Research library

Protocol: what they did

Compared low/moderate-intensity continuous endurance training (ET), high-intensity interval or continuous training (HIT), and sprint interval training (SIT) on skeletal muscle mitochondrial content and capillarization, with VO2max, fibre cross-sectional area and fibre type as secondary outcomes. Meta-regression on fitness level, age, sex and disease.

What was found

• Training load (volume x intensity) predicts change in mitochondrial content and VO2max. The relationship is less clear for capillary adaptation.
• SIT produced faster initial gains; HIT showed slower but steady improvement over a greater number of training weeks.
• The magnitude of change is largely determined by initial fitness level, with greater changes in individuals with lower initial fitness.
• The ability to adapt to exercise training is maintained throughout life, irrespective of sex and presence of disease.

Why it matters

This is the actionable T1 finding in the whole library. Everything else filed under mitochondria is observational or theoretical; this establishes an intervention that measurably increases mitochondrial content in humans, with a dose relationship.

Two details carry most of the practical weight. The less fit you are, the more you gain, which inverts the usual assumption that adaptation favours the already trained. And adaptation is preserved across age, sex and disease, which removes the main reasons people assume it will not work for them.

Limits

Search closed February 2022. Mitochondrial content is a structural measure and does not by itself establish improved clinical outcomes, though VO2max moved alongside it. Heterogeneous protocols and biopsy methods. No head-to-head prescription optimum falls out of it beyond "load matters".

What was found

• HbA1c -0.29%, with the largest effect at 3 months.
• Fasting blood glucose -7.12 mg/dL.
• Weight loss greatest at 3 months and attenuated thereafter.
• At 3 months, lower actual carbohydrate intake related to larger HbA1c and weight reductions (dose-response).
• Conclusion: modest, short-term metabolic benefits, diminishing over time.

The adherence finding, which matters more than the effect size

"No arm sustained carbohydrate intake < 10% of total energy; therefore, findings generalize to low-to-moderate carbohydrate rather than very-low-carbohydrate."

Across 27 randomised trials and 2,870 people, not one arm maintained actual ketogenic-level restriction. This is the single most useful sentence in the diabetes literature for interpreting real-world claims: what gets studied, and what people actually do, is low-to-moderate carbohydrate. Trials described as ketogenic are frequently not, measured by intake.

Why it matters

Establishes both that low-carbohydrate eating works in type 2 diabetes and the size and duration of the effect: real, modest, front-loaded, fading. The dose-response at 3 months supports carbohydrate reduction as the active variable at short horizons; the attenuation by later timepoints is what the long-term syntheses confirm (see 39046308_ichikawa-2024_...).

Limits

Heterogeneous definitions of "low carbohydrate", which is the problem the paper itself documents. Cultural subgrouping is exploratory. No cardiovascular or mortality endpoints. Attenuation over time cannot be separated from declining adherence.

What was found

Across 685 randomised trials, creatine supplementation showed no greater prevalence of adverse events than placebo, including the perennial accusations: no excess gastrointestinal, renal or muscle-cramp events.

Why it matters

The safety question is the practical barrier for most people, and this is as close to a complete answer as supplement science gets. Caveat carried honestly: the author group overlaps with the field's advocates and industry funding exists in the underlying literature.

Limits

Trial durations mostly weeks to months; healthy or athletic populations dominate; renal disease patients excluded by the underlying trials.

Podcast context

Filed from the Candow episode (Diary of a CEO, 2026-06-15), where the safety myths (kidneys, cramps, hydration) were the core theme and accurately described.

What was found

Creatine plus resistance training added about +1.14 kg of lean mass versus resistance training alone in the pooled randomised evidence.

Why it matters

A real, modest, highly replicated effect - the reference point for what "works" means in supplementation. Note the direction of the arithmetic: creatine amplifies training; it does nothing without the stimulus.

Limits

Modest absolute size; heterogeneous training programs; lean mass includes intracellular water, which creatine raises by mechanism.

Podcast context

Filed from the Candow episode. The guest's rendering was accurate, including the point that training does the work.

Protocol: what they did

Network meta-analysis comparing classic ketogenic diet, modified Atkins diet (MAD) and other dietary therapies against care as usual, and indirectly against each other.

What was found

Versus care as usual, for seizure reduction of 50% or higher:

• MAD: OR 11.3 (95% CI 5.1-25.1)
• Ketogenic diet: OR 8.6 (95% CI 3.7-20.0)

For 90% or higher reduction: ketogenic diet OR 6.5 (2.3-18.0), MAD OR 5.1 (2.2-12.0).

Against that:

• Discontinuation from adverse events was significantly higher: ketogenic diet OR 8.6 (1.8-40.6), MAD OR 6.5 (1.4-31.2).
• Indirectly, no significant difference between the dietary therapies themselves.
• Authors' conclusion: MAD may be the sounder option, with better tolerability and comparable efficacy.

Why it matters

This is the answer to "does keto work". In drug-resistant epilepsy in children it works, the effect is large, and it is randomised. Any discussion of ketogenic diets that leaves this out is incomplete, and any discussion that generalises from it to other indications is overreaching.

The second finding matters as much as the first: the diet that performs best is the less strict one. Adherence and tolerability are part of efficacy.

Limits

Children only, short term, and the trials pooled are individually small. See 32588435_martin-mcgill-2020_cochrane-ketogenic-drug-resistant-epilepsy.md: Cochrane grades the same body of evidence low to very low certainty because of small samples and risk of bias. Two credible syntheses, one reporting large odds ratios and one flagging low certainty, are both correct and describe the same underlying trials.

Protocol: what they did

Melanopic irradiance controlled independently of display luminance and colour, 4 h before habitual bedtime. Each subject exposed to one of four luminance levels (dim light, smartphone, tablet, computer screen illuminance) at either a low or a high melanopic setting.

What was found

Low melanopic light versus high, at matched luminance:

• shortened time to fall asleep
• attenuated evening melatonin suppression
• reduced morning melatonin
• advanced evening melatonin onset
• decreased alertness

Dose-dependent increases in sleep latency, reductions in melatonin and delays in melatonin onset as a function of melanopic irradiance. Subjective alertness was not dose-dependent.

Why it matters

This is the paper that makes the others actionable. It identifies melanopic irradiance (melanopic EDI), not lux and not "blue light", as the parameter that governs the effect, and shows the display can be spectrally tuned without compromising visual appearance. That converts the problem from "use screens less" into "set the right metric on the screen". It also confirms the Cajochen 2011 tradeoff in reverse: the low-melanopic setting that helps sleep also makes you less alert in the evening.

Limits

All male, healthy, single evening, laboratory. Melanopic EDI is still poorly exposed by consumer hardware, so applying it requires measurement or manufacturer data that usually is not published.

Protocol: what they did

Integrated data from 17 cohorts of mitochondrial disease patients (n = 690) to test resting energy expenditure, then reproduced the phenomenon longitudinally in patient-derived fibroblasts, disrupting OxPhos genetically and pharmacologically.

What was found

• Mitochondrial disorders increase resting energy expenditure: hypermetabolism.
• Disrupting OxPhos approximately doubles cellular energy expenditure.
• This happens despite near-normal OxPhos coupling efficiency, which excludes uncoupling as the general mechanism.
• Hypermetabolism is associated with mtDNA instability, activation of the integrated stress response (ISR), and increased secretion of age-related cytokines and metabokines including GDF15.
• OxPhos defects accelerate telomere erosion and epigenetic aging per cell division.

Why it matters

The empirical backbone of the whole "energy" framework. It inverts the intuitive model: mitochondrial disease patients are not simply short of ATP, they are burning more energy, and the excess expenditure is what tracks with accelerated aging. The exclusion of uncoupling is what makes this more than an association. It supplies the mechanism that the 2024 Nat Metab commentary and the 2024 brain-body model build on.

Limits

The 690-patient figure is pooled across heterogeneous cohorts with differing methods. Cell work is fibroblasts. "Reduce lifespan" in the title mixes cellular lifespan and patient outcomes, which are not the same endpoint. Generalisation from primary mitochondrial disease to ordinary aging is an extrapolation, and is presented as such.

What was found

Gastroparesis (OR 4.64), dysphagia (3.58), IBS without diarrhoea (3.53) and constipation (3.32) preceded Parkinson's disease specifically - versus matched controls, Alzheimer's and cerebrovascular disease - with the associations confirmed prospectively within 5 years.

Why it matters

The strongest human evidence for gut-first Parkinson's, made credible by its specificity (the same syndromes do not precede Alzheimer's) and by animal mechanism: vagotomy blocks gut-to-brain alpha-synuclein spread in mice (Kim 2019, Neuron, PMID 31255487, not filed).

Limits

Records-based; GI syndromes are common and the absolute risk translation is modest; prodromal reverse causation cannot be fully excluded.

Podcast context

Filed from the Spector episode. The direction is well supported; his "90% had gut problems 10 years before" appears in no retrieved paper - the evidence is elevated relative risk, not 90% prevalence.

What was found

LT4/LT3 combination versus LT4 monotherapy:

• No difference in clinical status, quality of life, psychological distress or depressive symptoms.
• Patient preference favoured combination therapy (43% vs 23%).
• Adverse events similar.

Why it matters

This is the reference synthesis for the question. The pattern it fixes - null on every measured outcome, yet a preference gap - is the core puzzle of this topic: whatever patients are responding to, the trial instruments do not capture it, or it is not specific to the T3 content.

Limits

Included trials are short (weeks to months), dosing regimens heterogeneous (mostly once-daily plain LT3, which cannot mimic physiological T3 rhythm), and none enriched for the dissatisfied-on-LT4 population the question is really about. Preference is vulnerable to period effects and incomplete blinding.

What was found

• Ketogenic diets could demonstrate effectiveness in children with drug-resistant epilepsy.
• Evidence in adults remains uncertain.
• Certainty of evidence: low to very low, because of small sample sizes, risk of bias and imprecision. In Cochrane's words, "the true effects of the diets could be substantially different to that reported".
• Adverse effects: in the adult sub-studies, specific events (constipation, diarrhoea, change in seizure pattern) each reached 17.5% to 20% of participants; the review reports no pooled review-wide adverse-effect rate.
• More palatable variants such as MAD may have similar effect on seizure control with fewer adverse effects, but this needs more investigation.
• Practical conclusion retained: for people with drug-resistant epilepsy or who are unsuitable for surgery, ketogenic diets remain a valid option.

Why it matters

Filed alongside the JAMA Pediatr network meta-analysis on purpose. These two syntheses cover largely the same trials and read very differently: one reports odds ratios of 8 to 11, the other says low to very low certainty.

Both are right. Large effect sizes measured in small, imprecise, partly biased trials are exactly what produces this pattern. It is the clearest example in the library of why effect size and certainty are separate axes, and why an impressive odds ratio is not by itself a strong evidence base.

Limits

Cochrane's inclusion criteria are strict, so it captures fewer studies than broader reviews. 2020, so it predates several later trials.

Protocol: what they did

Systematic search of four databases from inception to June 2016, on the amount and timing of light exposure versus objective and subjective sleep outcomes.

What was found

• Bright light (>1000 lux) is associated with better objectively assessed sleep than dim (<100 lux) or moderate (100-1000 lux) light, and with better subjective sleep than moderate light.
• Amount is moderated by timing: brighter morning light shifts the sleep period earlier, brighter evening light shifts it later.
• Subjectively: brighter morning light associated with self-reported sleep improvement, brighter evening light with worse self-reported sleep.

Why it matters

The counterweight to a purely restrictive reading of the other papers. Light is not the problem, mistimed light is. Daytime bright-light exposure is a positive intervention with its own evidence, so the practical shape is "more light early, less light late", not "less light".

Limits

Search ends June 2016, so it predates most melanopic-metric work. Observational studies are pooled with experiments. Healthy adults only.

Protocol: what they did

Systematic search on light and circadian rhythm (melatonin, REM sleep). 128 articles collected. Melatonin studied in 122, REM sleep in 13. Quality filter applied: minimum 20 subjects and reported light spectrum/wavelength plus illuminance/irradiance/photon density. Only 13 melatonin studies and 2 REM studies survived.

