5α-reductase and conversion to allopregnanolone: individual variability and neurological consequences

Project: L7-2
Stage: REPORT_EN
Evidence cutoff: 2026-07-31
Classification: progesterone metabolomics, neurosteroids, and female neurobiology
Core population: women aged 35–55 during the late reproductive transition–perimenopause
Nature: scientific research report; not medical advice, diagnosis, a prescription, or an assessment of an individual case.

1. Executive scientific abstract

5α-reductase initiates the conversion of progesterone (P4) to 5α-dihydroprogesterone (5α-DHP); a second 3α step produces allopregnanolone (Allo), a positive allosteric modulator of GABA_A receptors. That biochemistry is real, but the common explanation—“less 5α-reductase produces less Allo and therefore worse brain function”—has not been demonstrated in women. There is no longitudinal human measurement of brain P4→5α-DHP→Allo flux, nor a validated quantitative blood–CSF/brain bridge in women aged 35–55. A serum concentration or ratio integrates substrate, source, formation, transport, conjugation, and clearance; it does not identify enzyme activity or neural inhibition.

The most informative human evidence favors two serial filters. The first is metabolic: P4 availability and trajectory, 5α reduction, 3α partitioning/back-oxidation/epimerization, transport, and elimination determine local exposure to Allo and isoallopregnanolone (ISO). The second is neural: GABA_A composition, assembly, surface abundance, phosphorylation, and the chloride gradient determine how that exposure becomes function. In PMDD, controlled studies do not find a large mean peripheral separation in Allo, whereas suppression/addback, TMS, MRS, and pharmacological challenges show phase- and circuit-dependent sensitivity. Dutasteride demonstrates that the 5α pathway is perturbable, not that brain Allo is the only causal molecule. The larger multicenter sepranolone trial did not meet its prespecified primary endpoint.

The revised primary hypothesis is a trajectory–adaptation mismatch: at equal Allo peak and AUC, a rapid or irregular trajectory can produce a different function if the receptor adapts on different timescales. This hypothesis is interpretable only if E_GABA is hyperpolarizing and stable; otherwise, incomplete neuronal maturation, KCC2/NKCC1, or the electrophysiological configuration can simulate hysteresis. The competing hypothesis that best explains current human data is systemic/metrological: phase, P4, source, pharmacokinetics, selection, and matrix explain or render the peripheral signal non-identifiable, without demonstrating that a central effect is null. The translational hypothesis—a minimal phenotype containing one exposure measure and one response measure—remains parked until causality, compartment concordance, reliability, and external gain are demonstrated.

The connection to female longevity remains open, not established. Low cortical Allo in Alzheimer disease is a post-mortem association compatible with secondary loss or redistribution. Preclinical evidence shows different signs for acute/intermittent and continuous exposure, with additional dependence on age, model, and sex. The only defensible bridge to brain health is prospective: causal mechanism → validated central exposure → specific proximal function → multiyear trajectory → disease-free years. No available source completes that chain.

Concrete scientific delta of REPORT_EN: the final synthesis replaces the uniform-deficiency narrative with an identifiable model of local metabolism × exposure trajectory × receptor adaptation × chloride homeostasis; it establishes that the first decisive experiment is assay-first and that no bridge to brain health, Pharma, or longevity should open without separating stereochemistry, flux, compartment, and the functional sign of GABA_A.

2. Scientific question and relevance

The exact question is:

In adult women exposed to endogenous changes in P4, especially during the late reproductive transition–perimenopause, is there reproducible variability in P4→5α-DHP→Allo flux that modifies a proximal GABA_A function, or is neurological heterogeneity better explained by ovulatory architecture, source/clearance, downstream metabolism, stereochemistry, hormonal trajectory, and receptor adaptation?

The relevance to women's health arises from a temporal coincidence: before final menopause, anovulation, luteal-phase variability, and P4 irregularity increase. This may change not only the amount of substrate, but also the slope, duration, and frequency of neurosteroid exposure. Sleep, stress response, affect, acute cognition, and seizure threshold are plausible proximal functions, but they are not interchangeable. The longevity question is whether repeated windows of dysfunction leave a persistent trajectory of brain vulnerability; this is currently a hypothesis, not a conclusion.

Six links must be identified separately:

1. Substrate: bioavailable P4 and its temporal trajectory.
2. Flux: the fraction passing through 5α and then 3α.
3. Source/compartment: periphery, local neural synthesis, entry, and exit.
4. Fate: Allo, ISO, 5β/20α products, conjugates, and back-oxidation.
5. Response: GABA_A assembly, chloride, network state, and exposure history.
6. Health trajectory: proximal function, multiyear persistence, and a separate clinical outcome.

3. Scope, population, and life stage

The primary population is nonpregnant women approximately 35–55 years old, followed within person from the late reproductive stage through early and late perimenopause. Stage must be defined by repeated menstrual history and STRAW+10 criteria; ovulation requires an orthogonal method and cannot be defined by P4 alone. Chronological age is not equivalent to perimenopause.

