Ovarian androgens: testosterone and androstenedione as estrogen precursors during the active cycle

Stage: REPORT_EN
Evidence cutoff: 2026-09-04
Primary population: adult women with active spontaneous ovarian cycles
Target causal unit: woman–cycle–follicle
Overall maturity: H1 for the theca–granulosa topology; H0 for integrated physiological control and translation to healthspan
Document type: mechanistic scientific report; does not constitute diagnosis, prescribing, or clinical advice

1. Executive scientific abstract

Ovarian testosterone (T) and androstenedione (A4) are more than “androgen levels”: they form a C19 carbon pool whose fate is determined within a theca–granulosa–oocyte unit that depends on follicular stage. The theca, stimulated mainly by LH and modulated by the metabolic environment, produces A4 and T. With FSH and maturation, granulosa acquires CYP19A1 capacity to aromatize A4 to estrone (E1) and T to estradiol (E2); HSD17B enzymes redistribute A4/T and E1/E2. However, the linear model “more precursor → more estrogen → better function” is insufficient.

Human evidence supports the dominant theca→granulosa topology and the capacity of both C19 substrates to supply E1/E2 with moderate–high confidence. It does not quantify the physiological fraction of C19 carbon reaching C18 in an intact human follicle, nor demonstrate that normal variation in this fraction controls an independent follicular function. Serum concentrations or ratios such as E2:T mix ovarian and adrenal production, peripheral interconversion, SHBG, phase, and clearance; they are not flux measurements.

The cumulative synthesis produced four corrections:

1. Architecture and stage are the leading competitor. Granulosa mass and identity, follicle selection, FSH/LH, and oocyte state explain much of the covariation among C19, C18, and function.
2. AR and aromatase are not orthogonal branches. Data from 2025 in mice, mouse granulosa, and KGN suggest that AR can cooperate with FOXL2 at CYP19A1 promoter II. The corresponding human evidence—13 women with hyperandrogenic PCOS and 13 controls undergoing IVF—is correlational; the pathway is therefore a priority hypothesis, not an established physiological fact.
3. Direct AR action remains separate. Mouse and human ovarian cortex ex vivo support functional AR capacity, but do not prove that an independent AR signal sustains human antral granulosa when E1/E2 are clamped.
4. The experimental model drifts rapidly. Human granulosa isolated from small antral follicles loses CYP19A1 and HSD17B1 within 6–12 h and acquires luteal features at 24–48 h. A delayed monoculture assay may measure adaptation to culture instead of physiology.

The health connection is conditional and multiscale: local C19→C18 flux would first have to change ER/AR engagement and follicular function; then a repeated systemic waveform; finally, organ-specific exposure and function. Near-total aromatase deficiency and pharmacological inhibition after menopause show that estrogen signaling matters for development and bone, whereas circulating E1, E2, A4, and T are associated with premenopausal breast cancer risk. These fragments point to trade-offs, not a monotonic relationship with longevity. There is no direct evidence that the physiological fraction of ovarian aromatization during the active cycle determines healthspan or survival.

The revised primary hypothesis, L8-5-AR-H1F, proposes C19→AR–FOXL2→CYP19A1 gain feedback. The competing L8-5-AR-H2A attributes covariation to architecture/state. The translational hypothesis L8-5-AR-HT1 remains parked until local causality, reliability, and temporality are demonstrated. The smallest discriminator is sequential: isotope metrology, state qualification, acute AR perturbation with temporal ordering, and testing of a direct AR branch while free C18 is held constant.

2. Scientific question and relevance

Central question

What mechanisms connect ovarian A4 and T, as substrates for E1/E2 during the active cycle, to follicular function and—if the proximal chain survives—to women's health and longevity?

The useful question is not whether a T or A4 concentration is “high” or “low,” but:

• which source produced the C19;
• how much C19 reached granulosa;
• what fraction was aromatized;
• how it was partitioned between E1 and E2;
• what carbon remained available for AR signaling, export, or alternative pathways;
• whether partitioning changed an independent function;
• whether that function subsequently altered an organ outcome.

We define:

• J_C19: absolute A4+T input or production attributable to the thecal compartment within a defined window;
• J_C18: absolute formation per unit time of E1+E2 retaining carbon from the C19 precursor;
• φ_C18: the recovered fraction of traced C19 appearing as C18, with interconversion, conjugation, adsorption, and cell mass made explicit;
• exposure: the free fraction integrated over time, not a single total-hormone measurement;
• function: a follicular output independent of the steroid concentration itself.

Relevance to longevity derives from a possibility, not a result: if robust C19→C18 control contributes to the endocrine waveform over years, it could influence ER/AR-sensitive organs. Yet the same exposure may support bone or vasculature and increase proliferative signaling in hormone-sensitive tissues. “More estrogen” or “more androgen” has no universally favorable meaning.

3. Scope, population, and life stage

The primary estimand concerns adult premenopausal women with functional ovaries, spontaneous cycles, and endogenous follicular activity. The ideal human unit is woman–cycle–follicle, with within-person repetition.

