A human induced pluripotent stem cell model for the holistic study of epithelial-to-mesenchymal transitions Authors: Caroline Hookway, Antoine Borensztejn, Leigh K. Harris, Tiffany Barszczewski, Sara Carlson, Gokhan Dalgin, Suraj Mishra, Ellen M. Adams, Julie C. Dixon, Renske J. Dupar, Jacqueline H. Edmonds, Erik A. Ehlers, Alexandra J. Ferrante, Margaret A. Fuqua, Clare R. Gamlin, Philip Garrison, Janani Gopalan, Benjamin W. Gregor, Maxwell J. Hedayati, Victoria L. Hurless, Kyle N. Klein, Chantelle L. Leveille, Sean L. Meharry, Ricardo Mercado, Haley S. Morris, Gouthamrajan Nadarajan, Nivedita Nivedita, Sandra A. Oluoch, Serge E. Parent, Amber Phan, Brock Roberts, Ashwin Samudre, Emmanuel E. Sanchez, M. Filip Sluzewski, Lev S. Snyder, Derek J. Thirstrup, Hannah F. Thorp, John Paul Thottam, Julia R. Torvi, Gaea Turman Published: 2026 ABSTRACT The epithelial-to-mesenchymal transition (EMT) is a widely studied cell state change, yet differences in model design and measurement approaches limit comparison across studies. Addressing this challenge requires experimental model systems and analysis frameworks that support standardization across contexts. Here, we show that human induced pluripotent stem (hiPS) cells in defined cell culture geometries, two-dimensional colonies and three-dimensional lumenoids, enable multimodal measurements of EMT dynamics within a single experimental platform. Using fixed-cell and live-cell image-based assays, we quantify changes in cell migration, EMT-related molecular markers, cell–cell junction organization and interactions with the basement membrane, a specialized form of the extracellular matrix, during EMT induced in hiPS cells. We identify cell culture geometry-dependent differences in the timing of migration onset and show that basement membrane integrity can be quantitatively linked to these differences. Together, these results establish an imaging-based framework for analysis of cell state transitions and provide accessible datasets and tools. Main Cell state transitions are fundamental to biological processes but are difficult to define because cell states can be described by diverse functional, organizational, molecular, and environmental properties1,2. The epithelial-to-mesenchymal transition (EMT), which occurs in contexts ranging from embryonic development to cancer metastasis, exemplifies this challenge3,4. EMT has been described functionally as the loss of epithelial barrier properties and acquisition of migratory and invasive behavior5,6, organizationally as a transition from apicobasal polarity to front-rear organization with accompanying changes in cell adhesion and cell–cell junctions7,8, and molecularly by changes in the expression of EMT-related marker proteins and transcription factors (EMT markers)7. EMT also involves altered interactions with the cellular environment, including interactions between cells and with the basement membrane, a specialized sheet of extracellular matrix (ECM), which mesenchymal cells breach as they migrate to new environments9,10. However, individual experimental studies often define EMT using only one set of functional, organizational or molecular observations, and the relationships among these types of observations4 remain poorly understood. We established human induced pluripotent stem (hiPS) cells as an experimental platform for integrated analysis of the EMT. hiPS cells are a normal human cell type that can be cultured without immortalization or karyotypic abnormalities11. They are derived from reprogramming human somatic cells12, which involves a mesenchymal-to-epithelial transition (MET)13,14. Variability in MET15 and retention of epigenetic memory16 from the cell of origin suggest that hiPS cells may retain some mesenchymal features and could be primed to undergo EMT. EMT can be reproducibly induced in human embryonic stem cells and hiPS cells by activation of the Wnt signaling pathway using growth factors or small molecules17,18; whether the robustness of EMT induction is related to any starting mesenchymal character of hiPS cells remains unclear. The resulting EMT dynamics resemble gastrulation following primitive streak formation19. As such, hiPS cell differentiation provides a tractable system for studying coordinated EMT dynamics relevant to early human development. hiPS cells can also be cultured in defined cell culture geometries, including two-dimensional (2D) colonies composed of flat epithelial-like sheets and three-dimensional (3D) lumenoids, which are hollow, single-cell layer epithelial spheres with apical surfaces facing a central lumen and basal surfaces oriented outward. Application of dilute mouse-derived basement membrane proteins, such as Matrigel, enables conversion of 2D hiPS cell colonies into adherent 3D lumenoids20–22. Together, these defined cell culture geometries constitute an experimentally standardized model system for studying EMT across distinct observational perspectives. This simplified framework avoids the confounding effects of tissue-scale patterning, mechanical stress and signaling gradients that complicate interpretation of EMT during gastrulation in vivo23,24. We used migration onset timing, the expression of EMT markers and the reorganization and loss of cell–cell junctions as representative readouts of functional, molecular and organizational changes, respectively, during EMT in 2D and 3D hiPS cell culture geometries. Migration onset showed a consistent dependence on starting geometry, occurring later in 3D than in 2D cultures. In contrast, relationships between EMT marker expression dynamics and initial cell culture geometry varied by marker. Using live imaging of hiPS cell-derived collagen type IV during EMT, we found that cells cultured in 3D produce a surrounding basement membrane shell that cells must cross to migrate away. Therefore, we examined how migration onset relates to changes in the integrity of the basement membrane. The presence of this basement membrane seems to have delayed migration onset during EMT in 3D lumenoids by restricting egress of cells that had likely already delaminated, an effect not observed in 2D EMT conditions where the basement membrane was less established. Together, these results demonstrate the utility of this hiPS cell-based model for studying EMT across multiple types of observations. Results An hiPS cell-based model system for comparing EMT in 2D and 3D environmental contexts We developed a chemically induced EMT in hiPS cells to compare EMT dynamics across distinct observational modalities in different starting cell culture geometries. We implemented a streamlined and reproducible live-cell confocal imaging assay (Fig. 1a) across six hiPS cell lines expressing mEGFP-tagged cellular structures or HaloTag-tagged or mEGFP-tagged EMT-related transcription factors (Methods). In addition to visualizing tagged cellular structures, we investigated hiPS cell-derived basement membrane dynamics using a fluorescently labeled, human-specific anti-collagen IV antibody25 that labels hiPS-derived basement membrane but not mouse-derived Matrigel (Methods). Labeling of an independent basement membrane component, nidogen 1 (ref. 26), revealed a similar spatial pattern (Extended Data Fig. 1), confirming that collagen IV labeling reflected the presence and dynamics of the basement membrane. This assay enabled simultaneous observation of cell rearrangements, basement membrane dynamics and cell–basement membrane interactions throughout EMT shown with H2B–mEGFP, which is representative of behavior across all cell lines (Figs. 1b,c and 2a–e and Supplementary Videos 1–9; full dataset described in Methods and ‘Data availability’).Fig. 1Overview of conditions for imaging assay and hiPS cell culture geometries.a. Schematic of hiPS cell media and treatment regimes (colored rows) used to generate 2D cell culture geometries (row 1), form 3D lumenoids (row 5) and induce EMT in 2D colonies (rows 2–4) and 3D lumenoids (row 6). EMT conditions compared throughout the remainder of the manuscript are starred. b, Representative example frames at 12-h intervals of hiPS cell 2D colony growth from 30-min interval time lapses showing flat expansion of the colony as seen by the nuclei (H2B–mEGFP; green) and little hiPS cell-derived basement membrane labeling (anti-collagen IV; magenta). XY views show the maximum-intensity Z projections. YZ views are single Z slices through the approximate middle. c, As in b, but for 2D to 3D lumenoid formation, showing transition from a flat epithelial sheet to a 3D lumenoid surrounded by hiPS cell-derived basement membrane when grown in culture medium containing dilute Matrigel (Methods). Scale bar, 50 µm. d, Schematic representation of hiPS cell 2D colony growth and 2D to 3D lumenoid formation with nuclei as navy ellipses, cell boundaries as black lines and basement membrane as magenta lines. Data are findable in BFF (Methods) through the ‘used for’ annotation corresponding to the panel in which the data are shown. Schematic illustrations in a created in BioRender; Gaudreault, N. https://biorender.com/gjsnted (2026).Fig. 2EMT is influenced by starting hiPS cell colony geometry, environmental Matrigel and substrate adhesiveness.Nuclei expressing mEGFP-tagged H2B (green) and the basement membrane labeled with anti-collagen IV antibody (magenta) are shown. a, Upon EMT induction, cells from 2D hiPS cell colonies plated on Matrigel substrate without Matrigel in solution (colony EMT no Matrigel, selected healthy example) disperse (as seen by nuclei) and migrate away, remaining flat and depositing little hiPS cell-derived basement membrane. XY views show the maximum-intensity Z projections. YZ views are single Z slices through the approximate middle. b, hiPS cells grown on Matrigel with dilute Matrigel added during EMT induction (2D colony EMT) increase basement membrane deposition and form a transient nascent lumenoid before migration. c, hiPS cells plated on PLF instead of Matrigel with dilute Matrigel added during EMT induction (2D PLF colony EMT) show similar behavior to b, but cells are more spread and the transient lumenoid is flatter. d, Cells from preformed 3D lumenoids plated on Matrigel substrate (3D lumenoid EMT) migrate through the basement membrane and away from the collapsing lumenoid. Scale bar, 50 µm. e, AGAVE 3D rendering of images in d. f, Schematic summarizing colony dynamics during EMT (YZ side views) with nuclei as filled navy ellipses, cell boundaries as black lines and basement membrane as magenta lines. Data are findable in BFF (Methods) through the ‘used for’ annotation corresponding to the panel in which the data are shown. Overview of conditions for imaging assay and hiPS cell culture geometries. a. Schematic of hiPS cell media and treatment regimes (colored rows) used to generate 2D cell culture geometries (row 1), form 3D lumenoids (row 5) and induce EMT in 2D colonies (rows 2–4) and 3D lumenoids (row 6). EMT conditions compared throughout the remainder of the manuscript are starred. b, Representative example frames at 12-h intervals of hiPS cell 2D colony growth from 30-min interval time lapses showing flat expansion of the colony as seen by the nuclei (H2B–mEGFP; green) and little hiPS cell-derived basement membrane labeling (anti-collagen IV; magenta). XY views show the maximum-intensity Z projections. YZ views are single Z slices through the approximate middle. c, As in b, but for 2D to 3D lumenoid formation, showing transition from a flat epithelial sheet to a 3D lumenoid surrounded by hiPS cell-derived basement membrane when grown in culture medium containing dilute Matrigel (Methods). Scale bar, 50 µm. d, Schematic representation of hiPS cell 2D colony growth and 2D to 3D lumenoid formation with nuclei as navy ellipses, cell boundaries as black lines and basement membrane as magenta lines. Data are findable in BFF (Methods) through the ‘used for’ annotation corresponding to the panel in which the data are shown. Schematic illustrations in a created in BioRender; Gaudreault, N. https://biorender.com/gjsnted (2026). EMT is influenced by starting hiPS cell colony geometry, environmental Matrigel and substrate adhesiveness. Nuclei expressing mEGFP-tagged H2B (green) and the basement membrane labeled with anti-collagen IV antibody (magenta) are shown. a, Upon EMT induction, cells from 2D hiPS cell colonies plated on Matrigel substrate without Matrigel in solution (colony EMT no Matrigel, selected healthy example) disperse (as seen by nuclei) and migrate away, remaining flat and depositing little hiPS cell-derived basement membrane. XY views show the maximum-intensity Z projections. YZ views are single Z slices through the approximate middle. b, hiPS cells grown on Matrigel with dilute Matrigel added during EMT induction (2D colony EMT) increase basement membrane deposition and form a transient nascent lumenoid before migration. c, hiPS cells plated on PLF instead of Matrigel with dilute Matrigel added during EMT induction (2D PLF colony EMT) show similar behavior to b, but cells are more spread and the transient lumenoid is flatter. d, Cells from preformed 3D lumenoids plated on Matrigel substrate (3D lumenoid EMT) migrate through the basement membrane and away from