Required strata:

• ovulatory cycles in the late reproductive stage;
• early perimenopause with documented ovulation;
• late perimenopause, separating ovulatory and anovulatory cycles;
• natural postmenopause as a boundary of low ovarian production, not as a substitute for transition;
• follicular, periovulatory, early/mid/late luteal, and withdrawal phases;
• rising, plateau, oscillatory, and withdrawal hormone exposures;
• hormone or drug use, hepatic/renal function, E2, sleep, HPA, temperature, and medication as prespecified sources or modifiers.

PMDD, PTSD, pregnancy/puerperium, and Alzheimer disease are used as separate mechanistic perturbations. They are not pooled because they differ in hormonal magnitude, placenta, pathology, age, selection, and circuit. iPSC models, cocultures, and organoids test general human mechanisms; reprogramming may erase age and endocrine history and does not by itself model acquired perimenopause.

Outside scope: prescribing, individual advice, comparative treatment evaluation, product or marketing proposals, use of current Lua data, and any direct inference from symptoms or a diagnosis to longevity.

4. Background knowledge and mechanism map

4.1 Metabolic pathway mechanism

ovulation/luteal architecture + peripheral/neural sources
        ↓
available P4(t)
        ↓  SRD5A1/2 + NADPH
5α-DHP(t)
        ├── 3α reduction → Allo (3α,5α) → positive GABA_A modulation
        ├── 3β reduction → ISO (3β,5α) → functional buffering of Allo
        ├── back-oxidation → 5α-DHP
        └── 20α products/conjugates/exit

Human temporal tissue from epilepsy surgery demonstrated local 5α/3α capacity with an androgenic substrate; SRD5A1 was detectable and SRD5A2 was not in that tissue. This supports regional capacity, not in-vivo P4 flux or universal dominance of one isoform. The downstream step is reversible and depends on enzyme, cofactor, and compartment. Recombinant biochemistry favors AKR1C2 as a positive control for Allo formation; AKR1C3/HSD17B6 are RNA-localized candidates, not demonstrated effectors. HSD17B10 can oxidize Allo, but its essential mitochondrial functions invalidate a simple knockout as a specific test.

Local concentration can be represented as:

[ C_{Allo}(t)=local\ formation+entry-back\text{-}oxidation-epimerization-conjugation-exit. ]

Therefore, Allo/P4 or Allo/5α-DHP are not enzyme activities. A 5α bottleneck predicts joint decreases in 5α-DHP and Allo; a downstream bottleneck predicts preserved or accumulated 5α-DHP with lower Allo; a transport defect can change medium/plasma without changing intracellular formation.

4.2 Receptor and chloride mechanism

Allo modulates pentameric GABA_A receptors. Extrasynaptic configurations containing a δ subunit often support tonic inhibition and show high neurosteroid sensitivity; γ2-containing receptors contribute to phasic signaling. However, GABRA4/GABRD/GABRG2 RNA does not prove coexpression, the associated β subunit, assembly, surface abundance, or current.

Functional effect depends on the chloride driving force. A GABA_A current is not automatically inhibitory: if E_GABA is depolarizing because of immaturity, the NKCC1/KCC2 relationship, or recording conditions, increasing conductance may not reduce excitability. Allo can also produce rapid modulation and sustained PKC-dependent changes in phosphorylation/trafficking. There is no basis for assuming a single adaptation timescale.

4.3 Dynamic mechanism model

Let E(t) denote effective neurosteroid exposure—free Allo, ISO, ambient GABA, and distribution—and R(t) the receptor state—assembly, surface abundance, phosphorylation, and chloride. Function is:

[ I(t)=F[E(t),R(t),E_{GABA},circuit]. ]

Mismatch arises when E(t) changes faster than one or more adaptations of R(t). This formulation generates a strong prediction: with equal free Allo, peak, AUC, and time above threshold, different temporal trajectories can produce different inhibitory gain; the effect must be preceded by an independent receptor readout and not be explained by E_GABA, residual Allo, or ambient GABA.

4.4 Bridge to brain health

The proposed bridge, still inferred, is:

repeated mismatch windows → specific alteration of sleep/excitability/stress → incomplete recovery in the same domain → functional trajectory.

Sedation, mood, TMS, MRS, or post-mortem Alzheimer findings cannot be skipped directly to healthspan or longevity. Recovery between windows may erase, attenuate, or reverse any accumulated burden.

5. Evidence method

This report integrates the verified SCOPING, EVIDENCE_MAP, EVIDENCE_VERIFICATION, MECHANISTIC_SYNTHESIS, COMPUTE_DECISION, COMPUTE_OPTIONAL, HYPOTHESIS_GENERATION, ADVERSARIAL_REVIEW, and EXPERIMENT_DESIGN artifacts. Decisive statements were checked against primary articles, PubMed/PMC, publisher DOI records, and trial registries. Recent searching was directed by the capacity to change hypotheses: PMDD 2024–2025, GABA maturation in iPSC, AKR1C2 catalysis, HSD17B10 pleiotropy, and stereospecific metrology.