The following are informative but treated separately:

• small preselection antral follicles and dominant pre-LH follicles;
• anovulatory cycles;
• late reproductive stage with intermittent ovulation;
• PCOS/hyperandrogenism, insulin resistance, and metabolic obesity;
• tissue obtained during IVF or stimulation;
• cortical, primary granulosa, organotypic follicle, animal, and cell-line models.

Pregnancy, lactation, established menopause, primary ovarian insufficiency, bilateral oophorectomy, hormonal contraception or therapy, ovulation induction, and major adrenal/gonadal disease are excluded from the primary physiological estimand. These states may serve as perturbations, but do not represent the spontaneous cycle.

Stage is defined by follicular function, independently corroborated ovulation, and STRAW+10, not age alone. Reproductive age modifies architecture and reserve, but does not by itself identify the capacity of the selected follicle.

4. Background knowledge, mechanism, and multiscale map

4.1 Thecal C19 production

The dominant biochemical chain is:

LH → LHCGR → cAMP/PKA → STAR → CYP11A1 → pregnenolone → CYP17A1/HSD3B2 with POR/CYB5A → DHEA/A4/T.

Immunolocalization in 35 premenopausal ovaries and compartmentalized cultures place CYP17A1 and most A4/T production in the theca interna. hCG/cAMP stimulates thecal output; FSH acts predominantly in granulosa. “Predominantly” does not mean absolute exclusivity.

Insulin/IGF, cholesterol availability, perfusion, and cell state may modify J_C19. SHBG alters free T without necessarily changing production. PCOS therefore informs system perturbation, but does not quantify average physiology.

4.2 Transfer and aromatization in granulosa

Granulosa from small follicles produces little E2 without C19 substrate. Exogenous A4 or T enables E1/E2 production, and FSH increases capacity as the follicle matures. The central pathway is:

A4 → CYP19A1 → E1 ↔ HSD17B → E2
T → CYP19A1 → E2 ↔ HSD17B → E1.

Historical human incubations demonstrate capacity, not intact in-vivo mass balance. The apparent preference for A4 over T comes from a small comparison and cannot rank physiological precursors. A4 and T must first be traced separately to exclude interference, and only then competitively.

4.3 Control by FSH, the oocyte, and subcellular organization

FSHR→cAMP/PKA→CREB regulates CYP19A1. Oocyte-derived GDF9 and BMP15 can potentiate human cumulus-cell responses to FSH through SMAD2/3 and IGF. AKAP8/AKAP95 can organize nuclear PKA and pCREB. These layers mean CYP19A1 is not a cellular constant.

However, the public-data reanalysis did not promote the oocyte layer to the first perturbation: in seven paired follicles, the GDF9/BMP15 ligand score barely correlated with aromatization and worsened out-of-sample error when added to a donor+identity model. This does not refute oocyte biology; it shows that, with the available data, stage/identity is a more stable descriptive explanation.

4.4 Alternative DHEA-S–STS input

Human granulosa ex vivo can desulfate DHEA-S:

DHEA-S → transport → STS → DHEA → HSD3B2 → A4 → CYP19A1/HSD17B → E1/E2.

Bonser et al. demonstrated capacity, mainly in luteinized/stimulated granulosa. STS+HSD3B2+CYP19A1 was co-detected in 40/42 antral/preovulatory samples in GSE107746 and in mural regions from two spatially profiled ovaries. This justifies measuring the pathway, not claiming a material contribution. Transport, activity, free-DHEA contamination, and magnitude relative to thecal supply remain unresolved.

4.5 Local C18 fate

Human granulosa-luteal cells can produce E1/E2 and glucuronides from A4. Total C18 formed may therefore differ from free C18 available to ER. UGT expression was weak or diffuse in the analyzed datasets; conjugates must remain in the mass-balance panel, but a UGT perturbation is not a priority.

4.6 AR–ER branching and coupling

Non-aromatized C19 can be converted, exported, or activate AR. Reducing CYP19A1 simultaneously changes two exposures:

CYP19A1↓ → C18/ER↓ and potentially increased residual C19.

AR may also regulate aromatase itself:

C19 → AR–FOXL2 → CYP19A1 promoter II → C18 capacity.

Causality for this arrow is supported in mice, mouse granulosa, and KGN; the available human component is correlational in PCOS-IVF. AR and CYP19A1 must not be treated as orthogonal factors without separating time and clamping C18.

4.7 Stage-dependent change in the control coefficient

Before dominant selection, a follicle needs sufficient granulosa mass/identity, FSH, oocyte signaling, C19, CYP19A1/HSD17B, and bioenergetics to cross a threshold. After selection, it has greater mass and capacity and may tolerate transient perturbations. In the Allaway pilot, a single letrozole dose after selection reduced E2, increased FSH/LH, and did not prevent overt ovulation. This was not an equivalence trial, did not measure φ_C18, and did not manipulate HPO compensation.

