betaFeedback

WorksPhysics and Astronomy

Mechanical acclimation of bacterial growth and division across generations

Abstract

Bacteria frequently encounter mechanical confinement in natural and engineered environments, yet how an abrupt change in mechanical conditions reorganizes growth and division across generations remains poorly understood. Using single-cell time-lapse microscopy, we tracked Escherichia coli across generations following transfer from liquid culture to glass-attached or glass-gel-confined conditions. This mechanical confinement causes an immediate reduction in elongation rate, followed by a multi-generational mechanical acclimation process whose dynamics depend systematically on confinement strength. During this relaxation, division behavior becomes more heterogeneous and cells temporarily depart from canonical adder-like size regulation, exhibiting timer-like correlation signatures. Using a stochastic cell-cycle model that resolves the generation-dependent correlation signatures of different size-control strategies, we show that changes in elongation rate alone cannot account for the observed transition in growth-division coordination. Despite this transient reorganization, the birth-size distribution remains bounded and largely preserves its shape throughout acclimation. Notably, adder-like regulation is restored before the elongation rate fully recovers, revealing distinct relaxation timescales for growth and size-control dynamics. These results establish bacterial size control under mechanical confinement as a dynamical nonequilibrium process in which growth and division can transiently reorganize while global size homeostasis is maintained.

Introduction

Microbes thrive throughout the biosphere, from the atmosphere to deep oceans and subsurface rocks [1, 2, 3]. Many of their habitats are spatially heterogeneous and mechanically confined, including soils [4, 5, 6], depleted gas reservoirs [7, 8], host tissues and mucus-rich environments [9, 10, 11, 12], and marine snow and biologically-active sediments [13, 14, 15, 16]. In such environments, bacterial cells experience combinations of compression, adhesion, steric restriction, and crowding, all of which can influence microbial growth and survival [17, 18]. Understanding how single cells respond to these mechanical constraints is therefore essential for connecting microbial ecology with quantitative biophysics.

Bacteria can sense and respond to mechanical cues across scales, from changes in single-cell physiology to the collective organization of growing populations. At the single-cell level, surface attachment and external mechanical stress—such as hydrostatic pressure and soft-matter confinement—can alter cell metabolism [19, 20, 21], change envelope mechanics [22], suppress cell elongation [23, 24], and initiate biofilm formation [25]. Intense mechanical constraints can further disrupt cell morphology and division [26, 27, 28, 29, 30], leading to irregular cell shapes [31, 32], delayed division, and in extreme cases, cell death [26, 33]. Beyond these single-cell responses, mechanical forces, confinement, and crowding can shape colony organization, spatial structure, topology, and collective behavior in growing microbial populations [24, 34, 35, 36, 37, 38, 39, 40, 41, 42]. These studies establish the mechanical environment as an active regulator of bacterial growth, division, morphology, and collective organization. However, majority of previous studies have characterized cells after they have already adapted to a sustained stressed or confined state, leaving the dynamics of the initial acclimation process largely unresolved.

Under steady growth conditions, E. coli and Bacillus subtilis exhibit robust size control in both microfluidic and agarose-pad assays [43, 44, 45, 46, 47]. Bacterial size homeostasis relies on tight coupling between growth and division across generations. Single-cell measurements have shown that this coordination is often well described by the adder principle, in which cells add an approximately constant size between birth and division, largely independent of their birth size [45, 46]. Recent mechanistic studies have further suggested that division control can emerge from threshold accumulation of key division proteins, such as FtsZ, through balanced biosynthesis [48, 49]. These studies have focused primarily on cell-size control under steady-state growth. How this coordination is reshaped across generations following an abrupt switch in the mechanical environment remains poorly understood.

Here, we use single-cell time-lapse microscopy to track E. coli across successive generations after the onset of mechanical confinement. We quantify elongation rates, division timing, and birth-size distributions and use generation-resolved correlations to characterize size control. We further develop a minimal stochastic model to compare the observed dynamics with idealized adder-, sizer-, and timer-like reference behaviors. Our results show that confinement induces not only an immediate slowdown in growth, but also a transient transformation of canonical adder-like growth–division coupling.

Switch in mechanical environments initiates multi-generational growth acclimation

To investigate how mechanical confinement influences bacterial growth and size regulation, exponentially growing E. coli cells were transferred from lysogeny broth (LB) liquid culture into three confinement conditions of increasing mechanical strength (Figure 1(a–Figure 1d)): a glass-attached state in liquid LB medium (glass–liquid), confinement between glass and a 1.5% agarose pad (glass–soft gel), and confinement between glass and a 6% agarose pad (glass–hard gel). In all cases, colonies grew from isolated founder cells and expanded within chambers approximately 500 μm thick. All confinement chambers were pre-equilibrated at 30°C to minimize temperature perturbation during transfer.

Figure 1. Growth-cycle metrics under single-cell confinement. (a)–(d) Experimental workflow for single-cell confinement assays. Exponentially growing E. coli cells were transferred from liquid culture into confinement chambers for phase-contrast time-lapse imaging. (e) Definitions of growth-cycle metrics. Cell-length trajectories of representative cells during the first three generations following confinement. Black curves show representative cell-length records during the first three generations following confinement. Birth length, division length, added length, division time, division rate, and elongation rate are denoted by LbL_b, LdL_d, ΔL=Ld−Lb\Delta L = L_d - L_b, τd\tau_d, λd=ln⁡2/τd\lambda_d = \ln2/\tau_d, and λe\lambda_e, respectively. Elongation rates were obtained by fitting the measured cell lengths within each cell cycle to an exponential growth model; representative measurements used for fitting are indicated by red crosses.

