Progesterone and cognitive function: neuroprotection versus neurodestruction by route of administration

Stage: REPORT_EN
Closing date: 2026-08-11
Question: What mechanisms connect progesterone and cognitive function—neuroprotection versus neurodestruction by route of administration—with women's health and longevity, and what new hypotheses can be falsified?
Framework: mechanistic and translational report; it is not a clinical recommendation and does not prescribe a route, formulation, or regimen.

1. Executive scientific abstract

The available evidence does not support the simple dichotomy “oral progesterone is neurodestructive, non-oral progesterone is neuroprotective.” It does support a narrower statement: route, formulation, and dose markedly change peripheral partitioning between parent progesterone (P4) and neuroactive metabolites. In the small crossover by de Lignières et al. (n=9), 200 mg orally in a soft-gel capsule and 400 mg vaginally in a suppository produced relatively similar plasma P4 AUCs over 0–8 h, but very different exposures to allopregnanolone (Allo) and pregnanolone. Recalculation from the tables gives (AUC_Allo + AUC_pregnanolone)/AUC_P4 = 3.127 orally and 0.1184 vaginally: an aggregate separation of approximately 26.4× [1]. The same general direction appears in the equal-dose crossover by Nahoul et al. (n=6), although the estradiol co-intervention and its route prevent a clean causal replication [2].

That plasma difference is not a brain dose. No human data connect, within one experiment, P4 route → free stereopanel in plasma and cerebrospinal fluid (CSF) → receptor occupancy or function → persistent cognition → cognitive trajectory. The oral–vaginal comparison also confounds route with dose and formulation; the vaginal curves remained on a plateau or were still rising at 8 h; and the maximum of a mean curve is not the mean of individual maxima. Therefore, the 26.4× ratio is evidence of aggregate peripheral separation, not a validated waveform, a comparison of brain AUC, or a causal effect size for memory.

The strongest mechanistic branch is acute: Allo and other 3α-reduced neurosteroids positively modulate GABA_A receptors; during encoding or consolidation, greater inhibition can cause drowsiness, slowing, and a selective impairment of episodic memory without neuronal injury. Intravenous Allo administration to healthy women supports this time-dependent and domain-specific pharmacodynamics [4,5]. This signal does not demonstrate neurodestruction: lower performance during exposure, structural damage, and persistent decline are different outcomes.

A second, slower and context-dependent branch assigns parent P4 actions through PGR, PAQR7, and PGRMC1 to survival, inflammation, metabolism, plasticity, and injury response. Most of that evidence comes from cells or animals under a defined injury. The two phase III progesterone trials in traumatic brain injury (PROTECT III, n=882; SYNAPSE, n=1195) were null despite favorable preclinical signals [17,18]. This discordance prohibits using “animal neuroprotection” as a substitute for human cognitive benefit.

After a neurosteroid is removed, receptor–network modification may persist through phosphorylation, trafficking, or changes in GABA_A subunit composition. Its direction, however, is not universal. Preclinical systems have shown both withdrawal-associated adaptation and sustained PKC/β3-dependent enhancement of inhibition after washout of selected neurosteroids [12–14]. “Excitatory rebound” must therefore not be assumed; the appropriate hypothesis is a brief metabotropic memory dependent on ligand, assembly, exposure history, and network state.

The explanation that best fits the current evidence is: route changes a peripheral mixture; that mixture may acutely and reversibly modulate GABA_A function; route-dependent persistent brain injury or protection has not been demonstrated. The highest-priority competitor is that the blood–brain barrier, local metabolism, and tissue sequestration attenuate or reconfigure the plasma mixture before it reaches the synapse. The smallest decisive experiment is not another total-P4 study: it is a multiligand surface, within human concentration ranges and with stereospecific metrology, tested first in two human GABA_A assemblies and then in qualified human networks. Only if those gates are passed is a human crossover study with aligned compartments and times justified.

The connection to women's health and longevity remains open, not established. An acute effect on encoding could transiently affect performance and daily function; a persistent alteration would matter to healthspan only if it recurs, survives washout, changes a validated mediator, and predicts cognitive slope or years free of disability. No available study completes that bridge. Current Pharma relevance is discovery-stage H0–H1: mixture pharmacology and exposure shapes, not an asset, biomarker, formulation, or indication ready for clinical translation.

Concrete scientific delta of this report

This report replaces a binary route classification with a falsifiable multiligand surface + central filter + receptor–network state model, establishes that the ≈26.4× plasma separation is technically robust but causally insufficient, elevates central attenuation/reconfiguration to the priority competitor, and redefines “neurodestruction” as a chain that requires persistence and injury rather than as a cognitive decrement during sedation.

