The Estrogen Signal

What the scanner actually counts

A PET image is a map of positron annihilations per unit volume per unit time. Nothing in the physics distinguishes an annihilation from a receptor-bound tracer molecule from one dissolved in myelin, circulating in a capillary, or attached to a breakdown product of the tracer.

The tracer here is FES, 16α-[¹⁸F]fluoro-17β-estradiol: estradiol with fluorine-18 at the 16α position. Fluorine-18's 109.8-minute half-life is what makes a 90-minute dynamic scan practical. The molecule strongly prefers ERα — one binding study reported 94 nM for ERα with negligible ERβ binding . That selectivity is why the tracer is useful, and it is the first boundary on meaning: an FES image is ERα-weighted, blind by design to most of the ERβ and membrane-receptor biology the same tissue is doing.

What the camera records in a region is a time–activity curve: radioactivity concentration, frame by frame, for ninety minutes. That one curve is a sum of at least four physically distinct contributions, and the whole interpretive problem of this chapter lives in the fact that they reach the detector indistinguishable from one another.

First, specific binding: tracer in an estrogen receptor. Second, nondisplaceable uptake: tracer free in tissue water plus tracer partitioned into lipid. FES is lipophilic and this component is large in human brain — the white-matter signal that dominates a clinical FES head image is the textbook case, and it was not reduced by drug in the one human challenge that looked . Third, labelled metabolites: in seven women with arterial sampling, intact parent was 81 ± 7% of plasma radioactivity at five minutes and 14 ± 3% at ninety. By the end of a scan, most radioactivity passing the brain is no longer the tracer. Fourth, tracer still in blood, inside the vessels of the region rather than its tissue.

Every quantitative method applied to these data is a strategy for removing some contributions and tolerating the rest. A metabolite-corrected arterial input removes the blood term and yields a distribution volume. A reference region removes the nondisplaceable term — but only insofar as the reference holds no target and behaves identically in the groups compared. A late static window removes nothing and hopes the early contributions have washed out. The choice is not cosmetic: it decides which alternative explanations remain available afterwards.

The brightest thing in the picture is outside the brain

In every species examined, FES produces one dominant intracranial focus, and it sits in the pituitary — a gland whose blood supply is not protected by the blood–brain barrier. Everything convincingly demonstrated about specific FES binding has been demonstrated there.

In rats, pituitary uptake exceeded every other brain region by 4.4- to 18.9-fold, with hypothalamus a distant second; co-injected estradiol lowered pituitary and hypothalamic uptake and left the rest alone . In rhesus monkeys with arterial input and metabolite correction, estradiol pre-treatment cut pituitary V_T/f_P from 164 to 59 in one animal and 186 to 50 in the other, while twelve other regions moved a few per cent in no consistent direction. In seven women with full arterial kinetics, the only region a drug reduced was, again, the pituitary. Across mice, rats and 31 patients imaged with two estradiol-based tracers, the pituitary was the one reliably discernible brain focus in all three species.

That convergence is the strongest evidence in the field that FES binds estrogen receptors in living tissue. It is also the wrong kind of evidence for the question this report is about. The anterior pituitary is supplied by fenestrated capillaries; molecules reach it without crossing the barrier that protects cerebral tissue. A tracer that labels the pituitary has shown it binds the receptor. It has not shown it can do so on the other side of that barrier, at the receptor densities cerebral tissue actually has .

The counter-argument is well supported and deserves equal weight: delivery is probably not the limiting problem. In rats the influx constant K₁ sat in the ordinary range for brain tracers, which is not what an efflux-limited molecule looks like . In humans, peak whole-brain uptake reached SUV 3.47 ± 0.37 against a pituitary peak of 4.82 ± 0.61; by eighty minutes whole brain had fallen to 0.40 ± 0.04 while the pituitary held 1.17 ± 0.24. The tracer enters the brain freely. It simply does not stay anywhere in particular. The likelier diagnosis is a ratio problem — cerebral receptor concentrations low relative to the tracer's affinity and to how much molecule is lipophilically parked around them — and that calls for a probe with less nonspecific uptake, not a more permeable one .

A week of a degrader is not an occupancy test

The only human pharmacological challenge to brain FES found for this review used a selective estrogen receptor degrader given daily for seven days. That design answers a question about drug action; it does not cleanly answer how many receptors the tracer was bound to.

Ghazanfari 2024 reanalysed brain scans acquired inside a phase-1 trial of elacestrant . Seven healthy postmenopausal women had a 90-minute dynamic scan with arterial sampling and metabolite-corrected input; four were rescanned four hours after the last of seven daily 500 mg doses. Pituitary V_T fell from 2.98 ± 1.30 to 1.65 ± 0.59, and the authors' Lassen plot put the reduction in availability at 62 to 64% . No cerebral region showed a significant reduction. Three features of the design govern how that cerebral null should be read, and none is a criticism of a careful study that remains the best human kinetic dataset for this tracer.

The drug degrades its target. A SERD both competes for the pocket and drives the receptor to be destroyed. After seven days, a reduced signal could mean the pocket was occupied or that there were fewer pockets. The image cannot separate them. Target engagement is the honest label; occupancy is not .

Cerebral exposure was never demonstrated at the target. In the parent trial, the median ratio of elacestrant in cerebrospinal fluid to plasma was 0.126% at 500 mg and 0.205% at 200 mg . Cerebrospinal fluid is a poor proxy for unbound drug at a receptor inside tissue. A blocking study whose blocker may not have reached an effective free concentration cannot distinguish "no receptor here" from "no drug here" .

Several cerebral estimates moved the wrong way. Temporal lobe V_T rose from 0.70 ± 0.29 to 1.18 ± 0.27, cingulate from 0.85 ± 0.31 to 1.28 ± 0.36 and frontal lobe from 0.83 ± 0.29 to 1.39 ± 0.57, all flagged significant in the published table . A competitive block does not raise distribution volumes. Whatever produced those increases, they are a reminder that these estimates are fragile, and that reading the non-significant regions as a clean negative credits the data more than the positive-direction regions allow .

The menopause observation, and the region it is divided by

The finding that motivates the field is real and well executed: in 54 carefully screened midlife women, cerebellum-referenced FES retention rose stepwise from pre- to peri- to postmenopause. The open question is what the denominator of that ratio was doing.

Mosconi 2024 scanned 54 women aged 40 to 65, eighteen per menopause stage, with a brain-dedicated 90-minute dynamic protocol and co-registered MRI, and no arterial sampling . Distribution volume ratios came from Logan analysis against a custom cerebellar region, adjusted for age, plasma estradiol and SHBG — which differed sharply across groups, from 159 ± 29 pg/mL estradiol premenopause to 18 ± 26 postmenopause. The differences survived adjustment: pituitary DVR was 36% higher postmenopause, posterior cingulate and caudate also reached significance, and the regional pattern classified every participant's stage correctly.

Now the denominator. The reference was not simply "the cerebellum": it was built — restricted first to the outermost part of cerebellar crus II on the argument that this tissue expresses little ERα, then refined by supervised clustering to the sub-portion whose uptake was invariant across the menopause classes .

That second step is where the difficulty becomes concrete. The classifier that chose the reference voxels was trained on these images to discriminate the same three groups whose differences are the study's result. Choosing a denominator for its flatness across the contrast you are about to test makes the denominator a function of the result rather than an independent measure of background . The supplement states the procedure openly, so this is a transparency point and not a hidden flaw — but the reference's invariance was selected for, not tested.

