The Estradiol Story, Re‑Synthesized

The full narrative — molecule, enzymes, routes, hepatic ER, VTE, and the prodrug attempt — rebuilt after fact‑checking, recalibrated against a working v4 PK model, and folded together with the deep dives.

This is the integration of everything — the original story plus what the fact-check passes corrected and what the new quantitative model showed. The narrative is unchanged in its big arc: it follows estradiol from atoms (the C3 phenol and C17 hydroxyl) through enzymes, through hepatic concentration effects, through clinical outcomes, ending at the rational drug-design strategies the field has been attempting. The arc is the same. The numbers, in several specific places, are not. Where they changed, the change is called out explicitly — those are the load-bearing corrections.

1One molecule, two vulnerable positions

Everything about estrogen pharmacology comes down to two atoms on a single molecule. Look at the 17β-estradiol skeleton:

The C3-OH is a phenol. Its pKa is ~10.7, so at physiological pH the molecule is essentially fully protonated and uncharged — passive diffusion across membranes works at every body site (gastric, intestinal, transdermal, sublingual). The aromatic A-ring is also what makes C3 vulnerable to conjugation: SULTs and UGTs grab the phenolic oxygen and attach a sulfate or glucuronic acid group, converting the hormone into a water-soluble inactive form that the receptor doesn't see.

The C17-OH is an aliphatic secondary alcohol on the saturated D-ring. Its chemistry is completely different: there's an α-hydrogen on C17 itself, which lets an oxidoreductase remove that hydrogen and convert the alcohol to a ketone — turning E2 into E1. This is the chemistry HSDs perform. C17 is also accessible to one UGT isoform (UGT2B7) for glucuronidation, but the dominant fate of C17 is redox, not conjugation.

Two atoms, two completely different stories. Understanding which enzymes can act on which atom is the foundation for everything else.

2The enzymes that act on those positions

Four enzyme families do almost all the work. Each one has a clear preferred position and a clear directionality.

The directionality of HSD17B1 vs HSD17B2 is the masterstroke. Same chemical reaction; different tissues; different cofactor pools:

• Liver, gut, endometrium express HSD17B2 with NAD⁺ cofactor. Cytoplasmic [NAD⁺]/[NADH] in fed-state hepatocyte ~700 (Williamson 1967) — strongly drives E2 → E1 (deactivation). This is why a hepatocyte sits at ~5–10:1 E1:E2 at equilibrium, with the higher end of that range being a thermodynamic estimate the old AI built (sound arithmetic, fragile inputs — see §13) rather than a measured number.
• Ovary, breast, target tissues express HSD17B1 with NADPH cofactor. Cytoplasmic [NADPH]/[NADP⁺] ~100 — drives the opposite direction: E1 → E2. Circulating E1 can be reactivated to E2 inside ER-target cells.

Estrogen activity isn't just plasma concentration — it's which tissue you're in. The same E1 in plasma can be reduced to E2 inside a breast cell while being oxidized back to E1 in a hepatocyte. The body uses tissue-specific enzyme expression to create directional gradients, and target tissues actively pull active hormone from the inactive reservoir.

3The oral path: where 1 mg actually goes

Take 1 mg of micronized oral estradiol. Within 24 hours the dose has been transformed into a specific distribution of metabolites. Central estimates synthesized from Kuhl 2005, Stanczyk 2013/2024, in vitro UGT1A10/UGT1A1 activity studies, the new v4 model fits, and clinical PK data:

The clinical numbers this produces are well-established:

• Plasma total E2: 30–50 pg/mL at steady state on 1 mg/d (peak ~8 h post-dose)
• Plasma total E1: 150–300 pg/mL — about 5–7× the E2 level
• Plasma E1S: 2,000–4,000 pg/mL — about 50–100× the E2 level
• Apparent half-life: 13–20 hours (vs the true half-life of free E2, which is only 1–2 hours when measured by IV bolus)

The apparent half-life is the non-obvious feature. The 13–20 h figure isn't how fast free E2 is metabolized — it's how fast the E1S reservoir releases free hormone back to plasma. Without the reservoir, oral E2 would need q2h dosing; with it, once daily is enough.