Single-cell growth and division were monitored by phase-contrast time-lapse microscopy at 2 min intervals for 3–7 h at 30°C. Cell contours, lineages, and growth dynamics were extracted using automated segmentation and tracking pipelines, with manual corrections if necessary (see Supplementary Fig. S1). Elongation rates were obtained by fitting the measured cell lengths within each cell cycle to an exponential growth model. Additional experimental and analysis details are provided in the Supplementary Material.

Following transfer into the confinement chamber, approximately 10–15 min was required to identify isolated founder cells and initiate time-lapse acquisition, which is shorter than a typical cell cycle (∼36\sim36 min at 30°C). Cells present in the first recorded frame are therefore defined as generation Gen. 0, and their first division marks entry into Gen. 1.

For each cell cycle, the cell length immediately after division is defined as the birth length LbL_b, and the length immediately before the subsequent division as the division length LdL_d. The added length is ΔL=Ld−Lb\Delta L = L_d - L_b, and the division time is denoted by τd\tau_d. The corresponding division rate is λd=ln⁡2/τd\lambda_d = \ln2/\tau_d. Single-cell elongation rates, λe\lambda_e, were obtained by fitting cell-length trajectories to an exponential growth model (Figure 1(e)). Under ideal exponential growth with constant λe\lambda_e and Ld=2LbL_d = 2L_b, the division and elongation rates satisfy λd=λe\lambda_d = \lambda_e.

We observed a pronounced reduction in the single-cell elongation rate λe\lambda_e when exponentially growing E. coli cells were transferred from liquid culture into three different confinement conditions (Figure 2, Supplementary Fig. S2). Dispersed cells in liquid culture elongated at an average rate of ⟨λe⟩liq≈0.032 min−1\langle\lambda_e\rangle_{\mathrm{liq}} \approx0.032\,\mathrm{min}^{-1}, whereas cells in the initial confined generation (Gen. 0) elongated at reduced rates of approximately 0.019 min−10.019\,\mathrm{min}^{-1} in cells attached to glass in liquid LB medium (glass–liquid medium), 0.018 min−10.018\,\mathrm{min}^{-1} under glass–soft gel confinement and 0.013 min−10.013\,\mathrm{min}^{-1} under glass–hard gel confinement (Figure 2(d,Figure 2e)). Notably, cells continued to elongate exponentially after transfer, indicating that the confined environment remained permissive for growth. However, the substantial reduction in λe\lambda_e demonstrates that the transition into confinement, often regarded as a passive imaging condition in bacterial physiology experiments, represents an abrupt change in the mechanical environment that induces a pronounced physiological response, even in the absence of gel compression.

Figure 2. Growth acclimation after switch in mechanical environment. (a)–(c) Experimental conditions of increasing confinement strength: (a) glass–liquid medium (glass-attached state in liquid LB medium), (b) glass–soft gel (1.5% agarose), and (c) glass–hard gel (6% agarose). The representative colony morphologies at 150 min, around Gen. 4. (d) Violin plots of single-cell elongation rate λe\lambda_e across generations following transfer into glass–soft gel confinement. Cells with outlier values beyond three standard deviations (z-score > 3) were excluded. Black crosses mark the median, while vertical end markers denote the minimum and maximum values. Cells exhibit an initial reduction in elongation rate followed by gradual recovery over approximately 3–4 generations. The mean elongation rate of dispersed cells in liquid culture is indicated by the red dashed line. (e) Recovery of elongation dynamics under different confinement strengths. All the elongation rates measured from Gen. 0 to Gen. 5 were fitted with the exponential recovery function λe(t)=λ∞−Δλe−t/τ\lambda_e(t) = \lambda_\infty- \Delta\lambda e^{-t/\tau} for cells grown under glass–liquid medium, glass–soft gel and glass–hard gel confinements. Mean λe±\lambda_e \pm s.d. is shown for each generation, with data points from different confinement conditions slightly shifted along the xx axis slightly shifted to improve visual clarity.

Across confinement conditions, elongation rates recovered approximately exponentially over multiple generations toward a plateau, consistent with a single-timescale acclimation process, λe(t)=λ∞−Δλe−t/τ\lambda_e(t) = \lambda_\infty- \Delta\lambda e^{-t/\tau}. Increasing confinement strength generally reduced the recovered elongation rate λ∞\lambda_\infty and increased the magnitude of the initial growth suppression Δλ\Delta\lambda. In contrast, the characteristic recovery timescale τ\tau remained on the order of a few generations and did not show a clear monotonic dependence on confinement strength (see Supplementary Fig. S3). The recovery was most pronounced during the first 3–4 generations following confinement.

At later generations, such as Gen. 6 and Gen. 7, λe\lambda_e decreases again in a subset of cells, likely reflecting the collective mechano-response associated with increasing colony density [24]. Furthermore, stress buildup in the colony center can rotate cells and extrude them out of the monolayer [34, 38, 50, 51]. This mechanically driven monolayer-to-multilayer transition marks an early step toward biofilm-like colony architecture and can compromise reliable single-cell tracking. We therefore focus our analysis on the early acclimation phase preceding substantial multilayer growth.

A second hallmark of the confinement-induced stress response is filamentation [26, 27, 29, 30], characterized by continued cell elongation despite delayed or impaired division. Beginning at Gen. 2, a subpopulation of cells (∼7%\sim7\% in area fraction, see Supplementary Fig. S9) fails to divide within the expected cell cycle and continues to elongate while neighboring cells undergo multiple generations of division, before eventually exhibiting irregular division events. Representative are shown in Supplementary Fig. S8(a–e). These observations indicate that cell elongation can continue despite substantial disruption of division under confinement. Similar filamentous phenotypes have previously been observed under diverse stress conditions and are generally associated with temporary suppression of division while growth continues. Additional analyses of filamentation, division-length distributions, and added-length statistics are provided in the Supplementary Figs. S5–S9.