Two details compound it. The parametric images retained only voxels with binding potential above zero — defensible for making readable maps, and it also means the voxelwise analysis could only see tissue brighter than the denominator . And one reported region, the hypothalamus, has a DVR below 1 in the pre- and perimenopausal groups (0.955 and 0.984) and only 1.027 postmenopause: no more retention than the reference, in a structure rodent work places near the top of the ERα rank order.

None of this makes the menopause contrast an artifact. The authors' defence is strong on its own terms: nonspecific binding, being independent of receptor expression, should not generate a stepwise gradient across three groups . That argument rests on one assumption — that nonspecific uptake really is invariant across endocrine states. Tissue lipid composition, white-matter structure and plasma protein binding all change across midlife, and FES is a lipophilic molecule carried by SHBG, whose concentration halved across these same groups. The assumption is plausible. It has not been measured .

Publication is not replication

Three human FES papers sit at the centre of this literature. They represent two groups of participants, not three, and the two that share participants also share a denominator.

Nerattini 2025 asked whether a short static acquisition could stand in for the full dynamic analysis. Comparing five twenty-minute SUVR windows against the Logan DVR, the 30–50 minute window agreed best — pituitary intraclass correlation 0.973 — and the paper recommends a 30–60 minute static protocol for clinical use. It is clear about its own scope, limiting the validated interpretation to relative pituitary retention and calling the other regions preliminary.

Its 55 participants are the 54 of the earlier study plus one postmenopausal woman whose laboratory data arrived later . SUVR and DVR are computed from the same scans, over overlapping windows, normalised to the same cerebellar region. Their agreement is close to a mathematical identity: two summaries of one curve, divided by the same number, will track each other. The study calibrates a simplification; it cannot independently confirm what the signal is, because nothing in its design is independent of the study it calibrates against . The same structure appears on the pharmacological side: the seven-participant challenge is a reanalysis of the phase-1 imaging cohort, so counting the two reports as two pieces of evidence double-counts seven people.

The disagreement in print

This is one of the few places in the report where specialists have publicly contradicted each other about what an image means. The exchange repays close reading, because each side is right about something different.

Biegon 2024, a letter co-signed by three senior imaging scientists, argued that the reported regional pattern does not match the known distribution of brain estrogen receptors but does match white matter: hypothalamus, which should rank near the top, came out lowest, while thalamus and caudate — low in receptor, high in adjacent white matter — came out near the top. The letter also held that labels in the published figure point at white-matter structures . Mosconi 2025 replied that those leader lines indicate general anatomy rather than sampled boundaries, that regions were placed on gray matter in a standard atlas, and that the figure displays SUV while the analysis used DVR. On the prior literature the reply is largely correct: the rodent studies cited report low but non-zero uptake outside the pituitary, not absence, and one states explicitly that FES uptake is not limited by the blood–brain barrier.

Both can hold at once, and our reading is that they do. The rank order is a genuine problem for the claim that these values are proportional to regional receptor density. The group gradient is a genuine problem for the claim that the signal is inert background. The reconciling possibility — that something varies with endocrine state which is not cerebral receptor number — is the one neither letter tests . Candidates are easy to list and none was measured in these participants: partial-volume mixing changing with atrophy, white-matter composition, plasma protein binding, metabolite production, or the reference region itself. What the exchange demonstrably did not add was new evidence: no blocking experiment, no arterial input, no test–retest, no partial-volume correction, no independent reference validation .

A second tracer, and what it did and did not settle

4FMFES was designed to fix the specific weakness that limits FES in brain — too much nonspecific signal. It demonstrably improves contrast. Improving contrast and demonstrating cerebral specific binding are different achievements.

Paquette 2020 pooled preclinical and clinical datasets to compare the two tracers head to head . In 31 breast cancer patients scanned with both, pituitary SUV_max was actually lower with 4FMFES (0.73 ± 0.21) than with FES (1.15 ± 0.22), but cortical background was about half, so pituitary-to-cortex contrast was better: 2.61 ± 0.73 against 2.00 ± 0.47. The authors attribute the lower background to roughly 2.5-fold greater resistance to metabolism and almost no plasma-globulin binding. In rats, co-injected estradiol reduced pituitary 4FMFES uptake from 0.45 ± 0.13 to 0.02 ± 0.01 %ID/g — a near-complete block, and the cleanest displacement result in this chapter — while ovariectomy raised rat pituitary FES uptake about 1.7-fold.

Two boundaries on that package. There was no human blocking experiment: the patient scans establish where the tracer goes, not what holds it. And in those 31 patients, pituitary uptake did not differ significantly between the 9 premenopausal and 22 postmenopausal women with either tracer .

That is not a contradiction of the menopause finding: the designs differ in almost every way that matters — an oncology cohort scanned whole-body with the head off-centre, uneven groups, no standardisation to cycle phase, uptake-guided regions, and a frontal-cortex reference. Any of those could hide an effect this size. But it is the only other human dataset we found that examines pituitary uptake by menopause status, and a reader deserves to know it did not see the same thing .

The same paper makes a resolution argument that bears on every cerebral FES result. Clinical PET resolves about 4.8 mm, roughly 111 µL; the human pituitary is 460 to 520 µL, only four to five times that. The small-animal scanner used here reached 0.74 mm, about 0.4 µL, and with it the authors separated a second focus anterior to the pituitary in rodents plus three further foci in dissected brain — structures that do not exist as separable objects in a human scan . The regions where estrogen receptors are densest are, in humans, at or below the size a clinical scanner can hold apart from their neighbours.

The clinical-development thread is thin but real. A registered study, MOSAIC at the Université de Sherbrooke — the group that developed 4FMFES — proposes to combine the tracer with carbon-11 acetoacetate and FDG in 45 participants across the menopausal transition. Checked directly on 2 October 2026, the record still reads "not yet recruiting", was last updated in June 2025, and gives an estimated start of July 2025 that has passed . That is a registry record, not recruitment and not results. An earlier attempt to validate brain FES quantification against arterial input was terminated with three participants enrolled and nothing posted — a recruitment outcome, not a technical one.

Engineered receptors are a bridge to a different question

Two studies show FES imaging an estrogen receptor in the brain with excellent contrast. In both, the receptor was put there on purpose.

ChRERα fuses a light-activated ion channel to the ligand-binding domain of ERα, delivered by viral vector. The pocket is preserved, so FES binds it; the channel component carries it to the cell membrane, at densities and in places no native tissue provides. Bonaventura 2023 tracked it in rats and two squirrel monkeys for up to 80 weeks, injection-site binding potential rising from 0.12 to 0.47 and a connected parietal site from 0.1 to 1.1, with the resulting anatomical map overlapping a functional connectivity map from a separate cohort . Li 2024 took the system to three rhesus monkeys, where injection-site uptake rose 70% and 86% in the two animals that expressed the construct and — the more interesting result — correctly identified the third, in which transduction had failed, before any behavioural experiment was wasted on it .

This belongs in the chapter rather than a methods appendix because of what the same papers say when they are not imaging the reporter. Li 2024's native baseline phase is the best-controlled estrogen-receptor blocking study in a primate brain that we found: full kinetic modelling, metabolite-corrected arterial input, estradiol pre-treatment. Pituitary V_T/f_P dropped by about two-thirds; the cerebellum and eleven other regions did not move. The authors concluded that endogenous expression in monkey brain is negligible enough for the cerebellum to serve as a reference . Read from the other direction, that is a negative result about native cerebral FES binding in a primate, obtained with better methods than any human study has used.

