Demir, Ali Aslan; Combriat, Thomas Michel Daniel; Heyward, Catherine Anne; Tiainen, Hanna; Carlier, Aurélie & Dysthe, Dag Kristian
(2026).
A cellular automaton model of osteogenic differentiation reveals identifiability limits of endpoint assays.
Frontiers in Systems Biology.
6.
doi: 10.3389/fsysb.2026.1872774.
Full text in Research ArchiveShow summary
Introduction Standard differentiation assays sample cell states only at discrete time points, while the underlying progression unfolds continuously and heterogeneously across cells. As a result, different combinations of proliferation, commitment, and maturation dynamics can converge to similar endpoint measurements. This many-to-one mapping between latent trajectories and observable readouts constitutes a partially observed inverse problem that limits mechanistic interpretation. Although this ambiguity is inherent to many experimental systems, it is rarely examined using models that connect cell-state dynamics to assay-level quantities. Methods We present OsteoMin, a coarse-grained cellular automaton that links stochastic transitions between pre-osteoblast and osteoblast states to experimentally measurable readouts of alkaline phosphatase activity, collagen deposition, and mineralization. Model parameters were constrained using literature-reported kinetics and evaluated against dexamethasone and menaquinone-4 perturbations. The framework enables systematic analysis of how cell-state progression, matrix maturation, and external perturbations shape differentiation outcomes. Results OsteoMin reproduces qualitative assay trends and quantifies the identifiability limits of endpoint assays. Distinct perturbation families often produce similar endpoint responses (macro-F1 ≈ 0.42$), indicating limited discriminative power. Incorporating temporal trajectories improves separability (macro-F1 ≈ 0.78$), demonstrating that most identifiable information resides in marker dynamics rather than terminal measurements. Sobol analysis shows that early markers depend on proliferation timing, whereas late mineralization is governed by nonlinear matrix maturation and parameter interactions. Discussion These results show that endpoint assays constrain overall progression but do not uniquely identify underlying mechanisms. OsteoMin provides a framework linking differentiation dynamics to assay observables and a basis for assessing identifiability in endpoint-driven systems.
Cells in epithelial monolayers are commonly quantified using a projected area and implicitly assuming constant cell volume and prism-shaped cells. These 2.5D assumptions ignore the volume and mass dynamics that may accompany density fluctuation in confluent space-filling tissues but remain inaccessible to the labelling- and intensity-based imaging techniques used to date. Here, we use time-lapse quantitative phase imaging (QPI) to obtain spatially and temporally resolved maps of height, volume, and dry mass in MDCK epithelial monolayers under physiological conditions. Three findings follow. (i) Cellular dry mass concentration is maintained to within ∼ 4.5 % across the monolayer and over time, even during large-amplitude pulsations, ruling out fluid (water) transport as the dominant driver of height and area dynamics. (ii) The mean height of the monolayer increases from ∼ 5.5 to ∼ 9 µm and the mean cell volume decreases by ∼ 25 % as the cell density doubles, evidence of contact inhibition of cell size rather than constant cell volume. Cell heights vary up to 30%, oscillate with a ∼ 5 h period, and follow gamma-shaped distributions. (iii) At cellular scales, both segmented-cell tracking and continuum mass-flux analysis show that projected cell volume is not conserved; mass conservation is recovered only after coarse-graining over ∼ 2 cell diameters and ∼ 1.6 h. Two non-exclusive mechanisms may explain this: non-prismatic cell geometry produces apparent volume fluctuations even at constant true volume, and periodic mass exchange between cells and the ECM produces genuine cyclic changes in cellular dry mass. A propagating elastic mode in the dry-mass structure factor ( c s ∼ 15 µm/h) further indicates that the monolayer behaves as a viscoelastic active medium rather than a plug-flow fluid. Each of these results requires simultaneous, time-resolved access to height and dry mass at the cellular scale and is therefore newly visible with QPI applied to epithelial monolayers. Together, they question the common 2.5D assumptions of prism-shaped cells, constant volume, plug-flow kinematics, and argue that quantitative continuum and cell-based models of epithelial dynamics must incorporate dry-mass-density regulation and 3D cell geometry.
Recent clinical trial data show curative potential of cell therapy for diabetes, however the cells required are a bottleneck. Cell differentiation exhibits substantial variability, even among clones of stem cells generated from the same patient. Human experts struggle to see the difference between highly- and lowly-efficient cell clones early. We therefore propose an image-based deep learning model to guide the selection of the most efficient clones. We apply different deep learning models to learn the morphological differences between good and bad stem cell clones and classify them based on phase-contrast imaging. To gain insight into the learned features, we use layer-wise relevance propagation, and Fourier-based frequency analysis. Using an EfficientNet-V2-S model, we obtain a novel early prediction for the outcome of the differentiation process from patient-derived stem cells to β -cells using imaging. Clone level accuracy is 96.7 % at 53 hours after start of differentiation. The analysis of learned features shows that the structure of the cell population is an important predictive feature. This study is a proof-of-concept that deep learning combined with label-free imaging can be highly predictive and guide selection of stem cell clones, thereby reducing cost of β -cell production.
Låstad, Silja Borring; Abbasova, Nigar; Combriat, Thomas Michel Daniel & Dysthe, Dag Kristian
(2026).
Quantitative Phase Imaging of the Dynamics of Epithelial Monolayers.
