Telomeres Rose, Then the Senescent Cells Came Back

A dark clinical illustration of two diverging data traces rising from a shared baseline, one climbing steadily while the other dips and rebounds, over a faint hyperbaric chamber interior

The curve a longevity clinic would put on a poster is in a paper published in Immunity & Ageing in August, and it is almost too tidy. Relative telomere length in peripheral blood mononuclear cells, measured at four points across a course of hyperbaric oxygen: 1.000 before the first session, 1.138 after ten, 1.202 after twenty, 1.299 after thirty. A monotonic climb of 13.9 per cent, then 20.3, then 29.0, with a p below 0.001 at the end and a Cohen’s d of 0.70.

The other half of the same paper is a mess, and it is the half that decides what the first half means.

What was actually done

One hundred adults were recruited at the First Hospital of China Medical University between 2021 and 2023 and treated in a multiplace chamber (YC/Q3224-24, Huaxin). The protocol was twenty minutes of pressurisation to 2.5 ATA, sixty minutes breathing 100 per cent oxygen by mask with ten-minute air breaks every half hour, then thirty minutes of decompression. Thirty sessions, with a two-day gap after every ten. Venous blood was drawn four times: before the course, and after ten, twenty and thirty sessions.

Telomere data survived for 97 participants and immunophenotyping for 93. Ethics approval was 2021126, and the paper states registration in the Chinese Clinical Trial Registry as ChiCTR2500110103; the public registry record could not be retrieved when this was written, so what follows relies on the paper.

The model is a linear mixed-effects fit with time as a categorical fixed effect and participant-specific random intercepts, adjusted for baseline values and, in a second model, for age, sex and BMI, which is the right choice for repeated measures.

The line that never wobbled

Telomere length rose at every interval, and the confidence intervals are narrow: an estimate of 0.139 (95% CI 0.039 to 0.238) at ten sessions, 0.203 (0.103 to 0.302) at twenty, 0.290 (0.189 to 0.391) at thirty. Adjustment for age, sex and BMI changed nothing, and no interaction with any of the three reached significance. Whatever the assay was seeing, it saw it regardless of who was in the chamber.

Immune measurements are rarely that well behaved. Telomere length here was relative, not absolute: real-time quantitative PCR with SYBR Green, normalised to the single-copy reference gene 36B4, expressed as 2^−ΔΔCt with each participant’s own baseline sample as the internal calibrator. The 29 per cent is a fold-change against that person’s first draw, not a length in base pairs, and it carries the assumptions of every relative qPCR assay. A 2013 assay study in PLoS One showed how much of this measurement sits in the tube rather than the biology.

An average of a crowd

There is a deeper problem, and the authors raise it themselves rather than waiting to be asked. PBMCs are a mixture of lymphocytes and monocytes with very different replicative histories, and a bulk telomere measurement is a weighted average of that mixture. If the composition shifts, the average moves without any single cell’s telomeres changing.

Their own words: “a reduction in short-telomere, senescence-associated T-cell subsets could increase the average telomere signal measured in bulk PBMCs, even without uniform telomere elongation in individual cells.” They also concede that with no direct measurement of telomerase activity, oxidative stress, DNA damage responses or hypoxia signalling, “the mechanisms underlying the observed telomere changes remain speculative.”

The admission hands the reader a testable question: if the telomere rise is a composition artefact, the populations doing the shifting should track the telomere curve.

The fractions did not follow the same script

They do not. Take the four senescence-associated phenotypes, defined by loss of the co-stimulatory receptors CD28 and CD27, and watch the means across the same four timepoints.

CD28⁻CD8⁺ T cells, as a share of the CD8⁺ compartment, went 48.61 per cent, then 45.63, then 47.13, then 47.09. The model reports a significant time effect, but look at where the significance lives: a drop of 3.180 percentage points at ten sessions (p = 0.002), no significant change at twenty (p = 0.062), and a smaller 2.184-point drop at thirty (p = 0.039). The excursion is at session ten, and between ten and twenty the fraction climbs back most of the way.