2. Scientific question and relevance

The question contains three problems that must be separated:

1. Pharmacokinetics: what mixture of P4, Allo, pregnanolone, 5α-dihydroprogesterone (5α-DHP), isopregnanolone (ISO), and other steroids is produced by each route, formulation, and dose?
2. Pharmacodynamics: how does that mixture modify excitability, plasticity, encoding, consolidation, and injury response?
3. Trajectory: does any effect persist and alter sustained function, risk of decline, or years of independent cognitive life?

Conflating them creates two symmetrical errors. The first calls drowsiness or worse test performance at peak exposure neurotoxicity. The second calls a molecular survival signal neuroprotection without functional recovery, or treats an animal result that fails to replicate in humans as protection.

For women's health, the distinction matters because endogenous exposure changes across the cycle, pregnancy, postpartum, and the menopausal transition, whereas exogenous exposure depends on dose, formulation, coadministration of estrogen, time since menopause, and individual metabolism. Relevant longevity is not only survival but preservation of executive function, episodic memory, sleep, mobility, and autonomy. A longevity hypothesis requires, at minimum, a verifiable longitudinal mediation chain; it cannot be inferred from an isolated plasma AUC.

3. Scope, population, and life stage

The translational core is restricted to cognitively intact women aged 40–65 years, stratified by STRAW+10 and time since the final menstrual period. This range includes perimenopause and early postmenopause, when steroids, sleep, and symptoms fluctuate, while avoiding untested projection from healthy young women, pregnancy, postpartum, epilepsy, traumatic brain injury, or Alzheimer's disease.

Data from other populations are retained as mechanistic evidence or tests of transportability:

• healthy young women: acute Allo pharmacodynamics and memory;
• postmenopausal women: oral P4 pharmacokinetics and cognitive studies with co-interventions;
• men or mixed samples: mechanistic support only, never the primary estimate for women;
• animals and cells: plausibility, localization, and perturbation design;
• traumatic brain injury and neurodegeneration: translation stress tests, not equivalents of healthy aging.

“Oral,” “vaginal,” “transdermal,” or “intravenous” are not compared as abstract labels. The causal unit is route × formulation × dose × schedule × co-intervention × endocrine state. Micronized P4 is also not equated with synthetic progestogens.

4. Background knowledge and mechanism map

4.1 From administration to the peripheral mixture

Oral P4 passes through the gut and portal circulation, favoring first-pass generation of reduced metabolites. Vaginal administration avoids part of that first pass, changes absorption and local distribution, and may sustain parent P4 across a different window. Small human crossovers support a higher relative metabolite burden after oral P4 and a higher parent fraction after vaginal administration [1,2]. Formulation and dose prevent attribution of the entire contrast to route anatomy.

P4 is converted by 5α-reductase to 5α-DHP and by 3α-hydroxysteroid dehydrogenase to Allo; 5β pathways generate pregnanolone. ISO is a stereoisomer with different pharmacology and possible functional antagonism of some Allo effects. Summing concentrations ignores potency, efficacy, antagonism, protein binding, and nonlinearity. The simple ratio is useful for demonstrating partitioning, not for predicting a receptor response.

4.2 From plasma to the brain compartment

P4 and neurosteroids are lipophilic, but “crosses the barrier” does not imply plasma-to-synapse proportionality. Free fraction, albumin and globulin binding, endothelial uptake, efflux, endothelial/astrocytic/neuronal metabolism, local synthesis, membrane sequestration, and clearance all matter. Muk et al. measured nine steroids by LC-MS/MS in paired plasma and CSF from 204 participants across cognitive impairment, Alzheimer's disease, and vascular dementia [19]. The small decrease in plasma P4 among women with Alzheimer's disease does not identify causality, exogenous route, or a complete Allo stereopanel. It demonstrates compartmental measurement feasibility, not validation of plasma as a synaptic surrogate.

4.3 Fast GABA_A branch

Allo and pregnanolone potentiate GABA_A at sites distinct from the GABA site; the effect depends on concentration, receptor composition, and ambient GABA tone. α1β3γ2 represents a common synaptic configuration; α4β3δ represents an extrasynaptic configuration sensitive to neurosteroids. If the chloride gradient makes GABA hyperpolarizing, greater current lowers network gain. During a memory task, this can reduce attention, encoding, or consolidation without killing cells.