A note on the two stages of "first-pass". The ~40% gut-wall conjugation shown above and the "UGT1A10 is ~10× more active on E2 than hepatic UGT1A1" claim from §2 are the same fact stated two ways — the gut wall kills a big fraction of the dose because the enzyme there is so catalytically fast. But that alone wouldn't get oral bioavailability down to ~5%; it only gets it down to ~60%. The remaining ~580 μg of free E2 then hits the liver, where another ~90% of that is extracted on first pass (HSD17B2 oxidation, SULT1E1 sulfation, more glucuronidation, biliary export via MRP2). Gut and liver are two compounding filters, and "first-pass" properly means both. Only the product of the two losses gives the ~4% free-E2 bioavailability that ends up in systemic circulation.

What the "gut conjugates" actually are, and where they go

The 400 μg "conj" bucket in the mass balance is a mix of glucuronide and sulfate adducts, in roughly this composition for E2:

• E2-3-glucuronide (the major species; ~60–75% of gut conjugates). UGT1A10 attaches glucuronic acid at the C3 phenol. UGT1A10 has higher activity on this position than any liver isoform.
• E2-17-glucuronide (~20–30%). UGT2B7 attaches glucuronic acid at the C17 secondary alcohol. UGT2B7 is the only UGT that conjugates the C17 position of E2 with appreciable activity.
• E2-3-sulfate (~5–10%). SULT1E1 dominates this contribution at physiological [E2] (Km ~5–20 nM) despite having lower gut abundance than SULT1A1 (Km ~2.4 μM for E2); the higher-affinity enzyme wins at low substrate concentrations. (An earlier draft said the reverse — that was wrong; mechanism fact-check round 2 caught it.)

These conjugates have three possible fates: (a) refluxed back into the gut lumen across the apical enterocyte membrane via MRP2, (b) exported into portal blood across the basolateral membrane via MRP3, or (c) processed by hepatic enzymes once they reach the liver. Conjugates that make it into the colon can be deconjugated by bacterial β-glucuronidase — regenerating free E2 that can then be reabsorbed. This is the enterohepatic recirculation loop, and it's part of why the apparent half-life of oral E2 is longer than the E1S reservoir alone would predict.

Direct primary data on E2 in human gut (and what's missing)

The gut-wall first-pass fraction for E2 is inferred rather than directly measured. The primary data:

• Strassburg et al. 1998, 2000 — characterized UGT1A8 and UGT1A10 expression patterns in human intestinal microsomes; established that UGT1A10 is highly expressed in the gut wall (foundational work on intestinal UGT distribution).
• Basu et al. 2004 (J Biol Chem, PMID 15117964) — the load-bearing reference for "UGT1A10 ~10× more active on E2 than hepatic UGT1A1." Measured E2-glucuronidation kinetics across recombinant human UGTs.
• Cheng et al. 1999 — recombinant UGT1A regiospecificity for catechol estrogens; relevant for catechol-pathway minor metabolites rather than primary E2-3-G kinetics.
• Lépine et al. 2004 (JCEM) — comprehensive UGT activity table covering all 19 human UGTs against E2, E1, and other steroids. The C3 vs C17 regioselectivity comes from this work.
• Itäaho et al. 2008 (Drug Metab Dispos) — review and update of the above, with refined kinetic constants.
• Pichard et al. 1990 — characterized CYP3A4 expression in human gut; relevant for catechol estrogen formation (though minor pathway for E2 first-pass).
• Powers et al. 1985 — compared oral vs IV E2 PK in women, derived total first-pass extraction. Doesn't distinguish gut vs hepatic.

What's not in the literature: no human portal-vein cannulation study has been done on E2 the way Back & Rogers 1982 did for EE. So while the in vitro biochemistry on E2 in gut is well-characterized (and the basis for the 30–50% gut-conjugation estimate), there's no direct in vivo human measurement of the fraction of an oral E2 dose that's conjugated at the gut wall vs the liver. The number is built from the in vitro enzyme activities plus the observed oral/IV bioavailability ratio, not from a portal-vein measurement on humans dosed with E2.

What the three oral routes actually look like over time

Putting sublingual E2, oral E2, and oral EE on the same time-course makes the route-dependence visible. Single-dose curves, post-dose hour on the x-axis:

Primary-data anchor for the EE curve above. Reproduced from Goldzieher 1990 (Figure 1), originally Brody, Turkes & Goldzieher 1989, Contraception 40:269–84. Single-dose 2-tablet (= 70 μg EE) administration to n=24 women. Mean Cmax ~160 pg/mL at 1 h; note the highest-responder peak at ~300 pg/mL — 5× the lowest responder. The mean curve shows the characteristic 4–6 h enterohepatic-recirculation plateau, β-phase decay to ~35 pg/mL by 24 h, and ~10 pg/mL by 48 h. My synthetic curve in the previous figure is roughly Goldzieher mean × (30/70) — i.e., linear dose-scaling, which is approximately right for EE in this dose range (Stanczyk 2013; Goldzieher 1990). The huge highest/lowest-response spread (n=24, same drug, same dose) is the inter-individual variability the model curves do not show — real PK has 30–60% CV on most parameters, and you can see it directly here.

