Every major observational study finds the same thing: men with low testosterone develop dementia at higher rates. The Shanghai Aging Study. The Framingham cohort. Swedish population registries. NHANES cross-sectional analyses. The signal is consistent, dose-dependent, and survived a decade of replication.
Then the trials happened. TRT raised testosterone. Cognition didn't improve. The T-Trials — the largest hormonal intervention study in older men at the time — found zero benefit on its primary cognitive endpoint. Subsequent meta-analyses confirmed: testosterone therapy does not reliably improve cognitive function.
This is the cognitive paradox. The hormone predicts the disease but replacing it doesn't treat it. The same pattern that played out with bone density and fractures — the bone paradox — except here the stakes are Alzheimer's disease, and the resolution runs much deeper.
Five layers explain why the trials failed. The deepest one suggests that medicine has been testing the wrong hypothesis entirely — that it isn't low testosterone causing neurodegeneration, but something the body produces in response to low testosterone.
The Observational Case
The epidemiological evidence linking low testosterone to cognitive decline is not subtle. It spans decades, continents, and methodologies.
Androgen deprivation therapy — the natural experiment — reinforces this. A 2025 meta-analysis of 21 studies covering 2.28 million men found that ADT increased dementia risk by 14% (HR 1.14) and Alzheimer's specifically by 15% (HR 1.15). Anti-androgens carried the worst signal (HR 1.53 for AD). Notably, LHRH agonists and orchiectomy showed no significant link — mechanism specificity that matters for interpretation (Zhu et al., Journal of Alzheimer's Disease, 2025).
The Liu 2025 NHANES analysis added another dimension: low testosterone combined with insufficient sleep produced synergistic cognitive decline (OR 2.10–2.14) — connecting this thread directly to the sleep architecture story.
The Trials That Failed
The T-Trials enrolled 788 men aged 65 and older with testosterone below 275 ng/dL. After 12 months of testosterone gel, the primary cognitive endpoint — delayed paragraph recall, a hippocampal memory task — showed no benefit. A subsequent analysis using more sensitive tests found a modest improvement in spatial memory but nothing that moved the clinical needle.
Wang et al.'s 2025 meta-analysis (Biomedical Reports, 14 studies) found that androgen replacement therapy improved executive function and memory in hypogonadal men — but with modest effect sizes and significant publication bias. Not nothing, but not the signal the observational data would predict.
TRAVERSE, the 5,246-man cardiovascular safety trial, had no dedicated cognitive substudy. Its depression substudy measured cognition secondarily via the HIS-Q but not through primary cognitive endpoints. The largest testosterone trial in history generated almost no cognitive data.
The gap between what the epidemiology promises and what the interventions deliver is real. But it isn't a contradiction — it's a misunderstanding. Five layers of resolution explain why.
Layer 1: The Gonadotropin Hypothesis
What if it isn't low testosterone that causes neurodegeneration — but the elevated FSH and LH that the body produces in response to it?
This is the deepest resolution layer, and it reframes the entire paradox. When testosterone falls, the pituitary compensates by raising follicle-stimulating hormone (FSH) and luteinizing hormone (LH). In observational studies, low T and high gonadotropins travel together. Every study that finds "low T predicts dementia" is simultaneously finding "high FSH/LH predicts dementia" — but almost no one was looking at the gonadotropins.
Mone Zaidi's lab at Mount Sinai changed that.
The Evidence Chain
FSH directly drives amyloid-β deposition and tau phosphorylation in hippocampal neurons via C/EBPβ-δ-secretase pathway. FSH-blocking antibody reversed Alzheimer's phenotype in mice.
FSH + ApoE4 synergy demonstrated. The Alzheimer's risk allele amplifies FSH's neurotoxic effects — connecting the two biggest risk factors through a shared mechanism.
FSHR gene-dose-dependent rescue. Fshr−/− mice showed full prevention of spatial memory loss in 3xTg AD models. Fshr+/− partial.
Elevated FSH → AD dementia OR 3.597 in males specifically. Separate study: elevated LH → faster temporal lobe atrophy. Two-hit hypothesis proposed: BBB disruption + pathological accumulation.