Intravenous Allo produced acute pharmacokinetic and behavioral effects in healthy women [4]. In Kask et al., impairment was selective for components of episodic memory rather than a uniform loss of semantic or working memory [5]. A recent oral extended-release P4 study showed peak plasma Allo at approximately 2 h and transient sedation in 10 volunteers, but 9 were men, the assay used ELISA, and repeat-dose extrapolation was modeled [8]. It is weak temporal support, not central evidence in women.

4.4 Slow parent-P4 branch

P4 can activate nuclear PGR and pathways associated with PAQR7 and PGRMC1. In injury contexts, cellular and animal models describe changes in apoptosis, inflammation, mitochondria, myelin, BDNF, and AKT signaling. These pathways are biologically plausible, but three conditions limit interpretation:

• protection may exist only under a specific injury;
• molecular engagement is not functional recovery;
• a route with more plasma P4 does not guarantee more free P4 in the target brain cell.

The translation stress test is decisive: positive preclinical traumatic brain injury results did not translate into functional benefit in PROTECT III or SYNAPSE [17,18]. This does not show that P4 is neurodestructive; it lowers the credibility of generalizable neuroprotection.

4.5 Receptor–network memory after washout

P4/Allo exposure and withdrawal can modify GABA_A subunit expression, trafficking, or function in cells and animals [12,13]. In another system, Allo and the synthetic neurosteroid SGE-516—but not ganaxolone—left tonic potentiation after ≥30 min of washout; this depended on PKC, β3 phosphorylation, and increased surface receptors [14]. Every post-washout effect should therefore be treated as:

ligand × assembly × concentration × duration × exposure history × network state.

The direction may be inhibitory, excitatory, or null. Calling every effect “withdrawal” or “rebound” conceals the mechanism being tested.

4.6 Candidate multiscale chain

route/formulation/dose → free peripheral mixture → central filtering/reconfiguration → GABA_A occupancy and state + P4 receptors → network gain during exposure → possible post-washout memory or injury response → persistent cognitive performance → cognitive slope/healthspan.

The first two links have direct but limited human evidence. The acute receptor-to-performance branch has partial human evidence. Links from central filtering to persistence, and from persistence to longevity, are unvalidated.

5. Evidence assessment method

The prior scope, map, verification, mechanism, computational decision, computation, hypothesis, adversarial-review, and experimental-design artifacts were integrated. Recent or uncertain claims were checked against primary scientific sources. Each link was classified as HUMAN-DIRECT, HUMAN-INDIRECT, ANIMAL, IN-VITRO/EX-VIVO, COMPUTATIONAL, or INFERRED.

Four rules were applied:

1. a metabolite association does not establish mediation;
2. absence of significance does not establish equivalence;
3. a decrement during exposure does not establish injury;
4. a survival signal does not establish functional benefit or longevity.

Hypotheses were retained only when they specify population/system, exposure, window, comparator, outcome, margin, and kill criterion. Results with concomitant estrogen were interpreted as combined regimens, not isolated P4 effects.

6. Evidence map

7. Contradictory evidence, null findings, and adversarial evidence

7.1 The strongest pharmacokinetic contrast is confounded

de Lignières et al. compared 200 mg orally with 400 mg vaginally and used different formulations [1]. P4 AUC over 0–8 h was 46.16 versus 54.72 ng·h/mL, but Allo was 92.92 versus 5.11 and pregnanolone was 51.44 versus 1.37. The aggregate contrast is large, but it does not identify the pure effect of route. In Nahoul et al., P4 dose was equal, but P4 and E2 were given through opposite routes across cycles [2]. Directional agreement improves plausibility; it does not remove confounding.

7.2 Human cognition does not show a uniform signature

Repeated oral-P4 studies report null, small, or domain- and time-dependent results. Sedation can worsen performance without changing storage; practice, motor speed, sleep, and expectations can mimic a cognitive effect. Estradiol trials that include vaginal P4 only for participants with a uterus do not isolate P4's contribution [9]. A global average of “cognition” therefore combines incompatible functions and regimens.

7.3 Continuity and intermittency are unresolved

In Alzheimer's disease models, continuous Allo exposure was associated with adverse outcomes in some studies, whereas intermittent schedules have been linked to neurogenesis or recovery [11]. Sex, age, model, pathology, dose, AUC, pulse, and outcome change simultaneously. “Continuous is bad/intermittent is good” is a hypothesis generator, not a law.

7.4 TBI failure limits general neuroprotection

PROTECT III did not improve functional outcome (RR 0.95; 95% CI 0.85–1.06) and SYNAPSE was also null (OR 0.96; 95% CI 0.77–1.18) [17,18]. Although the population and injury differ from healthy aging, failure of two large programs shows that plausibility and animal efficacy are insufficient.

