A vertically integrated system for tracking and assessing cell-cycle-aware phenotypes under confinement

TL;DR

This paper introduces a new system to study how cell migration and cell cycle interact under spatial constraints, revealing changes in cell behavior and cycle progression.

Contribution

The novel contribution is a vertically integrated platform combining fluorescent reporters, photopatterned matrices, and automated imaging for live-cell tracking under confinement.

Findings

Planar confinement reduces cell area and cytoskeletal spread while altering cell cycle phase distributions.

Confined conditions increase abnormal cell cycle events like prolonged G1 and mitotic slippage.

Cell migration is faster during the G1 phase of the cell cycle under confinement.

Abstract

Quantitative cell biology often examines migration and cell-cycle (CC) progression separately, limiting insights into their interplay under spatial constraints. Here, we present a vertically integrated platform combining multiplexed fluorescent reporters for CC phases, actin, and tubulin with photopatterned extracellular matrix islands of defined sizes, alongside an automated imaging pipeline (Fab2Mic) for high-throughput, live-cell tracking of migration and CC dynamics under planar confinement. Using HT1080 fibrosarcoma cells, we observed that planar confinement progressively reduced cell area and cytoskeletal spread, altered CC phase distributions, and increased abnormal CC events, including prolonged G1 and mitotic slippage, which is unique to confined conditions. Dynamic imaging revealed CC-dependent motility variations, with faster migration in G1. This system enables systematic, CC-aware mechanobiology studies under controlled confinement, providing access to dynamic phenotypes inaccessible to static assays and offering a scalable approach for mechanistic investigations and screening applications.

Figures

Engineering and rescue of a four-reporter CALIPERS-enabled HT1080 reference line. (a) Schematic of the genetic engineering pipeline. HT1080 fibrosarcoma cells pre-engineered with EGFP LifeAct were genome-edited at the endogenous TUBA1B locus to introduce RFP-tubulin and transduced with the FUCCIplex cassette (CFP-G1 and iRFP-S/G2/M). Lentiviral infection and hygromycin selection yielded a mixed population expressing four channels (actin, tubulin, G1, and S/G2/M). High-resolution confocal imaging confirmed robust actin filaments, tubulin networks, and phase-marked nuclei. (b) Widefield imaging revealed anomalous cytoplasmic iRFP signal in many cells (example in inset 1) compared to correctly localized nuclear iRFP (inset 2). (c) Quantification of signal distribution in the mixed population (n = 898 cells, 26 FOVs) showed 55% ± 5% of total cells CFP+ and 45% ± 5% of total cells iRFP+. (d) Localization analysis confirmed 100% nuclear localization of CFP, whereas only 15% of the iRFP signal was nuclear, with the majority retained in the cytoplasm. (e) Rescue pipeline. (i) Post-scraping enrichment of cells with nuclear iRFP yielded approximately 3× higher nuclear signal. (ii) Clonal seeding of the scraped population in 96-well format enabled systematic high-throughput screening. (iii) An automated selector routine segmented tubulin as ground truth and generated a DAPI-equivalent mask from CFP and iRFP nuclear signals, calling a clone only when nuclear counts matched >80% of the tubulin-defined cell count. (iv) Manual inspection of brightfield and fluorescence confirmed nuclear expression of both CFP and iRFP in selected clones. Five reference lines were banked, with clone E4 chosen as the working HT1080 CALIPERS line.

CC-aware phenotyping of HT1080 CALIPERS in free space. (a)–(c) Representative four-channel fluorescence images of HT1080 CALIPERS cells in G1 (CFP, cell 1) and S/G2/M (iRFP, cells 2 and 3) with actin (EGFP LifeAct) and tubulin (RFP). Scale bars, 25 μm. (d) and (e) Quantification of static phenotyping at early (4 h, n = 60 cells) and late (24 h, n = 82 cells) time points. (d) CC phase distribution: 37% ± 4% G1 vs 58% ± 11% S/G2/M (early) and 39% ± 5% G1 vs 61% ± 5% S/G2/M (late). (e) Cell area: early G1 2039 ± 228 μm2, S/G2/M 2257 ± 178 μm2; late G1 1802 ± 184 μm2, S/G2/M 2025 ± 139 μm2. (f) Tubulin spread early G1 976 ± 82 μm2, S/G2/M 1216 ± 117 μm2; late G1 655 ± 110 μm2, S/G2/M 718 ± 62 μm2. *p < 0.05 between phases within conditions (two-way ANOVA; time/area effect n.s.). (g) Dynamic phenotyping tracks of two cells across 700 min showing G1–S/G2/M transitions. Population level 48 h time-lapse in supplementary material Movie S1. Dynamic quantification of (h) velocity (average 0.35 μm/min, mean path length 192 ± 22 μm, n = 53 tracks), (i) area (increase toward S/G2/M), and (j) tubulin spread (high variability, frequent 1.5-fold changes within 1 h).

Fab2Mic pipeline for correlative fabrication-to-microscopy of engineered ECM islands. (a) Schematic of Fab2Mic workflow translating fabrication layouts (.tif + .xml) into imaging coordinates (.xml). The example shows a 50 μm square mask (SQ_2500 μm2, 5 × 5 array) with centroids (red) and imaging FOVs (green). (b)–(d) Representative photomasks to impose graded planar confinement: 10 000 μm2 (100 × 100 μm), 2500 μm2 (50 × 50 μm), and 625 μm2 (25 × 25 μm). (e)–(g) Low-resolution multi-channel views showing single-cell occupancy on ECM island arrays for each confinement regime. The adhesive islands are visualized with FITC-labeled fibronectin, allowing direct identification of island boundaries and the surrounding non-adhesive background. (h)–(j) High-resolution acquisition illustrating cell spreading constrained by the island boundary and cellular readouts on individual islands.

CC-aware phenotyping of HT1080 CALIPERS under confinement reveals scaling of morphology and cell-cycle fidelity. Representative images of HT1080 CALIPERS cells seeded on engineered ECM islands of (a) 10 000, (b) 2500, and (c) 625 μm2, showing actin (EGFP LifeAct), tubulin (RFP), and nuclei in G1 (CFP) or S/G2/M (iRFP). Scale bars, 25 μm. (d)–(f) Static phenotyping across free space and confinement. (d) CC phase distribution: free space (40% G1, 60% S/G2/M) shifts under confinement, with G1 reduced to 26% ± 17% at 10 000 μm2 and 23.5% ± 9% at 2500 μm2, with variability peaking at 625 μm2 (33% ± 25%). (e) Cell area decreased progressively with confinement, from >2000 μm2 in free space to 711 ± 106 μm2 at 625 μm2. (f) Tubulin spread followed the same trend, halving from 659 ± 246 μm2 (G1, 10 000 μm2) to 383 ± 80 μm2 (G1, 625 μm2). *p < 0.05 between phases within conditions. Representative dynamic trajectories at 2500 μm2 show normal CC progression (g) and a prolonged G1 abnormality (Long g1) (h). (i) Frequency of abnormal CC events increased with confinement: 17% in free space, 48% at 10 000 μm2, 66% at 2500 μm2, and 78% at 625 μm2. Long G1 phases dominated at small islands (75% of abnormal events at 625 μm2), while S/G2/M–G1 slippage emerged but decreased with island size (21% at 10 000 μm2, 5% at 2500 μm2, and 4% at 625 μm2). Slippage was observed only under confined conditions and was interpreted as a confinement-associated vulnerability. Rare Long S/G2/M and Long Mitosis events were also observed, together accounting for about 5% of all abnormal events. Percentages were calculated over the total number of tracked cells pooled across three independent experiments. Cell death events were also recorded (5 in free space, 1 on 10 000 μm2 islands, 7 on 2500 μm2 islands, and 9 on 625 μm2 islands) but not included in the visualization for clarity of abnormal CC phenotypes. (j)–(l) Dynamic comparisons at 2500 μm2 between a normal and abnormal (Long G1) cell. (j) Instantaneous velocity was higher in normal G1 (0.44 ± 0.22 μm/min) than in Long G1 (0.29 ± 0.23 μm/min). (k) Abnormal cells maintained 2× larger G1 area (912 ± 127 vs 465 ± 98 μm2). (l) Tubulin spread expanded 2× during normal S/G2/M progression (309 ± 56 to 416 ± 98 μm2) but fluctuated broadly without net growth in Long G1 cells (489 ± 122 μm2).

Full text

INTRODUCTION

Recent estimates still report cancer as a leading cause of mortality, with 20 × 10^6^ new cases and 9.7 × 10^6^ deaths in 2022, the majority after metastatic spread.1 Beyond genetic alterations, tumor progression is critically shaped by the tumor microenvironment, where the extracellular matrix (ECM), stromal barriers, and tissue architecture impose physical constraints on cell behaviors. In vivo, these microenvironmental features govern how cells migrate, invade, and proliferate during dissemination.2 Among them, geometric confinement challenges cell migration,3 affects nuclear integrity,4,5 and can alter cell-cycle (CC) progression.6,7 However, in vitro assays generally measure either migration or proliferation: standard 2D assays, such as scratch and transwell assays,8–10 report collective migration or transmigration but do not detail cell-cycle phase; more advanced 3D in vitro models (e.g., spheroids, collagen invasion assays) better mimic tissue architecture but likewise seldom provide real-time information on cell-cycle stage.11 Conversely, end point readouts such as EdU and Ki-67 provide snapshots of the proliferative state of the cells but cannot track CC dynamics over time.12,13 Furthermore, ECM cues, such as substrate stiffness,13–15 chemical composition with force transmission,16 and cell shape and geometry,17 have been shown to regulate adhesion, cytoskeletal tension,18–20 and proliferation programs.21–24 Each of these has been linked to either proliferative or migratory phenotypes, underscoring the importance of contextual mechanical inputs in determining cell fate.