The emergence of filamentation demonstrates that mechanical confinement can transiently disrupt the coordination between growth and division during the acclimation phase. This observation raises the question in how far also the growth–division coordination of nonfilamentous cells might be affected and whether size homeostasis remains preserved under mechanical confinement. To this end, our further quantitative analysis was restricted to cells completing identifiable division cycles, while highly filamentous cells with unreliable contour extraction were excluded.

Birth-size statistics remain robust under mechanical confinements

To examine how bacterial cell size at birth responds to sudden mechanical confinement, we examined the distribution of birth lengths LbL_b across generations. Figure 3(a) shows violin plots of LbL_b from Gen. 1 to Gen. 7 (sample sizes ranging from 37 to 543 cells per generation). Despite the pronounced reduction in elongation rate following confinement, the mean birth length remains approximately constant throughout the mechanical acclimation process. The birth lengths generally follow a lognormal distribution (see Supplementary Fig. S5). This is consistent with stochastic multiplicative fluctuations during exponential growth, moderated by different size-control mechanisms [52].

Figure 3. Birth-size statistics remain robust under mechanical confinement. (a) Violin plots of birth length LbL_b across generations following transfer into glass–soft gel confinement, showing that the characteristic birth size remains bounded despite the transient reduction in elongation rate. White bars denote the mean, black crosses the median, and vertical whiskers the distribution extrema. Analysis was restricted to cells completing identifiable division cycles; highly filamentous cells with unreliable contour extraction were excluded. (b) Empirical cumulative distribution functions (ECDFs) of LbL_b after zz-score normalization. The approximate collapse of the normalized ECDFs indicates that the standardized birth-length distributions retain a similar shape across generations. The normalization is defined as z=(Lb−μ)/σz = (L_b - \mu)/\sigma, where μ\mu and σ\sigma are the mean and standard deviation of LbL_b for each generation, respectively. Inset: Coefficient of variation (CV) of LbL_b across generations. Variability increases transiently during the acclimation phase before partially recovering at later generations.

Mechanical confinement nevertheless transiently increases the cell-to-cell variability. As shown in the inset of Figure 3(b), the coefficient of variation increases during the early acclimation phase and subsequently decreases as elongation dynamics recover. The relatively low coefficient of variation observed at Gen. 1 may partially reflect sampling bias associated with cell selection at the onset of time-lapse imaging.

To determine whether mechanical confinement alters the distributional form of LbL_b, we computed empirical cumulative distribution functions (ECDFs) after applying a zz-score normalization. As shown in Figure 3(b), the normalized ECDFs collapse onto a common curve across generations, demonstrating that the underlying lognormal distributional shape of LbL_b remains largely invariant throughout the acclimation phase. Thus, the apparent generational differences in the acclimation phase merely reflect a transient heteroscedasticity, rather than systematic changes in the underlying distributional shape.

The preservation of bounded birth-size statistics with little changes in the mean across generations alongside the previously identified multi-generational growth acclimation necessitates to clarify the role of the underlying size-control mechanism. To this end, we established a general framework for generating transient size-control dynamics using numerical simulations of idealized adder, timer, and sizer mechanisms, as a quantitative tool to identify and model a rewired trigger of cell division during mechanical acclimation.

Classification framework for transient size-control dynamics

Different phenomenological strategies have been proposed to characterize bacterial size control, most notably the sizer, timer, and adder, each predicting distinct relationships among birth length LbL_b, division length LdL_d, added length ΔL\Delta L, and division time τd\tau_d. In an ideal sizer, cells divide upon reaching a characteristic size, such that LdL_d is independent of LbL_b and ΔL\Delta L decreases with increasing LbL_b. In an ideal timer, cells divide after a characteristic time, such that τd\tau_d is independent of LbL_b and, under exponential elongation, both LdL_d and ΔL\Delta L increase with LbL_b. In an ideal adder, cells add a characteristic length between birth and division, such that ΔL\Delta L is independent of LbL_b and LdL_d increases linearly with LbL_b with a slope of approximately unity (Figure 4). Under steady growth conditions, single-cell measurements have shown that adder-like behavior predominates across a wide range of bacterial species and growth environments [45, 46].

Figure 4. Unified classification of canonical size-control signatures. Two-dimensional density distributions of simulated populations generated under pure adder (left column: a, d, g), sizer (middle column: b, e, h), and timer (right column: c, f, i) control mechanisms. Dashed lines show linear regressions used to extract the slope of each variable pair. Heatmap colors indicate the sign of the correlation: green for positive, purple for negative, and gray for no correlation. Experimental data from representative adder-like cells (glass–soft gel confinement, Gen. 5) and timer-like cells (glass–soft gel confinement, Gen. 1) were non-dimensionalized by the mean birth length and doubling rate and overlaid onto the corresponding heatmaps as red dots, together with the linear regression lines (red solid lines). The characteristic slopes are stated in Eqs. (6) to (8).

These idealized correlation signatures provide a simple framework for distinguishing size-control mechanisms. More general stochastic frameworks describe division through a state-dependent division or hazard rate, providing a unified description from which sizer-, adder-, or timer-like behavior can emerge [47, 53, 54, 55]. In experimental populations, however, stochastic variations in growth, division, and cell size can broaden the underlying distributions and blur the characteristic signatures, particularly when sample sizes are limited [56, 57, 58, 59]. Their interpretation becomes even less straightforward during mechanical acclimation, when the elongation rate itself changes across generations. We therefore constructed a minimal stochastic cell-cycle model implementing pure sizer, timer, and adder control to determine how their characteristic correlation signatures are modified by noise and changing growth rate, and to establish a reference for interpreting the experimentally observed correlations.