Applied rules:

• distinguish human, animal, in-vitro/ex-vivo, computational, and inferred evidence;
• do not convert absence of significance into equivalence;
• preserve prespecified endpoints over post-hoc signals;
• keep blood, CSF, tissue, PBMC, TMS, MRS, and fMRI as distinct readouts;
• do not attribute an intervention that alters several steroids specifically to Allo;
• do not call a concentration, ratio, RNA, or protein a flux;
• do not extrapolate PMDD/puerperium/animal evidence to perimenopause without validation;
• do not call a symptom, organ, diagnosis, or short follow-up longevity.

Maturity scale: H0 mechanistic idea; H1 literature-supported links; H2 computational/dataset signal; H3 observational replication; H4 experimental/preclinical validation; H5 eligibility for human-reviewed partnering assessment. No integrated L7-2 hypothesis exceeds H0; some links reach H1 and transcriptomic localization reaches descriptive H2.

6. Evidence map

No source reaches high confidence for the complete chain. Evidence is strongest for biochemical capacity and perturbability; central individual variability, functional causality, and longevity remain unidentified.

7. Contradictory evidence and null findings

1. Similar peripheral Allo, but a perturbable pathway. PHASE and suppression/addback show no large mean Allo difference in PMDD; dutasteride reduces oscillation and symptoms in a small sample. Explanations: dynamics of several 5α steroids, a central subgroup, receptor sensitivity, or a small-study false positive. Smallest discriminator: tracer/bypass plus one proximal functional endpoint.

2. Phase-dependent function, but not in every circuit. MRS/TMS suggest adaptation; IV Allo preserved sedation in both groups and startle/PPI was null; a larger startle study was also null. This restricts the mechanism to circuit, dose, and window.

3. Initial positive sepranolone result, larger primary-negative trial. The 2021 trial did not separate treatment from placebo on the primary endpoint except for a limited signal; favorable results were post hoc, per protocol, cycle 3, and at 10 mg, not 16 mg. It does not validate simple ISO/Allo pharmacology.

4. Menopause and compartment. One GC/MS study found lower absolute serum Allo after menopause but an Allo/P4 ratio indistinguishable from the follicular phase; an earlier RIA study did not find a simple serum decline. Post-mortem brain levels were lower, but n=10 with confounding by age and cause of death. Method, source, and compartment remain explanations.

5. Allo can potentiate or reduce a current. In recombinant α4β2δ receptors, the effect depends on polarity, concentration, desensitization, and PKC. This does not negate general positive modulation, but it destroys the equivalence “more current = more inhibition.”

6. RNA and catalysis disagree. The atlas favors detectability of AKR1C3/HSD17B6; recombinant biochemistry demonstrates direct formation with AKR1C2 and weak or unshown activity of other candidates on the relevant substrate. Localization does not rank catalysis.

7. Allo and aging do not have a uniform sign. Low cortical Allo is associated with AD, but continuous exposure worsened some amyloid models and acute/intermittent exposure benefited other contexts. Reverse causality, regimen, age, genotype, and sex must compete.

8. An epistemic null is not a biological null. The current blood signal can disappear because of phase, PK, and measurement error even if a real brain difference exists. Only validated central exposure and reliable function could support biological equivalence.

8. Multiscale mechanism synthesis

8.1 Molecular/cellular mechanism

SRD5A1 is the most defensible initial isoform for an adult brain system. The downstream step must be resolved with labeled product: AKR1C2 as the forming control; AKR1C1/3, HSD17B6, and an SDR as partitioning rivals; HSD17B10 only with separation of function. The Allo/ISO balance may change functional sign at similar total 5α load, but the dominant enzyme and its human relevance are not established.

Transcriptomic distribution suggests two rival architectures: intraneuronal sufficiency or an astrocyte–neuron metabolic microcircuit. Coculture will count as a microcircuit only if it changes mass balance and a loss–rescue localizes the effect; a supernatant change may be transport or adsorption.

8.2 Receptor/network mechanism

The receptor is an adaptive, multiscale filter. Allo can modulate the channel immediately and, over hours, change phosphorylation, trafficking, and surface abundance. Response depends on E_GABA, KCC2/NKCC1, ambient GABA, β subunits, E/I balance, and circuit. An immature neural preparation can manufacture a paradoxical result without modeling any female vulnerability.

8.3 System/life-stage mechanism

In perimenopause, the central prediction is not a monotonic fall in Allo, but greater irregularity of P4(t) and downstream exposure. Ovulatory cycles, anovulatory cycles, and withdrawals must be analyzed separately. HPA, E2, and sleep may be confounders, mediators, or parallel pathways; the temporal DAG must determine their role before analysis, not after the result.

8.4 Health-trajectory mechanism

If windows repeatedly alter a function, they must precede the change, add beyond baseline function, and show incomplete recovery. Summing windows is not valid by default: adaptation and recovery may make effects nonadditive. Healthspan requires multiyear persistence and organ-specific outcomes; longevity requires disease-free years and competing risks. L7-2 has no such evidence.

9. Computational layer

CELLxGENE Census 2025-11-08 was used with Python 3.12.13, cellxgene-census 1.18.0, and tiledbsoma 1.17.1. The filter included primary brain tissue, female sex, normal disease status, and ages 35–55 when available. The final matrix contained 317,097 cells/nuclei, 24 datasets, and 48 dataset×donor pairs; observed ages were 36–38, 40–45, and 48–55. No stage contrasts were performed.