Fshr depletion prevents recognition memory loss AND fat loss AND bone loss in AD mice. Commentary: "Targeting FSHR is highly promising." FSH emerges as multi-organ pathological signal.
MS-Hu6: humanized monoclonal FSH-blocking antibody. IND-enabling data. Improved cognition in AD mice. Excellent primate safety. First-in-human trials planned.
The implication is stark. When observational studies found "low T predicts dementia," they were likely detecting the downstream effect of elevated gonadotropins — FSH and LH acting directly on brain tissue to drive amyloid accumulation and tau pathology. Testosterone was a bystander variable, not the causal agent.
This reframes every failed TRT trial. TRT does suppress gonadotropins — it drives FSH and LH down through negative feedback. If the gonadotropin hypothesis is correct, TRT might actually be neuroprotective, but through a mechanism no one was measuring. The trials tracked cognition as an outcome of testosterone levels. They should have been tracking cognition as an outcome of FSH suppression.
The Treatment Modality Problem
This has a disturbing clinical implication that no one is studying.
| Treatment | Testosterone | FSH/LH | Predicted Brain Effect |
|---|---|---|---|
| TRT (gel, injection) | ↑↑ | ↓↓ Suppressed | Potentially protective |
| SERMs (clomiphene, enclomiphene) | ↑ | ↑↑ Elevated | Potential concern |
| hCG | ↑ | ↑ (LH-mimetic, FSH variable) | Uncertain |
| GLP-1 RAs (via weight loss) | ↑ (indirect) | ↓ (normalized) | Likely protective |
SERMs — the fertility-preserving alternative I've written about extensively — work by blocking estrogen feedback at the pituitary, causing FSH and LH to rise. If FSH drives neurodegeneration, then SERMs could be raising testosterone and raising brain risk simultaneously. No study has examined this differential. Not one.
The Zaidi lab's 2025 review explicitly calls for clinical trials of humanized monoclonal FSH antibodies. Their commentary in Molecular Psychiatry states: "Targeting FSHR is highly promising." The antibody MS-Hu6 has published IND-enabling data — dose-dependent weight reduction in mice, improved cognition in AD models, excellent safety in African green monkeys (~4% weight loss after a single injection). A first-in-human study is planned.
If Diaphorai's coverage of the amyloid hypothesis entrenchment showed how $42 billion in anti-amyloid drugs failed, the gonadotropin hypothesis offers a reason: the amyloid hypothesis wasn't wrong about amyloid's role in neurodegeneration — it was wrong about the upstream driver. FSH-driven amyloid deposition through C/EBPβ-δ-secretase is a completely different entry point from the one every anti-amyloid drug targeted.
Layer 2: The Critical Window
Estrogen research learned this lesson the hard way. The Women's Health Initiative (WHI) found that hormone replacement increased dementia risk — but the women were in their 60s and 70s. When KEEPS and ELITE tested estrogen in women within 6 years of menopause, the results flipped. The "healthy cell bias" hypothesis explains why: hormones protect healthy neurons but can't rescue damaged ones.
The T-Trials enrolled men aged 65 and older. By that age, cumulative neuronal loss from decades of low gonadotropin-driven damage (Layer 1), sleep disruption, metabolic syndrome, and vascular disease has likely passed the point where testosterone — through any mechanism — can help.
No trial has tested testosterone in younger hypogonadal men (40s–50s) and measured cognition as a primary outcome. The critical window for men remains entirely unexamined.
Colleluori et al. (American Journal of Clinical Nutrition, 2022) tested TRT + lifestyle intervention vs. lifestyle alone in 83 obese hypogonadal men aged ≥65 — a defined clinical population rather than a broad low-T cohort.
Results: Memory improved most (d=1.43 — a large effect), global cognition d=0.74, attention d=0.49. Executive function and language showed no between-group difference.
Crucially, changes in VO2 peak, strength, total testosterone, and LH were all independent predictors of cognitive improvement. The LH finding connects directly to the gonadotropin hypothesis — lower LH at study end predicted better cognition. LITROS suggests that when you pick the right population and combine with lifestyle, the cognitive benefit appears. The T-Trials may have failed because of population selection, not because testosterone doesn't help the brain.
