Cellular senescence

Terminology

• Replicative senescence is the arrest that ends the division of human cells in culture after a limited number of divisions, the Hayflick limit. It is triggered by short telomeres (see telomeres and hayflick-moorhead-experiment).
• Oncogene-induced senescence is senescence triggered by an activated oncogene, such as mutant RAS or BRAF, independently of the number of divisions.⁵⁸
• Quiescence is a reversible pause in division, from which a cell returns when conditions change. Senescence differs in being stable, and in the cell’s changed shape, metabolism and secretions.¹
• The senescence-associated secretory phenotype (SASP) is the set of proteins senescent cells secrete, including inflammatory cytokines such as IL-6 and IL-8.¹⁹ (See sasp.)
• Senolytics are drugs that kill senescent cells selectively; senomorphics suppress the SASP without killing the cells. (See senolytics.)
• Markers. No single marker identifies senescent cells. The most used are the activity of a β-galactosidase enzyme detectable at pH 6 (SA-β-gal), high levels of the cell-cycle inhibitors p16 and p21, and the absence of markers of division; a 2019 consensus recommended combining several.²⁰²

Discovery

Hayflick and Moorhead described 25 strains of normal human fibroblasts derived from fetuses and listed their finite lifetime in culture as one of the criteria that distinguished them from cancer cell lines.³ In 1965 Hayflick reported that the time at which such strains stop dividing “is not a function of the number of subcultivations but rather of the number of potential cell doublings”, which he put at 50±10. Cells from adult donors stopped after about 20 doublings.²¹

The counting mechanism was found three decades later. Telomeres shorten as human cells divide in culture,²² and cells given telomerase, the enzyme that lengthens telomeres, continued dividing at least 20 doublings past their normal limit.²³ Fabrizio d’Adda di Fagagna and colleagues showed in 2003 that the arrest is a DNA-damage response: senescent human cells carry clusters of DNA-damage proteins at their chromosome ends, and inactivating the checkpoint kinases CHK1 and CHK2 lets them re-enter DNA replication.⁴

In 1997 Manuel Serrano, Scott Lowe and colleagues found that an activated RAS oncogene causes normal human and rodent cells to enter “a permanent G1 arrest” that is “phenotypically indistinguishable from cellular senescence”. Senescence, they concluded, “does not simply reflect the accumulation of cell divisions, but can be prematurely activated in response to an oncogenic stimulus.”⁵

Pathways and markers

Two tumour-suppressor pathways establish and maintain the arrest (see p53, p21, p16 and RB).

• p53 and p21. DNA damage, including damage at short telomeres, activates p53, which induces p21. p21 inhibits the cyclin-dependent kinases that drive the cell into DNA replication.
• p16 and RB. p16 inhibits the kinases CDK4 and CDK6, which leaves the RB protein active. Active RB represses the E2F transcription factors that cells need to replicate DNA.

The two pathways differ in how they hold the arrest. Christian Beauséjour and colleagues inactivated p53 in senescent human fibroblasts: cells with low p16 resumed growth, but cells with high p16 did not. Telomerase alone did not reverse the arrest. The arrest that follows telomere dysfunction is therefore maintained mainly by p53, and p16 provides a second barrier that, once established, makes it hard to reverse.²⁴

Senescent cells were first identified in tissue by Goberdhan Dimri and colleagues, who found that “several human cells express a beta-galactosidase, histochemically detectable at pH 6, upon senescence in culture”, and that cells staining for it become more common in human skin with age.²⁰ The stained enzyme is ordinary lysosomal β-galactosidase, whose activity at pH 6 reflects lysosomal mass,²⁵ and macrophages that express both it and p16 accumulate in old mice.¹⁵ Later work showed that senescent cells are heterogeneous. Comparing many transcriptomes, Alejandra Hernandez-Segura and colleagues found that the senescence programme depends on cell type, on the stress that caused it and on time since arrest, and identified 55 genes at its common core, none of them a classical marker such as p16 or p21.²⁶ The International Cell Senescence Association therefore recommended identifying senescent cells by several markers together.²

The secretory phenotype

Jean-Philippe Coppé, Judith Campisi and colleagues named the SASP in 2008. Human cells induced to senesce by DNA damage “secrete myriad factors associated with inflammation and malignancy”; the secretion developed over several days, only after damage severe enough to cause senescence, and was amplified by oncogenic RAS and by loss of p53.¹⁹ The same year, two groups showed that the SASP also acts on the senescent cell itself. Juan Carlos Acosta and colleagues found that chemokines acting through the receptor CXCR2 reinforce senescence,²⁷ and Thomas Kuilman and colleagues found that IL-6 is required for oncogene-induced senescence to begin and to persist.²⁸

The SASP makes senescence a property of tissues as well as cells. It recruits immune cells that clear senescent cells, and it can also promote inflammation and, in some contexts, the growth of neighbouring premalignant cells.¹⁹²⁹ (See sasp.)