In this model, each cell elongates exponentially with a generation-dependent specific growth rate,

dlidt=λi(g)li+σliξi(t),(1)\frac{\mathrm{d}l_i}{\mathrm{d}t} = \lambda_i(g)l_i + \sigma l_i \xi_i(t), \tag*{(1)}

where lil_i is the length of cell ii, gg labels its generation, λi(g)\lambda_i(g) is drawn at birth from a generation-dependent distribution, σ\sigma denotes the noise intensity of single-cell elongation fluctuations, and ξi(t)\xi_i(t) represents uncorrelated Gaussian white noise with zero mean and unit variance. The mean growth rate ⟨λ(g)⟩\langle\lambda(g)\rangle was chosen to follow the experimentally observed recovery after confinement, using the same generation-resolved data shown in Figure 2, so that the model explicitly incorporates the physiological growth acclimation. Division was then imposed by one of three idealized control rules: an adder, in which division occurs after adding a stochastic target length ΔLtarget\Delta L_{\mathrm{target}}; a sizer, in which division occurs at a stochastic target division length LtargetL_{\mathrm{target}}; or a timer, in which division occurs after a stochastic target interdivision time τtarget\tau_{\mathrm{target}}. At division, the mother cell of length LdL_{\mathrm{d}} was partitioned into two daughters with birth lengths fLdfL_{\mathrm{d}} and (1−f)Ld(1-f)L_{\mathrm{d}}, where ff fluctuates around 1/21/2.

To compare model predictions with experimental single-cell data across different growth conditions, we non-dimensionalize the cell-cycle variables using fixed reference scales. The characteristic length is defined as L0≡⟨ΔLadder⟩L_0 \equiv\langle\Delta L_{\mathrm{adder}}\rangle, the mean added length in the adder regime (Gen. 2 ∼ 5 for glass–soft gel confinement). The reference doubling time as τ0≡ln⁡(2)/λ∞\tau_0 \equiv\ln(2)/\lambda_{\infty}, where λ∞\lambda_{\infty} is the recovered elongation rate obtained from the exponential acclimation fit. The rescaled variables are

L∗=LL0,τd∗=τdτ0,λd∗=λdτ0.(2)L^{*} = \frac{L}{L_0}, \qquad\tau_{\mathrm{d}}^{*} = \frac{\tau_{\mathrm{d}}}{\tau_0}, \qquad\lambda_{\mathrm{d}}^{*} = \lambda_{\mathrm{d}}\tau_0. \tag*{(2)}

These reference scales are kept fixed across generations.

The three idealized size-control strategies predict distinct correlation signatures between cell-cycle observables. We characterize these signatures using the slopes

mΔ≡Slope⁡(ΔL∗,Lb∗),(3)m_{\Delta} \equiv\operatorname{Slope}\left(\Delta L^{*}, L_{\mathrm{b}}^{*}\right), \tag*{(3)}
md≡Slope⁡(Ld∗,Lb∗),(4)m_{\mathrm{d}} \equiv\operatorname{Slope}\left(L_{\mathrm{d}}^{*}, L_{\mathrm{b}}^{*}\right), \tag*{(4)}
mλ≡Slope⁡(λd∗,Lb∗)(5)m_{\lambda} \equiv\operatorname{Slope}\left(\lambda_{\mathrm{d}}^{*}, L_{\mathrm{b}}^{*}\right) \tag*{(5)}

expected from the (quasi-) linear relationship between selected pairs of normalized variables. For an ideal adder, the added length ΔL\Delta L is independent of birth length LbL_{\mathrm{b}}, while Ld=Lb+ΔLL_{\mathrm{d}} = L_{\mathrm{b}} + \Delta L. The corresponding slopes are

mΔ=0,md=1,mλ≈12.(6)m_{\Delta} = 0,\qquad m_{\mathrm{d}} = 1,\qquad m_{\lambda} \approx\frac{1}{2}. \tag*{(6)}

For an ideal sizer, division length LdL_{\mathrm{d}} is independent of LbL_{\mathrm{b}}, while ΔL=Ld−Lb\Delta L = L_{\mathrm{d}} - L_{\mathrm{b}}. This yields

mΔ=−1,md=0,mλ≈1.(7)m_{\Delta} = -1,\qquad m_{\mathrm{d}} = 0,\qquad m_{\lambda} \approx1. \tag*{(7)}

For an ideal timer, division time τd\tau_d is independent of LbL_b and, under exponential elongation, Ld=Lbexp⁡(λeτd)L_d = L_b \exp(\lambda_e\tau_d), therefore

mΔ≈1,md≈2,mλ=0.(8)m_{\Delta} \approx1,\qquad m_d \approx2,\qquad m_{\lambda} = 0. \tag*{(8)}

Note that the predicted values of mλm_{\lambda} for the adder and sizer correspond to local linear approximations near the reference birth length, as the division-rate relationships are generally nonlinear. Further details can be found in the Supplementary Material.

The idealized slopes in Eqs. (6) to (8) provide a compact classification scheme for distinguishing between the adder, sizer, and timer size-control mechanisms. They serve as benchmarks for interpreting the experimental correlation signatures by fitting linear regression models (see Figure 4), rather than as an exact description of every generation. Detailed derivations and statistical comparisons are provided in the Supplementary Material.

Our simulation framework allows the physiological recovery of growth rate and the structural rule of division control to be varied independently. We thus further examined whether correlations between elongation rate λe\lambda_e and birth length LbL_b could confound the classification of size-control dynamics. Numerical simulations show that adder- and sizer-like regulation remain robustly distinguishable over a broad range of coupling strengths, whereas timer-like signatures can mimic adder-like slopes but only under unrealistically strong negative λe\lambda_e–LbL_b correlations (Supplementary Fig. S18). These results demonstrate that the proposed classification framework is robust to the level of growth-rate heterogeneity observed experimentally.