Primary-data anchor for the oral and sublingual E2 curves. From the transfemscience.org sublingual E2 review, aggregating mean curves from Burnier 1981, Casper & Yen 1981, Fiet 1982, Kuhnz 1993, Gausau & Mahler 1993, Price 1997, Wiegratz 2001, Wren 2003, Pickar 2015 — single dose 0.25 to 2 mg micronized estradiol. The dotted black line is the premenopausal-average integrated estradiol level (~50 pg/mL, Verdonk 2019). This is the chart that should drive your intuition, not my smooth schematic above. Two things stand out: (a) the curve shape difference is real and large — oral is broad and modest, sublingual is sharp and high; and (b) the inter-study Cmax variability for 1 mg sublingual spans roughly 100–400+ pg/mL, a factor of ~4. My single 140 pg/mL central curve corresponds to the lower end of this range (it follows Doll 2022's LC-MS/MS measurement); other studies, often older RIA-based, report substantially higher peaks. The "true" central value depends on assay method, individual variability, and how much of the tablet was actually retained sublingually rather than swallowed. My schematic curve preserves the route physiology (biphasic decay, lower late tail than oral) but does not honestly show the Cmax uncertainty — read both figures together.

Where the long tail comes from — and why oral takes longer to peak than sublingual

You asked: is the oral peak late because gut absorption is slow? No — gut absorption is fast for both routes (~30–60 min for oral E2 to cross the enterocyte). The reason oral peaks at 8 h while sublingual peaks at 1–2 h is about where the dose goes, not about how long it takes to be absorbed:

• Sublingual — some uncertain fraction of the dose (older textbook estimates say ~20–25% direct; the v4 PK model elsewhere on this page uses ~5% direct calibrated against Doll 2022's AUC ratio; the only primary anchor is a marmoset study at 10%; human absolute direct sublingual bioavailability is not cleanly measured) absorbs across the sublingual mucosa straight into systemic circulation as free E2. That fraction reaches venous return → heart → systemic in roughly the time it takes a sip of liquid to dissolve. Plasma peaks at 1–2 h and decays fast (free-E2 t½ is only 1–2 h). The rest is swallowed and follows the oral path. Relative AUC of sublingual vs oral E2 at the same dose is ~1.8× (Doll 2022) up to ~2–5× (older studies).
• Oral — the dose goes GI lumen → enterocyte → portal vein → liver. Gut wall conjugates ~40% of E2 to glucuronides via UGT1A10. Of what enters the liver, ~90% is extracted on first pass: HSD17B2 oxidizes most of it to E1, SULT1E1 sulfates E1 to E1S, UGTs glucuronidate. Only ~5% of the dose comes out the other side as free E2. The plasma E2 you see at 8 h is mostly not the directly-absorbed fraction (that already came and went). It's the back-conversion: E1S that target tissues (and the liver in reverse) cleave to E1 via STS, which is then reduced to E2 by HSD17B1.

This is what the apparent half-life is measuring. Free E2 itself has an IV-bolus half-life of only 1–2 h. The 13–20 h "oral apparent half-life" is the decay rate of the E1S reservoir, which is what's continuously back-feeding the free-E2 pool. Specifically:

• E1S plasma half-life: 10–12 h (Ruder 1972). This is the rate-limiting step for clearance of the reservoir.
• For a 1 mg oral dose, ~32% (≈320 μg-equivalent) enters the systemic compartment as E1S. That reservoir then bleeds back over 12–24 h via STS reactivation.
• For a 1 mg sublingual dose, the E1S reservoir is smaller (because only the ~75% swallowed fraction goes through hepatic first-pass; the ~25% directly-absorbed fraction skips it). The sublingual long tail therefore has lower steady-state plasma E2 than oral.

So the apparent half-life difference between oral and sublingual isn't really about absorption rate. It's about how much E1S each route loads into the reservoir per dose. More E1S → flatter, longer-tailed plasma E2 curve. The E1S kinetics (10–12 h half-life, STS reactivation in target tissues) are the same in both cases; only the reservoir size differs.