7.5 The central competitor may erase the peripheral signal

The dominant inference assumes that the plasma relationship among compounds reaches the receptor. An equally plausible competitor is that endothelium, astrocyte, and neuron lower the amplitude, change ligand ratios, or generate new metabolites. If the central output surface is nearly proportional but small, route could matter quantitatively; if it is reconfigured, a peripheral ratio could even rank conditions incorrectly.

7.6 Neurodestruction requires stronger criteria

“Neurodestruction” is reserved for cell loss, structural damage, persistent synaptic loss, or functional decline that survives exposure, sleep, and response-speed effects. Sedation, poorer encoding during the peak, or reversible homeostatic plasticity do not meet that standard. Under this criterion, there is no direct human evidence of route-dependent neurodestruction.

8. Multiscale mechanistic synthesis

The integrative explanation requiring the fewest assumptions is temporal separation:

• minutes–hours: neuroactive metabolites modify GABA_A, drowsiness, speed, and encoding;
• hours–days: phosphorylation, trafficking, and subunits may retain a ligand-dependent post-washout state;
• days–weeks under injury: P4 receptors may modify survival and recovery, but benefit is contextual;
• months–years: only a persistent, repeated, and functionally relevant effect could alter cognitive slope or healthspan.

Route is not the proximal mechanism; it is an exposure generator. The receptor is not the outcome; it is a mediator. Acute episodic memory is not longevity. The model avoids these jumps and identifies the gates that must be closed:

1. identity and free concentration;
2. multiligand surface in a human receptor;
3. coherence in a human network;
4. preservation or transformation across the central compartment;
5. contemporaneous effect versus persistence;
6. recovery under injury;
7. external longitudinal prediction.

9. Computational layer

The published curves from de Lignières et al. were independently digitized at 300 and 600 dpi. The integrated ratio from tables was 3.127 orally and 0.1184 vaginally, or 26.409×; digitization reproduced approximately 26.39×, and sensitivity analyses maintained separation above 21.65×. This supports the existence of two clearly separable aggregate plasma mixtures.

Two gates failed:

• C0, calibration of maxima: the maximum of the mean curve underestimated the tabulated mean of individual Cmax values by 17.7% to 42%; it cannot be used as a representative individual Cmax.
• C2, temporal closure: vaginal P4 and Allo remained on a plateau or were rising at 8 h, and there was no complete temporal curve for 5α-DHP; AUC over 0–8 h is not total AUC.

Consequently, no D0 waveforms were issued as if they were brain doses. The computational layer authorizes only one claim: robust aggregate peripheral separation within the observed window. It does not authorize an exact temporal shape, total-P4 equivalence, synaptic composition, or a cognitive effect.

For the next stage, the following prespecified models will be compared:

• M0: scalar (Allo + pregnanolone)/P4;
• M1: linear model of absolute free concentrations;
• M2: nonlinear surface with interactions;
• M3: M2 + assembly and state/desensitization.

Comparison will use blind holdouts. If M0 predicts as well as or better than the other models, the new multiligand hypothesis dies or is simplified.

10. Primary hypothesis

L7-6-AR-H1A v1 — human multiligand surface

Status: proposed after adversarial split.
Maturity: integrated H0; partial components H1.
Prior confidence: 0.34.

Statement. In assembled human GABA_A receptors and qualified human neural networks, a prespecified nonlinear surface of free Allo, pregnanolone, ISO, and P4 concentrations predicts current and network gain better than the scalar (Allo + pregnanolone)/P4, total AUC, or route label.

Rationale. Ligands differ in potency, efficacy, stereoselectivity, and possible antagonism; synaptic and extrasynaptic receptors differ in sensitivity. A sum weighted after inspecting data would be flexible but not falsifiable, so models and holdouts must be frozen before analysis.

Predictions. (a) M2/M3 will reduce holdout RMSE by ≥20% relative to the best M0/M1; (b) mixture ordering will change between α1β3γ2 and α4β3δ; (c) the human network will reproduce the functional direction of current without cytotoxicity; (d) mixtures with the same scalar but different composition will produce different responses.

Kill criteria. It dies if the scalar predicts equally well or better in holdout; if there is no effect within the plausible human range; if the result appears only in one assembly, batch, or supranormal concentration; or if current and network are incoherent after controlling chloride, maturity, and toxicity.