CD27⁻CD8⁺ T cells went 42.27, 39.84, 39.24, 40.94. Significant at ten sessions (2.088 points, p = 0.026) and at twenty (2.824 points, p = 0.003), and not significant at thirty (p = 0.088) as the fraction rose again.

CD27⁻CD4⁺ T cells show no significant overall time effect at all (F = 2.407, p = 0.068), and their means travel 17.48, 15.87, 15.99, 17.15, ending within a third of a percentage point of where they started. CD28⁻CD4⁺ T cells barely move: 10.20, 9.43, 9.44, 9.65, no significant time effect.

Two of the four phenotypes produce a significant overall time effect, and in both the largest change arrives early, with the thirty-session value either non-significant or a partial rebound. The paper hedges accordingly, calling these populations “senescence-associated T-cell phenotypes rather than direct evidence of functional T-cell senescence,” because CD27 and CD28 expression was measured without any functional test.

Put the halves together. If bulk telomere length rose because short-telomere CD8⁺ cells were removed from the average, that removal needed to persist at session thirty, when the telomere signal peaked. It did not; the fractions were climbing back. If instead individual cells genuinely elongated their telomeres, the senolytic story that made this field famous would predict the senescent fractions continuing to fall as telomeres grew. They did not do that either. Two markers of the same process, running on different clocks, and the paper does not resolve the disagreement.

The arithmetic problem

None of this proves the finding wrong, but it is worth putting a number beside it.

A 2008 study in Heart measured mean telomere restriction fragment length by Southern blot in circulating leucocytes and found a decline of 27.5 base pairs per year of age, with a mean length of about 6.34 kilobases. Take a 29 per cent increase on a 6.34 kb average and you get roughly 1.84 kb, which is about sixty-seven years of ordinary attrition in the other direction, compressed into a course lasting weeks.

For scale, the largest randomised evidence anyone has on moving this biomarker: the VITAL telomere sub-study measured absolute telomere length by qPCR in 1,031 participants at baseline, year two and year four, and found that 2,000 IU of vitamin D3 daily reduced attrition by 0.14 kilobases over four years, about 140 base pairs, at p = 0.039. That was a real result from a four-year randomised trial.

Relative qPCR units are not base pairs and assays are not interchangeable, so this is an order-of-magnitude sanity check rather than a conversion. Stated plainly: the reported effect is roughly an order of magnitude larger than the best randomised result in the field, arriving in a fraction of the time, in an uncontrolled cohort. That is not impossible. It is the kind of claim that needs a control arm before anyone builds a price list on it.

Who was in the chamber

The paper’s first table repays a close read. Seventy-three of the hundred participants were patients being treated for a clinical indication; twenty-seven were healthy or apparently healthy people having hyperbaric oxygen for what the paper calls health management. Among the patients, 23 were treated for osteomyelitis, 19 for tinnitus or hearing disorders, 11 for osteonecrosis of the femoral head, 10 for insomnia. Sixty-one were women. Forty-four were between 60 and 79 years old, 42 between 40 and 59, and fourteen between 20 and 39.

Fourteen. The headline subgroup finding, that young adults had the largest reduction in CD28⁻CD8⁺ cells (11.025 percentage points, p < 0.001) while the 60-to-79 group had none at any timepoint, rests on an age band containing fourteen people, thirteen in the immunophenotyping set. The sex finding, that women responded and men did not, splits 61 against 39. The paper is honest that these stratified analyses were “exploratory and not powered for definitive subgroup-specific conclusions,” and several reported interactions sit at p = 0.052, 0.078, 0.082 and 0.083. Bonferroni adjustment was applied within pairwise comparisons of models that were already significant, not across four phenotypes times three stratifiers times three timepoints.