The pseudobulk unit was dataset×donor×region×class, with sensitivities of ≥50, ≥100, and ≥200 cells. SRD5A1 was robust in major classes; SRD5A2 did not pass the conservative gate. AKR1C3 and HSD17B6 were actionable in neurons/astrocytes; AKR1C4 had no detectable unit. In neurons at the ≥100 cutoff, GABRA4, GABRD, and GABRG2 had median positive fractions of 79.1%, 40.2%, and 87.4%.

Interpretation: prioritize SRD5A1, compare neurons with astrocyte–neuron coculture, and retain AKR1C3/HSD17B6 as localization candidates. This layer does not demonstrate protein, coexpression, catalysis, redox direction, Allo/ISO, GABA_A, life stage, or longevity. SEA-AD replication was not run because the version and object size changed; no external replication was fabricated.

10. Primary hypothesis: L7-2-AR-H1 v2

Role: narrow primary mechanistic hypothesis.
Status: weakened_narrowed.
Confidence: 0.30.
Maturity: integrated H0; H1 for the general trajectory/plasticity links.

Falsifiable statement: in human neurons with demonstrated hyperpolarizing GABA, stable E_GABA, free Allo, ambient GABA, and steroid metabolism, equal-peak/equal-AUC Allo trajectories will modify inhibitory gain only if they induce an independently measured receptor/trafficking change; the pattern must predict a held-out trajectory and cannot be explained by chloride or residual concentration.

Mechanism: rapid/irregular exposure → receptor phosphorylation and turnover on different timescales → transient mismatch between expected conductance and inhibitory gain → circuit-specific change in excitability.

Predictions:

1. E_GABA will remain within ±5 mV of baseline during contrasts.
2. A phosphorylation/surface estimate obtained in independent plates will predict the trajectory with the larger effect before electrophysiology is opened.
3. Temporal order will be receptor → inhibitory gain → excitability.
4. A held-out oscillatory waveform will replicate without refitting timescales.
5. Equalizing Allo/ISO and blocking a validated metabolic effector will not erase the trajectory×receptor interaction.

Evidence for: human endocrine manipulation, phase-dependent MRS/TMS, contextual pharmacology, and animal δ/α4 causality.
Evidence against: circuit-specific nulls, primary-negative sepranolone trial, no isolated human manipulation, maturation/chloride, and no central perimenopausal cohort.

Kill criteria: equivalence between trajectories in two backgrounds and replication; failed held-out prediction; effect explained by E_GABA, residual Allo, ISO, GABA, temperature, viability, or network composition; receptor change without a material functional change; or the need to refit a different timescale for every trajectory.

Smallest discriminator: Q1→D0C, described in Section 14.

11. Competing hypothesis: L7-2-AR-H3 v2

Role: systemic/metrological competitor that best explains the existing human evidence.
Status: strengthened_for_existing_human_data.
Confidence: 0.62 for the epistemic null; 0.20 for biological equivalence.
Maturity: epistemic H1; causal H0.

Falsifiable statement: in blood-based human associations, the signal attributed to 5AR/Allo will be explained or become non-identifiable after modeling phase, P4, source, PK, method, and selection; this does not imply central equivalence. A biological null will be accepted only if validated neural exposure and a reliable proximal function add no material effect.

Rival mechanism: ovulatory/energetic architecture and stage determine peripheral P4/Allo; E2, HPA, sleep, inflammation, and baseline health change neural function in parallel; method, matrix, clearance, and sampling add error.

Predictions:

1. Peripheral ratios will show low repeatability and phase/matrix sensitivity.
2. P4/ovulation, source, and PK will explain more blood variance than the clinical phenotype.
3. Blood will have insufficient or biased concordance with CSF in some women.
4. The neurosteroid block will not materially improve leave-one-cycle-out prediction beyond P4/E2, prior function, and temporal variables.
5. Any residual signal will be circuit-/stage-specific and will not transport to healthspan.

Evidence for: PMDD/peripartum nulls, PHASE, RIA–MS discrepancies, preserved follicular–postmenopausal ratio, and lack of blood–CNS pairs.
Evidence against: dutasteride, addback, brexanolone, and receptor models demonstrate perturbability in selected contexts.

Kill criteria: reproducible central flux or response precedes function, persists after prior confounders, and replicates; or a specific perturbation changes function at equivalent exposure/substrate.

Auxiliary rivals that must be measured within the same program:

• L7-2-AR-H1C v1: chloride/maturation explains apparent hysteresis; a priority control, not a complete human explanation.
• L7-2-AR-H2 v2: a local metabolic microcircuit changes Allo/ISO and function; weakened until catalysis, balance, and bypass are demonstrated.

12. Translational hypothesis: L7-2-AR-HT1 v2

Role: conditional metrological/translational hypothesis.
Status: parked; HUMAN_QA_REQUIRED.
Confidence: 0.02.
Maturity: H0; not eligible for H5.

Falsifiable statement: only after causality and metrology are validated separately will an exposure parameter with central concordance and a GABA_A response parameter improve external error and calibration for one proximal neurophysiological function versus phase, ovulation, P4/E2 AUC, baseline sleep, and prior function.

