although a complete characterization is outside the scope of this study, our results indicate broader physiological consequences of overpressurization and provide a basis for future research.

Importantly, uropathogenic E. coli also underwent rod shortening during confined growth, a feature previously described during bladder infection^29,31. The agreement between bacterial cell dimensions observed in the confiner and inside IBCs in vivo, as reported in previous studies³¹, strongly suggests that these bacteria are experiencing mechanical confinement and need to resist the osmotic pressure exerted by the host cell (estimated to be about 300 kPa for a typical trapped osmolyte concentration of 120 mM), especially if they are inside endocytic compartments⁶³. Therefore, bacterial confinement during infection and specific adaptation mechanisms to this mechanical environment might contribute to disease progression. It remains to be elucidated how cell shape change, modulated by the mechanochemical environment, regulates escape from the intracellular compartment and lower antibiotic susceptibility^32,80–82. We anticipate that confinement-induced physiological adaptation might be critical for bacterial survival through repeated cycles of pressure build-up and release during IBC growth and dispersal³². Future studies will determine whether confinement might act as an environmental cue influencing virulence, potentially increasing bacterial survival during antibiotic treatment, and whether this is dependent on Rcs pathway activation⁸³. More generally, we anticipate that our experimental approach and physical model might help to decipher bacterial virulence mechanisms associated with mechanical constraints.

To summarize, we showed that bacterial responses to mechanical confinement are largely determined by cell envelope architecture and growth rate. Gram-negative bacteria, such as E. coli, partially deform, distributing stress via the outer membrane; Gram-positives will likely resist deformation through thick peptidoglycan and controlled remodeling; and mycobacteria may deform minimally, adapting via polar growth and stress-response activation. These mechanical stresses trigger envelope-stress pathways such as Rcs and Cpx, which strengthen and remodel the envelope, potentially enhancing antibiotic tolerance. Based on these findings, we anticipate that slow-growing species like mycobacteria may be particularly protected under confinement, thanks to gradual stress responses that preserve envelope integrity and enhance tolerance.

In conclusion, this work uncovers novel molecular and physical mechanisms responsible for bacterial adaptation to mechanical constraints with important implications for bacterial evolution, antibiotic resistance, and bacterial infections. Future studies on bacterial adaptation to confinement will provide new insights into these mechanisms across diverse species.

Methods

Bacterial strains

The bacterial strains, plasmids, and primers used in this study are described in Tables 1–3, respectively. Genetically modified E. coli strains were derivatives of the following E. coli strains: the K-12 MG1655, TB28, BW27783 strains, one from the XL1-Blue background and the uropathogenic cystitis isolate UTI89 strain⁵⁷. Bacteria were grown in liquid or solid Luria-Bertani medium (LB, BD Difco) at 37 °C or 30 °C when grown to perform allelic exchange experiments. Antibiotics were used at the following concentrations: ampicillin (Ap), 100 μg/ml, chloramphenicol (Cm), 25 μg/ml, kanamycin (Km), 25-75 μg/ml, tetracycline (Tc), 2.5 µg/ml.

Table 2.

Plasmids used in this study

Plasmids	Description	Relevant information	Source or reference
pPrcsA-GFP	PrcsA-gfpmut2		56
pKOBEG	pSC101tsΩPBAD-redγβα for λ Red recombination, CmR		84
pKOBEGA	pSC101tsΩPBAD-redγβα for λ Red recombination, ApR		84
pBR322	ColE1 origin, ApR TcR	source of tet gene	85
pCDNA3.1-pCMV-PfV-GS-Sapphire	Fusion of the gene encoding for Pyrococcus furiosus encapsulin with the Sapphire fluorophore, ApR	source of GEM40	39 Addgene #116933
pMGC10	Cloning vector for GEM40, LacI repressor, and Plac, ApR CmR		35
pREP4	p15A origin, LacI repressor, KmR		Qiagen
pZE1R-GFPAAV
pEZZ-Cutinase	gfp (AAV) under the control of the constitutive PR promoter, ApR ZZ–cutinase fusion directed to the periplasm, β-lactamase, ApR		C. Beloin51,88

Deleted genes are in italics.

Table 1.