4All the routes side‑by‑side

Route determines two distinct quantities: (1) how much drug reaches systemic circulation, and (2) how much of it the liver sees during first-pass. These can diverge dramatically.

Several non-obvious patterns:

Oral E2 valerate behaves essentially as oral E2. The valerate ester at C17 is hydrolyzed by esterases in the gut wall and plasma within 30 min, regenerating free E2 plus valeric acid. From that point on, pharmacokinetics are identical to oral E2 with one molecular-weight correction (E2V is ~76% E2 by mass; 1.5 mg E2V ≈ 1.15 mg E2).

The E1:E2 ratio is a route signature. Premenopausal cycling women have a ratio near 1:1 (the ovary makes both, with some peripheral conversion). Transdermal and injectable routes produce ~1:1 because they deliver E2 systemically without portal-vein concentration. Oral routes produce 5:1 or higher because hepatic HSD17B2 runs the NAD⁺-driven equilibrium during first-pass. The E1:E2 ratio is a direct readout of how much hepatic exposure the hormone has had.

Sublingual is the awkward middle. The peak is 4× oral, the ratio is closer to physiologic at peak, and it bypasses the first-pass bolus. So it looks like a clean compromise. The actual evidence (§7) tells a different story.

5Hepatic ER‑α: receptor, threshold, and where the old framing oversimplified

The mechanism connecting all of this to clinical outcomes is hepatic ER-α — the estrogen receptor expressed in hepatocytes. When activated, it transcriptionally upregulates dozens of liver-synthesized proteins, including the coagulation cascade.

The Kd is fine for plotting where the receptor is mostly bound. It's not the right parameter for predicting which routes elevate clotting factors. The right framing is:

• Receptor occupancy tracks free hepatic E2 with a steep curve centered around ~68 pg/mL (Kd).
• SHBG induction tracks free hepatic E2 with a much shallower, higher-threshold curve centered around ~1500 pg/mL (calibrated EC50 in the v4 model) and a slow time constant (τ ≈ 3 days from SHBG protein turnover).
• Coagulation factor induction probably tracks something in between, with protein-specific sensitivity. Protein S is unusually peak-responsive (Bar 2024) — different shape from SHBG.

The qualitative claim — that route, not dose, is what governs hepatic effects — is still right. The quantitative framing in the original (single Kd, single Hill curve, threshold at 68 pg/mL) was too crisp.

What hepatic ER-α does when activated:

The net hemostatic shift is procoagulant. Importantly, this is a local effect of the hepatocyte — plasma protein levels track the concentrations the hepatocyte sees, which depends on route, not on systemic E2.

6Why VTE risk doesn't scale with plasma E2

If hepatic protein induction is the mechanism, why doesn't the procoagulant effect scale linearly with plasma E2? Pregnancy gives the cleanest natural experiment: term plasma E2 is ~250× cycling baseline, yet pregnancy VTE risk is only ~3–10× baseline, not 250×. Several mechanisms compound to attenuate the signal:

The clinical VTE numbers — and the load-bearing fact-check correction here:

7Sublingual: where the model breaks

I want to flag what the original synthesis got wrong, because correcting it sharpens the whole story. The intuitive model was: "sublingual bypasses portal vein, so it bypasses first-pass, so it avoids hepatic effects." That's right for the first step (the PK is genuinely different) but the conclusion (no hepatic effects) doesn't follow. The data:

• Cirrincione 2021 (LGBT Health 8:125): sublingual E2 produces plasma E1 levels much closer to oral than to transdermal/injectable. If sublingual were really liver-sparing, E1 would be at transdermal levels. It isn't — meaning sublingual E2 reaches the hepatocyte and gets oxidized there.
• Pines 1999 and Lim 2019: sublingual produces hepatic protein synthesis changes (SHBG, lipids) similar in magnitude to oral, not similar to transdermal.
• Bar 2024 (ECE conference, abstract EP592): sublingual E2 produces a clinically significant decrease in free protein S — the same pro-thrombotic shift oral E2 produces. The kinetics suggest protein S responds to peak exposure rather than AUC, which would mean sublingual is at least as bad as oral per unit dose, possibly worse.

Why does the intuitive model fail? Because the liver doesn't only see drugs on first-pass. It sees them on every blood circuit. Cardiac output ~5 L/min, hepatic blood flow ~1.5 L/min — the liver sees the entire systemic blood volume every ~4 minutes. For a drug with high hepatic extraction (E2 has ~70% extraction per pass), most of any systemically delivered drug ends up in the liver over a few hours; it just arrives in many small servings instead of one big one. Sublingual peaks lower than oral on each pass but the hepatocyte sees moderately elevated E2 every blood circuit, accumulating substantial integrated exposure.