Three qualifications keep it from being decisive. The blocking dose was deliberately conservative — 0.1 mg/kg estradiol, a third of the rodent dose, chosen for safety. The sample is two animals. And "no detectable specific binding at this sensitivity" is not "no receptors": the same paper notes that the comparison reporter produced better signal-to-background despite binding its target about fifteen times more weakly, so affinity was not the limiting quantity . The rhesus data were deposited publicly and inventoried in September 2026 as eight files, about 2.5 GB, under a CC BY 4.0 licence; the archive was not downloaded, so its contents are unverified .

Which observation would do the most work

The useful question is not whether to believe brain FES images. It is which single piece of evidence, obtainable at reasonable cost, would most sharply separate specific cerebral binding from the alternatives that currently explain the same data.

Four explanations are on the table for the menopause contrast, and they are not equally hard to test. Roughly in order of what it would take to exclude each one:

The denominator moved. Cheapest to probe and the most consequential, because it needs no new participant. The published group means can be turned around arithmetically: for each region, how much would the reference have to change, with the target held exactly constant, to produce the difference that was reported ?

Partial-volume mixing changed. Gray matter thins across midlife, so a fixed region samples a different gray-to-white mixture in an older brain, and white-matter FES uptake is high and non-specific. Testable on the existing images: reconstruct with resolution modelling, or correct using the MRI segmentations already acquired, and report the sensitivity .

The input changed. Plasma protein binding, metabolism and free fraction plausibly differ across endocrine states, and SHBG — which protects FES from rapid metabolism — halved across these groups. A reference-tissue method is blind to this by construction. Testing it needs blood, not more scanning: parent fraction and free fraction in a subset would bound the effect .

Cerebral receptor availability changed. The explanation the field would most like to be true, and the only one needing a displacement experiment. The design constraints are now clear: a competitor whose free concentration at the target is demonstrated rather than assumed, given acutely so competition is not confounded with degradation, in enough participants to detect a small effect against a large nondisplaceable background, read against an arterial input rather than a reference chosen for flatness .

The order is deliberate. The first three can be attempted on data that already exist; only the fourth needs a new scan, and it is also the one most likely to fail for reasons unrelated to biology. The figure below holds the first explanation in your hands — three quantities in arbitrary units, no fitted values, no study's data — because that algebra is what this chapter has been circling.

A last point about sequencing, which is really about cost. Everything solid in this chapter was obtained in the pituitary, and the pituitary is not a consolation prize: it is hormone-responsive, receptor-dense, it carries the strongest menopause signal, and it is the one place a human FES measurement has a demonstrated molecular meaning. A study built around it as a positive control, with cerebral regions as the exploratory arm, is more defensible than one that treats a cortical DVR as a receptor density from the start . The field's own diagnosis points the same way: a perspective from this literature argued for brain-first estrogen-receptor imaging and for a subtype-selective, brain-penetrant ligand that has not yet been demonstrated — a chemical limit rather than a conceptual one. The next chapter steps one layer upstream, to the enzyme that makes the hormone.

Upstream of the receptor is a different measurement problem

Chapter 1 asked what a receptor tracer is bound to. This chapter asks something a receptor tracer cannot answer: where the hormone is made, where it is destroyed, and how fast.

Aromatase is the enzyme that makes estrogens. It converts C19 androgens into C18 estrogens by aromatising the A ring of the steroid: androstenedione becomes estrone, testosterone becomes 17β-estradiol. It is a cytochrome P450 anchored in the membrane of the endoplasmic reticulum, and it carries out the final step of estrogen synthesis — which is why it is also the target of the aromatase inhibitors used in oncology, and why a labelled inhibitor was available to turn into a tracer in the first place.

In the human brain, the enzyme is concentrated in limbic structures. Imaging and post-mortem work place the highest levels in the thalamus, with clear signal in the amygdala, the hypothalamic and preoptic region and the medulla, and low but detectable amounts in white matter, putamen and cerebellum . That regional pattern is species-specific in a way that matters for reading the literature: in rodents the highest expression is in the medial amygdala and the bed nucleus of the stria terminalis, in non-human primates the amygdala leads, and in humans the thalamus does . A rodent finding and a human finding about "the aromatase-rich region" are not findings about the same place.

That promise is the reason to be careful. What a tracer reports is the number of binding sites it can reach under the conditions of the scan. An enzyme has at least three quantities attached to it, and they come apart: how much enzyme is present, how much of it a ligand can occupy, and how fast it is turning substrate into product. The first two are stocks, the third is a rate. A catalyst's abundance sets a ceiling on the rate; it does not set the rate. Throughput also depends on how much substrate arrives, on the supply of reducing equivalents the P450 reaction needs, on whatever else is competing for the active site, and on what happens to the product once it is made.

Immenschuh 2024, the thesis that collects most of the human cetrozole work, states the separation in its own methods discussion: availability measured by a radiotracer, catalytic activity measured by the rate at which labelled water is released as an androgen is converted, protein measured by immunohistochemistry, and messenger RNA measured by hybridisation each answer a different question, at a different spatial resolution . None of them substitutes for another. Imaging adds a fifth property that the bench assays do not have and cannot avoid: it is done through the blood, in a living person, in about an hour.

What the human cetrozole work established, exactly

One tracer, one research programme and one group of ten women carry the human evidence. What that work established is a defensible way to quantify binding, not a reading of synthesis.

The tracer is [¹¹C]cetrozole, a non-steroidal aromatase inhibitor labelled with carbon-11. It is described as binding a single site on the enzyme with higher affinity, better signal-to-noise and better metabolic stability than the earlier [¹¹C]vorozole — a characterisation this review takes from the studies that cite Takahashi 2014 rather than from the original, so the assay conditions behind it were not checked. Carbon-11 decays with a half-life of 20.4 minutes, which fixes much of the design: the tracer has to be made next to the scanner, and the whole measurement has to be finished inside about ninety minutes.

Jonasson 2020 is the quantification study. Thirteen women were recruited and three were excluded for incomplete data or dropout, leaving ten analysed participants who contributed thirty scans across a baseline and two challenge sessions, eight of them with arterial sampling . A two-tissue compartment model fitted the data well, but the individual rate constants that would describe binding directly were unstable, which is a common and honest outcome and the reason the field reaches for reference-tissue models. The authors compared those simpler models against the compartmental result, concluded that a sixty-minute acquisition read with a simplified reference-tissue model was adequate, and showed that a static uptake ratio overestimated binding relative to it .

Two things follow that the study itself is careful about and a retelling tends to lose. First, the reference models were judged against an outcome that was itself normalised to the cerebellum . Agreement between two readings that share a denominator is internal consistency; it cannot show that the denominator is free of specific binding, and it cannot show that the denominator stays put when something changes. Second, agreement between a truncated and a full acquisition is not a test–retest experiment. No test–retest figure for this tracer — the same people scanned twice with nothing changed — appears in the sources behind this chapter, and that figure is the yardstick any later endocrine comparison would have to be measured against .

A challenge that moves the signal, and what it can carry

Nicotine lowered thalamic binding in eight of nine women. That is a real perturbation result, and it is routinely asked to carry more than its design can bear.

