Senescent cells are one of the hallmarks of aging and their elimination is being pursued for therapeutic purposes. The idea that cellular senescence cannot be reverted has been stated multiple times. Here's a selection of papers that show up in Google Scholar, each of which having being cited over a hundred times.

Judith Campisi (2001)

Because telomerase, the enzyme that can synthesize telomeric DNA de novo, is not expressed by most human cells, telomeres shorten with each cell cycle. When the telomeres erode from their maximum size of 10–15 kb (in the germ line) to an average size of 4–6 kb, human cells irreversibly arrest growth, producing a characteristic (senescent) phenotype [...] Thus, cellular senescence appears to be a mechanism for irreversibly arresting the growth of cells at risk for tumorigenesis

Herbig et al. (2006)

Mammalian somatic cells in culture display a limited proliferative life span, at the end of which they undergo an irreversible cell cycle arrest known as replicative senescence.

Watanabe et al. (2017)

Senescent cells are essentially irreversibly arrested in either the G1 or G2/M phase of the cell cycle and are no longer able to divide, despite remaining viable and metabolically active for long periods

Ogrodnik et al. (2017)

Cellular senescence is a state of irreversible cell-cycle arrest, which can be induced by a variety of stressors, including telomere dysfunction and genotoxic and oxidative stress

More recently the irreversibility is stated in more hedged conditions, but is still included as a hallmark of cellular senescence.

Gorgoulis et al. (2019)

Cellular senescence is a cell state triggered by stressful insults and certain physiological processes, characterized by a prolonged and generally irreversible cell-cycle arrest [...] One common feature of senescent cells is an essentially irreversible cell-cycle arrest that can be an alarm response instigated by deleterious stimuli or aberrant proliferation[...] Cell-cycle arrest is generally considered irreversible during senescence and terminal differentiation, although cell-cycle re-entry can occur under certain conditions

And finally in Nature last year we just got this article referencing a review by Lee & Schmidt (2019) which in turn points us to some articles I mention at the end of this post.

Before going into what is going on it's worth mentioning what counts as senescence and what does not. Most cells do not normally divide but could divide given the right conditions, for example fibroblasts exposed to growth factors in vitro will readily divide (until they reach senescence).

The irreversible growth arrest can be said to be THE defining characteristic of senescent cells. Whether or not a cell is senescent is usually assessed by looking at the expression of p16 or SA-β-gal, but not all cells that express that are senescent; conversely not all cells that do not replicate are senescent (they can be quiescent), so senescence has to be assessed by multiple markers.

But what are those those conditions?

Choi et al. (2010) study whether interactions with the extracellular matrix (ECM) can reverse senescence. In their words, so far there is no known method for restoring aged, senescent cells to the young state

  • Were the cells really senescent? 80% of the fibroglasts were SA‐β‐gal stained
  • What was the intervention? Putting senescent cells on ECM derived from young fibroblasts.
  • What was observed? Within the first week of plating in young ECM, senescent cells were proliferating again and were indistinguisable morphologically (visual inspection) from young cells. To make sure they were not looking at leftover young cells in the ECM, they repeated the measurement again wih neonatal fibroblast ECM and "a different source of cells"; finding that both had the same effect on senescent cells. DNA fingerprinting confirmed that the origin of those multiplying cells were from not from the young ECM donor. They also tagged with red fluorescent protein the cells to track how they adopt a healthy phenotype. In contrast, almost no change was observed in old ECM. To make sure the cells had not mutated and were actually cancerous, they cultivated the restored cells in soft agar (HeLa cells form colonies there, fibroblasts are not supposed to, and they did not) Cells showed:
    • Lower levels of SA‐β‐gal and ROS (comparable to young cells)
    • Complete recovery of mitochondrial membrane potential (JC-1 dye)
    • Reduced levels of p21 and p53 expressions (comparable to young cells, Western blot)
    • When estimulated with EGF, levels of phosporylation of ERK/MAPK (downstream of EGF receptor) are similar to young cells (Western blot)
    • The amount of DNA breaks was reduced to that of young cells (comet assay) after exposing them
    • Telomeres elongated (not to the levels of the young cells, but suficiently that average telomere length was longer than the longest seen in the senescent cells, Southern blot)
    • There was no activation of telomerase as far as they could see (TRAP and RT-PCR assays) so they especulate the lenghtening happened via ALT or very short lived telomerase activation
  • Other things they did: Young fibroblasts grow well on old ECM, but not as well as on young one. Restoration of senescent cells was dependent on Ku70, a protein that plays a role on DNA damage repair, as well as SIRT1 (which activates Ku70).
  • Risk of bias: The soft agar assay if done improperly can fail to detect cancer, but they also did the assay with HeLa cells and they did manage to get them forming colonies, so assuming they did the same both times, this is addressed. And also the cells stopped dividing and became senescent again after 25 doublings, so the cells had not become cancerous. The bias that non-senescent cells were expanding is there, but given that they did a 2x2 study, the control group should have a similar amount.
  • Assessment: The paper convincingly shows that young ECM can indeed un-senesce young fibroblasts. This has a limit: eventually they stop replicating; the likely explanation is that telomeres shortened again: Young ECM can restore telomeres, but not to the same level as young cells; it is a one-off boost but not a complete repair. From this, it seems that the decision to undergo senescence is not only a function of telomere length, oncogene presence, and DNA damage, but also ECM state. Rather than X OR Y OR Z, senescence may happen probabilistically depending on the degrees of damage it is exposed to. For example, if there is a low-grade activation of an oncogene the cell may not senescence, but if exposed to an old ECM it might (while a regular unmutated cell won't).