Predictions: each component passes stability, LLOQ, ICC/CCC/CV, and specificity before combination; the two-parameter model improves leave-one-cycle-out validation and external replication; gain persists beyond simple comparators; it does not require contemporaneous symptoms or a post-hoc panel.

Evidence for: the two filters are mechanistically distinct and isolated concentrations repeatedly fail.
Evidence against: no standardized assay, central bridge, test–retest reliability, or external validation exists; risk of overfitting and phase leakage is high.

Kill criteria: failure of either component, insufficient plasma–CSF concordance, no external gain, phase/batch instability, need for more than two components, or failure of the corresponding causal mechanism.

13. Falsifiable predictions and kill criteria

These are decision criteria. A valid null is preserved and stops expansion; it is not rescued by adding variables, subgroups, or endpoints after the data are opened.

14. Discriminating experiment

14.1 Minimal sequence

Q0 metrology → Q1 chloride/maturity
      └────────→ D0A-0 catalysis → D0A-1 flux/bypass
Q1 positive ─────────────────────→ D0C trajectory/receptor
positive functional D0 → D1X plasma–CSF → D1 two cycles

14.2 Q0 — metrology

Validate LC-MS/MS in buffer, cell/medium, plasma, and CSF. Separate P4, 5α/5β-DHP, Allo, ISO, pregnanolone, epipregnanolone, 20α products, conjugates, and tracers. Six runs over ≥3 days and two operators. Gates: bias/imprecision ≤15% in QC and ≤20% at LLOQ; neighboring isomer contribution <10%; spike balance 80–120% and ≥70% of label explained in biology or explicitly recorded as partial flux. If it fails, only profiles—not flux—are allowed.

14.3 Q1 — neural model

Six unrelated female iPSC backgrounds, three differentiations per background. Requirement: at least five backgrounds and two differentiations per background with E_GABA ≥10 mV more negative than rest, test–retest ±5 mV, GABA_A pharmacology, functional KCC2/NKCC1, firing/network maturity, acute Allo response, and an assembled surface receptor containing a β subunit. Gramicidin/perforated patch is mandatory. If it fails, change the model and do not interpret D0C.

14.4 D0A-0/D0A-1 — metabolism

Blind screen with labeled 5α-DHP, AKR1C2 as positive control, AKR1C1/3, HSD17B6, and verifiable SDRs as rivals; HSD17B10 only in a catalytic system or with separation of function. At most one former and one oxidant/epimerase advance if its IC90 clears blank, reaches ≥10% of AKR1C2 or produces a qualitatively distinct product in ≥5/6 runs, and replicates in a held-out run.

In cells, compare neuronal monoculture with astrocyte–neuron coculture, 40–70% partial reduction, rescue, and non-targeting control. Labeled P4 pulse, 5α-DHP bypass, and Allo/ISO bypass. Biological unit: 18 differentiations (6×3). SESOI_flux=max(20%,3×analytical CV). Require direction in ≥5/6 backgrounds. A metabolite without a functional effect kills the neural extension.

14.5 D0C — trajectory/receptor

In independent plates, estimate phosphorylation and surface abundance at 0, 0.5, 2, 6, 24, and 48 h and freeze a one- or two-timescale model before the endpoint. Deliver a slow trajectory, rapid withdrawal, and a held-out oscillatory waveform with free Allo, peak, AUC, and time above threshold equivalent within ±10%. Select one receptor node and compare intact, partial reduction, and rescue.

Primary: inhibitory gain by perforated current clamp with E_GABA within ±5 mV of baseline. SESOI_function=max(15%,2×test–retest CV). Hierarchical model by background/differentiation; TOST with IC90 for equivalence. The held-out waveform is evaluated once, without refitting.

14.6 D1X/D1 — conditional human bridge

D1X: 50 women aged 35–55 already scheduled for a procedure with clinical access to CSF; no puncture is performed solely for research. Plasma and CSF within 15 minutes, before intrathecal anesthetic. Primary: absolute Allo CCC. A no-go closes blood as a central proxy, not brain biology.

D1: only after a functional mechanism and compartment bridge. Sixty women not selected by symptoms, 30 late reproductive and 30 early perimenopausal, followed for two cycles. Six samples per cycle and TMS/SICI in follicular, mid-luteal, and withdrawal phases. One primary endpoint; systemic, +trajectory, and +response models validated leave-one-cycle-out. It is not powered for a stage interaction. A positive result demonstrates a proximal human association, not healthspan or definitive causality.

15. Candidate measures and stratification

No biomarker is validated. The following are candidate research measures:

Human stratification must be by stage/ovulation, ovulatory/anovulatory cycle, phase/slope, method/matrix, hormone medication, and hepatic/renal function. PMDD, PTSD, puerperium, and AD are not interchangeable strata. GABRD in PBMC is an experimental proxy and must not replace brain receptor.