4.8 Multiscale map

Architecture/reserve + LH/FSH + metabolism + stage
→ thecal production and DHEA-S input
→ C19 transfer
→ FSH/CREB gain, FOXL2, and oocyte signals
→ CYP19A1/HSD17B
→ free and conjugated E1/E2 and waveform
↔ AR–FOXL2 and direct AR branch
→ follicular function
→ repeated systemic exposure
→ organ-specific engagement
→ organ-specific outcome
→ only ultimately, years free of disease/disability.

Every arrow downstream of follicular function remains open.

5. Evidence method

All verified project artifacts were integrated: scope, evidence map and verification, mechanistic synthesis, computational decision, reproducible computation, hypothesis generation, adversarial review, and experimental design. Primary human studies were prioritized with:

• ovarian vein, follicular fluid, or ovarian tissue;
• tracers, perturbation, or anatomical gradients;
• identifiable phase/stage and donor;
• LC–MS/MS or an explicit method;
• null results and contradictions;
• preservation of the human statistical unit.

Evidence was classified as human in vivo, human ex vivo/in vitro, extreme human genetics, causal animal, cellular, computational, or inferred. RNA, protein, concentration, ratio, and capacity were kept separate from flux and function.

The recent update verified primary sources:

• Tsai et al. 2025, AR–FOXL2–CYP19A1: causality in mouse/KGN models and human PCOS-IVF correlation;
• Jensen et al. 2025, human granulosa drift/luteinization within 6–12 h;
• Sequeira et al. 2025, STS/OATP2B1 in poor IVF response, with sample-size discrepancies and age/reserve confounding;
• Sousa et al. 2026, a human 3D spheroid as feasibility evidence, not demonstrated physiology.

The absence of a complete chain is phrased as “not located in the audited evidence,” not as proof of universal nonexistence.

QA correction record for the previous draft: the spheroid reference incorrectly attributed first authorship to Dadashzadeh. The primary record identifies Maria João Sousa as first author and Arezoo Dadashzadeh as a coauthor. Attribution and the bibliographic title are corrected in both twins; DOI, PMID, results, limitations, and conclusions are preserved.

Bilingual review correction: the project state records failures in mechanism-heading recognition and numerical parity. The required term is made explicit in the corresponding heading of both twins, and the representation of every quantity in the factual manifest is preserved during translation. The scientific interpretation and hypothesis maturity do not change.

6. Evidence map

7. Contradictory evidence and null findings

7.1 Aromatization is necessary, but an acute reduction did not collapse ovulation

Near-total CYP19A1 deficiency causes profound developmental and reproductive abnormalities. By contrast, letrozole after selection did not prevent overt ovulation. Competing explanations:

1. a threshold exists: the follicle had already crossed the gate;
2. the HPO axis increased gonadotropins and compensated;
3. AR/residual C19 sustained some function;
4. the assessed function was too coarse;
5. intrafollicular inhibition was incomplete.

The smallest discriminator is confirmed flux reduction before and after selection, under fixed gonadotropins, with free C18 and AR measured.

7.2 Higher CYP19A1 RNA with age, but unknown flux per cell

Women aged 36–45 years with regular cycles showed follicular/periovulatory E2 similar to women aged 22–34 and approximately threefold higher CYP19A1 mRNA. This may represent compensation per cell, higher FSH, selection of a surviving follicle, or different composition. Without cell counts, protein, and tracer, it is not evidence of higher φ_C18.

7.3 PCOS can show high A4 with preserved or increased aromatase

Granulosa from ovaries with polycystic morphology did not show a simple aromatase failure; some subgroups were hyperresponsive to FSH. The finding contradicts “high androgen = low conversion.” PCOS combines insulin, AMH, reserve, gonadotropins, follicular state, and stimulation, and cannot serve as an isolated C19 experiment.

7.4 Architecture predicts, but the computational analysis is small

The architecture model outperformed the oocyte-augmented model in seven follicles. The augmented model had six coefficients for seven observations, 24 possible permutations, and three donors. The result demonstrates overfitting in that dataset, not causal sufficiency of architecture or flux equivalence.

7.5 AR may sustain aromatase, but human evidence is correlational

The AR–FOXL2–CYP19A1 pathway resolves a design non-orthogonality; it must not become a human conclusion. DHT at 1 μM, KGN, PCOS-like mice, and IVF tissue do not define physiological exposure in the spontaneous cycle.

7.6 Circulating androgens and bone

In SWAN, T, free androgen index, and DHEAS did not significantly predict bone loss as FSH did. This does not refute a local C19 function; it challenges using an annual serum hormone as a sufficient substitute for architecture and flux.

7.7 Systemic signals with opposite directions

Very low estrogen exposure may be adverse for bone; higher circulating E1/E2/A4/T exposures are associated with greater breast cancer risk. The contradiction is addressed through specificity of organ, window, receptor, and source, not by averaging outcomes into a longevity index.

8. Multiscale mechanistic synthesis

Molecular level

CYP19A1 is a necessary chemical gate, but is insufficient to define J_C18. Its output depends on accessible substrate, FSH/PKA/CREB, FOXL2, subcellular organization, HSD17B, cofactors, and product fate. AR may modify both CYP19A1 transcription and parallel functions.