11. Competing hypothesis

L7-6-HG-H2N v2 — contemporaneous acute effect, without demonstrated persistence

Status: strengthened relative to alternatives, still unproven.
Maturity: acute H1; post-washout equivalence H0.
Prior confidence: 0.50.

Statement. After controlling compound residue, sleep, drowsiness, practice, speed, baseline activity, and estradiol, cognitive and network differences associated with a neurosteroid-rich mixture are concentrated within the exposure window; after washout, receptor, network, and structural results remain within prespecified equivalence margins.

Rationale. This is the minimal explanation for coincidence between the neurosteroid peak, sedation, and episodic alteration. It does not deny receptor–network memory; it requires that memory to be demonstrated with measured free exposure and formal equivalence testing.

Predictions. (a) an effect at peak proportional to free exposure and GABA_A pharmacodynamics; (b) once residue disappears, the functional area after threshold crossing will be small; (c) without injury there will be no persistent cellular, synaptic, or functional loss; (d) TOST will support post-washout equivalence within ±20% only if assay reliability permits it.

Kill criteria. It dies if, with residue-free medium and cells, a reproducible effect outside the margin appears at 24 h, replicates in four additional female backgrounds, shows an engagement mechanism, and is accompanied by functional alteration not explained by chloride, maturity, stress, or nonspecific toxicity.

Associated adversarial hypothesis. L7-6-HG-H3 v2 proposes a brief metabotropic memory of unspecified direction. It has confidence 0.22 and is activated if the effect persists after genuine washout. L7-6-AR-H4C v1, confidence 0.42, proposes that the central compartment attenuates or reconfigures the mixture; it is the priority competitor before any route interpretation.

A separate mechanistic alternative, L7-6-AR-H1B v1 (confidence 0.18), restricts parent-P4 protection to a defined injury and receptor. It does not compete with acute pharmacodynamics in healthy tissue: it dies if engagement is not accompanied by loss of effect after receptor perturbation, rescue, and functional recovery.

12. Translational hypothesis

L7-6-HG-HT1 v2 — functional signature of exposure and trajectory

Status: parked and weakened.
Maturity: H0.
Confidence: 0.05.
Control: HUMAN_QA_REQUIRED before any clinical interpretation.

Statement. Only if receptor, network, compartment, and persistence gates survive will a low-dimensional functional vector—frozen before validation—that combines free exposure and receptor–network response add external prediction of cognitive function beyond P4, E2, route, sleep, speed, menopausal stage, and baseline performance.

Predictions. The vector will reproduce across laboratories and batches, retain calibration in an external cohort, and improve incremental prediction without depending on a single route. Its association with cognition will persist after adjustment for drowsiness and speed.

Kill criteria. It dies if any prior gate fails; if the vector adds no external prediction; if calibration fails by batch or reproductive stage; if the signal disappears after adjusting for sleep/speed; or if it predicts only the acute peak and not a persistent measure.

This hypothesis does not authorize inference about dementia prevention or longevity. Those claims would later require a longitudinal cohort with cognitive slope and years free of disability as outcomes, with therapeutic and endocrine confounding controlled.

13. Falsifiable predictions and decision rules

p>0.05 will not be interpreted as equivalence. Margins will be fixed by technical reliability and biological relevance; if test–retest variability exceeds 10%, the ±20% margin is not defensible and the gate remains inconclusive.

14. Smallest discriminating experiment and gate sequence

The sequence is:

Q0 metrology → D−1a receptor → D−1b network → B0 compartment → E0 human PK/PD → D0 persistence → D1 injury → E1 trajectory.

14.1 Q0: stereospecific metrology

Before biology, LC-MS/MS will verify identity, purity, stereospecificity, free concentration, stability, and adsorption of P4, 5α-DHP, Allo, pregnanolone, and ISO. Criteria: intra-assay CV ≤10%, inter-assay CV ≤15%, recovery 80–120%, carryover ≤1%, and explicit mass balance. Free and total fractions will be measured in medium and cells. Failure prevents interpretation of any surface.

14.2 D−1a: human receptor surface—the smallest decisive experiment

α1β3γ2 and α4β3δ will be used with seven individual concentration curves, a D-optimal design of 24 mixtures, and eight blind holdouts, with GABA at EC10. Primary outcomes will be current or open probability, steady-state/pulse response, and desensitization. Four independent expression batches per assembly, at least six cells per condition, and confirmation in two new batches are required.

M0–M3 will be compared without redefining them after viewing holdouts. M2/M3 must improve RMSE by ≥20%; if performance is within 5% of the simpler model, parsimony is chosen. Absence of a material effect within ±20% across human ranges kills the extension “route separation is functionally material.”

