https://www.sciencedirect.com/science/article/pii/S2589004226001768
chatGPT(5.5paid):
Paper
Passanisi V, Spencer SL. “Replicative senescence induction in single cells is not predicted by telomere length, dysfunction, or oxidation.” iScience, 2026.
Overall assessment
This is a technically strong and conceptually important single-cell study. Its central finding is not that telomeres are irrelevant to replicative senescence, but that the telomere measurements currently available—average or distributional telomere length, telomere-associated 53BP1, and telomeric 8-oxoguanine—are poor predictors of which individual fibroblast is close to permanent cell-cycle withdrawal.
The study therefore challenges a simple, deterministic version of the telomere-driven Hayflick-limit model. It instead supports a model in which replicative senescence emerges gradually from multiple interacting cellular changes, including altered cell-cycle control, lysosomal/autophagic activity, chromatin organisation, cell enlargement and p21 expression.
1. Summary
Research question
The conventional model proposes that replicative senescence is triggered when one or more telomeres become critically short or dysfunctional, activate a DNA-damage response and induce p53–p21-mediated cell-cycle arrest.
The authors ask a more precise question:
In a heterogeneous population of ageing cells, can telomere length, telomere-associated DNA damage or telomere oxidation identify which individual cells are closest to senescence?
This differs from asking whether telomeres shorten on average with passage number. The study tests whether telomere features predict the behaviour of the same individual cell.
Experimental strategy
The authors combine:
- long-term live-cell imaging;
- single-cell lineage tracking;
- a fluorescent CDK2 activity reporter;
- post-imaging immunofluorescence;
- telomere quantitative FISH;
- three-dimensional confocal imaging;
- computational matching of cells across different microscopes.
The CDK2 reporter distinguishes cells actively cycling from cells in a CDK2-low state. The authors use the cumulative time spent CDK2-low during a 72–96-hour observation period as a measure of proximity to senescence.
They study several fibroblast systems:
- Hs68 human dermal fibroblasts at early, intermediate and late population doublings.
- Fibroblasts obtained from a 79-year-old donor.
- IMR90 lung fibroblasts aged under 3% oxygen.
- Additional longitudinal fibroblast samples from donors in the Baltimore Longitudinal Study of Aging for the oxidation experiments.
Main findings
A. Entry into senescence is gradual rather than an abrupt switch
As fibroblasts age, they do not simply divide normally until suddenly becoming permanently arrested. Instead:
- the probability of immediately recommitting to another cell cycle after mitosis declines;
- cells spend progressively longer intervals in the CDK2-low state;
- fast-cycling, slow-cycling and non-cycling cells coexist within early-passage populations;
- conventional senescence-associated features increase gradually with CDK2-low duration.
The heatmaps and lineage traces in Figure 1 illustrate a progressive lengthening of non-cycling intervals rather than a uniform binary transition.
The authors therefore propose a “gradual induction” model, in which cells move along a continuum from frequent proliferation through increasingly prolonged quiescence to effectively irreversible senescence.
B. Conventional senescence markers correlate with withdrawal, but none is definitive
The authors examine:
- SA-β-galactosidase;
- LAMP1;
- p21;
- nuclear area;
- cytoplasmic area;
- Lamin B1;
- variation in Hoechst DNA staining;
- nuclear 53BP1.
These features differ on average among fast-, slow- and non-cycling cells, but their distributions overlap substantially.
Several nevertheless show moderate discriminatory ability, with receiver-operating-characteristic areas under the curve of approximately 0.66–0.84. Lysosomal content, cell size, chromatin architecture and p21 perform better than the measured telomere features.
This supports the view that senescence cannot be identified reliably by a single marker.
C. Telomere-length distributions do not predict senescence proximity
For each tracked cell, the authors calculate:
- median telomere fluorescence;
- mean telomere fluorescence;
- variation in telomere length;
- fifth-percentile telomere length.
None reliably predicts prolonged CDK2-low withdrawal. The reported areas under the curve are no greater than about 0.59, close to random classification.
This result is reproduced across early and late Hs68 cells, donor-derived fibroblasts and hypoxia-aged IMR90 cells.
There may be weak population-level trends—for example, slightly shorter median telomeres in slower-cycling cells—but the overlap between individual cells is too large for telomere length to serve as a useful single-cell predictor.