Functions

Tumour suppression

In 2005 four groups found senescence in early tumours. Manuel Collado and colleagues found senescence markers abundant in premalignant lung adenomas driven by K-ras in mice, and essentially absent from the adenocarcinomas that develop from them.⁷ Chrysiis Michaloglou and colleagues found that sustained mutant BRAF arrests human melanocytes, and that congenital naevi (moles), most of which carry this mutation, are “invariably positive” for SA-β-gal without telomere attrition.⁸ Later studies found that the usual senescence markers do not distinguish naevus cells from normal or malignant melanocytes, and in a mouse model naevus cells showed no evidence of senescence, so whether naevi are held by senescence is disputed.³⁰³¹ Zhenbang Chen and colleagues found that loss of Pten in the mouse prostate induces p53-dependent senescence, so that invasive cancer develops early only when p53 is also lost.³² Melanie Braig and colleagues found that oncogene-induced senescence is a barrier to lymphoma in mice.³³

The immune system removes senescent premalignant and tumour cells. Wen Xue and colleagues found that restoring p53 in mouse liver carcinomas caused complete regressions, mainly by senescence followed by “an innate immune response that targeted the tumour cells”.³⁴ Tae-Won Kang and colleagues found that hepatocytes made senescent by oncogenic Nras are cleared by an immune response that requires CD4+ T cells, which limits the development of liver cancer.³⁵

Repair and development

Senescent cells appear early in skin wounds and speed their closure by secreting the growth factor PDGF-AA; clearing them delays healing.⁹ In damaged mouse liver, activated stellate cells become senescent, make less scar tissue and are killed by natural killer cells, which limits fibrosis; mice that cannot activate senescence develop more fibrosis.¹⁰

Senescence also occurs in normal embryos. Daniel Muñoz-Espín and colleagues found it in the mesonephros (a transient embryonic kidney) and the endolymphatic sac of the inner ear of mouse and human embryos, dependent on p21 but independent of DNA damage and p53, and followed by clearance by macrophages; they proposed that this programme “is the evolutionary origin of damage-induced senescence”.⁶ Mekayla Storer and colleagues found senescent cells in the signalling centre that directs limb growth, where loss of p21 caused defects in limb patterning.³⁶

Senescent cells and ageing

Accumulation

In tissue arrays from five donors per age group, cells staining for p16 or p21 increased with age in skin, kidney and pancreas, did not increase in lung, and were undetectable in skeletal and cardiac muscle.³⁷ A meta-analysis of 38 human studies found a modest overall correlation between senescence markers and age (r = 0.28; 95% prediction interval −0.12 to 0.60). It was strongest for DNA-damage markers (r = 0.69) and absent for proliferation markers (r = 0.08), and the level of senescence “varies considerably depending upon tissue type, tissue section, and marker used to detect senescence”.¹¹

How long senescent cells persist depends on how fast they are removed. Omer Karin, Uri Alon and colleagues reported that senescent cells “turn over rapidly in young mice, with a half-life of days, but slow their own removal rate to a half-life of weeks in old mice”. Their model, fitted to whole-body p16-reporter signals from 33 mice, predicted half-lives of about 5 days at 3 months and 25 days at 22 months. After inducing senescence in lung epithelium, they measured about 4.7 days in young mice and about 18 days, with wide uncertainty and only two sampling days, in 22-month-old mice.³⁸ Mice with impaired cytotoxic immune cells, lacking perforin, accumulate more senescent cells and have shorter lives.³⁹