Transient departure from canonical adder regulation

Applying the classification framework to experimental data reveals a confinement-strength-dependent remodeling of bacterial size-control dynamics (Figure 5). The canonical adder behavior, with ΔL\Delta L largely independent of LbL_b and the LdL_d–LbL_b slope remaining near unity, is observed under all confinement conditions apart from an initial transient under gel confinement (Figure 5(j–Figure 5l)). Under weak confinement (glass–liquid medium), cell-size statistics show these typical adder signatures throughout the experiment (Figure 5(d,Figure 5g), Supplementary Fig. S10). Consistently, statistical tests of the generation-resolved regression slopes show no significant departure from the ideal adder prediction for added length, while rejecting the timer prediction across all measured generations.

Figure 5. Mechanical-stress-dependent rewiring of size-control signatures. (a)–(c) Experimental conditions of increasing confinement strength: glass–LB, glass–1.5% agarose gel, and glass–6% agarose gel. (d)–(f) Correlations between ΔL\Delta L and LbL_b during the early acclimation phase (Gen. 1). (g)–(i) Corresponding correlations after recovery (Gen. 4). (j)–(l) Generation-resolved evolution of mΔm_{\Delta}, mdm_d, and mλm_{\lambda}. Note the rescaled variables λd∗=λd⋅τ0\lambda_d^{*}=\lambda_d\cdot\tau_0, Lb∗=Lb/⟨ΔL⟩L_b^{*}=L_b/\langle\Delta L\rangle. Intermediate and strong confinement transiently shift cells away from canonical adder regulation before recovery at later generations. See Supplementary Figs. S10–S13 for the full generation-resolved correlation signatures.

By contrast, glass–soft gel (1.5% agarose) confinement induces a clear transient timer-like shift during Gen. 1: ΔL\Delta L becomes positively correlated with LbL_b, the LdL_d–LbL_b slope exceeds unity, and the correlation between τd\tau_d and LbL_b is weakened compared to all later generations (Figure 5(e, Figure 5j–Figure 5l), Supplementary Figs. S11–S12). This transient shift is statistically supported by the added-length slope (mΔ=1.61m_{\Delta} = 1.61, 95% CI [0.74, 2.47]), which significantly departs from the adder prediction (mΔ=0m_{\Delta} = 0, p<0.001p < 0.001) while remaining compatible with the ideal timer prediction (mΔ=1m_{\Delta} = 1). The rescaled division-rate slope (mλ=−0.10m_{\lambda} = -0.10) provides complementary evidence, significantly departing from the adder reference value (mλ≈0.5m_{\lambda} \approx0.5) while remaining statistically compatible with the zero value of a timer. These signatures indicate a transient departure from canonical adder regulation, consistent with partial decoupling between growth and division. Previous studies have shown that bacterial size-control behavior can vary with physiological state: adder-like regulation is commonly observed during faster growth, whereas slower growth can produce more sizer-like behavior [60, 61, 62]. In our experiments, however, the transient departure from adder regulation is timer-like rather than sizer-like.

Under glass–hard gel (6% agarose) confinement, a transient shift toward timer-like regulation is observed in both the first and second generations (Figure 5(f, Figure 5j–Figure 5l), Supplementary Fig. S13). Both generations exhibit positive added-length slopes (mΔ=0.65m_{\Delta} = 0.65 and 1.00, respectively), together with rescaled division-rate slopes (mλ=0.22m_{\lambda} = 0.22 and 0.17, respectively) approaching the ideal timer prediction (mλ=0m_{\lambda} = 0). Although the limited sample size (n=26n = 26) and substantial variability in Gen. 1 result in wide confidence intervals that prevent a statistically significant distinction between the two ideal mechanisms, Gen. 2 shows significant departures from the adder hypothesis for both observables (p<10−4p < 10^{-4} for mΔm_{\Delta} and p=0.002p = 0.002 for mλm_{\lambda}), while remaining statistically consistent with the timer hypothesis. These results support a timer-like response extending over two generations. This prolonged transient and the increased heterogeneity may reflect stronger mechanical confinement and its effects on elongation and division dynamics. Moreover, the smaller cell size observed under hard-gel confinement (Supplementary Fig. S5) is consistent with the growth slowdown and size reduction previously reported in crowded bacterial colonies [24].

Across all three confinement conditions, adder-like size regulation is observed or recovered at different generations depending on confinement strength: from Gen. 1 under glass–liquid conditions, Gen. 2 under glass–soft gel confinement, and Gen. 3 under glass–hard gel confinement. In these regimes, the ΔL\Delta L–LbL_b slope approaches zero (Figure 5(g–Figure 5j)), while the LdL_d–LbL_b slope approaches one (Figure 5(k)). A modest negative added-length slope in Gen. 7 under hard-gel confinement suggests a possible late-stage sizer-like tendency. Overall, these results indicate that stronger confinement prolongs the transient departure from adder-like regulation, followed by its gradual recovery. Notably, this recovery precedes the full recovery of the elongation rate (Figure 5(j–Figure 5l)), indicating that size-control coordination and bulk growth physiology relax on distinct timescales. Such a separation is consistent with nutrient-shift studies showing that division-related parameters can adjust on a timescale distinct from that of the overall growth rate [63, 64, 65, 66].

We further considered whether these transient correlation dynamics could simply arise from the gradual recovery of the elongation rate, rather than reflecting a genuine departure from the underlying adder size-control mechanism. To test this possibility, we simulated populations in which the experimentally measured generation-dependent elongation-rate statistics were imposed, but the division rule was fixed as an adder, sizer, or timer. Although these models reproduced the changing growth and division timescales, they failed to predict the observed evolution of the experimental correlation signatures (Figure 6), demonstrating that growth-rate recovery alone cannot account for the experimental observations.

