Development of a quantitative Cellular AGing Evaluation system

While 4ME substrate–based assays have been proposed for quantifying cellular senescence at 96- or 384-well HTS scales [19, 20], these methods also fail to account for cell number variations, potentially leading to misinterpretation of the anti-aging-like efficacy of certain compounds.

To address these limitations, we developed a qCAGEs, a dual-parameter microwell-based screening platform that simultaneously measures aging-associated enzymatic activity and cell viability (Fig. 1A). In this system, adherent cells or organoids were treated with compounds in either 2D or 3D formats, followed by parallel quantification of α-L-fucosidase activity (4ME readout) and cell number (nuclear staining for 2D; ATP-based viability for 3D). These paired measurements were integrated into a two-dimensional coordinate system, enabling the precise positioning of each drug response. To directly evaluate the aging-associated enzyme activity relative to cell number, we introduced the Cell count–Aging activity Reference Line (CARL (yref)), which represents the expected relationship between cell count and 4ME activity in untreated controls. By comparing each sample’s position relative to the initial cell count at day 0 (×0) and the CARL, drug responses were classified into four operational phenotypic response classes: anti-aging-like, pro-aging-like, senolytic-like, and cytotoxic. Furthermore, we developed a quantitative Cellular Aging Index (qCAI) to quantify the magnitude of aging-modulatory effects within each classification category.

Fig. 1

Schematic overview of the quantitative Cellular AGing Evaluation system (qCAGEs) workflow. A Adherent cells or organoids are treated with compounds in either two-dimensional (2D) or three-dimensional (3D) formats, followed by quantification of α-L-fucosidase activity (4ME) and cell number (nuclear staining for 2D; ATP-based viability for 3D). These dual-parameter measurements were integrated into the qCAGEs, enabling the positioning of each drug. The Cell count–Aging activity Reference Line (CARL) represents the relationship between the cell count and 4ME activity in untreated controls. By comparing the each sample relative to the cell count on day 0 (x0) and the CARL (yref), drug responses were classified into four categories. B Schematic overview of the qCAGEs classification and quantification system. Each drug response is positioned by its cell count (xi) and 4ME activity (yi) relative to the initial cell count (x0) and CARL (yref = mxi + b). Responses are classified into four categories: anti-aging-like (xi ≥ x0, yi < yref), pro-aging-like (xi ≥ x0, yi > yref), senolytic-like (xi < x0, yi < yref), and cytotoxic (xi < x0, yi > yref). The quantitative Cellular Aging Index (qCAI) is calculated as the perpendicular distance from each response coordinate to the CARL

In the classification step (Fig. 1B), drug responses were categorized based on the position of each response coordinate (xi, yi) relative to two references: the initial cell count on day 0 (x0) and CARL (yref = mxi + b). First, samples were stratified by viability: those with xi ≥ x0 (maintained or increased cell number) were further evaluated for aging-associated enzymatic activity relative to the CARL. Samples positioned below CARL (yᵢ < y_ref) were assigned to the anti-aging-like class, defined operationally as a reduced 4ME signal relative to the CARL-predicted value without loss of cell viability. Samples positioned above CARL (yᵢ > y_ref) were assigned to the pro-aging-like class, defined operationally as an elevated 4ME signal relative to the CARL-predicted value. Samples with reduced viability (xᵢ < x0) were similarly subdivided: those below CARL were assigned to the senolytic-like class—a profile compatible with preferential loss of senescent cells but not in itself proof of selective senescent cell elimination [7,8,9]; those above CARL were assigned to the cytotoxic class, defined operationally as cell loss accompanied by a 4ME signal that remains disproportionately high, a profile compatible with non-selective cytotoxicity. We emphasize that these are phenotypic response classes defined relative to the empirical CARL reference, rather than direct mechanistic identification.