A model of senescent-cell dynamics

Karin and colleagues proposed a model in which senescent cells are produced at a rate that rises in proportion to age and removed at a rate that saturates when they are numerous, as when the immune cells that remove them are fully occupied. Random fluctuations are added, and the authors assumed, for illustration, that death occurs when the burden of senescent cells first exceeds a threshold; they noted that “the precise relationship is currently unknown”. With production, removal and noise chosen to be consistent with the reporter data and to match the mean and spread of C57BL/6J lifespans, the model gives a mortality rate that rises roughly exponentially with age, as in the Gompertz law, and slows at the oldest ages.³⁸ The fit to mortality does not on its own show that the burden is senescent cells: the same equation, with the burden read as generic damage, fits the survival curves of flies, nematodes and other species.³⁸⁴⁰

Figure 1 simulates the model for 20,000 mice. With the maximum removal rate of 0.15 a day used in the authors’ mortality fit (their fit to the reporter trajectories gave 0.27 a day), the median mouse lives 28.4 months, and the monthly mortality rate rises about tenfold between 15 and 21 months. Drag the removal rate down to 0.12 a day and the median lifespan falls to 24.5 months; raise it to 0.18 and it rises to 32.4. A 20% change in the maximum removal rate, applied throughout life, shifts the median lifespan by about 14% in either direction.

{
  "$schema": "https://vega.github.io/schema/vega-lite/v6.json",
  "description": "Figure 1. The saturating-removal model of Karin and colleagues, simulated for 20,000 mice. Senescent cells are produced at a rate that rises with age and removed at a rate that saturates when they are numerous; random fluctuations are added, and a mouse dies when its senescent-cell burden first crosses a threshold. The first panel shows the burden among mice still alive (median, with the 10th to 90th percentile shaded) and the threshold. The second shows the resulting mortality rate per month, on a logarithmic scale. The slider sets the maximum removal rate β; the other parameters are those the authors used to fit C57BL/6J mortality (production slope 0.084 per day per year, half-saturation 0.5, noise 0.16 per day, threshold 17; burden in units of the mean burden of young mice).",
  "params": [{"name": "B", "value": 0.15, "bind": {"input": "range", "min": 0.12, "max": 0.18, "step": 0.01, "name": "Maximum removal rate β, per day"}}],
  "hconcat": [
    {
      "width": 290,
      "height": 230,
      "layer": [
        {
          "data": {"url": "/attachments/senescent-cell-model.csv"},
          "transform": [{"filter": "abs(datum.beta - B) < 0.001 && isValid(datum.xmed) && datum.xmed !== ''"}],
          "layer": [
            {"mark": {"type": "area", "style": "observed", "opacity": 0.35}, "encoding": {"y": {"field": "xlo", "type": "quantitative"}, "y2": {"field": "xhi"}}},
            {"mark": {"type": "line", "style": "model"}, "encoding": {"y": {"field": "xmed", "type": "quantitative"}}}
          ],
          "encoding": {"x": {"field": "month", "type": "quantitative"}}
        },
        {"data": {"values": [{}]}, "mark": {"type": "rule", "style": "guide"}, "encoding": {"y": {"datum": 17, "type": "quantitative"}}},
        {
          "data": {"values": [{"month": 1, "y": 17, "label": "threshold"}]},
          "mark": {"type": "text", "style": "observed-label", "align": "left", "dy": -8},
          "encoding": {"x": {"field": "month", "type": "quantitative"}, "y": {"field": "y", "type": "quantitative"}, "text": {"field": "label"}}
        }
      ],
      "encoding": {
        "x": {"type": "quantitative", "scale": {"domain": [0, 45], "nice": false}, "axis": {"title": "Age, months", "values": [0, 6, 12, 18, 24, 30, 36, 42]}},
        "y": {"type": "quantitative", "scale": {"domain": [0, 18], "nice": false}, "axis": {"title": "Senescent-cell burden", "values": [0, 5, 10, 15]}}
      }
    },
    {
      "width": 290,
      "height": 230,
      "layer": [
        {
          "data": {"url": "/attachments/senescent-cell-model.csv"},
          "transform": [{"filter": "abs(datum.beta - B) < 0.001 && isValid(datum.hazard) && datum.hazard !== '' && datum.hazard >= 0.002"}],
          "layer": [
            {"mark": {"type": "line", "style": "model", "strokeWidth": 2}},
            {"mark": {"type": "circle", "style": "model-point", "size": 22}}
          ],
          "encoding": {"x": {"field": "month", "type": "quantitative"}, "y": {"field": "hazard", "type": "quantitative"}}
        },
        {
          "data": {"url": "/attachments/senescent-cell-model.csv"},
          "transform": [
            {"filter": "abs(datum.beta - B) < 0.001"},
            {"aggregate": [{"op": "max", "field": "median", "as": "median"}]},
            {"calculate": "'Median lifespan ' + format(datum.median, '.1f') + ' months'", "as": "label"},
            {"calculate": "1", "as": "lx"},
            {"calculate": "0.8", "as": "ly"}
          ],
          "mark": {"type": "text", "style": "model-label", "align": "left", "baseline": "top"},
          "encoding": {"x": {"field": "lx", "type": "quantitative"}, "y": {"field": "ly", "type": "quantitative"}, "text": {"field": "label"}}
        }
      ],
      "encoding": {
        "x": {"type": "quantitative", "scale": {"domain": [0, 45], "nice": false}, "axis": {"title": "Age, months", "values": [0, 6, 12, 18, 24, 30, 36, 42]}},
        "y": {
          "type": "quantitative",
          "scale": {"type": "log", "domain": [0.001, 1], "nice": false},
          "axis": {"title": "Deaths per month among survivors", "values": [0.001, 0.01, 0.1, 1], "labelExpr": "datum.value >= 1 ? '1' : '1 in ' + format(1 / datum.value, ',')"}
        }
      }
    }
  ]
}