14.3 D−1b: human network

Four female iPSC neuron–astrocyte backgrounds will be studied: two for discovery and two for blind replication, with two differentiations and four wells per condition. Before inclusion, cultures will be qualified for E_GABA by gramicidin perforated patch, KCC2/NKCC1, maturity, and MEA stability. The primary outcome is network gain; patch-clamp is confirmatory. Advancement requires coherent current and network effects without toxicity.

14.4 B0: endothelial–astrocyte–neuron filter

Four female backgrounds and three devices per background will receive selected mixtures. Sampling will occur at 0, 0.5, 2, 6, 12, and 24 h; mass balance must be 80–120%. Central output below one-third of input supports strong attenuation. Proportionality within 0.80–1.25 refutes the strong reconfiguration version. Newly formed metabolites, not only disappearance of the parent compound, will be measured.

14.5 E0: human PK/PD crossover, conditional on prior gates

Participants: cognitively intact women aged 40–65 years, STRAW+10 +1a/+1b/+1c, within 10 years of the final menstrual period. Randomized, double-blind, double-dummy, three-period design: oral P4, vaginal P4, and placebo, with doses/formulations justified by exposure rather than nominal equivalence. Up to 36 would be recruited to obtain 30 analyzable participants.

Samples and PD would be collected at 0, 1, 2, 3, 4, 6, 8, 12, 24, and 48 h. The free and total stereopanel, exclusion panel, E2, and modifiers would be measured. Episodic memory would be the exploratory primary cognitive PD; sleep, drowsiness, and speed would be modeled separately. With 30 participants, the design detects approximately a standardized within-person effect of 0.53 with 80% power; it is not powered to demonstrate post-washout equivalence or small trajectory effects.

14.6 D0: genuine persistence

Four discovery and four replication backgrounds will be used with a single mixture or ligand derived from E0. Windows are 0, 0.5, 2, 6, 12, and 24 h; 48 h only if a signal exists at 24 h. The primary outcome will be the area of functional deviation after crossing the exposure threshold. Residue will be measured in medium and cells; E_GABA, baseline activity, stress, viability, and cellular composition will be controlled. TOST with a 90% CI and a ±20% margin is used only if test–retest CV ≤10%.

14.7 D1: protection under a defined injury

One injury and one receptor will be prespecified, without an open exploratory panel. Four female discovery and four replication backgrounds, three differentiations, and four wells will be used. Functional recovery is primary. Engagement, signal loss by antagonism/knockdown, rescue, and recovery must align. BDNF or pAKT alone does not constitute neuroprotection.

14.8 E1: trajectory

Only after D0/D1 is it justified to validate the frozen vector in an external cohort and then estimate cognitive slope. The longevity outcome would be years free of functional decline, not a baseline concentration. This phase would require independent human oversight and ethical/statistical review (HUMAN_QA_REQUIRED).

15. Biomarkers and stratification

Exposure biomarkers

• free and total P4, 5α-DHP, Allo, pregnanolone, and ISO by stereospecific LC-MS/MS;
• AUC, individual Cmax, time above threshold, and rise/fall rate;
• mass balance and plasma-to-central-output relationships, not a single ratio.

Engagement biomarkers

• GABA_A current, desensitization, and extrasynaptic tone;
• β3 phosphorylation, surface trafficking, and subunit composition only as secondary mediators;
• for parent P4: occupancy/perturbation of the selected receptor and loss–rescue evidence.

Functional outcomes

• episodic memory separated into encoding, consolidation, and retrieval;
• response speed, drowsiness, and sleep as separate covariates/outcomes;
• network gain and functional recovery in cellular models;
• cognitive slope and everyday function only in a longitudinal phase.

Minimum stratification

STRAW+10, years since final menstrual period, uterus/ovaries, prior hormonal exposure, contemporaneous E2, baseline sleep, chronotype, GABAergic medications, liver function, alcohol, body composition, and variants/expression of 5α-reductase/3α-HSD enzymes when feasible. Stratification is used for prespecified heterogeneity, not to rescue a null primary outcome.

16. Individual and life-stage variability

The same dose can produce different profiles because of absorption, first pass, microbiota, liver function, protein binding, and enzyme activity. Brain response may vary with chloride gradient, GABA_A composition, sleep state, stress, and prior endocrine exposure. During perimenopause, endogenous fluctuation may increase variance and modify receptor–network adaptation; in stable postmenopause, endocrine variance may decrease, but age, comorbidity, and time since hormone loss change.