Cellular level

Granulosa operates as a variable C19 sink. Its identity changes with stage and drifts rapidly upon isolation. The oocyte/cumulus may modulate gain; the theca may limit substrate; STS may open an alternative pathway. No isolated cell alone reproduces transfer, matrix, gradients, or feedback.

Follicular level

Dominant selection creates a control transition. Before selection, small coordinated variations may determine whether the threshold is crossed; afterward, mass and reserve may buffer perturbations. The same percentage reduction in flux may therefore have different functional elasticities.

Systemic level

The HPO axis responds to E2/inhibins. A local decrease may increase FSH/LH and partially restore production without correcting the original lesion. Peripheral concentrations also integrate adrenal glands, adipose tissue, liver, kidney, SHBG, and conjugation.

Organ and life-course level

Only a repeated, source-resolved free waveform could mediate organ effects. Bone, vasculature, metabolism, brain, breast, and endometrium require separate mediators and endpoints. Benefit in one organ does not authorize a healthspan gain if another organ incurs a cost.

Causal judgment

The best-supported mechanism today is distributed control by architecture/stage with a regulatable aromatase gate, not dominance of a single androgen. AR–FOXL2 is the most informative molecular perturbation; architecture/state is the causal rival to confront; direct AR action and STS are active alternatives of lower maturity.

9. Computational layer

Three public human datasets were used:

• GSE107746: 80 oocytes and 71 granulosa samples after reconciliation; a five-stage trajectory;
• GSE186504: 113 QC+ cells from 14 follicles/3 donors; a paired core of seven healthy follicles;
• GSE260686: 257 valid regions from two ovaries, including theca, mural granulosa, and cumulus.

Main results:

• in GSE186504, leave-one-follicle-out MSE was 0.0598 for architecture versus 0.1778 after adding oocyte ligand and SMAD/IGF response; relative worsening 197%;
• oocyte-ligand rho with aromatization was −0.071; identity 0.643; FSH–PKA/CREB gain 0.679;
• in GSE107746, the CYP19A1/HSD17B1 module increased with stage, rho 0.803;
• in GSE260686, the thecal module was maximal in theca and aromatization higher in granulosa/cumulus in both donors;
• STS+HSD3B2+CYP19A1 was co-detected in 40/42 antral/preovulatory GC samples;
• UGT2B7/UGT2B15 were detectable but diffuse.

Interpretation: architecture/stage gains experimental priority; the oocyte layer is deferred, not refuted; STS merits a later module; UGT does not merit an initial perturbation. No computational result measures protein, catalysis, receptor signaling, function, or health.

BioNeMo was omitted because a sequence, structure, or embedding model does not solve the small donor count, absent flux, or lack of perturbation. Computation changed ranking, not causal maturity.

10. Primary hypothesis

L8-5-AR-H1F v1 — AR–FOXL2–CYP19A1 gain feedback

Role: revised mechanistic primary.
Status: proposed priority.
Maturity: H1 in mouse/KGN models; H0 for human active-cycle physiology.
Confidence: low–moderate.

Falsifiable statement: in non-tumoral human antral granulosa, with FSH, state, and free C19 fixed, AR activation will first increase AR–FOXL2 engagement and nascent CYP19A1 transcription, followed by protein and J_C18; consequently, the principal component of the functional effect of sustained AR loss will be mediated by lower C18.

Mechanism: C19 activates AR; AR cooperates with FOXL2 at CYP19A1 promoter II; aromatase capacity increases; C18 activates ER and contributes to function.

Predictions:

1. AR/FOXL2 and promoter occupancy change before nascent RNA, protein, and J_C18.
2. Competitive antagonism and rapid AR degradation converge in direction.
3. Chronic AR loss reduces J_C18; downstream C18 rescues the mediated component without restoring AR.
4. Acute AR loss, applied after clamping free E1/E2, leaves a smaller residual functional effect if feedback dominates.
5. The effect is stage-dependent and disappears or attenuates outside antral granulosa.

Evidence for: multimethod causality in mouse/KGN; correlation of AR, CYP19A1, and E1/E2 in human PCOS-IVF; historical plausibility of stage-dependent androgen regulation.

Evidence against: supraphysiological DHT dose, tumor cell line, PCOS-like models, stimulated human samples, absence of loss–rescue in normal human granulosa, and absence of tracer.

Kill criteria: two AR perturbations with target engagement do not change nascent RNA or J_C18 beyond equivalence; temporal ordering fails; the effect appears only in KGN/mouse/PCOS/IVF; or identity, viability, and luteinization explain it.

11. Competing hypothesis

L8-5-AR-H2A v1 — Architecture/state as a common cause

Role: leading competitor.
Status: strengthened as a priority, not demonstrated as sufficient.
Maturity: H1 for links; H0 for the integrated causal thesis.
Confidence: moderate.

Falsifiable statement: maturity, mass, pretreatment identity, selection, and FSH/LH drive will explain most covariation among C19, C18, and function; adding source-resolved J_C18 will not materially improve prediction outside the donor.

Mechanism: reserve/stage determines which follicle is selected and how much competent granulosa exists; the same state increases CYP19A1/E2 and function, generating covariation without donor-level variation in φ_C18 being the dominant control.