D. Telomeric DNA-damage foci correlate only weakly with withdrawal
Ageing increases both telomeric and non-telomeric 53BP1 foci.
Telomere-associated 53BP1 shows some correlation with longer CDK2-low periods in early-passage Hs68 cells, but:
- the relationship is inconsistent across cell sources;
- telomeric foci do not classify cycling versus non-cycling cells well;
- maximum areas under the curve are approximately 0.61 for telomere-associated measures;
- total nuclear 53BP1 performs better in one early-passage population, but poorly in other populations.
The findings suggest that integrated DNA-damage signalling may sometimes be more informative than whether a focus is classified as telomeric.
E. Telomere oxidation does not accumulate with ageing or senescence
The authors quantify 8-oxoguanine, an oxidative DNA lesion, both genome-wide and at telomeres.
Contrary to the proposed model of progressive telomeric oxidation:
- proliferating cells have more genomic 8-oxoguanine than non-cycling cells;
- genomic oxidation declines with replicative age;
- cycling cells generally have more oxidised telomeres than non-cycling cells of the same passage;
- the number and intensity of oxidised telomeres do not progressively increase with age;
- longitudinal donor-derived fibroblasts show no consistent age-dependent accumulation.
The likely explanation offered is that cycling cells have greater metabolic and replicative activity and therefore a higher steady-state oxidative burden.
Authors’ interpretation
The authors conclude that replicative senescence is a complex systems-level transition. Telomere shortening or damage may contribute, but measured telomere features do not act as a simple, dominant “timer” that determines the immediate proliferative fate of individual cells.
They suggest that senescence proximity is determined by the integrated burden of multiple processes, including:
- DNA damage;
- p21 and tumour-suppressor signalling;
- autophagy and lysosomal expansion;
- chromatin reorganisation;
- mitochondrial dysfunction;
- altered metabolism;
- impaired proteostasis.
2. Novelty
2.1 Linking prior cell behaviour to telomere measurements in the same cell
The strongest novelty is methodological.
Previous studies generally compared:
- population-average telomere measurements with passage number;
- senescence-marker-positive and -negative populations;
- fixed cells at a single time point.
This study links several days of observed cell-cycle behaviour to subsequent high-resolution telomere measurements in the same individual cells.
The cross-microscope mapping pipeline is particularly valuable because long-term, high-throughput live-cell imaging and three-dimensional telomere imaging normally require different instruments and acquisition conditions.
This makes the study much more informative than a simple cross-sectional correlation between passage number and telomere length.
2.2 Behavioural definition of senescence proximity
The authors use prolonged CDK2 inactivity as a behavioural ground truth rather than defining senescence solely through SA-β-gal, p16, p21 or morphology.
This is conceptually useful because conventional senescence markers can also arise during reversible quiescence. The approach shifts the question from:
Does the cell express a senescence-associated marker?
to:
What has the cell actually been doing over the preceding several days?
That is a significant improvement in single-cell senescence research.
2.3 Evidence for a graded single-cell trajectory
Population growth curves have long implied progressive loss of proliferative capacity, but the study directly demonstrates that individual cells increasingly dwell in a non-cycling state before terminal withdrawal.
The proposed continuum—
fast cycling → slow cycling/intermittent quiescence → prolonged withdrawal → senescence
—is a useful refinement of the binary proliferating-versus-senescent framework.
The idea is consistent with earlier transcriptomic work, but this study gives it direct dynamic support.
2.4 Direct testing of several telomere hypotheses
The paper does not examine only mean telomere length. It separately tests:
- central tendency of telomere length;
- lower-tail telomere length;
- cell-to-cell length heterogeneity;
- telomeric 53BP1 burden;
- total DNA-damage signalling;
- telomeric 8-oxoguanine;
- normoxic and hypoxic ageing conditions.
This breadth makes the negative conclusion more persuasive than a study relying on one telomere assay or one cell line.
2.5 Challenging the telomere-oxidation integrator model
The oxidation results are especially novel. They argue against the idea that telomeres progressively accumulate stable oxidative lesions that integrate lifetime oxidative stress and eventually trigger senescence.
The finding that oxidation is higher in cycling than non-cycling cells changes the likely causal interpretation: steady-state 8-oxoguanine may reflect current metabolic and proliferative activity more than accumulated cellular age.