Strains used in this study

Strains	Description	Relevant information	Source or reference
E. coli
MG1655	Wild-type E. coli K-12		Lab collection
TB28::attHKpHC503	MG1655 ∆lacIZYA PzapA-ZipA(TM)-mCherry2, ApR	source of ZipA-mCherry2	34
CHN128 = MG1655 GFP	intC::PR-meGFP CmR		Ivan Matic’s lab
SJ156 = HU-GFP PsulA-mCherry	hupA-GFP KmR PsulA-mCherry	source of HU-GFP	38
MG1655 PcpxP-meGFP	galK::cpxP-meGFP CmR		Ivan Matic’s lab
MG1655 PrecA-YFP	intC::PrecA-YFP CmR		Ivan Matic’s lab
MG1655 mKate	∆intC::PR-mKate-cat CmR	source for mKate	88
BW25113 slmA	∆slmA::KmR	source for slmA	89
BW25113 rcsB	∆rcsB::KmR	source for rcsB	89
BW25113 wcaJ	∆wcaJ::KmR	source for wcaJ	89
BW25113 cpxR	∆cpxR::KmR	source for cpxR	89
BW27783 FtsZ-mNG	PftsZ-G55-ftsZ-Q56-mNeonGreen		37
XL1-Blue	recA1 endA1 gyrA96 thi-1 hsdR17 supE44 relA1 lac [F ´ proAB lacIq Z∆M15 Tn10] TcR		Stratagene
MG1655 ZipATM-mCherry	tet-PzapA-ZipA(TM)-mCherry2 TcR
MG1655 GFP
ZipATM-mCherry	intC::PR-meGFP CmR tet-PzapA-ZipA(TM)-mCherry2 TcR	attBHK022 site	This study
MG1655 ZipATM-mCherry GFPAAV	tet-PzapA-ZipA(TM)-mCherry2 TcR intC::PR-GFPAAV CmR		This study
MG1655 HU-GFP	hupA-GFP KmR		This study
MG1655 HU-GFP ZipATM-mCherry	hupA-GFP KmR tet-PzapA-ZipA(TM)-mCherry2 TcR		This study
BW27783 FtsZ-mNG ZipATM-mCherry	PftsZ-G55-ftsZ-Q56-mNeonGreen tet-PzapA-ZipA(TM)-mCherry2 TcR		This study
MG1655 PcpxP-meGFP ZipATM-mCherry	tet-PzapA-ZipA(TM)-mCherry2 TcR galK::cpxP-meGFP CmR		This study
MG1655 PrecA-YFP ZipATM-mCherry	tet-PzapA-ZipA(TM)-mCherry2 TcR intC::PrecA-YFP CmR		This study
XL1-Blue GEM40	cat-Plac-PfV-GS-Sapphire CmR		This study
MG1655 GEM40	cat-Plac-PfV-GS-Sapphire CmR	attBλ site	This study
MG1655 GEM40 ZipATM-mCherry	cat-Plac-PfV-GS-Sapphire CmR tet-PzapA-ZipA(TM)-mCherry2 TcR	attBλ site	This study
MG1655 ZipATM-mCherry slmA	tet-PzapA-ZipA(TM)-mCherry2 TcR ∆slmA::KmR		This study
MG1655 ZipATM-mCherry rcsB	tet-PzapA-ZipA(TM)-mCherry2 TcR ∆rcsB::KmR		This study
MG1655 ZipATM-mCherry wcaJ	tet-PzapA-ZipA(TM)-mCherry2 TcR ∆wcaJ::KmR		This study
MG1655 ZipATM-mCherry cpxR	tet-PzapA-ZipA(TM)-mCherry2 TcR ∆cpxR::KmR		This study
in UTI89 (UPEC)
UTI89 RFP	aphA-marsRFP KmR		57
UTI89		StrR	57
UTI89 mKate	∆intC::PR-mKate-cat CmR	StrR	This study
UTI89 mKate rcsB	∆intC::PR-mKate-cat CmR ∆rcsB::KmR	StrR	This study

Bacterial species and deleted genes are in italics.

Table 3.

Primers used in this study

Primer name	Sequence (5’ → 3’) (enzymes' restriction sites are underlined)
HK022-att-Tc-F	CCATCCAGAGTCTTCGGGTCAGGGTTAAATTCACGGTCGGTGCACTTTAGTTTGCGCATTCACAGTTCTC
Tc-R	AGTGGTAGTCTCATGTTTGACAGCTTATCATCG
Tc-PzapA-F	GATAAGCTGTCAAACATGAGACTACCACTTTGGGCCCTGG
HK022-att-mCherry-R	ATAAGGCTTTATGCTAGATGCATTCCGCTTTGCGACTCAACCTTTTTCACAGAAAGGCCGGGAAATACC
P1	GGAATCAATGCCTGAGTG
P4	GGCATCAACAGCACATTC
GEM40-XhoI-F	CTCGAGAGGAGTAATTTTATGCTCTCAATAAATCCAACC
GEM40-PmeI-R	GTTTAAACTTATTTGTACAATTCATCAATACCA
GEM40-seq1-F	CCAAAAAGATATGCAAAGTTGCT
GEM40-seq2-R	ACCCACTACCGCTGCCAGACCA
lambda-att-Cm-F	CTTTTTGTCTTTTTACCTTCCCGTTTCGCTCAAGTTAGTACAAGAATTGCCGGCGGAT
Cm-UGB1479-R	TGAGACGTTGATCGGCACGTAAG
Cm-UGB1479-F	GCCGATCAACGTCTCAGACATCATAACGGTTCTGGCAAA
lambda-att-Sapphire-R	ATGAAATAGAAAAATGAATCCGTTGAAGCCTGCTTTTTTATTATTTGTACAATTCATCAATACCATGG
lambda-att-ext5-F	GGCGATAAATTGCCGCATCG
lambda-att-ext3-R	TGCCACCATCAAGGGAAAGCCC
intC-PR-GFPAAV-F	CGTAGATTTACAGTTCGTCATGGTTCGCTTCAGATCGTTGACAGCCGCACCTAACACCGTGCGTGTTGAC
GFPAAV-R	GCTTAAACTGCTGCAGCGTAGTTTTC
GFPAAV-Cm-F	GCTGCAGCAGTTTAAGCCGGAAGGGCATCGATGG
intC-Cm-R	ATAGTGTTAAGGTCGCTCACTCCACCTTCTCATCAAGCCAGTCCGCCCA
intC-F	ACGCCAAAAAGCAGAACGTGC
intC-R	TGCACTGGATTGCAAGACTTT
hupA-F	ATGAACAAGACTCAACTGATTGAT
yjaH-R	GAAGAGTTATGACTACAGGCAGTG
slmA-F	GGTCACTCTGGTCGTCAGTA
slmA-R	TGATTACAACTGCAAAGTGATCT
wcaJ-F	ATGGCATCAGCATGACCTTT
wcaJ-R	TCAGAAAGCGTATCTGCCAG
cpxR-F	TAACTATGCGCATCATTTGC
cpxR-R	GACATGCTGCTCAATCATCA
mKate-F	ACGCCAAAAAGCAGAACGTGC
mKate-R	TGCACTGGATTGCAAGACTTT
rcsB-F	CTTAAAGGCGTATTTGCCAT
rcsB-R	GATAAAACAGACGCTGACGTTA