Predictions:

1. pretreatment state explains more functional variance than J_C18;
2. at fixed state, donor-level J_C18 shows low repeatability or a small functional association;
3. partial CYP19A1 reduction moves C18/proximal ER without moving function until a threshold;
4. E1/E2 restores ER but not function if the limitation is architectural;
5. serum ratios lose signal after resolving source, phase, SHBG, and clearance.

Evidence for: human maturity/mass data, the bounded letrozole result, Shaw/Wang, transcriptomic pattern and spatial topology, and substantial serum instability.

Evidence against: aromatase is necessary at extremes; equivalence has not been tested with flux and architecture fixed; identity RNA may be a consequence.

Kill criteria: reproducible J_C18 precedes function, adds at least 10% prespecified improvement in out-of-donor error/calibration, responds to two perturbations, is rescued by C18, and replicates in an independent human context.

Associated null rival: L8-5-AR-H2N v1

A physiologically anchored reduction in J_C18 will change product/proximal ER but leave function within equivalence. It is killed if two methods produce functional change outside the margin and C18 provides a specific bypass.

12. Translational hypothesis

L8-5-AR-HT1 v1 — Conditional source-resolved parameter

Role: translational, parked.
Status: not ready for assessment.
Maturity: H0.
HUMAN_QA_REQUIRED: yes.

Falsifiable statement: only if a local mechanism survives and a single source-resolved parameter achieves blinded reliability will that parameter improve external error and calibration for an organ-specific outcome beyond stage, reserve, baseline function, and standard hormones.

The previous “conversion reserve + HPO cost” version is narrowed because “cost” lacks a validated unit. The future parameter will be neither a serum ratio nor a “hormonal age.” It must first:

1. pass ICC/CCC/CV and drift gates;
2. remain stable across cycles and laboratories;
3. precede the outcome;
4. add out-of-sample performance;
5. survive adjustment for phase, AMH/reserve, SHBG, adrenal context, liver/kidney, and baseline function;
6. concern one organ, without combining opposite directions.

It is killed if any of these gates fails or the local mechanism is refuted.

13. Falsifiable predictions and kill criteria

The direct AR branch, L8-5-AR-H1D, remains active but weakened. Its statement is that acute AR loss modifies function before changing CYP19A1 or identity when C18 is clamped. STS reserve, L8-5-AR-H3, remains H1 for capacity/presence and H0 for material contribution.

14. Discriminating experiment

14.1 Q0-MASS: metrology before biology

Qualify [2,3,4-13C3]A4 and [2,3,4-13C3]T separately, subject to experimental confirmation of retention during aromatization. Avoid exchangeable or lost deuterium. Internal standards must have non-overlapping masses.

Measure precursor, E1, E2, HSD17B interconversion, conjugates, free fraction, adsorption, and cell+medium balance. Controls: recombinant CYP19A1, cell-free medium, inactivated cells, spiked matrix, blanks, and carryover.

Provisional gate: closure 85–115%, CV ≤15% within the quantifiable range, and blank/carryover <20% of LLOQ. If it fails, the subsequent result is metrologically uninformative, not a biological refutation.

14.2 Q1-STATE: finding a valid human window

Pilot in eight donors with fresh granulosa at 0, 2, and 6 h. Measure viability/mass, CYP19A1, HSD17B1, FSHR, FOXL2, STAR, CYP11A1, LHCGR, P4:E2, J_C18, and FSH response.

The window qualifies only if markers, flux, and mass remain within 0.80–1.25 of baseline, no luteal increase exceeds MDC, median viability is ≥85% with no analyzed sample <70%, and functional response is retained. If no window exists, move to intact follicles; do not arbitrarily extend culture.

14.3 D0A-ACUTE: AR–FOXL2–CYP19A1

Twenty-four evaluable donors, up to 30 to allow for logistical failures, with vehicle, AR antagonist, and rapid degrader. Traced A4 and T remain in separate aliquots; FSH, density, and free C19 are fixed.

Time points:

• 30–60 min: target engagement, AR translocation, and AR–FOXL2 interaction;
• 2 h: nascent CYP19A1 transcription;
• 6 h: protein and J_C18, within Q1.

Readouts: PLA/co-IP, targeted CUT&RUN/CUT&Tag at promoter II, introns/4sU, protein, and flux. The complete chain, not RNA alone, confirms H1F. Tool discordance triggers an off-target audit; the favorable tool is not selectively chosen.

14.4 D0B-DIRECT: AR branch with C18 clamped

In sister aliquots, free E1/E2 is matched to vehicle before AR perturbation. Primary endpoint: FSH-induced cAMP AUC over 0–60 min per viable mass; pCREB is confirmatory.

The margin is the narrower of a 0.80–1.25 ratio and half the deterioration induced by a functional control above MDC. Equivalence requires the complete CI90 for both tools to fall within the margin. An acute effect does not prove chronic or systemic benefit.