the collapsing lumenoid. Scale bar, 50 µm. e, AGAVE 3D rendering of images in d. f, Schematic summarizing colony dynamics during EMT (YZ side views) with nuclei as filled navy ellipses, cell boundaries as black lines and basement membrane as magenta lines. Data are findable in BFF (Methods) through the ‘used for’ annotation corresponding to the panel in which the data are shown. hiPS cells cultured on Matrigel-coated glass formed epithelial-like 2D monolayer colonies with basal surfaces contacting the glass (‘hiPS cell 2D colony growth’ Fig. 1a row 1, and Methods). As these colonies grew, they produced little detectable collagen IV (Fig. 1b,d top row, and Supplementary Videos 1 and 2). Exposure of 2D colonies on Matrigel-coated glass to dilute Matrigel in the surrounding medium resulted in the formation of 3D lumenoids from 2D colonies that were firmly anchored to the substrate, facilitating microscopy (‘2D to 3D lumenoid formation’ Fig. 1a row 5, and Methods). Time-lapse imaging during this process revealed progressive basement membrane deposition beginning at colony edges, coincident with cells lifting their basal surfaces away from the glass. Edge cells then cinched together to enclose a central lumen, forming a nascent lumenoid, which subsequently inflated into a spherical, fully formed lumenoid as cells continued to proliferate and grow (Fig. 1c,d bottom row, and Supplementary Videos 3 and 4). Lumenoid closure occurred approximately 12–24 h after Matrigel addition. In contrast to 2D colonies, fully formed lumenoids were surrounded by a bright, continuous collagen IV shell, indicating robust basement membrane deposition. We induced EMT in these 2D and 3D hiPS cell cultures by activating the Wnt signaling pathway through small-molecule inhibition of GSK3β (CHIR99021; Methods). In 2D colonies cultured on Matrigel-coated glass (‘2D colony EMT no Matrigel’ Fig. 1a row 2, and Methods), cells contracted and increased in height for ~20–30 h before migrating away from the initial colony (Fig. 2a,f top row, and Supplementary Video 5). In this condition, collagen IV labeling was restricted to the glass surface. Frequent cell death led us to exclude this condition from further study. In contrast, when EMT was induced in 2D colonies in the presence of dilute Matrigel (‘2D colony EMT’ Fig. 1a row 3, and Methods), cell death was consistently reduced (Fig. 2b,f second row, and Supplementary Video 6). Under these conditions, collagen IV labeling was seen at the colony periphery and a nascent lumenoid was formed within the first ~12–24 h after EMT induction (Fig. 2b 12-h and 24-h columns), consistent with dilute Matrigel driving lumenoid formation. This nascent lumenoid was weakly coated with collagen IV and did not inflate before cells began migrating outward at ~20–30 h (Fig. 2b 24-h column), suggesting that EMT-associated migration interrupted progression toward a fully formed lumenoid because of the overlapping timescales of lumenoid formation and EMT. We also induced EMT in 2D colonies plated on a different substrate, poly(D-lysine), laminin and fibronectin (PLF), in the presence of dilute Matrigel (‘2D PLF colony EMT’ Fig. 1a row 4, and Methods). Colonies grown on PLF appeared flatter and more spread than those on Matrigel, consistent with greater substrate adhesiveness27 (Fig. 2a–c 0-h panels). Overall, dynamics in 2D PLF EMT (Fig. 2c,f third row, and Supplementary Video 7) resembled those observed in 2D colony EMT (Fig. 2b), including the timing of nascent lumenoid formation and cell migration. However, unlike 2D colony EMT, the nascent lumenoid in 2D PLF colony EMT remained flat (Fig. 2c 24-h panels). Lastly, EMT was induced in preformed lumenoids (Fig. 1a row 6, and Methods). Following EMT induction, lumenoids continued to grow as hollow spheres for ~24 h (Fig. 2d 0-h to 24-h panels), after which the lumen disappeared and cells migrated across the basement membrane and onto the glass (Fig. 2d, the 36 and 48 h panels, and Fig. 2f, bottom row, and Supplementary Video 8). A 3D rendering using AGAVE (Methods) revealed additional cells migrating through basement membrane regions not directly contacting the glass (Fig. 2e and Supplementary Video 9). The basement membrane surrounding preformed lumenoids was brighter and thicker than that associated with nascent lumenoids in 2D colony EMT or 2D PLF colony EMT (Fig. 2b–d 0-h panel and 24-h panels); the 2-day exposure of colonies to dilute Matrigel to preform the lumenoids in 3D lumenoid EMT (Fig. 1a row 6) likely provided more time for cells to deposit basement membrane before EMT induction than in the 2D conditions. By manipulating the substrate adhesiveness and Matrigel exposure, we generated three standardized starting cell culture geometries: flat 2D colonies on Matrigel (2D colony), flatter 2D colonies on PLF (2D PLF colony) and spherical 3D lumenoids formed in dilute Matrigel on a Matrigel substrate (3D lumenoid). These served as defined starting points for EMT induction, yielding the 2D colony EMT, 2D PLF colony EMT and 3D lumenoid EMT conditions, respectively (Fig. 1a, starred rows 3, 4 and 6, and Fig. 2b–d). Although EMT throughout this study was induced using GSK3β inhibition, comparable EMT phenotypes were also observed in all three culture geometries following induction with the TGFβ family member BMP4 (Methods and Extended Data Fig. 2a). BMP4-induced EMT was delayed or blocked by BMP receptor inhibition (Methods and Extended Data Fig. 2b,c). We used GSK3β inhibition for subsequent experiments because of its robust downstream activation of Wnt signaling and reduced variability relative to upstream receptor-mediated signaling. To support hiPS cells as an effective experimental platform for studying EMT, we generated a comprehensive dataset comprising 95 single-plane and 795 3D Z-stack live-cell time-lapse videos, together with 2,648 3D Z-stack images of fixed cells, all accompanied by standardized metadata and annotations (Methods). Cells migrate later in 3D lumenoid EMT than in 2D EMT conditions We developed a quantitative assay to follow up on the initial observation that cell migration began later after EMT induction in 3D lumenoids than 2D colonies (Fig. 2b–d). Migration onset time was measured from 3D time-lapse data as the time point immediately preceding a sharp increase in cell footprint area at the glass, calculated from the bottom two Z slices of a 3D all-cells mask generated from the bright-field channel (Fig. 3a–d, Extended Data Fig. 3 and Methods). We found that in H2B–mEGFP cells, migration onset occurred later in 3D lumenoid EMT (median: 27.0 h, min–max: 20.0–32.5 h) than in either 2D EMT condition, which were statistically indistinguishable (2D colony EMT, median: 21.0 h, min–max: 18.5–25.0 h; 2D PLF colony EMT, median: 22.5 h, min–max: 18.5–25.0 h; Fig. 3d and Methods). Across five additional cell lines, migration onset in 3D lumenoid EMT was consistently delayed relative to both 2D EMT conditions, whereas differences between the two 2D geometries were variable (Extended Data Fig. 4). These results suggest that initial colony geometry influences migration onset timing during EMT.Fig. 3Migration onset time differs for 2D colony and 3D lumenoid geometries.a, The 3D renderings of all-cells masks with the bottom two Z slices in magenta and all others in cyan. Scale bars, 50 µm. b, Area of the maximum-intensity projection of the bottom two Z slices (area at the glass) over time trajectories. The median (line) and interquartile range (shaded) are shown for 2D PLF colony EMT (n = 16), 2D colony EMT (n = 25) and 3D lumenoid EMT (n = 21) from the H2B–mEGFP cell line. c, Representative single-trajectory examples are shown from b with annotated migration onsets (black squares). d, Migration onset times for all traces in b. Box-and-whisker plots show the median and interquartile range; whiskers extend to values within 1.5× the interquartile range. Significance between conditions is denoted as follows: ****P < 0.0001 and ***P < 0.001; NS, not significant (P ≥ 0.05). e, The 3D rendered example showing nuclear centroids inside (yellow) and outside (cyan) the basement membrane mesh (gray) at selected time points (0, 30 and 35 h), illustrating no centroids outside the basement membrane, initial centroid exit and mixed inside–outside centroids, respectively. Scale bars, 100 µm. f, Fraction of nuclei with centroid outside the basement membrane over time. The median (line) and interquartile range (shaded) are shown for H2B–mEGFP 3D lumenoid EMT (n = 20). g, Representative single-trajectory example from f, corresponding to the example in e, with annotated migration onset (black square). h, Bland–Altman comparison of migration onset timing measured by the inside–outside basement membrane method in f and the footprint area method in b (n = 20). The separation between the bias-corrected fit (solid line) and the unity reference (dotted line) shows differences between the two measurement methods (Methods; details in Supplementary Table 3 and 4). Data are findable in BFF (Methods) through the ‘used for’ annotation corresponding to the panel in which the data are shown. Migration onset time differs for 2D colony and 3D lumenoid geometries. a, The 3D renderings of all-cells masks with the bottom two Z slices in magenta and all others in cyan. Scale bars, 50 µm. b, Area of the maximum-intensity projection of the bottom two Z slices (area at the glass) over time trajectories. The median (line) and interquartile range (shaded) are shown for 2D PLF colony EMT (n = 16), 2D colony EMT (n = 25) and 3D lumenoid EMT (n = 21) from the H2B–mEGFP cell line. c, Representative single-trajectory examples are shown from b with annotated migration onsets (black squares). d, Migration onset times for all traces in b. Box-and-whisker plots show the median and interquartile range; whiskers extend to values within 1.5× the interquartile range. Significance between conditions is denoted as follows: ****P < 0.0001 and ***P < 0.001; NS, not significant (P ≥ 0.05). e, The 3D rendered example showing nuclear centroids inside (yellow) and outside (cyan) the basement membrane mesh (gray) at selected time points (0, 30 and 35 h), illustrating no centroids outside the basement membrane, initial centroid exit and mixed inside–outside centroids, respectively. Scale bars, 100 µm. f, Fraction of nuclei with centroid outside the basement membrane over time. The median (line) and interquartile range (shaded) are shown for H2B–mEGFP 3D lumenoid EMT (n = 20). g, Representative single-trajectory example from f, corresponding to the example in e, with annotated migration onset (black square). h, Bland–Altman comparison of migration onset timing measured by the inside–outside basement membrane method in f and the footprint area method in b (n = 20). The separation between the bias-corrected fit (solid line) and the unity reference (dotted line) shows differences between the two measurement methods (Methods; details in Supplementary Table 3 and 4). Data are findable in BFF (Methods) through the ‘used for’ annotation corresponding to the panel in which the data are shown. We further validated that the delayed migration onset observed in 3D lumenoids was not an artifact of measuring migration at the glass rather than in 3D. Migration onset times detected using the change in the area of the footprint were compared to those obtained with a second method that detected cells exiting the basement membrane at regions at and above the glass (inside–outside basement membrane method; Methods, Fig. 3e–g and Supplementary Video 10). Migration onset times measured by each method were in close agreement (Fig. 3h and Methods), suggesting that, in 3D lumenoid EMT, the onset of cell spreading at the glass coincides temporally with cells exiting the basement membrane at and above the glass. Therefore, we used the footprint-area-based method for all subsequent analyses. The expression timing of canonical EMT markers is variable with respect to starting hiPS cell culture geometry The geometry-dependent differences in migration onset timing motivated an examination of whether EMT-associated molecular changes also varied with cell culture geometry. First, we confirmed that canonical EMT marker proteins exhibited expected expression trajectories upon EMT induction in hiPS cells. To do this, we fixed samples every 4 h across all 2D and 3D EMT conditions and labeled with a panel of eight antibodies consisting of transcription factors Brachyury, Eomesodermin (Eomes), Snail, and Twist 1; junctional proteins E-cadherin and N-cadherin; intermediate filament marker Vimentin; and an epigenomic marker H3K36me2 (Supplementary Table 1). Bulk fluorescence intensities measured within 3D all-cells masks (Methods) revealed that all eight proteins changed over the course of EMT (Extended Data Fig. 5 and Supplementary Figs. 1–3). The general pattern of changes in bulk expression occurred as expected for most of the EMT markers in the panel (Fig. 4, Extended Data Fig. 5 and Methods). For example, the classic switch from E-cadherin to N-cadherin expression over the course of EMT28–31 was observed across all conditions (Fig. 4a–c, Extended Data Fig. 5 and Supplementary Figs. 1–3). While the