Pregnancy as a dynamical paradox: robustness, control and birth onset

TL;DR

The study reframes labor onset as a controlled dynamical transition in a spatial network of uterine cells, governed by sparse adaptive feedback and modulated by noise and coupling. It combines a 2D lattice model with an adaptive control term and stochasticity to predict both Alvarez and Braxton-Hicks contractions and preterm birth as a boundary-crossing failure of regulation. A key contribution is the identification of a cost-aware operating point near criticality, where small fluctuations can sample impending instability while diffusion and multiple controlled sites reduce energetic cost. The work further introduces sentinel early-warning monitoring that uses trend metrics on local activity to anticipate labor transitions, offering testable predictions and potential therapeutic implications for mitigating preterm birth risk.

Abstract

The timing of human labor is among the most critical determinants of neonatal survival, yet the mechanisms that govern the transition from uterine quiescence to coordinated contractions remain elusive. Here we present a dynamical-systems framework that models the pregnant uterus as a spatially extended network of electrically excitable cells regulated by sparse adaptive feedback mimicking hormonal and mechanical influences. This approach reveals how stability during gestation and sensitivity near parturition can be simultaneously maintained through the interplay of control, network structure, and noise. Our analysis shows that spontaneous contractions such as Braxton-Hicks and Alvarez waves are not epiphenomena, but functional components that reduce control effort and preserve responsiveness. Moreover, we identify preterm labor as a boundary-crossing phenomenon arising when control fails to correctly interpret early-warning signals. These results establish a unifying mechanistic theory for labor onset, yield testable predictions, and suggest new therapeutic strategies to mitigate preterm birth risk.

Figures (13)

• Figure 1: Schematic representation of the 2D uterine network: active cells (big circles) can be either controlled (pink) or not (blue). They are coupled to each other with strength $D$. Each active cell is connected to a random number of passive cells (small black cirles) with coupling $C_r$.
• Figure 2: Sparse control in the global–synchronization regime. Left column (deterministic). (a) Phase boundary separating controlled ($f_o=0$) and non-controlled ($f_o>0$) phase, as a function of $n_c$ and $D$; (b–d) temporal dependence of the membrane‑potentials in the uncontrolled ($f_o=1$), 'almost' controlled ($f_o \simeq 0$) and fully controlled ($f_o = 0$) case respectively. Right column (stochastic). (e) Filled contours of control effort in the $n_c{-}\alpha$ plane for $D=1$ and noise amplitude $\sigma= 0.05$. Superimposed on the heatmap is a white dashed curve indicating the phase boundary $\alpha_c(n_c; D=1)$, i.e., the set of $(n_c,\alpha)$ values that separates the fully controlled regime ($f_o=0$) from the oscillatory regime ($f_o>0$) as determined in Fig. \ref{['figDS:main']}. (f) CCDF of the spatial size of spatio‑temporal clusters of oscillating nodes ("wave clusters") for $\alpha=\alpha_c(n_c)+0.01$ for different values of $n_c$ and noise amplitude $\sigma= 0.05$.
• Figure 3: Dynamic bifurcations. (a) Voltage over time for one isolated cell (Eq. \ref{['eq:sys1dPRIMA']}) with slowly varying $\alpha$. The dashed lines indicate, respectively from left to right, the static bifurcation point, the actual onset of oscillation, and the point at which oscillations are suppressed again. Note how the first and third $\alpha$ values are virtually identical, implying that the Hopf bifurcation delay is present only in one direction. (b) Power of the electric potential $V$ over a sliding window as a function of $\alpha$, first slowly decreasing and then increasing back, for different starting points $\alpha_0$. Parameters for (a) and (b) are the same ($r=10^{-5}$, $C_r=1$, $n_p=1$). (c) Square of the mean bifurcation shift, $(\Delta\alpha)^2$, versus the noise amplitude $\sigma$ (standard deviation of the additive noise) on a logarithmic $x$-axis for fixed $\alpha_0$. Here $\Delta\alpha = \langle \alpha_{\mathrm{onset}} \rangle - \alpha_c$ is averaged over 100 independent realizations (error bars show standard deviations). (d) Hysteresis in the spatially extended system. Parameters: $n_c=0.7$, $D=1$, $r=10^{-5}$. The histograms show the onset–oscillation time on the $\alpha$-decreasing branch with and without noise.
• Figure 4: Classification pipeline for detecting imminent parturition. The monitoring module processes the time series of membrane potentials from individual cells to extract a one-dimensional summary metric. It then computes the trend using the Kendall tau coefficient. If the trend exceeds a predefined threshold, the instance is classified as indicating imminent parturition.
• Figure 5: Performance of the sentinel module treated as a binary classifier. (a) Area under the ROC curve (AUC) as a function of $n_c$ for $D=1$. Each curve corresponds to a different subset of sites observed by the module: controlled cells, uncontrolled cells, or all cells. The AUC summarizes the trade-off between true positive rate and false positive rate when the decision threshold on $\tau_K$ is varied. (b) False negative rate (FNR) when the false positive rate (FPR) is constrained to be below 10%. Here, $\mathrm{FNR} = \frac{\# \text{ imminent parturitions not detected}}{\# \text{ imminent parturitions}}$ and $\mathrm{FPR} = \frac{\# \text{ quiescent cases incorrectly flagged}}{\# \text{ quiescent cases}}$. The notation “FNR @ FPR” means the FNR measured under the constraint that FPR $< 0.1$. (c–d) Box plots of the Kendall rank correlation coefficient$\tau_K$ between the average power per cell $\overline{P}_{\mathcal{S}}(t)$ (Eq. \ref{['eq:avg_pow_cell']}) and time $t$, computed over the most recent $W$ points of each time series. Panel (c) shows the box plots of $\tau_K$ when $\mathcal{S}$ is the set of controlled cells; panel (d) shows box plots of $\tau_K$ when $\mathcal{S}$ is the set of uncontrolled cells. In each panel, box plots are shown separately for the “approaching parturition” (positive) and “quiescent” (negative) classes, illustrating the degree of overlap between the two classes in $\tau_K$ space and thus how easily they can be separated by a threshold.