Establishment and validation of cellular senescence models

To prepare biological models for the dual-parameter aging assay, we established two senescence models using HDFs (Fig. 2A). For the RS model, HDFs were serially passaged beyond passage 42. For the DDIS model, young HDFs were treated with low-dose doxorubicin (0.5 µM for 24 h) [29, 30]. SA-β-gal staining confirmed a marked increase in senescent cells in both the RS and DDIS models compared to young HDFs (mean SA-β-gal-positive cells: Young, 6.95%; RS, 58.75%; DDIS, 75.35%; p = 0.0079 for Young vs. RS; p = 0.0079 for Young vs. DDIS) (Fig. 2B). Cellular ATP levels measured on days 0 and 7 demonstrated a significant reduction in proliferative capacity in both senescence models compared to young controls (Fig. 2C). Consistently, qRT-PCR analysis revealed significant upregulation of CDKN1A (p21) expression in RS and DDIS cells relative to young controls (p = 0.00016 for both comparisons) (Fig. 2D) [31]. To place the 4ME readout on equal footing with the classical senescence markers in the model-validation step, we present the 4ME level in the same Young, RS, and DDIS HDF models (Fig. 2E). The 4ME value was significantly elevated in RS (vs Young, p = 0.029) and DDIS (vs Young, p = 0.029; Wilcoxon rank-sum test), demonstrating that 4ME discriminates senescent from young HDF cells with a magnitude comparable to SA-β-gal and p21, and validating its suitability as the principal aging-associated readout of the qCAGEs framework.

Fig. 2

Establishment of cellular senescence models in human dermal fibroblasts (HDFs). A Schematic of replicative senescence (RS) and DNA damage–induced senescence (DDIS) induction protocols, with representative senescence-associated β-galactosidase (SA-β-gal) staining images. RS was induced by serial passaging beyond passage 42, and DDIS was generated by treatment with 0.5 µM doxorubicin for 24 h. B Quantification of SA-β-gal-positive cells in the Young, RS, and DDIS models. P value was obtained using the Wilcoxon rank-sum test (n = 5). C Cellular ATP levels in the Young, RS, and DDIS models on days 0 and 7, demonstrating reduced proliferative capacity in senescent cells. D Quantitative real-time polymerase chain reaction (qRT-PCR) analysis of CDKN1A (p21) expression in the Young, RS, and DDIS models. E Cellular 4ME levels in the Young, RS, and DDIS models, demonstrating elevated 4ME in RS and DDIS models, compared to young HDFs. P value was calculated using the Wilcoxon rank-sum test (n = 9)

Assay reproducibility and robustness validation

Based on these findings, we sought to develop an integrated HTS system that simultaneously considers both 4ME activity and cell number for cellular aging evaluation. To validate the assay performance, we assessed the correlation among 4ME, ATP, and cell count measurements across a range of cell densities (250–4000 cells/well, twofold range) in a 384-well format (Fig. 3A). Strong positive correlations were observed between 4ME and cell count (Spearman ρ = 0.97, p < 2.2 × 10−16) (Fig. 3B), and between 4ME and ATP (Spearman ρ = 0.96, p < 2.2 × 10−16) (Fig. 3C), indicating that the 4ME values exhibited robust correlations with both viability indicators.

Fig. 3

Performance validation of the dual-parameter assay for quantifying cellular aging response. A Pairwise correlation matrix among 4ME, ATP, and cell count measurements. Lower panels display smoothed scatter plots; upper panels report Spearman rank correlation coefficients (ρ). B and C Scatter plot showing Spearman’s correlation between 4ME and cell count (B), and 4ME and ATP levels (C). D Coefficient of variation (CV, %) for 4ME, cell count, and ATP measurements across five cell density conditions. The dashed line indicates a 15% threshold. E Spearman’s correlations of 4ME values measured using black clear-bottom versus white opaque 384-well plates under Young (left), RS (middle), and DDIS (right) conditions

Assay precision was evaluated by calculating the coefficient of variation (CV) for 4ME, cell count, and ATP across the five cell density conditions. All CV values remained below 15% at densities of 250 cells/well or greater (Fig. 3D), demonstrating acceptable precision for HTS applications [11]. To identify the optimal plate format for 4ME measurements, we compared fluorescence readings between black clear-bottom and white opaque 384-well plates across a range of cell densities (375–3000 cells/well). Strong correlations were observed between plate types for all models (Young: Spearman ρ = 0.96, p = 6.24 × 10−12; RS: Spearman ρ = 0.96, p = 1.2 × 10−13; DDIS: Spearman ρ = 0.94, p = 9.46 × 10−12) (Fig. 3E). These results demonstrate that the 4ME assay exhibits consistent performance across different plate formats, providing flexibility in the experimental design without compromising data quality.