average intensities of some EMT markers did not change as expected, qualitative EMT-related changes in organizational and spatial distribution of these markers were observed (Supplementary Figs. 1–3). These results underscore the need for multiple types of observations, such as protein expression level and cellular organization, to be considered in describing EMT4.Fig. 4The expression of canonical EMT marker proteins changes during EMT similarly across different plating geometries and reveals differences in starting state.a, Representative immunofluorescence images of anti-E-cadherin labeling in samples fixed at 8-h intervals across EMT induced in all three hiPS cell culture geometries. Maximum-intensity Z projections are shown. Scale bar, 50 µm. b, As in a, but labeled with an anti-N-cadherin antibody. c, Heat map showing fluorescence intensity of eight canonical EMT markers quantified from immunolabeled samples fixed at 4-h intervals over the course of EMT in 3D Z stacks from 2D PLF colony EMT (n = 1,041), 2D colony EMT (n = 1,026) and 3D lumenoid EMT (n = 1,040). Data are findable in BFF (Methods) through the ‘used for’ annotation corresponding to the panel in which the data are shown. The expression of canonical EMT marker proteins changes during EMT similarly across different plating geometries and reveals differences in starting state. a, Representative immunofluorescence images of anti-E-cadherin labeling in samples fixed at 8-h intervals across EMT induced in all three hiPS cell culture geometries. Maximum-intensity Z projections are shown. Scale bar, 50 µm. b, As in a, but labeled with an anti-N-cadherin antibody. c, Heat map showing fluorescence intensity of eight canonical EMT markers quantified from immunolabeled samples fixed at 4-h intervals over the course of EMT in 3D Z stacks from 2D PLF colony EMT (n = 1,041), 2D colony EMT (n = 1,026) and 3D lumenoid EMT (n = 1,040). Data are findable in BFF (Methods) through the ‘used for’ annotation corresponding to the panel in which the data are shown. EMT marker antibody labeling also revealed cell culture geometry-dependent differences in marker expression patterns. E-cadherin expression provides a clear example: As expected for a classical EMT trajectory, 3D lumenoids displayed high initial E-cadherin levels that then decreased over the course of EMT. However, 2D colonies began with relatively low E-cadherin expression, which transiently increased just after EMT induction before declining to low and ultimately nearly absent levels by 48 h (Fig. 4a,c). The increase in E-cadherin expression within the first ~20 h in 2D EMT conditions may be a result of the nascent lumenoid formation that occurred, in the presence of Matrigel, around the same time (Fig. 2b,c). These results raise the possibility that hiPS cells may have different initial epithelial, mesenchymal or hybrid states in these different cell culture geometries and subsequently take different EMT trajectories. Immunolabeling enabled characterization of EMT markers across distinct cell populations at each time point. To characterize EMT marker expression dynamics within individual colonies and lumenoids, we performed live-cell imaging of hiPS cell lines expressing HaloTag-tagged or mEGFP-tagged versions of four EMT markers. Three of these markers, Brachyury, Eomes and E-cadherin, overlapped with the antibody panel. We additionally included mEGFP-tagged SOX-2, a stemness marker that is downregulated during EMT32 and cell differentiation33,34. Fluorescence intensities were quantified using the same approach as for the antibody panel (Methods) in images of hiPS cells undergoing EMT in each of the three cell culture geometries. As expected, SOX-2 intensity decreased monotonically following EMT induction in all geometries (Fig. 5a,b and Supplementary Videos 11–13), aside from a transient, noninterpretable spike (Fig. 5b,c, gray regions, and Methods). Consistent with immunolabeling results, Brachyury and Eomes intensities increased and then declined during EMT (Figs. 4c and 5f,g,k,l and Supplementary Videos 14–19), and E-cadherin intensity decreased in all geometries (Figs. 4c and 5p,q and Supplementary Videos 20–22). In addition, the initial expression levels of these markers differed by cell culture geometry, including higher E-cadherin intensities in 3D lumenoids compared to 2D colonies (Fig. 5p,q).Fig. 5Comparison of EMT marker expression dynamics across cell culture geometries and their timing relative to migration onset.a,f,k,p, Maximum-intensity Z projections show SOX-2–mEGFP (a), Brachyury–HaloTag (f), Eomes–mEGFP (k) and E-cadherin–mEGFP (p) in 2D PLF colony EMT (top row), 2D colony EMT (middle row) and 3D lumenoid EMT (bottom row). Dashed white lines outline colony or lumenoid perimeters where shown. Scale bars, 50 µm. b,g,l,q, Average fluorescence intensity trajectories (mean, line; interquartile range, shaded) are shown for SOX-2 (b; n = 5, 17 and 13), Brachyury (g; n = 10, 40 and 27), Eomes (l; n = 33, 40 and 38) and E-cadherin (q; n = 11, 23 and 30) for 2D PLF colony EMT, 2D colony EMT and 3D lumenoid EMT, respectively. A transient, noninterpretable fluorescence spike is indicated where present (gray regions; Methods). c,h,m,r, Representative single-trajectory examples are shown with marker-specific timing features indicated by black squares: time of half-maximal intensity for SOX-2 (c), time of maximal intensity for Brachyury (h) and Eomes (m) and time of inflection for E-cadherin (r). Gray regions indicate transient, noninterpretable fluorescence spikes where present (Methods). d,i,n,s, Timing summaries presented as box-and-whisker plots showing the mean and interquartile range; whiskers extend to values within 1.5× the interquartile range for SOX-2 half-maximal intensity (d), Brachyury maximal intensity (i), Eomes maximal intensity (n) and E-cadherin inflection time (s). Significance between conditions is denoted as follows: ****P < 0.0001 and ***P < 0.001; NS (P ≥ 0.05). e,j,o,t, Time of migration onset is plotted versus marker-specific timing metrics for SOX-2 (e), Brachyury (j), Eomes (o) and E-cadherin (t). Unity lines are shown (gray dotted). Correlation text is colored by cell culture geometry; gray values indicate correlations when all EMT conditions are considered together (details in Supplementary Tables 3 and 4). Data are findable in BFF (Methods) through the ‘used for’ annotation corresponding to the panel in which the data are shown. Comparison of EMT marker expression dynamics across cell culture geometries and their timing relative to migration onset. a,f,k,p, Maximum-intensity Z projections show SOX-2–mEGFP (a), Brachyury–HaloTag (f), Eomes–mEGFP (k) and E-cadherin–mEGFP (p) in 2D PLF colony EMT (top row), 2D colony EMT (middle row) and 3D lumenoid EMT (bottom row). Dashed white lines outline colony or lumenoid perimeters where shown. Scale bars, 50 µm. b,g,l,q, Average fluorescence intensity trajectories (mean, line; interquartile range, shaded) are shown for SOX-2 (b; n = 5, 17 and 13), Brachyury (g; n = 10, 40 and 27), Eomes (l; n = 33, 40 and 38) and E-cadherin (q; n = 11, 23 and 30) for 2D PLF colony EMT, 2D colony EMT and 3D lumenoid EMT, respectively. A transient, noninterpretable fluorescence spike is indicated where present (gray regions; Methods). c,h,m,r, Representative single-trajectory examples are shown with marker-specific timing features indicated by black squares: time of half-maximal intensity for SOX-2 (c), time of maximal intensity for Brachyury (h) and Eomes (m) and time of inflection for E-cadherin (r). Gray regions indicate transient, noninterpretable fluorescence spikes where present (Methods). d,i,n,s, Timing summaries presented as box-and-whisker plots showing the mean and interquartile range; whiskers extend to values within 1.5× the interquartile range for SOX-2 half-maximal intensity (d), Brachyury maximal intensity (i), Eomes maximal intensity (n) and E-cadherin inflection time (s). Significance between conditions is denoted as follows: ****P < 0.0001 and ***P < 0.001; NS (P ≥ 0.05). e,j,o,t, Time of migration onset is plotted versus marker-specific timing metrics for SOX-2 (e), Brachyury (j), Eomes (o) and E-cadherin (t). Unity lines are shown (gray dotted). Correlation text is colored by cell culture geometry; gray values indicate correlations when all EMT conditions are considered together (details in Supplementary Tables 3 and 4). Data are findable in BFF (Methods) through the ‘used for’ annotation corresponding to the panel in which the data are shown. We developed automated methods to quantify the timing of EMT marker expression from individual fluorescence intensity traces (Methods). To assess SOX-2 loss, we measured the time from induction to half-maximal intensity decrease for each trajectory across all conditions (Fig. 5b,c and Methods) and found similar loss kinetics of SOX-2 across conditions (Fig. 5d). We quantified Brachyury and Eomes dynamics by detecting the time of peak fluorescence intensity (Fig. 5g,l,h,m and Methods). Across EMT conditions, significant differences in peak timing were detected for both markers, although the differences varied by specific condition and marker. For Brachyury, peak timing was the same in 2D colony EMT and 3D lumenoid EMT, whereas Brachyury peaked earlier in both of those EMT conditions compared to 2D PLF colony EMT (Fig. 5i). For Eomes, on the other hand, peak timing was the same for 2D PLF colony EMT and 3D lumenoid EMT, while Eomes peaked later in both of those EMT conditions compared to 2D colony EMT (Fig. 5n). We quantified E-cadherin loss by automated detection of the inflection point preceding the rapid decrease in fluorescence intensity (Fig. 5q,r and Methods) and observed significant differences in loss kinetics among all three EMT conditions (Fig. 5s). Thus, while migration onset was consistently delayed in 3D lumenoid EMT compared to 2D EMT conditions (Fig. 3d and Extended Data Fig. 4), culture geometry-dependent EMT marker expression dynamics were marker specific (Fig. 5d,i,n,s). These results indicate that geometry-dependent differences in EMT marker expression dynamics alone do not explain delayed migration onset in 3D lumenoid EMT relative to 2D EMT conditions, motivating a within-sample comparison of marker expression dynamics and migration onset timing. To directly assess the relationship between EMT marker expression dynamics and migration onset, we compared migration onset time to the corresponding marker expression change within the same time-lapse images (Fig. 5e,j,o,t). When all EMT conditions were pooled, moderate to very strong correlations were observed between migration onset time and expression changes of Brachyury, Eomes and E-cadherin, in order of increasing strength (Fig. 5j,o,t and Methods). When correlations were evaluated within each EMT condition, most EMT marker–migration onset relationships remained statistically significant. Those that were not significantly correlated with migration onset included Brachyury expression in both 2D EMT conditions (Fig. 5j,o,t). In contrast, SOX-2 loss was never correlated with migration onset (Fig. 5e). As indicated by the relative positions of data points with respect to the unity line, the timing of EMT marker dynamics relative to migration onset differed by marker; E-cadherin inflection occurred slightly before migration onset, Eomes peaked near migration onset and Brachyury peaked after migration onset, whereas half-maximal SOX-2 loss occurred substantially earlier (Fig. 5t,o,j,e). Correlations between EMT marker expression dynamics and migration onset timing motivated tests of whether marker expression has a causal role in determining migration onset. We tested the effect of CRISPR interference (CRISPRi)-mediated TBXT or SNAI1 knockdown on migration onset timing in 3D lumenoid EMT. Across all knockdown conditions, migration onset was delayed relative to controls, although only a subset of single guide RNA sequences produced statistically significant effects (Extended Data Fig. 6). This demonstrates that it is possible, at least in some contexts, for marker expression to influence migration timing. Organizational changes in hiPS cell–cell junctions during EMT indicate the contraction of apical faces and cell delamination Because E-cadherin exhibited the most consistent relationship between migration onset and expression change across EMT conditions and because its loss consistently preceded migration onset (Fig. 5t), we reasoned that changes in E-cadherin might help explain differences in migration onset timing between 2D and 3D EMT conditions. Cadherins, core components of adherens junctions, regulate intercellular adhesion35,36 and change in relative expression over the course of EMT28, with a central role at cell–cell junctions. Changes in and subsequent loss of cell–cell junctions are hallmark organizational features of EMT. E-cadherin undergoes changes in expression level during subcellular relocalization (Fig. 4a), complicating its use as a readout of junctional remodeling. Therefore, we examined ZO-1, a component of tight junctions37, which are known to be reorganized and ultimately lost during EMT as cells lose apicobasal polarity38,39 but maintain