2.6 Comparative evidence that non-telomeric markers perform better
The paper does not merely report poor telomere prediction. It directly shows that p21, lysosomal content, cell size and chromatin-related measurements classify prolonged withdrawal more successfully.
This comparative element makes the paper’s central conclusion stronger: telomeres are not merely imperfect; under these conditions they are measurably less informative than several broader cellular-state indicators.
3. Critique
Strengths
3.1 Excellent single-cell design
The core strength is the pairing of dynamic and fixed-cell measurements. By following thousands of individual cells before telomere analysis, the authors substantially reduce the inferential limitations of population averaging.
3.2 Large numbers of tracked cells
Several analyses include thousands of cells, producing high statistical power for detecting even weak associations. The absence of strong predictive relationships is therefore unlikely to be explained simply by small sample size.
3.3 Multiple ageing contexts
The use of:
- normoxia-aged Hs68 cells;
- hypoxia-aged IMR90 cells;
- fibroblasts from an older donor;
- longitudinal donor-derived samples,
makes the results more robust than a single-cell-line study.
3.4 Appropriate emphasis on predictive performance
The use of ROC curves is helpful. A statistically significant difference between group means can have almost no practical ability to predict an individual cell’s state. The paper correctly distinguishes statistical association from useful discrimination.
3.5 Transparent acknowledgement of limitations
The authors explicitly recognise that Q-FISH cannot reliably identify the very shortest telomere and that telomere–DDR colocalisation can produce false-positive assignments because large 53BP1 domains may overlap telomeres by chance.
These are central limitations, and the authors do not conceal them.
Important limitations and interpretive concerns
3.6 The study does not exclude a shortest-telomere mechanism
This is the most important limitation.
Replicative senescence may be triggered not by mean or median telomere length but by one or a very small number of critically short, uncapped telomeres. The authors acknowledge that Q-FISH has limited sensitivity for extremely short telomeres.
Therefore, the paper supports:
The telomere distribution metrics measurable by this Q-FISH method do not predict cell-cycle withdrawal.
It does not conclusively establish:
Critically short telomeres do not trigger replicative senescence.
The title is somewhat broader than the experimental evidence. “Telomere length” may be read as including the shortest causal telomere, when that is precisely the measurement the method cannot reliably make.
A more tightly qualified title would have referred to “Q-FISH-derived telomere-length metrics.”
3.7 Prediction is not the same as causation
A causal trigger need not be a strong contemporaneous classifier.
For example, a short-lived telomere lesion could initiate a durable p53–p21 or epigenetic programme and then be repaired, hidden or no longer detectable by the time the cell is measured. Similarly, downstream state variables such as p21 or lysosomal expansion would naturally correlate more strongly with current withdrawal than an upstream initiating event.
Thus, the fact that p21 predicts withdrawal better than telomere length does not prove p21-associated processes are the original cause.
The study is strongest as a biomarker/predictive analysis and weaker as a causal refutation of telomere-driven senescence.
3.8 CDK2-low duration is an imperfect surrogate for irreversible senescence
The study defines proximity to senescence by cumulative or prolonged CDK2-low time over 72–96 hours. This is reasonable, but CDK2-low encompasses both:
- reversible quiescence;
- irreversible senescence.
A cell that remains non-cycling throughout a 72-hour movie might still re-enter the cell cycle later under altered culture conditions.
The authors’ continuum model partly acknowledges this, but the analysis would be stronger if irreversible arrest were validated by:
- longer follow-up;
- mitogen rechallenge;
- replating at low density;
- serum-restimulation experiments;
- clonogenic assays;
- demonstration that individual cells cannot resume division.
Without this, the primary endpoint is best described as prolonged withdrawal or presumed senescence proximity, not definitive senescence.
3.9 The definition of “proximity” assumes the gradual model being tested
The authors infer that longer CDK2-low duration means a cell is closer to terminal senescence. This is plausible and supported by passage-level trends, but it remains partly an assumption.
Some cells may undergo repeated but reversible quiescent episodes without being on a monotonic path to senescence. Other cells may transition abruptly after a specific lesion despite previously short CDK2-low dwell times.
Ideally, individual cells would be tracked until their eventual irreversible arrest, allowing retrospective calculation of actual “time to senescence.” That would directly test whether a biomarker predicts the future event rather than correlating with current behaviour.