To study how confinement reshapes CC–migration coupling,7,25–27 it is crucial to integrate confined migration assays with live CC readouts. However, open technical challenges are preventing this integration. First, most fluorescent CC reporters rely on GFP and red fluorescent protein (RFP),28 the same spectral channels used by common structural and functional sensors. This competition limits multiplexing, making it difficult to track the CC and migration-driving cytoskeletal dynamics simultaneously. Second, photopatterning approaches such as LIMAP and PRIMO now allow fabrication of engineered ECM islands with precise geometry.29 However, aligning these patterns with microscope fields of view (FOVs) still requires manual trial-and-error, reducing throughput and reproducibility.30 Finally, it has been difficult to obtain accurate long-term quantification when cells can move out of the field of view over hours of imaging.31–33 These obstacles span distinct competencies: reporter engineering, photopatterned microfabrication, and high-content imaging and analysis, which require different skill sets and infrastructure. Bringing them together within one workflow imposes a substantial coordination and tooling burden, which has constrained routine, large-throughput execution of end-to-end CC-aware confinement assays.

To address this gap, we introduce a vertically integrated platform that simultaneously tracks migration and assesses the cell cycle, unifying multiplexed reporters, photopatterned microfabrication, and high-content live-cell imaging. This system builds directly on our CALIPERS34 framework, which leverages FUCCIplex, a cell-cycle sensor spectrally multiplexable with diverse structural and functional reporters. Here, we extend this concept to a migratory HT1080 fibrosarcoma model and integrate it with Fab2Mic. This fabrication-to-microscopy correlative pipeline engineers ECM islands and registers them to acquisition coordinates, enabling automated imaging of single-cell migration and proliferation on each island. Custom image analysis scripts segment and track individual cells and nuclei over time, allowing quantification of how 2D geometric confinement modulates migration–CC coupling.

RESULTS

Microscopy-based automatic selection for engineered reporter line

We set out to engineer HT1080 fibrosarcoma cells into a four-reporter reference line that simultaneously encodes CC state and a migration-supporting cytoskeletal structure [Fig. 1(a)]. In this work, we used HT1080 cells carrying the EGFP LifeAct construct, edited the endogenous TUBA1B locus with a red fluorescent protein (RFP) tag via CRISPR/Cas9 editing, and introduced the FUCCIplex cassette via lentiviral infection under the EF1α promoter (Methods). The FUCCIplex design replaces the RFP and GFP markers for the G1 and S/G2/M phases in the traditional FUCCI sensor28 with cyan and far-red variants (CFP and iRFP), enabling simultaneous readout of CC state alongside actin and tubulin reporters.

Engineering and rescue of a four-reporter CALIPERS-enabled HT1080 reference line. (a) Schematic of the genetic engineering pipeline. HT1080 fibrosarcoma cells pre-engineered with EGFP LifeAct were genome-edited at the endogenous TUBA1B locus to introduce RFP-tubulin and transduced with the FUCCIplex cassette (CFP-G1 and iRFP-S/G2/M). Lentiviral infection and hygromycin selection yielded a mixed population expressing four channels (actin, tubulin, G1, and S/G2/M). High-resolution confocal imaging confirmed robust actin filaments, tubulin networks, and phase-marked nuclei. (b) Widefield imaging revealed anomalous cytoplasmic iRFP signal in many cells (example in inset 1) compared to correctly localized nuclear iRFP (inset 2). (c) Quantification of signal distribution in the mixed population (n = 898 cells, 26 FOVs) showed 55% ± 5% of total cells CFP+ and 45% ± 5% of total cells iRFP+. (d) Localization analysis confirmed 100% nuclear localization of CFP, whereas only 15% of the iRFP signal was nuclear, with the majority retained in the cytoplasm. (e) Rescue pipeline. (i) Post-scraping enrichment of cells with nuclear iRFP yielded approximately 3× higher nuclear signal. (ii) Clonal seeding of the scraped population in 96-well format enabled systematic high-throughput screening. (iii) An automated selector routine segmented tubulin as ground truth and generated a DAPI-equivalent mask from CFP and iRFP nuclear signals, calling a clone only when nuclear counts matched >80% of the tubulin-defined cell count. (iv) Manual inspection of brightfield and fluorescence confirmed nuclear expression of both CFP and iRFP in selected clones. Five reference lines were banked, with clone E4 chosen as the working HT1080 CALIPERS line.

Following the selection of engineered cells using hygromycin, we confirmed that actin filaments, microtubules, and nuclei, marked in G1 (CFP) or S/G2/M (iRFP), were fluorescently tagged using high-resolution microscopy and PCR [Fig. 1(a), supplementary material Figs. S1(A) and S1(C)]. In terms of nuclear markers, the mixed population featured CFP localized robustly to nuclei. In contrast, iRFP frequently exhibited cytoplasmic retention [Fig. 1(b)]. In cancer cells, such as HT1080, nuclear localization failures can arise from multiple causes, ranging from altered nuclear import machinery4,5,35 to deregulated protein degradation.36

Experimentally, across 898 annotated cells (26 fields of view), 55% ± 5% expressed CFP and 45% ± 5% expressed iRFP [Fig. 1(c)]; however, only 15% ± 5% of iRFP^+^ cells showed predominantly nuclear signal [Fig. 1(d)], indicating that correct FUCCI nuclear localization represented a limiting quality attribute of the engineered population.

Because downstream phenotyping assays rely on four-channel live-cell imaging, we used our vertically integrated microscopy workflow to qualify and select a working HT1080 CALIPERS clone under the same operational regime used for prolonged live-cell phenotyping. A microscopy-guided enrichment increased the frequency of cells with nuclear iRFP by ∼3-fold [Fig. 1(e-i)]. We then performed automated high-throughput clonal screening and applied an image-based quality filter that retained only clones with concordant tubulin-based cell counts and FUCCIplex-derived nuclear masks [Fig. 1(e); full screening logic and thresholds in the section on Methods]. More than half of the screened clones passed this automated qualification, and manual inspection of brightfield/fluorescence overlays confirmed the algorithm's calls. We banked five validated four-color lines and selected clone E4 as the working HT1080 CALIPERS line for all subsequent experiments [supplementary material Fig. S1(E)].

Baseline simultaneous assessment of cell-cycle (CC) progression and migration

To establish a reference baseline for CC-aware migratory phenotyping in free space, we first imaged HT1080 CALIPERS cells seeded on fibronectin-coated glass substrates.

Qualitatively, our analysis highlights how cellular heterogeneity emerges from the combined influence of cell-cycle state and morphology. Cell 1, shown in Fig. 2(a) and captured in G1, displays a relatively smaller projected area. In contrast, cell 2 in Fig. 2(b), which is in S/G2/M, exhibits a markedly larger projected area. Conversely, cell 3 in Fig. 2(c) shares the same S/G2/M phase as cell 2 but shows a substantially smaller projected area, underscoring the variability that exists even within a single cell-cycle phase.

*CC-aware phenotyping of HT1080 CALIPERS in free space. (a)–(c) Representative four-channel fluorescence images of HT1080 CALIPERS cells in G1 (CFP, cell 1) and S/G2/M (iRFP, cells 2 and 3) with actin (EGFP LifeAct) and tubulin (RFP). Scale bars, 25 μm. (d) and (e) Quantification of static phenotyping at early (4 h, n = 60 cells) and late (24 h, n = 82 cells) time points. (d) CC phase distribution: 37% ± 4% G1 vs 58% ± 11% S/G2/M (early) and 39% ± 5% G1 vs 61% ± 5% S/G2/M (late). (e) Cell area: early G1 2039 ± 228 μm2, S/G2/M 2257 ± 178 μm2; late G1 1802 ± 184 μm2, S/G2/M 2025 ± 139 μm2. (f) Tubulin spread early G1 976 ± 82 μm2, S/G2/M 1216 ± 117 μm2; late G1 655 ± 110 μm2, S/G2/M 718 ± 62 μm2. p < 0.05 between phases within conditions (two-way ANOVA; time/area effect n.s.). (g) Dynamic phenotyping tracks of two cells across 700 min showing G1–S/G2/M transitions. Population level 48 h time-lapse in supplementary material Movie S1. Dynamic quantification of (h) velocity (average 0.35 μm/min, mean path length 192 ± 22 μm, n = 53 tracks), (i) area (increase toward S/G2/M), and (j) tubulin spread (high variability, frequent 1.5-fold changes within 1 h).

We then quantified CC distribution and morphology across three independent experiments at early (4 h) and late (24 h) time points to control for seeding bias. A total of 60 cells were analyzed in the early condition and 82 in the late condition. G1 and S/G2/M comprised, respectively, approximately 37%–39% and 58%–61% of the population at both time points [Fig. 2(d)], consistent with near-steady-state cycling, where the fraction of cells observed in each phase reflects the relative time that cells spend in that phase across the population, which indicates ergodicity. Normality (p = 0.822) and variance (p = 0.418) tests were passed; two-way ANOVA with Tukey correction confirmed statistical differences only between phases within each condition (^*^p < 0.05). The average cell area trended downward from early to late time points but without significance: early G1 2039 ± 228 μm^2^ vs late G1 1802 ± 184 μm^2^; early S/G2/M 2257 ± 178 μm^2^ vs late S/G2/M 2025 ± 139 μm^2^ [normality p = 0.492, variance p = 0.907; no factor-level effect, Fig. 2(e)]. Tubulin spread was consistently higher in S/G2/M than in G1 but declined between early and late: early G1 976 ± 82 μm^2^ vs late G1 655 ± 110 μm^2^; early S/G2/M 1216 ± 117 μm^2^ vs late S/G2/M 718 ± 62 μm^2^ [Fig. 2(f)]. These data suggest that tubulin remodeling is the most sensitive morphological feature to the passage of time in culture.

To capture how migration evolves through phases, we performed long-term CC-aware live imaging for 15 h at 30 min intervals (three independent experiments, ≥3 FOVs each). Representative trajectories [Fig. 2(g)] show two cells tracked across CFP-iRFP transitions. Overall, across 54 free space tracks, cells spent on average 4.5 ± 2.0 h in G1 and 5.2 ± 2.0 h in S/G2/M, respectively. Velocity traces revealed an average velocity of 0.37 ± 0.20 μm/min in G1 and 0.32 ± 0.20 μm/min in S/G2/M, with a mean path length of 192 ± 22 μm. Instantaneous velocity peaked during G1, decreased at the G1–S/G2/M transition, and rose again before mitosis [Fig. 2(h)], indicating that migration speed is coupled to cell-cycle state, potentially reflecting higher motility during growth, reduced dynamics during DNA replication, and cytoskeletal reorganization preceding division. In parallel, projected cell area increased toward S/G2/M, consistent with the static imaging results [Fig. 2(i)]. Importantly, these phase-dependent trends are not limited to the representative traces shown in Figs. 2(h)–2(j): a 48 h free space acquisition enabling quantification of 50 complete cell cycles reproduced the same temporal modulation of velocity and area when computed as a population average with phase-aligned trajectories [supplementary material Figs. S3(D) and S3(E); supplementary material Movie S1].