localization to apical faces until junctions are lost (Extended Data Fig. 7). An mEGFP–ZO-1 cell line was used to visualize the tight-junction network dynamics (Fig. 6a, Extended Data Fig. 7a and Supplementary Videos 23–28). Before migration onset, the ZO-1 network shrank in area in the 2D EMT conditions and in luminal surface area in 3D lumenoid EMT (Fig. 6a, XY views, and Extended Data Fig. 7a). Simultaneously, the bottom layer of cells in 2D colony EMT and the lumenoid wall in 3D lumenoid EMT thickened, whereas cells in 2D PLF colony EMT remained relatively flat (Fig. 6a, YZ view, and Supplementary Video 23).Fig. 6Junctional reorganization and EMT-associated cell delamination across cell culture geometries in relation to basement membrane integrity.a, Maximum-intensity Z projections through five Z slices (XY views) and Y projections through 50 pixels (YZ views), centered on the ZO-1 network. Where upward displacement is observed, yellow bars indicate the distance from the glass to the apical ZO-1 network. Scale bar, 50 µm. b, Heat maps corresponding to the examples in a show the mean ZO-1 fluorescence intensity within all-cells masks as a function of Z height above the glass and time, with black dots indicating time points shown in a, yellow bars denoting the same network heights and vertical gray dashed lines marking migration onset. Black arrows indicate the accumulation of ZO-1 fluorescence at the glass around the time of cell migration. c, Schematic of proposed model showing the effect of different basement membrane environments in 2D colony EMT versus 3D lumenoid EMT on the relationship between the timing of delamination and migration onset. d, Maximum-intensity Z projection of basement membrane labeling (gray) showing the formation of perforation-like or hole-like openings (derived from the same image stack as in Fig. 2d,e). e,f, Migration onset times are summarized as box-and-whisker plots showing the median and interquartile range; whiskers extend to values within 1.5× the interquartile range. e, Migration onset time following addition of type IV collagenase at the onset of EMT induction in H2B–mEGFP cells in the 3D lumenoid EMT condition; sample sizes were n = 11, 18, 23, 20, 23 and 25 from left to right on the x axis. f, Migration onset time following addition of the pan-MMP inhibitor ilomastat at the onset of EMT induction in H2B–mEGFP cells in the 3D lumenoid EMT condition; sample sizes were n = 3 per group from left to right on the x axis, representing one biological replicate. Two additional biological replicates are shown in Supplementary Fig. 4 (details in Supplementary Tables 3 and 4). Data are findable in BFF (Methods) through the ‘used for’ annotation corresponding to the panel in which the data are shown. Junctional reorganization and EMT-associated cell delamination across cell culture geometries in relation to basement membrane integrity. a, Maximum-intensity Z projections through five Z slices (XY views) and Y projections through 50 pixels (YZ views), centered on the ZO-1 network. Where upward displacement is observed, yellow bars indicate the distance from the glass to the apical ZO-1 network. Scale bar, 50 µm. b, Heat maps corresponding to the examples in a show the mean ZO-1 fluorescence intensity within all-cells masks as a function of Z height above the glass and time, with black dots indicating time points shown in a, yellow bars denoting the same network heights and vertical gray dashed lines marking migration onset. Black arrows indicate the accumulation of ZO-1 fluorescence at the glass around the time of cell migration. c, Schematic of proposed model showing the effect of different basement membrane environments in 2D colony EMT versus 3D lumenoid EMT on the relationship between the timing of delamination and migration onset. d, Maximum-intensity Z projection of basement membrane labeling (gray) showing the formation of perforation-like or hole-like openings (derived from the same image stack as in Fig. 2d,e). e,f, Migration onset times are summarized as box-and-whisker plots showing the median and interquartile range; whiskers extend to values within 1.5× the interquartile range. e, Migration onset time following addition of type IV collagenase at the onset of EMT induction in H2B–mEGFP cells in the 3D lumenoid EMT condition; sample sizes were n = 11, 18, 23, 20, 23 and 25 from left to right on the x axis. f, Migration onset time following addition of the pan-MMP inhibitor ilomastat at the onset of EMT induction in H2B–mEGFP cells in the 3D lumenoid EMT condition; sample sizes were n = 3 per group from left to right on the x axis, representing one biological replicate. Two additional biological replicates are shown in Supplementary Fig. 4 (details in Supplementary Tables 3 and 4). Data are findable in BFF (Methods) through the ‘used for’ annotation corresponding to the panel in which the data are shown. To examine changes in the height of the ZO-1 network (along the Z dimension), we took advantage of the localization of ZO-1 to apical junctions in hiPS cells, which face upward in hiPS cell colonies on glass. This geometry enabled monitoring of the behavior of the connected apical junctional network before polarity loss. We then quantified ZO-1 fluorescence intensity within corresponding all-cells masks at each time point and Z slice and used this metric as a proxy for the vertical position of the apical network over time (Methods). This analysis revealed an upward shift of the apical network before migration onset in both 2D colony EMT and 3D lumenoid EMT, but not in 2D PLF colony EMT (Fig. 6b), and this pattern was highly reproducible (Extended Data Fig. 7b). This upward shift is consistent with cells becoming taller and/or delaminating from the epithelial-like sheet, such that cells exiting the network displaced the remaining junctional network upward. Single-cell analysis revealed constriction and closure of ZO-1-delineated apical faces before disappearance, supporting delamination through apical constriction (Extended Data Fig. 7c and Supplementary Videos 29–31). Comparing the timing of these junctional changes to migration onset revealed that cells began migrating in the 2D EMT conditions at approximately the same time that the ZO-1 network shrank (Fig. 6a, 22-h panel), whereas, in 3D lumenoid EMT, migration occurred after network remodeling (Fig. 6a, 27-h and 32-h panels). These results suggest a decoupling between the timing of cell delamination from the ZO-1 network and subsequent migration onset in 3D lumenoid EMT but not in the 2D EMT conditions. Together, the relationship between cell delamination and migration onset time and the difference in apparent amount of basement membrane present in 2D versus 3D hiPS cell culture geometries led us to consider the following model. In 2D EMT conditions, cells migrate away as soon as delamination occurs, consistent with the absence of a sufficiently developed basement membrane to confine their motility. In contrast, in 3D lumenoid EMT, cells may delaminate without the ability to immediately migrate away from the lumenoid, as they remain confined by a more well-developed basement membrane; migration away from the lumenoid occurs later, once cells are able to penetrate or otherwise overcome this basement membrane barrier (Fig. 6c). Imaging of the basement membrane in 3D lumenoid EMT revealed changes in its shape and consistency. Before cell migration, the basement membrane became wavy (Fig. 2e), and an upward detachment from the substrate at the center of the lumenoid footprint (‘dimpling’) occurred in a size-dependent manner (Extended Data Fig. 8 and Supplementary Videos 32–46). Concomitant with cell migration, discrete openings formed that resembled holes or perforations described in other morphogenetic and invasive contexts, including morphogenesis40, cancer cell invasion41 and mouse embryo gastrulation42 (Fig. 6d and Supplementary Video 47). Cells were frequently observed moving through these openings, including transient back-and-forth movements (Extended Data Fig. 9a–c and Supplementary Videos 48–50). These observations suggested that basement membrane integrity might constrain the timing of cell exit. We perturbed basement membrane integrity by either facilitating or impeding its breakdown. We facilitated basement membrane breakdown using type IV collagenase to digest collagen IV, a major component of the basement membrane (Methods). If the basement membrane delays migration by trapping delaminated cells, collagenase treatment would be expected to hasten migration onset. Consistent with this expectation, addition of collagenase produced a negative dose-dependent monotonic decrease in migration onset time (Extended Data Fig. 10b and Fig. 6e). Reduced anti-collagen IV labeling confirmed effective basement membrane digestion under collagenase treatment (Extended Data Fig. 10c). We also inhibited the digestion of the basement membrane by matrix metalloproteinases (MMPs), which digest the basement membrane and are regulated during EMT10,43, using the pan-MMP inhibitor ilomastat (Methods). This produced a positive dose-dependent monotonic increase in migration onset time in two of three biological replicates (Extended Data Fig. 10a, Fig. 6f, Supplementary Fig. 4 and Methods), suggesting that migration may be delayed when basement membrane digestion was impeded. Together, these results suggest basement membrane integrity affects migration onset timing and may in part explain the delayed migration observed in 3D lumenoid EMT, where a well-established basement membrane is present, in contrast to 2D EMT conditions, where there is relatively little basement membrane. Discussion EMT is a fundamental cell state transition in development and disease, yet it remains difficult to define consistently because of varied emphasis on cellular function, organization, molecular components and/or environment. Building on prior work demonstrating the utility of hiPS cells for integrating measurements of cell organization in 2D colonies44, we establish hiPS cells herein as a model system and experimental platform for quantitative analysis of EMT during differentiation across distinct observational modalities and across both 2D and 3D culture geometries. EMT induction in hiPS cells occurs reproducibly across distinct cell culture geometries and through multiple induction strategies, including downstream activation of the Wnt signaling pathway through GSK3β inhibition and upstream activation of BMP receptors using BMP4. hiPS cells are, thus, a tractable model system in which EMT can be robustly initiated under diverse conditions. The use of individual-derived or genetically perturbed hiPS cell lines would enable quantitative links between specific genetic alterations and changes in EMT dynamics, thereby expanding the platform for disease modeling. In this study, we focused on how the different technical constraints and biological implications of 2D and 3D cell culture geometries contributed to EMT dynamics. The 2D cell cultures are experimentally tractable and well suited for live imaging but constrain cells to an artificially flat geometry, whereas in vivo-like 3D systems are often difficult to manipulate and interpret because of cellular heterogeneity and complex environments45. The adherent 3D hiPS cell lumenoid bridges this gap by providing a physiologically relevant epithelial geometry while retaining experimental accessibility. Lumenoids exhibit uniform apicobasal polarity and a shared luminal environment, they are formed with dilute Matrigel rather than gel embedding. Furthermore, when combined with anti-collagen IV antibody, they enable live visualization of basement membrane dynamics. Comparing EMT across defined 2D and 3D hiPS cell culture geometries revealed differences consistent with both distinct initial cellular states and subsequent EMT trajectories. Differences in E-cadherin expression dynamics between 2D and 3D EMT conditions suggest variations in epithelial or hybrid character at EMT onset. In 2D EMT, transient increases in E-cadherin expression coincided with nascent lumenoid formation. Elevated cell death in the absence of dilute Matrigel further suggests that nascent lumenoid formation may stabilize an epithelial-like state and contribute to EMT robustness. Variability between 2D conditions was greatest on highly adhesive PLF substrates, where extreme flattening may disrupt nascent lumenoid closure and alter EMT timing. Together, these results show that controlled variation in cell culture geometry can be used to interrogate hybrid states and alternative EMT trajectories within a single human cell model. This framework could be leveraged to systematically relate differences in EMT dynamics observed across culture conditions, model systems and biological contexts. We characterized molecular changes during EMT in the hiPS cell model using antibody labeling and cell lines expressing fluorescently tagged EMT markers. Most markers exhibited expression patterns consistent with canonical EMT trajectories, supporting the suitability of this system for capturing core molecular features of EMT, although some deviated from classical expectations. These observations underscore