Entry 01 · Biomarkers
Epigenetic clocks and “biological age”
“You can measure your biological age — and this intervention reversed it by several years.”
What the clocks are
DNA methylation clocks read chemical marks at specific sites in the genome and return a number. They are real, reproducible, and genuinely predictive. What they predict depends entirely on what they were trained on, and the differences are not cosmetic.
Horvath’s multi-tissue clock — the original
Horvath S. Genome Biology 2013;14(10):R115 · 7,844 non-cancer samples, 82 datasets, 51 tissues and cell types
353 methylation sites, trained directly on chronological age, predicting it in the test data with a median error of 3.6 years across tissues.
What this meansTrained on the calendar, its ceiling is reproducing the calendar. It was never designed to measure health, and Horvath reports it is poorly calibrated in breast, endometrium, fibroblasts, skeletal muscle and heart tissue.
PhenoAge and GrimAge — trained on death, not on age
Levine ME et al. Aging 2018;10(4):573–591 · Lu AT et al. Aging 2019;11(2):303–327 (Framingham Offspring, n=2,356)
PhenoAge was built from nine clinical biomarkers through a mortality model; GrimAge from methylation surrogates for seven plasma proteins and smoking, regressed on time-to-death. Both are deliberately worse at guessing chronological age and better at predicting mortality. Only 41 of PhenoAge’s 513 sites overlap Horvath’s 353.
Read carefullyThese are different instruments measuring different things. A person can be “younger” on one and “older” on another without either being wrong. A result quoted without naming the clock is not a result.
DunedinPACE — a rate rather than an age
Belsky DW et al. eLife 2022;11:e73420 · Dunedin birth cohort; n=817 with both methylation and a 20-year pace measure
Built against nineteen organ-system biomarkers tracked in the same people at ages 26, 32, 38 and 45. It reports how fast someone is ageing per calendar year, not how old they are.
LimitsOne modestly sized, single-country, predominantly white cohort. The authors state that future work must still show the measure responds to interventions and that changes in it track health outcomes.
The best test of whether a clock can be moved
Two years of caloric restriction, measured on four clocks
Waziry R et al. Nature Aging 2023;3:248–257 · randomised, 2 years, 197 CALERIE participants with methylation data (128 restricted, 69 control)
DunedinPACE slowed: d = −0.29 at twelve months and −0.25 at twenty-four (P < 0.003), which the authors put at a 2–3% reduction in the pace of ageing.
PhenoAge, GrimAge, Horvath and Hannum showed no significant change at all, and the point estimates for several of them pointed very slightly the other way.
What the authors sayThe two groups’ DunedinPACE trajectories overlap by close to 90%. And, verbatim: “It is therefore unclear if the changes in DunedinPACE observed during the 2-yr intervention will translate into reduced morbidity and mortality over the long term.”
The study behind the “age reversal” headlines
TRIIM
Fahy GM et al. Aging Cell 2019;18(6):e13028 · 10 men enrolled, 9 completed, ages 51–65
Twelve months of recombinant human growth hormone with DHEA and metformin. Reported epigenetic age about 2.5 years lower than the untreated projection, with similar movement on four clocks.
The designNine men. All male. No control group — single-arm and open-label, so the comparison is against expected drift, not against anybody. The authors themselves write that the results must be replicated in an appropriately powered study, and note that epigenetic age “is not synonymous with aging itself.” Their own headline claim is explicitly about a predictor, not about measured lifespan. This is described here because it is widely cited, not because it is a protocol worth copying — growth hormone is a prescription drug with its own risk profile.
What this adds up to
DemonstratedMethylation clocks predict mortality and disease risk at population level, reproducibly, and better than chronological age alone.
PlausibleThat a sustained intervention can slow one clock. Caloric restriction moved DunedinPACE by 2–3% in a randomised trial — a real result, and a small one.
Not establishedThat moving a clock reading changes how long or how well anyone lives. The Biomarkers of Aging Consortium states it plainly: there are no formally validated surrogate endpoint biomarkers of ageing, and a reduction in epigenetic age is meaningful only if it turns out to track a clinical benefit. That has not been shown.
Not establishedThat any individual’s clock reading is actionable. These instruments were built and validated on groups. Nothing establishes what a single person should do differently on receiving a number.
Consortium statement: Moqri M et al. Cell 2023;186(18):3758–3775. Methodological review: Bell CG et al. Genome Biology 2019;20:249.
Entry 02 · Intervention
Caloric restriction, fasting and autophagy
“Fasting triggers autophagy, the body’s cellular renewal process, and caloric restriction extends lifespan.”
The best human trial
CALERIE Phase 2
Kraus WE et al. Lancet Diabetes & Endocrinology 2019;7(9):673–683 · 218 randomised (143 restricted, 75 control), healthy non-obese adults aged 21–50, two years
Significant improvements against control in LDL cholesterol, the total-to-HDL ratio, systolic and diastolic blood pressure, C-reactive protein, insulin sensitivity and a metabolic syndrome score. Mean weight loss 7.5 kg, 71% of it fat.