SPIE Quantitative Phase Imaging XII.
13861.
doi: 10.1117/12.3084036.
Full text in Research ArchiveShow summary
This study uses two different quantitative phase imaging techniques (QPI) and for the first time measures the height, volume, and mass dynamics of Madin-Darby Canine Kidney (MDCK) monolayers. We demonstrate novel methods to determine the height of confluent monolayers of cells from 2D and 3D QPI data and validate that the two methods agree. We developed a novel cell segmentation method adapted to QPI images of confluent cell layers and present a robust measure of relative error. We also demonstrate that height statistics of cells can be obtained without segmenting the images. We obtain the following precisions of cell density (1 %), height (3 %), area (5 %) and volume (6 %). Cell height varies 15-25 % over a monolayer and increases 50-100 % when cell density doubles. The average refractive index and the dry mass fraction of the cells, on the other hand, are constant over the entire density range.
Gastruloids are 3D stem cell aggregate models for early embryogenesis that provide a unique platform to study how collective cell dynamics drive tissue symmetry breaking and axial elongation. Using 3D light sheet imaging, we show that a pulse of Chiron, a Wnt activator, induces coherent alignment of cell polarity during elongation. While nuclear elongation occurs with or without treatment, only Chiron-treated gastruloids exhibit quasi-long-range alignment of nuclear axes, linking cell polarity coherence to tissue-scale remodeling. A minimal physical model of polarized cells, incorporating alignment-dependent torques and polarity-mediated adhesion, reproduces symmetry breaking and elongation, demonstrating that local cell polarity alignment alone can drive tissue-scale convergence-extension flows.
Ho, Richard David Jun-Geet; Bøe, Stig Ove; Dysthe, Dag Kristian & Angheluta, Luiza
(2024).
Role of tissue fluidization and topological defects in epithelial tubulogenesis.
Physical Review Research (PRResearch).
6(2),
p. 1–10.
doi: 10.1103/PhysRevResearch.6.023315.
Full text in Research ArchiveShow summary
Cellular rearrangements, as primary sources of tissue fluidization, facilitate topological transitions during tissue morphogenesis. We study the role of intrinsic cell properties such as cell polarity and cell-cell adhesion in shaping epithelial tissues using a minimal model of interacting polarized cells. The presence of a vortex in the cell polarization poses the topological constraint that induces an inwards migration with the formation of a conical shape. Local rearrangements at the tip of the cone lead to the onset of tube formation. Switching between collective migration and structural rearrangements is key for balancing the contrasting tendencies, such as the tissue rigidity needed to preserve shape and the tissue fluidity allowing for topological transitions during tissue morphogenesis.
Låstad, Silja Borring; Abbasova, Nigar; Combriat, Thomas Michel Daniel & Dysthe, Dag Kristian
(2026).
Three Dimensional Dynamics of Epithelial Monolayers.
Full text in Research Archive
Låstad, Silja Borring; Abbasova, Nigar; Ho, Richard David Jun-Geet; Combriat, Thomas Michel Daniel & Dysthe, Dag Kristian
(2025).
Segmentation-free measurement of cell motion from quantitative phase imaging.
Full text in Research Archive
Låstad, Silja Borring; Abbasova, Nigar; Combriat, Thomas Michel Daniel & Dysthe, Dag Kristian
(2025).
Cell height, area and volume fluctuations in MDCK monolayers.
Full text in Research Archive
Schöb, Franziska Jessica; Dysthe, Dag Kristian; Scholz, Hanne; Binder, Alexander & Malthe-Sørenssen, Anders
(2026).
From Pixels to Phenotypes: Deep Learning on Label-Free Microscopy Data for Cell Fate Prediction How Algorithms Help Us See More.
Universitetet i Oslo.
Full text in Research ArchiveShow summary
In this thesis, we used supervised deep learning on label-free microscopy data to predict cell fate and state from purely visual information, without relying on fluorescent labeling or manual segmentation. Across three studies, we combined phase-contrast and holotomographic imaging with convolutional neural networks to tackle problems that are central to stem cell biology and cancer research: predicting the differentiation efficiency of induced pluripotent stem cell clones, identifying germ-layer lineages in differentiating cultures, and detecting macrophage-induced cancer cell death. By building segmentation-free pipelines that operated directly on image patches, we reduced the need for explicit cell masks and made our methods easier to deploy in realistic, noisy experimental settings.
The overarching results showed that biological label-free images contained sufficient morphological information for deep networks to recover biologically meaningful patterns, but also revealed clear limits to what static morphology could predict. In stem cell differentiation, our models already distinguished high- from lowefficiency clones early in the protocol, and learned image cues that matched intuitive notions of colony coverage and uniformity. In germlayer prediction, endpoint-trained models decoded lineage identity from holotomographic morphology and reconstructed a plausible timeline of lineage divergence when applied retrospectively, while struggling more in mixed cultures with overlapping lineages. In the cancer cell death study, deep learning distinguished dying from alive cells at the time of death, yet early prediction beyond a short pre-death window remained difficult, indicating that death morphology emerged abruptly rather than gradually over many hours.
Together, these findings support the view that deep learning on label-free microscopy could provide non-destructive, scalable readouts of cell fate, while emphasizing the need for time-aware models, richer training labels, and improved domain generalization to fully exploit biological label-free imaging in biomedical workflows.