that no single set of proteins universally defines EMT4 and that molecular EMT programs are varied across cell lines46. The reproducibility of EMT in hiPS cells enabled comparison of EMT trajectories across different cell culture geometries using a broad set of fluorescently tagged reporters. Across these measurements, changes in Brachyury, Eomes and E-cadherin, but not SOX-2, were associated with migration onset. However, this study does not establish causal or temporal ordering of molecular events during EMT, and additional processes, including cell death and mitosis, may influence migration timing and EMT progression. Such factors are known to regulate epithelial integrity and, in the case of mitosis, the timing of cell ingression47,48. The large, annotated and standardized dataset generated here provides a resource for future studies aimed at resolving these relationships. Our data indicate that differences in migration onset timing between 2D colony EMT and 3D lumenoid EMT are influenced by differences in basement membrane integrity acting as a physical barrier to cell migration. Accelerating basement membrane degradation with collagenase hastened migration onset, whereas inhibiting basement membrane degradation with an MMP inhibitor seemed to delay migration onset in 3D lumenoid EMT. Additional mechanisms of basement membrane remodeling may also contribute to how cells cross the basement membrane49 and, thus, influence migration timing. For example, the brightening of collagen IV labeling at the edges of basement membrane openings observed during 3D lumenoid EMT suggests local remodeling by mechanical displacement, analogous to the accumulation of fabric at the edges of a pulled curtain and similar to mechanisms reported in vivo50. Further analysis of basement membrane deposition and remodeling during 2D colony to 3D lumenoid formation, before EMT induction, may also provide insight into how basement membrane organization subsequently shapes EMT-associated migration during development51. This study underscores the value of integrating distinct observational perspectives of EMT to gain holistic insights. The hiPS cell model system offers a tractable experimental platform to quantitatively probe the dynamic changes in cell function, organization, molecular components and environment that accompany EMT, as well as discover relationships between these distinct types of observations. For example, the standardized imaging and analysis workflows might provide a testbed for developing and validating computational methods that could infer state or otherwise predict one type of dynamic change given another, applicable to EMT and other types of state change more broadly. Therefore, the openly accessible quantitative tools, analysis workflows and accompanying extensive imaging dataset constitute a valuable resource for studying EMT and other related complex cell state transitions. Methods Experimental methods Cell lines All work with hiPS cell lines was approved by internal oversight committees and performed in accordance with applicable National Institutes of Health, National Academy of Sciences and International Society for Stem Cell Research guidelines52. The parental WTC-11 hiPS cell line, derived from a healthy male donor53, was used as the background for generating all endogenously tagged reporter lines used in this study (https://www.allencell.org). Detailed information about plasmids, cell lines and genome editing is provided in Supplementary Note 1.1. Cell culture Cells were maintained in mTeSR1 (StemCell Technologies, 85850) supplemented with 1% penicillin–streptomycin (hiPS cell maintenance medium) on surfaces precoated with growth-factor-reduced Matrigel (Corning, 356231). hiPS cell maintenance medium was changed daily. A detailed protocol is available online (https://www.allencell.org/). Steps were performed either manually or by automated cell culture system (Hamilton STAR robot)54. 2D and 3D cell culture on glass-bottom imaging plates Cells were seeded onto 96-well glass-bottom imaging plates at 750–2,000 cells per cm2. Before seeding, glass surfaces were coated with either PLF or Matrigel. The 2D conditions received daily hiPS cell maintenance medium. To initiate 3D lumenoid formation, cells were cultured as 2D colonies for 2–4 days, after which medium was replaced with hiPS cell maintenance medium containing dilute Matrigel (1:60) (Supplementary Note 1.2). EMT induction was performed in 2D colonies and preformed 3D lumenoids at 4 days after seeding (Supplementary Notes 1.3 and 1.4). A schematic timeline of base EMT treatment conditions and controls is presented in Fig. 1a. EMT induction Unless otherwise noted, EMT was induced by switching cells to an EMT induction medium containing the small-molecule Wnt signaling activator CHIR99021, diluted in RPMI-1640 medium supplemented with a 1:50 dilution of B27 supplement devoid of insulin and dilute Matrigel (1:60). Using this Wnt-based EMT induction, three EMT conditions—2D PLF colony EMT, 2D colony EMT and 3D lumenoid EMT—were generated across three initial culture geometries and form the experimental base for the majority of this study (Supplementary Notes 1.2 and 1.3). In a subset of experiments, EMT was induced using BMP4 instead of Wnt activation (Supplementary Note 1.4). Detailed EMT induction medium compositions are provided in Supplementary Notes 1.3 and 1.4 and EMT induction media were supplemented as described below depending on the specific experiment. Visualizing Brachyury with HaloTag For experiments using the TBXT–HaloTag cell line (AICS-0143 cl.31), Janelia Fluor 646 HaloTag (JF646–HaloTag) ligand (Promega, HT106A) was added to EMT induction medium at a concentration of 3 nM. EMT perturbations A series of targeted perturbation experiments were performed, consisting of inhibition of BMP signaling with LDN-193189 (LDN), transcription factor knockdown by CRISPRi, inhibition of matrix metalloproteases (MMPs) with ilomastat and enzymatic digestion of basement membrane with type IV collagenase. Chemical and enzymatic perturbations were implemented by supplementing the EMT induction medium with perturbation-specific reagents, whereas CRISPRi perturbations were performed using dedicated knockdown cell lines; imaging parameters were adjusted as needed for specific perturbations. Detailed experimental procedures for each perturbation are provided in Supplementary Note 1.5. Visualization of basement membrane To visualize the basement membrane, eFluor 660-conjugated anti-collagen IV antibody (Thermo Fisher Scientific, 50-9871-82) was added at a dilution of 1:600 in hiPS cell maintenance medium or EMT induction medium depending on the experimental condition before imaging. The anti-collagen IV antibody was included in all experiments except those using only transmitted light (bright-field imaging): BMP inhibitor, MMP inhibitor and CRISPRi knockdown experiments or experiments using the TBXT–HaloTag line, where the far-red channel was used to visualize the JF646–HaloTag ligand. Fixation and labeling for immunofluorescence H2B–mEGFP cells across 2D colony EMT, 2D PLF colony EMT and 3D lumenoid EMT conditions were fixed in 4% paraformaldehyde at 4-h intervals over 48 h. They were then permeabilized with Triton X-100 overnight at 4 °C, blocked at room temperature for 30 min and immunolabeled using antibodies in blocking buffer as outlined in Supplementary Table 1. Primary and secondary antibodies were incubated sequentially overnight at 4 °C. Nidogen 1 and collagen IV labeling required an additional fixation step after their (primary antibody) incubation. Full fixation, labeling, washing and reagent details are provided in Supplementary Note 1.6. Image acquisition Imaging position selection A ×10 bright-field well overview was used to select 2D colonies and 3D lumenoids for imaging to ensure consistent colony sizes throughout the dataset. Following ×10 overview acquisition, plates were removed from the microscope and the medium was exchanged to EMT induction medium or hiPS cell maintenance medium supplemented depending on the specific experiment and condition. Plates were then placed back on the microscope and usually imaged within 1 h. Acquisition of standard 60-h, 3D time-lapse videos at 30-min intervals Unless otherwise noted, live-cell imaging throughout this study was performed using a standardized 3D time-lapse imaging protocol consisting of 60-h acquisitions collected at 30-min intervals. Imaging was performed on ZEISS Axio Observer Z1 spinning-disk confocal microscopes equipped with ×20 (numerical aperture (NA): 0.8) Plan-Apochromat objectives and environmental control to maintain cells at 37 °C and 5% CO2. Fluorescence and transmitted-light channels were acquired using fixed acquisition settings across experiments. The mEGFP and transmitted-light channels were acquired simultaneously using dual cameras, while the far-red channel (used to image fluorescent anti-collagen IV antibody or HaloTag ligand) was acquired sequentially on a single camera. Z stacks were collected with a nominal 2-µm spacing (actual: 2.88 µm), with each stack comprising 30 planes. Laser powers were held constant across experiments and verified at the start of each acquisition to ensure consistency. This standard time-lapse imaging protocol was used to acquire the majority of datasets in this study and underlies the representative images, plots and renderings shown in Figs. 1–3, 5 and 6 (excluding Figs. 5f–j and 6f) and Extended Data Figs. 3, 7, 8 and 10 (excluding Extended Data Fig. 7c). These datasets include time-lapse imaging of mEGFP-tagged and HaloTag-tagged cell lines (H2B–mEGFP, Brachyury–HaloTag, Eomes–mEGFP, SOX-2–mEGFP, E-cadherin–mEGFP, and mEGFP–ZO-1); additional time-lapse datasets are described below. In specific experiments, this standard protocol was adapted by modifying the combination of channels acquired or the Z-stack configuration to accommodate experimental constraints (for example, increased photosensitivity or channel-specific labeling). Full microscope configurations, acquisition parameters, and detailed descriptions of protocol adaptations are provided in Supplementary Note 2.1; optical controls are described in Supplementary Note 2.4. Acquisition of fixed immunofluorescence images Fixed samples were imaged using the same spinning-disk confocal microscope configuration and optical settings described for live-cell imaging (Supplementary Note 2.1), except that images were acquired as static Z stacks without time-lapse acquisition. Fixed immunofluorescence imaging included detection of mEGFP, Alexa Plus 555 and Alexa Plus 647. The additional bandpass filter used for Alexa Plus 555 detection was 600/50 nm (Chroma); filters for mEGFP and far-red fluorophores were as described in Supplementary Note 2.1. Laser power settings for antibody-based imaging were optimized for each antibody and are provided in Supplementary Table 1. Imaging of mEGFP-tagged histone H2B was performed using 488-nm excitation at 2.96 mW. Each fluorescence channel was acquired individually with an exposure time of 100 ms. Z stacks were collected using the same spacing and depth as the standard live-cell imaging configuration. Optical controls were performed as described in Supplementary Note 2.4. To minimize variability across fixation time points, samples labeled with the same antibody solution across plates fixed at different time points were imaged on the same microscope on the same day. Additional time-lapse imaging on other systems Higher-temporal-resolution (3-min interval) 3D time lapses of mEGFP-tagged ZO-1 (Extended Data Fig. 7c) and mEGFP-tagged H2B (Extended Data Fig. 9) were acquired on an Eclipse Ti2 Nikon spinning-disk confocal microscope. Lower-temporal-resolution (60-min interval) widefield imaging of BMP inhibition experiments (Extended Data Fig. 2a,b) were acquired on a Leica DMi8 inverted microscope. Details are provided in Supplementary Notes 2.2 and 2.3. Refractive index mismatch correction for 3D images Where ×20 (NA: 0.8) Plan-Apochromat objectives were used (Supplementary Notes 2.1 and 2.2), the effects of spherical aberration on Z-step size accuracy arising from the refractive index mismatch between the air objective and the culture medium were estimated using previously established methods55. This analysis indicated a correction factor of approximately 1.44. Accordingly, nominal Z-step sizes specified in the acquisition software were converted to physically accurate Z-step sizes, such that nominal 2-µm steps corresponded to an actual step size of 2.88 µm and nominal 0.5-µm steps corresponded to an actual step size of 0.72 µm. Following this adjustment, fluorescent microspheres imaged as part of the optical controls appeared spherical rather than axially compressed. Quality control and dataset curation Time-lapse videos were manually annotated by trained microscopists using strict, predefined quality-control criteria, including cell health, presence of EMT-associated migration and absence of confounding events within the field of view at migration onset. Only time lapses passing all quality-control criteria were included in downstream analyses. Full annotation procedures and dataset curation criteria