Dubol 2023 gave ten healthy non-smoking women, aged 22 to 33 with regular cycles, two ninety-minute dynamic scans one menstrual cycle apart, with 1 mg of nicotine nasal spray given five minutes before the second . The estimated dose reaching the circulation was about 0.007 mg/kg — chosen to resemble a cigarette, and for ethical reasons not exceeded. Three different scanners were used with reconstruction settings matched to a resolution of roughly 6 mm, and binding was read as a cerebellum-referenced Logan outcome in three prespecified regions.

The result: a region-by-session interaction, driven by the thalamus, where binding fell in eight of the nine participants included in that comparison, by between 2 and 36 per cent, with a within-person effect size of −0.99 . The nominal p value was 0.021, which became q = 0.063 under the authors' stated false-discovery threshold of q < 0.1 — it meets the threshold the authors declared, and it would not meet the more common one. Amygdala and hypothalamus were null. The dose-response hint, a negative association between salivary cotinine and thalamic binding, involved seven participants and did not reach significance .

The authors' reading is direct enzymatic blockade, and it is a reasonable one: nicotine and its metabolites inhibit aromatase in cell preparations, an effect relieved by an excess of the enzyme's own substrate (Barbieri 1986) , and an earlier primate study with the predecessor tracer reported the same direction, in the region where that species carries most of the enzyme (Biegon 2010) . Convergence across species and across two tracers is genuinely encouraging for the claim that the signal is aromatase-related.

It is a different claim from the one the chapter is about. A fall in binding is consistent with occupancy of the site by something else; it is not a measurement of a change in estradiol production. The paper says so. What is harder to keep in view is the second interpretation, and it comes from the study's own supplement.

In the one participant who had arterial sampling in both sessions, tracer volume of distribution fell after nicotine in every region examined — including the cerebellum, the tissue used as the reference . The authors read this as evidence that the reference-based effect is an underestimate, because a reference that falls alongside the target hides part of the target's change inside the ratio. That reading is sound. It is also exactly the observation a moving denominator would produce, and with one participant neither reading is established . This is the single most load-bearing gap in the chapter: the supplementary material that contains it was cited by the article and could not be obtained here, so it stays a named gate rather than a resolved question .

One more source-level detail stays visible rather than being tidied away: the article describes nicotine as the second session for everyone, while the thesis figure describing the same protocol says the order of the two sessions was randomised . For a within-person design with one cycle between visits, session order is not a cosmetic detail, and the two descriptions cannot both be complete.

Four questions about one enzyme, and no ladder between them

Expression, protein, reachable sites and catalytic rate are four different experiments. Each is informative. None of them delivers the next, and a study that passes one has not passed the others.

The expression work behind the first column is in rats, and it is unusually specific: aromatase transcript is sex-, region- and cell-type-dependent, found mostly in inhibitory neurons, with higher expression in males across several limbic regions . That is a result about cells, in a species whose aromatase geography differs from ours. It tells us what kind of cell the human signal might be coming from; it does not tell us that the human signal comes from there.

The clearest illustration of the gap between columns is the hippocampus. Aromatase protein has been reported there in human tissue, and yet neither of the two human PET studies in this programme could measure reliable availability in that region: the signal sat below the threshold they treated as usable . Nothing is contradictory about that. A tracer that cannot resolve a low-density region is reporting its own detection limit. The error to avoid is the easy one: a failure to quantify is not a demonstration of absence, in this region or any other.

The third column, availability, is also where the multimodal arm of the same programme sits. In twenty-one women, aromatase availability in the thalamus, amygdala and hypothalamus correlated with regional grey-matter volume, and inversely with the volume and thickness of several prefrontal subregions; hypothalamic availability was related to the volume and microstructure of the fornix . Those correlations are interesting and they are not an endocrine result: no relation was found between availability and menstrual cycle phase, or between availability and circulating hormone levels. That study was an unpublished manuscript at the time of the thesis, and this review could not establish its current publication status .

A peripheral negative result, read as the four experiments it was

The endometriosis study is the most useful negative result in this chapter — and only while its four separate experiments stay separate.

Zhang 2026 set out to find a PET tracer for endometriotic lesions, and had three candidates for three different features of the disease: aromatase, which is raised in lesions and nearly absent from healthy endometrium; neutrophil elastase, as a marker of the inflammatory infiltrate; and a receptor marking the fibrotic process. Ten biopsies from nine patients were sectioned and exposed to the tracers with and without a blocking agent, alongside histology on nearby sections. One tracer was then taken into a small imaging study in patients.

The aromatase row is the one this chapter needs, and it is a careful negative. Binding was visible on the sections, but it did not fall when the site was blocked — and the test was run only in the first, four-patient stage, with a handful of blocking conditions. The authors say plainly that visible uptake without blockable binding cannot be called specific aromatase binding. What makes the result hard to interpret further is the missing step: the confirmatory stain for aromatase was not performed, so there is no independent evidence about whether the target was present in the sections where the blocking failed . Staining performed for other targets does not substitute.

Four gates are visible in this one paper, and they fail independently: is the target plausibly there (expression and pathology say yes for aromatase in lesions); is it confirmed in this tissue (not tested); does the tracer bind it displaceably (not shown here); and does the tracer reach the tissue in a living person (not shown, for a different tracer, in three people). A failure at one gate says nothing about the others .

Three sentences are therefore unavailable, and all three are easy to write by accident. That the aromatase tracer failed in patients: it was never given to a patient in this study. That brain aromatase imaging is in doubt because a peripheral assay was negative: lesion tissue and brain tissue are not interchangeable, and an autoradiography assay is not a scan. And that the negative imaging pilot tells us anything about aromatase: it used a different tracer against a different target . The authors are explicit about the structure of their own study; it is the summary of it that tends to collapse.

What the study does establish is worth stating positively, because it is the template this chapter recommends. Two of the three tracers showed specific, blockable binding with histological confirmation on adjacent material. For one of them, that in-vitro success was followed by an in-vivo null with internal controls behaving correctly, which is exactly the configuration that isolates delivery as the next thing to measure — and the authors propose doing so with a perfusion tracer rather than asserting it .

Inactivation is the other half, and here it is a hypothesis

Estrogen is also switched off locally, by sulfonation. The imaging evidence in this branch is computational, it is early, and it points at a reference region rather than at estrogen.

Sulfotransferases attach a sulfonate group to hormones, neurotransmitters and xenobiotics, which is one of the ways a tissue takes a molecule out of play. One member of the family, SULT1E1, is the estrogen-specific one, and it is the reason this branch belongs in a chapter about local estrogen handling: the enzyme that makes estradiol and the enzyme that inactivates it together set what a region is exposed to, independently of what is in the blood .

The route into imaging is unexpected. Earlier chemistry work identified this estrogen-inactivating enzyme as a binding partner of the amyloid tracer scaffold built on thioflavin — Cole 2010, which this review could reach only as indexed excerpts, an access gap in this work rather than a statement about the article. Miccoli 2026 asked whether the fluorinated derivatives in clinical use share that liability, using public expression data, molecular docking and 300-nanosecond simulations against deposited and predicted enzyme structures.

Two of its findings matter here. The first is a correction to the premise: in the transcript datasets analysed, the estrogen-specific isoform was negligibly expressed in the adult human cortex, while three other family members were abundant — the opposite of what the earlier work reported, which the authors attribute partly to antibodies that may not distinguish between closely related family members . The second is where those three abundant isoforms sit. They are expressed in the cerebellum: the region used as the reference for amyloid imaging .