Tubulin spread fluctuated substantially, often changing by 1.5-fold within an hour, reflecting intracellular remodeling that could not be captured from static snapshots [Fig. 2(j)]. A small G0/G1-arrested subpopulation was present but excluded from this analysis, as arrested cells remain outside the proliferative cycle and thus do not engage in the dynamic remodeling observed in cycling populations; these are addressed in Fig. 4.

Taken together, these results suggest that in free space, HT1080 CALIPERS cells maintained a near steady-state distribution of approximately 40% in G1 and 60% in S/G2/M, in line with previous reports27,37,38 and the detailed FUCCIplex cross-validation at the population level reported in supplementary material Figs. S2 and S3. Dynamic assays complement the picture by directly linking motility to cell-cycle transitions. Existing in vivo and ex vivo imaging studies39 demonstrate that CC-dependent regulation of invasion occurs. At the same time, in vitro longitudinal tracking approaches like ours34,37,40 are valuable, as they can capture instantaneous velocity, trajectory, and intracellular variability, which are inaccessible to static end points.

Implementation and characterization of a correlative pipeline for photofabrication and imaging

Having characterized our HT1080 CALIPERS reporter line in free space, we set out to study how cell migration and morphology evolve under confinement by engineering adhesive ECM patterns via photofabrication that constrain cell migration over time. We realized that manually performing photofabrication and imaging as separate steps was both error-prone and inefficient. Mapping fabricated arrays and imaging fields of view (FOVs) via manual trial-and-error could lead to inaccuracy, which wasted imaging or analysis time. Instead, we developed Fab2Mic, a correlative fabrication-to-microscopy pipeline that directly translates photofabrication layouts into imaging coordinates, ensuring efficient and reproducible alignment between fabrication and live imaging [Fig. 3(a)].

Fab2Mic pipeline for correlative fabrication-to-microscopy of engineered ECM islands. (a) Schematic of Fab2Mic workflow translating fabrication layouts (.tif + .xml) into imaging coordinates (.xml). The example shows a 50 μm square mask (SQ_2500 μm2, 5 × 5 array) with centroids (red) and imaging FOVs (green). (b)–(d) Representative photomasks to impose graded planar confinement: 10 000 μm2 (100 × 100 μm), 2500 μm2 (50 × 50 μm), and 625 μm2 (25 × 25 μm). (e)–(g) Low-resolution multi-channel views showing single-cell occupancy on ECM island arrays for each confinement regime. The adhesive islands are visualized with FITC-labeled fibronectin, allowing direct identification of island boundaries and the surrounding non-adhesive background. (h)–(j) High-resolution acquisition illustrating cell spreading constrained by the island boundary and cellular readouts on individual islands.

The algorithm takes the fabrication design (.tif) and metadata (.xml) as input, extracts feature centroids and array geometry, and exports an xml file that can be used immediately for multi-point acquisition in most microscopes via commercial software or Micro-Manager. Smaller features, less than 500 μm in size, are centered in the imaging FOV, whereas larger features are tiled seamlessly to minimize redundancy. We validated this method on a 50 μm square design (5 × 5 array, 15 mJ/mm^2^ exposure) and confirmed that centroids (red) and imaging FOVs (green) matched the fabricated array on inspection [Fig. 3(a)]. As a proof-of-concept, we also fabricated a 2D grayscale rendering of a picture of the Ponte Coperto in Pavia, coated it with FITC-fibronectin, and imaged a composite of 4 × 3 stitched FOVs [supplementary material Figs. S6(A) and S6(B); supplementary material Movie S2]. The Fab2Mic routine automatically generated the coordinate map and enabled acquisition without any manual adjustment.

We then applied this pipeline to fabricate engineered ECM islands of three sizes (10 000, 2500, and 625 μm^2^) arranged as 25 × 25 arrays [Figs. 3(b)–3(d)]. Widefield 477 nm imaging of FITC-fibronectin confirmed sharp island boundaries and reproducible array geometry. Importantly, the surrounding inter-island background remained passivated and non-adhesive. Thus, adherent cells localized to ECM-positive islands rather than attaching in unpatterned regions, as shown in Figs. 3(e)–3(j). For all ECM-confinement analyses, we included only single-cell islands and excluded rare boundary-straddling or off-island adherent cells by manual curation of the dataset.

Because single-cell occupancy is essential for unbiased phenotyping, we adopted a probabilistic seeding strategy that deliberately favors single-cell islands. On average, more than 50% of the occupied islands contain a single cell, although this comes at the cost of an overall occupancy of about 20% of the available islands [supplementary material Figs. S6(D) and S6(E)]. This trade-off was strategic: by combining probabilistic seeding with the automated Fab2Mic pipeline, we ensured reproducible access to single-cell islands while substantially reducing manual settings and registration time. After seeding HT1080 CALIPERS cells at optimized densities (4× array size for 625 and 2500 μm^2^, 2× for 10 000 μm^2^), cells adhered selectively to the islands. Finally, high-resolution imaging centered on individual islands revealed actin filaments aligned with fibronectin boundaries, microtubule networks spanning the cytoplasm, and CC phases marked by CFP (G1) or iRFP (S/G2/M).

Taken together, these results validate Fab2Mic as a correlative pipeline that reduces alignment errors in multi-point automated acquisitions, enables multiplexed imaging, and streamlines assay execution by saving time. By demonstrating robust single-cell occupancy across engineered ECM islands of decreasing size, we established the feasibility of combining photofabrication and CC-aware live imaging for quantitative studies of confined migration.

Quantifying confinement effects on CC progression and proliferation

The combination of a validated reporter line, free space baselines, and the Fab2Mic pipeline for high-content correlation between fabrication and imaging enables the systematic study of the influence of planar confinement on both migration and proliferation. Using engineered ECM islands of 10 000, 2500, and 625 μm^2^, we quantified CC distribution, morphology, and cytoskeletal dynamics under defined geometric constraints, comparing directly with free space conditions. Here, we set 2D planar confinement exclusively by restricting the adhesive area on a passivated substrate, such that cells are laterally constrained by the island boundary while remaining uncompressed in the vertical dimension. The three island sizes span distinct confinement regimes relative to the free space projected area: 625 μm^2^ enforces strong footprint restriction, 2500 μm^2^ approximates a near-physiological spread area but prevents long-range displacement, and 10 000 μm^2^ permits intra-island exploration while still bounding migration to a finite domain. From the outset, imaging was restricted to islands occupied by single cells, ensuring unbiased phenotyping at the individual cell level.

Cells adapted to the adhesive island boundaries based on their free space area: cells retained large spread morphologies on 10 000 μm^2^ islands [Fig. 4(a), roughly five times their free space area], while at 625 μm^2^ the cytoskeleton filled the available area [Fig. 4(c)]. Interestingly, the CC distribution and cytoskeletal architecture show a scaling in confinement [Figs. 4(d)–4(f)]. This scaling suggests that confinement acts as a continuum rather than a binary on/off state, with gradual modulation of both morphology and CC fidelity. In free space, approximately 40% of cells were in G1 and about 60% in S/G2/M. On the engineered ECM islands, the G1 fraction decreased to 26% ± 17% (10 000 μm^2^) and 23.5% ± 9% (2500 μm^2^), with partial recovery to 33% ± 25% at 625 μm^2^, suggesting a relative accumulation in pre-mitotic phases under confinement. Notably, variability increased with confinement, particularly in G1, where standard deviation rose to ±25% at 625 μm^2^ [Fig. 4(d)]. Cell area followed a similar trend: from over 2000 μm^2^ in free space to 1663 ± 639 (G1) and 1483 ± 425 μm^2^ (S/G2/M) at 10 000 μm^2^, 1803 ± 867 and 1559 ± 137 μm^2^ at 2500 μm^2^, and just 711 ± 106 and 688 ± 128 μm^2^ at 625 μm^2^ [Fig. 4(e)]. Tubulin spread also decreased with confinement, roughly halving from 659 ± 246 μm^2^ (G1, 10 000 μm^2^) to 383 ± 80 μm^2^ (G1, 625 μm^2^) [Fig. 4(f)]. Dynamic phenotyping revealed how these morphological effects intersect with CC fidelity [Figs. 4(g)–4(i), supplementary material Fig. S9, and supplementary material Movie S3].

*CC-aware phenotyping of HT1080 CALIPERS under confinement reveals scaling of morphology and cell-cycle fidelity. Representative images of HT1080 CALIPERS cells seeded on engineered ECM islands of (a) 10 000, (b) 2500, and (c) 625 μm2, showing actin (EGFP LifeAct), tubulin (RFP), and nuclei in G1 (CFP) or S/G2/M (iRFP). Scale bars, 25 μm. (d)–(f) Static phenotyping across free space and confinement. (d) CC phase distribution: free space (40% G1, 60% S/G2/M) shifts under confinement, with G1 reduced to 26% ± 17% at 10 000 μm2 and 23.5% ± 9% at 2500 μm2, with variability peaking at 625 μm2 (33% ± 25%). (e) Cell area decreased progressively with confinement, from >2000 μm2 in free space to 711 ± 106 μm2 at 625 μm2. (f) Tubulin spread followed the same trend, halving from 659 ± 246 μm2 (G1, 10 000 μm2) to 383 ± 80 μm2 (G1, 625 μm2). p < 0.05 between phases within conditions. Representative dynamic trajectories at 2500 μm2 show normal CC progression (g) and a prolonged G1 abnormality (Long g1) (h). (i) Frequency of abnormal CC events increased with confinement: 17% in free space, 48% at 10 000 μm2, 66% at 2500 μm2, and 78% at 625 μm2. Long G1 phases dominated at small islands (75% of abnormal events at 625 μm2), while S/G2/M–G1 slippage emerged but decreased with island size (21% at 10 000 μm2, 5% at 2500 μm2, and 4% at 625 μm2). Slippage was observed only under confined conditions and was interpreted as a confinement-associated vulnerability. Rare Long S/G2/M and Long Mitosis events were also observed, together accounting for about 5% of all abnormal events. Percentages were calculated over the total number of tracked cells pooled across three independent experiments. Cell death events were also recorded (5 in free space, 1 on 10 000 μm2 islands, 7 on 2500 μm2 islands, and 9 on 625 μm2 islands) but not included in the visualization for clarity of abnormal CC phenotypes. (j)–(l) Dynamic comparisons at 2500 μm2 between a normal and abnormal (Long G1) cell. (j) Instantaneous velocity was higher in normal G1 (0.44 ± 0.22 μm/min) than in Long G1 (0.29 ± 0.23 μm/min). (k) Abnormal cells maintained 2× larger G1 area (912 ± 127 vs 465 ± 98 μm2). (l) Tubulin spread expanded 2× during normal S/G2/M progression (309 ± 56 to 416 ± 98 μm2) but fluctuated broadly without net growth in Long G1 cells (489 ± 122 μm2).