Genetically modified E. coli strains were constructed either by λ-red recombination or P1 phage transduction, as described in the following. The primers were synthesized by Eurogentec, and the sequencing was done by Eurofins.

For λ-red recombination, PCR products (detailed below) were transferred by the λ-red recombination method into E. coli K-12 MG1655 and derivatives or UTI89 and derivatives transformed with the pKOBEGA or pKOBEG plasmids⁸⁴, which contain the λ-Red operon under the control of the arabinose-inducible araBAD promoter. PCR fragments were dialyzed on a 0.025 μm porosity filter (Millipore) and electroporated after induction with 0.2 % arabinose. Bacteria were spread on LB agar plates with the appropriate selection antibiotic and incubated overnight at 37 °C. Transformants were checked by PCR and sequencing, and then tested for ampicillin or chloramphenicol resistance to test loss of pKOBEGA or pKOBEG, respectively.

The PCR products that have been generated are described below:

– ZipA_TM-mCherry2-TcR: A fragment containing ZipA_TM-mCherry2 and a Tc^R cassette was created in two steps. First, the Tc^R cassette was amplified from pBR322 vector⁸⁵ with primers HK022-att-Tc-F/ Tc-R, and the ZipA_TM-mCherry2 fusion was amplified from E. coli TB28(attHKpHC503)³⁴ with Tc-PzapA-F/HK022-att-mCherry-R. The latter PCR product, plus the amplified Tc^R cassette were used for a two-way overlap PCR with primers HK022-att-Tc-F and HK022-att-mCherry-R. The tet gene was oriented in the opposite direction to ZipA_TM-mCherry2. Primers HK022-att-Tc-F and HK022-att-mCherry-R carried 50 bp of homology with the insertion region at each end (HK022 att site). The expected chromosome structure in the HK022 att site was confirmed with primers P1 and P4 (Table 3)⁸⁶. The PCR product was inserted into MG1655, MG1655 GFP, MG1655 HU-GFP, MG1655 PcpxP-meGFP, MG1655 PrecA-YFP, MG1655 GEM40 and BW27783 FtsZ-mNG.

– GFP_AAV: The fragment P_R-GFP_AAV containing a Cm^R cassette was created in two steps⁸⁷. First, a P_R-GFP_AAV fragment was amplified from the pZE1R-GFP_AAV plasmid (kind gift from C. Beloin) with intC-P_R-GFP_AAV-F/GFP_AAV-R primers, and the Cm^R gene was PCR amplified from genomic DNA of E. coli MG1655 GFP ZipA_TM-mCherry with GFP_AAV-Cm-F/intC-Cm-R primers. The P_R-GFP_AAV PCR product plus the amplified Cm^R cassette were used for a two-way overlap PCR with intC-P_R-GFP_AAV-F/intC-Cm-R. The cat gene is oriented in the direction opposite that of P_R-GFP_AAV. Primers intC-PR-GFP_AAV-F/intC-Cm-R carried 50 bp of homology with the insertion region at each end (intC site). The PCR product was electroporated into E. coli MG1655 ZipA_TM-mCherry, and the expected chromosome structure in the intC site was confirmed with primers IntC-F and intC-R.