The authors are careful about what this is, and the carefulness should survive the retelling. Simulations assess whether a pose is physically plausible and stable; they do not establish that a molecule gets into a cell at a sufficient concentration, that it competes successfully with whatever is already in the pocket, or that any of it registers in an image. Transcript abundance is not protein abundance, protein abundance is not occupancy, and occupancy is not signal. The paper states explicitly that it cannot determine whether these interactions affect reference-region activity or the scales derived from it .

The supporting material for this branch is thin in a way worth marking rather than hiding. The chemistry work that anchors the whole branch was available here only in indexed form. A human study that followed it, Surmak 2020, is known to this review as a citation inside the later paper and nothing more; no outcome from it is used here . The computational work's own deposited data — a single processed project file of binding energies and trajectory deviations, under a permissive licence — was inventoried but not opened or reanalysed .

Which quantity would an endocrine contrast move?

Before designing a study of ovarian state and local synthesis, it is worth saying out loud which of the three quantities the design moves, and which of them the scanner can see.

The evidence so far is quietly consistent, and it points the wrong way for an easy study. Across the cycle, Biegon 2015 reported that ovarian tracer uptake varied while no corresponding brain difference was detected . In the twenty-one-woman multimodal study, availability was unrelated to cycle phase and to circulating hormones . And in Dubol 2026, a later challenge with the same tracer, one week of transdermal testosterone — an exposure that raises the enzyme's own substrate — produced no statistically significant corrected change in binding in ten women .

Three negatives of that kind invite one of two readings, and the honest position is that both remain open. Either the enzyme pool is genuinely not moved much by ordinary endocrine variation on these timescales — in which case aromatase availability is a trait-like measure rather than a state measure, which would be useful to know and would change what it is good for — or the measurement is not sensitive to what did change. None of these studies performed an equivalence analysis, and none had the sample size to support one, so "no detected difference" is as far as they go .

The interactive below is a toy, and it exists to make one relationship concrete: the levers that move an enzyme's output and the levers that move its tracer signal are not the same levers, and one of them moves them in opposite directions.

The uncomfortable case is the last of those three. An endocrine manipulation that changes how much androgen is available to the enzyme changes its output directly, and may change the tracer signal in the opposite direction, because the tracer and the substrate are competing for the same pocket . A design that measures only binding cannot tell that case apart from a change in the amount of enzyme. This is the same structure as the competition problem in chapter 1, with the roles reversed: there, the endogenous hormone competes with the tracer at a receptor; here, the endogenous substrate competes with the tracer at an enzyme.

What would make the next measurement decisive

Four observations would separate the readings this chapter has kept apart. None of them requires a new tracer, and two of them are already half-done.

Reference behaviour under the challenge, in more than one person. The single arterial dataset inside the nicotine study is the cheapest discriminator in the chapter, and it is also the one piece of evidence that would decide between "the effect is underestimated" and "the denominator moved" . Obtaining the supplement is a source-level task; acquiring arterial or image-derived input in a handful of participants is a small methodological study, not a new programme.

A test–retest in the relevant population. Without a figure for how much the measure moves on its own, an endocrine group difference has no yardstick, and a null has no interpretation . This is the step that would turn the three negatives above into either a useful trait finding or a statement about sensitivity.

An orthogonal handle on synthesis, measured at the same time. The whole argument of this chapter is that binding and production are different quantities; the way to stop arguing about it is to measure both. Substrate and product concentrations in blood, a tissue activity assay where tissue is available, or an exposure designed to move the substrate rather than the enzyme would each give the binding measurement something to be checked against .

Target confirmation in whatever tissue carries the assay. The endometriosis work makes the case by omission: a blocking failure with no confirmation of the target is an uninterpretable result, where the same experiment with a stain attached would have been a clean one .

There is also a question that stays entirely inside what already exists, which makes it a plausible opening for a collaboration rather than a proposal for a new study: whether the existing cetrozole sessions can be re-read with the reference region treated as an unknown instead of a constant. That is a modelling question asked of data that are already collected. It is not an open-data question — the participants in this series did not consent to public sharing of their data, which the methods paper states — so it would have to be asked of the investigators who hold them, and framed as a methods collaboration rather than a request for a dataset .

There is no single “low versus high estrogen” contrast

Seven familiar endocrine states and manipulations change different hormones over different timescales. Lining them up on one low-to-high axis hides what each design can estimate.

Natural cycles move estradiol and progesterone together over weeks; a menopausal transition stretches irregular cycling over years. Bilateral ovariectomy removes the ovaries in one operation, so the loss is abrupt by construction. Chemical follicular depletion with VCD leaves the ovaries in place and empties the follicle reserve over months: in one rat study, cycles stayed regular in the cohorts examined at one and two months and became irregular only at three .

Pharmacological designs perturb an axis rather than one hormone. A gonadotropin-releasing-hormone agonist implant produced what Frokjaer 2015 describe as a biphasic fluctuation, stimulation followed by suppression . A combined oral contraceptive suppresses the natural cycle while adding a synthetic estrogen and a progestin; the Pill Project's formulation is 30 µg ethinylestradiol with 150 µg levonorgestrel . Replacement raises estradiol from a low baseline: FEMME titrated a transdermal patch toward a target serum range . Postpartum studies scan in the days after delivery, where their authors place the fall in estrogen that follows late pregnancy .

None of these is better than the others in general; each supplies a different comparison. “Low estrogen” a fortnight into suppression, decades after surgical menopause and three months after VCD are three different biological states, in different species, under different measurement conditions.

The scan moments matter as much as the curves. Khayum 2020 scanned the same rats 13 and 60 days after surgery; Ramli 2024 scanned a different set of rats at each of three times; the menopause-stage and postpartum studies scanned each person once. A within-animal change, a between-cohort difference and a between-person difference are three different estimands, even with the same tracer and region .

Two rodent experiments, two different questions

Khayum 2020 tests what estradiol given at the moment of surgery prevents. Ramli 2024 compares stages of gradual depletion in separate cohorts. Each leaves out the experiment the other seems to supply.

In Khayum 2020 all 32 young adult female Wistar rats were ovariectomized (OVX) on day 0 and given an estradiol or placebo pellet at once, then crossed with chronic mild stress or control housing: four groups of eight . FDG PET followed on days 13 and 60, with stress from day 17. There was no sham or intact group and no delayed estradiol . The design identifies estradiol present versus absent in animals that have all lost their ovaries, a prevention contrast. It is not OVX versus sham, and it does not test whether estradiol reverses a state that has already developed .

Ramli 2024 asked something else. Rats aged three months received VCD or vehicle daily for 15 days, ovaries left in place, and were split into separate cohorts examined one, two or three months after the final injection: 30 rats, five per arm per timepoint . Cytology followed every group, but serum hormones and follicle counts were measured only at three months, where the authors report lower estradiol, progesterone and anti-Müllerian hormone, higher FSH and fewer follicles . The earlier stages have no measured hormone levels, and because each stage is a different set of animals, the design compares stages between cohorts rather than following a transition within any animal . The authors chose this layout to avoid learning effects from repeated water-maze testing .

