Brain & Cognitive Aging: What the Primary Evidence Supports

Exercise is the best-evidenced lever for the aging brain. Across 39 RCTs, exercise improved cognition in adults 50+ with an overall effect of g = 0.29 [1], and an earlier meta-analysis found executive function, the domain most vulnerable to aging, showed the largest gains, with combined aerobic-plus-strength training outperforming aerobic alone [2]. The strongest structural evidence is mechanistic as well as behavioral: one year of moderate aerobic walking increased hippocampal volume ~2%, reversing an estimated 1-2 years of atrophy, and the gain tracked both improved spatial memory and rising serum BDNF [3]. When exercise is embedded in a multidomain program (diet, cognitive training, vascular-risk monitoring), the FINGER trial showed a significant slowing of cognitive decline in at-risk older adults (NTB composite difference 0.022/year) [4].

Blood pressure control protects cognition, within limits. In hypertensive adults, intensive systolic control (<120 vs <140 mmHg) significantly reduced mild cognitive impairment (HR 0.81) and the combined MCI-or-dementia endpoint (HR 0.85) [5]. The probable-dementia endpoint alone did not reach significance (HR 0.83), because the trial stopped early for cardiovascular benefit and was underpowered for that slower outcome [5]. The underlying blood-pressure biology is settled upstream: each 10 mmHg lower systolic BP cuts stroke ~27% [6], and the <120 target itself cut all-cause mortality ~27% in high-risk adults, at the cost of more hypotension, syncope, and acute kidney injury [7].

Sleep clears the brain, and the clearance machinery is real. The glymphatic pathway, periarterial CSF influx, interstitial exchange, perivenous waste efflux, is AQP4-dependent, established when Aqp4 deletion cut CSF, ISF exchange ~65% and slowed amyloid-β clearance ~55% [8], and re-confirmed across knockout lines after a replication dispute [9]. Sleep expands the interstitial space and accelerates amyloid clearance [10], and in humans NREM slow-wave activity drives pulsatile CSF flow, physiological corroboration of sleep-dependent clearance [11]. Causally, a single night of selective slow-wave disruption raised next-morning CSF amyloid-β in responders [12]. The direction is well-supported; the human magnitudes are not, Fultz shows CSF flow coupled to sleep, not a measured clearance outcome, and much of the quantitative glymphatic picture remains rodent.

APOE4 is the dominant common-variant risk gene, and its effect is dose-dependent. Each ε4 allele adds Alzheimer's risk and lowers age of onset [13]; against ε3/ε3, ε3/ε4 carries OR ~3.2 and ε4/ε4 OR ~14.9 [14]. ε4/ε4 homozygosity behaves as a near-fully-penetrant, distinct genetic form of AD, abnormal biomarkers from ~age 55, most amyloid-positive by 65, though the cross-sectional, research-cohort penetrance may be overstated [15]. The allele sets a near-linear LDL-C gradient (ε2 lowest → ε4 highest) [16], while ε2 confers better survival independent of AD and CV factors [17] and lower coronary risk (OR 0.87) [18].

The omega-3 / APOE4 brain story is mechanistically rich but clinically null so far. Young ε4 carriers paradoxically show greater regional DHA uptake on PET (gray-matter K* +16%), an exploratory n=22 finding [19]. The ε4 brain hyperactivates cPLA2 (via p38 MAPK), extracting arachidonic acid and driving LTB4-mediated neuroinflammation, confirmed in human postmortem AD brain, mouse, and cell culture [20]. But the interventions fail: 2 g/day DHA over 18 months did not slow decline in Alzheimer's disease, and the ε4 subgroup showed no benefit on any measure [21]. The pre-symptomatic PREVENT-E4 trial (ages 55, 80) showed DHA does reach the brain, CSF DHA:AA rose significantly, independent of genotype, yet cognitive and structural endpoints were null at 24 months [22]. Delivery without benefit shifts the question from "can it reach the brain?" to "does it help after it arrives?", and at 24 months, the answer is no.

Omega-3 gives a small memory signal, not a general cognitive boost. In non-demented adults, EPA+DHA produces a mild memory benefit (g ≈ 0.31) with no effect on global cognition (g ≈ 0.02), concentrated in older/MCI participants [23]. Good evidence, small effects, no miracles.