Figure 6. A timer-like-to-adder-like transformation reproduces the experimental transient correlation dynamics. Comparison between experimental correlation slopes and simulations with candidate static and switching control models. Allowing the growth rate to recover over generations is insufficient to reproduce the measured slope trajectories when the control rule is fixed. Among the tested two-state transitions, only the timer-like-to-adder-like transformation captures the experimentally observed shift from early timer-like correlations to late adder-like regulation.

We also extended our models by allowing the effective size-control rule to change dynamically. Among all two-state transitions between adder-, sizer-, and timer-like regulation, only a transient timer-like-to-adder-like transition reproduced the experimental evolution of all three classification observables (Figure 6, purple solid line). The model captures both the early timer-like signatures, characterized by positive ΔL\Delta L–LbL_b correlations and superunity LdL_d–LbL_b slopes, and the subsequent restoration of canonical adder behavior. These results indicate that mechanical confinement does not merely rescale bacterial growth dynamics, but transiently remodels the effective growth–division coupling before canonical adder regulation is re-established.

Finally, we demonstrated that the colony can maintain the population-level size homeostasis despite the transient timer-like behavior inferred from the correlation analysis. To this end, we simulated a generalized timer-to-adder transition occurring beyond the first generation. Although the transient timer state results in a temporary broadening of the birth-size distribution, it remains bounded and rapidly converges back to a universal steady-state after the adder behavior is restored (Supplementary Figs. S16–S17). Therefore, a transient timer-like regulation persisting over only a few generations does not destabilize the global birth-size statistics, establishing this dynamic size-control mechanism as an effective strategy for mechanical acclimation to new environments.

Discussion

Bacteria frequently inhabit mechanically structured soft environments in which growth occurs under surface contact, crowding, and confinement. We have found that entry into such environments initiates a multi-generational reorganization of growth and division dynamics, highlighting mechanical acclimation as an important aspect of bacterial physiology beyond steady-state conditions. We use controlled surface attachment and agarose confinement to isolate this transition and follow its consequences across generations.

Our results show that transferring exponentially growing E. coli from liquid culture to surface-associated or confined environments induces an immediate reduction in elongation rate, followed by multi-generational acclimation of growth and division dynamics. Even glass attachment in liquid medium produces a measurable response, which becomes stronger and more prolonged with increasing stiffness of contacting gels. These findings demonstrate that surface attachment and confinement are not merely passive imaging conditions but physical perturbations that trigger dynamic physiological responses.

Under gel confinements, mechanical acclimation is also accompanied by a transient shift from canonical adder-like regulation toward timer-like correlations, followed by the restoration of adder-like behavior. Under soft-gel confinement, the shift was detected in Gen. 1, whereas under hard-gel confinement, the slope estimates suggested a two-generation transient.

Despite this transient reorganization, birth-size distributions remain bounded throughout acclimation. Simulations further show that timer-like regulation lasting only a few generations does not produce the size divergence expected under a sustained timer mechanism. Thus, transient changes in growth–division coordination can occur without disrupting size homeostasis.

Notably, adder-like regulation is restored before the elongation rate fully recovers, indicating distinct recovery timescales for size-control coordination and cellular growth. Our correlation-based framework provides a general approach for characterizing such transient responses under non-steady growth conditions. More broadly, our results highlight bacterial size control as a dynamical process beyond its conventional steady-state description: following an environmental change, growth–division correlations can be transiently reorganized and subsequently restored on timescales distinct from those governing the recovery of cellular growth.

Acknowledgments

We acknowledge financial support from the Institute for Advanced Studies, University of Luxembourg (AUDACITY Grant IAS-21/CAMEOS to A.S.), the Marie Skłodowska-Curie Actions Individual Fellowship (Project 101110587-MINIMA to C.J.), and the Luxembourg National Research Fund (FNR) through ATTRACT Investigator Grant (A17/MS/11572821/MBRACE) and a CORE Grant (C19/MS/13719464/TOPOFLUME/Sengupta) to A.S. A.S. thanks Nicola Storelli and Francesco Origgi for insightful discussions on bacterial physiology. G.H.P.N., R.W., and H.L. acknowledge funding from the Deutsche Forschungsgemeinschaft (DFG, German Research Foundation) through the Priority Programme SPP 2265 (Grant Nos. WI 5527/1-2, and LO 418/25-2).

AI Usage Disclosure: OpenAI ChatGPT (GPT-5.6 Sol) was used during manuscript preparation to assist with translation, language editing, condensation, organization of the text, and refinement of the presentation and logical structure of the arguments. All AI-assisted content was critically reviewed, verified, and revised by the authors, who take full responsibility for the manuscript. Generative AI was not used for experimental data analysis or figure generation.

Data Availability Statement

The data supporting the findings of this study, including the numerical data underlying the figures, have been deposited in Zenodo [67]. During peer review, the data are available to the editors and reviewers through a private repository link. The dataset will be made publicly available through Zenodo upon publication of this article.