And critically: hepatic ER responds to integrated exposure for SHBG-like (slow-induction) effects, but possibly to peaks for protein S. So sublingual could be worse than oral on protein-S-driven thrombosis even at lower AUC. Bar 2024 is suggestive of exactly this.

The v4 model captures sublingual as ~5% direct sublingual absorption (Doll 2022 AUC) + ~75% swallowed → normal oral first-pass. That gives an "intermediate" hepatic load — between oral and transdermal, but much closer to oral. The model's sublingual prediction is the only validation scenario sitting outside 2× (×2.18 on E1S), and the literature anchor there is itself uncertain.

8Ethinyl estradiol: the structural outlier (and three SHBG pathways)

Everything above describes E2 and E2-equivalent prodrugs. Ethinyl estradiol breaks the pattern.

EE is E2 with a 17α-ethinyl group (–C≡CH attached to C17). This single substitution changes everything:

• HSD17B2 can't oxidize EE. Oxidation of E2 to E1 requires HSD17B2 to abstract the C17 carbinol hydrogen (the H attached to the same carbon as the 17β-OH). In EE that carbon is quaternary — bearing the 17β-OH, the 17α-ethinyl, and the two ring carbons C16/C13 — with no hydrogen available to abstract. So HSD17B2 cannot make an "E1-equivalent" from EE. The C17-OH is still there (it's a 17β-OH); the oxidation pathway is what's gone.
• EE persists as EE. Without HSD17B2 oxidation, EE is cleared only by sulfation, glucuronidation, and CYP hydroxylation — all slower than the HSD17B2 arm for E2. EE half-life is 18–24 h at steady state vs 13–20 h for oral E2.
• The gut wall does ~44% of EE's first-pass. Back & Rogers 1982 cannulated the portal vein in humans and measured first-pass extraction of EE directly: ~0.44 of the absorbed dose is conjugated in the enterocyte before reaching the portal blood. (This is the load-bearing primary measurement for the "gut-wall first-pass is large" claim — it was made on EE specifically, and shouldn't be uncritically transferred to E2.)
• EE recirculates extensively. Its conjugates (EE-3-sulfate) are reversible via STS; the drug recirculates many times before being permanently cleared. Enterohepatic recirculation is real in humans (large in rats, modest in humans, but not zero).
• Per-mole hepatic SHBG induction is large but endpoint-specific. Kuhl 2005 reports oral EE vs oral E2 on a weight basis, and the apparent disproportion depends on which hepatic protein you measure: ~100× for SHBG, larger for some other hepatic proteins, smaller for others. The "~100×" used as shorthand throughout this page is a modeling shorthand averaged over hepatic-protein endpoints, not a universal measured constant. At the peripheral ER, EE is only ~1.5× as potent as E2 per mole — the disproportion is hepatic, not general.
• EE does not bind SHBG meaningfully. Kd for SHBG is ~50× weaker than E2 (Kuhl 2005). So when EE drives SHBG up via hepatic ER, the rising SHBG does not trap the EE — but it crashes the free fraction of endogenous E2. This is why EE in COCs maintains biological activity while suppressing ovarian E2 to near-castrate levels.

The mass balance for an oral EE dose looks dramatically different from the oral E2 mass balance in §3 — both quantitatively (much higher F) and structurally (no E1 equivalent, smaller sulfate reservoir):

And here's where the new SHBG model adds insight. The v4 model splits SHBG induction into three parallel pathways, each calibrated against a different scenario:

The three pathways add additively (capped at full saturation). The EE pathway is the most potent per-dose because EE's first-pass hepatic concentration during absorption is enormous and the drug persists once inside the hepatocyte.

The empirical SHBG-rise table on combined OCs makes the point quantitatively (Stegeman 2013, Kuhl 2005):

The progestin matters — androgenic progestins (LNG) blunt SHBG rise; antiandrogenic ones (drospirenone, cyproterone) amplify it. Diane-35 (EE + cyproterone) lands at roughly pregnancy-level SHBG. The estrogen choice (EE vs E2-valerate) matters at least as much.