The saturating-removal model

The senescent-cell burden $X$ changes as

$$\frac{\mathrm{d}X}{\mathrm{d}t}=\eta t-\frac{\beta X}{\kappa +X}+\sqrt{2\epsilon }\xi (t),$$

where $\eta t$ is the production rate, rising with age $t$; $\beta$ is the maximum removal rate; $\kappa$ is the burden at which removal runs at half its maximum; and $\xi (t)$ is white noise of strength $\epsilon$. Death occurs when $X$ first reaches $X_c$.³⁸ Figure 1 uses $\eta =0.084$ per day per year, $\kappa =0.5$, $\epsilon =0.16$ per day and $X_c=17$, with $X$ in arbitrary units, and integrates the equation in half-day steps.

Clearing senescent cells in mice

Darren Baker, Jan van Deursen and colleagues built mice in which cells expressing p16 can be killed with a drug (see baker-ink-attac-experiment). In mice with a progeroid mutation, life-long removal of these cells delayed age-related disease in fat, muscle and eye, and removal late in life “attenuated progression of already established age-related disorders”.⁴¹ In normal mice, removal twice a week from 12 months of age extended median lifespan by 27% in one genetic background and 24% in another, delayed tumour formation and attenuated the deterioration of kidney, heart and fat.¹² Clearance was partial and did not reach liver or colon; maximum lifespan was not extended in C57BL/6 mice; and the authors noted that vehicle-injected male C57BL/6 controls were short-lived, “suggesting that repetitive vehicle injection stress may have negatively impacted C57BL/6 male longevity”.¹² Removing cells that express high levels of p21, once a month from 20 months of age, extended both median and maximum lifespan.⁴²

Transplanting small numbers of senescent cells into young mice caused persistent physical dysfunction and spread senescence to the host’s tissues. Ming Xu and colleagues gave the senolytic combination of dasatinib and quercetin every two weeks to C57BL/6 mice from 24–27 months of age; median survival from the start of treatment was 36% longer than in controls, and the mortality hazard was 65% of theirs.⁴³ No independent laboratory has reported repeating this lifespan result.

Not all cells with high p16 are harmful to remove. In p16 reporter mice, most such cells that appear from about 10–12 months were vascular endothelial cells, mostly in the liver sinusoids, with fewer macrophages and adipocytes. Removing them, continuously or acutely, disrupted blood–tissue barriers, caused liver and perivascular fibrosis and worsened health, and the endothelial cells were not replaced.¹³ In the lung, p16-expressing fibroblasts in the basement membrane form a niche that supports repair of the epithelium.¹⁴ Clearing senescent cells from skin wounds delays healing.⁹

Senolytic drugs

Yi Zhu, James Kirkland and colleagues found that senescent cells depend on survival pathways that, when blocked, kill them but not dividing or quiescent cells. Dasatinib, a cancer drug, and quercetin, a plant flavonoid, each killed some types of senescent cells, and together they reduced the senescent-cell burden in old and progeroid mice.⁴⁴ Navitoclax (ABT-263), an inhibitor of the BCL-2 family of survival proteins, cleared senescent blood and muscle stem cells in old and irradiated mice.⁴⁵ Fisetin, another flavonoid, was reported to extend median and maximum lifespan when fed to old mice, in groups of only eight or nine animals.⁴⁶ The National Institute on Aging’s Interventions Testing Program, feeding fisetin from 20 months of age to genetically diverse mice at three sites, found no effect on lifespan in either sex and no reduction in p16-positive cells.⁴⁷

Human studies are early.