What was found

• 39 of 128 studies had 10 or fewer subjects. Very few studies in children or older adults; most subjects were in their twenties or thirties.
• 2 h of evening blue light (460 nm) suppresses melatonin; the maximum effect is at the shortest wavelengths (424 nm, violet), not at 460 nm.
• Melatonin recovers within about 15 min after the exposure stops, suggesting a short-term rather than lasting suppressive effect.
• Melatonin secretion and suppression decline with age, but the light-induced phase advance is not impaired with age.
• Even 631 nm (red) light and intermittent exposures produce circadian resetting.
• 5-10 lux at night with eyes closed still produces a circadian response.

Why it matters

The most important corrective document in this folder. Three things it changes: the "blue light" framing is too narrow (red and intermittent light also reset the clock); the suppression is short-lived once light stops; and eyes-closed exposure at very low levels still counts, which makes bedroom darkness a real variable. It also documents how thin the evidence base is, which should temper every other claim here.

Limits

Written partly to support life cycle assessment methodology for light pollution, so its framing is environmental as well as clinical. Restricted to melatonin and REM sleep as outcomes.

Protocol: what they did

Systematic review restricted to experimentally controlled studies, so that causal inference about induced psychological stress on mitochondria was possible.

What was found

• 23 studies met criteria. All of them were in male laboratory animals.
• 19 showed adverse effects of stress on mitochondria, 4 showed increases in function or size.
• Effects were particularly concentrated in the brain.
• In humans: only 6 observational studies, none experimental, and most measured markers that do not reflect mitochondrial function, chiefly mtDNA copy number.

Why it matters

The companion piece to the framework paper (PMID 29389735) and the reason to trust the author. Published simultaneously with his own conceptual model, it documents that the human evidence for that model did not yet exist. It is the baseline against which the following years of work should be measured.

Limits

The restriction to controlled designs is what makes it rigorous and also what makes it nearly empty on the human side. Male-animal-only evidence base. Now dated: much of the human work it calls for has since been done, largely by this group. See ../authors/picard-martin.csv for what followed.

What was found

Partial sleep deprivation increased next-day energy intake by a pooled +385 kcal/day (95% CI ~250-500) with no change in energy expenditure - the energy gap is all appetite.

Why it matters

The quantitative bridge between short sleep and weight: people do not burn less when sleep-deprived, they eat more. Converges with Covassin 2022 (the causal visceral-fat trial) and gives the sleep topic its T1: the intake effect is synthesis-level and replicated.

Limits

Laboratory settings, short protocols; free-living extrapolation approximate; the widely quoted "250-400 kcal" range is this study's confidence interval, not a range of separate findings.

Podcast context

Filed from the St-Onge Huberman episode (2026-06-08), where the guest quoted her own field accurately to within rounding.

Protocol: what they did

Compared reading an electronic book on a light-emitting device with reading a printed book in the hours before bedtime, under inpatient conditions. Endpoints: sleep latency, evening sleepiness, melatonin, circadian phase, and next-morning alertness.

What was found

Reading the light-emitting eReader, versus print:

• took longer to fall asleep
• reduced evening sleepiness
• suppressed melatonin secretion
• delayed circadian phase
• reduced next-morning alertness

Why it matters

The most complete causal chain in this set: it does not stop at the melatonin biomarker, it carries through to how the person functions the following morning. That is what makes it the standard citation for evening screen advice. It also isolates the light variable cleanly, since the cognitive activity (reading) is held constant and only the medium changes.

Limits

Tablet at close range in a controlled inpatient setting, several hours per night. Real-world exposure varies enormously in distance, duration and brightness, so effect sizes should not be transferred directly.

What was found

Each 5 kg lower grip strength carried a hazard ratio of 1.16 for all-cause mortality after adjustment, and grip strength out-predicted systolic blood pressure for death.

Why it matters

The strongest simple field measure of the strength-mortality association, in 17 countries across income levels. It complements this topic's training evidence: Molmen 2025 shows training rebuilds capacity; Leong shows how much capacity predicts.

The marker trap applies: grip strength predicts outcomes because it proxies whole-body strength and health. Training the grip itself optimises the gauge, not the engine - the same logic the library applies to GDF15.

Limits

Observational; cannot separate strength from everything strength correlates with; 4-year median follow-up.

Podcast context

Filed from the Galpin episode, where it was cited accurately, including the warning against training the test.

Protocol: what they did

Compared daily melatonin profiles in people living under room light (<200 lux) versus dim light (<3 lux) for the 8 h preceding bedtime.

What was found

• Room light delayed melatonin onset in 99.0% of individuals.
• Melatonin duration shortened by about 90 minutes.
• Light during the usual hours of sleep suppressed melatonin by more than 50% in 85% of trials.

Why it matters

The key number in this whole folder. The threshold is not "bright light", it is ordinary domestic room lighting, under 200 lux, which almost nobody thinks of as a light exposure. This is the paper that makes evening ambient light, not just screens, the target. The authors flag downstream implications for sleep, thermoregulation, blood pressure and glucose homeostasis, though the study itself measured none of those.

Limits

Retrospective pooling of two studies rather than a purpose-built randomised trial. Young adults only, and light sensitivity declines with age. The health outcomes in the conclusion are inferred, not measured.

What was found

Across 11 randomised trials, no difference between combination and monotherapy in bodily pain, depression, anxiety, fatigue, quality of life, body weight or lipids. Adverse events similar. Conclusion: LT4 monotherapy should remain the treatment of choice.

Why it matters

The original null. Every later synthesis (Millan-Alanis 2021, Nassar 2024) has reproduced it, which is what earns the group-level claim T1: not one large paper, but the same answer from independent teams across twenty years.

Limits

Trials from the era of once-daily LT3 substitution with variable dose ratios; several underpowered alone. None targeted patients with persistent symptoms, the group for whom the question is actually asked.

What was found

Patients on 'adequate' LT4 (normal TSH) showed impaired psychological well-being versus matched controls: GHQ caseness 34.4% vs 25.6%.

Why it matters

This established the real phenomenon underneath all combination-therapy marketing: a measurable minority feels worse on standard treatment despite normal biochemistry. The library's trial evidence then shows adding T3 has not fixed this at the group level - the complaint is validated, the proposed remedy is not.

Limits

Cross-sectional: cannot separate the effect of therapy from the effect of having a condition requiring therapy. Autoimmunity, comorbidity and diagnosis labelling are all candidate confounders.

Protocol: what they did

Full-field monochromatic light exposures between 02:00 and 03:30 with pupils dilated. Blood melatonin before and after. Each subject tested at seven or more irradiances of a single wavelength, at least one week between nights. 627 nighttime melatonin suppression tests across 420-600 nm.

What was found

• Eight univariant sigmoidal fluence-response curves, R2 = 0.81-0.95.
• The action spectrum fits an opsin template (R2 = 0.91) peaking at 446-477 nm, distinct from rod and cone photopigments.
• Conclusion: a separate retinaldehyde-based photopigment mediates circadian photoreception in humans.

Why it matters

This is the foundational paper. Every later "blue light" claim traces back to this action spectrum. It is the reason melanopsin and ipRGCs became the unit of analysis, and the reason evening-light advice targets short wavelengths specifically rather than brightness alone.

Limits

Pupils were pharmacologically dilated and light was monochromatic and full-field, i.e. a deliberately extreme laboratory condition, not a domestic one. Young adults only. It establishes the spectrum, not the size of any real-world effect.

Protocol: what they did

Plasma GDF15 measured at baseline in UK Biobank participants, followed for a median of 14 years. Cox proportional hazards models for incident brain disorders. Mediation analysis over biochemical and haematological pathways. One-sample Mendelian randomization to test for causality.

What was found

Hazard ratios per 1-unit increase in log2-transformed GDF15, all P < 0.001:

Mediation: lipid- and inflammation-related markers partially mediate. HDL-C accounted for ~7.5% of the depression association and ~11.5% of the sleep-disorder association; neutrophil count showed larger mediation estimates across multiple outcomes.

Mendelian randomization did not support a direct causal effect of GDF15.

Why it matters

Two findings that pull in opposite directions, and both belong in this topic.

The association is real, large and prospective. GDF15 measured once predicts dementia, stroke, Alzheimer's, epilepsy, depression, anxiety, Parkinson's and sleep disorders more than a decade later. That is the strongest evidence filed for GDF15 as a marker of something systemically important, and it is consistent with the framing in 41056945_picard-2025_energy-resistance-principle.md, where GDF15 rises with disease-causing stressors.

The causality test is negative. MR does not support GDF15 causing these outcomes. That distinction is the whole practical question. If GDF15 is the signal by which the brain learns that energy is short, then it is a readout, and lowering it treats the gauge rather than the engine. This is the empirical version of the caution the Picard group themselves raised in "Potential Risks of Blocking GDF15-Based Brain Energy Sensing" (J Am Geriatr Soc 2026), listed in ../authors/picard-martin.csv.

The mediation result adds a third thread: HDL-C and neutrophil count carry part of the association. HDL is also the one marker that moves consistently under carbohydrate restriction across the diabetes syntheses in ../topics/ketogenic-diet-by-indication.md. Whether those are the same story is untested and should not be assumed.

Limits

Single cohort. UK Biobank is large but is a volunteer sample, healthier and less deprived than the general population, and overwhelmingly of European ancestry, which limits generalisation and also constrains the MR.

One-sample MR is weaker than two-sample MR with independent samples, and a null MR is not proof of no causal effect: it can also reflect weak instruments, insufficient power, or a causal effect confined to a subgroup or a time window. Read it as "causality is not supported", not "causality is excluded".

Exact sample size, event counts and covariate adjustment are not stated in the abstract and were not verified against the full text. HRs are per log2 unit, which is not an intuitive scale and should not be read as per-unit clinical risk.

What was found

• "RCTs showed no significant OS/PFS differences versus controls", though exploratory analyses suggested better outcomes with lower glucose levels.
• Ketosis was commonly achieved; glucose reductions were variable.
• GKI was rarely reported and target values were seldom reached outside fasting periods.
• Weight and BMI decreased modestly or stayed stable; adverse events mostly mild; tolerability acceptable.
• Conclusion: evidence is heterogeneous and limited; ketosis and safety are achievable but survival or quality of life benefits remain unproven.

Why it matters

Two things worth carrying forward.

First, it confirms on the whole randomised literature what ERGO2 found in one trial. Second, and less often noticed: the Glucose Ketone Index was seldom reported and its targets were rarely reached. The metric proposed in 25798181_meidenbauer-2015_glucose-ketone-index.md as the way to monitor therapy is largely absent from the trials that would validate it, and where measured, the target was not achieved outside fasting.

The glucose signal recurring here, as in ERGO2 and in Seyfried's own 2003 mouse experiment, is the most consistent thread in this topic.

Limits

Systematic review without meta-analysis for survival. Heterogeneous studies, varying diet compositions. 2026 and not yet independently appraised.

Protocol: what they did

Used RNA-seq from fibroblasts under eleven metabolic perturbations, including genetic and pharmacological OxPhos defects, to define the gene network of the integrated stress response (ISR). Built an ISR-GDF15 index, validated it against optogenetic and small-molecule activation of ISR kinases, then applied it across 44 postmortem human tissues.

What was found

• The ISR has multiple arms; the GDF15 arm can be isolated from the rest.
• The index has rapid kinetics, preceding GDF15 gene expression.
• It correlates with age across human tissues.
• Upregulated in the heart of people with acute causes of death in the emergency room; upregulated in the brain of people who died after protracted inpatient stays.

Why it matters

Turns GDF15 from a single circulating biomarker into a resolvable transcriptional program with tissue specificity, and provides a usable instrument (the index). The acute-versus-protracted death contrast is the intriguing result: the same stress axis engages different organs depending on how the stress unfolded in time.