The measurement conditions differ as much as the endocrine design

Khayum 2020 gave FDG intravenously under brief isoflurane and let the rats take it up mostly awake, reporting SUV; Ramli 2024 gave it intraperitoneally after a fast, under continuous isoflurane, reporting SUVR against the cerebellum . A fasting or glucose-correction step was not identified in the Khayum PET methods, and its stress schedule includes food and water deprivation . The “physiological” estradiol exposure attributed to the pellets is the authors' description, not a measured serum level. Each of these differences is a separate route by which two FDG numbers can diverge . Body weight diverged too: by the end, control-housed placebo rats had gained 56.2 ± 5.7 g against 10.4 ± 7.0 g with estradiol, which matters for a weight-normalised SUV without showing that the findings are artefacts .

Documented discrepancies stay visible

The full texts contain internal inconsistencies, recorded here as found, checked against the rendered pages and not repaired. None alone invalidates a study; together they are why the regional values are neither pooled nor digitised .

A human study that asks about fuel, not only glucose

FEMME paired FDG with carbon-11 acetoacetate before and after eight weeks of transdermal estradiol. It is a valuable measurement precedent with a small, uncontrolled denominator, not evidence of an estradiol effect on fuel choice.

FEMME (Female Estrogen Menopause Mind and Energy; NCT03681691) studied postmenopausal women aged 60 to 80, with or without type 2 diabetes, at Wake Forest . Its rationale was the healthy-cell-bias idea that estradiol may help glucose-using cells but act differently on metabolically stressed cells that lean on ketone bodies . Both groups wore an estradiol patch for eight weeks, titrated after a week-2 serum check, with PET at baseline and week 8. The protocol states there was no randomisation or blinding, because diabetes cannot be assigned .

The imaging is the interesting part: carbon-11 acetoacetate with a 30-minute dynamic acquisition, a washout, then FDG for 60 minutes, with venous samples from a heated forearm . Two tracers in one sitting can ask whether a change in glucose uptake comes with a change in ketone uptake, which one FDG image cannot .

The posted results show how far the analysed sample sits from the plan. Twenty women were planned; 12 enrolled (8 with diabetes, 4 without), 9 completed, and the primary PET outcomes report 8, six with diabetes and two without .

Four limits follow from the design and the posting, not from any judgement of the investigators. With no untreated comparison, a week-8 difference cannot be separated from time or repeat scanning. Diabetes was not randomised, and the group without it is two women in the primary PET outcomes. The registry gives timepoint means and standard deviations but no paired change, so the paired covariance needed for within-person change is unavailable. And the protocol framed acetoacetate relative to FDG as a way “to find potential areas of compensatory ketone use”, yet the pre-specified ketone-to-glucose ratio has no posted result . Nothing posted demonstrates a compensatory fuel shift .

The dual-fuel idea reappears in a registry lead. MOSAIC (NCT07021664), sponsored by the Université de Sherbrooke, plans a cross-sectional comparison of about 45 women aged 35 to 60 across the menopausal transition, with carbon-11 acetoacetate and FDG PET and fluorine-18 4FMFES receptor PET as a secondary measure . Its record still reads “not yet recruiting” with a passed estimated start, so it shows neither recruitment nor results; by design it would identify stage associations, not an estradiol effect .

Suppression and contraception: controlled, but never estradiol alone

Randomised suppression and randomised contraception give real causal leverage, but each moves the whole reproductive axis. The concrete resources here are a stale registry entry and a controlled-access dataset.

The Pill Project (NCT05212389), at the Neurobiology Research Unit in Copenhagen, registered a randomised, participant-blinded comparison of a combined pill with placebo over three cycles in healthy women aged 18 to 22, with change in neostriatal, neocortical and hippocampal 5-HT4 binding as the primary outcome and an estimated 40 participants . It follows cross-sectional work in which pill users had lower global 5-HT4 binding (Larsen 2020), an association that cannot separate the pill from who chooses it . The registry status is UNKNOWN, last verified February 2022, with no results posted; individual data are expected from December of 2026, after publication and subject to approval and a signed agreement . The investigator page gives ages 18 to 25, and the discrepancy is kept. Even with results, the randomised contrast is the whole formulation: suppression plus a synthetic estrogen and a progestin.

Frokjaer 2015 randomised 63 healthy women to a goserelin implant or placebo, double-blind; 60 completed, with carbon-11 DASB serotonin-transporter PET in the follicular phase at baseline and about 16 days after the intervention began . The authors report that depressive symptoms emerged, tracked the net fall in estradiol, and were associated with increases in neocortical transporter binding relative to placebo . It is one of the strongest designs in the human map, and still a suppression of the axis after an initial stimulation, not an isolated estradiol withdrawal.

A controlled-access repository, PN000018, lists this study with 63 participants, anatomical and PET data and 39.54 GB, behind a Data User Agreement . It is an access route to an existing study, not a new cohort or 63 proven scan pairs; its variables are unknown, its displayed BIDS version is malformed, and no access has been requested .

Publication is not replication here either

The contraception preprint Kauffmann 2025 states that its altanserin sample includes the original 29 participants of Frokjaer 2009 plus ten more, so it is not an independent replication of that comparison . A 2009 conference abstract, Sacher 2009, describes 15 women at days 3 to 6 and 8 at weeks 2 to 8 postpartum, reporting that monoamine oxidase A binding normalised after the first week; the journal article, Sacher 2010, reports only the early comparison. Whether the later eight were rescans, a separate sample or ever published is unresolved .

Which contrast does each design identify?

Sorted by what they can identify, the designs spread across six contrasts. No design fills every column, and random assignment of a hormone is rare.

Figure 3.3 places twelve designs against six contrasts: exposure (hormone present or absent at the scan), suppression (pharmacological lowering of the axis), withdrawal (a transition observed across the fall), replacement (hormone added back to a low state), adaptation (change with time in a state) and association (states nobody assigned). Marks come from assignment, timing and comparator, not from evidence counts or effect sizes .

Two patterns stand out. Filled marks for exposure, suppression and replacement come from designs that assign the hormone; association is filled by designs that do not. And withdrawal is the hardest contrast to isolate: the randomised discontinuation of Rasgon 2014 comes closest, while the GnRHa and postpartum designs reach it only partly, because the whole axis, or all of postpartum physiology, moves with estradiol . Neither rodent study gives estradiol after a delay, and no row compares early with late initiation in one model, which a timing question would need.

The human study map holds 73 rows: 26 hormone-therapy or administration rows (plus four experimental-administration and estrogen-use rows), 14 menstrual-cycle, 10 menopause-stage, 6 contraception, 4 pregnancy and postpartum, 3 gender-affirming, 3 direct-target validation, 2 surgical menopause and 1 ovarian suppression. Twenty-nine are associations (27 cross-sectional, 2 comparisons of treatment users); only five involve random assignment, and those come from four trials, because two rows share the KEEPS trial (Kantarci 2016; Kantarci 2026) .

Prevention, withdrawal, adaptation and rescue are different experiments

Four questions hide inside “what does estrogen loss do to the brain?”, and each needs its own comparison and its own scan timing.

Prevention asks what changes when the hormone is kept present from the moment it would have been lost. Withdrawal asks what happens across the fall, which needs the same animal or person measured before and after, ideally against a group that does not fall. Adaptation asks what changes with time in the low state, which needs repeated measurement after the fall. Rescue asks whether restoring the hormone after a delay reverses what adaptation produced, which needs the adapted state shown first and a late-start arm compared with an untreated one .

The rodent pair shows why this matters. Lower uptake at day 13 in placebo-implanted rats, compared with estradiol-implanted ones, is a prevention contrast at a short interval. Lower hippocampal SUVR in a three-month VCD cohort, compared with its vehicle cohort, is a depletion-stage contrast between animals. A day-13 to day-60 change in the placebo arm would be adaptation, but half of those animals were stressed between scans. None is a rescue contrast, and agreement or disagreement in direction between them cannot by itself identify a mechanism .