To test whether the reduced G1 fraction observed in static imaging reflected an artifact, we performed dynamic imaging as described in Fig. 2 and evaluated cells with normal and abnormal CC events separately. Across 156 tracked cells (31 at 10 000 μm^2^, 58 at 2500 μm^2^, and 67 at 625 μm^2^), 42 cells exhibited normal CC progression (supplementary material Fig. S8), defined by canonical FUCCI phase progression with a single monotonic G1–S/G2/M transition and no abnormal CC events, with average phase durations of ∼39% G1 and ∼61% S/G2/M, even under confinement. Notably, this phase partition is consistent with prior HT1080 FUCCI time-lapse reports, which describe G1 lasting ∼4–6 h and S/G2 lasting ∼8–10 h with ∼1 h mitosis.37 However, abnormal CC events rose with increasing confinement, from a baseline of 17% in free space to 78% at 625 μm^2^ [Fig. 4(i)]. Abnormal events were classified as Long G1, Long S/G2/M, Long Mitosis, or S/G2/M–G1 slippage (see the classification of abnormal cell cycles in the section on Methods). While Long G1 and S/G2/M–G1 slippage dominated under confinement, rare instances of Long S/G2/M and Long Mitosis were also observed, together accounting for approximately 5% of all abnormal events. Long G1 phases dominated confinement-induced anomalies, increasing from 37% of abnormal events in free space to 75% at 625 μm^2^. Conversely, S/G2/M–G1 mitotic slippage emerged,41,42 but its incidence decreased with tighter constraints (21% at 10 000 μm^2^ to 4% at 625 μm^2^). In the context of this study, S/G2/M–G1 slippage was observed only under our confined conditions and was interpreted as a confinement-associated vulnerability. Representative trajectories at 2500 μm^2^ [Figs. 4(g) and 4(h)], similar to the projected area of cells in free space, highlight the divergence: a normal cell progressed from G1 to S/G2/M with G1 velocity averaging 0.44 ± 0.22 μm/min, while a Long G1 cell had a lower motility (0.29 ± 0.23 μm/min) [Fig. 4(j)]. Morphologically, abnormal G1 cells maintained 2× larger areas (912 ± 127 vs 465 ± 98 μm^2^) [Fig. 4(k)] and broader tubulin spread (489 ± 122 vs 309 ± 56 μm^2^), lacking the 2× cytoskeletal expansion observed in normal S/G2/M transitions (416 ± 98 μm^2^) [Fig. 4(l)].

Taken together, these results indicate that confinement modulates both morphology and CC fidelity, consistent with prior reports7,19,43 showing that reduced adhesive area compresses cytoskeletal assembly. It increases the frequency of abnormal phenotypes, particularly prolonged G1 states (Long G1).44 Notably, while static imaging suggested a reduced representation of normal G1 under confinement, dynamic imaging revealed two distinct populations: normal and abnormal cycling cells. This indicates that ergodicity assumptions must be applied with caution. Automated cell tracking and cell-cycle phase assignment, as used here, provide critical value in distinguishing these subpopulations.

DISCUSSION

This study establishes a vertically integrated platform that unifies multiplexed CC reporters, engineered ECM islands, and correlative high-content imaging to resolve the coupling between migration and CC under geometric confinement. By combining spectral reallocation through FUCCIplex (freeing the green/red channels for structural phenotyping), bridging the registration-throughput gap with Fab2Mic, and enabling longitudinal single-cell tracking, we convert confinement into a tunable parameter, consistent with our results. Indeed, as the adhesive area decreases, the G1 fraction declines, cytoskeletal organization becomes compressed, and variability increases. Embedding CC resolution directly into high-content morphological analysis provides a practical framework for systematic phenotyping of migration and proliferation under constraint, extending classic adhesion-22,23 and geometry-dependent control of proliferation17 including nuclear volume G1 coupling on micropatterns.25,26 Our results also support the importance of including live-cell imaging experiments to provide dynamic phenotyping, which revealed single-cell confinement-associated behaviors invisible to static imaging, including Long G1 or Long S/G2/M phases and S/G2/M–G1 mitotic slippage. These phenotypes are consistent with reports that mechanical stress impairs mitotic progression and promotes slippage-like outcomes under constraint.42,45,46

More broadly, our platform extends the pioneering framework established by Ingber, who demonstrated how adhesion geometry and cytoskeletal tension regulate proliferation.17 By integrating CC resolution through FUCCIplex and confinement control via Fab2Mic, we embed this principle into a high-content, imaging-based screening. This provides a framework to interrogate how physical context may shape proliferative fate decisions. The approach is compatible with large-scale image-based screens, where CC-aware analysis can be applied to evaluate compound effects on CC–migration coupling or to design microenvironments that guide morphogenesis in engineered tissues.47 Time-resolved high-content screening workflows have precedent (e.g., genome-scale live-cell imaging of mitosis), demonstrating the feasibility and value of temporal phenoprints.48–50 However, scaling dynamic, multi-channel assays demands a strengthened quality-control framework. Segmentation errors, tracking identity swaps, and channel registration drift remain partial bottlenecks, and manual curation was still necessary in this study for a subset of the data (∼20%). Future integration of AI-based segmentation and standardized quality control (QC) benchmarks (e.g., error rates, reproducibility metrics) will be essential to translate CC-aware assays into robust pharma-ready pipelines, including multi-metric artifact detection and image-level QC (focus/illumination/saturation checks), which are being democratized via open guidance/toolkits for large-scale screens.51 For example, cell-level QC workflows utilize clustering coupled with one-class support vector machines (SVMs) to flag segmentation artifacts and recover valid cells from partially corrupted images.52 Regardless of these future improvements, we believe our work bridges classical mechanobiology with next-generation tissue engineering, providing both a conceptual extension and a practical toolkit for probing how cells “read” geometry and confinement to balance migration and growth.

To contextualize these findings, it is important to acknowledge key limitations of the present work and the most direct paths for improvement. First, our confinement model intentionally simplifies the in vivo mechanical landscape: cells are constrained on 2D ECM-patterned substrates with defined adhesive boundaries. This design isolates geometry-dependent effects but does not recapitulate the hallmark features of 3D tissues, including matrix porosity, viscoelasticity, degradability, and multicellular remodeling. In this context, the purpose of this 2D model is to generate quantitative, well-controlled single-cell phenotypes that can prioritize hypotheses and readouts for in vivo ground-truthing rather than serve as a replacement for in vivo validation. Specifically, the confinement-associated CC signatures observed here (e.g., prolonged G1/slow-cycling fractions and slippage-like outcomes) suggest candidate biomarkers that can be tested using established in vivo proliferation and arrest-associated readouts (Ki-67, EdU/BrdU incorporation, and p27/p21 as CDK inhibitors associated with G1 arrest/quiescence-like programs), complemented by single-cell profiling to test concordant transcriptional programs. These measurements can be paired with lineage tracing and/or intravital imaging in relevant microenvironmental niches to determine whether similar CC states occur in vivo and whether they are associated with dormancy or drug-tolerant residual disease.

Second, we primarily interrogate a single-cell type and a limited set of ECM architectures; therefore, the quantitative phase-migration relationships reported here should be interpreted as platform-specific phenotypes within a controlled design space rather than universal rules across tissues or microenvironments. Third, dynamic multi-channel imaging remains constrained by practical tradeoffs among throughput, phototoxicity, segmentation/tracking fidelity (especially in dense fields), and reporter-specific limitations (spectral crosstalk, expression variability, and robust phase calling over long acquisitions). These constraints motivated partial manual curation in this study.

Accordingly, future extensions should prioritize translating confinement into more physiologically relevant settings (2.5D/3D matrices, microchannel-based confinement, and patterned ECM networks with tunable stiffness/viscoelasticity); integrating additional mechanical and biochemical cues (e.g., strain, shear, stiffness gradients, and co-culture contexts); and strengthening the computational stack through automated QC, event-aware tracking, and uncertainty-aware reporting to enable unbiased, scalable, CC-aware phenotypic profiles.

Despite these limitations, the present work provides practical, extensible toolkits for CC-aware phenotyping within a defined design space. We introduce an integrated, open-format workflow that links engineered multi-reporter cells, ECM microenvironment fabrication-to-imaging validation, and phase-resolved quantitative readouts into a single experimental–computational pipeline. By keeping both the readouts modular and the analysis transferable across experiments, this platform can be readily extended to additional cell types, ECM architectures, and higher-dimensional microenvironments.

In the cancer context, our study contributes by introducing a four-color reporter HT1080 line that enables systematic exploration of CC–migration coupling. We deliberately selected conditions that emphasize long-term behaviors, whereas actual metastatic migration in vivo is typically a transient, short-lived event. The value of our approach lies less in reproducing metastatic dissemination per se and more in providing a robust system to study how CC dynamics interact with migratory states. This perspective aligns with the “go-or-grow” framework,22,40,53 and with pathology observations that invasive fronts are often slow cycling while cores are proliferative.24,54,55 In vivo, metastatic cells “go” to a distant site and then “grow,” whereas in our system, cells “go” but remain confined, leading to gradual accumulation of CC stress.