are described in Supplementary Notes 3.1 and 3.2. Image analysis methods Camera alignment matrix generation Two-camera alignment was performed using fluorescent field-of-rings optical control images acquired simultaneously on both cameras (Supplementary Note 2.4). Ring positions in each channel were segmented and matched by linear sum assignment optimization and a similarity transformation matrix (without scaling) was computed. This transformation was applied for all analyses requiring spatial correspondence between channels acquired on different cameras. It was applied to map images and associated coordinate data into a common camera reference frame using nearest-neighbor interpolation (details in Supplementary Note 4.1). All-cells mask generation from bright-field The 3D all-cells masks were generated from the bright-field channel of Z stacks acquired using the standard imaging settings to distinguish cell-containing regions of the stack from background. Ground-truth masks were generated from cytoplasmic and nuclear mEGFP fluorescence (AAVS1 cytoplasmic and nuclear mEGFP cell line) using Otsu thresholding. A convolutional neural network56 was then trained to predict these masks directly from bright-field images (Extended Data Fig. 3 and Supplementary Videos 51–52). To improve segmentation accuracy, multiscale patch-based inference was used with predictions from three patch scales merged by logical disjunction. Because all-cells mask prediction relied exclusively on bright-field input, this approach was generalizable across cell lines and fluorescent labels, enabling robust mask generation for all standard live time-lapse datasets and all fixed immunofluorescence image stacks analyzed in this study (detailed parameters in Supplementary Note 4.2). Denoising method Denoised images were generated using the Cellpose3 denoising model57 to improve downstream segmentation and visualization where required. Images of mEGFP-tagged H2B were denoised before nuclear segmentation and mEGFP-tagged ZO-1 images were denoised before apical face annotation (Extended Data Fig. 7c; technical details in Supplementary Note 4.3). 3D nuclear instance segmentation and centroid identification A 3D nuclear instance segmentation pipeline was developed for standard time-lapse images of EMT in the mEGFP-tagged H2B cell line. Raw images were rescaled and denoised before segmentation with a fine-tuned Cellpose ‘cyto3’ model57. Segmentation masks were refined using adaptive 3D clipping and hole-filling operations. Nuclear centroids were extracted from these masks and used to represent nuclear position in downstream analyses (detailed parameters and validation in Supplementary Note 4.4). Basement membrane segmentation and 3D mesh generation To identify the basement membrane surface, collagen IV was segmented using a dynamic UNet model56 trained on manually curated ground-truth masks. Postprocessed segmentations were converted to 3D meshes by generating a point cloud, creating an initial inverted-dome mesh and registering it to the point cloud contours using nonrigid iterative closest points58. Final meshes were converted to a watertight format59 for downstream analysis to determine the inside–outside classification (Extended Data Fig. 3h) of nuclear centroids with respect to the location of the basement membrane shell (detailed parameters and validation in Supplementary Note 4.5). Data analysis methods Determination of migration onset time based on the change in area of cells at the glass (footprint-area-based semiautomated detection method) Migration onset time was defined by the inflection point before the rapid rise in footprint area of cells over time. The footprint area was calculated from the bottom two Z slices of the all-cells mask (Fig. 3a) and smoothed using a Savitzky–Golay filter; the inflection point was manually identified by experts (Fig. 3c). Interexpert validation confirmed high reproducibility (mean difference: 0.5 h, n = 21; details in Supplementary Note 5.2). Determination of migration onset timing using visual inspection of images (manual) For MMP and BMP inhibition experiments, migration onset timing could not be measured using the footprint-area-based method because the images of these were not the standard Z stacks needed to generate all-cells masks (exceptions in Supplementary Note 2.4). Migration onset was therefore manually scored for these experiments (Extended Data Fig. 2c–e, Fig. 6f and Supplementary Fig. 4). Experts identified the ‘first cell time’ when the first cell became isolated from a colony or lumenoid. Interexpert reproducibility was validated with average differences of 1.78–2.25 h (details in Supplementary Note 5.3). Determination of migration onset time based on nuclei location relative to basement membrane (inside–outside basement membrane semiautomated detection method) Nuclear centroids were classified as inside or outside the lumenoid on the basis of their position relative to 3D basement membrane meshes after alignment of the corresponding image channels using a camera alignment transformation. Migration onset time was identified by experts as the inflection point before the rise in the fraction of nuclei outside the lumenoid over time (Fig. 3g; details in Supplementary Note 5.4). Assessment of agreement between methods of measuring migration timing Agreement between the footprint-area-based and inside–outside basement-membrane-based methods was assessed using Bland–Altman analysis, including estimation of mean bias, limits of agreement and assessment of proportional bias by linear regression of measurement differences against measurement means. Full statistical details and diagnostic tests are provided in Supplementary Note 5.5 and Supplementary Table 3. Calculation of mean fluorescence intensity The mean fluorescence intensity for a given image stack was calculated as the ratio of the summed fluorescence intensity within the all-cells mask across the bottom ten Z slices to the summed all-cells mask area across those same Z slices. Before this calculation, an alignment correction was performed if the bright-field channel (from which the all-cells mask was derived) and the fluorescence channel to be quantified were not acquired on the same camera. Mean intensities were calculated from fixed-cell images acquired at discrete time points. Fixed-cell measurements were analyzed as described below. In live cells, continuous intensity trajectories were obtained from time-lapse imaging. Calculated mean intensity trajectories were smoothed using Savitzky–Golay filtering and analyzed using marker-specific approaches to identify the time of half-maximum decay for SOX-2, time of peak intensity for Eomes and Brachyury and the inflection point of decay for E-cadherin (Supplementary Note 5.6). Comparison of expression timing of canonical EMT marker immunolabeling Mean fluorescence intensity was calculated across Z-stack images of labeled canonical EMT marker proteins. Intensity values were normalized to the initial (t = 0) intensity for each antibody and condition, pooled across replicate experiments and min–max-normalized (0–100) to focus on relative timing rather than intensity magnitude. Normalized intensities were visualized as heat maps (Fig. 4c; details in Supplementary Note 5.8). Characterization of apical dynamics using ZO-1 To quantify the axial position of the ZO-1-defined apical junctional network over time, the mean ZO-1 fluorescence intensity within the corresponding all-cells mask was measured at each Z slice and time point in standard time-lapse images after alignment of the corresponding image channels using a camera alignment transformation and visualized as heat maps (Fig. 6b and Extended Data Fig. 7b; details in Supplementary Note 5.9). To quantify apical face dynamics and closure behavior, higher-temporal-resolution ZO-1 time-lapse images were analyzed using maximum Z projections of denoised image stacks, with apical faces manually outlined (Extended Data Fig. 7d). Detailed analysis parameters are provided in Supplementary Note 5.9. Statistical analysis and reporting All measurements are based on independent time-lapse videos or static images, except for droplet digital PCR (ddPCR) assessment of CRISPRi knockdown, which was performed on individual cell lysates. Replicate types and exact sample sizes (n) for each experimental group are reported in Supplementary Table 4; unless otherwise indicated, experiments included at least three biological replicates. For live-cell time-lapse experiments, repeated measurements were obtained from the same field of view over time. Where indicated in the figure legends, the same time-lapse videos were used for multiple analyses to enable direct comparison across measurement types. Statistical tests used for group and paired comparisons are described in the Supplementary Notes. Parametric and nonparametric tests were selected on the basis of data characteristics and are reported by name. Measures of central tendency and variability are indicated in the figures and legends (for example, mean or median together with interquartile range). Additional estimates of uncertainty, including s.d. and confidence intervals, are reported in the Supplementary Notes and Supplementary Tables where applicable. For null-hypothesis testing, test statistics, effect sizes and exact P values are reported. Where Spearman and Pearson correlation analyses were performed, two-sided correlation tests were used; correlation coefficients and 95% confidence intervals were estimated using a nonparametric bootstrap approach (2,000 resamples with replacement) and are reported in Supplementary Table 3. Detailed statistical results are provided in Supplementary Table 3 and relevant Supplementary Notes. Visualization Vol-E (Volume Explorer) All OME-Zarr datasets associated with this paper can be viewed without downloading using the web-based volume viewer Vol-E (https://vol-e.allencell.org/). Vol-E enables interactive 3D time-series visualization with on-the-fly rendering of raw microscopy images overlayed with their corresponding segmentations (details in Supplementary Note 6.1). Vol-E uses multiresolution rendering optimized for web-based performance; users who require access to the full native spatial resolution of the datasets may instead download and view it locally using a compatible OME-Zarr viewer such as AGAVE. AGAVE rendering The 3D volumetric renderings were generated using AGAVE. AGAVE uses a physically based lighting and rendering algorithm to produce high quality views of volumetric data (https://www.allencell.org/pathtrace-rendering.html and https://github.com/allen-cell-animated/agave). BioFileFinder (BFF) The BFF60 is an open-source web application for searching, organizing and sharing biological image datasets using structured metadata. BFF enables interactive filtering and browsing of image collections defined through simple spreadsheet-based annotations, with visualization and downstream analysis available through connected web-based and desktop tools (https://bff.allencell.org)60. Links to data released with this manuscript in BFF are provided. Reporting summary Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article. Online content Any methods, additional references, Nature Portfolio reporting summaries, source data, extended data, supplementary information, acknowledgements, peer review information; details of author contributions and competing interests; and statements of data and code availability are available at 10.1038/s41592-026-03096-9. Supplementary information Supplementary InformationTable of contents, Supplementary figure legends, notes and tables. Reporting Summary Supplementary Table 4This table reports the number of samples plotted, together with the breakdown of associated biological and technical replicates. The file is organized into three sheets: the ‘readme’ sheet, which describes the structure of the document and the definitions used, the ‘live imaging data’ sheet, which reports the number of replicates for each unique combination of cell line and experimental condition, and the ‘fixed imaging data’ sheet, which reports, for each antibody analyzed in the manuscript, the number of replicates per experimental condition and time point. Supplementary Video 1Video corresponding to the image in Fig. 1b showing YZ views of H2B hiPS cell 2D colony growth. Frames were acquired every 30 min for a duration of 48 h, displayed at ten frames per second. Timestamps are displayed as hh:mm. Supplementary Video 2Video corresponding to the image in Fig. 1b showing XY views of H2B hiPS cell 2D colony growth. Frames were acquired every 30 min for a duration of 48 h, displayed at ten frames per second. Timestamps are displayed as hh:mm. Supplementary Video 3Video corresponding to the image in Fig. 1c showing YZ views of H2B hiPS cell 2D to 3D lumenoid formation. Frames were acquired every 30 min for a duration of 48 h, displayed at ten frames per second. Timestamps are displayed as hh:mm. Supplementary Video 4Video corresponding to the image in Fig. 1c showing XY views H2B hiPS cell 2D to 3D lumenoid formation. Frames were acquired every 30 min for a duration of 48 h, displayed at ten frames per