– GEM40-Cm^R: The sequence encoded for the 40 nm-GEM was amplified using GEM40-XhoI-F/GEM40-PmeI-R primers from the plasmid pCDNA3.1-pCMVPfV-GS-Sapphire (Addgene #116933)³⁹ and cloned into pMGC10 vector³⁵ with XhoI and PmeI restriction sites. The plasmid contained the open reading frame encoding the Pyrococcus furiosus encapsulating 196 fused with the T-Sapphire fluorophore under the control of the inducible Plac promoter. Then, a fragment 40 nm-GEM containing a CmR cassette was created in two steps. First, the Cm^R gene was PCR amplified from E. coli TG1_λATT-Cm-GFP (UGB1479), kind gift from C. Beloin, and the 40 nm-GEM was amplified from the previously constructed pMGC10 plasmid with Cm-UGB1479-F_λATT-Sapphire-R. The latter PCR product, plus the amplified Cm^R cassette were used for a two-way overlap PCR with λATT-Cm-F/λATT-Sapphire-R. The cat gene was oriented in the opposite direction to 40 nm-GEM. Primers λATT-Cm-F/λATT-Sapphire-R carried 41 bp of homology with the insertion region at each end (λATT site). The expected chromosome structure in the λATT site was confirmed with primers λATT-ext5-F and λATT-ext3-R. The PCR product was electroporated into both XL1-Blue and MG1655 backgrounds. In the latter, the low-copy plasmid pREP4 (Qiagen) was transferred into E. coli MG1655 ZipA_TM-mCherry strain expressing 40 nm-GEMs, constitutively expressing the LacI repressor protein, and tightly regulating recombinant protein.

– HU-GFP, slmA, wcaJ, cpxR, rcsB: PCR fragments containing a Km^R cassette were amplified from E. coli SJ156³⁸ or from the Keio collection⁸⁸, respectively, with primers listed in Table 3.

– mKate: a PCR fragment containing a Cm^R cassette was amplified from E. coli MG1655 mKate⁸⁹ with primers mKate-F and mKate-R (Table 3).

P1 allelic transduction from the Keio collection⁸⁸ was used to delete the rcsB gene into E. coli MG1655 ZipA_TM-mCherry. Transductants were verified by PCR and sequencing, with the following primers listed in Table 3.

Bacterial culture

Bacteria were streaked from −80 °C frozen stocks onto fresh LB agar plates supplemented with the appropriate antibiotics and incubated overnight at 37 °C. For the experiments carried out with the MG1655 strain, a few bacterial colonies from plates were resuspended in 5 ml filtered LB Miller and incubated for 7–12 h at 37 °C with agitation at 180 RPM. A bacterial subculture was then started in pre-warmed filtered LB Miller at an optical density of 0.01 and incubated for 2 h at 37 °C with agitation at 180 RPM. For the experiments carried out with the UTI89 strain, a few bacterial colonies from the plate were resuspended in 20 ml filtered LB Miller and incubated overnight at 37 °C without agitation.

Bacterial stress reporter validation

Bacterial precultures were diluted at an optical density of 0.1, loaded in 96-well black glass-bottom plates, submitted to a range of chemical stress inducers, and incubated in the plate reader (Cytation5, Biotek) under agitation at 37 °C with automated acquisition of optical density and fluorescence every 30 min for 18 h. Cephalexin (CEP) and ciprofloxacin (CIP) were used to induce envelope and DNA damage, respectively. CEP was used at 0, 5, 15, 20 µg/ml, and CIP at 0, 0.01, 0.1, 0.5, 1, 10 µg/ml. For all the reporters, fluorescence intensity increased with inducer concentration at a given optical density, validating reporter responsiveness.

Bacterial proliferation on agar pads

Bacterial proliferation on agar pads was used to assess bacterial growth in classical conditions. For this, exponentially growing bacteria diluted at an optical density of 0.015 were trapped between a glass coverslip and a 2 % agarose gel composed of LB Miller or M9 minimal media, placed under the microscope, and imaged at 37 °C (The Cube).

Microfabrication

Master containing the chip design was fabricated in the clean room at the LAAS-CNRS using two-layer stepper-based high-resolution photolithography. The designs of the silicon master molds were created using Clewin software by rescaling the dimensions of the chip previously developed to confine yeasts¹⁶. Briefly, the first layer defining the nutrient channels was patterned on a 1.1 µm layer of ECI positive photoresist using a UV projection stepper (Canon FPA 3000i4/i5, dose = 400 J/m²) and etched at a depth of 400 nm by high-density plasma etching. The second layer, defining the growth chambers and the loading channels, was patterned on SU8 negative photoresist at a height of 2.6 µm. For this, the silicon wafer previously etched was coated with SU8 negative photoresist, aligned with the first layer, and insolated in the stepper using a dose of 3000 J/m². Characterization by profilometer and scanning electron microscope was performed between each step to ensure a correct aspect ratio. Before being taken out of the clean room, the masters were passivated with silane. For chip preparation, a 15:1 mixture of PDMS (Sylgard 184, Dow corning) and curing agent was deposited in drops of roughly 100 µl on top of the chamber features and cured at 65 °C for 30 min. A 10:1 mixture was then poured and cured overnight at 65 °C. Then, PDMS was cut, punched, cleaned with isopropanol, bonded to pre-cleaned glass coverslips (Marienfeld, #1.5) using oxygen-plasma (Femto Science, power = 60 W, exposure time = 40 s), cured at 65 °C for 5 h, sterilized using UV ozone cleaner (UVO Cleaner, Jelight) for 30 min, and finally cured at 65 °C for 30 min. Prior to all experiments, the integrity of the nanochannels was checked.