Although caution is warranted when extrapolating to clinical applications, the CC abnormalities revealed under prolonged confinement indicate an increased frequency of abnormal cell-cycle events, a state that, in principle, could prevent the development of a large metastatic mass. By contrast, recent findings show that confinement-induced states can also be more drug-tolerant, as HMGB2-high melanoma cells under mechanical confinement resist Taxol and BRAF/MEK inhibitors.54 Reconciling these observations highlights a critical dependency on therapeutic context: most cytotoxic drugs preferentially target rapidly proliferating cells, so confinement-driven cell-cycle exit may simultaneously suppress metastatic outgrowth (beneficial) but blunt the efficacy of conventional chemotherapy (detrimental). Mechanistically, Taxol triggers mitotic arrest; therefore, non-dividing, confined cells are inherently less affected, and confinement-driven HMGB2 programs slow cycling and tolerance, with HMGB2 overexpression impairing BRAF/MEK response in vivo.7 This tension highlights the importance of considering CC–migration coupling within the broader “go-or-grow” framework and evaluating confinement as both a source of mechanistic vulnerability and a driver of therapy resistance. Notably, some in vitro studies do not observe a strict trade-off: FUCCI-based analyses reported comparable motility in G1 vs S/G2/M and even under pharmacologic arrest,40 and cross-line surveys found migration can correlate with proliferation in certain cancers.53

Our results suggest that the appropriate geometric confinement can be leveraged to play along the go-or-grow continuum. This framing opens the door to thinking of designing confinement with a therapeutic goal. Analogous to immuno-engineered hydrogels designed to capture or prime immune cells,56–58 one could envision biomaterials tailored as exploratory “capture sites” for metastatic cells. While speculative, this concept illustrates how our vertically integrated platform for CC–migration coupling assessment may inform both mechanistic mechanobiology studies and exploratory therapeutic screenings.

METHODS

Cell line generation

HT1080 human fibrosarcoma cells (Ibidi, #HT-1080-LifeAct-TagGFP2, catalog# 40101) were genome-engineered to generate a four-color CALIPERS reporter line.

CRISPR/Cas9 tubulin tagging

Endogenous α-tubulin (TUBA1B, NM_006082.3) was tagged with tagRFP using the Thermo Fisher Scientific TrueTag system. The donor sequence and primers with locus-specific homology arms for RFP tagging of the N-terminus of α-tubulin (TUBA1B, NM_006082.3) were designed with the assistance of the TrueDesign Genome Editor tool (Thermo Fisher Scientific).59 The linear TrueTag dsDNA donor was amplified by PCR using Phusion Flash High-Fidelity PCR Master Mix (TrueTag Donor DNA Kit, RFP, Thermo Fisher Scientific, catalog# A42993). The ribonucleoprotein (RNP) complex was assembled by combining TrueGuide synthetic sgRNA (Thermo Fisher Scientific; target sequence: GCACGGCTTACTCACCATAG) with TrueCut Cas9 Protein V2 (Thermo Fisher Scientific, catalog# A36496). TrueCut Cas9 Protein V2 (2 μg, 12 pmol), sgRNA (400 ng, 12 pmol), and the TrueTag donor (500 ng) were resuspended in 10 μl of Resuspension Buffer R (Thermo Fisher Scientific, catalog# MPK1025) and subsequently mixed with 8 × 10^4^ HT1080 cells resuspended in 8 μl of the same buffer, yielding a final reaction volume of 18 μl. A 10 μl aliquot of this mixture was electroporated using the Neon Transfection System (Thermo Fisher Scientific, catalog# NEON1S) with the following parameters: 1200 V, 20 ms, 4 pulses. Following electroporation, cells were seeded into 24-well plates containing 0.5 ml of Dulbecco's Modified Eagle Medium/Ham's F-12 Nutrient Mixture without phenol red (DMEM F-12, Gibco, catalog# 21041-025), supplemented with 10% heat-inactivated fetal bovine serum (FBS, Gibco, catalog# 10270-106). Cultures were incubated at 37 °C in a humidified CO_2_ atmosphere. RFP-positive cells were subsequently isolated and expanded in a clonal manner. Edited clones were validated via live-cell fluorescence microscopy and PCR [supplementary material Figs. S1(A) and S1(C)].

FUCCIplex sensor integration

The FUCCIplex cassette was derived from pBOB-EF1-FastFUCCI-Puro (Addgene, plasmid #86849) by replacing mKO2 (Cdt1 degron) with mTurquoise2 and hmAzami Green (Geminin degron) with miRFP670. The cassette was cloned into a third-generation lentiviral backbone (pLV[Exp]-Hygro-EF1A, VectorBuilder, catalog# VB210825-1211nbd) with hygromycin resistance. Lentiviral particles were produced in HEK293T packaging cells (ATCC, catalog# CRL-1573), seeded at 2.5 × 10^6^ cells per 10 cm dish pre-coated with 0.2% gelatin (Sigma-Aldrich, catalog# G1393). Transfection was performed with Lipofectamine 2000 (Thermo Fisher Scientific, catalog# 11668019). Viral supernatants were collected at 48 h, clarified by centrifugation, and applied to HT1080 cells. In more detail, HT1080 cells were plated at a density of 6 × 10^4^ cells per well in 12-well plates containing complete DMEM F-12 medium. After 24 h, the culture medium was replaced with fresh complete medium supplemented with 8 μg/ml polybrene (Sigma-Aldrich, catalog# TR-1003-G). Concentrated FUCCIplex lentiviral particles (>10^8^ TU/ml, VectorBuilder; 10 μl resuspended in 500 μl of complete medium) were then added directly to the wells. Cells were incubated overnight at 37 °C in a humidified 5% CO_2_ atmosphere. The following morning, the viral medium was aspirated and replaced with fresh complete medium. After 4 days, positively transduced cells were selected using Hygromycin B (Invitrogen, catalog# 10687010, 20 μl/ml from a 50 mg/ml stock in PBS). Resistant populations were clonally seeded and expanded. PCR and confocal fluorescence imaging verified correct integration and reporter expression.

PCR and RT-qPCR

Genomic DNA was extracted from HT1080 EGFP LifeAct and FUCCIplex clones using the DNeasy Blood and Tissue Kit (Qiagen, catalog# 69504) according to the manufacturer's instructions. For genomic PCR, two different primer pairs were used to amplify the TUBA1B WT locus and the RFP-tagged gene. Specifically, primer pair 1–3 recognizes the WT locus, while primer pair 2–3 generates an amplicon only in the condition where the gene has been edited. PCR products were analyzed by agarose gel (1%) electrophoresis. PCR was carried out using Platinum PCR SuperMix, High Fidelity (Thermo Fisher Scientific, catalog# 12532016).

Total RNA was isolated using TRIzol reagent (Thermo Fisher Scientific, catalog# 15596026), treated with DNase I (Thermo Fisher Scientific, catalog# EN0521), and reverse-transcribed using SuperScript™ IV Reverse Transcriptase (Thermo Fisher Scientific, catalog# 18090200). RT-qPCR was performed with primers specific for miRFP670, Hygromycin, and FUCCIplex. Relative expression levels were determined using the ΔΔCt method, with HPRT as the internal control.

Cell culture

HT1080 CALIPERS cells were maintained in Dulbecco's Modified Eagle Medium/Ham's F-12 Nutrient Mixture without phenol red (DMEM F-12, Gibco, catalog# 21041-025), supplemented with 10% heat-inactivated fetal bovine serum (FBS, Gibco, catalog# 10270-106) and 1% penicillin–streptomycin (HiMedia, catalog# A001-100ML). Cells were cultured at 37 °C in a humidified 5% CO_2_ incubator and routinely split at 70% confluency. For experiment preparation, cells were detached using Trypsin/EDTA 0.25% (Thermo Fisher Scientific, catalog# 25200056), resuspended in complete DMEM F-12, and properly counted according to the experimental needs.

CC phase assignment with DNA-content profiling and FUCCIplex-based flow cytometry

To benchmark baseline DNA-content fractions and validate FUCCIplex-derived phase occupancy in multicolor-engineered HT1080 lines, flow cytometry was performed on three asynchronous cultures: parental HT1080 EGFP, a partially edited HT1080 EGFP LifeAct—Actin^+^/RFP—Tubulin^+^ line (FUCCIplex-negative control), and the fully edited HT1080 CALIPERS line (EGFP LifeAct—Actin, RFP-Tubulin, and FUCCIplex CFP/iRFP). Cells were harvested as single-cell suspensions and analyzed using sequential gating to exclude debris and doublets/aggregates while retaining a single-cell population. For DNA-content analysis, samples were processed using a standard PI/RNase staining workflow, and PI DNA-content histograms were used to quantify the fractions of cells in G0/G1, S, and G2/M. Due to the multicolor nature of the edited lines and the observed sensitivity of PI readouts to instrument settings and spectral overlap with RFP in the PI detection channel, PI-based DNA-content fractions are reported here only for the parental HT1080 EGFP line as a robust baseline reference [supplementary material Fig. S2(A)]. In parallel, FUCCIplex-based phase assignment in the HT1080 CALIPERS line was performed by bivariate analysis of CFP (G1) and iRFP (S/G2/M) fluorescence [supplementary material Fig. S2(B)]. CFP and iRFP thresholds were defined using the FUCCIplex-negative EGFP-LifeAct^+^/RFP-Tubulin^+^ control to set the negative baseline and account for autofluorescence and residual spillover/bleedthrough from structural reporters. Using these thresholds, the CALIPERS population was partitioned into CFP/iRFP quadrants to quantify reporter-positive fractions and derive the FUCCI-based phase distribution. All PI and FACS analyses were performed in FlowJo Software (FlowJo™ Software, v. 11),60 as shown in supplementary material Fig. S2(A).

Cytoplasmic vs nuclear CFP/iRFP evaluation

To evaluate CFP and iRFP signal expression and localization, HT1080 CALIPERS were screened with high-resolution widefield fluorescence microscopy with a Nikon CFI Plan Apo Lambda S 40× silicone oil objective (Silicone oil immersion, NA 1.25, WD 0.30 mm, catalog# MRD73400). See the section on Microscopy acquisitions for detailed imaging settings.

Samples were prepared by seeding 5 × 10^4^ cells on 35 mm #1.5 glass-bottom dishes (Ibidi, catalog# 81158) pre-coated with 20 μg/ml human plasma fibronectin (Sigma-Aldrich, catalog# F0895) in standard supplemented DMEM F-12 without phenol red. After 24 h, the medium was refreshed, and the cells were imaged.