14.5 D0C-DELAYED: only if a 3D model preserves state

A donor-resolved spheroid or organotypic model must preserve CYP19A1/HSD17B1/FSHR/FOXL2, J_C18, P4:E2, viability, and FSH response at 24 h. Only then:

• sustained AR loss;
• J_C18 reduction of 20–30% within an anchored range;
• FSH-stimulated inhibin B per viable mass;
• free C18 add-back.

This adjudicates H2N. A luteinized model is not a negative result; it is an uninterpretable system.

14.6 D1-ARCH and D2-STS

D1 compares pretreatment architecture predictors against architecture+J_C18 using leave-one-donor-out validation. It does not adjust for post-treatment identity because it may be a mediator.

D2 opens only after Q0/Q1/D0: 16 paired donors, high/low thecal supply, DHEA-S-13C, uptake, STS loss through two methods, and DHEA bypass. The contribution is material only if it exceeds the maximum of analytical MDC and the smallest C18 change that moved ER/function in D0C.

14.7 Sample size and analysis

The donor is n; the follicle is nested; the well is a technical replicate. D0A/B starts with 24 evaluable donors and allows blinded variance re-estimation after eight, with a cap of 36. Under conventional assumptions, 24 pairs offer approximately 80% power for a within-donor standardized effect of 0.60, but the confidence interval governs because no reliable prior human effect size exists.

Hierarchical model on the log scale with fixed condition/time effects, donor intercept, and nested follicle. Gatekeeping: target→interface/promoter→nascent RNA→flux→function. Nonsignificance is not equivalence.

15. Biomarkers and stratification

No validated biomarker of ovarian φ_C18 exists. A4, T, E1, E2, SHBG, or their ratios are contextual observables, not flux.

Before any translational candidacy, the following must be resolved:

• ovarian versus adrenal/peripheral source;
• free fraction;
• phase and independent ovulation;
• repeatability across cycles;
• production and clearance;
• reserve/architecture;
• laboratory, batch, and LLOQ;
• out-of-sample performance.

Design stratifiers, not biomarkers:

• follicular diameter and stage;
• pretreatment identity/luteinization;
• FSH/LH, AMH, and inhibins;
• DHEA-S and 11-oxC19 as partial controls of adrenal context;
• SHBG/albumin, liver, and kidney;
• adiposity, insulin, and PCOS;
• stimulation, surgical indication, and ischemia.

A future source-resolved parameter must pass metrology before being associated with outcomes. The biomarker designation remains blocked.

16. Individual variability

Variability may arise at multiple levels:

1. Architecture: reserve, follicle number, selection, and granulosa mass.
2. Stage: preselection, dominant pre-LH, and luteinization have different coefficients.
3. HPO axis: FSH/LH drive and E2/inhibin feedback.
4. Metabolism: insulin/IGF, adiposity, SHBG, hepatic and renal function.
5. Adrenal source: DHEA-S and classical C19 do not necessarily change in synchrony with the ovary.
6. Cell state: FOXL2, FSHR, bioenergetics, atresia, and response to isolation.
7. Method: immunoassay versus LC–MS/MS, total versus free, batch, preanalytical time.
8. Clinical context: IVF, PCOS, surgery, fertility preservation, ovarian disease.
9. Population: ancestry and Mexico/LATAM context require later validation; effect sizes from white/Asian cohorts are not extrapolated.

Heterogeneity must not rescue a null primary result through post-hoc subgroups. Modifiers are prespecified and tested after a valid mechanism and assay.

17. Pharma relevance and maturity

AR–FOXL2/CYP19A1

If H1F replicates in spontaneous-cycle granulosa, the interface offers a target for state-dependent aromatase regulation. It is not yet a therapeutic opportunity: AR is systemic, FOXL2 maintains granulosa identity, and CYP19A1 affects multiple organs. The Pharma priority is a human causal assay and a platform that preserves state, not a compound.

STS

STS becomes a target only if D2 demonstrates uptake, material carbon contribution, loss–rescue, and function. Systemic inhibition would redistribute multiple sulfated steroids; ovarian selectivity and reproductive safety would be major requirements.

Aromatase

Known druggability does not imply relevance to longevity during the active cycle. Aromatase inhibition reduces C18 and reroutes C19; organ effects may oppose each other. Its current use is as a calibrated experimental perturbation.

Asset with the lowest scientific risk

The most mature asset is a donor-resolved C19→C18 flux platform with preserved state, free exposure, mass balance, and a functional readout. It can validate targets and safety without claiming clinical efficacy.

Overall Pharma maturity: H0, target discovery/validation. Advancement requires primary human causality, reversible exposure–response, replication, selectivity, target engagement, identity preservation, and organ-specific assessment. Any clinical transition requires HUMAN_QA_REQUIRED.