Microfluidic device operation

The microfluidic chips were filled in the following order to prevent air bubbles within the channels using sterile equipment¹⁷. First, exponentially growing bacteria were loaded at an optical density of 0.3–0.4 in the growth chambers. Second, the loading channel was filled with pre-warmed filtered LB Miller using a Fluigent pressure controller (MFCS, Fluigent) set at 1 bar during the whole course of the experiments. Third, a plug was inserted at the outlet of the loading channel to allow continuous medium renewal within the growth chambers. The chips were then placed in the microscope thermoregulated chamber set at 37 °C.

Microfluidic device calibration for pressure measurements

Growth-induced pressure was inferred from the deformation of the chambers as previously done⁹⁰. To optimize the detection of the chamber walls, the rigidity of the PDMS was adjusted, and the channel edges were fluorescently stained by perfusing LB Miller supplemented with FM1-43 at 2 µg/ml. Chamber deformation was semi-automatically tracked using homemade Fiji and Python scripts. The relationship between deformation and growth-induced pressure was then calibrated as follows: the deformation of the loading channel was measured at the top of the channel under a range of known pressures (0, 200, 400, 600, 800, 1000, 1500, 2000, 3000, 4000 mbar). The applied pressure was represented as a function of the measured deformation and fitted with a linear regression curve (y = 282.6x, R² = 0.99) (Supplementary Fig. 1H). This calibration curve was established for all the chips where force measurements were performed. To facilitate force measurements while looking at other fluorescent reporters, we also calibrated the growth-induced pressure as a function of the number of bacteria in the 2D focal plane of observation. This provided a strategy to infer the pressure in the chamber by using the number of bacteria as a readout and determine the corresponding phases of confinement (Supplementary Fig. 2C). Here is a full list of figures where growth-induced pressure was measured by tracking chamber deformation, either automatically or manually based on bright field images as previously done¹⁶: Figs. 1D, 2B–D, 4C, D, M, N, 6B–D and Supplementary Figs. 1F–J, 3I, 5B, D, F–H, 6N.

Osmotic shocks

Hyperosmotic shocks were performed by increasing the osmolarity of the culture medium using sorbitol (Sigma-Aldrich) at various concentrations (0.25, 0.5, 1 M) as previously done^41,91, either using Ibidi channels or CellASIC ONIX commercial chambers when continuous flow perfusion was required.

In the first assay, exponentially growing bacteria were introduced slowly within Ibidi channels (sticky-slide VI 0.4, Ibidi) coated with 3-Aminopropyltriethoxysilane (APTES, Sigma-Aldrich) at 2 % in ethanol and incubated for 10 min at 37 °C to promote bacterial attachment. LB medium was then rinsed out and replaced with fresh medium supplemented with sorbitol at various concentrations (0.25, 0.5, 0.75, and 1 M). Image acquisition was launched 30 s after changing the medium. Note that E. coli XL1Blue bacteria were used for these experiments.

In the second assay, exponentially growing E. coli bacteria fluorescently labeled at the inner membrane and DNA were loaded in B04A microfluidic perfusion plates (CellASIC), and pre-warmed medium was exchanged using the ONIX microfluidic platform (CellASIC). The culture medium was switched to culture medium supplemented with sorbitol at 0.5 M first, and then 1 M to progressively increase cytoplasmic crowding. The culture medium was always supplemented with wheat germ agglutinin-AlexaFluor 642 (WGA-AF642, Invitrogen) to fluorescently label the cell wall. Bacterial division was monitored based on the inner membrane signal and confirmed with cell wall fluorescent staining in a time window of 3 min after the hyperosmotic shock, as previously done⁴¹. For quantification, bacteria with a clear membrane invagination and delimitation at mid-cell were considered as dividing. The fraction of dividing cells was determined by manually counting the number of dividing bacteria over the total number of bacteria.

Fluorescent probes

Several fluorescent probes were perfused in the bacterial confiner to characterize bacterial aggregates upon confinement. SYTOX Green (Invitrogen) was supplemented in the culture medium at 5 µg/ml to fluorescently stain nonviable bacteria characterized by a permeable envelope. FITC-Dextran 10 kDa (Sigma-Aldrich) was supplemented in the culture medium at 1 mg/ml to assess the kinetics of medium perfusion within confined aggregates. The amphiphilic dye FM1-43 (AAT Bioquest) was supplemented in the culture medium at 2 µg/ml to fluorescently stain PDMS walls.