Limits

Fibroblast-derived index applied to postmortem tissue, where cause and manner of death are strong confounders and the acute/protracted comparison is not randomised. 2026 publication with no independent replication. GDF15's role remains correlational in humans; see the companion caution in the same year's "Potential Risks of Blocking GDF15-Based Brain Energy Sensing" (J Am Geriatr Soc), listed in ../authors/picard-martin.csv.

What was found

Weighted mean differences, all in favour of the ketogenic diet:

FSH did not change significantly.

Why it matters

Convergence is what makes this strong. Three further independent meta-analyses reach the same conclusions on the same outcomes:

• Gynecol Obstet Invest 2025 (PMID 39978319), see the companion file
• Reprod Biol Endocrinol 2025 (PMID 40394635)
• Clin Nutr 2026 (PMID 41483483)

Four separate teams, four separate syntheses, same direction on weight, insulin resistance and androgens. Each individual meta-analysis is T2; the claim they jointly support is T1. That distinction is the reason the tier rates papers rather than claims, and it is worth reading ../topics/ketogenic-diet-by-indication.md for the aggregate view.

Limits

Underlying trials are small and short. Heterogeneity was high for most outcomes in the companion reviews (I2 up to 90% for insulin resistance in the Clin Nutr synthesis). Populations are overweight or obese women, so weight loss itself may mediate much of the hormonal effect, and none of these reviews establish that ketosis rather than caloric deficit is responsible. No long-term or fertility-outcome data.

What was found

Daily multivitamin-multimineral supplementation slowed second-generation epigenetic clocks (PCGrimAge, PCPhenoAge) versus placebo in the COSMOS ancillary cohort.

Why it matters

Together with Bischoff-Ferrari 2025, this makes two independent randomised trials in which cheap, low-risk supplementation nudged epigenetic aging measures. Two trials, two different supplements, the same class of surrogate endpoint: suggestive of clock responsiveness, not yet of health benefit.

Limits

Ancillary endpoint in under 1,000 of COSMOS's 21,000+ participants; clocks are surrogates; the parent trial's cognitive findings are separate.

Podcast context

Filed from the Rhonda Patrick episode, where it was cited as part of the case that micronutrient sufficiency matters for aging.

What was found

A 30+ ingredient whole-food plant blend changed the abundance of 57 species-level genome bins versus 14 (control) and 4 (probiotic), and improved self-reported indigestion, constipation, heartburn, flatulence and energy. The microbiome health ranks of the changing species did not differ significantly between groups.

Why it matters

A real randomised test of a food-diversity intervention with an active comparator - rare in this field. More species moved, but the study's own health-direction metric could not distinguish the arms: movement is proven, benefit is not yet.

Limits

Sponsor-run (ZOE); 6 weeks; symptom outcomes self-reported and unblindable; the health-ranking instrument is the sponsor's own (Asnicar 2026).

Podcast context

Filed from the Spector episode, where it was presented as dramatic microbiome improvement with mood gains; the paper's own primary contrast is null on the health-ranked direction and mood is not among the significant outcomes. The host disclosed on air that he is a ZOE investor.

What was found

A ranking of gut microbial species most favourably and unfavourably associated with diet and health markers (the "Microbiome Health Ranking"), reproducible across US/UK cohorts and against public datasets.

Why it matters

Currently the largest species-level association map between the microbiome and health markers. The authors state plainly that causal inference is not possible from this design.

Limits

Associational; health markers, not disease outcomes; the ranking is also the sponsor's product-evaluation instrument, which circularises its use in the BIOME trial (Creedon 2026).

Podcast context

Filed from the Spector episode, where it was described as "50 good and 50 bad bugs" and treated as an outcome measure - the paper frames it as hypothesis-generating.

What was found

• PGC-1alpha was the most consistently reported marker of mitochondrial biogenesis.
• Pooled effect after endurance exercise: Hedge's g = 1.17 (95% CI 0.14-2.19), a large effect, I2 = 84.5%.
• Subgroups: interval g = 1.29, continuous g = 1.01, no significant difference between modalities (p > 0.05).

Why it matters

Supplies the molecular mechanism under Mølmen 2025's structural finding. Together they cover the same claim at two levels: exercise raises PGC-1alpha signalling, and mitochondrial content increases.

The practical read is the null result between modalities: interval and continuous training both work, and the choice can be made on preference and adherence rather than on physiology.

Limits

PGC-1alpha is a surrogate, an expression marker rather than a functional or structural outcome, and acute expression changes do not guarantee lasting adaptation. I2 = 84.5% is substantial heterogeneity, and the confidence interval (0.14-2.19) is wide enough that the true effect could be modest. Healthy individuals only.

Protocol: what they did

Physically partitioned a frozen human coronal hemisphere section into 703 voxels at neuroimaging resolution (3 x 3 x 3 mm) and profiled each one: OXPHOS enzyme activities, mtDNA and volume density, mitochondria-specific respiratory capacity. Then built a backwards linear regression model integrating neuroimaging modalities to extrapolate a brain-wide map.

What was found

• The human brain contains diverse mitochondrial phenotypes driven by both topology and cell type.
• Grey matter contains over 50% more mitochondria than white matter, and those mitochondria are biochemically optimised for energy transformation, particularly in recently evolved cortical regions.
• The model predicted mitochondrial characteristics in an independent region of the same brain.
• Output is the MitoBrainMap, public at humanmitobrainmap.bcblab.com

Why it matters

Bridges the scale gap between cognitive neuroscience (voxels) and cell biology (organelles) by making them literally the same unit. That is the methodological contribution: it lets mitochondrial biology be related to existing neuroimaging data rather than requiring new modalities.

Limits

One donor brain, one hemisphere section. The internal validation predicts another region of the same brain, which tests the model but not between-person generalisation. Postmortem enzyme activities are not in vivo respiration. Individual variation, sex differences and disease states are all still open. Note there is a published Author Correction to this paper (2025).

What was found

Standardised mean differences, all significant:

• FSH raised (SMD -0.76, 95% CI -1.25 to -0.28, P = 0.002)
• LH/FSH ratio decreased (SMD 2.04, 95% CI 1.04-3.03, P < 0.001)
• Free testosterone decreased (SMD 0.57, 0.28-0.86, P < 0.001)
• Total testosterone decreased (SMD 0.54, 0.28-0.80, P < 0.001)
• SHBG increased (SMD 0.79, P < 0.001)

Conclusion: substantial benefit across all anthropometric, metabolic and hormonal parameters evaluated.

Why it matters

Filed specifically as corroboration. Its value is that it is a different team, different journal, different statistical approach (SMD rather than WMD) reaching the same conclusion as PMID 41249157 and two others.

In a library where most topics rest on a single group, four independent confirmations is the strongest pattern available, and PCOS is currently the only place it occurs.

Limits

Data on estrogens, progesterone, DHEAS and pregnancies were too scarce to meta-analyse, so the reproductive endpoints that matter most to patients are still open. Overweight and obese women only. Same unresolved question as the companion review: how much is ketosis and how much is weight loss.

What was found

Versus non-ketogenic diets, ketogenic diets significantly:

• reduced fat mass, visceral fat, insulin, blood glucose, fatigue and insomnia
• improved LDL and total cholesterol, TSH, protein uptake, ketosis events, emotional function and social function
• raised beta-hydroxybutyrate, confirming the intervention was delivered

CRP improvement was greater when the intervention ran more than 12 weeks, with a composition of 2-4% carbohydrate, 16-18% protein, 80-85% fat.

Why it matters

This is the evidence that a ketogenic diet helps cancer patients, and it is distinct from the claim that it treats cancer. Fatigue, insomnia, emotional and social function are outcomes patients care about directly, and they improved.

Filed because the topic would otherwise read as though ketogenic diets do nothing in cancer, which is not what the literature says. The correct statement is narrower and still meaningful: measurable benefit to symptoms, metabolic markers and body composition; no demonstrated survival benefit.

Limits

No survival endpoint. Pooled across cancer types and diet protocols. Unblindable interventions, so patient-reported outcomes such as fatigue, emotional and social function carry expectancy effects that cannot be separated out. Fat mass reduction in cancer patients is not unambiguously good, given that unintentional weight loss and cachexia predict mortality.

What was found

Liothyronine at regulated doses was not associated with excess mortality or adverse events across randomised and large cohort sources.

Why it matters

Removes the standard safety objection to properly dosed LT3, which shifts the combination question from "is it dangerous" to "does it help" - where the blinded evidence remains null. Minor heart-rate effects on DTE (Hoang 2013, Shakir 2021) sit outside this reassurance.

Limits

Cohort data dominate the long-term estimates, with the usual confounding; unregulated compounding and high-dose use are not covered.

What was found

52% preferred LT4+LT3 or DTE versus 24% LT4 (RR 2.20, 95% CI 1.38-3.52).

Why it matters

The maximal published estimate of the preference effect. The gap to Akirov 2019's chance-level estimate comes from different included trials and methods; until a preference-primary blinded trial exists, the honest position is that patients may genuinely prefer T3-containing regimens and the size of that effect is unknown.

Limits

Preference is subjective and partially unblindable; heterogeneous trials; network geometry thin for DTE.

What was found

For equivalent associations with outcomes, 1 minute of vigorous activity equated to roughly 4 minutes of moderate activity for mortality (about 8 for cardiovascular disease, 9 for type 2 diabetes, 3.5 for cancer endpoints).

Why it matters

An empirical exchange rate between intensity and time from objective measurement. It complements Molmen 2025's load principle: load is volume x intensity, and this quantifies how far intensity can substitute for time for people whose constraint is the calendar.

Limits

Observational associations from one cohort; device-measured but self-selected activity; not a randomised intensity trial.

Podcast context

Filed from the Rhonda Patrick episode, where the mortality figure was cited accurately.

What was found

Omega-3 supplementation (1 g/day) slowed several epigenetic aging clocks by 2.9 to 3.8 months over 3 years; vitamin D and exercise added small additional effects in combination.

Why it matters

One of the first randomised (rather than observational) links from a common supplement to epigenetic aging measures. The honest size: roughly one month of clock slowing per year of supplementation, on surrogate clocks whose clinical meaning is still being established.

Limits

Ancillary endpoint; adults over 70; epigenetic clocks are biomarkers, not outcomes; single trial.

Podcast context

Filed from the Rhonda Patrick episode, cited in support of omega-3; the clock-months effect size is the part podcast renditions tend to drop.

What was found

Toothbrushing was associated with slower global cognitive decline and lower dementia risk. Flossing, mouthwash, toothpicks and denture cleaning were not associated with dementia risk.

Why it matters

The oral-health-to-brain link exists in the data, but it runs through the toothbrush. Filed as the standing correction to "flossing halves dementia risk" claims: the best synthesis finds no flossing association at all.

Limits

Observational cohorts; oral hygiene self-reported; residual confounding by health behaviour is the obvious alternative explanation.

Podcast context

Filed from the Spector episode as the direct correction to its flossing claim ("not flossing doubles dementia risk"), which this synthesis contradicts.

What was found

Twelve weeks of creatine at standard doses produced no significant change in DHT, DHT:testosterone ratio, or measured hair parameters versus placebo in resistance-trained men.

Why it matters

The hair-loss fear traces to a single 2009 rugby study's DHT blip that was never replicated. This trial tested the claim directly - hormones and hair - and found nothing. Myth-testing trials are rare enough to be worth shelf space on their own.

Limits

12 weeks, young trained men, hair endpoints over a short window; genetic baldness plays out over years.

Podcast context

Filed from the Candow episode, where the myth's genealogy was told accurately.

What was found

Maternal diabetes was associated with RR 1.28 for any neurodevelopmental disorder in offspring (autism 1.25, ADHD 1.30). The authors state explicitly that causality is not established.

Why it matters

The definitive size estimate for a real association - and a masterclass in the gap between "associated" and the podcast framing ("rewiring the baby's brain"). Confounding by shared genetics, obesity and healthcare contact is unresolved by design.

Limits

Observational inputs; heterogeneous diabetes definitions; residual confounding acknowledged by the authors.

Podcast context

Filed from the Inchauspé episode, where the host quoted the numbers correctly and the causal framing exceeded the paper's own.