Validation of cell number–normalized aging index and identification of single-parameter limitations

Following model establishment, we performed chemical screening using 39 compounds in the RS model. Both 4ME activity and cell count were measured in parallel, enabling a direct comparison of aging-associated enzymatic activity and cell viability. Drug responses were normalized to the RS control and plotted as deviations from reference values (Fig. 4A and B). Based on the 4ME measurements, 20 compounds (51%) exhibited increased values compared to the RS control, whereas 19 compounds (49%) showed decreased values (Fig. 4A). For cell count measurements, three compounds (8%) demonstrated increased values, whereas 36 compounds (92%) showed decreased values relative to the RS control (Fig. 4B).

Fig. 4

Validation of cell number–normalized aging index and single-parameter drug response classification. A Waterfall plot showing the deviation of normalized 4ME levels from the RS control for each compound. Bars above the dashed line indicate increased 4ME activity relative to the control, and bars below indicate reduced activity. B Waterfall plot showing the deviation of the normalized cell count from the RS control. Positive deviations indicated increased cell numbers, whereas negative deviations indicated reduced viability. C Sankey diagram illustrating the classification of drug responses based on combined 4ME and cell count deviations. Responses were categorized as 4ME up/cell number up (pattern #4), 4ME up/cell number down (pattern #2), 4ME down/cell number up (pattern #3), or 4ME down/cell number down (pattern #1). The line width represents the proportion of compounds in each category. D Stability of cell number–normalized aging metrics across seeding density conditions in 384-well microplates. Ratios of 4ME/Count (left panel) and 4ME/ATP (right panel) were calculated across three representative cell seeding densities (250, 500, and 1000 cells/well) spanning the operational range of the platform. Statistical significance was calculated using the Wilcoxon rank-sum test. E Discrimination of senescence states using the cell number–normalized aging index (4ME/Count ratio). Relative 4ME/Count ratios were quantified in three cellular states: young proliferative HDFs (baseline, normalized to 1.0), replicative senescence (RS, high-passage cells), and DNA damage–induced senescence (DDIS, doxorubicin-treated cells). Statistical significance was calculated using the Wilcoxon rank-sum test (n = 6 per group). F Bar plot showing the 4ME/cell count ratio for the 39 tested compounds and the RS control. The upper and lower dashed lines represent the RS control ratio ± 15% (i.e., 0.85× and 1.15× of the RS control value), defining the neutral classification window. Red bars indicate pro-aging-like compounds (ratio above the upper threshold), blue bars indicate anti-aging-like compounds (ratio below the lower threshold), and white bars indicate neutral compounds (ratio within the ± 15% window of the RS control). G Bar graphs (left panel) showing 4ME/cell count ratio (upper) and cell count quantified by DAPI staining (lower) for RS cells at day 0 and day 14, and after treatment with indicated drugs. Cell count changes were quantified from three independent wells and statistical significance was calculated using the Wilcoxon rank-sum test. Representative DAPI fluorescence images for each group are shown in the right panel. Scale bar = 200 µm

When both parameters were integrated, drug responses were categorized into four patterns based on binary outcomes (Up or Down relative to RS control) for each measurement. Pattern #1 (4ME Down/Count Down) comprised 18 compounds (46%); Pattern #2 (4ME Up/Count Down) included 17 compounds (44%); Pattern #3 (4ME Down/Count Up) contained one compound (2%); and Pattern #4 (4ME Up/Count Up) encompassed three compounds (8%) (Fig. 4C). Notably, 46% of the compounds exhibited discordant patterns (#2 and #3), where 4ME and cell count showed opposing trends. These discordant phenotypes demonstrate that relying on a single assay (4ME or cell count alone) is insufficient to accurately interpret drug-induced aging responses, underscoring the necessity of a dual-parameter screening approach [11].