VTE risk per-10,000-woman-years (Stegeman 2013, ASH 2024): no exposure ~1–5 · oral E2 HRT ~9–15 · transdermal E2 ~5 · EE + LNG ~5–7 · EE + drospirenone ~10–12 · E2V + dienogest (Qlaira) ~4–6 · old high-dose EE COCs ~15–20. The progestin matters, but the estrogen choice (EE vs E2-valerate) is at least as important.

9What the fact-checks moved

The original synthesis was built on top of a 700-page conversation that drifted in several specific places. Two independent fact-checking agents (a Claude run and a GPT/codex run, via the codex CLI in a separate tmux session) plus a structured literature scan converged on the same major issues. Where they agreed, I've folded the correction into the narrative above. The mechanically load-bearing ones:

10The v4 quantitative model

The new model is a 5-state (E2, E1, E1S, EE, SHBG) ODE system integrated with scipy LSODA, with 12 calibrated scenarios. It replaces the original AI's JavaScript Euler integrator (which the original conversation admitted was hand-iterated and not actually run). The validation grid below shows model predictions vs literature anchors:

Validation across 12 scenarios. Model output (colored bars) vs literature anchors (gray bars) for E2 / E1 / E1S / EE / SHBG. 10 of 12 scenarios within 2× on all targets. The two outliers: sublingual E1S (×2.18, against a literature anchor that's itself uncertain) and pregnancy SHBG (×0.60, because the model lacks an explicit placental SHBG production term). Generated by v4_with_EE.py → plot_v4.py.

EE vs oral E2: same plasma estrogen, very different SHBG

A 30 μg/d EE dose gives plasma estrogen (EE) similar to what oral 2 mg E2 gives, but SHBG rises ~50% more than with oral E2. Per molar at the hepatic level, EE is ~100× more potent than E2 at SHBG induction. The mechanism is the 17α-ethinyl group blocking the HSD17B2 inactivation step that "uses up" oral E2 on first pass. This is the visualization of the central pharmacology lesson: EE is not 100× more estrogenic; it is 100× more hepatically estrogenic per mole.

Oral dose-response

Model predictions vs Kuhl 2005 / Lindberg 2005 anchors. E2/E1/E1S linear across the 1–4 mg range (the SULT1E1 substrate-inhibition term starts to bite at the upper end, which is why E1S barely keeps pace at 4 mg). SHBG response is sub-linear (Hill saturation through pathway 2 of the three-pathway induction model).

How to actually run the model and play with it

• Interactive viewer — pick a scenario, scrub through time, watch concentrations and metabolic fluxes update live. Green/red live ratio-vs-anchor coding. Self-contained HTML; no server needed.
• Sankey flux diagrams — open any sankey_*.html in ./model/figures/v4/. Steady-state mass flow (μg/day) from sources → plasma species → metabolic sinks. Hover for exact flux values. The oral E2 Sankey makes the "most of the dose enters as already-conjugated estrogens, not as free E2" point visually: the largest flow on the diagram is into E1S, not into free E2.
• Source code — Python in ./model/: v1_basic.py, v2_dynamics.py, v3_calibrated.py, v4_with_EE.py. Each is runnable; each prints its own validation table.
• Writeups — 01_fact_check_summary.md (the headline corrections), 04_v4_extensions.md (what v4 added).

What the model still can't do

• No real tissue compartments. Adipose, breast, brain, uterus all lumped into systemic plasma. Tissue concentrations can't be predicted.
• No portal vein as a real compartment. Oral/SL first-pass amplification is a 50× multiplier on hepatic free E2 during absorption — a coarse heuristic.
• No CYP catechol pathway. 2-OH-E2, 4-OH-E2, 16α-OH-E2 lumped into "other clearance".
• No enterohepatic recirculation. Gut bacterial β-glucuronidase reactivation step is absent. Probably under-predicts oral apparent half-life by some amount.
• No estriol pathway. Pregnancy is lumped placental E2/E1 input; the actual DHEA-S → E3 chain isn't there.
• No inter-individual variability. Single deterministic trajectory. Real PK has CV ~30–60% on most parameters.
• SHBG model is rough. Three-pathway scheme calibrated, but pregnancy SHBG still under-predicts by 40%; would need an explicit placental SHBG production term (some of pregnancy SHBG rise is HCG-driven, not E2-driven).

11The prodrug solution: EC508 and the sulfamate strategy

Now the chemistry from the very beginning of the original conversation makes sense in context. The pharmacology problem is:

We want an oral estradiol formulation that delivers physiological systemic E2 levels without producing the hepatic ER-α saturation that drives VTE risk. Transdermal works but is inconvenient and expensive. Can we design a prodrug that's taken orally but acts pharmacokinetically like transdermal?