Live imaging

Most of the experiments were performed using a spinning-disk confocal microscope (Ti Eclipse, Nikon) equipped with a spinning-disk module (CSU-X1, Yokogawa) and a 100X oil objective (Plan Fluor, NA = 1.45) coupled to a Live-SR SIM-like module (Gataca Systems). The microscope stage was kept at 37 °C using a temperature controller (The Cube). Fluorescence was recorded using a CMOS camera (95B Prime, Photometrics). The laser power was set at 30 %, the exposure time was set at 300 ms, and the focus was maintained using a Perfect Focus System (PFS, Nikon). Image acquisitions were computer-controlled using MetaMorph software and custom-made scripts (Molecular Devices). The CellASIC experiments (Fig. 3H–K) were imaged using a confocal microscope (LSM710, Zeiss) equipped with an Airyscan module (Zeiss) and a 63X oil objective (Plan Apochromat, NA = 1.4 DIC M27). Super-resolution 3D imaging of the chips (Supplementary Fig. 2A) was performed at a fixed time point using a SIM microscope (Zeiss Elyra7, Lattice SIM) equipped with the 63X oil objective described above. Similar Z-stacks acquisitions were performed using 1 µm-wide beads to correct optical aberrations (Tetraspeck, Invitrogen).

Unless otherwise stated, bacterial proliferation in the bacterial confiner was monitored with a time-lapse of 30 min frame rate over 12–17 h and Z-stacks of 0.1 µm step over 1 µm centered around the focal plane corresponding to the nanochannels. For force measurement, a MetaMorph script was used to acquire Z-stacks of 0.2 µm over 2.6 µm to monitor chamber walls at the height of maximal deformation. For tracking analyses, bacterial proliferation was monitored at 5 min frame rate for 5–6 h only to avoid photobleaching. Bacterial proliferation in agar pads was monitored with a time-lapse of 5 min frame rate over 2.5 h to avoid nutrient depletion. 40 nm-GEMs diffusion was imaged in stream mode every 50 ms over 2.5 s using a laser power set at 95 %.

Assessment of bacterial metabolic activity

A live-cell enzymatic assay based on the Förster resonance energy transfer (FRET) probe 7-hydroxycoumarin-cephalosporin-fluorescein acetoxymethyl ester (CCF2/AM) was used to assess bacterial metabolic activity^51,92. The probe, originally developed for eukaryotic cells, was adapted for bacterial use inside the bacterial confiner (Fig. 4J). This substrate is membrane-permeable, and removal of AM groups based on the activity of an esterase (i.e., cutinase) allows retention of the probe within the cell. The CCF2 molecule is fluorescent when excited with a blue laser (λ = 405 nm), with an emission spectrum that depends on enzymatic cleavage of the β-lactam linker between the coumarin (donor) and fluorescein (acceptor) fluorophores. In metabolically active cells expressing β-lactamase, CCF2/AM is converted, impacting FRET efficiency and therefore producing a fluorescence shift from green to blue. Because E. coli lacks efficient periplasmic esterase activity to remove AM groups, we expressed a periplasmic cutinase in addition to β-lactamase to promote local de-esterification and retention of the fluorescent product within the periplasm⁵¹. Cells expressing pEZZ-cutinase or not were loaded with CCF2/AM under constant perfusion, and fluorescence was recorded by live microscopy in the green (fluorescein, data not shown) and blue (coumarin) channels (Fig. 4K). We further validated these results with medium-switch experiments in which CCF2/AM was loaded at the start of Phase 3 (Supplementary Fig. 5K).

Antibiotic perfusion to arrest protein synthesis

Sub-MIC spectinomycin concentrations that reduce protein synthesis while maintaining cell viability were determined by measuring OD₆₀₀ and fluorescence intensity from the P_R-GFP reporter across a range of concentrations using a plate reader (Cytation5, Biotek), as shown in Supplementary Fig. 5L, M. Finally, a concentration that partially inhibited translation while maintaining growth was selected (5 µg/ml). In the bacterial confiner, perfusion was switched to a medium containing 5 µg/ml spectinomycin at the beginning of Phase 2 (i.e., ∼1 h after confinement onset). GIP and fluorescence intensity from the P_R-GFP reporter were monitored in parallel.

Image analysis

All analysis scripts and representative datasets required to reproduce the analyses, including those for the model, are publicly available (see the “Data Availability” and “Code Availability” sections for the repository link).

Force measurements

Fluorescent images of chamber walls at a height of 2.6 µm (i.e., the top of the chamber at the beginning of the experiment—but also the height at which the chamber is the most deformed during the whole course of the experiment) were binarized using the Otsu auto-threshold in Fiji and verified manually. The distance along the x-axis between the chamber walls was measured using a Python script and used to compute the deformation over time with respect to its initial size. A similar analysis was performed on the calibration channel to establish the calibration curve. The deformation of the chamber was then related to the corresponding growth-induced pressure by using the equation established during the calibration.

Bacterial single-cell segmentation and tracking

Strains with membrane labeling were segmented using ilastik⁹³ (v1.3.3) pixel-classification workflow, trained specifically on the bacterial inner membrane fluorescent signal. The model file generated from manual annotations was then used in TrackMate, using its ilastik integration³⁵ (v7.11.1) to segment bacteria with a threshold on the probability map of 0.6. For strains having only a cytoplasmic labeling (UTI89 strains), bacteria were segmented with TrackMate, using its Omnipose⁹⁴ integration with parameters: pretrained model bact_fluor_omni, cell diameter 2 µm. For the morphology measurements reported in Figs. 2A–E, 5D, and 6D, E, G, L, and in Supplementary Figs. 3B, C, G, I, 4R, and 6O, P, segmentation mistakes were manually corrected using Fiji on the exported mask images. To measure bacterial division and growth rates (Fig. 2F, Supplementary Fig. 3D–F), the segmented bacteria were tracked with TrackMate using the overlap tracker with parameters: min IoU 0.1, scale 1.3. Tracking mistakes were manually corrected in TrackMate.