This heterogeneity favors within-person designs and dense temporal sampling. It does not justify individualized conclusions from one metabolite or extrapolation among pregnancy, postpartum, luteal phase, and menopause. A group mean may conceal opposite profiles; exposure shape and interactions must therefore be prespecified and externally replicated.

17. Pharma relevance and translational maturity

What is scientifically actionable

• a discovery platform for mixture pharmacology in human receptors and networks;
• metrology and compartment criteria for comparing formulations without treating total P4 as a surrogate;
• identification of whether an exposure shape produces a contemporaneous effect, post-washout memory, or injury-specific protection;
• early stop rules that prevent advancement of a nonfunctional peripheral signal.

What is not ready

• no asset or formulation has demonstrated cognitive superiority;
• no validated biomarker of brain benefit or harm exists;
• there is no cognitive-prevention indication;
• there is no basis for ranking routes by “neuronal safety”;
• no evidence chain reaches longevity.

Overall maturity is H0–H1. The acute GABA_A hypothesis reaches partial H1; persistence, human protection, and trajectory remain H0. Any human study after the gates requires HUMAN_QA_REQUIRED, ethical review, and an explicit safety plan; this report does not recommend clinical use.

18. Limitations

1. Direct route studies are small, old, and confound route with dose, formulation, or E2.
2. The 0–8 h AUC from de Lignières does not capture the entire vaginal exposure; digitization cannot recover individual data.
3. Historical immunoassays may cross-react and do not separate stereoisomers with LC-MS/MS precision.
4. Plasma is not central free fraction, membrane, or synapse; complete human P4/Allo/pregnanolone/ISO pairs are missing.
5. IV Allo studies demonstrate acute sufficiency but not equivalence with oral-P4 metabolism.
6. Withdrawal and receptor–network memory evidence is preclinical and ligand/system-dependent.
7. Injury, TBI, and Alzheimer's disease models do not represent healthy cognitive aging; animal-to-human translation has failed materially.
8. Cognitive studies commonly mix domains, windows, and co-interventions and rarely separate sedation from memory.
9. The translational functional vector does not yet exist; proposing it does not demonstrate utility.
10. No longitudinal study connects route, central exposure, persistence, and years of independent cognitive function.

19. Conclusions

1. Route of administration materially modifies the peripheral mixture, but the published magnitude cannot be attributed exclusively to route or transported to the brain.
2. The most coherent human evidence supports acute, selective, and potentially reversible GABA_A modulation of sedation/encoding; it does not demonstrate neurodestruction.
3. Neuroprotection by parent P4 is contextual and mostly preclinical; two null phase III TBI trials prevent generalization.
4. Post-washout persistence is plausible, but its direction depends on ligand and receptor; it must be demonstrated with residue, chloride, network, and equivalence controlled.
5. The central competitor—attenuation or reconfiguration between plasma and synapse—is a priority causal gap.
6. The primary falsifiable hypothesis is that a multiligand surface predicts receptor/network better than route, AUC, or ratio. The competing hypothesis is that effects disappear after genuine washout. The translational hypothesis remains parked until those gates are passed.
7. The bridge to women's health and longevity remains closed: current evidence cannot infer benefit or harm to cognitive trajectory from a P4 route.

