Regenerative & personalized
Cognitive performance optimisation
Adenosine blockade, the brain's energy buffer and a genotype-dependent dopamine optimum — the mechanisms that survive scrutiny, and the null replications that are findings in their own right.
The cleanest way to judge this category is a single distinction: does the intervention restore a function that has dropped for some reason, or does it raise a rested, healthy person above their own baseline? Almost everything with a reproducible mechanism belongs to the first kind.
Adenosine: a real mechanism, but the effect is a return to baseline
Caffeine is a competitive antagonist at adenosine A1 and A2A receptors. Adenosine accumulates in the brain across waking and is the molecular carrier of homeostatic sleep pressure; by blocking its binding, caffeine removes an inhibition rather than adding a resource. That produces the field’s central methodological problem: in habitual consumers, much of the measured “improvement” is reversal of overnight withdrawal. Studies that recruit caffeine-naive participants report substantially smaller effects on vigilance and sustained attention than the early literature, and close to nothing on complex executive tasks.
Creatine follows similar logic. The brain spends ATP, and phosphocreatine is the fast buffer for resynthesising it. Supplementation does raise brain creatine, but reproducible cognitive gains appear mainly under metabolic stress — sleep deprivation, hypoxia, or vegetarians with low baseline stores. In a rested omnivorous adult the effect is close to nothing.
Where genotype genuinely enters
The most substantive version of “personalisation” here rests on the COMT Val158Met polymorphism. Catechol-O-methyltransferase degrades dopamine, and its contribution is unusually large in prefrontal cortex, where reuptake is weak. The Val allele gives a more active enzyme and lower tonic dopamine; the Met allele the opposite. Because working-memory performance depends on prefrontal dopamine as an inverted U, a dopaminergic intervention can improve one genotype and impair the other. The mechanism is well grounded and has been demonstrated — but effect sizes are small, replications uneven, and the leap from that finding to commercial nutrigenomic panels is not supported by prospective controlled trials. The polymorphism accounts for a fraction of a percent of population variance in cognitive measures.
The null results that should be named
Computerised working-memory training reliably improves the trained task and close variants of it. Meta-analyses of transfer to unrelated tasks, to intelligence and to academic attainment — that is, to everything the training is sold for — consistently return effects indistinguishable from zero. Ginkgo biloba was tested in the large randomised GEM trial in more than 3,000 older adults and reduced neither dementia incidence nor the rate of cognitive decline. Transcranial direct-current stimulation produces results sensitive to skull anatomy and to task state, and its meta-analyses disagree with one another.
The practical conclusion is narrow rather than nihilistic: sleep, aerobic exercise and treating the underlying illness move cognitive measures reproducibly. Everything else in this category should be read as a small effect conditional on a pre-existing deficit, and a seller should be asked to name the compound, the dose and the specific task on which the effect was measured.