Creatine helps the stressed brain, not the healthy one. A high single dose improved cognition during ~21 h of sleep deprivation while raising cerebral high-energy phosphates [24], consistent with earlier work showing a benefit on a central-executive task only under extreme sleep deprivation [25]. But a 2024 systematic review found the brain-energetics mechanism does not translate to a consistent general cognitive benefit [26]. The defensible statement is narrow: creatine may aid cognition under acute stress; a healthy-population memory benefit is not established.

Two metabolic routes to Alzheimer's cognition, one modest, one null. MCT/ketones reliably raise brain ketone metabolism and give a modest, inconsistent, often APOE4-negative cognitive effect, not "reversal" [27]. NAD-precursor supplements do not improve human cognition: the highest-quality trial to date doubled NAD+ in amnestic MCI (target engagement confirmed) yet was null on cognition, cerebral blood flow, and blood pressure [28]. The preclinical promise has not translated.

Vascular disease reaches the brain. In an ApoE-knockout mouse model with carotid stenosis, foam-cell-derived exosomes worsened ischemic white-matter injury via a miR-101-3p → Nrf2 → Slc2a1 axis transferred to microglia [29], directionally linking atherosclerosis to white-matter pathology, though human extrapolation remains early.

Where the Evidence Is Contested

Observational data and randomized trial data disagree about whether the MIND diet protects cognition. Morris 2015 ([30]) found that high adherence was associated with roughly 7.5 years of slower cognitive aging, with a dose-response pattern. The randomized trial by Barnes 2023 in NEJM ([31]), a 3-year study, was null: the difference in global cognition was +0.035 (95% CI −0.022, 0.092, p=0.23). Both findings are well-supported. The structural explanation for the divergence is healthy-user confounding in the observational data.

The evidence on creatine and memory turns on a methodological dispute. The meta-analysis by Prokopidis 2023 ([32]) reported a positive effect, with an SMD of 0.29 (p=0.02). Eckert and Pascher 2023, in a Letter ([33]), raised a double-counting critique, and Verhoeven ([34]) replicated that critique. A narrower effect, confined to sleep deprivation and stress, survives this challenge ([25]/[24]).

Whether NAD+ helps in Alzheimer's remains contested across the gap between animal and human work. Hou 2018 ([35]) reported a reversal in mice, but this is preclinical. Human RCTs of NAD precursors have been null ([36]; Martens [28]). This fits the classic pattern of preclinical-to-human translation failure; the MCTs show only modest effects ([27]).

Two further claims are contested against their own primary data. On APOE4 and DHA transport, the prior model from Yassine held that transporter failure prevented delivery. The PREVENT-E4 primary analysis ([22]) found instead that ε4 carriers do achieve CSF DHA delivery, a null result for the transport-failure hypothesis at this age window. Separately, Walker's framing in Why We Sleep presents sleep's effects as sensitive to specific decades of life. The primary data in Winer 2019 ([37]) do not match this: the published decade windows, with Aβ in the 50s and 70s and tau in the 60s only, differ from Walker's claim.