20. References

1. de Lignières B, Dennerstein L, Bäckström T. Influence of route of administration on progesterone metabolism. Maturitas. 1995;21:251–257. PMID: 7616875. DOI: 10.1016/0378-5122(94)00882-8.
2. Nahoul K, et al. Profiles of plasma estrogens, progesterone and their metabolites after oral or vaginal administration of estradiol or progesterone. Maturitas. 1993. PMID: 8515718. DOI: 10.1016/0378-5122(93)90064-O.
3. Andréen L, Spigset O, Andersson A, Nyberg S, Bäckström T. Pharmacokinetics of progesterone and its metabolites allopregnanolone and pregnanolone after oral administration of low-dose progesterone. Maturitas. 2006. PMID: 16406399. DOI: 10.1016/j.maturitas.2005.11.005.
4. Timby E, et al. Pharmacokinetic and behavioral effects of allopregnanolone in healthy women. Psychopharmacology. 2006. PMID: 16177884. DOI: 10.1007/s00213-005-0148-7.
5. Kask K, et al. Allopregnanolone impairs episodic memory in healthy women. Psychopharmacology. 2008. PMID: 18551282. DOI: 10.1007/s00213-008-1150-7.
6. Freeman EW, Purdy RH, Coutifaris C, Rickels K, Paul SM. Anxiolytic metabolites of progesterone: correlation with mood and performance measures following oral progesterone administration to healthy female volunteers. Neuroendocrinology. 1993. PMID: 7904330.
7. Schüssler P, Kluge M, Yassouridis A, Dresler M, Held K, Zihl J, Steiger A. Progesterone reduces wakefulness in sleep EEG and has no effect on cognition in healthy postmenopausal women. Psychoneuroendocrinology. 2008. PMID: 18676087. DOI: 10.1016/j.psyneuen.2008.05.013.
8. Hughes-Medlicott NJ, Nguyen H, Glue P, Barak Y. Allopregnanolone Concentrations After Ascending Single Dose Administration of Progesterone to Healthy Volunteers. Human Psychopharmacology. 2025. PMID: 40699217. DOI: 10.1002/hup.70012.
9. Henderson VW, et al. Cognitive effects of estradiol after menopause: A randomized trial of the timing hypothesis. Neurology. 2016. PMID: 27421538.
10. Wang JM, et al. Allopregnanolone reverses neurogenic and cognitive deficits in mouse model of Alzheimer's disease. Proc Natl Acad Sci USA. 2010. PMID: 20231471. DOI: 10.1073/pnas.1001422107.
11. Bengtsson SKS, et al. Primary continuous-exposure series: (a) Chronic Allopregnanolone Treatment Accelerates Alzheimer's Disease Development in AβPPSwePSEN1ΔE9 Mice. Journal of Alzheimer's Disease. 2012;31:71–84. PMID: 22495347. DOI: 10.3233/JAD-2012-120268; (b) Brief but Chronic Increase in Allopregnanolone Cause Accelerated AD Pathology Differently in Two Mouse Models. Current Alzheimer Research. 2013;10:38–47. PMID: 23157375. DOI: 10.2174/1567205011310010006; (c) Long-term continuous allopregnanolone elevation causes memory decline and hippocampus shrinkage, in female wild-type B6 mice. Hormones and Behavior. 2016;78:160–167. PMID: 26497250. DOI: 10.1016/j.yhbeh.2015.10.010.
12. Follesa P, et al. Allopregnanolone synthesis in cerebellar granule cells: roles in regulation of GABA(A) receptor expression and function during progesterone treatment and withdrawal. PMID: 10825399.
13. Mostallino MC, Mura ML, Maciocco E, Murru L, Sanna E, Biggio G. Changes in expression of the delta subunit of the GABA(A) receptor and in receptor function induced by progesterone exposure and withdrawal. Journal of Neurochemistry. 2006. PMID: 16879715.
14. Modgil A, Parakala ML, Ackley MA, Doherty JJ, Moss SJ, Davies PA. Endogenous and synthetic neuroactive steroids evoke sustained increases in the efficacy of GABAergic inhibition via a protein kinase C-dependent mechanism. Neuropharmacology. 2017. PMID: 27743930.
15. Tateiwa H, et al. The mechanism of enantioselective neurosteroid actions on GABA_A receptors. Biomolecules. 2023;13:341. DOI: 10.3390/biom13020341.
16. Sugasawa Y, et al. Site-specific effects of neurosteroids on GABA_A receptor activation and desensitization. eLife. 2020;9:e55331. DOI: 10.7554/eLife.55331.
17. Wright DW, et al. Very early administration of progesterone for acute traumatic brain injury (PROTECT III). N Engl J Med. 2014. PMID: 25493974. DOI: 10.1056/NEJMoa1404304.
18. Skolnick BE, et al. A clinical trial of progesterone for severe traumatic brain injury (SYNAPSE). N Engl J Med. 2014. PMID: 25493978. DOI: 10.1056/NEJMoa1411090.
19. Muk T, et al. Steroid hormones in Alzheimer’s disease: a molecular analysis of blood and cerebrospinal fluid. Alzheimers Res Ther. 2026. PMID: 42458616. DOI: 10.1186/s13195-026-02117-1.
20. Castelnovo LF, Thomas P. Progesterone exerts a neuroprotective action in a Parkinson’s disease human cell model through membrane progesterone receptor α (mPRα/PAQR7). Frontiers in Endocrinology. 2023;14:1125962. PMID: 36967764. DOI: 10.3389/fendo.2023.1125962.

Evidence closure. The negative conclusion is specific: no human evidence demonstrates route-dependent persistent cognitive injury or superior cognitive neuroprotection. The positive conclusion is also specific: a large peripheral separation and plausible acute pharmacology justify the sequential falsification program described here, beginning with metrology and a multiligand surface, not a clinical recommendation.