References

References

  1. [1]H.-C. Flemming and S. Wuertz, Nat. Rev. Microbiol. 17, 247 (2019).
  2. [2]Y. M. Bar-On, R. Phillips, and R. Milo, Proc. Natl. Acad. Sci. USA 115, 6506 (2018).
  3. [3]C. Jin and A. Sengupta, Biophys. Rev. 16, 173 (2024).
  4. [4]A. Ebrahimi and D. Or, Water Resour. Res. 50, 7406 (2014).
  5. [5]G. Kondakova, N. Verkhovtseva, and S. Ostroumov, Mosc. Univ. Biol. Sci. Bull. 68, 119 (2013).
  6. [6]L. Philippot, C. Chenu, A. Kappler, M. C. Rillig, and N. Fierer, Nat. Rev. Microbiol. 22, 226 (2024).
  7. [7]N. Heimann, J. Alcalde, J. M. Miocic, S. J. T. Hangx, J. Kallmeyer, C. Ostertag-Henning, A. Hassanpouryouzb and, E. M. Thaysen, G. J. Strobel, C. Schmidt-Hattenberger, et al., Energy Environ. Sci. 14, 853 (2021).
  8. [8]N. S. Muhammad, B. Haq, D. Al Shehri, A. Al-Ahmed, M. M. Rahman, and E. Zaman, Energy Rep. 8, 461 (2022).
  9. [9]J. Kirch, A. Schneider, B. Abou, A. Hopf, U. Schaefer, M. Schneider, C. Schalch, C. Wagner, and C. Lehr, Proc. Natl. Acad. Sci. USA 109, 18355 (2012).
  10. [10]K. Wheeler, G. Cárcamo-Oyarce, B. Turner, S. Dellos-Nolan, J. Co, S. Lehoux, R. Cummings, D. Wozniak, and K. Ribbeck, Nat. Microbiol. 4, 2146 (2019).
  11. [11]R. Bansil and B. Turner, Adv. Drug Deliv. Rev. 124, 3 (2018).
  12. [12]C. M. Wu, K. M. Wheeler, G. Cárcamo-Oyarce, K. Aoki, A. McShane, S. S. Datta, J. L. Mark Welch, M. Tiemeyer, A. L. Griffen, and K. Ribbeck, npj Biofilms Microbiomes 9, 11 (2023).
  13. [13]H. Dang and C. R. Lovell, Microbiol. Mol. Biol. Rev. 80, 91 (2016).
  14. [14]H. Alcolombri, F. J. Peaudecrf, V. I. Fernandez, L. Behrendt, K. S. Lee, and R. Stocker, Nat. Geosci. 14, 775 (2021).
  15. [15]B. Borer, I. Zhang, A. Baker, G. O’Toole, and A. Babbin, PNAS Nexus 2, pgac311 (2023).
  16. [16]U. Alcolombri, A. Nissan, J. Slomka, S. Charlton, E. Secchi, I. Short, K. S. Lee, F. J. Peaudecerf, D. A. Baumgartner, A. Sichert, et al., Nat. Commun. 16, 3290 (2025).
  17. [17][17*] A. Persat, C. D. Nadell, M. K. Kim, F. Ingreamau, A. Siryaporn, K. Drescher, N. S. Wingreen, B. L. Bassler, Z. Gitai, and H. A. Stone, Cell 161, 988 (2015).
  18. [18]Y. F. Dufrêne and A. Persat, Nat. Rev. Microbiol. 18, 227 (2020).
  19. [19]T. E. Kimkes and M. Heinemann, FEMS Microbiol. Rev. 44, 106 (2020).
  20. [20]A. Siryaporn, S. L. Kuchma, G. A. O’Toole, and Z. Gitai, Proc. Natl. Acad. Sci. USA 111, 16860 (2014).
  21. [21]Y. Zhou, X. Zhou, J. Zhang, Y. Zhao, Z. Ye, F. Xu, and F. Li, Anal. Chem. 97, 5517 (2025).
  22. [22]P. Pelech, P. P. Navarro, A. Vettiger, L. H. Chao, and C. Allolio, Proc. Natl. Acad. Sci. USA 122, e2424441122 (2025).
  23. [23]H. H. Tuson, G. K. Auer, L. D. Renner, M. Hasebe, C. Tropini, M. Salick, W. C. Crone, A. Gopinathan, K. C. Huang, and D. B. Weibel, Mol. Microbiol. 84, 874 (2012).
  24. [24]R. Wittmann, G. P. Nguyen, H. Löwen, F. J. Schwarzen-dahl, and A. Sengupta, Commun. Phys. 6, 331 (2023).
  25. [25]E. K. Chu, O. Kilic, H. Cho, A. Groisman, and A. Levchenko, Nat. Commun. 9, 4087 (2018).
  26. [26]P. Kumar and A. Lichaber, Biophys. J. 105, 783 (2013).
  27. [27]S. Nepal and P. Kumar, Phys. Rev. E 97, 052411 (2018).
  28. [28]L. Le Blanc, B. Alric, R. Rollin, L. Xénard, M. Av-dovic, S. Goussard, L. Mazencq, M. Piel, J.-Y. Tinevez, M. Delaure, G. Duménil, and D. Bonazzi, Nat. Commun. 10.1038/s41467-026-76492-6 (2026).
  29. [29]O. B. Aguilar-Luviano, F. Santos-Escobar, S. Orozco-Barrera, and R. Peña-Miller, J. R. Soc. Interface 23, 20250959 (2026).
  30. [30]J. Cayron, A. Dedieu-Berne, and C. Lesterlin, Mol. Microbiol. 119, 237 (2023).
  31. [31]J. Männik, R. Driessen, P. Galajda, J. E. Keymer, and C. Dekker, Proc. Natl. Acad. Sci. USA 106, 14861 (2009).
  32. [32]F. Si, B. Li, W. Margolin, and S. X. Sun, Sci. Rep. 5, 11367 (2015).
  33. [33]P. Mañas and B. M. Mackey, Appl. Environ. Microbiol. 70, 1545 (2004).
  34. [34]R. J. Allen and B. Waclaw, Rep. Prog. Phys. 82, 016601 (2019).