Bacterial width and length measurements

Bacterial width and length were measured using a custom approach in Python (Supplementary Fig. 2D). Each bacterium contours were extracted from the TrackMate file. From these contours, the skeleton was computed with the scikit-image skeletonize function using the Zhang algorithm⁹⁵ that proved to be more robust than the Lee algorithm⁹⁶ in the case of small and irregularly shaped bacteria. When needed, spurious small branchings were pruned from the skeleton that was then simplified using the Douglas-Peucker algorithm⁹⁷ to obtain the bacterium medial line. The length of the bacterium was computed as the length of the medial line plus the value of the distance map at the skeleton tips. The bacterial width was defined as the median of the width distribution measured along the medial line, again using the distance map. The shape irregularity index was measured as the standard deviation of this distribution normalized by its mean value for the wild-type strain (Supplementary Fig. 6O, P).

Karyoplasmic ratio

The DNA occupancy in single bacteria was determined by segmenting the DNA channel in images (Fig. 3D, E and Supplementary Fig. 4K, L). As above, we used the ilastik pixel classification workflow and trained a model on the DNA channel that was further used in TrackMate with a threshold on the probability map of 0.7. The Python script to measure the karyoplasmic ratio from the bacterial and DNA masks is given in the Supplementary Information.

Single bacteria measurements

All the bacterial segmentation data generated with TrackMate and ilastik, with or without tracking data, were analyzed subsequently with dedicated Python scripts, tailored for the specific feature to extract. Fluorescence intensity measurements and bacterial area were extracted from the numerical features in TrackMate objects.

Growth and division rates

Single-cell growth (g_A) and division (g_D) rates (in min⁻¹) were computed from single-cell tracks using the following formulae: $g_A=\frac{A_{division}-A_{birth}}{t_{division}-t_{birth}}$ and $g_D=\frac{1}{t_{division}-t_{birth}}$; with A_t the area of the bacterium at time t (i.e., birth or division time). Values were recorded at the division time to generate single-cell curves (Supplementary Fig. 3E), subsequently pulled together and time-averaged to obtain the graphs in Fig. 2F.

Protein production rate

The protein production rate (PPR) of a fluorescent protein was measured over time as the variation of the protein total fluorescent intensity (TFI) over the whole chamber divided by the number of bacteria N, i.e., $\mathrm{PPR}=\frac{1}{{\mathrm{N}}^{*}}\left[\frac{1}{{\mathrm{TFI}}^{*}}\frac{\left(\mathrm{d}\left(\mathrm{TFI}\right)\right)}{\mathrm{d}t}\right]$. This measurement cannot be done at the single-cell scale because of the volume changes induced by the reductive divisions.

Determination of transcriptional reporter activation

Bacteria expressing fluorescent transcriptional reporters (P_rcsA-GFP, P_cpxP-meGFP, P_recA-YFP) were classified as activated or non-activated based on a threshold on the fold-change of cytoplasmic mean fluorescence intensity, with respect to movie frames where bacteria were not confined. Fold change values from all movies were pooled into two histograms (non-confined vs. confined). Reporters were considered activated when the two distributions exhibited two discernible peaks. In this case, a manual threshold was defined to quantify the percentage of activated bacteria (Supplementary Fig. 6C, F, I).

Analysis of unstable GFP

For each chamber, the mean GFP fluorescence intensities were measured over time in the largest regions of interest inscribed within each chamber, using Fiji. Then, the background GFP fluorescence intensity was measured over time in an area outside the chamber and away from any visible structure. This background GFP fluorescence intensity was subtracted from the mean GFP fluorescence intensity of the chamber. These measures were performed independently for the stable GFP chambers and the unstable GFP_AAV ones (Supplementary Fig. 4G, H). Based on mean fluorescence intensities of GFP and GFP_AAV as a function of growth-induced pressure, we estimated corresponding protein production rate (k_production) based on the following equation:

$$\frac{dMFI\left(t\right)}{dt}=\left[k_{production}-\left(k_{degradation}+k_{growth}\right)\right]MFI\left(t\right)$$

with MFI indicating here the total mean fluorescence intensity at the chamber scale, k_growth the growth rate, (also named g_A in the theoretical model and the rest of the article) and k_degradation the degradation rate. Since we measured k_growth, this also allows us to estimate the net production rate k_production − k_degradation. (Supplementary Fig. 4I, J and Supplementary Note 1).

Analysis of cutinase

The measurements of coumarin mean fluorescence intensities were performed as described above in the section “Analysis of unstable GFP”, i.e., over time within the chambers after background fluorescence intensity subtraction (Fig. 4K).