Human analogues exist for parts of this grid: randomised estradiol add-back during GnRHa suppression would turn suppression into replacement within one axis state; randomised discontinuation identifies withdrawal; repeated postpartum windows in the same women would separate the early fall from what follows. These are hypotheses about design, not proposals for anyone's facilities .

What a first collaboration could map

The cheapest useful next step is a ledger of what existing studies actually manipulated and measured, built before any new scan is planned.

For a small set of original studies, such a ledger would record exposure composition, timing relative to the transition, which hormones were measured and when, tracer and signal model, normalisation and reference, paired completeness, controls and cohort genealogy . It needs no scanner or participant data; its output would be the contrasts the literature identifies and two or three hypotheses worth testing.

Resources sit in different states: the FEMME protocol is public and its results are marginal summaries; the GnRHa repository is controlled and its variables unknown; the Pill Project may share data from December of 2026, under conditions. GSE268557 deposits hippocampal arrays from VCD and control mice in a 3xTg-AD model, with 28 sample titles in groups of 8, 6, 6 and 8; nothing has been downloaded, and it is target biology, not PET . Which of these a collaboration should take up, as a synthesis, an analysis or a new design, depends on its interests and access, which this review does not assume.

Seventy-three rows, read as a map

The human study map is a coverage instrument. It shows where molecular imaging has looked at ovarian state, not how strongly any answer is supported.

It was assembled in an earlier pass of this project from a 2025 systematic review of human non-clinical studies (Walsh 2025), extended with clinical and more recent work, and cross-checked against primary sources. Each row records one study or analysis: endocrine state, design, tracer, estimand, sample size, finding, attribution limit, cohort links and reading depth. The rows were not re-read for this report; later source checks are flagged inside the rows they change.

The map is uneven: hormone therapy or administration holds 26 rows, ovarian suppression one. It is mostly observational: 29 association rows and 25 within-person contrasts against five randomized rows. And it is mostly read at abstract level: 47 rows rest on primary abstracts alone and one on an abstract with partial methods, against 20 read in full text and five from review extraction or metadata.

Rows are not independent studies. Both Jovanovic 2009 rows come from one paper, as do the FDG and amyloid rows of Mosconi 2021. Moses 2000 and Moses-Kolko 2003 analyse the same five women, and three rows each descend from one Ottowitz infusion programme, from the NCT00097058 trial and from KEEPS, the Kronos Early Estrogen Prevention Study. A later check found that the Kauffmann 2025 preprint reuses the 29 participants of Frokjaer 2009 .

The sections below walk the atlas by system, giving the strongest or most instructive rows more room, then turn to three branches the map predates: inhibitory signalling, AMPA receptors and synaptic density.

Each tracer class keeps its own clock

Receptor snapshots, challenge displacement, enzyme trapping, transporter binding, glucose use and amyloid accumulation answer different questions, each on its own timescale.

Baseline availability (BPND or VT) is a snapshot at the scan. A difference between states can reflect receptor number, affinity, the tracer’s free fraction, delivery, the reference tissue and, for some tracers only, competition from an endogenous ligand. The map records antagonist 5-HT2A tracers as generally insensitive to acute serotonin release, and [11C]DASB as not validated as a release assay .

Challenge displacement differs in kind: a change in binding during a stimulus within one session (amphetamine, sustained pain), read as release, on a clock of minutes. A baseline contrast between cycle phases is not a release assay. Enzyme tracers split again: [11C]5-HTP and [18F]FMT are trapped by synthesis, [11C]PMP reports a hydrolysis rate, and [11C]harmine VT reflects reversible enzyme binding. FDG reports glucose uptake over one uptake period, sensitive to task, fasting, delivery and normalization. Amyloid and tau tracers integrate years to decades of pathology.

Figure 4.2 sets the classes against the endocrine exposures the map’s studies actually used. A measurement taken in one session can sit days or years away from the exposure it is meant to reflect, and the design has to say why that interval suits the biology.

Serotonin: the densest band and the best design

Serotonin holds 24 rows and the map’s strongest causal design, but its receptor, transporter, synthesis and enzyme tracers are four different measurements.

The band has 13 receptor rows (5-HT1A, 5-HT2A, 5-HT4), five transporter rows, one synthesis row and five monoamine oxidase A rows, and touches every endocrine context in the map.

The strongest design is Frokjaer 2015: 63 healthy women randomized to a GnRH agonist or placebo, 60 completing, with neocortical SERT binding measured at baseline and about 16 days later . There was no simple global effect, but depressive responses tracked SERT change. Randomization makes the contrast credible without changing what [11C]DASB binding is, and a GnRH agonist suppresses the whole axis rather than isolating estradiol. A controlled-access record linked to this study (PN000018) declares 63 participants; no access has been requested.

Sacher 2023 is the most detailed cycle study: 118 scans in 30 women with premenstrual dysphoric disorder (PMDD) and 29 controls. Midbrain SERT BPND rose about 18% premenstrually in PMDD against a roughly 10% fall in controls : a regional, symptom-linked change in availability, not a healthy-population hormone effect.

The receptor rows are smaller. Kranz 2014 randomized 30 postmenopausal women to estradiol with progesterone, estradiol alone or placebo for at least eight weeks and found no significant arm-by-time change in 5-HT1A BPND ; ten per arm leaves wide uncertainty. Moses 2000 and Moses-Kolko 2003 analyse the same five women, Kugaya 2003 (ten women, ten weeks of estradiol) had no control arm, and the nominal difference in Compton 2008 (1.17 versus 1.11, p = 0.02) did not survive correction. Kauffmann 2025 pooled two tracers: null overall, with lower binding in pill users in the agonist [11C]Cimbi-36 subgroup .

Larsen 2020 found roughly 9 to 12% lower global 5-HT4 binding in 16 oral-contraceptive users among 53 women ; a randomized pill-versus-placebo 5-HT4 study (NCT05212389) shows status unknown and no results. Among the [11C]harmine rows, Sacher 2010 found higher MAO-A binding 4 to 6 days postpartum, Rekkas 2014 higher binding in a perimenopausal-age group (age and stage not separated), and in gender-affirming treatment testosterone lowered binding (Kranz 2021) while Handschuh 2024 detected no change after about four and a half months.

Dopamine: availability, release and synthesis are three measurements

Eight dopamine rows ask three questions, and the clearest positive concerns synthesis capacity in contraceptive users.

Within the cycle, Nordstrom 1998 had only four women with a true cross-phase contrast and no difference beyond test–retest variability ; Petersen 2021 (16 women, [18F]fallypride) and Best 2005 ([123I]β-CIT) were also null. Taylor 2023 found higher synthesis capacity ([18F]FMT) in 15 contraceptive users than in 21 naturally cycling women, without differences in baseline D2/3 availability or stimulated release , and the exploratory amphetamine analyses of Smith 2019 agreed on release. Synthesis is a trapping estimate, BPND is availability, and amphetamine displacement is the release assay.

Gardiner 2004 is the one administration row: 13 postmenopausal women received conjugated equine estrogens (CEE) for four weeks, then two more with medroxyprogesterone, and anterior-putamen transporter uptake rose . With no placebo and a progestin added mid-study, estradiol is not isolated. Moses-Kolko 2012 (postpartum and depression) and Pohjalainen 1998 (a secondary eight-woman menopause contrast) complete the band.