Four things usually left outThe word “exploratory” is in the paper’s own title — these were not the pre-specified primary outcomes, which were resting metabolic rate and core body temperature. Participants were prescribed 25% restriction and achieved 11.9%. The trial cannot separate restriction from the 7.5 kg weight loss itself. And there were costs: roughly 2% loss of bone mineral density at spine and hip, with withdrawals for that and for reduced haematocrit.
The primate studies that disagreed
Wisconsin and the NIA reached opposite conclusions
Colman RJ et al. Science 2009;325:201–204 (76 monkeys) · Mattison JA et al. Nature 2012;489:318–321 (121 monkeys)
Wisconsin found 30% restriction reduced age-related mortality roughly threefold and cut diabetes, cancer and cardiovascular disease. The NIA, running what looked like the same experiment, reported that restriction “has not improved survival outcomes.”
What the 2017 reconciliation foundThe controls were not equivalent. NIA control monkeys were not free-fed — they were portioned to published guidelines, so that study effectively compared moderate restriction against more restriction. Diets differed sharply: sucrose was under 7% of carbohydrate at NIA and 45% at Wisconsin. And age at onset differed. Mattison JA et al. Nature Communications 2017;8:14063.
Time-restricted eating, tested directly
TREAT — 16:8 against three ordinary meals
Lowe DA et al. JAMA Internal Medicine 2020;180(11):1491–1499 · 116 randomised, 12 weeks
The time-restricted group lost 0.94 kg; the control group eating three structured meals lost 0.68 kg. The difference between groups was 0.26 kg and was not statistically significant (P = 0.63). No significant between-group difference in fasting insulin, glucose, HbA1c or lipids.
And the composition of what was lostThe restricted group lost significantly more lean mass than controls (P = 0.005). Twelve weeks only; adherence self-reported; body-composition data from a subgroup. Readers should also consult the clarification regarding participant messaging that the journal published in 2021 alongside this trial.
Autophagy — where the claim comes from, and where it stops
Yoshinori Ohsumi’s 2016 Nobel Prize was awarded for discovering the mechanisms of autophagy in baker’s yeast. It is foundational cell biology. It establishes nothing about fasting schedules in people, and it is routinely cited as though it does.
The measurement problem
Sargeant TJ, Bensalem J. Trends in Molecular Medicine 2021;27(12):1091–1094
Meaningful measurement requires autophagic flux — material actually sequestered and degraded over time — not a static snapshot. In humans this can currently be measured only in blood cells. There is no established biomarker of autophagic flux in people, and none at all for solid tissue.
ConsequenceAutophagic flux has never been measured in a living human liver, brain, heart or muscle — the very tissues the longevity claims are about.
When intermittent fasting was actually tested in human muscle, markers went down
Chaudhary R et al. Nutrition 2022;101:111662 · 50 women, 24-hour fasts three days a week for 8 weeks, with muscle biopsies
The authors’ conclusion, verbatim: “In humans, autophagy markers in muscle were reduced, likely in response to weight loss.” The same study found intermittent fasting did activate autophagy markers in mouse liver.
NoteThis is the most direct human muscle-biopsy test of intermittent fasting and autophagy available. It did not show induction. Two small 2025 pilot studies measuring flux in blood cells report possible increases; both are described by their own authors as exploratory or preliminary, and one was funded by a commercial fasting-product manufacturer.
What this adds up to
DemonstratedTwo years of moderate caloric restriction improves cardiometabolic risk markers in healthy non-obese adults — with measurable bone density loss alongside.
PlausibleThat restriction slows some ageing processes in primates. The two monkey studies agree on health benefits; they disagree on survival, and the disagreement is explained by design rather than resolved in favour of either.
Not establishedThat any fasting schedule induces autophagy in human tissue. The mechanism is solid in yeast and rodents. In human muscle it has been tested and the markers fell. There is no human evidence for a threshold fasting duration — the widely repeated “autophagy begins at sixteen hours” has no human tissue data behind it.
Not establishedThat 16:8 time-restricted eating outperforms eating three ordinary meals. Tested head to head, it did not — and more of what it did take off was lean mass.
Not establishedThat caloric restriction extends human lifespan. No trial has tested it and none realistically can: CALERIE ran two years, adherence reached less than half the prescribed dose, and ageing takes decades. The largest randomised test of sustained calorie restriction against hard outcomes — Look AHEAD, 5,145 people, median 9.6 years — was stopped for futility with a cardiovascular hazard ratio of 0.95 (95% CI 0.83–1.09).
Look AHEAD Research Group. New England Journal of Medicine 2013;369:145–154. Note that Look AHEAD tested a combined diet-and-exercise package in people with type 2 diabetes, not caloric restriction alone in healthy adults.
Entry 03 · Practice
Meditation, stress and the body
“Meditation lowers cortisol, lengthens telomeres and changes gene expression.”
This is the entry where I have the most to lose by being accurate, since I teach these practices. It is also the one where accuracy matters most, for the same reason.