A critical requirement for robust senescence quantification in HTS applications is that the biomarker signal reflects per-cell aging status rather than simply being proportional to total cell number [11, 19]. Therefore, we sought to validate whether the 4ME signal, when normalized to cell number, provides a density-independent measure of per-cell aging status that enables robust discrimination between young and senescent populations across different seeding conditions.

To examine whether the aging status per cell remained stable across different seeding densities in the 384-well microplate format, we calculated the ratios of 4ME/Count and 4ME/ATP (Fig. 4D) across three representative cell density conditions (250, 500, and 1000 cells/well) that span the operational range of the platform. Both ratios remained highly consistent across all tested conditions (coefficient of variation < 10% for both metrics), confirming that the 4ME signal reflects aging capacity on a per-cell basis rather than being influenced by total cell number. The stability of the 4ME/Count and 4ME/ATP ratios across different densities validates the use of these normalized indices as quantitative metrics of per-cell senescence burden, independent of population size.

Given the strong positive correlation between 4ME activity and cell number demonstrated in Fig. 3B and C (Spearman ρ > 0.96), we hypothesized that a cell number–normalized index (4ME/Count ratio) would more robustly represent cellular aging status by accounting for variations in cell density while preserving biological signal. To test this hypothesis, we compared the 4ME/Count ratios across three cellular states: young HDFs (low passage, proliferative), replicative senescence (RS, high passage), and DNA damage–induced senescence (DDIS, doxorubicin-treated) [29, 30]. Analysis revealed that the 4ME/Count ratio was significantly elevated in RS cells (6.33-fold increase relative to young controls; p = 0.0028, Wilcoxon rank-sum test) and DDIS cells (6.58-fold increase; p = 0.0028) (Fig. 4E), demonstrating that this normalized metric provides robust quantitative discrimination of senescence status [32].

Based on these findings, we further evaluated whether the 4ME/cell number ratio could discriminate drug-induced aging status changes using single-parameter measurement. Analysis of the 39-compound screening dataset using this ratio-based approach revealed that KU-60019 and dabrafenib exhibited anti-aging-like effects, whereas the majority of compounds (32 out of 39, 82%) showed pro-aging-like patterns (Fig. 4F). However, this single-parameter ratio-based approach exhibited substantial limitations in distinguishing genuine aging effects from cytotoxicity when temporal changes in cell viability were not accounted for. For instance, four compounds—oxaliplatin, vemurafenib, ulixertinib, and vismodegib—all exhibited 4ME/cell number ratios around 4000, suggesting increased aging compared to RS control cells (Fig. 4G). Critically, oxaliplatin and ulixertinib demonstrated statistically significant reductions in cell count relative to day 0 baseline (oxaliplatin: p = 0.031; ulixertinib: p = 0.0002), indicating that the elevated 4ME/cell number ratio likely reflects cytotoxic effects rather than genuine pro-aging activity. These findings demonstrate that while the 4ME/cell number ratio successfully discriminates between young and senescent cell populations, single-parameter ratio measurements fail to accurately capture the dynamics of drug-induced aging responses, particularly in distinguishing pro-aging effects from cytotoxicity-driven ratio increases.

CARL-based classification and quantitative indexing of drug responses

To overcome these limitations, we developed a dual-parameter framework that (i) incorporates cell number dynamics before and after drug treatment and (ii) employs distance-based analysis from a Cell count–Aging activity Reference Line (CARL) derived from multiple cell densities, rather than simple comparison to control cells. Using this framework, cell count and 4ME data from 39 drug-treated samples on day 14, along with senescence model controls (Young and DDIS), were visualized on a quadrant map (Fig. 5A).

Fig. 5

CARL-based classification and quantitative indexing of drug responses. A Quadrant map displaying drug responses (39 compounds) on day 14, with Young and DDIS controls. The shaded regions denote the four CARL-defined operational phenotypic response classes. B Representative SA-β-gal staining images and quantification of RS control and selected compounds: KU-60019 (anti-aging-like), folinic acid, and vemurafenib (pro-aging-like). P values were obtained using the Wilcoxon rank-sum test (n = 5). C Within-class qCAI ranking of the 39 tested compounds, grouped by CARL-based phenotypic class and ordered by increasing qCAI from left to right. 4ME activity and cell count are shown alongside qCAI for each compound. Exact values are listed in Supplementary Table 4