The strategy: design a prodrug that's absorbed orally but sequestered into red blood cells as it crosses the gut wall, so it's carried past the liver in the cellular compartment rather than as free drug in plasma. Once in systemic circulation, slow plasma esterase hydrolysis releases active E2 at low concentrations that the liver clears one circuit at a time without saturation.

The mechanism for RBC sequestration: bind tightly to carbonic anhydrase II (CAII), which sits at high concentrations inside erythrocytes. The CAII-binding pharmacophore is a sulfamoyl group (–SO₂NH₂) — the sulfamate nitrogen ligates the active-site zinc of CAII.

The first attempt: E2MATE (estradiol-3-sulfamate). The C3 phenolic oxygen esterified with a sulfamate group, so the molecule both (a) is protected from C3 conjugation by SULT/UGT and (b) binds CAII. It worked for absorption and CAII binding. But the C3-O-SO₂NH₂ bond is also recognized by steroid sulfatase (STS) — and instead of being cleaved cleanly to release E2, the sulfamate group irreversibly inhibited STS via mechanism-based covalent modification: STS uses an active-site formylglycine residue (a modified cysteine) for catalysis, and the sulfamoyl group transfers onto that residue and stays there, blocking the active site permanently. This destroyed STS activity throughout the body, preventing both the drug's own activation to E2 and the normal reactivation of the body's E1S reservoir. E2MATE was repurposed as an STS inhibitor for endometriosis.

The redesign: EC508 = estradiol-17β-(1-(4-(aminosulfonyl)benzoyl)-L-proline). Move the sulfamoyl group off the C3 oxygen — where STS attacks — and onto a separate handle attached at C17 via an esterase-cleavable linker. The estradiol is intact; the sulfamoyl head is held away on a proline spacer; plasma esterases slowly release E2 from the C17 ester bond.

The proline linker is what the original conversation synthesized in the Boc/Cbz protection chemistry section. The Steglich esterification joins L-proline's COOH to estradiol's C17-OH; the 4-sulfamoylbenzoyl group is attached to proline's nitrogen. Boc and Cbz both protect proline's N during the Steglich step, then come off later — Boc by TFA, Cbz by H₂/Pd hydrogenolysis. The synthesis chemistry exists specifically because this molecule is one of the few rational solutions to the hepatic-saturation problem.

The disappointing update: development status EC508 was developed by Evestra Pharmaceuticals. As of 2021, the transfemscience.org EC508 article notes "No Further Development." It never progressed to human clinical trials. The reasons aren't entirely public, but small-pharmaceutical-company funding constraints are the most likely explanation. The chemistry works in principle and in animals; the molecule simply didn't find a developer willing to take it through Phase I–III trials.

12The whole picture in one diagram

13Rebuilt confidence: what we know, what we don't

Closing with explicit confidence levels for each major claim, refreshed after the fact-checks:

High confidence well-established

• The molecular pharmacology of estradiol: structure, IUPAC numbering, the two vulnerable hydroxyls, their pKa and reactivity.
• The major enzyme families and basic kinetics: HSD17B1/2, SULT1E1, UGT1A1/1A10/2B7, CYPs, STS. Tissue distribution, cofactor requirements, substrate specificity.
• Oral E2 pharmacokinetics: ~5% bioavailability for free-E2 AUC; ~13–20 h apparent half-life; dominant E1S reservoir.
• The E1:E2 ratio as a route signature: ~1:1 with transdermal/cycling, 5:1+ with oral.
• Hepatic ER-α Kd ~0.25 nM (Yager 1989). Note: rat hepatocyte; not necessarily the same in human, and not the SHBG-induction EC50.
• ESTHER (Canonico 2007) VTE numbers: oral 4.2×, transdermal 0.9×. Replicated direction (not magnitude) across populations.
• The pregnancy fact pattern: plasma E2 ~250× baseline; SHBG 5–10×; fibrinogen 2×; VTE incidence 3–10× during pregnancy proper.
• The EE structural advantage that creates its disadvantage: 17α-ethinyl blocks HSD17B2 oxidation; ~45% oral bioavailability; ~100× per-mole hepatic SHBG potency (Kuhl 2005).
• SHBG Kd for E2 ~20 nM (Hammond, Avvakumov 2010). The 1 nM number is for DHT, not E2.
• Postmenopausal endogenous E1 production ~40 μg/day (Longcope 1986; Grodin 1973). The 80 μg/day figure used by the old AI was about 2× too high.
• The SULT1E1 ↔ STS sulfation/desulfation cycle as the tissue-switching mechanism; E1S as the inactive currency.