The dataset was manually curated to identify cells expressing either CFP or iRFP signals. For each population, signal localization was assessed to distinguish between exclusive nuclear vs cytoplasmic expression. GraphPad Prism (version 10.4.2) was used to generate the graphics of this panel set. Results are reported in bar graphs with mean ± SEM.

Scraping and clonal seeding

To enhance the yield of nuclear iRFP localization of HT1080 CALIPERS, a gentle scraping method was employed. A total of 5 × 10^4^ cells were seeded on two 35 mm #1.5 glass-bottom dishes (Ibidi, catalog# 81158) pre-coated with 20 μg/ml human plasma fibronectin (Sigma-Aldrich, catalog# F0895) in standard supplemented DMEM F-12 without phenol red medium.

After 24 h, the medium (DMEM F-12, Gibco, catalog# 21041-025) was refreshed before the imaging session. Widefield fluorescence microscopy was used to manually annotate regions of interest (ROIs) corresponding to nuclear iRFP-positive cells directly on the dish.

Subsequently, adherent populations were gently detached using a sterile cell scraper.

The scraped cell sample was resuspended in fresh medium and replated for expansion and subsequent clonal seeding. Once the population reached 70% confluency, cells were trypsinized with Trypsin/EDTA, resuspended in complete DMEM F-12 medium, counted, and diluted to a final concentration of ∼0.8 cells per well. Clonal seeding was performed in a sterile 96-well plate (Ibidi, catalog# 89626), where 100 μl was dispensed per well to favor single-cell deposition. To ensure optimal fluorescence imaging, 48 wells were seeded, avoiding the outer rows/columns of the plate. The 96-well plate was incubated under standard culture conditions (37 °C, 5% CO_2_), and wells were monitored to confirm clonal origin. Of the 48 initially seeded wells, 8 remained empty, leaving 40 available clones for evaluation. The medium was refreshed every other day with standard supplemented DMEM F-12 until day 7 of clonal expansion.

Automated FUCCIplex clonal screening

A high-throughput live-cell imaging JOBS was developed using Nikon NIS-Elements AR software (version 5.42.03)61 to scan a multi-well plate through a multi-channel serial acquisition and large-image composition per well. The JOBS wizard enabled selection of wells of interest, definition of imaging channel stacks, and specification of the reference channel for automated focus adjustment, which was based on contrast optimization.

The implementation utilized 638 nm for iRFP (S/G2/M), 546 nm for RFP (Tubulin), 477 nm for EGFP LifeAct (Actin, used as the reference plane), 446 nm for CFP (G1), and brightfield acquisition. Each well is acquired as a large image of 6 × 6 fields within a Nikon CFI Plan Fluor DIC 10× (air objective, NA 0.3, WD 16 mm, catalog# MRH00105). The entire well is then reconstructed by stitching with a 10% overlap and stored as a merged file.

FUCCIplex selector routine

An automated FUCCIplex selector routine was developed using the General Analysis 3 module in Nikon NIS-Elements AR software (version 5.42.03).61 The routine was based on tubulin and FUCCIplex nuclear signals. Specifically, FUCCIplex nuclear signals were combined into a DAPI-equivalent (DAPIEq) mask that captured both CFP and iRFP expression. The mask generated from CFP and iRFP signals was created within thresholds defined by object intensity and size to exclude cytoplasmic signals. CFP-positive nuclei were selected with a mean intensity >800 and a nuclear diameter <15 μm. iRFP nuclear-localized expression was selected with an intensity >600 and a nuclear diameter <15 μm. Using the tubulin and DAPIEq masks, the algorithm considered the number of cells detected by tubulin as the ground truth. As a selection criterion, the DAPIEq object count was required to reach >80% of the tubulin-based cell count for a well to be classified as an HT1080 CALIPERS four-reporter clone. The threshold was relaxed to 80% because early G1- or M-phase cells may be missed by the FUCCIplex selector but still appear in tubulin-based counts, leading to slight mismatches.

HT1080 CALIPERS synchronization by nocodazole mitotic arrest

HT1080 CALIPERS cells were seeded and allowed to adhere for 8 h before synchronization. Cells were then treated overnight with nocodazole (Sigma-Aldrich, M1404-50MG, 100 ng/ml) to induce mitotic arrest. The following day, mitotic cells were collected by gentle shake-off, nocodazole was removed by washout, and the harvested cells were reseeded in an Ibidi 35 mm round glass-bottom dish in fresh complete medium.62 Cells were allowed to re-adhere for 4 h before initiating live-cell imaging. Four-channel fluorescence time-lapse imaging was performed for 18 h at low magnification (large field of view) to enable tracking of the full population and quantification of cell-cycle progression and dynamics during release from nocodazole-induced arrest (supplementary material Figs. S4 and S5).

Photofabrication

Substrate cleaning and passivation

35 mm #1.5 glass-bottom dishes (Ibidi, catalog# 81158) were washed in 2% Hellmanex III solution (Hellma Analytics, catalog# 9-307-0110), rinsed with DI water, and dried. Dishes were exposed to UV-ozone for 30 min (BioForce Nanosciences ProCleaner). Substrates were coated with 200 μl of 1 mg/ml poly-L-lysine (PLL, Sigma-Aldrich, catalog# P4707) for 30 min, rinsed with HEPES buffer (pH 8.3–8.6; Sigma-Aldrich, catalog# H3375), and incubated with freshly prepared 70 mg/ml PEG-SVA (Laysan Bio, catalog# SVA-PEG-5000) for 1 h, followed by three DI water washes.

For #1.5 Ibidi 8-well glass plates (Ibidi, catalog# 80807), volumes were downscaled according to the single-well volume, and UV-ozone exposure was replaced with 200 μl of 1M NaOH treatment for 30 min, to restrict substrate functionalization to individual wells.

ECM islands and validation designs

Square engineered ECM island designs of 625, 2500, and 10 000 μm^2^ were generated in Inkscape (version 1.3)63 and converted into 8-bit grayscale format with a white square island foreground. Validation images to test the Fab2Mic routine were also generated to test both the small design with array repetition and a large photopatterned region. For validation, a large region was generated from an 8-bit grayscale .tif image of the “Ponte Coperto in Pavia” [supplementary material Figs. S6(A) and S6(B)].

Photopatterning

35 mm round substrates were coated with a photoinitiator solution consisting of 66 μl ethanol (VWR, catalog# 20821.330) mixed with 4 μl PLPP gel (Alvéole, catalog# LPPG-100) and dried in the dark. For #1.5 8-well glass Ibidi plates, a surfactant solution (Surfactant Mix, Alvéole) was required to allow the photoinitiator to spread evenly across the entire well bottom and prevent meniscus formation. The solution was therefore prepared by mixing 97% (vol./vol.) DI water, 2.5% PLPP gel, and 0.5% Surfactant Mix. A volume of 150 μl was dispensed into each well and allowed to dry completely.

Photopatterning was performed using the PRIMO system (Alvéole)29 coupled to a Nikon Ti2 inverted epifluorescence microscope. The integrated 375 nm UV source and a Nikon CFI S Plan Fluor ELWD 20× (air objective, NA 0.45, WD 6.9–8.2, catalog# MRH08230) provided illumination. Selected designs were projected through the optical path using a Digital Micromirror Device (DMD), enabling the transfer of user-defined designs onto the substrate with a resolution of 0.28 μm/px. Fabrication layouts were tuned by selecting the array size (5 × 5 or 25 × 25) and inter-pattern spacing (50 or 150 μm). Pattern alignment and exposure control were managed through the Leonardo software package (Alvéole, v.5.2).64

ECM coating

Photopatterned substrates were incubated with 20 μg/ml FITC-fibronectin (Sigma-Aldrich, catalog# FNR01-A) in PBS for 30 min for validation tests or with human plasma fibronectin (Sigma-Aldrich, catalog# F0895) for 10 min. Three PBS washes removed excess coating.

Cell seeding on ECM photopatterned islands

HT1080 CALIPERS cells were seeded on pre-coated engineered ECM islands with human plasma fibronectin (Sigma-Aldrich, catalog# F0895) at a density adjusted for single-cell occupancy: 4 × array size for 625/2500 μm^2^ and 2 × array size for 10 000 μm^2^ patterns [supplementary material Figs. S6(C)–S6(E)]. Cells were dispensed in complete DMEM F-12 filled photopatterned Ibidi 8-well glass plates (Ibidi, catalog# 80807) or 35 mm #1.5 glass-bottom dishes (Ibidi, catalog# 81158). After 15 min of settling under the hood, cells were incubated at 37 °C and 5% CO_2_ for 2 h before replacement with fresh phenol-red-free DMEM F-12 for imaging.

Fab2Mic—Correlative fabrication-to-microscopy pipeline

A Python pipeline (Fab2Mic) was developed to translate photofabrication templates directly into imaging coordinates. The routine accepts as input a binary fabrication mask (.tif) and a photofabrication report (.xml). The xml file encodes essential metadata, including the scale factor, substrate position, grid size, spacing, and fabrication objective specifications. These parameters were parsed to reconstruct the layout and to generate coordinate maps compatible with microscope control software.

For high-resolution acquisitions (≤500 μm field of view), imaging positions were assigned to the centroid of each fabricated feature to preserve positional fidelity. For larger fields of view (>500 μm), redundancy was minimized by computing a tiling grid, with fields automatically centered and spaced by their dimensions to ensure complete coverage without overlap [supplementary material Figs. S6(A) and S6(B)]. The pipeline, therefore, enables direct one-to-one registration between fabrication and imaging, scaling seamlessly from single-feature to array-level designs without manual intervention.

Microscopy acquisitions

Acquisitions were performed on a widefield Nikon Ti2 inverted microscope with a CrestOptics X-Light V3 spinning disk confocal unit and an environmental chamber (Okolab Bold Line; 37 °C, 5% CO_2_, humidity). The microscope is equipped with a Nikon linear-encoder motorized stage with a Mad City Labs 100 μm range Z-piezo insert (Mad City Labs, catalog# NI-2-C312). Illumination was provided by the Lumencor Celesta light engine (405, 446, 477, 520, 546, 638, and 740 nm; up to 800 mW), routed through multiband dichroics and appropriate emission filters.