What was found

At more than 12 months, long-term low-carbohydrate versus control diets:

• HbA1c: no significant difference (SMD -0.11, 95% CI -0.33 to 0.11, P = 0.32).
• No significant difference in weight loss, blood pressure or LDL.
• HDL higher (SMD 0.22, 0.04-0.41, P = 0.02).
• Triglycerides lower (SMD -0.19, -0.37 to 0.02, P = 0.03).

Authors' conclusion: efficacy for dyslipidaemia, but they do not recommend long-term low-carbohydrate diets for glycaemic control.

Why it matters

The direct complement to the 3-month result in 40930317_mongkolsucharitkul-2025_.... Short term the diet lowers HbA1c and weight; past a year the glycaemic and weight advantages are gone and only the lipid changes persist.

That is a genuinely useful shape to know: it reframes low-carbohydrate eating in diabetes from a glycaemic treatment into a short-term glycaemic tool with a durable lipid effect. It also means the honest question is not "does it work" but "for how long, and compared with what".

Limits

Only 6 trials. A null result from 6 studies is weaker than it looks and could reflect insufficient power as much as absent effect. Long-term trials are also the ones most affected by declining adherence, so this measures the diet as actually followed rather than as prescribed, which is arguably the more relevant question but is not the same one.

Protocol: what they did

Combined longitudinal antemortem psychosocial assessments with postmortem brain proteomics (dorsolateral prefrontal cortex) in older adults, then repeated the analysis per cell type using single-nucleus RNA-seq.

What was found

• Higher well-being was linked to greater abundance of the mitochondrial OxPhos machinery; higher negative mood to lower OxPhos protein content.
• Combined positive and negative psychosocial factors explained 18 to 25% of the variance in complex I abundance, the entry point that energises brain mitochondria.
• Cell-type resolution reversed the picture: strong positive associations in glia, opposite associations in neurons.
• Because these cancel out, the associations were masked in bulk RNA-seq, meaning bulk-tissue studies underestimate true psychobiological effect sizes.

Why it matters

The human-brain evidence the 2018 systematic review said did not exist. The psychosocial exposure was measured years before death, which rules out reverse causation from the terminal period, though not from lifelong confounding. The glia-versus-neuron reversal is methodologically the most important part: it explains why earlier bulk studies found nothing, and it warns that a null result in bulk tissue is not a null result.

Limits

Observational. Association, not causation: the design cannot separate "well-being supports mitochondria" from "better mitochondria support well-being", and the authors present both directions as live. One brain region, older adults, postmortem tissue. Self-reported psychosocial measures.

What was found

Glycaemic control (SUCRA ranking):

1. Mediterranean diet, 88.15%
2. Moderate-carbohydrate diet, 83.3%
3. Low-carbohydrate diet, 55.7%

Anthropometric measurements:

1. Low-carbohydrate diet, 74.6%
2. Moderate-carbohydrate, 68.7%
3. Vegetarian, 57%

Important qualifier from the authors: differences between diets on anthropometric measures were "mostly small and often trivial".

Why it matters

Answers the comparative question the single-diet meta-analyses cannot. Low-carbohydrate eating is not the best option for glycaemic control in type 2 diabetes; Mediterranean ranks well above it. Where low-carbohydrate leads is weight and body composition, and even there the margin is described as often trivial.

This is the corrective to reading "low-carb lowers HbA1c versus usual care" as "low-carb is the best diet for diabetes". Both statements can be true of the same literature, and only the network analysis distinguishes them.

Limits

SUCRA rankings are probabilistic and routinely over-read: a difference in rank is not a demonstrated difference in effect, especially when the authors say the differences are trivial. Overweight and obese patients only. Indirect comparisons inherit the weaknesses of every trial in the network. Short follow-up throughout.

What was found

Combination therapy raised serum T3 and lowered T4/FT4, as expected. GHQ-28 scores (psychological distress) were lower on combination; heart rate, lipids and quality-of-life scores showed no significant differences.

Why it matters

Confirms the biochemical premise (monotherapy and combination produce different hormone profiles) while mostly reproducing the outcome nulls. Its isolated GHQ-28 finding is the one patient-centred signal in any synthesis, and it did not appear in the two larger meta-analyses; treat it as a lead, not a result.

Limits

Outcome sets heterogeneous across trials; the GHQ-28 pool is a subset of studies; no dose-response analysis; DTE represented by only two trials.

Protocol: what they did

Totally thyroidectomised patients randomised to LT4+LT3 or LT4+placebo for 24 weeks, double-blind (the LEVOLIO trial).

What was found

No significant differences in tissue markers of thyroid hormone action or in quality of life.

Why it matters

Thyroidectomised patients have zero endogenous T3 secretion, so if combination therapy helped anyone at the group level it should be visible here. It was not. This is the strongest recent blinded evidence that the older nulls were not an artefact of trial design vintage.

Limits

24 weeks; n=141; participants were not selected for persistent symptoms, which is the enrichment the 2021 consensus says a decisive trial needs.

What was found

A single high dose of creatine (0.35 g/kg) during overnight sleep deprivation improved cognitive performance and processing speed and shifted brain high-energy phosphates (31P-MRS) toward recovery.

Why it matters

The cleanest demonstration that oral creatine acutely reaches and affects the sleep-deprived brain's energetics. It anchors the plausible half of the brain-creatine story - acute rescue under energetic stress - as distinct from the unproven half (chronic cognitive enhancement in rested, replete adults).

Limits

n=15, single dose far above daily supplementation doses, sleep-deprivation context only; a lower 0.2 g/kg follow-up showed a smaller effect.

Podcast context

Filed from the Candow episode, where it was described accurately; his passing dismissal of the lower-dose follow-up slightly undersells its real but smaller effect.

What was found

Six weeks of mild sleep restriction (about 1.5 h less, within what many people do voluntarily) impaired insulin sensitivity in women, independently of changes in adiposity.

Why it matters

Covassin and Al Khatib test severe short-term restriction; this trial tests the dose people actually live at - and the metabolic cost is already measurable at six weeks. The women-only design is a feature (the population most under-represented in sleep-metabolism trials), and a limit on generalisation.

Limits

n=36, women only, six weeks, insulin sensitivity is a surrogate.

Podcast context

Filed from the St-Onge episode; on air the sex-specific design of this and its companion blood-pressure trial went unmentioned - the file keeps it.

What was found

People whose first ~1000 days (conception to age 2) fell under UK sugar rationing had ~35% lower type 2 diabetes and ~20% lower hypertension risk decades later; roughly a third of the protection traces to the in-utero window, the rest to infancy.

Why it matters

The strongest causal-leaning evidence that early-life sugar environment shapes adult metabolic disease - a natural experiment where exposure was set by policy, not by choice. The partition matters: most of the effect is post-natal, which is actionable without moralising pregnancy.

Limits

One historical setting (1950s Britain, low baseline sugar); outcome cohorts from UK Biobank with its selection profile; exposure inferred from birth timing.

Podcast context

Filed from the Inchauspé episode (Diary of a CEO, 2026-02-26), where it was quoted as "15% lower diabetes" framed as an in-womb effect - the paper's headline is ~35% for the full 1000-day exposure, mostly postnatal.

What was found

Ketogenic diet versus control diet:

• No significant difference in glycaemic control.
• No significant difference in body weight.
• HDL higher (SMD 0.19, 95% CI 0.02-0.37, I2 = 0%, moderate-quality evidence).
• Triglycerides lower (SMD -0.41, -0.64 to -0.18, I2 = 0%, low-quality evidence).

Conclusion: may improve lipid profiles, no additional benefit for glycaemic control or weight loss compared with control diets.

Why it matters

This is the sharpest result in the diabetes set because it isolates ketogenic rather than low-carbohydrate generally, and finds no glycaemic or weight advantage over control diets. Read alongside 40930317_..., which found low-to-moderate carbohydrate restriction does lower HbA1c short term, the pattern suggests the benefit does not scale with going all the way to ketosis.

Note also I2 = 0% on both lipid outcomes: the lipid effects are highly consistent across trials, which is unusual in this literature and makes them the most reliable finding here.

Limits

11 trials, studies published 2001-2021, minimum duration only 14 days. Control diets vary. No hard endpoints. As with every diet trial, unblindable, and adherence to a true ketogenic diet is the recurring problem documented in 40930317_mongkolsucharitkul-2025_....

What was found

Two years of creatine with supervised exercise in postmenopausal women: no significant effect on bone mineral density (the primary outcome); bone geometry measures at the proximal femur were better preserved.

Why it matters

The bone story is where creatine marketing runs ahead of data. The honest result: density unchanged over two years; structural geometry preserved. The earlier "reduced hip BMD loss" claim belongs to a smaller 2015 trial and did not replicate here as density.

Limits

Single trial population (postmenopausal women); geometry endpoints are secondary; exercise co-intervention in both arms.

Podcast context

Filed from the Candow episode - his own trial, and his on-air account partly hedged the density/geometry distinction. The file keeps it sharp.

What was found

14 days of experimental sleep restriction (about 4 h in bed) versus control: energy intake rose ~300 kcal/day, and visceral abdominal fat increased 11% (CT-measured), a gain that persisted into recovery sleep despite weight partially normalising.

Why it matters

The cleanest human evidence that short sleep causes visceral fat accumulation rather than merely correlating with it - and the persistence after recovery is the uncomfortable part. This is the strongest single argument in the library for treating sleep duration as a metabolic lever, not a comfort.

Limits

n=12, young and healthy, 14 days, imposed restriction rather than habitual short sleep; one study.

Podcast context

Filed from the Rhonda Patrick episode (Diary of a CEO, 2026-03-30), where it was cited accurately.

Protocol: what they did

Systematic search for evening wear of blue-blocking (amber) glasses, for sleep outcomes and for mood disorders.

What was found

• Mechanism: reduced activation of intrinsically photosensitive retinal ganglion cells (ipRGCs), inducing dim-light melatonin onset.
• Sleep (24 of 29 publications): substantial evidence for reduced sleep onset latency in sleep disorders, jet lag and variable shift work.
• Bipolar disorder: only 1 case study and 1 RCT in acute mania, both showing substantial symptom reduction. Preliminary, described by the authors as high-yield to study further.
• Major and postpartum depression: 3 publications, heterogeneous and conflicting on efficacy.
• Authors conclude blue-blockers are a viable recommendation for insomnia or delayed sleep phase.

Why it matters

The only intervention paper in this folder. Everything else describes exposure; this one tests a countermeasure. It also sets the honest ceiling on the claim: the reliable outcome is falling asleep faster, not mood, not sleep quality, not duration.

Limits

453 patients across 16 RCTs is roughly 28 per trial. Conference abstracts were included alongside peer-reviewed papers. Blinding is close to impossible with amber glasses, so expectancy effects are uncontrolled. Lens transmission spectra vary widely between products and are usually unreported.

What was found

LT4 vs LT4+LT3 vs DTE, all blinded: no differences in primary or secondary outcomes for the whole group; minor heart-rate increase on DTE. The most symptomatic third at baseline preferred T3-containing therapy and improved on several measures while on it.

Why it matters

Reproduces the group-level null with all three regimens head-to-head, and supplies the best current version of the enrichment hypothesis: if anyone benefits, it is the most symptomatic minority. That is exactly the population the 2021 consensus says the decisive trial must enrol - prospectively, not post-hoc.

Limits

The subgroup analysis is post-hoc with roughly 25 patients per tertile; preference again partially unblindable through heart rate and weight.

What was found

On average only about 72% of activity calories add to total daily expenditure - roughly 28% is compensated away by reductions elsewhere - and compensation is larger at higher adiposity.

Why it matters

Total energy expenditure is partly constrained: the body absorbs some of what exercise spends. This explains why exercise alone underperforms for weight loss while remaining fully effective for fitness and mitochondrial adaptation (Molmen 2025) - the training benefits in this topic do not depend on net calorie arithmetic.

Limits

Cross-sectional, model-based decomposition; individual compensation varies widely; says nothing about any individual's response.

Podcast context

Filed from the Galpin episode. The guest implied compensation can fully erase a 500-kcal workout; the paper shows ~28% average compensation, with full offset only at the extreme.