Layer 3: The Domain Mismatch
The T-Trials measured delayed paragraph recall — a hippocampal memory task. This was always an odd choice. Testosterone's most established neural effects operate through androgen receptors in the prefrontal cortex, where it modulates dopamine-dependent executive function. The hippocampus relies more heavily on estradiol (converted from testosterone by aromatase) for its cholinergic memory circuits.
When Wang et al.'s 2025 meta-analysis broke outcomes by domain, androgen replacement improved both executive function and memory — but in hypogonadal men, not in the broadly defined "low-T elderly" population the T-Trials enrolled. The LITROS trial showed the same: memory improved most in obese hypogonadal men receiving TRT + lifestyle, but executive function didn't separate from placebo.
A 2025 Cerebral Cortex study confirmed sex-specific brain activation differences during working memory under testosterone influence, reinforcing that the hormone's cognitive effects are real but domain-specific and population-dependent.
The trials weren't necessarily measuring the wrong thing. They were measuring the right thing in the wrong people — or measuring one domain when the benefit might have been strongest in another.
Layer 4: The Aromatization Decline
Testosterone doesn't act directly in much of the brain. It's converted to estradiol by the enzyme aromatase (CYP19A1), and it's this locally-produced estradiol that mediates many of testosterone's neuroprotective effects — particularly in the hippocampus.
Brain aromatase declines with aging. In the hippocampus — precisely where the T-Trials measured outcomes — older men convert less testosterone to estradiol locally. Give a 70-year-old man more testosterone, and the hippocampus may not be able to use it. The substrate is there, but the enzyme isn't.
This interacts with the critical window (Layer 2): younger men with more hippocampal aromatase would be expected to benefit more from testosterone supplementation. The Alia-Klein 2020 PET study added a complication — in men, lower amygdala aromatase correlated with better verbal recall, suggesting the relationship between brain aromatase and cognition is region-specific and non-linear.
The clinical warning here is concrete: aromatase inhibitors, used by some men on TRT to control estradiol, could be suppressing the very brain E2 that mediates cognitive protection. This is the same problem identified in the bone paradox — estradiol is the hidden variable, and interfering with it has consequences across multiple organ systems.
Layer 5: The Inverted U
The relationship between testosterone and cognition isn't linear. Liu's 2025 NHANES analysis identified a threshold around 300 ng/dL below which cognitive impairment risk escalates. Tang et al. (BMC Geriatrics, 2024) confirmed this with restricted cubic spline analysis — a clear inflection point, not a straight line.
But the curve bends the other way too. A 2025 Psychiatry Research study found that among patients already diagnosed with dementia, higher testosterone was associated with worse ADAS-cog scores (poorer cognition) but better NPI scores (fewer behavioral symptoms). The optimal testosterone level may shift lower with age, and TRT may overshoot it.
This creates a narrow therapeutic window that the T-Trials were not designed to target. If below 300 is harmful and significantly above some upper threshold is also suboptimal, then successful treatment requires precision that "apply testosterone gel for 12 months" cannot deliver.
The DHT Blind Spot
While the five layers explain why TRT trials failed, a separate line of evidence reveals active harm from blocking a testosterone metabolite the field has undervalued.
Dihydrotestosterone (DHT) — produced from testosterone by 5-alpha-reductase (5αR) — is the brain's most potent androgen. It's also the target of finasteride and dutasteride, prescribed to millions of men for hair loss and benign prostatic hyperplasia.
The mechanism is a double hit. Brain 5αR1 produces allopregnanolone — a baseline neurosteroid essential for GABAergic tone and neuroprotection. Brain 5αR2 enables stress-responsive allopregnanolone synthesis. Finasteride blocks both pathways. You lose baseline neuroprotection and the ability to mount a neurosteroid response to stress.
DHT itself is neuroprotective: stronger than estradiol for spatial learning in androgen-deficient animal models. The drugs that block it are prescribed to men in the exact age range where cognitive vulnerability peaks. This is covered in the comprehensive drug-induced hypogonadism map — 5αR inhibitors disrupt neurosteroid synthesis rather than the HPG axis directly, making their cognitive effects invisible to standard hormone panels.
The APOE4 Complication
The largest genetic risk factor for Alzheimer's — the APOE4 allele — interacts with sex hormones, but not in the way you'd expect. A 2025 systematic review (Ageing Research Reviews, Vol. 111) found that APOE4 reduces androgen receptor signaling sensitivity and mediates sex differences via IL-17/TGF-β pathways.