16. Individual variability

Variability may arise at at least ten levels:

1. Ovulatory architecture: ovulation probability, luteal mass/duration, and P4 withdrawal.
2. Source: ovary, adrenal, periphery, or local neural synthesis.
3. 5α isoform: SRD5A1/2 by cell/region, still without a replicated female functional phenotype.
4. Downstream partitioning: formation, back-oxidation, ISO epimerization, and 20α diversions.
5. Cofactors/redox: may change net direction, with insufficient human in-vivo evidence.
6. Transport/conjugation/clearance: alter blood and medium without changing local flux.
7. Compartment: equal blood can coexist with different neural exposure.
8. Receptor: subunits, assembly, surface abundance, and phosphorylation.
9. Chloride/network: KCC2/NKCC1, ambient GABA, E/I balance, and circuit.
10. Temporal history and metrology: prior exposure, frequency, misdated phase, matrix, batch, and LLOQ.

A difference will be called “individual” only if it is reproducible within woman/background and exceeds analytical and phase variation. Genotype, RNA, or a single sample is insufficient. Six experimental backgrounds permit detection of large mechanisms and consistency, not population prevalence.

Life stage changes the distribution of these sources: phase dominates an ovulatory cycle; irregularity probably increases in perimenopause; substrate and sources change after menopause; extreme withdrawal dominates the puerperium; and tissue composition/pathology changes in AD. Sharing a metabolite does not imply sharing a mechanism.

17. Pharma relevance and maturity

Pharmacological relevance is conditional and limited to research hypotheses:

A positive D0A would yield a flux target and target-engagement assay, not a drug. A positive D0C would yield an experimental regimen rule, not clinical efficacy. A positive D1 would identify a proximal physiological signal, not an indication. All require expert human review; there is no readiness for safety, efficacy, companion diagnostics, or partnering.

Pharma no-go: peripheral-only effect; pleiotropic perturbation; result dependent on immature chloride; lack of steroidomic balance; benefit inferred from symptoms; or continuous exposure without a safety window. HUMAN_QA_REQUIRED remains active.

18. Limitations

1. There is no longitudinal P4→5α-DHP→Allo brain flux measurement in living women.
2. There is no validated quantitative blood–CSF/brain bridge in the target women; CSF is not equivalent to a synapse.
3. Human brain tissue comes from epilepsy surgery or post mortem.
4. PMDD dominates dynamic human evidence but does not automatically transport to perimenopause.
5. Mechanistic human samples are small; several studies have n=8–30 and were not equivalence studies.
6. Dutasteride affects multiple 5α steroids; P4 produces multiple metabolites; sepranolone modifies response rather than formation.
7. RIA, MS, serum/plasma, and mixed epimers create material heterogeneity.
8. TMS, MRS, fMRI, and PBMC are indirect, circuit-/tissue-specific readouts.
9. The computational atlas mixes platforms/regions, uses age rather than stage, and lacks an executed SEA-AD replication.
10. RNA is not protein, assembly, cofactor, direction, or activity.
11. iPSC may erase age/hormonal history and maintain depolarizing GABA.
12. HSD17B10 and other candidates have pleiotropic functions that complicate attribution.
13. Animal studies differ by species, regimen, age, sex, and model; some AD results use males.
14. Low cortical Allo in AD is cross-sectional and compatible with reverse causality.
15. SESOIs and experimental gates are operational and must be frozen without viewing biological contrasts.
16. Healthspan and longevity have not been evaluated; cycle follow-up or 12 months would not do so.

19. Conclusions

The 5α pathway is necessary to form Allo from P4, but it has not been demonstrated that an individual 5α-reductase deficiency explains female neurological variability. Human literature supports local capacity, phase dependence, and perturbability while rejecting a uniform peripheral deficiency as a general model. The best-supported human hypothesis today is epistemic: blood, phase, P4, source, PK, and method prevent identification of the central mechanism.

The best experimental hypothesis is more specific: a neurosteroid trajectory may exceed receptor adaptation and alter inhibitory gain at equal exposure. Its primary threat is biophysical—chloride and maturation—and must be measured before hysteresis is attributed. Its metabolic rival survives only if tracer, balance, loss–rescue, and double bypass localize 5α/3α/ISO partitioning and connect metabolite to function.

Validation order matters: stereochemistry → chloride/maturity → catalysis/flux → functional consequence → trajectory/receptor → human compartment → within-person series. Jumping directly to a blood cohort or pharmacological trial would reproduce the existing ambiguity.

For health and longevity, the result is deliberately negative and productive: there is no evidence that increasing or decreasing Allo improves female longevity. The AD association does not establish cause, and preclinical evidence does not show a uniform sign. The bridge can open only if a mechanism precedes a specific function over years and then connects to disease-free years while preserving recovery, organ, sex, stage, and competing risks.

Claims added by this report

• L7-2-REPORT-EN-C1: the final model requires four separate dimensions—flux, trajectory, receptor adaptation, and chloride; omitting any one makes functional sign non-identifiable.
• L7-2-REPORT-EN-C2: current human data favor a peripheral epistemic null, not central equivalence; the decisive test must begin with metrology and a cellular model, not a larger blood cohort.
• L7-2-REPORT-EN-C3: the transcriptomic ranking prioritizes where to measure, but downstream catalysis will be assigned only by labeled product, balance, loss–rescue, and bypass.
• L7-2-REPORT-EN-C4: extension to healthspan and Pharma remains closed until causal proximal function, compartment concordance, and external replication are demonstrated.