18. Limitations

• Much of the mechanistic human evidence comes from historical cultures, IVF, infertility, surgery, or luteinization.
• Some historical sample sizes and donor/follicle units have not been fully recovered.
• Separate cultures demonstrate capacity, not intact transfer.
• Exogenous hormones may exceed physiological free exposure.
• Isolated granulosa drifts within 6–12 h.
• KGN, mice, and PCOS do not transport effect magnitude to healthy cycles.
• DHT is not a pure AR intervention because it can modify aromatase and state.
• AR and FOXL2 cannot be perturbed without monitoring identity.
• RNA/protein do not substitute for flux; neither does an incorrectly positioned tracer.
• Follicles and wells from one donor are not independent human replicates.
• The computational analysis has seven follicles in the main contrast and two or three donors in validations.
• The letrozole pilot was not an equivalence study and did not measure flux.
• Congenital deficiency cases represent lifelong extremes.
• SWAN and postmenopausal inhibitors concern another stage.
• Breast associations do not separate source, receptor, or causality.
• Repeated source-resolved exposure, organ mediation, and survival evidence are absent.
• A net longevity effect cannot be calculated from opposing directions.
• No result authorizes prescribing or individual inference.

19. Conclusions

Ovarian A4 and T connect to female physiology through a partitioning network: they are E1/E2 substrates, ligands or precursors of the AR branch, and nodes subject to sources, architecture, and feedback. The theca–granulosa topology is robust; the physiological control coefficient of each component is not.

The project's central finding is methodological and mechanistic: aromatization and androgen signaling are not independent branches. AR may regulate CYP19A1 through FOXL2 and also exert a direct function; only temporal separation with C18 clamped can distinguish them. At the same time, architecture/state is the leading competitor to any flux-based explanation.

The health connection remains unresolved because the causal chain is incomplete, despite biological plausibility. Estrogen extremes and organ associations do not identify physiological variation in φ_C18. The required trajectory is source→flux→receptor→function→waveform→organ→event. Until it is completed, longevity remains a conditional hypothesis and trade-offs across organs must be retained.

Scientific delta of the report: the problem is reduced to a verifiable discriminator among AR–FOXL2→CYP19A1 feedback, direct AR action, and an architectural common cause, within a human 0–6 h window and using a chemically valid tracer. This replaces both the linear precursor→estrogen→health narrative and the non-orthogonal CYP19A1×AR factorial.

Final adjudicated claims

• L8-5-REPORT-ES-C1 — supported, moderate–high confidence: the dominant human topology is thecal C19→granulosa C18, but no modern intact balance quantifies physiological φ_C18.
• L8-5-REPORT-ES-C2 — supported, high confidence: serum concentrations/ratios identify neither ovarian source nor flux.
• L8-5-REPORT-ES-C3 — supported in models, not established in the human cycle: AR may regulate CYP19A1 through FOXL2; AR and aromatase are not orthogonal perturbations.
• L8-5-REPORT-ES-C4 — supported, high confidence: granulosa drift within 6–12 h invalidates delayed inference unless state preservation is demonstrated.
• L8-5-REPORT-ES-C5 — supported as a priority, not as sufficiency: architecture/stage is the leading causal competitor.
• L8-5-REPORT-ES-C6 — not established: no direct chain connects physiological φ_C18 to healthspan/longevity; any extension must be organ-specific and model competing risks.