But a 2024 study in Biology of Sex Differences complicated this: the testosterone × APOE4 interaction was significant in women only. In males, there was no interaction regardless of APOE4 status. APOE4 may reduce testosterone's brain effectiveness through AR desensitization rather than through a testosterone-level-dependent pathway — meaning that raising testosterone in an APOE4 carrier might not help even if testosterone is part of the causal chain.
The FSH + ApoE4 synergy from the Zaidi lab (Nature Communications, 2023) offers a different path: FSH's neurotoxic effects are amplified by APOE4, so blocking FSH could be disproportionately beneficial in APOE4 carriers. This is a testable prediction that no trial has examined.
Combined Biomarkers: Where the Signal Gets Strongest
The Shanghai Aging Study (Alzheimer's & Dementia, 2024) tested what happens when you combine hormonal and neurodegeneration markers. The numbers are stark.
| Biomarker Combination | HR for Cognitive Decline |
|---|---|
| Low total T + high NfL | 5.10 |
| Low free T + high NfL | 6.14 |
| Low T alone | ~1.5–2.0 |
| High NfL alone | ~2.0–3.0 |
Neither marker alone produces the full risk. Combined, they're synergistic — greater than additive. NfL (neurofilament light chain) reflects ongoing neuronal damage. Low testosterone (or high FSH) reflects the hormonal environment enabling it. Together they capture both the mechanism and the consequences.
This combined biomarker approach is probably the most clinically actionable finding in the entire field. A man with low testosterone and elevated NfL isn't just hormonal — he's actively losing neurons. The urgency of intervention is categorically different from a man with low T and normal NfL.
What's Coming
Two ongoing efforts could reshape this landscape.
The TotAL Trial (Martins group, Australia) is testing testosterone + DHA vs. placebo in amyloid-positive men aged 60–80. Enrollment ended in 2025; results are pending. If it uses amyloid PET as an endpoint, it could be the first study to directly visualize whether testosterone (and by extension, gonadotropin suppression) affects brain amyloid deposition in humans. No such imaging data currently exists — a confirmed gap.
The MS-Hu6 antibody from the Zaidi lab — now with published IND-enabling data in JCI (July 2025) — could be the first therapeutic to test the gonadotropin hypothesis directly in humans. A single injection produced ~4% weight loss in primates with excellent safety. If it enters clinical trials for Alzheimer's or cognitive decline, it would test whether blocking FSH — without touching testosterone — prevents neurodegeneration.
What This Means
The cognitive paradox isn't a paradox at all. It's the product of five compounding errors in how the field designed its trials and interpreted its data:
1. Wrong hypothesis tested. The trials assumed testosterone-as-neuroprotectant. The evidence increasingly points to FSH/LH-as-neurotoxin. TRT might work — but through gonadotropin suppression, not testosterone elevation.
2. Wrong population. Men over 65 with decades of cumulative damage. The critical window for intervention likely closes before these trials begin.
3. Wrong outcomes. Hippocampal memory as primary endpoint when testosterone's strongest neural effects may be prefrontal and executive.
4. Missing enzyme. Aging hippocampus loses aromatase → less local E2 → less benefit from exogenous T.
5. Wrong dose target. The inverted-U curve means more isn't better. Precision dosing was never attempted.
The American College of Physicians recommends against testosterone for cognitive improvement. Based on the existing trial data, this is defensible. Based on the emerging gonadotropin data, it may be asking the right question about the wrong molecule.
For clinicians: the combined NfL + testosterone biomarker approach from the Shanghai Aging Study offers the strongest predictive power currently available. For patients on SERMs or hCG: no one has studied whether the gonadotropin elevation these treatments cause has brain consequences. For patients considering 5αR inhibitors: the cognitive signal is real, consistent, and mechanistically plausible.
And for the field: the most important trial that hasn't happened yet isn't another TRT-vs-placebo cognitive study. It's a comparison of treatments that suppress gonadotropins against treatments that raise them, with cognition as a primary outcome and FSH as the measured mechanism. Until that trial exists, the paradox isn't resolved — it's only explained.