Among the tested compounds, 27 (69%) maintained or increased cell counts relative to day 0 (xi ≥ x0), whereas 12 (28%) showed reduced viability (xi < x0). Within the viable population, 26 compounds (96%) were classified as pro-aging-like (above CARL), and one compound (4%) was classified as anti-aging-like (below CARL). Among the compounds with reduced viability, 10 (83%) were classified as cytotoxic and 2 (17%) as senolytic-like. Young HDF cells were positioned in the anti-aging-like region. The DDIS model, although it exhibited a robust senescent phenotype as assessed by SA-β-gal, p21, ATP, and 4ME (Fig. 2), was also localized within the anti-aging-like region of the RS-referenced CARL map (Fig. 5A)—reflecting the fact that DDIS cells, despite being senescent, have a lower 4ME activity per cell than the RS-derived reference at matched cell number, consistent with known heterogeneity in 4ME and SA-β-gal expression across senescence inducers and cumulative damage exposure. This observation directly illustrates a key interpretive property of the framework: CARL-defined classes are relative phenotypic regions defined by deviation from the chosen reference state (here, RS HDF), not absolute biological categories.

To validate our classification system, we performed SA-β-gal staining following treatment with representative compounds from each category [15]. KU-60019, an ATM inhibitor previously reported to ameliorate senescence phenotypes [33], was classified as an anti-aging-like agent and demonstrated significantly reduced SA-β-gal positivity (p = 0.0079). In contrast, compounds classified as pro-aging-like, including folinic acid (p = 0.032) and vemurafenib (p = 0.056, trending toward significance), increased SA-β-gal staining (Fig. 5B). These results confirm the biological validity of the qCAGEs classification system.

Quantitative analysis using qCAI revealed substantial variation in the effect magnitude within each category (Fig. 5C). KU-60019 exhibited the strongest anti-aging-like effect, whereas lenvatinib demonstrated modest senolytic-like activity. Among the pro-aging-like compounds, sotorasib showed the highest qCAI value, whereas irinotecan exhibited the strongest cytotoxic effect (Supplementary Table 4). These results demonstrate that the qCAI provides a quantitative metric for ranking compound potency within each classification category, information that cannot be obtained from 4ME or cell count measurements alone.

Micropillar-based 3D platform adaptation for organoid-compatible senescence screening

Recent advances in organoid technology have generated an increasing demand for senescence-modulating drug screening platforms compatible with 3D culture systems [22,23,24]. However, implementing HTS-scale senescence screening in 3D formats presents unique technical challenges that have limited widespread adoption. First, accurate cell enumeration within 3D matrices remains difficult, as conventional nuclear staining–based counting methods are incompatible with dome-shaped constructs and matrix-encapsulated cells [22, 24, 25]. Second, automated dispensing of cell–matrix suspensions in sub-microliter volumes with acceptable precision poses significant technical barriers for miniaturized HTS formats [25, 34]. Third, biochemical readouts compatible with 3D culture require either complete construct lysis or the use of viability indicators that can penetrate matrix environments, adding complexity to assay workflows [24, 34]. These limitations have hindered the development of robust 3D senescence screening platforms, despite the growing demand for such systems in drug discovery [11, 22, 25]

To address these challenges, we adapted our dual-parameter screening platform to a 3D culture format using 384-well micropillar plates [26,27,28]. The micropillar platform design incorporates several key engineering features that enable HTS-compatible 3D screening, providing a biochemical readout compatible with matrix-encapsulated cells [26, 28, 34]. Cell–matrix suspensions were precisely dispensed onto pillar plates and overlaid with a drug-containing medium, followed by lysis, substrate reaction, and fluorescence readout, analogous to the 2D workflow (Fig. 6A).