second. Timestamps are displayed as hh:mm. Supplementary Video 5Video corresponding to the image in Fig. 2a showing H2B 2D colony EMT no Matrigel. Frames were acquired every 30 min for a duration of 48 h, displayed at ten frames per second. Timestamps are displayed as hh:mm. Supplementary Video 6Video corresponding to the image in Fig. 2b showing H2B 2D colony EMT. Frames were acquired every 30 min for a duration of 48 h, displayed at ten frames per second. Timestamps are displayed as hh:mm. Supplementary Video 7Video corresponding to the image in Fig. 2c showing H2B 2D PLF colony EMT. Frames were acquired every 30 min for a duration of 48 h, displayed at ten frames per second. Timestamps are displayed as hh:mm. Supplementary Video 8Video corresponding to the image in Fig. 2d showing 3D lumenoid EMT. Frames were acquired every 30 min for a duration of 48 h, displayed at ten frames per second. Timestamps are displayed as hh:mm. Supplementary Video 9Video corresponding to the image in Fig. 2e showing an AGAVE rendering of H2B 3D lumenoid EMT. Frames were acquired every 30 min for a duration of 48 h, displayed at ten frames per second. Timestamps are displayed as hh:mm. Supplementary Video 10Video corresponding to the image in Fig. 3e showing a representative 3D rendering of nuclear centroids inside (yellow) and outside (cyan) the basement membrane mesh (gray). Frames shown every 30 min for a duration of 45 h, displayed at ten frames per second. Timestamps are displayed as hh:mm. Supplementary Video 11Video corresponding to the image in Fig. 5a showing SOX-2 2D PLF colony EMT. Frames were acquired every 30 min for a duration of 48 h, displayed at ten frames per second. Timestamps are displayed as hh:mm. Supplementary Video 12Video corresponding to the image in Fig. 5a showing SOX-2 2D colony EMT. Frames were acquired every 30 min for a duration of 48 h, displayed at ten frames per second. Timestamps are displayed as hh:mm. Supplementary Video 13Video corresponding to the image in Fig. 5a showing SOX-2 3D lumenoid EMT. Frames were acquired every 30 min for a duration of 48 h, displayed at ten frames per second. Timestamps are displayed as hh:mm. Supplementary Video 14Video corresponding to the image in Fig. 5f showing Brachyury 2D PLF colony EMT. Frames were acquired every 30 min for a duration of 48 h, displayed at ten frames per second. Timestamps are displayed as hh:mm. Supplementary Video 15Video corresponding to the image in Fig. 5f showing Brachyury 2D colony EMT. Frames were acquired every 30 min for a duration of 48 h, displayed at ten frames per second. Timestamps are displayed as hh:mm. Supplementary Video 16Video corresponding to the image in Fig. 5f showing Brachyury 3D lumenoid EMT. Frames were acquired every 30 min for a duration of 48 h, displayed at ten frames per second. Timestamps are displayed as hh:mm. Supplementary Video 17Video corresponding to the image in Fig. 5k showing Eomes 2D PLF colony EMT. Frames were acquired every 30 min for a duration of 48 h, displayed at ten frames per second. Timestamps are displayed as hh:mm. Supplementary Video 18Video corresponding to the image in Fig. 5k showing Eomes 2D colony EMT. Frames were acquired every 30 min for a duration of 48 h, displayed at ten frames per second. Timestamps are displayed as hh:mm. Supplementary Video 19Video corresponding to the image in Fig. 5k showing Eomes 3D lumenoid EMT. Frames were acquired every 30 min for a duration of 48 h, displayed at ten frames per second. Timestamps are displayed as hh:mm. Supplementary Video 20Video corresponding to the image in Fig. 5p showing E-cadherin 2D PLF colony EMT. Frames were acquired every 30 min for a duration of 48 h, displayed at ten frames per second. Timestamps are displayed as hh:mm. Supplementary Video 21Video corresponding to the image in Fig. 5p showing E-cadherin 2D colony EMT. Frames were acquired every 30 min for a duration of 48 h, displayed at ten frames per second. Timestamps are displayed as hh:mm. Supplementary Video 22Video corresponding to the image in Fig. 5p showing E-cadherin 3D lumenoid EMT. Frames were acquired every 30 min for a duration of 48 h, displayed at ten frames per second. Timestamps are displayed as hh:mm. Supplementary Video 23Video corresponding to the image in Fig. 6a showing ZO-1 2D PLF colony EMT. Frames were acquired every 30 min for a duration of 48 h, displayed at ten frames per second. Timestamps are displayed as hh:mm. Supplementary Video 24Video corresponding to the image in Fig. 6a showing ZO-1 2D colony EMT. Frames were acquired every 30 min for a duration of 48 h, displayed at ten frames per second. Timestamps are displayed as hh:mm. Supplementary Video 25Video corresponding to the image in Fig. 6a showing ZO-1 3D lumenoid EMT. Frames were acquired every 30 min for a duration of 48 h, displayed at ten frames per second. Timestamps are displayed as hh:mm. Supplementary Video 26Video corresponding to the image in Extended Data Fig. 7a showing ZO-1 2D PLF colony EMT. Frames were acquired every 30 min for a duration of 48 h, displayed at ten frames per second. Timestamps are displayed as hh:mm. Supplementary Video 27Video corresponding to the image in Extended Data Fig. 7a showing ZO-1 2D colony EMT. Frames were acquired every 30 min for a duration of 48 h, displayed at ten frames per second. Timestamps are displayed as hh:mm. Supplementary Video 28Video corresponding to the image in Extended Data Fig. 7a showing ZO-1 3D lumenoid EMT. Frames were acquired every 30 min for a duration of 48 h, displayed at ten frames per second. Timestamps are displayed as hh:mm. Supplementary Video 29Video corresponding to the image in Extended Data Fig. 7c showing ZO-1 2D PLF colony EMT. The video starts at 20:48 (hh:mm) after EMT induction. Frames were acquired every 3 min and are displayed at three frames per second. Timestamps are displayed as hh:mm. Supplementary Video 30Video corresponding to the image in Extended Data Fig. 7c showing ZO-1 in 2D colony EMT. The video starts at 20:18 (hh:mm) after EMT induction. Frames were acquired every 3 min and are displayed at three frames per second. Timestamps are displayed as hh:mm. Supplementary Video 31Video corresponding to the image in Extended Data Fig. 7c showing ZO-1 3D lumenoid EMT. The video starts at 21:00 (hh:mm) after EMT induction. Frames were acquired every 3 min and are displayed at three frames per second. Timestamps are displayed as hh:mm. Supplementary Video 32Video corresponding to the image in Extended Data Fig. 8a showing an XY view of fluorescent collagen IV antibody during 2D PLF colony EMT. Frames were acquired every 30 min for a duration of 48 h, displayed at ten frames per second. Timestamps are displayed as hh:mm. Supplementary Video 33Video corresponding to the image in Extended Data Fig. 8a showing a YZ view of fluorescent collagen IV antibody during 2D PLF colony EMT. Frames were acquired every 30 min for a duration of 48 h, displayed at ten frames per second. Timestamps are displayed as hh:mm. Supplementary Video 34Video corresponding to the image in Extended Data Fig. 8b showing an XY view of fluorescent collagen IV antibody during 2D colony EMT. Frames were acquired every 30 min for a duration of 48 h, displayed at ten frames per second. Timestamps are displayed as hh:mm. Supplementary Video 35Video corresponding to the image in Extended Data Fig. 8b showing a YZ view of fluorescent collagen IV antibody during 2D colony EMT. Frames were acquired every 30 min for a duration of 48 h, displayed at ten frames per second. Timestamps are displayed as hh:mm. Supplementary Video 36Video corresponding to the image in Extended Data Fig. 8c showing an XY view of fluorescent collagen IV antibody during 3D lumenoid EMT (small lumenoid). Frames were acquired every 30 min for a duration of 48 h, displayed at ten frames per second. Timestamps are displayed as hh:mm. Supplementary Video 37Video corresponding to the image in Extended Data Fig. 8c showing YZ view of fluorescent collagen IV antibody during 3D lumenoid EMT (small lumenoid). Frames were acquired every 30 min for a duration of 48 h, displayed at ten frames per second. Timestamps are displayed as hh:mm. Supplementary Video 38Video corresponding to the image in Extended Data Fig. 8d showing an XY view of fluorescent collagen IV antibody during 3D lumenoid EMT (small lumenoid). Frames were acquired every 30 min for a duration of 48 h, displayed at ten frames per second. Timestamps are displayed as hh:mm. Supplementary Video 39Video corresponding to the image in Extended Data Fig. 8d showing YZ view of fluorescent collagen IV antibody during 3D lumenoid EMT (small lumenoid). Frames were acquired every 30 min for a duration of 48 h, displayed at ten frames per second. Timestamps are displayed as hh:mm. Supplementary Video 40Video corresponding to the image in Extended Data Fig. 8e top row showing YZ view of E-cadherin and collagen IV antibody in 3D lumenoid EMT. Frames were acquired every 30 min for a duration of 48 h, displayed at ten frames per second. Timestamps are displayed as hh:mm. Supplementary Video 41Video corresponding to the image in Extended Data Fig. 8e second row showing YZ view of E-cadherin and collagen IV antibody in 3D lumenoid EMT. Frames were acquired every 30 min for a duration of 48 h, displayed at ten frames per second. Timestamps are displayed as hh:mm. Supplementary Video 42Video corresponding to the image in Extended Data Fig. 8e third row showing YZ view of E-cadherin and collagen IV antibody in 3D lumenoid EMT. Frames were acquired every 30 min for a duration of 48 h, displayed at ten frames per second. Timestamps are displayed as hh:mm. Supplementary Video 43Video corresponding to the image in Extended Data Fig. 8e forth row showing YZ view of E-cadherin 3D and collagen IV antibody in 3D lumenoid EMT. Frames were acquired every 30 min for a duration of 48 h, displayed at ten frames per second. Timestamps are displayed as hh:mm. Supplementary Video 44Video corresponding to the image in Extended Data Fig. 8e fifth row showing YZ view of E-cadherin 3D and collagen IV antibody in 3D lumenoid EMT. Frames were acquired every 30 min for a duration of 48 h, displayed at ten frames per second. Timestamps are displayed as hh:mm. Supplementary Video 45Video corresponding to the image in Extended Data Fig. 8e sixth row showing YZ view of E-cadherin and collagen IV antibody in 3D lumenoid EMT. Frames were acquired every 30 min for a duration of 48 h, displayed at ten frames per second. Timestamps are displayed as hh:mm. Supplementary Video 46Video corresponding to the image in Extended Data Fig. 8e seventh row showing YZ view of E-cadherin and collagen IV antibody in 3D lumenoid EMT. Frames were acquired every 30 min for a duration of 48 h, displayed at ten frames per second. Timestamps are displayed as hh:mm. Supplementary Video 47Video corresponding to the image in Fig. 6d showing the collagen IV shell of H2B 3D lumenoid EMT. Frames were acquired every 30 min for a duration of 48 h, displayed at ten frames per second. Timestamps are displayed as hh:mm. Supplementary Video 48Video corresponding to the image in Extended Data Fig. 9a showing an example of a nucleus passing through a small basement membrane opening accompanied by an apparent change in nuclear shape in 3D lumenoid EMT. The video starts 29:57 (hh:mm) after EMT induction. Frames were acquired every 3 min and are displayed at two frames per second. Timestamps are displayed as hh:mm. Supplementary Video 49Video corresponding to the image in Extended Data Fig. 9b showing an example of a nucleus exiting through a larger basement membrane opening with minimal apparent change in nuclear shape in 3D lumenoid EMT. The video starts 33:33 (hh:mm) after EMT induction. Frames were acquired every 3 min and are displayed at three frames per second. Timestamps are displayed as hh:mm. Supplementary Video 50Video corresponding to the image in Extended Data Fig. 9c showing a nucleus that crosses the basement membrane in both directions over time in 3D lumenoid EMT. The video starts 29:48 (hh:mm) after EMT induction. Frames were acquired every 3 min and are displayed at two frames per second. Timestamps are displayed as hh:mm. Supplementary Video 51Video corresponding to the image in Extended Data Fig. 3d showing a representative overlay of bright-field images (gray) and corresponding all-cells masks (green) shown at a single Z slice for 2D colony EMT from the H2B–mEGFP cell line. Frames were acquired every 30 min for a duration of 48 h, displayed at ten frames per second. Timestamps are displayed as hh:mm. Supplementary Video 52Video corresponding to the image in Extended Data Fig. 3d showing a representative overlay of bright-field images (gray) and corresponding all-cells masks (green) shown at a single Z slice for 3D lumenoid EMT from the H2B–mEGFP cell line. Frames were acquired every 30 min for a duration of 48 h, displayed at ten frames per second. Timestamps are displayed as hh:mm. Table of contents, Supplementary figure legends, notes and tables. Reporting Summary This table reports the number of samples plotted, together with the breakdown of associated biological and technical replicates. The file is organized into