While this paper didn't look into what was it in the young ECM that caused the effect, Roy et al. (2020) and Roy et al. (2018), growing the cells into small wells (so they clump on top of each other) managed to make fibroblasts into cells whose transcriptomes resembled embryonary stem cells with no other inetrvention, pointing merely to mechanical pressure or cell-to-cell contact as the key element.

Another early attempt is Lapasset et al. (2011), using one of my favorite interventions: epigenetic reprogramminging (Yamanaka+other 2 genes, with a conversion efficiency into iPSCs of 0.06%). What they did was take fibroblasts from a 74 year old, and passage them until they stopped dividing.

  • Were the cells really senescent? There is an increase in p16 by Western blot, 99% vs 4% of cells showing SA-β-gal (staining and visual count), and formation of SAHF (immunofluorescence with antibody against histone H3K9me3, visual inspection). This happened after 51 population doublings. They waited two months to see if there were any changes in cell numbers and no. There are probably some non-senescent cells around but I don't think there will be enough to bias the analysis.
  • What was the intervention? Lentivirus-delivered cocktail of reprogramming factors.
  • What was observed? A week after, disappearance of SAHFs, 3 weeks after new start of proliferation. After a month, cells reached the iPSC state. They selected some of these and they were able to passage them over >35 times. Then, the cells were also able to differentiate into various kinds of cell types. When looking at p16 and p21 gene expression, they looked just like embryonic stem cells (And clearly different from the initial fibroblasts). Telomere length increased (measured by TRF) and after 110 population doublings, no decrease of telomeres was observed. Karyotypes looked normal (visual inspection), transcriptomes didn't look the same as embryonic-derived iPSCs though (there is a leftover "signature of aging"), mitochondrial membrane potential recovered to the same level as an embryonary stem cell (JC-1 dye, fluorescence intensity ratio).
  • Other things they did: They took some senescent-derived stem cells and non-senescent-derived stem cells (also through reprogramming) from the same cells earlier on. They also repeated the entire experiment on embryonic fibroblasts induced into senescent, and same results. Then they repeated the experiment from fibroblasts from individuals aged >96 (Here they didn't make sure they were senescent though). Once a iPSC had been made into a fibroblast, this fibroblast was able to enter senescence, showing that the pathways that lead to senescence stayed intact. The transcriptomes of these differentiated fibroblasts were closer to the fibroblasts derived from embryonic stem cells than the aged fibroblasts.
  • Risk of bias: They selected a small number of cells for analysis so it could have been that those cells happened to be leftover non-senescent cells. The probability of this happen should be small, assuming 1% leftover non-senescent cells and picking 3 of them, the odds of picking 3 non-senescent cells are 1e-6. The cells were most likely not merely quiescent (They sould have shown low levels of p16, Marthandan et al., 2014). They didn't look at SASP, that would have added extra confidence, but the paper did enough. Sa-b-gal is not only present in senescent cells, it is also present in melanocytes or cells grown to confluence; this wasn't the case here (Debacq-Chainiaux et al. 2009)
  • Assessment: The paper does what it says it does: cells can be induced out of replicative senescence by epigenetic reprogramming and the resulting -stem- cells are fine.

Latorre et al. (2017) look at mRNA splicing factors instead (These are the ones that take pre-mRNA and help separate introns from the exons that will ultimately make proteins) and how resveratrol-like compounds ("resveralogues") affect them, and through them, senescence, as the authors note mRNA splicing is impaired in senescent cells.