20. References

1. Sasano H, et al. Immunolocalization of aromatase, 17α-hydroxylase and side-chain-cleavage cytochromes P-450 in the human ovary. DOI: 10.1530/jrf.0.0850163. PMID: 2644425.
2. McNatty KP, et al. Concentrations of oestrogens and androgens in human ovarian venous plasma and follicular fluid throughout the menstrual cycle. DOI: 10.1677/joe.0.0710077. PMID: 978120.
3. McNatty KP, Baird DT. Relationship between FSH, granulosa-cell number and follicular steroids in human ovaries. DOI: 10.1677/joe.0.0760527. PMID: 632724.
4. Moon YS, et al. Production of estradiol by human granulosa cells and dependence on androgen substrate/FSH. DOI: 10.1210/jcem-47-2-263. PMID: 122400.
5. Hillier SG, et al. Aromatization of androgen by human granulosa cells and dependence on follicular maturity. DOI: 10.1210/jcem-50-4-640. PMID: 6767736.
6. Tsang BK, Armstrong DT, Whitfield JF. Steroid biosynthesis by isolated human ovarian follicular cells in vitro. DOI: 10.1210/jcem-51-6-1407. PMID: 6255006.
7. Moon YS, Duleba AJ. Effects of follicular cells on metabolism of testosterone by human granulosa cells. DOI: 10.1016/0039-128X(82)90066-6. PMID: 7179351.
8. Wouters W, et al. Kinetics and selective inhibition of aromatase in human granulosa cells. DOI: 10.1016/0022-4731(90)90113-7. PMID: 2362449.
9. Mason HD, et al. Aromatase activity in granulosa cells from normal and polycystic ovaries. DOI: 10.1210/jcem.79.5.7962330. PMID: 7962330.
10. Hobeika E, et al. Oocyte-secreted factors synergize with FSH to promote aromatase expression in primary human cumulus cells. DOI: 10.1210/jc.2018-01705. PMID: 30541132.
11. Gu Y, et al. Role of AKAP95 in regulation of CYP19A1 in human ovarian granulosa cells. DOI: 10.1071/RD17313. PMID: 29397057.
12. Bonser J, et al. Human granulosa cells are a site of sulphatase activity and can utilize DHEA-S for oestradiol production. DOI: 10.1677/joe.0.1670465. PMID: 11115773.
13. Prévost J, Bélanger A. Formation of estrogen glucuronides by human granulosa-luteal cells. DOI: 10.1095/biolreprod38.5.975. PMID: 3408786.
14. Abraham GE. Ovarian and adrenal contribution to androgen production measured with constant infusion tracers. DOI: 10.1172/JCI106027. PMID: 5774107.
15. Handelsman DJ, et al. Circulating androgens during the normal menstrual cycle measured by ID-LC-MS/MS. DOI: 10.1016/j.steroids.2012.10.010. PMID: 23127814.
16. Skiba MA, et al. Androgens during the normal menstrual cycle in 588 healthy women. DOI: 10.1210/jc.2019-01357.
17. Charoensri S, et al. Classic and 11-oxygenated androgens in ovarian/peripheral veins and after bilateral salpingo-oophorectomy. DOI: 10.1210/clinem/dgae420. PMID: 38885296.
18. van Winden L, et al. Changes in circulating steroids after risk-reducing salpingo-oophorectomy. DOI: 10.1210/jendso/bvac069. PMID: 35592509.
19. Allaway HCM, et al. Effects of a single 20 mg dose of letrozole on ovarian function after dominant follicle selection. DOI: 10.1186/s13048-017-0303-x. PMID: 28107821.
20. Ito Y, et al. Molecular basis of aromatase deficiency in a 46XX patient. DOI: 10.1073/pnas.90.24.11673. PMID: 8265607.
21. Gagliardi L, et al. Aromatase deficiency associated with a CYP19A1 duplication. DOI: 10.1186/1472-6823-14-16. PMID: 24552606.
22. Horie K, et al. Immunohistochemical localization of androgen receptor in the human ovary. DOI: 10.1093/oxfordjournals.humrep.a137614. PMID: 1577930.
23. Otala M, et al. Effect of testosterone on apoptosis in human ovarian tissue. DOI: 10.1016/j.fertnstert.2004.06.032.
24. Sen A, Hammes SR. Granulosa-cell-specific androgen receptors are critical regulators of ovarian development and function. DOI: 10.1210/me.2010-0006. PMID: 20501640.
25. Prizant H, et al. Androgen actions in granulosa cells involving FSHR and follicular survival. DOI: 10.1073/pnas.1318978111. PMID: 24516121.
26. Wada K, et al. Stage-dependent effects of dihydrotestosterone on estradiol production in human granulosa cells. DOI: 10.1620/tjem.154.253.
27. Tsai Y-R, et al. Hyperandrogenism in polycystic ovary syndrome augments estrogen synthesis through AR–FOXL2-mediated activation of aromatase. DOI: 10.1186/s13048-025-01790-4. PMID: 40898340.
28. Jensen LB, et al. Distinct luteinization profiles of cultured human granulosa cells from small antral and preovulatory follicles. DOI: 10.1210/clinem/dgaf218. PMID: 40186482.
29. Sequeira K, et al. Steroid sulfatase and the transporter of sulfated steroids are upregulated in granulosa cells from women of POSEIDON group 4. DOI: 10.1016/j.steroids.2025.109626. PMID: 40318834.
30. Vazakidou P, et al. The profile of steroid hormones in human fetal and adult ovaries. DOI: 10.1186/s12958-024-01233-7. PMID: 38778396.
31. Sousa MJ, et al. Long-term 3D culture of human cumulus granulosa cell spheroids in PEGylated fibrin: a preclinical model for reproduction and ovarian research. DOI: 10.1186/s13048-026-02079-w. PMID: 41857626.
32. Shaw ND, et al. Estrogen negative feedback and ovarian aromatase in older reproductive-age women. DOI: 10.1210/JC.2015-2191. PMID: 26126208.
33. Wang J, et al. Age and steroid/factor concentrations in small human antral follicular fluid. DOI: 10.1093/humrep/deaf017. PMID: 39922201.
34. Zhang Y, et al. Transcriptome landscape of human folliculogenesis; GSE107746. DOI: 10.1016/j.molcel.2018.10.029. PMID: 30472193.
35. Fan X, et al. Single-cell analysis of paired human follicles; GSE186504. DOI: 10.3390/ijms222111955. PMID: 34769386.
36. Jones RL, et al. Spatial transcriptomics of the human ovary; GSE260686. DOI: 10.1126/sciadv.adm7506. PMID: 38578993.
37. Finkelstein JS, et al. Bone mineral density changes during the menopause transition in SWAN. DOI: 10.1210/jc.2005-1836. PMID: 16403818.
38. Key TJ, et al. Sex hormones and risk of breast cancer in premenopausal women: collaborative reanalysis of seven prospective studies. DOI: 10.1016/S1470-2045(13)70301-2. PMID: 23890780.
39. McCloskey EV, et al. Effects of aromatase inhibitors on bone turnover and lipid metabolism in healthy postmenopausal women. DOI: 10.1016/j.ejca.2007.08.029. PMID: 18029171.