20. References

1. Steckelbroeck S, Watzka M, Reichelt R, et al. Characterization of the 5α-reductase–3α-hydroxysteroid dehydrogenase complex in the human brain. J Clin Endocrinol Metab. 2001;86(3). DOI: 10.1210/jcem.86.3.7325. PMID: 11238528.
2. Stoffel-Wagner B, Watzka M, Steckelbroeck S, et al. Expression of 5α-reductase isoenzymes in the human temporal lobe. J Clin Endocrinol Metab. 1998;83(10). DOI: 10.1210/jcem.83.10.5157.
3. Bixo M, Andersson A, Winblad B, Purdy RH, Bäckström T. Progesterone, 5α-pregnane-3,20-dione and 3α-hydroxy-5α-pregnan-20-one in specific regions of the human female brain in different endocrine states. Brain Res. 1997. DOI: 10.1016/S0006-8993(97)00455-1. PMID: 9295207.
4. Genazzani AR, Petraglia F, Bernardi F, et al. Circulating levels of allopregnanolone in humans: gender, age, and endocrine influences. J Clin Endocrinol Metab. 1998;83(6). DOI: 10.1210/jcem.83.6.4905. PMID: 9626145.
5. Hamidovic A, Karapetyan K, Serdarevic F, et al. Higher circulating levels of allopregnanolone follow progesterone across the menstrual cycle. Biomolecules. 2023;13:652. DOI: 10.3390/biom13040652. PMID: 37189398.
6. Hamidovic A, et al. Neuroactive steroid hormone trajectories across the menstrual cycle in PMDD: the PHASE study. Molecular Psychiatry. 2024. DOI: 10.1038/s41380-024-02566-w. PMID: 38664491. ClinicalTrials.gov: NCT03862469.
7. Nguyen TV, et al. Progesterone-derived neurosteroids in women with PMDD under ovarian suppression and hormone addback. Transl Psychiatry. 2017. DOI: 10.1038/tp.2017.146. PMID: 28786978.
8. Martinez PE, Rubinow DR, Nieman LK, et al. 5α-reductase inhibition prevents the luteal increase in plasma allopregnanolone and mitigates symptoms in PMDD. Neuropsychopharmacology. 2016. DOI: 10.1038/npp.2015.246. PMID: 26272051.
9. Schmidt PJ, et al. Ovarian suppression and estradiol/progesterone addback in PMDD. Am J Psychiatry. 2025. DOI: 10.1176/appi.ajp.20240596. PMID: 41030005.
10. Sundström I, et al. Altered response to pregnanolone in women with premenstrual syndrome. 1998. DOI: 10.1159/000054307. PMID: 9508043.
11. Kask K, et al. Allopregnanolone challenge, sedation and startle in PMDD. Pharmacol Biochem Behav. 2009. DOI: 10.1016/j.pbb.2009.02.014.
12. Epperson CN, et al. Cortical GABA across the menstrual cycle in healthy women and PMDD. Arch Gen Psychiatry. 2002;59:851–858. DOI: 10.1001/archpsyc.59.9.851. PMID: 12215085.
13. Smith MJ, et al. Menstrual-cycle modulation of cortical inhibition in healthy women. Neurology. 1999. DOI: 10.1212/WNL.53.9.2069. PMID: 10599783.
14. Smith MJ, et al. Cortical excitability and inhibition by menstrual phase in PMS. Biol Psychiatry. 2003. DOI: 10.1016/S0006-3223(02)01924-8. PMID: 14512217.
15. Bixo M, et al. Treatment of PMDD with the GABA-A modulating steroid antagonist sepranolone. Psychoneuroendocrinology. 2017. DOI: 10.1016/j.psyneuen.2017.02.031. PMID: 28319848.
16. Bäckström T, et al. A randomized, double-blind study on efficacy and safety of sepranolone in PMDD. Psychoneuroendocrinology. 2021. DOI: 10.1016/j.psyneuen.2021.105426. PMID: 34597899.
17. Stiernman L, et al. Transcription of GABAA receptor subunits in circulating monocytes and association to emotional brain function in PMDD. Transl Psychiatry. 2025. DOI: 10.1038/s41398-025-03465-6. PMID: 40701967.
18. Kimball A, et al. Neuroactive steroid levels are elevated in the follicular phase and predict premenstrual depression and anxiety symptom severity in women with menstrually related mood disorder. Arch Womens Ment Health. 2025. DOI: 10.1007/s00737-024-01532-3. PMID: 39601877.
19. Comasco E, et al. Emotion-induced brain activation across the menstrual cycle in PMDD and associations with progesterone-derived neurosteroids. Transl Psychiatry. 2023. DOI: 10.1038/s41398-023-02424-3. PMID: 37055419.
20. Kanes S, et al. Brexanolone for severe postpartum depression: a randomized controlled trial. Lancet. 2017. DOI: 10.1016/S0140-6736(17)31264-3. PMID: 28619476.
21. Deligiannidis KM, et al. Allopregnanolone and peripartum depressive symptoms: individual participant data meta-analysis. Molecular Psychiatry. 2024. DOI: 10.1038/s41380-024-02747-7. PMID: 39511449.
