Estrogen Signalling Gap Filling

Estrogen receptor 1 is among the most heavily assayed targets in ToxCast and the receptor the US endocrine-disruptor screening statute was written around (the evidence for both claims is in Endocrine Disruptor Screening History below). It has almost no presence in the dismech pathograph. This project closes that gap, and this page is the scope and worklist for it.

The measurement behind that claim is the coverage census, regenerated with just estrogen-census. The numbers below move with every curation PR, so read the report for current figures rather than trusting a number quoted here.

What the census found

Six tiers over kb/disorders/, loosest to strictest. When the gap was first measured, 115 entries mentioned estrogen somewhere and 57 built a pathophysiology node around it, but only 12 bound GO:0030520 estrogen receptor signaling pathway, and exactly one put a receptor gene on a pathograph node.

The gap was between those last two: the pathway was recorded and the receptor driving it was not. That is what the ToxCast mapping needs and what this project was mostly about. The pathograph worklist below closed it, and the node-bound count has moved off one. Read the census report for current figures.

Endocrine Disruptor Screening History

The opening claim of this page, that estrogen receptor 1 is among the most heavily assayed ToxCast targets and the receptor the US screening statute was written around, rests on five independent lines of evidence. They are worth keeping separate because they carry different weight. Every identifier below was confirmed against PubMed for identifier and exact title on 2026-09-29.

The screening mandate names estrogen in the statute. The Food Quality Protection Act of 1996 amended the Federal Food, Drug, and Cosmetic Act to require EPA to develop a screening program determining whether substances "may have an effect in humans that is similar to an effect produced by a naturally occurring estrogen" (21 U.S.C. 346a(p)(1), quoted in 80 FR 35350, 2015-06-19). Androgen and thyroid screening entered later, through the advisory committee that designed the program, not through the enabling text. Estrogen is the endocrine axis the regulation was written around.

The receptor has the deepest validated assay stack of any endocrine target. OECD maintains three guidelines for it: TG 455 for receptor transactivation, TG 493 for human recombinant receptor binding, and TG 440 for the rodent uterotrophic bioassay. Two of the three are performance-based, meaning the pathway is characterised well enough that new assays can be validated against a standard rather than adopted individually. TG 440 records that the bioassay "originated in the 1930's and was first standardized for screening by an expert committee in 1962", and its modern international ring trial was published across four papers in Environmental Health Perspectives: phase 1 (PMID:11564613), the programme overview (PMID:12948895), the phase 2 dose-response studies (PMID:12948896) and the phase 2 coded single-dose studies (PMID:12948897). A fifth paper covers the dietary phytoestrogen confound (PMID:12948898).

ToxCast built its first pathway model on this receptor, and it is the only endocrine target where high-throughput data replaced a guideline animal study. Judson and colleagues integrated 18 in vitro assays spanning receptor binding, dimerisation, chromatin binding, transactivation and receptor-dependent proliferation into a single pathway model (PMID:26272952), closing with the observation that the approach "is generalizable to any molecular pathway" — the estrogen receptor was the proving ground. Browne and colleagues then reported that "EPA is accepting ToxCast ER model data for 1812 chemicals as alternatives for EDSP Tier 1 ER binding, ER transactivation, and uterotrophic assays" (PMID:26066997; cite alongside its correction, PMID:28767231). The androgen receptor model that followed used 11 assays (PMID:27933809). Assay-minimisation work ran in the same order, estrogen in 2017 (PMID:28993267) and androgen in 2020 (PMID:32798611).

The modelling consortia ran estrogen first and named the successor after it. CERAPP pooled 17 groups across the US and Europe, 48 models, and screened 32,464 structures against a training set drawn from the ToxCast data (PMID:26908244). CoMPARA applied the same methodology to the androgen receptor four years later (PMID:32074470). No comparable consortium exists for any other endocrine receptor.

The receptor dominates the mechanistic pathway literature. In the AOP-Wiki snapshot of 2026-09-15, 26 of 597 Adverse Outcome Pathways declare an estrogen-receptor-related molecular initiating event, drawn from 8 distinct key events — more than the androgen receptor (16) or the aryl hydrocarbon receptor (23), and comparable to the whole thyroid axis (25) despite that axis spanning seven target proteins. This is cited as evidence of how heavily the receptor has been studied. Mapping those pathways onto dismech entries is not in this project's scope, and is tracked separately alongside the ToxCast work in #12858 and #12682.