References

1. Northey JM, Cherbuin N et al. 2018 (BJSM) — 39-RCT meta-analysis; 333 effect sizes; n=4,884 (50+ y); g=0.29 (95% CI 0.17–0.41, p<0.01); aerobic,… · PubMed 28438770
2. Colcombe S, Kramer AF. 2003 (Psychol Sci) — 18 RCTs, 1966–2001; overall ~0.5 SD across cognitive tasks; executive-control processes showed largest… · PubMed 12661673
3. Erickson KI, Voss MW, Kramer AF et al. 2011 (PNAS) — n=120 older adults; 1-yr aerobic walking vs stretching control; hippocampal volume +2%… · PubMed 21282661
4. Ngandu T, Solomon A, Kivipelto M et al. 2015 (Lancet) — FINGER RCT; n=1,260; 2yr; Finland; multidomain (diet + exercise + cognitive training +… · PubMed 25771249
5. Williamson JD et al. (SPRINT MIND Investigators). 2019 (JAMA) — Intensive BP control (target SBP <120) vs standard (<140); n=9,361; median 3.26yr;… · PubMed 30688979
6. Ettehad 2016, Lancet — BP lowering per 10 mmHg SBP (613,815 participants) · PubMed 26724178
7. SPRINT 2015, NEJM — intensive BP <120 cuts events + mortality (HR 0.75/0.73) · PubMed 26551272
8. Iliff 2012, Sci Transl Med — defined the glymphatic pathway (AQP4-dependent, clears Aβ) · PubMed 22896675
9. Mestre 2018, eLife — AQP4-dependence of glymphatic transport reaffirmed · PubMed 30561329
10. Xie et al. 2013, Science — sleep drives glymphatic clearance · PubMed 24136970
11. Fultz 2019, Science — human NREM slow-waves drive CSF pulsation · PubMed 31672896
12. Ju et al. 2017 Brain — SWS disruption → ↑CSF Aβ40 in responders (r=0.610, p=0.009; NOT full group); sub-acute home sleep quality → ↑CSF tau… · PubMed 28899014
13. Corder 1993, Science — APOE4 gene dose & AD risk (dose-dependent) · PubMed 8346443
14. Farrer LA, Cupples LA, Haines JL et al. 1997 (JAMA) — 40-consortium meta-analysis; n=5,930 AD + 8,607 controls; ε2/ε3 OR 0.6 (0.5–0.8) for AD vs… · PubMed 9343467
15. Fortea 2024, Nat Med — APOE4/4 as a distinct genetic form of AD · PubMed 38710950
16. Bennet 2007, JAMA — APOE genotype, LDL & CHD meta-analysis · PubMed 17878422
17. Shinohara M et al. 2020 (eLife) — NACC (n=24,661; 5,413 deaths) + apoE-TR mouse survival cohort (n=118); APOE2 survival advantage persists after… · PubMed 33074098
18. Kuo CL et al. 2020 (Aging Albany NY) — UK Biobank PheWAS; n=379,703 European-descent, 9.4y follow-up; ε2/ε3: LDL −0.43 SD, CAD OR 0.87 (0.83–0.90),… · PubMed 32511104
19. Yassine HN et al. 2017 Alz Res Ther — DHA brain uptake PET ([1-¹¹C]-DHA); n=22; young APOE4 carriers: global GM K* +16% (p=0.04), posterior… · PubMed 28335828
20. Wang S, Yassine HN et al. 2022 Mol Neurodegen — cPLA2 hyperactivation in APOE4: p38 MAPK → cPLA2 → arachidonic acid extraction → LTB4 →… · PubMed 35705959
21. Quinn JF et al. — DHA 2g/day × 18mo in mild-to-moderate AD (JAMA 2010; n=295; overall null; APOE4 subgroup null; non-E4 exploratory trend only, post-hoc) · PubMed 21045096
22. Yassine HN et al. 2026 (eBioMedicine) — PREVENT-E4 RCT peer-reviewed; n=368; 2g DHA/day × 24mo; primary: CSF DHA delivery … · PubMed 42315445
23. Omega-3 & cognition meta-analysis · PubMed 31841161
24. Gordji-Nejad 2024, Sci Rep — creatine & sleep-dep cognition · PubMed 38418540
25. McMorris et al. 2006 — creatine, sleep deprivation & cognition · PubMed 17046034
26. 2024 systematic review — creatine cognition (skeptical) · PubMed 38582412
27. MCT/ketones & AD cognition — meta-analysis · PubMed 31870908
28. Martens CR et al. 2026 (Alz & Dementia) — NR in aMCI pilot (n=42, 12 wk, DB-RCT); NAD+ doubled (target engagement confirmed); cognitive, CBF, BP… · PubMed 42478598
29. Zhang H et al. 2025 Cell Metab — foam cell exosomes → miR-101-3p → Nrf2 → Slc2a1 → microglial metabolic dysfunction → white matter injury (ApoE−/−… · PubMed 40345179
30. Morris MC, Tangney CC, Wang Y et al. 2015 (Alzheimer's & Dementia) — MIND diet score (0–15); n=923, age 58–98y, mean 4.7yr follow-up; top tertile… · PubMed 26086182
31. Barnes LL, Dhana K, Liu X et al. 2023 (NEJM) — MIND diet RCT; n=604 (301 MIND, 303 healthy control); adults ≥65, BMI ≥25, family history of AD,… · PubMed 37466280
32. Prokopidis et al. 2023, Nutrition Reviews — creatine & memory meta-analysis (disputed)
33. Eckert & Pascher 2023 — Letter: double-counting in Prokopidis meta-analysis · PubMed 36644917
34. Physionic — "Creatine Expert uncovers why 5 grams isn't Enough!" (video 06)
35. Hou/Fang 2018, PNAS — NAD+ normalizes AD (mouse)
36. Nicotinamide riboside in MCI — RCT (null cognition)
37. Winer et al. 2019 J Neurosci — SO-spindle→MTL tau (r=−0.47); SWA→Aβ (r=−0.36); lifespan: Aβ sensitive 50s+70s, tau sensitive 60s only; n=101;… · PubMed 31209175