The dichroic mirror wheel was selectively positioned with either a hard-coated Full Multi Band Penta CELESTA -DA/FI/TR/Cy5/Cy7-A (Nikon, catalog# MXR00543) or a Full Multiband dual CELESTA -CFP/YFP-A (Nikon, catalog# MXR00544). Depending on the excitation line, the beam was subsequently filtered through either a Multiband Full Penta FF01-391/477/549/639/741 (Semrock, catalog# FL-416877) or a Full Multiband Dual FF01-449/520 (Semrock, catalog# FL-411981).

Emission was collected using single-band filters mounted in the wheel: FF01-484/561 (Semrock, catalog# FL-412124) for CFP, FF01-685/40–25 nm (Semrock, catalog# FL-011482) for the iRFP signal, FF01-595/31 (Semrock, catalog# FL-004391) for RFP, or FF01-511/20-25 (Semrock, catalog# FL-004306) for EGFP.

Fluorescence was captured with the same Teledyne Photometrics Kinetix CMOS camera (6.5 μm pixels; 16-bit; native resolution 3200 × 3200 pixels, cropped to 2700 × 2700 pixels). Acquisition and hardware synchronization were controlled through NIS-Elements AR (v.5.42.03).61 Data were stored as .nd2 files.

Low-resolution clone selection

Static widefield and brightfield acquisitions were performed using a CFI Plan Fluor DIC 10× (air objective, NA 0.3, WD 16 mm, catalog# MRH00105). Clone selection was run with the custom automated CALIPERS clonal screening JOBS in which each well was acquired as a large image of 6 × 6 fields within a CFI Plan Fluor DIC 10× (air objective, NA 0.3, WD 1.6 mm, catalog# MRH00105). Channel acquisition was sequentially ordered from 638 nm to 446 nm, with brightfield acquired last. The high-throughput JOBS covered 40 wells, with the shutter closed during stage movements. Illumination settings were:

• •iRFP (S/G2/M): 638 nm, exposure 300 ms, laser power 28.71 ± 3 mW
• •RFP (Tubulin): 546 nm, exposure 100 ms, laser power 15.79 ± 1.6 mW
• •EGFP (Actin): 477 nm, exposure 100 ms, laser power 15.21 ± 1.5 mW
• •CFP (G1): 446 nm, exposure 300 ms, laser power 3.12 ± 0.3 mW
• •Brightfield: Dia lamp exposure 5 ms

Fab2Mic validation imaging

Fab2Mic validation was performed on square ECM designs of 625, 2500, and 10 000 μm^2^, as well as on a photopatterned image of the “Ponte Coperto in Pavia.” Designs were photopatterned and FITC-fibronectin-coated. Widefield fluorescence microscopy was used to validate the algorithm using NIS-Elements software.61 Imaging coordinates were imported from the xml file returned by Fab2Mic. Acquisitions of the photopatterned FITC-fibronectin square arrays were performed with a Nikon CFI Plan Fluor DIC 10× (air objective, NA 0.3, WD 16 mm, catalog# MRH00105) and the 477 nm line set at laser power 60.2 ± 6 mW with a 30 ms exposure. Acquisitions of the “Ponte Coperto in Pavia” pattern were performed with a Nikon CFI Plan Apo 20× (air objective, NA 0.75, WD 1 mm, catalog# MRD00205) and the 477 nm line set at laser power 27.8 ± 2.8 mW with a 300 ms exposure.

Low-resolution static live imaging

Static widefield acquisitions were performed using both Nikon CFI Plan Fluor DIC 10× (air objective, NA 0.3, WD 16 mm, catalog# MRH00105) and Nikon CFI Plan Apo 20× (air objective, NA 0.75, WD 1 mm, catalog# MRD00205). Multi-point acquisitions were conducted sequentially from 638 nm to 446 nm, with the shutter closed during stage movements.

For 10× imaging, sequential four-channel acquisitions were carried out with excitation settings as follows:

• •iRFP (S/G2/M): 638 nm, 4 × 4 binning, exposure 100 ms, laser power 28.71 ± 3 mW
• •RFP (Tubulin): 546 nm, 2 × 2 binning, exposure 50 ms, laser power 15.79 ± 1.6 mW
• •EGFP (Actin): 477 nm, 2 × 2 binning, exposure 50 ms, laser power 15.21 ± 1.5 mW
• •CFP (G1): 446 nm, 4 × 4 binning, exposure 30 ms, laser power 3.12 ± 0.3 mW

For 20× imaging:

• •iRFP (S/G2/M): 638 nm, 4 × 4 binning, exposure 300 ms, laser power 21.4 ± 2.1 mW
• •RFP (Tubulin): 546 nm, exposure 50 ms, laser power 20.0 ± 2 mW
• •EGFP (Actin): 477 nm, exposure 50 ms, laser power 15.8 ± 1.6 mW
• •CFP (G1): 446 nm, 4 × 4 binning, exposure 20 ms, laser power 3.1 ± 0.3 mW

High-resolution static confocal Z-stack

High-resolution static confocal imaging was performed with a Nikon CFI SR HP Plan Apo Lambda S 100× silicone oil objective (Silicone oil immersion, NA 1.35, WD 0.31 mm, catalog# MRD73950). Illumination settings were:

• •iRFP (S/G2/M): 638 nm, 2 × 2 binning, exposure 2 s, laser power 5.61 ± 0.6 mW
• •RFP (Tubulin): 546 nm, exposure 300 ms, laser power 1.8 ± 0.2 mW
• •EGFP (Actin): 477 nm, exposure 300 ms, laser power 1.9 ± 0.2 mW
• •CFP (G1): 446 nm, 2 × 2 binning, exposure 30 ms, laser power 1.4 ± 0.1 mW

Z-stacks were acquired using an MCL Piezo Z-device by capturing each channel across a 10 μm range within a step size of 0.3 μm.

High-resolution static widefield imaging

Static widefield imaging was performed with a Nikon CFI SR HP Plan Apo Lambda S 100× silicone oil objective (Silicone oil immersion, NA 1.35, WD 0.31 mm, catalog# MRD73950), Nikon CFI Plan Apo Lambda S 40× silicone oil objective (Silicone oil immersion, NA 1.25, WD 0.30 mm, catalog# MRD73400).

For 100× imaging, sequential four-channel acquisitions were carried out with the following settings:

• •iRFP (S/G2/M): 638 nm, exposure 50 ms, laser power 14.8 ± 1.5 mW
• •RFP (Tubulin): 546 nm, exposure 30 ms, laser power 6.7 ± 0.7 mW
• •EGFP (Actin): 477 nm, exposure 30 ms, laser power 5.9 ± 0.6 mW
• •CFP (G1): 446 nm, exposure 2 ms, laser power 0.8 ± 0.1 mW

For 40× imaging, acquisitions were performed with:

• •iRFP (S/G2/M): 638 nm, exposure 100 ms, laser power 28.8 ± 3 mW
• •RFP (Tubulin): 546 nm, exposure 40 ms, laser power 14.8 ± 1.5 mW
• •EGFP (Actin): 477 nm, exposure 40 ms, laser power 13.5 ± 1.4 mW
• •CFP (G1): 446 nm, exposure 60 ms, laser power 2.2 ± 0.2 mW

Low-resolution dynamic live imaging

Long-term (48 h) widefield acquisitions were performed using a Nikon CFI Plan Fluor DIC 10× (air objective, NA 0.3, WD 16 mm, catalog# MRH00105). Multi-point time-lapse imaging was conducted sequentially from 638 to 446 nm, with the shutter closed during stage movements.

For 10× imaging, sequential four-channel acquisitions were carried out with excitation settings as follows:

• •iRFP (S/G2/M): 638 nm, 4 × 4 binning, exposure 100 ms, laser power 28.71 ± 3 mW
• •RFP (Tubulin): 546 nm, 2 × 2 binning, exposure 50 ms, laser power 15.79 ± 1.6 mW
• •EGFP (Actin): 477 nm, 2 × 2 binning, exposure 50 ms, laser power 15.21 ± 1.5 mW
• •CFP (G1): 446 nm, 4 × 4 binning, exposure 30 ms, laser power 3.12 ± 0.3 mW

The time series spanned 48 h with a frame interval of 30 min. Stage movements followed an optimized multi-point path. Nikon Perfect Focus remained engaged throughout overnight acquisitions to minimize drift. Samples were maintained in phenol-red-free DMEM F-12 medium.

Low-resolution dynamic live imaging

Long-term widefield acquisitions were performed using a Nikon CFI Plan Apo 20× (air objective, NA 0.75, WD 1 mm, catalog# MRD00205). Multi-point time-lapse imaging was conducted sequentially from 638 to 446 nm, with the shutter closed during stage movements.

For 20× imaging, sequential four-channel acquisitions were carried out with excitation settings as follows:

• •iRFP (S/G2/M): 638 nm, 4 × 4 binning, exposure 300 ms, laser power 21.4 ± 2.1 mW
• •RFP (Tubulin): 546 nm, exposure 50 ms, laser power 20.0 ± 2 mW
• •EGFP (Actin): 477 nm, exposure 50 ms, laser power 15.8 ± 1.6 mW
• •CFP (G1): 446 nm, 4 × 4 binning, exposure 20 ms, laser power 3.1 ± 0.3 mW

The time series spanned 15 h with a frame interval of 30 min. Stage movements followed an optimized multi-point path. Nikon Perfect Focus remained engaged throughout overnight acquisitions to minimize drift. Samples were maintained in phenol-red-free DMEM F-12 medium.

Image preprocessing and visualization

Raw acquisitions were denoised and deconvolved in NIS-Elements using the Richardson–Lucy algorithm (10 iterations). A custom Fiji65 macro was applied for visualization, including a rolling-ball background subtraction (50 px radius) and channel-specific intensity level subtraction (446 nm = 50; 477 nm = 20; 546 nm = 20; 638 nm = 100). Scale bars were set at 25 μm.

HT1080 CALIPERS CC-aware bioimage analysis

Static and dynamic microscopy datasets underwent a multi-step analysis pipeline to extract CC-aware dynamics, motility, and macro- and micro-scale morphology features.