The design cannot escape one more confounder. Chronic osteomyelitis is treated with weeks of antibiotics, and antibiotics reshape the circulating immune compartment. Being unwell, being treated and being in a chamber all moved together in seventy-three of these hundred people, so the immune-ageing signal and the treatment-response signal sit in the same variable.

The null that got dressed up

The cytokine arm of the study is the most instructive failure. Tumour necrosis factor alpha, interleukin-1 beta and interleukin-10 were measured at all four timepoints by multiplex bead assay, and none showed a significant time-dependent change across the cohort. That is a null result.

The abstract calls it “dynamic modulation of circulating inflammatory profiles,” which is technically defensible and rhetorically generous. What produced the phrase was subgroup work: men and middle-aged participants showed a fall in TNF-alpha at ten sessions, with the paper noting that “HBOT-associated TNF-alpha modulation was primarily observed during the early treatment phase.” A transient early dip that has gone by the end of the course is not rejuvenation of the inflammatory milieu. It might be a chamber effect, an assay effect, or ordinary regression. Without a control arm, the three cannot be separated.

Where the regulators actually stand

None of this changes the regulatory picture, which the marketing tends to omit.

Hyperbaric oxygen devices are Class II devices cleared through the 510(k) pathway under product code CBF, and the cleared indications are a specific list: decompression sickness, carbon monoxide poisoning, gas gangrene, crush injury, selected non-healing diabetic wounds, compromised grafts and flaps, refractory osteomyelitis, delayed radiation injury, severe anaemia, intracranial abscess and a handful more. Ageing is not on the list. Neither is immunosenescence, telomere length, or health management.

Medicare’s national coverage determination 20.29 is blunter. It limits reimbursement to fifteen named conditions, requires thirty days of failed standard wound therapy before a diabetic wound qualifies, and mandates reassessment every thirty days with coverage withdrawn if healing is not measurable. The words ageing, telomere, longevity and senescence appear nowhere in it. The Undersea and Hyperbaric Medical Society’s fifteenth edition adds detail but not scope: seventeen numbered entries across fifteen indications, from air or gas embolism to avascular necrosis, and no indication for ageing.

The safety file should end the argument about where this belongs. The FDA’s August 2025 letter to health care providers states that the agency is aware of fires involving hyperbaric oxygen devices that resulted in serious injuries and deaths, that the root cause is not known, and that users should follow manufacturer instructions on grounding, training, monitoring and prohibited electrical or static devices, with cotton rather than wool or synthetics. The UHMS, in the same document that lists its indications, warns about low-pressure “mild hyperbaric chambers” sold in wellness centres and spas, often not physician-prescribed, in vessels that are not approved and buildings that do not meet fire and construction standards, putting those exposed at risk of serious injury and death.

What would actually settle it

The study is not worthless. It is prospective, it took four timepoints, it uses a model that respects the repeated-measures structure, and its authors write about their limitations more candidly than most. It is the first cohort in this line of work to report telomere, phenotype and cytokine data together, and the finding that responses differ by age, sex and body mass is worth following up, even if these subgroups are too small to trust.

What would move this from interesting to settled is not more sessions or a bigger poster. It is a parallel control or sham arm; cytomegalovirus serostatus, because the expansion of CD28⁻CD8⁺ cells is driven substantially by latent CMV and studies now show ageing and CMV acting on the T-cell pool both separately and jointly; sorted subsets or single-cell telomere measurement instead of a bulk average; direct telomerase activity; and a blood draw months after the last session to see whether any of it held.

Until then, the accurate sentence is narrow. Thirty sessions of hyperbaric oxygen at 2.5 ATA were associated with a rise in a relative telomere measure in a heterogeneous, uncontrolled cohort, alongside transient and partly reversing changes in two senescence-associated T-cell phenotypes and no overall change in three circulating cytokines. That is a hypothesis with a plausible mechanism and an implausible effect size. It is not a rejuvenation protocol, and no regulator in the United States treats it as one.

Sources

← Back to blog