Fig. 6

Micropillar-based adaptation of the dual-parameter platform to 3D cultures. A Schematic workflow of the 3D screening platform. Cell–matrix suspensions were precisely dispensed onto 384-well micropillar plates and overlaid with a drug-containing medium. Following incubation, pillar plates were coupled with bottom plates to enable automated liquid handling, then cells were lysed on-pillar, and 4ME substrate was added. The reactions were terminated with pH 9.0 stop buffer, and fluorescence was measured. ATP quantification was used as a viability parameter compatible with matrix-encapsulated cells. B Pairwise correlation matrix between 4ME and ATP measurements in 3D cultures. Lower panels display smoothed scatter plots; upper panels report the Spearman correlation coefficients (ρ). C Scatter plot showing the Spearman correlation between 4ME and ATP across multiple cell density conditions. D Ratio of 4ME/ATP across different cell seeding densities (100–800 cells/well), demonstrating stable per-cell aging detection in the micropillar-based 3D format. The Wilcoxon rank-sum test (n = 6 per group). E Coefficient of variation (CV, %) for 4ME and ATP measurements across four cell density conditions. The dashed line indicates a 15% threshold, demonstrating miniaturized HTS format compatibility. F CARL-based quadrant map of chemical responses on day 14, with Young and DDIS control. The shaded regions represent the four response categories enabled by the micropillar platform design. G Cosine similarity analysis comparing 2D and 3D drug response profiles. Compounds with cosine similarity > 0.5 indicate high platform concordance; values between 0 and 0.5 indicate moderate concordance; and negative values indicate opposing effects between microplate and micropillar platforms

As nuclear staining–based cell counting is not feasible in 3D dome structures, ATP quantification was employed as the viability parameter for the dual-parameter framework. Consistent with our 2D findings, 4ME and ATP measurements exhibited a strong positive correlation in the 3D system (Spearman ρ = 0.94, p = 2.1 × 10−6) (Fig. 6B and C). The 4ME/ATP ratio remained highly consistent across cell densities ranging from 100 to 800 cells/well (Fig. 6D), confirming reliable per-cell aging detection in the 3D format of the assay. Precision analysis demonstrated that the CV values for 4ME remained below 15% at seeding densities of 200 cells/well or higher (Fig. 6E), establishing an acceptable operating range for 3D screening applications.

Drug responses in 3D-cultured HDFs on day 14 were visualized on a quadrant map using the qCAGEs framework (Fig. 6F). Among the tested compounds, 32 (82%) maintained or increased cell viability relative to day 0, whereas 7 (18%) showed reduced viability. Within the viable population, 31 compounds (97%) were classified as pro-aging-like, and one compound (3%) was classified as anti-aging-like. Among the compounds with reduced viability, seven (100%) were classified as cytotoxic and 0 (0%) as senolytic-like.

To evaluate platform concordance, cosine similarity was calculated between the 2D (4ME + Count) and 3D (4ME + ATP) drug response profiles for each compound (Fig. 6G). The majority of compounds (30, 77%) exhibited high similarity between platforms (cosine similarity > 0.5), indicating the robust conservation of aging response profiles across culture formats. However, five compounds (13%), including sotorasib, demonstrated moderate concordance (cosine similarity 0–0.5), whereas four compounds (10%), including carboplatin, showed negative cosine similarity values, indicating opposing effects between the 2D and 3D conditions. Three factors should be considered when comparing 2D and 3D qCAGEs outputs. First, in 3D the viability axis is operationally redefined: ATP serves as a metabolic surrogate for cell abundance rather than a direct cell count, so per-cell metabolic shifts under drug treatment can move 3D coordinates independently of true cell number changes. Second, the observed 2D–3D phenotypic concordance is substantial but not uniform (77% concordant, 13% moderate, 10% opposing; Fig. 6G), indicating that a clinically meaningful minority of compounds behave differently in matrix-embedded culture. Third, no compound was classified as senolytic-like in the 3D format. This may reflect (a) compound-specific differences in matrix permeability that attenuate cytotoxic exposure in 3D; (b) matrix-derived survival or anti-apoptotic signals that protect senescent cells in 3D; or (c) ATP-signal compression at low cell numbers, which can obscure the disproportionate-loss pattern that defines the senolytic-like region. Collectively, these results demonstrate that the qCAGEs platform is technically transferable to 3D culture formats with acceptable analytical robustness, while its biological outputs in 3D should be interpreted as a related but not equivalent phenotypic classification—expanding applicability to more physiologically relevant screening contexts but requiring orthogonal validation, particularly for senolytic-like classification, before mechanistic conclusions are drawn.