3.10 Limited cell-type generalisability
All principal models are fibroblasts.
Fibroblasts are standard for replicative-senescence work, but the balance among telomere erosion, oxidative damage, mitochondrial dysfunction and tumour-suppressor pathways may differ in:
- epithelial cells;
- endothelial cells;
- mesenchymal stem cells;
- immune cells;
- hepatocytes;
- tissue-resident progenitors.
The conclusions should therefore not automatically be extended to all human somatic-cell senescence.
3.11 In-vitro replicative ageing is not equivalent to organismal ageing
Serial passage introduces distinctive pressures:
- repeated enzymatic detachment;
- artificial substrate stiffness;
- high nutrient availability;
- absence of normal extracellular matrix;
- altered oxygen exposure;
- selection of subclones;
- lack of immune and stromal interactions.
The donor-derived cells add relevance, but they are still expanded and measured ex vivo.
The paper provides strong evidence about cultured fibroblast replicative senescence, not necessarily about the origin of senescent cells in intact ageing tissues.
3.12 Donor diversity is limited
The principal in-vivo-aged comparison appears to rely heavily on one 79-year-old donor. The longitudinal oxidation analysis includes three donors, which is useful but still small.
Interindividual variation in:
- inherited telomere length;
- telomerase regulation;
- oxidative-stress handling;
- DNA-repair capacity;
- sex;
- tissue site;
- environmental exposure,
could substantially influence the telomere–senescence relationship.
More donors across age groups would be needed before drawing population-level conclusions.
3.13 Telomere counting raises technical concerns
The authors report more telomeric objects than the expected 92 chromosome ends in diploid human G1 cells in some populations.
Possible causes include:
- cells in S/G2;
- polyploidy or aneuploidy;
- fragmented FISH signals;
- sister telomeres resolved separately;
- segmentation artefacts;
- merged or split three-dimensional objects.
The paper reports no relationship between the number of detected telomeres and withdrawal, but object-detection variability could still affect lower-percentile and variance metrics.
A more detailed validation against cell-cycle phase and karyotype would improve confidence.
3.14 Cross-platform imaging complicates quantitative comparison
Some samples were imaged on different confocal systems with different detector technologies. The authors appropriately warn that absolute values cannot be compared directly.
Nevertheless, platform-specific dynamic range, point-spread functions and segmentation behaviour could influence:
- measured telomere intensity;
- detection of dim/short telomeres;
- 53BP1 focus classification;
- oxidation colocalisation.
The internal within-population analyses remain valid, but cross-population replication is not identical to a fully standardised multisample experiment.
3.15 ROC analyses are largely univariate
The paper compares individual markers, but senescence is explicitly proposed to be multivariate and systems-level.
It would have been valuable to assess:
- combined telomere features;
- combined non-telomere features;
- telomere plus cellular-state models;
- cross-validated logistic regression or machine learning;
- out-of-sample prediction across cell lines or donors.
Telomere features might have little univariate predictive value but add incremental information to a multivariable model.
Conversely, a multivariable model might establish that telomere measurements add essentially nothing once p21, chromatin and cell size are included.
3.16 Potential circularity in downstream markers
Markers such as p21 are mechanistically close to CDK2 inhibition. It is therefore unsurprising that p21 predicts a CDK2-low state better than an upstream lesion.
Similarly, cell enlargement and lysosomal accumulation increase as a consequence of prolonged withdrawal. Their stronger prediction does not necessarily mean they are more fundamental drivers.
The study should distinguish more explicitly between:
- predictors of the current arrested phenotype;
- predictors of future irreversible arrest;
- causal initiators;
- consequences of time spent arrested.
3.17 Interpretation of 8-oxoguanine depends on steady-state abundance
The 8-oxoguanine assay measures the amount present at fixation, which reflects the balance of:
- damage generation;
- base-excision repair;
- replication-associated processing;
- chromatin accessibility;
- antibody accessibility.
A lack of accumulation does not necessarily mean oxidative telomere damage is unimportant. High lesion turnover or efficient repair could conceal repeated oxidative insults.
The authors acknowledge the possibility of short-lived lesions, but the conclusions regarding oxidative causation should remain qualified.
A more direct approach might measure:
- lesion formation and repair kinetics;
- OGG1 recruitment;
- telomere replication stress;
- transient telomere fragility;
- oxidative lesions in specific cell-cycle phases.