three sheets: the ‘readme’ sheet, which describes the structure of the document and the definitions used, the ‘live imaging data’ sheet, which reports the number of replicates for each unique combination of cell line and experimental condition, and the ‘fixed imaging data’ sheet, which reports, for each antibody analyzed in the manuscript, the number of replicates per experimental condition and time point. Video corresponding to the image in Fig. 1b showing YZ views of H2B hiPS cell 2D colony growth. Frames were acquired every 30 min for a duration of 48 h, displayed at ten frames per second. Timestamps are displayed as hh:mm. Video corresponding to the image in Fig. 1b showing XY views of H2B hiPS cell 2D colony growth. Frames were acquired every 30 min for a duration of 48 h, displayed at ten frames per second. Timestamps are displayed as hh:mm. Video corresponding to the image in Fig. 1c showing YZ views of H2B hiPS cell 2D to 3D lumenoid formation. Frames were acquired every 30 min for a duration of 48 h, displayed at ten frames per second. Timestamps are displayed as hh:mm. Video corresponding to the image in Fig. 1c showing XY views H2B hiPS cell 2D to 3D lumenoid formation. Frames were acquired every 30 min for a duration of 48 h, displayed at ten frames per second. Timestamps are displayed as hh:mm. Video corresponding to the image in Fig. 2a showing H2B 2D colony EMT no Matrigel. Frames were acquired every 30 min for a duration of 48 h, displayed at ten frames per second. Timestamps are displayed as hh:mm. Video corresponding to the image in Fig. 2b showing H2B 2D colony EMT. Frames were acquired every 30 min for a duration of 48 h, displayed at ten frames per second. Timestamps are displayed as hh:mm. Video corresponding to the image in Fig. 2c showing H2B 2D PLF colony EMT. Frames were acquired every 30 min for a duration of 48 h, displayed at ten frames per second. Timestamps are displayed as hh:mm. Video corresponding to the image in Fig. 2d showing 3D lumenoid EMT. Frames were acquired every 30 min for a duration of 48 h, displayed at ten frames per second. Timestamps are displayed as hh:mm. Video corresponding to the image in Fig. 2e showing an AGAVE rendering of H2B 3D lumenoid EMT. Frames were acquired every 30 min for a duration of 48 h, displayed at ten frames per second. Timestamps are displayed as hh:mm. Video corresponding to the image in Fig. 3e showing a representative 3D rendering of nuclear centroids inside (yellow) and outside (cyan) the basement membrane mesh (gray). Frames shown every 30 min for a duration of 45 h, displayed at ten frames per second. Timestamps are displayed as hh:mm. Video corresponding to the image in Fig. 5a showing SOX-2 2D PLF colony EMT. Frames were acquired every 30 min for a duration of 48 h, displayed at ten frames per second. Timestamps are displayed as hh:mm. Video corresponding to the image in Fig. 5a showing SOX-2 2D colony EMT. Frames were acquired every 30 min for a duration of 48 h, displayed at ten frames per second. Timestamps are displayed as hh:mm. Video corresponding to the image in Fig. 5a showing SOX-2 3D lumenoid EMT. Frames were acquired every 30 min for a duration of 48 h, displayed at ten frames per second. Timestamps are displayed as hh:mm. Video corresponding to the image in Fig. 5f showing Brachyury 2D PLF colony EMT. Frames were acquired every 30 min for a duration of 48 h, displayed at ten frames per second. Timestamps are displayed as hh:mm. Video corresponding to the image in Fig. 5f showing Brachyury 2D colony EMT. Frames were acquired every 30 min for a duration of 48 h, displayed at ten frames per second. Timestamps are displayed as hh:mm. Video corresponding to the image in Fig. 5f showing Brachyury 3D lumenoid EMT. Frames were acquired every 30 min for a duration of 48 h, displayed at ten frames per second. Timestamps are displayed as hh:mm. Video corresponding to the image in Fig. 5k showing Eomes 2D PLF colony EMT. Frames were acquired every 30 min for a duration of 48 h, displayed at ten frames per second. Timestamps are displayed as hh:mm. Video corresponding to the image in Fig. 5k showing Eomes 2D colony EMT. Frames were acquired every 30 min for a duration of 48 h, displayed at ten frames per second. Timestamps are displayed as hh:mm. Video corresponding to the image in Fig. 5k showing Eomes 3D lumenoid EMT. Frames were acquired every 30 min for a duration of 48 h, displayed at ten frames per second. Timestamps are displayed as hh:mm. Video corresponding to the image in Fig. 5p showing E-cadherin 2D PLF colony EMT. Frames were acquired every 30 min for a duration of 48 h, displayed at ten frames per second. Timestamps are displayed as hh:mm. Video corresponding to the image in Fig. 5p showing E-cadherin 2D colony EMT. Frames were acquired every 30 min for a duration of 48 h, displayed at ten frames per second. Timestamps are displayed as hh:mm. Video corresponding to the image in Fig. 5p showing E-cadherin 3D lumenoid EMT. Frames were acquired every 30 min for a duration of 48 h, displayed at ten frames per second. Timestamps are displayed as hh:mm. Video corresponding to the image in Fig. 6a showing ZO-1 2D PLF colony EMT. Frames were acquired every 30 min for a duration of 48 h, displayed at ten frames per second. Timestamps are displayed as hh:mm. Video corresponding to the image in Fig. 6a showing ZO-1 2D colony EMT. Frames were acquired every 30 min for a duration of 48 h, displayed at ten frames per second. Timestamps are displayed as hh:mm. Video corresponding to the image in Fig. 6a showing ZO-1 3D lumenoid EMT. Frames were acquired every 30 min for a duration of 48 h, displayed at ten frames per second. Timestamps are displayed as hh:mm. Video corresponding to the image in Extended Data Fig. 7a showing ZO-1 2D PLF colony EMT. Frames were acquired every 30 min for a duration of 48 h, displayed at ten frames per second. Timestamps are displayed as hh:mm. Video corresponding to the image in Extended Data Fig. 7a showing ZO-1 2D colony EMT. Frames were acquired every 30 min for a duration of 48 h, displayed at ten frames per second. Timestamps are displayed as hh:mm. Video corresponding to the image in Extended Data Fig. 7a showing ZO-1 3D lumenoid EMT. Frames were acquired every 30 min for a duration of 48 h, displayed at ten frames per second. Timestamps are displayed as hh:mm. Video corresponding to the image in Extended Data Fig. 7c showing ZO-1 2D PLF colony EMT. The video starts at 20:48 (hh:mm) after EMT induction. Frames were acquired every 3 min and are displayed at three frames per second. Timestamps are displayed as hh:mm. Video corresponding to the image in Extended Data Fig. 7c showing ZO-1 in 2D colony EMT. The video starts at 20:18 (hh:mm) after EMT induction. Frames were acquired every 3 min and are displayed at three frames per second. Timestamps are displayed as hh:mm. Video corresponding to the image in Extended Data Fig. 7c showing ZO-1 3D lumenoid EMT. The video starts at 21:00 (hh:mm) after EMT induction. Frames were acquired every 3 min and are displayed at three frames per second. Timestamps are displayed as hh:mm. Video corresponding to the image in Extended Data Fig. 8a showing an XY view of fluorescent collagen IV antibody during 2D PLF colony EMT. Frames were acquired every 30 min for a duration of 48 h, displayed at ten frames per second. Timestamps are displayed as hh:mm. Video corresponding to the image in Extended Data Fig. 8a showing a YZ view of fluorescent collagen IV antibody during 2D PLF colony EMT. Frames were acquired every 30 min for a duration of 48 h, displayed at ten frames per second. Timestamps are displayed as hh:mm. Video corresponding to the image in Extended Data Fig. 8b showing an XY view of fluorescent collagen IV antibody during 2D colony EMT. Frames were acquired every 30 min for a duration of 48 h, displayed at ten frames per second. Timestamps are displayed as hh:mm. Video corresponding to the image in Extended Data Fig. 8b showing a YZ view of fluorescent collagen IV antibody during 2D colony EMT. Frames were acquired every 30 min for a duration of 48 h, displayed at ten frames per second. Timestamps are displayed as hh:mm. Video corresponding to the image in Extended Data Fig. 8c showing an XY view of fluorescent collagen IV antibody during 3D lumenoid EMT (small lumenoid). Frames were acquired every 30 min for a duration of 48 h, displayed at ten frames per second. Timestamps are displayed as hh:mm. Video corresponding to the image in Extended Data Fig. 8c showing YZ view of fluorescent collagen IV antibody during 3D lumenoid EMT (small lumenoid). Frames were acquired every 30 min for a duration of 48 h, displayed at ten frames per second. Timestamps are displayed as hh:mm. Video corresponding to the image in Extended Data Fig. 8d showing an XY view of fluorescent collagen IV antibody during 3D lumenoid EMT (small lumenoid). Frames were acquired every 30 min for a duration of 48 h, displayed at ten frames per second. Timestamps are displayed as hh:mm. Video corresponding to the image in Extended Data Fig. 8d showing YZ view of fluorescent collagen IV antibody during 3D lumenoid EMT (small lumenoid). Frames were acquired every 30 min for a duration of 48 h, displayed at ten frames per second. Timestamps are displayed as hh:mm. Video corresponding to the image in Extended Data Fig. 8e top row showing YZ view of E-cadherin and collagen IV antibody in 3D lumenoid EMT. Frames were acquired every 30 min for a duration of 48 h, displayed at ten frames per second. Timestamps are displayed as hh:mm. Video corresponding to the image in Extended Data Fig. 8e second row showing YZ view of E-cadherin and collagen IV antibody in 3D lumenoid EMT. Frames were acquired every 30 min for a duration of 48 h, displayed at ten frames per second. Timestamps are displayed as hh:mm. Video corresponding to the image in Extended Data Fig. 8e third row showing YZ view of E-cadherin and collagen IV antibody in 3D lumenoid EMT. Frames were acquired every 30 min for a duration of 48 h, displayed at ten frames per second. Timestamps are displayed as hh:mm. Video corresponding to the image in Extended Data Fig. 8e forth row showing YZ view of E-cadherin 3D and collagen IV antibody in 3D lumenoid EMT. Frames were acquired every 30 min for a duration of 48 h, displayed at ten frames per second. Timestamps are displayed as hh:mm. Video corresponding to the image in Extended Data Fig. 8e fifth row showing YZ view of E-cadherin 3D and collagen IV antibody in 3D lumenoid EMT. Frames were acquired every 30 min for a duration of 48 h, displayed at ten frames per second. Timestamps are displayed as hh:mm. Video corresponding to the image in Extended Data Fig. 8e sixth row showing YZ view of E-cadherin and collagen IV antibody in 3D lumenoid EMT. Frames were acquired every 30 min for a duration of 48 h, displayed at ten frames per second. Timestamps are displayed as hh:mm. Video corresponding to the image in Extended Data Fig. 8e seventh row showing YZ view of E-cadherin and collagen IV antibody in 3D lumenoid EMT. Frames were acquired every 30 min for a duration of 48 h, displayed at ten frames per second. Timestamps are displayed as hh:mm. Video corresponding to the image in Fig. 6d showing the collagen IV shell of H2B 3D lumenoid EMT. Frames were acquired every 30 min for a duration of 48 h, displayed at ten frames per second. Timestamps are displayed as hh:mm. Video corresponding to the image in Extended Data Fig. 9a showing an example of a nucleus passing through a small basement membrane opening accompanied by an apparent change in nuclear shape in 3D lumenoid EMT. The video starts 29:57 (hh:mm) after EMT induction. Frames were acquired every 3 min and are displayed at two frames per second. Timestamps are displayed as hh:mm. Video corresponding to the image in Extended Data Fig. 9b showing an example of a nucleus exiting through a larger basement membrane opening with minimal apparent change in nuclear shape in 3D lumenoid EMT. The video starts 33:33 (hh:mm) after EMT induction. Frames were acquired every 3 min and are displayed at three frames per second. Timestamps are displayed as hh:mm. Video corresponding to the image in Extended Data Fig. 9c showing a nucleus that crosses the basement membrane in both directions over time in 3D lumenoid EMT. The video starts 29:48 (hh:mm) after EMT induction. Frames were acquired every 3 min and are displayed at two frames per second. Timestamps are displayed as hh:mm. Video corresponding to the image in Extended Data Fig. 3d showing a representative overlay of bright-field images (gray) and corresponding all-cells masks (green) shown at a single Z slice for 2D colony EMT from the H2B–mEGFP cell line. Frames were acquired every 30 min for a duration of 48 h, displayed at ten frames per second. Timestamps are displayed as hh:mm. Video corresponding to the image in Extended Data Fig. 3d showing a representative overlay of bright-field images (gray) and corresponding all-cells masks (green) shown at a single Z slice for 3D lumenoid EMT from the H2B–mEGFP cell line. Frames were acquired every 30 min for a duration of 48 h, displayed at ten frames per second. Timestamps are displayed as hh:mm.