  • Were the cells really senescent? The cells were fibroblasts (of 3 kinds). They assessed senescence when growth slowed to less than 0.5 population doublings per week (This happened around 65-64 doublings). 75% of the cells were Sa-b-gal stained.
  • What was the intervention? Resveratrol and resveralogues delivered into the culture medium
  • What was observed? Alteration of key SASP components (Including IL6, TNFa or IL1b). Resveratrol reduced all of them, the other compounds had weaker effects, in some cases increasing the secretion of inflammatory proteins (Measured by ELISA, changes assessed by coputer vision). Decrease in p16 transcript activity (relative to other genes they claim do not change with senescence). Sa-b-gal activity decreased to 25% of the cells. Some cells of the population underwent mitosis, while a control group of senescent cells did not. Telomeres lengthened (measured by qPCR, qPCR is not great for this purpose), but they did with all the resveralogues so maybe there's something to it.
  • Risk of bias: Here there were a lot of potentially non-senescent cells, and the passaging number wasn't as high as in the previous paper. The possibility that those cells are dividing when presented with resveratrol (instead of the senescent cells) exists. To mitigate this bias, that the authors acknowledge, they cultivated cells in a medium that lacks the factors needed to induce cells to replicate, and they triplicated this part of the experiment, and they noted that here too there was a decrease in Sa-b-gal. In total, 15% of the senescent cells reverted back. Another potential bias is that senescent cells die, increasing the % of non-senescent cells; they addressed this by looking for LDH which would only be present in the medium if cells undergo apoptosis. Relative to control, cells treated had less or similar LDH activity at the compounds concentrations studied. This wasn't true for resveratrol though. They also did a TUNEL assay to measure apoptosis, counting at least 400 cells each time, showing lower levels of apoptosis, including for resveratrol. This sounds like a contradiction with the LDH assay, but there they used 10 uM and in TUNEL they used 5. It would have been nicer had they also used 5 to make sure both tests point in the same direction at the right concentration. In an additional apoptosis test they did (Caspase 3 and 7 assays) for, say, resveratrol, where TUNEL had shown a 50% drop in apoptosis, these tests showed a slight increase.
  • Assessment: Between the high number of potentially non-senescent cells and the equivocal results, I think this paper is not good evidence that they managed to un-senescence the cells. Resveratrol may have increased apoptosis, and if we take resveratrol out, the other compounds didn't clearly reduce all the components of the SASP. On the other hand they did manage to reduce Sa-β-Gal; but I'm not sufficiently confident that it was not due to apoptosis of senescent cells. The paper didn't look at more detailed measures of function (like mitochondrial function).Note that I don't say that they didn't; just that it wouldn't be strong evidence that they did, the data is not as clear as I'd like.

So cellular senescence can, after all be reverted, but not in all cases. If telomeres are sufficiently short then interventions other than reprogramming (and that has potential teratogenic effects), or gene editing (As in Beasejour et al., 2003, Yu et al. 2018 or Sage et al. 2003, but note they are knocking off p53 which is a tumor suppressor) won't undo senescence. It may well be that small molecules also work; the Latorre paper shows some promise, and there is other work on applying small molecules to make iPSCs out of senescent cells (Borgohain et al., 2018), so it is plausible that some combination could go half-way and rejuvenate a cell without de-differentiation.

So far, the only cases where rescue from senescence has been observed are in telomere-induced senescence; this makes sense because these cells are fine in any regard other than having short telomeres, it is the growth arrest that is causing the cell to damage itself, it's like clamping a car while it is still trying to move (senescent cells repeatedly try and fail to undergo mitosis), removing the clamp and the cell goes back to normal. A hypothesised function of senescence is as an anti-tumor mechanism: following the rule "if your telomeres are short, stop replicating & your telomeres shorten in every replication" is a general rule that will stop cancers (unless cancers manage to get their hands on sweet sweet telomerase or ALT). But that rule will also cause lots of cells that are perfectly fine to undergo senescence, and so there is probably nothing bad in rescuing them.

But for oncogene-senescence other than with targeted gene editing it has not yet been shown if they can be rescued, and undoing their senescence in any other way wouldn't be great for the organism hosting the cell, so for anti-aging purposes it's a bad move.

Rejuvenating the ECM does show promise. If it can be made elastic again then those senescent cells that could still go back to normal will do so, reducing SASP-induced inflammation, and given that the ECM seems to be similarly linked to other hallmarks (Fedintsev & Moskalev, 2020), targeting it makes sense; however targeting it to some extent means breaking glucosepane, which is quite difficult. At the moment there is a single startup working on that, Revel Pharmaceuticals.