  35. [35]N. A. Araújo, L. M. Janssen, T. Barois, G. Boffetta, I. Cohen, A. Corbetta, O. Dauchot, M. Dijkstra, W. M. Durham, A. Dussutour, et al., Soft Matter 19, 1695 (2023).
  36. [36]O. J. Meacock, A. Doostmohammadi, K. R. Foster, J. M. Yeomans, and W. M. Durham, Nat. Phys. 17, 205 (2021).
  37. [37]A. Sengupta, Front. Phys. 8, 184 (2020).
  38. [38]J. Dhar, A. L. Thai, A. Ghoshal, L. Giomi, and A. Sengupta, Nat. Phys. 18, 945 (2022).
  39. [39]G. Rani, G. H. P. Nguyen, R. Wittmann, H. Löwen, and A. Sengupta, Nascent biofilms on soft surfaces, arXiv:2512.21870 [physics.bio-ph].
  40. [40]M. Sreepadmanabh, M. Ganesh, P. Sanjenbam, C. Kurzthaler, D. Agashe, and T. Bhattacharjee, Nat. Commun. 15, 9561 (2024).
  41. [41]A. M. Hancock and S. S. Datta, Biophys. J. 123, 957 (2024).
  42. [42]P. Zimmer, P. Bittihn, and Y. G. Pollack, New J. Phys. 28, 084402 (2026).
  43. [43]S. Jun, F. Si, R. Pugatch, and M. Scott, Rep. Prog. Phys. 81, 056601 (2018).
  44. [44]P. Wang, L. Robert, J. Pelletier, W. L. Dang, F. Taddei, A. Wright, and S. Jun, Curr. Biol. 20, 1099 (2010).
  45. [45]M. Campos, I. V. Surovtsev, S. Kato, A. Paintdakhi, B. Beltran, S. E. Emebier, and C. Jacobs-Wagner, Cell 159, 1433 (2014).
  46. [46]S. Taheri-Araghi, S. Bradde, J. T. Sauls, N. S. Hill, P. A. Levin, J. Paulsson, M. Vergassola, and S. Jun, Curr. Biol. 25, 385 (2015).
  47. [47]A. S. Kennard, M. Osella, A. Javer, J. Grilli, P. Nghe, S. J. Tans, P. Cicuta, and M. Cosentino Lagomarsino, Phys. Rev. E 93, 012408 (2016).
  48. [48]F. Si, G. Le Treut, J. T. Sauls, S. Vadia, P. A. Levin, and S. Jun, Curr. Biol. 29, 1760 (2019).
  49. [49]A. Miguel, M. Zietek, H. Shi, A. Sueki, F. Corona, L. Maier, J. Verheul, T. den Blaauwen, D. Van Valen, A. Typas, et al., mBio 16, e02281 (2025).
  50. [50]Z. You, D. J. Pearce, A. Sengupta, and L. Giomi, Phys. Rev. X 8, 031065 (2018).
  51. [51]Z. You, D. J. Pearce, A. Sengupta, and L. Giomi, Phys. Rev. Lett. 123, 178001 (2019).
  52. [52]K. Hosoda, T. Matsuura, H. Suzuki, and T. Yomo, Phys. Rev. E 83, 031118 (2011).
  53. [53]M. Osella, E. Nugent, and M. Cosentino Lagomarsino, Proc. Natl. Acad. Sci. USA 111, 3431 (2014).
  54. [54]J. Grilli, M. Osella, A. S. Kennard, and M. C. Lagomarsino, Phys. Rev. E 95, 032411 (2017).
  55. [55]C. Nieto, J. Arias-Castro, C. Sánchez, C. Vargas-García, and J. M. Pedraza, Phys. Rev. E 101, 022401 (2020).
  56. [56]G. Facchetti, B. Knapp, F. Chang, and M. Howard, Biophys. J. 117, 1728 (2019).
  57. [57]G. Le Treut, F. Si, D. Li, and S. Jun, Front. Microbiol. 12, 721899 (2021).
  58. [58]L. Luo, Y. Bai, and X. Fu, Phys. Rev. Res. 5, 013173 (2023).
  59. [59]A. Genthon and P. Thomas, Cell size control in bacteria is modulated through extrinsic noise, single-cell- and population-growth, arXiv:2601.05193 [q-bio.PE].
  60. [60]M. Wallden, D. Fange, E. G. L. D. Lundius, Ö. Baltekin, and J. Elf, Cell 166, 729 (2016).
  61. [61]S. Lee, L. J. Wu, and J. Errington, MicrobiologyOpen 8, e876 (2019).
  62. [62]C. Nieto, C. A. Vargas-García, J. M. Pedraza, and A. Singh, npj Syst. Biol. Appl. 10, 61 (2024).
  63. [63]M. Panilio, J. Grilli, G. Tallarico, I. Iuliani, B. Sclavi, P. Cicuta, and M. Cosentino Lagomarsino, Proc. Natl. Acad. Sci. USA 118, e2016391118 (2021).
  64. [64]H. Shi, Y. Hu, P. D. Odermatt, C. G. Gonzalez, L. Zhang, J. E. Elias, F. Chang, and K. C. Huang, Nat. Commun. 12, 1975 (2021).
  65. [65]F. Büke, J. Grilli, M. C. Lagomarsino, G. Bokinsky, and S. J. Tans, Curr. Biol. 32, 870 (2022).
  66. [66]S. Pahujani, Y. Zhang, M. G. Stetter, and J. Krug, Genetics 234, iyag168 (2026).
  67. [67]C. Jin and A. Sengupta, Mechanical confinement induces a transient timer-like shift in bacterial size control, version 2, Zenodo (2025).

Paper details

vlnc:W_ugaz6m3q
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 0

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Physics and Astronomy

GPT-5.6 Sol (bare)

View statement
Looking for related papers…
Contents