1. The one intervention that grows brain tissue in a randomised trial
Almost nothing in this topic can claim a structural change to the brain in a controlled trial. Exercise can.
Erickson and colleagues, PNAS, 2011. 120 older adults randomised to aerobic exercise or a stretching control. The exercise group's anterior hippocampal volume increased by about 2% over a year, against the expected age-related decline in the control group — effectively reversing one to two years of volume loss, with measurable improvement in spatial memory. The gain tracked serum BDNF, a protein involved in the survival and growth of neurons.
The hippocampus is the structure that shrinks with age and is hit early in Alzheimer's disease, which is why this trial is quoted so heavily. Graded Moderate: one good randomised trial with a structural endpoint and a small cognitive one, not a demonstration that anyone avoided dementia.
2. The correction, because the attribution is wrong nearly everywhere
That trial is now routinely cited as evidence for high-intensity interval training. It did not test HIIT. The intervention was moderate-intensity walking, working up to about 40 minutes three times a week at moderate effort.
This matters practically rather than pedantically. The population that most needs this result is older, often deconditioned, and frequently has cardiovascular disease. Telling them the evidence requires interval training misdescribes the trial and puts a harder, riskier and less sustainable version of it in front of exactly the people least able to take it up. The evidence is for walking.
None of which is an argument against interval training. It has good evidence for cardiorespiratory fitness, which is itself associated with better cognitive outcomes, and single sessions of vigorous exercise acutely raise lactate, glutamate and BDNF. If you enjoy it and it is safe for you, do it. Just do not believe you are failing to protect your brain by walking instead.
3. What the dose looks like
- Aerobic activity is the better-evidenced half for the hippocampus specifically. Roughly 150 minutes a week of moderate activity is the standard target, and the largest gains in risk are between doing nothing and doing something.
- Resistance training earns its place for different reasons — muscle, bone, insulin sensitivity and independence later. See the exercise chapter and the bone chapter.
- The vascular route is probably doing much of the work. Exercise improves blood pressure, glucose handling and endothelial function, and the dementia chapter shows how much of dementia risk is vascular. That makes exercise the clearest case on this site of one intervention paying into two accounts.
4. Cold water, and what the study actually measured
Brief cold exposure is widely promoted as a brain intervention on a hormesis argument: a short sharp stressor forces the nervous system to adapt.
The study underneath the claim did not measure cognition. The foundational work is a small physiological study of human responses to immersion at different water temperatures. It found what it looked for — large acute rises in metabolic rate and in circulating catecholamines including noradrenaline. There is no cognitive outcome in it, no long-term follow-up, and no brain imaging.
A rise in noradrenaline is a real physiological event that feels like alertness. Calling it reversal of brain aging is a claim nobody has tested.
And the caution that belongs on a cardiovascular site. Sudden cold immersion triggers a cold shock response — involuntary gasp, hyperventilation, a sharp rise in heart rate and blood pressure. The people this caution is for, specifically: anyone with known coronary disease, a known arrhythmia, or uncontrolled high blood pressure. For them it is a genuine risk rather than a theoretical one. Drowning after an involuntary gasp is also the mechanism behind a meaningful share of cold-water deaths in the first minutes.
Two rules, and they are not optional: enter gradually, and never alone.
Graded Emerging for mood and alertness, and ungraded for cognition, because the question has not been asked.