3.18 Colocalisation does not establish that damage originated at the telomere
As the authors recognise, 53BP1 domains can extend well beyond the original break. Larger foci are more likely to overlap a telomere by chance.
Although the computational three-dimensional analysis improves on simple Pearson correlation, colocalisation remains an indirect measure of telomere dysfunction-induced foci.
Orthogonal assays—such as chromosome-orientation FISH, metaphase telomere-dysfunction analysis or live recruitment to engineered telomeres—would strengthen the assignment.
3.19 The “gradual induction” model may coexist with threshold mechanisms
The paper sometimes contrasts a binary model against a gradual model. These need not be mutually exclusive.
A biologically plausible model is:
- Multiple cellular stresses gradually accumulate.
- Compensatory capacity progressively declines.
- A threshold is crossed.
- p53–p21 or p16–Rb feedback stabilises irreversible arrest.
Under this formulation, observable cell behaviour is graded while final commitment still involves a nonlinear threshold.
The data strongly refute a crude “normal cycling until an instantaneous switch” model, but they do not rule out threshold behaviour embedded within a gradual ageing trajectory.
3.20 Telomerase discussion is somewhat speculative
The authors correctly note that telomerase has non-canonical functions, but the suggestion that telomerase bypass of the Hayflick limit might partly reflect non-telomeric functions is not directly tested in this paper.
The established ability of telomerase to prevent replicative senescence remains important causal evidence for telomere involvement. To distinguish lengthening from non-canonical effects would require experiments comparing:
- wild-type telomerase;
- catalytically inactive telomerase;
- telomere-targeted versus non-targeted constructs;
- direct telomere elongation by alternative methods;
- temporally controlled telomerase induction.
The discussion appropriately raises this issue, but the results do not resolve it.
4. What the paper establishes—and what it does not
Strongly supported conclusions
- Ageing fibroblasts progressively spend longer in CDK2-low states.
- Replicative ageing is heterogeneous at the single-cell level.
- Conventional senescence markers overlap extensively between cycling and non-cycling cells.
- Mean, median, variance and fifth-percentile Q-FISH telomere metrics are poor classifiers of prolonged withdrawal.
- Telomere-associated 53BP1 is only weakly and inconsistently predictive.
- Stable telomeric 8-oxoguanine does not progressively accumulate in these models.
- Senescence is better represented as a multidimensional cellular-state transition than by any single biomarker.
Not established
- That telomeres have no causal role in replicative senescence.
- That the shortest telomere is not the relevant trigger.
- That prolonged CDK2-low status always represents irreversible senescence.
- That the findings generalise beyond cultured human fibroblasts.
- That non-telomeric cellular changes initiate senescence rather than occur downstream of withdrawal.
- That transient oxidative telomere damage is unimportant.
5. Suggested follow-up experiments
The most decisive next study would track individual cells until irreversible senescence and measure whether present-day features predict the future time to terminal arrest.
Useful additions would include:
- Longer live imaging followed by mitogen rechallenge or clonogenic testing.
- A method capable of identifying the shortest individual telomeres, rather than distribution summaries.
- Direct measurement of telomere replication stress, fragility and uncapping.
- Multivariable predictive models with held-out validation.
- Perturbation experiments manipulating p21, mitochondrial function, autophagy, mTOR and telomere protection.
- Studies across epithelial, endothelial and stem/progenitor cells.
- Larger numbers of donors across different ages.
- Temporal measurement of oxidative lesion formation and repair rather than fixed steady-state 8-oxoguanine.
Conclusion
This is a high-quality study whose most important contribution is to separate population-level association from single-cell prediction. Telomeres plainly change during replicative ageing, but the telomere features measured here do not tell us reliably which individual cell is about to withdraw from the cell cycle.
The work therefore weakens a simple model in which measurable telomere shortening, dysfunction or oxidation acts as the sole dominant clock of senescence. It supports a more integrated model in which gradual changes in cell-cycle regulation, chromatin, lysosomes, metabolism, mitochondrial function and proteostasis collectively shape the approach to senescence.
Its central caution is also its central limitation: the shortest and potentially decisive telomere remains inadequately measured. The paper should consequently be read as a strong critique of simple and measurable telomere biomarker models, rather than a definitive disproof of telomere-initiated replicative senescence.