Protocol: what they did

Not a Seyfried paper. An independent German group testing the hypothesis directly. 50 patients with recurrent malignant glioma randomised to reirradiation plus either an unrestricted diet or calorie-restricted ketogenic diet with intermittent fasting (3 days KD at 21-23 kcal/kg/d, 3 days fasting, 3 days KD).

Primary endpoint prespecified: progression-free survival at 6 months (PFS6).

What was found

• PFS6: 20% diet versus 16% control. Not significantly different.
• No difference in PFS, local PFS6, or overall survival.
• The intervention worked as intended: 17 of 20 completers developed ketosis and glucose fell significantly.
• Well tolerated. Modest weight loss (-2.1 kg). No severe adverse events attributable to the diet.
• Exploratory: within the diet group, patients whose day-6 glucose was below the median (83.5 mg/dL) had significantly longer PFS and OS (P < 0.05).

Why it matters

This is the paper that changes the tier of the whole topic. The hypothesis was tested prospectively, by people with no stake in it, in the disease where the claim is loudest, and it did not work. The diet was feasible and safe; it simply did not improve any survival endpoint.

The glucose subgroup signal is the one thread left. It points at glucose level, not ketosis, as the operative variable, which converges with Seyfried's own 2003 finding that restriction rather than diet type drove the mouse effect.

Limits

n=50 and underpowered for small effects. Recurrent, heavily pretreated patients, where any intervention has the least room to work. The diet schedule was short (9 days). A companion analysis (PMID 34487222) found the control group ate less than expected (21 rather than 30 kcal/kg/d), which narrows the contrast between arms and complicates interpretation. A longer schedule in newly diagnosed patients remains untested.

What was found

Pooled blinded preference for combination therapy: 46.2% (95% CI 40.2-52.4) - which the authors judge not distinguishable from chance.

Why it matters

Preference is the strongest recurring pro-combination signal, and this is the estimate that bounds it from below. Read together with de Lima Beltrao 2025 (RR 2.20 for preferring T3-containing therapy), the preference literature currently supports both "chance" and "twofold" depending on inclusion and analysis choices. That disagreement, not either number, is the finding.

Limits

348 patients pooled; preference elicitation differs across trials; period and carryover effects in crossovers.

What was found

An estimated 936 million adults aged 30-69 have obstructive sleep apnea (AHI >= 5), 425 million of them moderate-to-severe. Prevalence exceeds 50% in some countries.

Why it matters

Establishes sleep apnea as a population-scale condition rather than a niche diagnosis, which is what makes the underdiagnosis figure (Young 1997) a public health problem rather than a curiosity.

Limits

Model extrapolated from 17 prevalence studies; funded by a CPAP manufacturer; AHI >= 5 is a permissive threshold.

Podcast context

Filed from the Galpin episode (Diary of a CEO, 2026-08-13), where the guest's "about a billion people" matches the paper.

What was found

On an ad libitum ultra-processed diet matched to the unprocessed arm for presented calories, energy density, macronutrients, sugar, sodium and fibre, participants ate 508 +/- 106 kcal/day more and gained 0.9 kg in two weeks; they lost 0.9 kg on the unprocessed arm.

Why it matters

The only inpatient randomised test of ultra-processed food per se, with matching tight enough to isolate processing itself. It converts a mountain of UPF epidemiology into one causal data point: processing drives passive overconsumption.

Limits

n=20, two weeks, one site (NIH metabolic ward); mechanism (eating rate, energy density of what was actually eaten) not fully resolved.

Podcast context

Filed from the Spector episode (Diary of a CEO, 2026-01-26); his "about 25% overeating" is a fair paraphrase of +508 kcal/day.

What was found

The DIO2 Thr92Ala polymorphism showed no association with TSH, fT4, fT3, quality of life, or cognitive functioning in the general population or in thyroid hormone users.

Why it matters

The direct population-scale test of the Panicker 2009 hypothesis, and it found nothing. Read together, the genotype story is one positive post-hoc subgroup against one large null: exactly what T3-preliminary means.

Limits

Only 364 participants on replacement therapy, so the specific treated-plus-genotype interaction the hypothesis needs remains weakly tested; QoL instruments differ from WATTS.

What was found

Melatonin content ranged from -83% to +478% of the label across 31 products; more than 71% deviated by over 10%. Serotonin contamination was found in 8 of 31.

Why it matters

Whatever one's position on melatonin, dose control barely exists over the counter. The serotonin contamination is the sharper finding and the better reason for caution than any dosing debate.

Limits

Canadian products, 2017; one batch per product; regulation differs by country (in France melatonin above 2 mg is prescription territory).

Podcast context

Filed from the Galpin episode. The guest claimed content "up to 100x higher" than label; the measured maximum is about 5.8x. Direction right, magnitude wrong - the paper is filed, not the soundbite.

What was found

LT4 users with normal TSH had lower T3:T4 ratios than matched euthyroid controls and differed on 12 of 52 objective measures, including higher weight, despite normal TSH.

Why it matters

Shows LT4 monotherapy at a normal TSH is not biochemically identical to native euthyroidism. That is the mechanistic core of the combination argument - and it is genuine. What does not follow, on current trial evidence, is that restoring the ratio restores the outcomes.

Limits

Cross-sectional matching; 12/52 measures invites multiplicity caution; effect sizes small.

What was found

Across 168 studies with objectively recorded sleep, there was no significant change in healthy adults' sleep duration over more than 50 years.

Why it matters

"We sleep an hour less than our grandparents" is one of the most repeated sleep claims in popular media, and the objective record does not show it. Filed as a standing correction: sleep problems are real; a secular collapse in sleep duration is not established.

Limits

Healthy adults only; methods vary across decades; says nothing about sleep quality, timing or restriction in subgroups.

Podcast context

Filed from the Galpin episode as the direct correction to the guest's "60-minute drop over the last 60 years".

What was found

Desiccated thyroid extract versus LT4, blinded crossover:

• No differences in symptoms or neurocognitive measures.
• About 1.4 kg (3 lb) lower weight on DTE.
• Preference: 48.6% DTE, 18.6% LT4, 32.9% none.

Why it matters

DTE is the substance behind most testimonial marketing, and this is its only blinded test of size. The ~1.4 kg weight difference is the largest objective physical difference recorded anywhere in this literature - which is the number to hold against any before/after transformation claim.

Limits

Single centre; 16-week arms; weight and heart-rate changes could partially unblind preference; fixed conversion dosing.

Protocol: what they did

5 h evening exposure to a white LED-backlit screen emitting more than twice the 464 nm light of a white non-LED-backlit screen (0.241 vs 0.099 W/(sr x m2) in the 454-474 nm band). Melatonin, subjective and objective sleepiness (slow eye movements, frontal EEG 1-7 Hz), and cognitive tests.

What was found

• LED screen significantly suppressed the evening melatonin rise.
• Reduced subjective and objective sleepiness.
• Improved sustained attention (GO/NOGO), working memory/attention (explicit timing) and declarative memory in a word-learning task.
• Subjects rated screen quality and visual comfort as identical; the non-LED screen actually tended to be rated brighter.

Why it matters

The honest complication in this folder: the same evening blue light that damages circadian timing acutely improves cognitive performance. Any recommendation has to name that tradeoff instead of pretending evening blue light is simply bad. The comfort rating is the second useful point: the effect is invisible to the user, so subjective judgement is not a usable guide.

Limits

n = 13, all male, single evening. Cognitive gains are acute and say nothing about performance after the disrupted night. No sleep outcome measured.

Protocol: what they did

Narrowband blue LED (peak 469 nm, half-peak bandwidth 26 nm), freely reactive pupils, exposure 02:00-03:30, plasma melatonin before and after, across a range of irradiances. Ocular safety cleared against ACGIH exposure limits first.

What was found

• Increasing irradiance produced increasing melatonin suppression (P < 0.0001).
• Sigmoidal fluence-response curve, R2 = 0.99, ED50 = 14.19 microW/cm2.
• Compared with 40 microW/cm2 of 4000 K broadband white fluorescent light (the standard office fixture), narrowband blue LED appears stronger per unit irradiance at suppressing melatonin.

Why it matters

Moves the action spectrum from monochromatic laboratory light to actual LED hardware, and gives a numeric threshold (ED50) rather than a qualitative claim. The white-fluorescent comparison is the practical hook: spectrum beats raw lux.

Limits

n = 8. Middle of the biological night, which is the most sensitive window and not when most people are reading a phone. No sleep or next-day outcomes.

What was found

Five nights of 4 h in bed versus 9 h: participants ate ~300 kcal/day more under short sleep with no compensating rise in energy expenditure; women showed the clearer intake effect.

Why it matters

One of the primary trials inside the Al Khatib synthesis, filed for its design detail: inpatient, controlled feeding, measured (not reported) intake. The mechanism of sleep-loss weight gain is behavioural - appetite - not metabolic slowdown.

Limits

n=30, five nights, inpatient conditions.

Podcast context

Filed from the St-Onge episode; her account matched the paper.

Protocol: what they did

Shotgun lipidomics of cardiolipin content and composition in highly purified brain mitochondria from two inbred mouse strains and from five tumours derived from them (astrocytoma, ependymoblastoma, stem cell tumour, two microgliomas), with paired electron transport chain activity measurements.

What was found

• Major cardiolipin abnormalities in all five tumours, in content or composition.
• Compositional pattern: excess immature molecular species, deficiency of mature ones, indicating defects in cardiolipin synthesis and remodelling.
• Abnormalities were associated with significant reductions in individual and linked electron transport chain activities.

Why it matters

This is the paper that gives the theory a physical substrate. Warburg proposed irreversible injury to respiration but never identified the lesion. Cardiolipin is a specific lipid, confined to the inner mitochondrial membrane, required for respiratory chain function, and it is measurably abnormal here.

Measured biochemistry, not argument. It belongs in a different category from the theory papers that cite it.

Limits

Five mouse tumours from two inbred strains, one laboratory. Correlation between cardiolipin abnormality and reduced ETC activity does not establish that either one initiates tumourigenesis, which is the claim the theory papers make from it. Purified mitochondria from tumour tissue also raise the isolation-artefact question that Picard's own 2011 work flagged for this exact preparation.

What was found

Maternal glucose showed continuous associations with birth weight, cord C-peptide and adverse outcomes across the entire range below diagnostic diabetes - no threshold where risk begins.

Why it matters

The paper behind modern gestational-diabetes criteria: risk is a gradient, not a category. It legitimises attention to glycemia in pregnancy while cutting against scare framings - the gradient is smooth and modest per increment, not a cliff behind a "spike".

Limits

Observational; outcomes are perinatal, not long-term offspring health; treatment implications came from later RCTs, not this cohort.

Podcast context

Filed from the Inchauspé episode as the solid core beneath its pregnancy glycemia theme.

Protocol: what they did

Four arms, which is what makes this paper worth reading closely:

1. standard diet unrestricted (SD-UR)
2. ketogenic diet unrestricted (KD-UR)
3. standard diet restricted to 40% (SD-R)
4. ketogenic diet restricted to 40% (KD-R)

Tumour weight, plasma glucose, beta-hydroxybutyrate and IGF-1 measured at day 13.

What was found

• Tumour growth was rapid in both unrestricted arms, including the ketogenic one.
• Growth was reduced by about 80% in both restricted arms, regardless of diet type.
• Plasma glucose predicted tumour growth. Restriction of either diet lowered IGF-1.
• The authors' own conclusion: growth "is dependent more on the amount than on the origin of dietary calories".

Why it matters

This is the experiment the whole metabolic-therapy programme is built on, and it contains the control arm that later work rarely repeats. An unrestricted ketogenic diet did nothing. The therapeutic effect came from calorie restriction, present in the standard-diet arm too.

Anyone citing this paper as evidence that ketosis starves tumours is citing the opposite of its result. What it supports is glucose availability, and by extension caloric intake, as the operative variable.

Limits

Mouse, one implanted tumour line, one lab, 13 days. 40% caloric restriction is severe and is not a therapy that can be sustained in a cancer patient who is already at risk of cachexia. No survival endpoint.