Medium confidence defensible inference

• The quantitative attribution of the 250× → 3–10× pregnancy attenuation cascade. Each mechanism is well-supported; the multiplicative weights are my synthesis, not a single validated model.
• The gut-wall metabolism fraction of ~40% for E2. The 0.44 figure from Back 1982 is for EE in human portal vein, not E2 directly. Inferred to be similar or higher for E2; limited direct human portal-vein data.
• The SHBG induction EC50 of ~1500 pg/mL free hepatic E2 used in the v4 model. This is a calibration target chosen so the model reproduces oral E2 dose-response data (Lindberg 2005), not a directly-measured parameter. The HepG2 dose-response for E2 on SHBG (Mercier-Bodard 1999) suggests responses at 250 nM–2.5 μM — orders of magnitude higher than 1500 pg/mL (~5.5 nM). The model EC50 is best read as "the integrated-exposure threshold at which oral E2 produces measurable SHBG rises in humans," which is a different beast from an in vitro EC50.
• The hepatocyte intracellular [E1]/[E2] ratio of ~5–20 during oral first-pass. Plasma E1/E2 of ~5 is observed; intracellular has never been directly measured.
• The v4 model's portal-vein amplifier of 50× hepatic over systemic. Coarse heuristic; a real PBPK would integrate portal flow × concentration over the absorption time course.
• The "three-pathway SHBG induction" model. Each pathway is fit to its scenario; the additive composition is my framing.

Lower confidence limited or indirect evidence

• VTE risk for sublingual E2 specifically. No outcome studies. Inferred to be similar to oral from indirect markers (E1 levels, SHBG, protein S). The Bar 2024 protein S finding is preliminary (conference abstract).
• The theoretical performance of EC508 in humans. Animal data shows 100% bioavailability and minimal hepatic effects, but no human trials. Mechanism is sound; human PK could differ.
• The in-vivo Hill coefficient for hepatic ER-α → SHBG. No paper has fitted this. n ≈ 1.0 is the default assumption; actual value could be 0.7–1.5.
• The "above SHBG EC50 → hepatic effect" rule in detail. Hepatic estrogen effects are graded, depend on portal vs systemic concentrations, time integration, metabolite contributions (E1 also activates ER), and probably distinct genes' response shapes.
• Whether protein S responds to peak free E2 (Bar 2024) or to integrated AUC. The mechanism is plausible but unvalidated.
• The cytoplasmic [E1]/[E2] = 11 derivation from NAD⁺/NADH × Keq. The arithmetic is correct given the inputs, but both inputs (NAD⁺/NADH = 700; E°' for E1/E2 = −0.265 V) are shaky. The qualitative point (hepatocytes oxidize most E2 to E1) is right; the precise factor of 11 is not.

This re-synthesis integrates: the original synthesis (linked in the meta-strip above), the v4 PK model writeups (fact-check summary, v3 architecture, v3 diff, v4 extensions), the Claude and codex fact-check reports in ./research/, the sublingual correction, the e1s reservoir dynamics, the VTE quantitative model, the EE parameter scan, and the published PK/PBPK model survey. Key references: Kuhl 2005 (Climacteric); Stanczyk 2013, 2024 (Contraception); Yager & Williams 1989 (Cancer Res); ESTHER (Canonico 2007, Circulation); Hammond et al. / Avvakumov 2010 (SHBG affinity); Mercier-Bodard 1999 (HepG2 E2 dose-response on SHBG, replacing the previously-cited Selva & Hammond 2009 which was misattributed); Longcope 1968 (J Clin Invest); Ruder 1972 (E1S kinetics); Doll 2022 (Endocr Pract, sublingual LC-MS/MS); Bar 2024 (ECE EP592, protein S); Cirrincione 2021 (LGBT Health); Stegeman 2013 (JTH, SHBG in COCs); Lorbek 2018 PBPK; Klipping 2012 pop-PK; Wikipedia "Pharmacokinetics of estradiol" and "Ethinylestradiol" articles as well-referenced indices into primary literature; the transfemscience.org EC508 article. All numbers are central estimates; inter-individual variation in real populations is 4–6 fold for most PK parameters and substantial for clinical outcomes. AI-generated, AI-fact-checked, not clinically validated.