• Pulmonary fibrosis. In an open-label pilot study of 14 people with idiopathic pulmonary fibrosis, nine doses of dasatinib and quercetin over three weeks improved walking distance, gait speed and chair-standing time; lung function did not change. There was no placebo group.⁴⁸
• Diabetic kidney disease. In an uncontrolled study of 11 people with diabetic kidney disease given three days of dasatinib and quercetin, p16-positive cells, macrophages and crown-like structures in abdominal fat were fewer 11 days later (tissue from 9 people). A corrected analysis found no significant change in skin senescence markers, and among blood SASP factors only MMP-12 fell individually.⁴⁹⁵⁰
• Pulmonary fibrosis (randomised). In a placebo-controlled pilot trial of 12 patients, physical, pulmonary and frailty measures did not differ meaningfully between groups; the trial was not powered for efficacy, and the treated group had more non-serious adverse events.¹⁷
• Bone. In 60 postmenopausal women randomised to intermittent dasatinib plus quercetin or no treatment for 20 weeks, the bone-resorption marker CTx, the primary endpoint, did not differ between groups (P = 0.61). A transient rise in a bone-formation marker and benefits in women with high T-cell CDKN2A expression were exploratory.¹⁸
• Osteoarthritis. UBX0101, injected into the knee, reduced post-traumatic osteoarthritis in mice.⁵¹ In a placebo-controlled phase 2 trial of 183 patients with painful knee osteoarthritis, a single injection showed “no statistically significant difference” from placebo in pain at 12 weeks, and development was stopped.¹⁶

Several of the mouse and human studies above, and a 2022 review of the field,⁵² come from the group that developed dasatinib plus quercetin, whose members hold patents on senolytic drugs. None of the randomised trials published so far has shown that a senolytic improves a clinical outcome.

Senescence in theories of ageing

Campisi proposed in 2005 that senescence is antagonistically pleiotropic: it promotes survival early in life by suppressing cancer, and limits it later as senescent cells accumulate.²⁹ Silvia Giaimo and d’Adda di Fagagna questioned whether the evidence shows that the benefits fall mainly in youth and the costs mainly in old age.⁵³ Senescence also continues to suppress cancer in old organisms, which bears on whether removing senescent cells late in life is safe. (See Antagonistic pleiotropy and tumor-suppressor-theory-of-aging.)

History

Replicative senescence

Hayflick and Moorhead established in 1961 that normal human cells have a finite lifetime in culture, and Hayflick related it to the number of divisions in 1965.³²¹ In 1995 Dimri and colleagues found a marker that identified senescent cells in tissue,²⁰ and between 1990 and 2003 shortening telomeres and the DNA-damage response at chromosome ends were identified as the mechanism of the replicative limit.²²²³⁴

Senescence beyond the Hayflick limit

Oncogene-induced senescence was described in 1997,⁵ and in 2005 four studies found senescence in premalignant tumours.⁷⁸³²³³ The SASP was named in 2008,¹⁹ and senescence in normal embryonic development was described in 2013.⁶³⁶

Senescent cells as a target

In 2011 and 2016 Baker and colleagues showed that removing p16-expressing cells delays age-related disease and extends lifespan in mice,⁴¹¹² and the first senolytic drugs were reported in 2015.⁴⁴ The first human pilot studies were published in 2019,⁴⁸⁴⁹ and in 2020 the first placebo-controlled phase 2 trial of a senolytic failed.¹⁶ Randomised trials of dasatinib plus quercetin published in 2023 and 2024 were also negative on their primary or main outcomes.¹⁷¹⁸

Data and figures

Figure 1 was computed for this article by simulating the model of Karin and colleagues,³⁸ with the parameters in the note beneath it, for 20,000 mice at each of seven values of the maximum removal rate. Mortality rates are deaths in each month divided by mice alive at its start, shown where they are at least 0.002; burden percentiles are shown while at least 200 mice remain alive.