What was found

82% of men and 93% of women with moderate-to-severe sleep apnea syndrome had no clinical diagnosis in the Wisconsin Sleep Cohort.

Why it matters

The canonical underdiagnosis figure. Combined with Benjafield 2019, it implies most of a very large affected population does not know. Symptoms that overlap with ordinary tiredness are exactly the kind a health record never captures unless someone looks.

Limits

1990s data from one US cohort; diagnostic access has changed; syndrome definition required symptoms plus AHI.

Podcast context

Filed from the Galpin episode; the guest's "70-80%, women 90-95%" is a faithful rendering.

Protocol: what they did

Measured arteriovenous differences for acetoacetate, beta-hydroxybutyrate, glucose, lactic acid, glutamine and other amino acids across tumours growing in living rats fasted for two days. Blood entering the tumour versus blood leaving it, which is the direct way to ask what a tumour consumes.

What was found

• Arterial ketones in the fasted tumour-bearing rats: acetoacetate 0.52 mM, beta-hydroxybutyrate 1.82 mM.
• Both ketone bodies were utilized by the tumours, and the rates of utilization were directly related to the rates of supply.
• Mean utilization: acetoacetate 13.9 nmol/min/g, beta-hydroxybutyrate 24.7 nmol/min/g of tumour wet weight.

Why it matters

This is the direct counter-evidence to the premise underlying ketogenic metabolic therapy. Compare the stated rationale in 25798181_meidenbauer-2015_glucose-ketone-index.md:

"Brain tumour cells, unlike normal brain cells, cannot use ketone bodies effectively for energy when glucose becomes limiting."

Sauer and Dauchy measured tumours consuming ketones in living animals in 1983, at rates proportional to how much was supplied. The proportionality is the damaging part: raising blood ketones raised tumour ketone consumption.

Filed here so the topic contains the measurement that the framework has to answer, not only the framework.

Limits

Rat, and not a brain tumour. Three hepatomas and a sarcocarcinoma. The selectivity claim is made specifically for brain tumours, where the comparison is against neurons that are unusually good at using ketones, so this does not refute the brain case directly. It does show that tumour ketone utilization is real and supply-dependent, which is the general premise.

1983, and the tissue-isolated tumour preparation is not the same as an orthotopic tumour with native vasculature.

What was found

• Adherence high: over 75% maintained nutritional ketosis.
• Median overall survival 29.4 months in adherent cohorts versus 14.6 months in historical controls, with a 66.7% three-year survival rate.
• Well tolerated. Adverse events limited to mild gastrointestinal symptoms and fatigue. No grade 3/4 diet-related toxicities.
• Conclusion: supports KD as a safe, feasible, biologically rational adjunct with potential to prolong survival; calls for standardised phase III trials.

Why it matters

A doubling of median survival in glioblastoma would be the largest advance in the disease in decades. The comparator is the problem: historical controls, and the cohorts are defined by adherence.

Patients who sustain strict ketosis for months are not a random sample. They are younger, better performance status, better resourced, more likely to have favourable tumour biology. Comparing them to historical averages measures who can do the diet as much as what the diet does. This is the classic bias that randomisation exists to remove, and when ERGO2 removed it the difference was 20% versus 16%.

Filed deliberately next to 42451166_persiani-2026_..., published the same year on the same literature, which reports that randomised trials showed no significant OS or PFS difference. Two credible 2026 syntheses, opposite conclusions, and the entire difference is the comparator. That contrast is more instructive than either conclusion alone.

Limits

Historical-control comparison, adherence-defined cohorts, heterogeneous interventions pooled (4:1 classic, modified Atkins and calorie restriction are not the same exposure). The safety and feasibility findings are sound; the survival figure should not be quoted without its comparator.

What was found

Hypothyroid treatment containing liothyronine was associated with reduced risk of dementia and mortality compared with LT4 alone.

Why it matters

If causal, this would overturn the null consensus - which is precisely why its design matters: patients channelled to T3-containing therapy differ systematically (younger, healthier, more engaged) from those left on LT4. The finding earns a trial, not a prescription.

Limits

Retrospective; prescription channeling and healthy-user bias; no dose or adherence detail; outcome ascertainment from records.

What was found

Aging-associated decline of phosphatidylcholine (PC) synthesis disrupts the mitochondrial network and contributes to mitochondrial dysfunction in normal aging - identified in C. elegans (including long-lived clk-1 and isp-1 mitochondrial mutants) and echoed in human transcriptomic and metabolomic data. Boosting PC through diet restored late-life mitochondrial integrity in nematodes and reinstated metabolic resilience in human cell culture. Leibniz Institute on Aging (Jena); no competing interests declared.

Why it matters

A genuinely new candidate driver of "natural" mitochondrial aging - membrane lipid supply rather than mutation or damage - and one that is dietarily malleable in the model. It connects three library threads: the mitochondria topic's aging framework, the choline literature, and the supplement question. If the mechanism holds in humans, PC/choline status becomes a concrete, testable target.

Limits

• The intervention evidence is worms and cell culture. The human data are correlational omics, not a trial.
• No dosing, no human endpoint, no safety-in-context data.
• One study, published April 2026; no independent replication yet.
• The step from "restores worm mitochondrial network" to "slows human biological decline" is exactly the step this field has repeatedly failed to make (compare urolithin A, filed in this topic with its missed primary endpoint).

Media context

Filed from a circulating popular-science summary (submitted 24 August 2026) claiming mitochondria "could let us slow or even reverse biological decline". The paper itself claims a malleable trigger of mitochondrial aging in model systems; its authors explicitly caution against supplement conclusions. The summary also misattributes first authorship (Ermolaeva is senior author).

What was found

Higher prenatal BPA exposure was associated with autism spectrum disorder specifically in boys with low aromatase-pathway activity (a genetically defined subgroup), with supporting mechanistic evidence of brain aromatase disruption.

Why it matters

The best current version of the BPA-neurodevelopment claim, and a lesson in precision: the association lives in a defined subgroup of boys, not in children generally. Podcast renditions ("BPA causes autism", "6x risk") drop the conditioning that makes the finding credible.

Limits

One cohort; subgroup analysis; exposure measured by spot urine; not replicated; association plus mechanism, not causation in humans.

Podcast context

Filed from the Rhonda Patrick episode (the receipts/BPA segment). The 6x odds she cited apply only to the low-aromatase male subgroup; her separate "autistic kids are 30x less likely to excrete BPA" claim is contradicted by the underlying excretion paper.

Protocol: what they did

Four states in the same subjects: overnight-fasted baseline, 2-hour postprandial fed, 36-hour fasted, and 2-hour postprandial re-fed. Metabolic and immune markers, macrophage function, and plasma metabolomics. Plasma-borne factors then tested for lifespan effects in C. elegans.

What was found

36-hour fasting modulated macrophage function and raised multiple bioactive plasma metabolites, and those metabolites extended lifespan in C. elegans.

Why it matters

Filed for completeness, but note the two mismatches: it is about fasting, not a ketogenic diet, and it is about longevity and immunity, not cancer. It belongs to the caloric-restriction thread that runs through this library (Seyfried 2003, ERGO2's restriction component) rather than to the ketone thread.

Limits

n=20, healthy young adults, single 36-hour fast, mechanistic endpoints only. The lifespan extension is in worms, from human plasma factors, which is a long way from a human outcome. Registered as a trial but reported as a pilot.

What was found

Urolithin A (500 or 1,000 mg/day) improved muscle strength ~12% and some mitochondrial biomarkers; the primary endpoint (6-minute walk / peak VO2 improvement vs placebo) was not met.

Why it matters

Urolithin A is the flagship "mitophagy" supplement, and this is its best human trial: secondary wins, primary miss. The mechanistic story is appealing; the clinical proof is not yet there.

Limits

n=88, 4 months, sponsor-run (Amazentis); strength gains without performance gains invite multiple readings.

Podcast context

Filed from the Rhonda Patrick episode. Her mouse-lifespan and VO2max figures for urolithin A could not be located in any publication; this trial, with its missed primary endpoint, is what the human evidence actually is.

Protocol: what they did

Same trial, same registration IRCT20171105037259N2, same 80 patients, 12 weeks. This paper reports insulin, IGF-1, CEA, CA15-3, ESR, CRP, IL-10, TNF-alpha, plus imaging (sonography for locally advanced, CT/MRI for metastatic) and restaging after surgery.

What was found

• TNF-alpha decreased and IL-10 increased in the diet group (both P < 0.001).
• Lower adjusted serum insulin (P < 0.002).
• Tumour size reduction 27 vs 6 mm (P = 0.01) and significant downstaging in locally advanced disease.
• No significant difference in response rate in patients with metastases.

Why it matters

The single most-cited positive human finding in this literature. Tumour size reduction and downstaging are real clinical endpoints, not biomarkers.

Limits

Read it as what it is: a secondary-outcome paper from a trial whose registered primary outcomes were null, published separately a year later. That pattern warrants caution on its own, independent of anyone's intentions.

Everything was adjunctive to chemotherapy, so the design cannot attribute tumour shrinkage to the diet rather than to chemotherapy plus a diet-associated difference. n=80, single centre, unblindable, imaging not stated as blinded. And the arm where metabolic therapy is most often promoted, metastatic disease, showed no benefit.

Protocol: what they did

Developed hair pigmentation pattern (HPP) profiling along single hair shafts, turning one hair into a physical timeline of pigmentation. Combined with proteomics on the same individual hairs, plus a computational simulation.

What was found

• White/grey hairs naturally regain pigmentation, across sex, ethnicities, ages and body regions. Greying is therefore quantifiably reversible, not a one-way process.
• Grey hairs upregulate proteins related to energy metabolism, mitochondria and antioxidant defences.
• Greying and reversal can occur in parallel with psychological stressors.
• The simulation suggests a threshold mechanism for temporary reversibility.

Why it matters

The most striking and most over-quoted result in this bibliography. What is solidly established is that human hair greying is reversible and carries a molecular signature pointing at mitochondria. It is a visible, datable biological record of recent life history, which is genuinely novel as a method.

Limits

The stress association rests on a small number of hairs from a small number of people whose stress was self-reported and time-matched retrospectively. That is a case-series-level claim carried by a method paper. It is not evidence that stress causes greying or that reducing stress reverses it, and it is routinely reported as if it were. Treat the method as the contribution and the stress link as a hypothesis.

What was found

A high-fermented-food diet (up to ~6 servings/day) steadily increased gut microbiota diversity and decreased 19 of 93 inflammatory proteins; the high-fibre arm did neither over 17 weeks. The primary cytokine response score did not change in either arm.

Why it matters

The origin of the fermented-foods wave. The secondary findings are genuinely interesting and biologically coherent; they are also exactly that - secondary, in 36 people, with a null primary endpoint.

Limits

n=36, single centre, unblindable, multiple comparisons across 93 proteins; 17 weeks.

Podcast context

Filed from the Spector episode, where every quoted number was wrong (28 people for 36, 5 servings for 6, a month for 17 weeks, a "25% inflammation drop" that appears nowhere) and the null primary outcome went unmentioned.

Protocol: what they did

80 patients randomised to ketogenic diet or control for 12 weeks during chemotherapy. Quality of life, physical activity, and biochemical safety markers (thyroid, electrolytes, albumin, ammonia, ALP, lactate, ketones).

Registration: IRCT20171105037259N2.

What was found

• No significant difference in quality of life or physical activity at 12 weeks.
• Higher global quality of life and physical activity in the diet group at 6 weeks only (P = 0.02, P = 0.01), not sustained.
• Serum lactate and ALP decreased significantly in the diet group.
• No differences in thyroid hormones, electrolytes, albumin, LDH or ammonia, i.e. no safety signal.
• Compliance 66.7% to 79.2%.

Why it matters

This is one of only two randomised publications in the entire Seyfried bibliography, and its primary outcomes were null. It does establish that a ketogenic diet is tolerable alongside chemotherapy without metabolic harm, which is a real if modest finding.