22. Kimball A, et al. Circulating allopregnanolone and progesterone before and after menopause. Psychoneuroendocrinology. 2020. DOI: 10.1016/j.psyneuen.2019.104512.
23. Rasmusson AM, et al. Reduced CSF GABAergic neurosteroids in women with PTSD. Biol Psychiatry. 2006. DOI: 10.1016/j.biopsych.2006.03.026.
24. Pineles SL, et al. Menstrual phase, stress reactivity and progesterone-derived neurosteroids in women with PTSD. Psychoneuroendocrinology. 2018. DOI: 10.1016/j.psyneuen.2018.04.024.
25. Naylor JC, et al. Allopregnanolone levels are reduced in temporal cortex in Alzheimer disease. Biochim Biophys Acta. 2010. DOI: 10.1016/j.bbalip.2010.05.006. PMID: 20488256.
26. Hernandez GD, et al. Phase 1b/2a trial of allopregnanolone in early Alzheimer disease. Alzheimers Dement (N Y). 2020. DOI: 10.1002/trc2.12107.
27. Maguire JL, et al. Ovarian cycle-linked changes in GABAA receptors mediate tonic inhibition and seizure susceptibility. Nat Neurosci. 2005. DOI: 10.1038/nn1469. PMID: 15895085.
28. Smith SS, Gong QH. Neurosteroid withdrawal, α4-containing GABAA receptors and altered inhibition. J Physiol. 2005. DOI: 10.1113/jphysiol.2004.077297. PMID: 15705652.
29. Gong QH, Smith SS. Characterization of neurosteroid effects on hyperpolarizing current at α4β2δ GABAA receptors. Psychopharmacology. 2014. DOI: 10.1007/s00213-014-3538-x. PMID: 24740493.
30. Modgil A, et al. Endogenous and synthetic neuroactive steroids evoke sustained increases in GABAergic inhibition via a PKC-dependent mechanism. Neuropharmacology. 2017. DOI: 10.1016/j.neuropharm.2016.10.010. PMID: 27743930.
31. Ben Mahmoud M, et al. Multifactorial approach is needed to unravel maturation phases of human iPSC-derived neurons. Sci Rep. 2025. DOI: 10.1038/s41598-024-81140-4. PMID: 39837910.
32. Trauger JW, et al. Kinetics of allopregnanolone formation catalyzed by human AKR1C2. Biochemistry. 2002. DOI: 10.1021/bi026109w. PMID: 12416991.
33. Usami N, et al. Substrate specificity of human AKR1C enzymes for neurosteroids. Biol Pharm Bull. 2002. DOI: 10.1248/bpb.25.441. PMID: 11995921.
34. Belyaeva OV, et al. Oxidation and epimerization of 3α-hydroxysteroids by human microsomal SDR enzymes. Endocrinology. 2007. DOI: 10.1210/en.2006-1491. PMID: 17289849.
35. He XY, et al. Intracellular oxidation of allopregnanolone by human brain HSD17B10. Brain Res. 2005. DOI: 10.1016/j.brainres.2005.01.022. PMID: 15804423.
36. Rauschenberger K, et al. A non-enzymatic function of HSD17B10 is required for mitochondrial integrity and cell survival. EMBO Mol Med. 2010. DOI: 10.1002/emmm.200900055. PMID: 20077426.
37. Bengtsson SKS, et al. Continuous allopregnanolone exposure in amyloid mouse models. Curr Alzheimer Res. 2013. DOI: 10.2174/1567205011310010006. PMID: 23157375.
38. Singh C, et al. Allopregnanolone restores neurogenesis and cognitive function in an Alzheimer mouse model in an age- and disease-dependent manner. Neurobiol Aging. 2012. DOI: 10.1016/j.neurobiolaging.2011.06.008. PMID: 21803451.
39. Chen X, et al. A brain cell atlas integrating single-cell transcriptomes across human brain regions. Nat Med. 2024;30:2679–2691. DOI: 10.1038/s41591-024-03150-z.
40. CELLxGENE Census. Release 2025-11-08; manifest and SOMA object. Official release manifest.
41. Gabitto MI, Travaglini KJ, et al. Integrated multimodal cell atlas of Alzheimer’s disease. Nat Neurosci. 2024;27:2366–2383. DOI: 10.1038/s41593-024-01774-5. Source of planned replication, not executed.
42. Mayne G, et al. LC-MS/MS measurement of allopregnanolone and progesterone derivatives/isomers. Anal Bioanal Chem. 2021. DOI: 10.1007/s00216-021-03523-0. PMID: 34279681.
43. Hamzah KA, et al. LC-MS/MS separation of allopregnanolone, pregnanolone, isoallopregnanolone and epimers in serum. Sci Rep. 2024. DOI: 10.1038/s41598-024-78807-3. PMID: 39537691.
44. Kaleta M, et al. UHPLC-MS/MS of progesterone, 5α-DHP, allopregnanolone and related steroids. ACS Chem Neurosci. 2024. DOI: 10.1021/acschemneuro.3c00824. PMID: 38655788.