Preprocessing

Low-resolution fluorescence acquisitions were first preprocessed by applying BaSiC flat-field correction66 to compensate for non-uniform illumination and background shading, yielding homogeneous image intensity. Denoising was performed using a Butterworth filter implemented in the dexp library (Royer Lab, GitHub).67,68 For all channels, parameters were calibrated on the first frame using calibrate_denoise_butterworth, and the resulting function was applied across all time points. Channel indexing and metadata were provided via a configuration file. Processed stacks were saved as 16-bit OME-TIFFs using AICSImageIO.69 Image manipulations were performed with scikit-image.70

Segmentation and tracking

Nuclear masks were generated with StarDist (version 0.8.3, pre-trained fluorescent nuclei model).31 Channel signals were merged into a single DAPI-equivalent image, from which segmentation masks were obtained. The resulting DAPIEq masks were used as labels for motility analysis. Cytoskeleton masks were generated with Cellpose v2.032 applied to RFP (tubulin) and EGFP LifeAct (actin) channels. The segmentation derived from cytoskeletal signals defined macro-scale morphological labels for HT1080 CALIPERS. Tubulin microstructures were labeled with Fiji Otsu-thresholding71 on background-subtracted (rolling ball 50 px) and Gaussian smoothed (σ = 2 px) RFP (tubulin) channel (supplementary material Fig. S7).

FUCCIphase with morphology cues

Cell-cycle phase assignment was performed using FUCCIphase,34 with newly calibrated intensity curves of the HT1080 CALIPERS cells. For each tracked nucleus, normalized CFP and iRFP mean intensities were measured within the DAPIEq mask. Phase classification was performed using fixed intensity thresholds derived from 24-h reference recordings, yielding G1 (CFP-high/iRFP-low) and S/G2/M (iRFP-high/CFP-low). Intermediate G1/S events were grouped with S/G2/M for the analyses. Time-resolved motility and morphology were computed by linking nuclear centroids in Fiji65 using the TrackMate plugin33 with LAP tracking.

The nuclear signal intensity of the two FUCCI channels was used for static and dynamic CC-phase assignment, and the centroid of the nuclear labels was used for cell motility parameter computation. Additionally, CC-related nuclear morphology features were directly computed from the nuclear labels [supplementary material Figs. S7(D)–S7(F) and S10]. In parallel, static or time-resolved cytoskeletal features were extracted from cytoskeleton mask centroids within the same pipeline. Additional morphological features, such as perimeter, circularity, and the ellipse aspect ratio (Ellipse AR), were evaluated [supplementary material Figs. S7(G)–S7(I) and S11]. The spread of tubulin microstructure was quantified through an automated pixel count macro applied to tubulin masks, with measurements converted to physical area (μm^2^).

For dynamic analyses, single-cell tubulin spread was computed per frame and paired with the concurrent FUCCIphase-derived cell track. To link nuclear information with cytoskeletal information, a custom MATLAB script (FUCCIphase_morphology, version R2023b, 23.2)72 was developed to map each nuclear centroid to its enclosing cytoskeletal mask centroid within a 20 μm search radius, minimizing incorrect associations. The output was a unified FUCCIphase_morphology table (per-cell, per-frame for dynamic analyses) containing time stamps, positions, phase labels, cell-scale morphology, and tubulin spread. The same pipeline was applied to both static and dynamic datasets, acquired under both free space and confined conditions.

Static HT1080 CALIPERS CC-aware phenotyping

For the free space condition, 5 × 10^4^ cells were seeded on two 35 mm #1.5 glass-bottom dishes (Ibidi #81158) pre-coated with 20 μg/ml human plasma fibronectin (Sigma-Aldrich, catalog# F0895) in standard supplemented DMEM F-12 without phenol red medium for Early (4-h post seeding) vs Late (24-h post seeding) time point evaluations. For confinement conditions, cells were seeded according to the engineered ECM island size and array dimension. See the section on Photofabrication for details on engineered ECM island seeding. In both conditions, the medium was refreshed 2 h before the imaging session. Three independent high-resolution widefield fluorescence experiments were performed with a CFI SR HP Plan Apo Lambda S 100× silicone oil objective (Silicone oil immersion, NA 1.35, WD 0.31 mm, catalog# MRD73950). Static datasets were analyzed with the HT1080 CALIPERS CC-aware bioimage analysis pipeline and stratified by cell-cycle phase (G1 = CFP; S/G2/M = iRFP) to extract principally the average cell-cycle phase distribution, cell area, and tubulin spread per experiment [Figs. 2(d)–2(f) and 4(d)–4(f)]. Additional information on nuclear morphology and cytoskeleton assembly is also quantified (supplementary material Fig. S7).

Static HT1080 CALIPERS population-scale CC phase distribution

To evaluate whether the FUCCIplex-defined cell-cycle phase distribution inferred from the baseline characterization is representative at the population level, a quantification of FUCCIplex phases in free-space conditions was performed. Ten independent FOVs were acquired using a 10× objective in fluorescence mode at low resolution across a large field of view [supplementary material Fig. S2(A)]. A total of n = 555 cells were quantified by classifying each nucleus as G1 (CFP^+^) or S/G2/M (iRFP^+^). Images were processed in Arivis Pro (Arivis Pro, v. 4.40),73 where nuclei were segmented and assigned to a FUCCIplex phase based on nuclear CFP and iRFP signal thresholds. The resulting phase fractions were aggregated across FOVs to yield a population-level estimate of the CC phase distribution [supplementary material Fig. S2(E)].

Dynamic HT1080 CALIPERS CC-aware phenotyping

For free space condition, 5 × 10^4^ cells were seeded on 35 mm #1.5 glass-bottom dishes (Ibidi #81158) pre-coated with 20 μg/ml human plasma fibronectin (Sigma-Aldrich, catalog# F0895) in standard supplemented DMEM F-12 without phenol red. For confinement conditions, cells were seeded according to the square ECM size and array dimension. See the section on Photofabrication for details on ECM island seeding. In both conditions, the medium was refreshed 2 h before the imaging session in each case. Three independent dynamic low-resolution fluorescence experiments were performed with a CFI Plan Apo 20× (air objective, NA 0.75, WD 1 mm, catalog# MRD00205). Dynamic datasets were analyzed with the HT1080 CALIPERS CC-aware bioimage analysis pipeline.

CC-labeled tracks were computed using a custom MATLAB script (v.R2023b, 23.2)72 that generated color-coded CC on fluorescence images and extracted CC-aware motility metrics over time, including velocity, cell area, and tubulin spread [Figs. 2(h)–2(j) and 4(j)–4(l)]. Comprehensive CC-related nuclear and cytoskeleton morphology features were evaluated (supplementary material Figs. S10 and S11). Additionally, cell-cycle phase duration, total path length, and net displacement were quantified for each cell-cycle phase (G1 = CFP; S/G2/M = iRFP).

Dynamic HT1080 CALIPERS 48 h CC-aware phenotyping at the population level

To strengthen the time-dependent analysis with a larger sample size and capture dynamic CC progression, extended (48 h) low-magnification, multi-channel live imaging over a large field of view was performed. Cells were tracked, and a subset of trajectories with reliable single-cell segmentation and complete FUCCIplex phase transitions was retained for analysis. Using Arivis Pro, a large dataset of per-cell instantaneous velocity (μm/min) and projected area (μm^2^) was extracted, and cell-cycle-aligned profiles were computed on a common 0%–100% average cell-cycle axis by normalizing each trajectory to its FUCCIplex-defined phase durations. This yielded population-level time-dependent curves computed over n = 50 complete cell cycles; the data in supplementary material Figs. S3(D) and S3(E) are reported as mean ± SEM across tracked cell profiles.

FUCCIplex cell-cycle-normalized population-level analysis

To compare motility and morphology across several cells with variable cycle lengths, a population-level analysis was performed using the same dynamic analysis's MATLAB script, in which trajectories were normalized to a common cell-cycle axis. For each trajectory, time was aligned to the first time point, and phase windows (G1 and S/G2/M) were identified to measure their respective durations. Population means ± SD of phase durations were computed, and a linear time-warping procedure was applied. The G1 and S/G2/M segments of each trajectory were independently rescaled to match phase-specific population windows defined as the mean + 2SD for each phase, used here as a robust inclusion window to accommodate inter-cell variability while limiting the influence of rare, outlier-long phases on the normalization axis. The resulting normalized axis was subsequently mapped back to an “average cell-cycle time” (minutes), defined as the sum of the mean G1 and mean S/G2/M durations. On this common axis, population-average instantaneous velocity (μm/min) and projected cell area (μm^2^) were calculated, yielding phase-colored (G1 vs S/G2/M) profiles that represent the average phenotypic evolution across a FUCCIplex-defined cell cycle (supplementary material Fig. S3).

Abnormal cell-cycle events analysis

Cell-cycle abnormalities were manually annotated across the three analyzed experiments of dynamic HT1080 CALIPERS CC-aware phenotyping. Five categories were defined. To classify abnormal CC events, “Long G1” (>9 h) and “Long S/G2/M” (>10 h) were pragmatically defined using a distribution-based threshold, set at two standard deviations above the mean of the free space distribution. This approach follows precedent from single-cell timing studies that applied distribution thresholds for outlier detection.37 Additional abnormal events included “Long Mitosis,” defined as persistence in the M phase beyond the expected 1-h duration. Finally, rare events of direct S/G2/M–G1 reentry without mitosis (slippage) were identified exclusively under confinement. To minimize the risk of annotation errors, trajectories were manually curated, and only cases confirmed by visual inspection were retained.

Statistical analysis

All statistical analyses were performed using GraphPad Prism v10.4.274 and SigmaPlot v14.75 Normality was assessed with the Shapiro–Wilk test, and equal variance with the Brown–Forsythe test. For parametric data, two-way ANOVA with Tukey's post hoc correction was applied; for nonparametric distributions, Kruskal–Wallis tests were used. All static and dynamic phenotyping assays were performed in three independent experiments (n = 3), each derived from separate cell passages and sample/substrate preparations, representing both biological and technical replicates. Unless noted, values are reported as mean ± SD. Exceptionally, for cytoplasmic vs nuclear CFP/iRFP localization, values were averaged per field of view (FOV) and reported as mean ± SEM across FOVs; for population-level profiled plots, data are likewise displayed as mean ± SEM across cell tracks (e.g., supplementary material Fig. S3). For abnormal CC event frequencies [Fig. 4(i)], percentages were calculated over the total number of tracked cells pooled within each confinement condition. A threshold of p < 0.05 was considered statistically significant. Graphical annotations in figures indicate significant comparisons (^*^p < 0.05).

SUPPLEMENTARY MATERIAL

See the supplementary material for Figs. S1–S12 and Movies S1–S3.