The strongest evidence, and its ceiling
The meta-analysis that set the standard
Goyal M et al. JAMA Internal Medicine 2014;174(3):357–368 · 47 randomised trials, 3,515 participants — every one with an active control
Against controls matched for time and attention, mindfulness meditation produced moderate evidence of benefit for anxiety (effect size 0.38 at eight weeks), depression (0.30) and pain (0.33). Effects persisted at three to six months, smaller.
Evidence was low for stress and distress, and low or insufficient for positive mood, attention, substance use, eating habits, sleep and weight. Mantra-based programmes, including Transcendental Meditation, showed no demonstrable effect on any outcome examined.
The sentence that mattersVerbatim: “We found no evidence that meditation programs were better than any active treatment (ie, drugs, exercise, and other behavioral therapies).” Meditation works about as well as exercise. It has not been shown to work better.
The biological findings, examined
Cortisol
Koncz A et al. Health Psychology Review 2021;15(1):56–84 · Rogerson O et al. Psychoneuroendocrinology 2024;159:106415
Blood cortisol showed a moderate effect (g = 0.62, 10 studies). Salivary cortisol — the larger and better-powered body of evidence, 21 studies and 1,163 people — was not statistically significant (g = 0.18, 95% CI −0.04 to 0.40). Correcting for likely publication bias moved the salivary estimate to below zero.
A larger 2024 review across 58 trials and 3,508 participants found a small-to-moderate effect for mindfulness and meditation (g = 0.345), with no evidence of publication bias, and — against expectation — stronger effects in trials with active controls than passive ones.
Fair statementA small, heterogeneous, measurement-dependent effect on cortisol. Not an established physiological fact, and the first authors describe their own findings as preliminary.
Telomerase — the study behind the headline
Jacobs TL et al. Psychoneuroendocrinology 2011;36(5):664–681 · 60 adults, three-month retreat vs wait-list
Telomerase activity was higher in retreat participants than controls at the end of the retreat (P < 0.05).
Two facts that change the readingBaseline telomerase was never measured — the authors say so. This is a single after-the-fact comparison between groups, not a demonstrated increase in anyone. And the control was a wait-list, not an active control: it holds nothing constant for expectancy, rest, diet, altitude or three months away from ordinary life. Blood sample attrition reduced the analysed groups to 20 and 26.
Gene expression after a day of practice
Kaliman P et al. Psychoneuroendocrinology 2014;40:96–107 · 19 experienced meditators, 21 untrained controls
After eight hours, meditators showed reduced HDAC2, HDAC3 and HDAC9, altered histone modification, and decreased RIPK2 and COX2 relative to controls. The groups were indistinguishable at baseline.
What it cannot separateExpert meditators were compared with untrained people, so the result is entangled with who those practitioners already were — years of practice, self-selection, temperament, lifestyle. Nineteen against twenty-one, across a panel of genes. And the association between RIPK2, HDAC2 and faster cortisol recovery held in both groups — three words routinely dropped when this study is summarised.
What the field says about itself
The sharpest critique of meditation research was written by meditation researchers.
“Mind the Hype”
Van Dam NT et al. Perspectives on Psychological Science 2018;13(1):36–61 · fifteen authors, several of them prominent in the field
There is no agreed technical definition of mindfulness — a five-minute app exercise and a three-month retreat carry the same label. Only 9% of mindfulness-intervention research has reached the stage of testing efficacy against an active control. Fewer than a quarter of meditation trials actively assess adverse events, and passive reporting can undercount them by more than twentyfold.
Publication bias, quantified
Coronado-Montoya S et al. PLOS ONE 2016;11(4):e0153220
Of 124 published trials, 108 reported at least one positive outcome — about 1.6 times the number expected. Of 21 registered trials, 13 remained unpublished thirty months after completion.
Adverse effects, when actually asked about
Britton WB et al. Clinical Psychological Science 2021;9(6) · structured interviews after eight-week programmes, n=96
83% reported at least one meditation-related side effect. 37% reported a negative impact on functioning. Between 6% and 14% reported lasting adverse effects.
Context, from the authorsThose rates are comparable to other psychological treatments — the finding is not that meditation is dangerous. It is that open-ended questioning fails to surface this, and most trials never ask. Anyone teaching these practices should know the number and should ask.
What this adds up to
DemonstratedMeditation produces small-to-moderate reductions in anxiety, depression and pain that survive comparison with controls matched for time and attention. That is a real result and better than most of what is sold.
PlausibleA small effect on cortisol. Real in blood, absent in saliva, and vulnerable to publication bias in the medium where most of the data sits.
Not establishedThat meditation outperforms exercise, relaxation training, cognitive behavioural therapy or medication. No meta-analysis has shown this.
Not establishedThat meditation raises telomerase or slows cellular ageing. The telomerase study measured no baseline and used a wait-list control. The gene-expression work compares experts with novices in groups of about twenty. Neither would support a causal claim in any other field.
Not establishedThat these practices are without cost. When asked properly, most people report some difficult effect, and a minority report lasting ones.
None of this is a reason to stop practising. The contemplative traditions did not make their claims in the language of randomised trials and are not answerable to them. What they are answerable to is honesty about which claims are which — and a practice defended with weak evidence is weaker than one defended with none.