Limits

n=80, single centre, unblindable, adjunctive to chemotherapy so it cannot test the diet as a treatment. Compliance under 80%. The transient 6-week difference that disappeared by 12 weeks is the kind of result that inverts on replication.

Read together with PMID 32703721: it is the same trial, same 80 patients, same registration number, reporting a different outcome set. Counting them as two trials doubles the apparent evidence base.

Protocol: what they did

Exposed healthy midlife adults to a brief psychological challenge on two separate occasions and measured serum circulating cell-free mitochondrial DNA (ccf-mtDNA) and cell-free nuclear DNA. Then tested in cells whether glucocorticoid signalling alone can cause mtDNA extrusion.

What was found

• 2 to 3-fold increase in ccf-mtDNA after psychological stress.
• No change in cell-free nuclear DNA, which establishes specificity rather than generic cell damage.
• In cells, glucocorticoid signalling alone was sufficient to induce mtDNA extrusion, on a timescale consistent with the human response.

Why it matters

Shows that a purely psychological stressor produces the same class of signal as physical injury, and identifies a candidate mechanism (cortisol) linking the two. ccf-mtDNA becomes a blood-measurable readout of psychological state. Later work in this bibliography extends it to saliva and to diurnal dynamics.

Limits

Small sample, healthy midlife adults, acute laboratory stressor. The authors state explicitly that further controlled work is needed to confirm the stress attribution and that the functional significance of circulating mtDNA is unknown. It is a marker whose meaning is still open.

Protocol: what they did

Built a Mitochondrial Health Index (MHI) for human leukocytes, integrating nuclear- and mtDNA-encoded respiratory chain enzyme activities with mtDNA copy number, so that mitochondrial content per cell and mitochondrial quality can be told apart. Applied it to high-stress caregivers versus controls, alongside daily mood diaries.

What was found

• The MHI outperformed individual mitochondrial measures.
• Elevated positive mood at night was associated with higher MHI, and nightly positive mood mediated the caregiving-to-MHI association.
• MHI correlated with positive mood on the days preceding the blood draw but not the days following it, the first human evidence that mitochondria may respond to proximate emotional states within days.
• Caregivers had higher perceived stress, lower positive and greater negative affect, and lower MHI. Not explained by nuclear/mitochondrial genome mismatch.

Why it matters

The paper that answers the 2018 systematic review's complaint. It replaces mtDNA copy number (a proxy that does not measure function) with a functional composite, and it establishes a plausible timescale: days, not years. The mood-precedes-MHI directionality is the most interesting detail and the one most in need of replication.

Limits

n = 91, all mothers, cross-sectional group comparison. The temporal ordering is suggestive of direction but is not an experiment. Leukocytes are accessible, not necessarily representative of tissue mitochondria elsewhere.

What was found

Among self-selected online respondents, treatment satisfaction ranked DTE (~7/10) > LT4+LT3 (~6) > LT4 (~5), with prominent overall dissatisfaction.

Why it matters

This is the closest published analogue to testimonial and social-media transformation claims, and its design shows why they mislead: unblinded, self-selected respondents recruited through patient communities. Valuable as a measure of how large and vocal the dissatisfied population is; unusable as a comparison of therapies.

Limits

Self-selection, no blinding, no denominators, satisfaction confounded by the choice and effort of obtaining non-standard therapy.

What was found

Vinegar with a meal attenuated postprandial glucose (SMD -0.60) and insulin (SMD -1.30) across small clinical trials.

Why it matters

The vinegar "hack" has a real acute pharmacology (acetic acid slows gastric emptying and glucose absorption). What it does not have: any evidence that blunting healthy-range post-meal spikes improves any health outcome, which is the unstated premise the hack rides on.

Limits

Small heterogeneous trials, many in metabolically impaired populations; acute surrogate endpoints; dental and GI tolerability unexamined at habit scale.

Podcast context

Filed from the Inchauspé episode; accurately described in the acute sense, sold well beyond it.

What was found

Greater fiber intake predicted more slow-wave (deep) sleep; more saturated fat predicted less; more sugar predicted more arousals - with diet effects visible on polysomnography after a single day.

Why it matters

The bidirectional claim needs both arrows: sleep loss worsens eating (Al Khatib, T1), and eating quality feeds back on sleep architecture - this is the best mechanistic datapoint for the second arrow, and it is small. It also gives the Mediterranean-pattern recommendation a sleep rationale on top of its metabolic one.

Limits

n=26, one-day dietary manipulation, sleep-architecture surrogates.

Podcast context

Filed from the St-Onge episode, where it anchored the fiber-for-deep-sleep advice.

Protocol: what they did

Combined three interventions in the VM-M3 metastatic mouse model: ketogenic diet, ketone supplementation, and hyperbaric oxygen therapy. Followed by cell work at high glucose, low glucose, beta-hydroxybutyrate, hyperbaric oxygen, or all combined.

What was found

• Combination therapy reduced tumour growth rate and metastatic spread, and mice lived twice as long as controls.
• Each intervention individually decreased VM-M3 cell proliferation and viability in culture.
• Hyperbaric oxygen increased ROS production in the cells.

Why it matters

The source of the "doubled survival" figure that circulates in the popular discussion of this work. It is a genuine result, in mice.

Limits

Mouse, one model, three interventions at once. A combination design cannot tell you which component did the work, and cell culture at "low glucose" is not a dietary state. Hyperbaric oxygen is a clinic-based intervention with its own risks, not a lifestyle change. Doubling survival in a VM-M3 mouse says nothing quantitative about a human with metastatic disease, and the paper does not claim otherwise: it calls for further investigation.

What was found

Eating vegetables and protein before carbohydrate in the same meal lowered post-meal glucose by ~29-37% at 30-60 min and insulin AUC in 11 metformin-treated type 2 diabetics.

Why it matters

This is the experimental foundation of the food-order "hack" - a real, sensible physiological effect (delayed gastric emptying) measured in eleven diabetic patients over one meal. Filed so the claim's global popularity can be weighed against its evidence base. In healthy people the effect exists in small studies but is smaller, and no outcome trial exists anywhere.

Limits

n=11, T2DM on metformin, single test meal, acute glucose endpoints only.

Podcast context

Filed from the Inchauspé episode; food order is the centrepiece of the guest's method, generalised to healthy people and pregnancy well beyond this design.

What was found

In a 20-week Mediterranean weight-loss program, late main-meal eaters (after 15:00) lost significantly less weight than early eaters at equal reported intake, activity and sleep.

Why it matters

The foundational human datapoint for "when you eat matters" - filed at its true strength: an association within one program, not a randomised timing trial. It connects the circadian topic to the diet topic without yet proving the lever.

Limits

Observational; self-reported intake; Spanish meal culture (lunch as main meal); no randomisation of timing.

Podcast context

Filed from the St-Onge episode, where meal timing was discussed as promising rather than proven - consistent with this design.

What was found

Creatine added to an SSRI (escitalopram) in women with major depressive disorder accelerated and enhanced response versus SSRI plus placebo over 8 weeks.

Why it matters

The most cited datapoint behind "creatine for mood". A real randomised signal, in one trial, in one sex, as an adjunct - the brain-energetics rationale is coherent, and the replication that would move this to T2 has not arrived in over a decade.

Limits

n=52, women only, single site, 8 weeks, adjunct to an antidepressant - not a standalone effect.

Podcast context

Filed from the Candow episode, where it was presented with roughly appropriate caution.

What was found

DIO2 rs225014 (Thr92Ala) CC homozygotes (~16%) had worse baseline well-being (GHQ) and greater improvement on combination therapy (2.3 GHQ points at 3 months, P=0.03). Circulating thyroid hormone levels did not differ by genotype.

Why it matters

The paper behind every "get your DIO2 gene tested" pitch. The effect, if real, is a plausible mechanism for a responsive subgroup - but it is a post-hoc finding at P=0.03 with no biochemical footprint, and the large population study (Wouters 2017) found no genotype association at all.

Limits

Post-hoc; multiple genes tested; no prospective genotype-stratified trial has confirmed it in the seventeen years since.

Protocol: what they did

Compared fluorescent light powered by conventional (magnetic) ballasts against high-frequency electronic ballasts, measuring subjective well-being, performance and physiological arousal (EEG) in a laboratory office.

What was found

• Few general effects across the whole sample.
• In individuals with high critical flicker fusion frequency (CFF), conventional ballasts produced pronounced attenuation of EEG alpha waves, increased speed and decreased accuracy of performance.
• Interpreted as heightened CNS arousal in response to the light modulation.
• Recommendation: power fluorescent lighting with good-quality electronic high-frequency ballasts.

Why it matters

The only paper here about a non-visible property of light that is not spectral. Flicker is a separate axis from melanopic content, it acts on arousal rather than on melatonin, and it affects a susceptible subgroup rather than everybody. Modern relevance is through PWM dimming on LED lamps, phone and laptop backlights, which reintroduced deep modulation that magnetic-ballast research had made obsolete.

Limits

Subgroup finding, not a main effect, in a sample of 37. Predates LED and PWM entirely, so the transfer to current hardware is an inference, not a result. CFF was measured but no dose-response for modulation depth or frequency was established.

Protocol: what they did

Takes the phrase "exercise as mitochondrial medicine" seriously and asks the question that follows from it: if exercise is a medicine, what is the correct prescription (type, dose, frequency, duration)? Evaluates exercise prescription variables against a systematic framework distinguishing domains of mitochondrial biology: molecular features, activities, integrated functions, behaviours, and cell-dependent phenotypes.

Why it matters

Filed for contrast with the other T4 papers in this library. The framework papers in mitochondria-psychobiology.md and cancer-metabolic-theory.md argue about what mitochondria are; this one asks what dose of a known-effective intervention to give. That is a question that generates trials rather than positions.

It also closes the gap flagged in ../topics/mitochondria-psychobiology.md, which noted that no exercise paper had been filed despite this one existing in Picard's bibliography. Note Picard is the third author here.

Limits

No new data. A narrative review, and it does not deliver a settled prescription: the honest conclusion is that prescription variables matter and the evidence is not yet sufficient to specify an optimum. For the quantitative dose relationship, use 39390310_molmen-2025_..., which reports that training load (volume x intensity) predicts the change.

Protocol: what they did

Proposes the energy resistance principle (ERP): a general framework describing how energy behaves and transforms in what the authors call the carbon-based circuitry of biology.

What was found

• Energy resistance (eR) is presented as the fundamental property enabling transformation, converting the unformed energy potential of food-derived electrons flowing toward oxygen into useful work.
• eR is required to sustain life, but excess eR directly causes reductive and oxidative stress, heat, inflammation, molecular damage and information loss, described as the hallmarks of disease and aging.
• Disease-causing stressors raise eR and circulating GDF15.
• Sleep, physical activity and restorative interventions minimise eR.
• Offered as a testable general framework across development, aging, and the health-disease continuum.

Why it matters

This is the unifying frame behind the public-facing version of the work, and the source of the "you cannot add energy, only reduce resistance" formulation. It also names the practical levers explicitly, which no other paper in this set does.

Limits

Read this as a research programme, not a result. A single quantity proposed to span diabetes, cancer metabolism, Alzheimer's and aging is doing a great deal of explanatory work, and eR is not yet operationalised as something measurable in a person. GDF15 is the nearest available proxy. The authors call it testable; at the time of filing it has not been tested. Its practical recommendations (sleep, physical activity) are already well supported by evidence that owes nothing to this framework, so the framework should not be treated as their justification.

What was found

The AHA statement defines multidimensional sleep health - duration, timing, regularity, satisfaction, efficiency, daytime function - and links each dimension to cardiometabolic risk, setting the research and clinical agenda.

Why it matters

Filed as the sleep topic's framework anchor: duration is only one of six dimensions, and the library's filed evidence so far touches mainly that one. A T4 by the library's rules - it changes questions, not conclusions.

Limits

A consensus statement; no new data; never citable as evidence for any outcome claim.

Podcast context

Filed from the St-Onge episode; the guest chaired the statement.