Analysis of spectinomycin-perfused chambers

The measures of GFP mean fluorescence intensities were performed as described above in the section “Analysis of unstable GFP”, i.e., over time within the chambers after background fluorescence intensity subtraction (Fig. 4M).

Spatial pattern of Rcs transcriptional response

In Fig. 5A, the contour delineating a zone of lower Rcs activation in the chambers was drawn manually to roughly include all bacteria for which RcsA intensity was below a threshold. This threshold was determined by the Otsu method on the distribution of RcsA intensity measured on all bacteria in 3 chambers. Then, bacteria were classified as Center or Edges depending on whether they were found respectively inside or outside this contour (Fig. 5D). The same spatial classification was used in Supplementary Fig. 3F to quantify potential heterogeneities in growth and division rate within the chamber.

40 nm-GEMs diffusion measurements

Single 40 nm-GEMs nanoparticles were tracked using the MOSAIC Fiji plugin using the following parameters: size = 3 pixels, cutoff = 0, link = 1 frame, displacement = 5 pixels per frame. Tracks longer than 10 points were analyzed using a MATLAB code from a previous study³⁹. The Mean Squared Displacement (MSD) was computed over time Δt and fitted linearly over the first 0.5 s. The diffusion coefficient D was then calculated as D = MSD/4Δt (Supplementary Fig. 4M).

Data visualization and statistical analysis

The number of technical replicates (N), tracks (n_tracks), bacteria (n_bacteria), and chambers (n_chambers) is specified in the corresponding legends. All the observations described in the bacterial confiner were highly reproducible and observed in more than 15 chambers. Complex single-cell analyses were completed on a representative subset of data (n_chambers = 2–4) for the sake of time. Quantifications still consider a large number of individual bacteria (3928 ≤ n_bacteria ≤ 50661). Details on statistical analyses are written in the figure legends. For all the experiments, data visualization and statistical analysis were performed in GraphPad Prism.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Author contributions

L.L.B., M.D., G.D. and D.B. designed the study. L.L.B. designed, performed, and analyzed most experiments and actively contributed to manuscript preparation by building figures and writing the manuscript. B.A., L.L.B. and M.D. designed the bacterial confiner and optimized the protocol for microfabrication. R.R. built up a theoretical model of bacterial growth under confinement and edited the manuscript. L.X. and L.L.B. built up the pipelines for bacterial segmentation, tracking, and post-processing, and performed image analysis of the microscopy datasets. M.A. performed the experiments for the revision process. S.G. constructed most of the bacterial strains used in this study, together with L.L.B., L.M. provided crucial help with microfabrication of the nanochannels in the confiner. M.P. gave essential conceptual feedback on the theoretical model, suggested key experiments, provided feedback throughout the study, and helped write and edit several parts of the manuscript. J.Y.T. supervised L.X., provided feedback on all the image analysis presented in this study, and helped with study conceptualization. M.D., G.D. and D.B. developed the initial hypothesis, supervised the study, and wrote the manuscript. All authors contributed to manuscript preparation.

Peer review

Peer review information

Nature Communications thanks the anonymous reviewers for their contribution to the peer review of this work. A peer review file is available.

Funding

This work was supported by the Integrative Biology of Emerging Infectious Diseases (IBEID) laboratory of excellence (ANR-10-LABX-62), the National Infrastructure France-BioImaging supported by the French National Research Agency (ANR-24-INBS-0005 FBI BIOGEN), la Fondation pour la Recherche Médicale (FRM, FDT202204015046 to LLB, EQU202203014654 to GD, and FDT202504020138 to LX), the Agence Nationale pour la Recherche (ANR-21-CE13-0033-01 MechaMenin to DB) and the European Union (ERC UnderPressure, grant agreement number 101039998 to MD; ERC Destop to GD, ERC HomePath to DB). LLB was a student from the FIRE PhD program funded by the Bettencourt Schueller foundation. LX received funding by the INCEPTION project (PIA/ANR-16-CONV-0005) and is a student from the FIRE PhD program funded by the Bettencourt Schueller foundation and the EURIP graduate program (ANR-17-EURE-0012). Views and opinions expressed are those of the author(s) only and do not necessarily reflect those of the European Union or the European Research Council. Neither the European Union nor the granting authority can be held responsible for them.

Data availability

All data supporting the findings of this study have been deposited in Zenodo and are publicly available at https://doi.org/10.5281/zenodo.20324554. All underlying numerical data are available as complete Prism files for each figure, including all information necessary to reproduce the statistical analyses. A representative source dataset is available for each analysis to enable testing and validation of the analysis code. Additional information required to interpret or reproduce the results is available within the article, its Supplementary Information, and the Zenodo repository, or from the lead contact upon request.

Code availability

All code generated in this study (including the one for the theoretical model) and required to reproduce the analyses presented in this study is available in the Zenodo repository (https://doi.org/10.5281/zenodo.20324554).

Competing interests

The authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at https://doi.org/10.1038/s41467-026-76492-6.