The historical reason for all of this is chemical promiscuity rather than regulatory accident. The founding xenoestrogen observations were incidental contamination findings — nonylphenol leaching from modified polystyrene (PMID:1935846) and bisphenol A from autoclaved polycarbonate (PMID:8504731) — and Katzenellenbogen set out the underlying argument the same decade in "The structural pervasiveness of estrogenic activity" (PMID:8593885): the receptor's ligand tolerance is broad enough that structurally unrelated industrial chemicals hit it. Shanle and Xu's review of endocrine disrupting chemicals targeting this signalling pathway is the best single entry point (PMID:21053929).

Two things not to claim

• These AOPs are not the most OECD-endorsed — they are the least. None of the 26 carries WPHA/WNT endorsement; the endorsed endocrine set is androgen receptor, aromatase, thyroperoxidase, deiodinase and sodium-iodide symporter. The strongest estrogen pathways sit at Under Review. "Most studied" and "most formally endorsed" point in opposite directions here.
• The 40-endpoint figure for ESR1 is ours, not the literature's. It is computed from the ToxCast assay annotation table. Present it as a derived count with the invitrodb version stated, never as a citation. No published source ranks ToxCast targets by endpoint count; the citable within-programme comparison is 18 assays in the estrogen model against 11 in the androgen model.

Scope

In scope:

• Binding ESR1 or ESR2 on the pathophysiology nodes that already carry GO:0030520, where the literature supports it.
• Wiring the receptor into the pathograph for entries that bind it in genetic: only.
• Curating estrogen resistance syndrome (MONDO:0014148), the one disease where the receptor sits on a mechanism node by definition. Curated as Estrogen Resistance Syndrome; its stub was added and retired in the same PR.
• Deciding whether estrogen receptor signalling needs its own mechanism module. Decided against; see Module decision below. Every item in scope is now worked.

Not in scope, and deliberately left to the issues that already own them: the estrogen-adjacent findings listed under Related issues below.

The pathograph worklist

Each row is a research task, not a mechanical backfill. A genes: descriptor asserts that this receptor drives this mechanism in this disease, so each one needs a source read at the time of writing and an evidence item quoting it.

Three traps apply throughout:

• A gene binding only has to be self-consistent to validate. just validate-terms checks that a term.label is HGNC's canonical label for the CURIE; it does not check that the gene is the one the node means. ESR1 (hgnc:3467) and ESR2 (hgnc:3468) are paralogues with different tissue distributions, so the wrong one passes every gate while making a false claim. Run just list-gene-term-mismatches on each changed file.
• The direction of the claim has to match. A receptor-activation node and a loss-of-signalling node are opposite claims about the same gene. Evidence for one does not support the other.
• A node that should stay unbound is a finished task, provided notes records the search actually run and what it returned.

Nodes binding the pathway with no gene

All twelve rows are worked. Six took a receptor binding and six are recorded declines; each decline carries its reason and the searches run in the node's notes.

The 46_XX_Gonadal_Dysgenesis row the issue also listed was not gene-less: it already carried genes: PSMC3IP, and that correction is recorded in #13132.

The six declines are exactly the six rows the census still reports as binding the pathway with no receptor gene on any node. That table is a record of decisions taken, not of work outstanding.

Bound in genetic: only

Five entries bound ESR1 without the gene reaching any mechanism node, and all five are now dispositioned.

A genotype-association paper does not by itself support a causal edge, so three of the five stop at a recorded reason rather than an edge. That was the expected outcome for this half of the worklist.

Already on the pathograph

46 XX Gonadal Dysgenesis was the single entry binding a receptor gene on a pathophysiology node, via ESR2, when this project started. It was the pattern the rest of the worklist was trying to reach, and it is no longer alone.

Modules

Six modules mention estrogen or the receptor. None binds either gene. Only sex steroid driven proliferation binds GO:0030520.