Protocol: what they did

Four-page argument that the prevailing view of mitochondrial disease as ATP deficiency is wrong, and that OxPhos defects instead trigger maladaptive stress responses which consume excess energy.

What was found

The claim: chronic hypermetabolism imposes energetic constraints, and those constraints, rather than an energy shortfall, cause mitochondrial disease pathophysiology. Calls for translational studies from organelle to organism.

Why it matters

The clearest short statement of the inversion at the centre of this body of work: the problem is not too little energy produced, it is too much energy spent. Useful as an entry point before the heavier Sturm 2023 paper, and notable for the co-author list, which brings in energetics researchers from outside mitochondrial medicine (Pontzer, Kempes).

Limits

A commentary of about three pages. It makes no measurements and its evidentiary weight is entirely borrowed from 36635485_sturm-2023_oxphos-hypermetabolism-lifespan.md.

Protocol: what they did

Addresses a genuine contradiction: molecular damage accumulation increases cellular energy expenditure with age, yet whole-body energy expenditure stays flat or falls. The model resolves it by casting the brain as broker of the organismal energy economy.

What was found

The brain-body energy conservation (BEC) cascade. Aging or senescent cells activate costly intracellular stress responses and secrete cytokines signalling increased demand (hypermetabolism). The brain perceives a shrinking energy budget and deploys energy conservation responses, suppressing low-priority processes.

The proposed consequence is that many features read as age-related decline are instead active conservation: fatigue, physical inactivity, blunted sensory capacity, immune alterations, endocrine "deficits".

Why it matters

This is the "fixed energy budget with a hierarchy of needs" idea in its peer-reviewed form, and the reframing is the substance: symptoms as budgeting decisions rather than as failures. It also states where interventions could act, which is what makes the model testable in principle.

Limits

A model, not a result. The central inference, that the brain reads peripheral hypermetabolism and responds by conserving, is assembled from separate literatures rather than demonstrated end to end in one organism. Cytokine signalling of "energy demand" to the brain is best evidenced for GDF15; generalisation beyond that is provisional.

Protocol: what they did

Reviews the enzymology of ketone use by tumour cells and proposes drug targets.

What was found

The keto-paradox of the title:

• In tumour cells, ketolysis via SCOT (succinyl-CoA:3-oxoacid-CoA transferase) and ACAT1 (acetyl-CoA acetyltransferase 1) is a major source of mitochondrial acetyl-CoA.
• Active ACAT1 tetramers are stabilised by tyrosine phosphorylation, which facilitates the SCOT reaction and ketolysis. The same phosphorylation stabilises inactive PKM2 dimers and, with acetylation of PDH, closes the glycolytic supply of acetyl-CoA.
• Tumour cells also turn off fatty acid degradation because they must synthesise membranes (malonyl-CoA brake).
• Therefore: supplying ketones supplies acetyl-CoA by the one route left open. The proposed therapy is to inhibit SCOT and ACAT1, plus acetyl-CoA synthetase to block acetate uptake.

Why it matters

The prescription is the opposite of ketogenic metabolic therapy. Where Seyfried's programme raises ketones to starve the tumour, this paper argues ketones are fuel and the enzymes that use them should be blocked.

Filed as the mechanistic counterpart to 6861121_sauer-1983_...: one measured tumours consuming ketones, this one names the enzymes that let them.

Limits

No new data. A mechanistic argument in a review, and MDPI J Clin Med is a rapid-turnaround venue. Two of three authors are listed as "Independent Researcher" with residential addresses and no institutional affiliation; the second author, Eric Berg, is a chiropractor and health-media figure rather than a cancer researcher. That does not make the SCOT/ACAT1 biochemistry wrong, and it is independently citable, but the paper should not be weighted as an institutional review.

The same author also circulates non-peer-reviewed promotional material on cancer prevention, so this review and that material are not independent sources.

Protocol: what they did

Argues explicitly that "the analogy of mitochondria as powerhouses has expired" and proposes the mitochondrial information processing system (MIPS): mitochondria as the processor of the cell rather than its battery.

What was found

A three-step model. Mitochondria:

1. sense and respond to endogenous and environmental inputs through morphological and functional remodelling;
2. integrate information via dynamic network-based physical interactions and diffusion;
3. produce output signals that tune other organelles and regulate physiology systemically.

This input-to-output transformation is what lets mitochondria transduce metabolic, biochemical and neuroendocrine signals into organismal adaptation.

Why it matters

The flagship theory paper, and the one to read first if reading only one. Everything in the psychobiology work depends on mitochondria being a signalling system rather than an energy supply, because a battery cannot transduce a psychosocial state and a processor can.

Limits

A review in a metabolism journal, not a test of anything. "Information processing" is doing significant work as a term and is not operationalised into a measurable quantity here. The strength of the framework is that it generates the specific measurable claims found elsewhere in this folder.

Protocol: what they did

Argues that principles of behaviour in social organisms apply to mitochondria: they communicate with each other and with the nucleus, form groups, are interdependent, synchronise their behaviour, and functionally specialise.

What was found

Not a results paper. The claim is that mitochondria are social organelles, and that importing social concepts into cell biology is a useful theoretical shift because it foregrounds communication and interdependence over isolated function.

Why it matters

This is the framing behind the popular-facing version of the work, including the 2025 Scientific American piece and the podcast appearances. Filed here because knowing it is a metaphor being deliberately deployed, not a finding, is necessary to read the rest of the bibliography correctly.

Limits

An analogy. Analogies generate hypotheses and also smuggle in assumptions. "Mitochondria are social" is not falsifiable as stated; the specific claims underneath it (nanotunnels, network organisation, functional specialisation) are testable and are tested elsewhere in this bibliography.

Protocol: what they did

Direct comparison of the mitochondrial metabolic theory (MMT) against the somatic mutation theory (SMT), arguing MMT accounts better for the hallmarks of cancer.

What was found

The argued position:

• Proliferating cancer cells need carbon and nitrogen for metabolites and ATP. Glucose carbons come via glycolysis and the pentose phosphate pathway; glutamine supplies nitrogen and carbon via glutaminolysis.
• Glutamine-dependent mitochondrial substrate level phosphorylation becomes the essential ATP source in cells that over-express PKM2, have deficient OxPhos, and can grow in hypoxia (0.1% oxygen) or in cyanide.
• Therapeutic implication: target glucose and glutamine simultaneously while raising non-fermentable ketone bodies.

Why it matters

The clearest and most current statement of the theory, and the most useful entry point for judging it. The glutamine argument is also the part that has moved most: targeting glutamine is a live area of mainstream cancer metabolism research independent of this theory.

Limits

No new data. The paper's own framing is explanatory power, not experiment, and a theory that explains existing facts better is a hypothesis-generating position, not a demonstration. The causal-order claim, that mitochondrial damage precedes and causes the mutations, is the crux and is not settled by argument.

Note the practical recommendation (simultaneous glucose and glutamine restriction plus ketones) has never been tested in a randomised human trial in this or any other paper in this topic.

What was found

The joint American, British and European thyroid associations do not endorse combination therapy as superior. They define the trial that would settle the question: enrol patients dissatisfied despite adequate LT4, use twice-daily or slow-release LT3, patient-reported primary outcomes, genotype substudies. 28 of 34 statements reached at least 75% agreement.

Why it matters

The referee's position. Anyone claiming the science is settled - in either direction - is contradicted by the three societies' own reading: group-level evidence negative, the decisive enriched trial not yet run.

Limits

A consensus process, not data. Filed T4 and never citable as evidence for any outcome claim.

Protocol: what they did

Argues that no major survival advance has been made in glioblastoma in nearly a century, that current standard of care (resection, radiotherapy, temozolomide, dexamethasone) itself raises glucose and glutamine availability in the tumour microenvironment, and that ketogenic metabolic therapy should therefore be considered as standard of care.

The stated evidence base, in the abstract's own words: "preclinical and case report studies".

Why it matters

Worth filing precisely because of the asymmetry. The question in the title asks for a change to standard of care in a lethal disease. The evidence offered is mouse work and uncontrolled case reports.

The critique of the standard of care is not baseless: glioblastoma outcomes are genuinely poor and dexamethasone genuinely raises blood glucose. A valid criticism of the incumbent is not evidence for the alternative.

Limits

No new data. Published in 2019; ERGO2, the randomised test, reported in 2020 and was negative on every survival endpoint. Any current reading of this paper has to be made against that result, which is filed as 32619561_voss-2020_ergo2-ketogenic-glioma-rct.md.

Protocol: what they did

Narrative review proposing that mitochondria sense, integrate and transduce psychosocial and behavioural factors into cellular and molecular changes, and that this is how psychological states get biologically embedded.

What was found

• Chronic psychological stress induces metabolic and neuroendocrine mediators that structurally and functionally recalibrate mitochondria. Picard and McEwen name this mitochondrial allostatic load.
• Primary mitochondrial defects affect brain, endocrine and immune systems, the same systems involved in psychosomatic processes, suggesting a shared mechanistic basis.
• Mitochondria feed back systemically by releasing mitokines and metabolites, influence gene expression and epigenetic marks, and modulate the rate of cellular aging.

Why it matters

This is the origin document for everything Picard has done since. It converts "stress is bad for you" from a correlation into a proposed mechanism with a named organelle and measurable outputs. The companion systematic review (PMID 29389736) is the honest audit of how little evidence existed for it in 2018, and the two should always be read together.

Limits

A theory paper, and it says so. Every mechanism proposed here needed testing; much of the rest of this author's bibliography is that testing. The concept of allostatic load itself is inherited from McEwen and is a framework rather than a measurable quantity.

Protocol: what they did

Borrows a term from palaeontology (press-pulse extinction events) to propose a therapeutic strategy: a chronic "press" of calorie-restricted ketogenic diet to lower glucose and raise ketones, combined with acute "pulses" of drugs, hyperbaric oxygen and other stressors timed to when tumour cells are most vulnerable.

Why it matters

This is the protocol behind most clinical claims made for metabolic therapy, and the source of the specific regimens people encounter. Filed so that the protocol and its evidence base can be looked at in the same place.

Limits

No new data, and no trial of the protocol as specified. The strategy is assembled from separate preclinical findings, several of them from the authors' own mouse work, and combined by argument. A multi-component protocol is also the hardest kind to evaluate: if it were trialled and worked, the design would not say which component mattered.

The nearest actual randomised test of the diet component in the target disease, ERGO2, was negative.

Protocol: what they did

Proposes the Glucose Ketone Index (GKI), the ratio of blood glucose to beta-hydroxybutyrate as a single number, on the argument that daily glucose fluctuates too much for either value alone to track therapy. Applied the calculator retrospectively to already-published glucose and ketone data in humans and mice.

What was found

The authors report "a clear relationship between the GKI and therapeutic efficacy" in the data they reanalysed.

Why it matters

The GKI is widely used in the metabolic therapy community and is the number people are told to track. Knowing where it came from is the point of filing it.

Limits

No new data. A ratio applied backwards to studies selected by the authors, with no prospective validation, no threshold established against a clinical outcome in a controlled setting, and no independent evaluation. That a metric correlates with efficacy in retrospectively chosen datasets is the weakest form of support a biomarker can have.

Note also that ERGO2 (PMID 32619561) found glucose alone carried the prognostic signal in its exploratory analysis, with ketosis achieved in nearly all patients and no survival benefit. That is evidence for the glucose half of the ratio, not for the ratio.

Protocol: what they did

States the central claim: cancer originates from impaired cellular energy metabolism, specifically damage to mitochondrial respiration, and the genomic mutations that define the somatic mutation theory are downstream consequences rather than causes. Revives and extends Warburg.

Why it matters

This is the origin document, and the reason the position is worth taking seriously rather than dismissing: it makes a falsifiable claim about causal order (mitochondrial damage first, mutations second) and points to the nuclear/cytoplasm transfer experiments as the test.

Limits

No new data. The causal-order claim is genuinely contested and the mainstream position remains that mitochondrial dysfunction in tumours is real but is one feature among many rather than the initiating event. Note the venue: Nutrition & Metabolism, a nutrition journal, not a cancer biology journal, which is where a claim of this scope would need to survive review.