Issue #12925 names the first two and deep_placentation_defect. The census finds three more. Two of those, cholestatic liver injury and deregulated nutrient sensing, are incidental mentions and are listed here only so a later reader does not have to re-derive that they were considered. deep placentation defect names ESR1 only inside a reference_title and a quoted snippet, so the module cites a paper about the receptor rather than modelling it.

osteoporosis bone resorption is the one worth attention. It treats estrogen withdrawal as a named upstream driver across several nodes, and it is a non-proliferative bone mechanism, which is exactly the territory the module decision below considers.

Module decision: no estrogen receptor signalling module

The question was whether estrogen receptor signalling needs its own module. sex steroid driven proliferation is the nearest existing module, but it is scoped to proliferation of a hormone-responsive tissue and tagged ONCOLOGY. The non-proliferative mechanisms in the corpus — myocardial signalling in Heart Failure, bone in Osteoporosis, the CNS mechanisms in Premenstrual Dysphoric Disorder — sit outside that scope.

The census adds one fact the question was framed without: bone already has a module home in osteoporosis bone resorption, which models estrogen withdrawal without binding the receptor. Whether that argues for a new module, for widening an existing one, or for neither is a curation call to make after the worklist above has been worked, since that is what shows whether these mechanisms cluster or are one-offs.

The worklist is now worked, so the question is answerable. What it produced: three non-proliferative receptor bindings in different tissues (myocardium in Heart Failure, bone lining cells in Osteoporosis, and the systemic loss-of-signalling state in Estrogen Resistance Syndrome), against four proliferative ones that already have sex_steroid_driven_proliferation as a home. Decided: no estrogen receptor signalling module in dismech. Three tissues and one monogenic receptor disease do not cluster tightly enough to earn one. Below the shared node the three diverge immediately - to profibrotic effectors in Heart_Failure, to RANKL/OPG dysregulation in Osteoporosis, and to six target-tissue arms in Estrogen_Resistance_Syndrome - so what they have in common is a single node and nothing after it. A module here earns its place by saving conforming entries from duplicating a shared chain, and one node is thinner than that. Estrogen_Resistance_Syndrome could not conform in any case, because the existing module's chain is sustained signalling driving mitosis and that disease is loss of signalling.

The receptor-keyed view is real, but its home is elsewhere. dismech organises on the disease, and its modules on the outcome; the receptor is the axis ToxCast assays and AOP initiating events are keyed on. That view is now tracked as EnviroMech issue #3, where the fan-out across tissues is the object of interest rather than a reason the cluster is too thin.

sex_steroid_driven_proliferation#Nuclear Steroid Receptor Hyperactivation accordingly keeps no receptor gene, and its node notes now records why: the receptor is the substitution a conforming entry makes, so naming one would fix the choice the module leaves open.

Related issues

Cross-referenced, not absorbed. Each is its own issue and none overlaps the pathograph work above.

Issue #11705 has a direct consequence for any treatment this project adds: use NCIT:C15986 Pharmacotherapy with a therapeutic_agent, or NCIT:C15445 where the axis is unspecified. Do not reintroduce the menopause-scoped NCIT:C15599 for a non-menopausal indication.

Related project pages

This project cross-references the estrogen mentions already in other project pages and supersedes none of them. Each is tracking something other than estrogen signalling.

• CHRONIC.md carries one table row for Endometriosis whose mechanism column reads "Ectopic endometrium, estrogen dependence". That is a chronic-disease tracking row, not receptor curation.
• GWAS_MECHANISMS.md names Oestradiol once, as one of 28 UK Biobank serum biomarkers. That is the hormone as a GWAS trait, not the receptor. Issue #12925 describes this as an estrogen mention, which it is, but it is the ligand rather than anything this project curates.
• REACTOME_DISEASES.md carries an unchecked box for "estrogen-receptor positive breast cancer (DOID:0060075)". ER Positive Breast Cancer is curated in dismech and is the first entry on this project's worklist. Whether that box is therefore checkable depends on what Reactome coverage means in that project, which is its owner's call, not this one's.

Regenerating the census

just estrogen-census                   # markdown to stdout
just estrogen-census --format tsv      # one row per entry
just estrogen-census --out docs/reports/estrogen-signalling-coverage-census-<date>.md

The script reads only committed YAML. It is offline, report-only, and exits 0.