Opioid: baseline availability and a pain challenge are different endpoints

One estradiol study measured two estimands in the same women, and they should not be merged.

Smith 1998 found no robust phase contrast in ten cycling women, although carfentanil binding related to luteinizing-hormone pulsatility.

Smith 2006 is the instructive example . Eight of ten recruited women were analysed, each scanned twice in the early follicular phase (days 2 to 9) and twice after 7 to 9 days of transdermal estradiol, which raised mean estradiol from 38 to 262 pg/ml. Each pair combined a control infusion with sustained jaw-muscle pain held 20 to 40 minutes after injection; eight men were scanned for comparison, against an occipital reference.

High estradiol was associated with higher baseline μ-opioid availability in some regions and with greater pain-evoked displacement (ΔBP), read as endogenous opioid release; the low-estradiol state showed reduced opioid activity in thalamus, nucleus accumbens and amygdala, with hyperalgesia . Baseline BPND is availability at rest; ΔBP is displacement during pain, and can also reflect trafficking or affinity. The estradiol scans always came second, so order cannot be excluded.

Cholinergic: four small windows on four targets

Four rows, four targets, and no randomized treatment.

Cosgrove 2007 found no phase effect on β2* nicotinic availability in nine non-smokers. Smith 2001 (16 hormone-therapy users, 12 non-users) saw no group difference in vesicular acetylcholine transporter binding, though longer therapy correlated with higher cortical binding; Norbury 2007 reported higher muscarinic binding in 11 estrogen users than in 11 never-users; Smith 2011a measured an acetylcholinesterase hydrolysis rate, not binding, in 50 women grouped by formulation . All three compare treatment chosen years earlier, so selection, formulation and duration are entangled.

Glucose use: downstream of nearly everything

FDG supplies 17 rows and some of the best designs, but glucose uptake also integrates perfusion, task, fasting and normalization.

The fastest clock belongs to Ottowitz 2008a: in eleven women receiving graded estradiol infusion, hypothalamic uptake fell at 24 hours with luteinizing-hormone suppression and pituitary uptake rose at 72 hours with the surge ; Ottowitz 2008b and Ottowitz 2008c reanalyse the same programme. Rasgon 2014a randomized women at elevated dementia risk to continue or stop long-term therapy for two years, and continuing relatively preserved frontal and parietal metabolism . That randomizes continuation, not initiation or formulation; Rasgon 2014b and Silverman 2011 analyse the same trial.

Mosconi 2017, Mosconi 2018 and the FDG arm of Mosconi 2021 (161 women) report lower metabolism in Alzheimer-vulnerable regions across the transition , but age and stage overlap only partly, and participant overlap between the papers is unresolved. Cycle rows (Reiman 1996, Tu 2009, Rapkin 2011) and naturalistic therapy studies (Rasgon 2001, Rasgon 2005, Kenna 2009, Eberling 2000, Eberling 2004) complete the band.

Amyloid and tau run on decades

Eleven pathology rows include two randomized follow-ups of one trial, read at different delays and with different results.

KEEPS randomized recently postmenopausal women to oral CEE, transdermal estradiol or placebo for four years. In 68 women scanned about three years after treatment, transdermal estradiol was associated with lower PiB SUVR than placebo (adjusted OR 0.31, 95% CI 0.11 to 0.83) and oral CEE was not (Kantarci 2016) . About ten years after the trial, 244 amyloid scans showed no significant difference among randomized groups (Kantarci 2026) . Differently timed and powered rather than contradictory, neither is an acute hormone-sensitive signal .

The observational rows are larger and harder to attribute: therapy timing with tau in the presence of amyloid (Coughlan 2023), faster tau accumulation after prior therapy in women over 70 (Coughlan 2025), lower tau biomarkers with therapy (Wang 2024), few PET associations (Lee 2024), and higher amyloid and tau after oophorectomy before age 46 (Kantarci 2025). Mosconi 2021 found no menopausal-group amyloid difference, Buckley 2022 found menopause moderating sex differences in tau, and Zeydan 2021 is a post hoc sleep analysis within KEEPS. Exposures, timing, ages and selection differ; no pooled effect would be meaningful.

Small nulls, postpartum and mixed regimens

Three habits would distort this map: reading small nulls as equivalence, treating postpartum as a model of menopause, and crediting estradiol with a mixed regimen’s effects.

Eight downstream cycle-phase rows found no clear phase effect (Nordstrom 1998, Smith 1998, Best 2005, Jovanovic 2006, both Jovanovic 2009 arms, Cosgrove 2007, Petersen 2021), each with four to sixteen women. They cannot show that phase leaves these targets unchanged, but they keep the chapter from resting only on positives.

Postpartum is not healthy menopause. All four postpartum rows are cross-sectional, three involve depression or postpartum crying, and the best-timed samples days 4 to 6, when hormones, lactation, sleep and mood change together. Moses-Kolko 2008 reported roughly 20 to 28% lower 5-HT1A binding in postpartum depression, a clinical contrast without a pre-pregnancy baseline. A 2009 conference abstract (Sacher 2009) adds eight women at weeks 2 to 8 with reported normalization of MAO-A binding ; its overlap with the published cohort is unresolved.

A mixed regimen can be the right clinical question and the wrong mechanistic one: estrogen in the rationale does not make estradiol the isolated cause.

The inhibitory perspective

Inhibitory signalling keeps the synaptic story from collapsing into AMPA alone, but its four common measurements are not one quantity.

No map row images GABA. Michopoulos 2013 used [18F]flumazenil PET in ovariectomized female rhesus monkeys, all receiving estradiol, to test whether social subordination alters GABA-A receptor BPND . Subordinate females had higher prefrontal binding than dominant ones, and three days of a CRH receptor antagonist (astressin B), compared with saline, removed the difference.

Estradiol was present in both conditions. The study varies stress signalling against a fixed estradiol background and says nothing about estradiol alone. The reading covers the abstract and indexed methods, not the full text.

Flumazenil binding is benzodiazepine-site availability: not GABA concentration, not inhibitory current, and not neurosteroid occupancy at other receptor sites. MRS GABA+ is a spectroscopy signal that includes co-edited macromolecules. The SAGE protocol (Jimenez-Balado 2026) plans spectroscopy, task fMRI, hormones and reproductive history in 300 adults aged 50 to 79, two-thirds women ; it has no outcomes yet and no PET. A flumazenil PET and TMS study notes that menstrual cycles were not accounted for (Steinholtz 2023). The question these motivate is whether an endocrine contrast moves receptor availability, GABA+ and excitability together or apart .

AMPA: a registered direction with a metabolite-mediated signal

Menopause-related AMPA imaging is a real, registered study, built on a tracer whose image is attributed mainly to a labelled metabolite and validated so far in reused and male cohorts.

The WISH record (jRCTs031250654) describes a single-arm, open-label diagnostic study of 60 women aged 40 to 59 , relating the Menopause Rating Scale to white-matter-referenced [11C]K-2 SUVR, with hormones, menopausal status and time since the last menstruation as secondary endpoints. It is interventional only because it gives an unapproved probe. Hormone therapy and ovarian or uterine removal are excluded, and the Japanese list also excludes recent GnRH-analogue use, an item missing from the English list, so WISH cannot address surgical menopause or treatment response. Enrolment began on 9 February 2026 with a planned end in March 2030, and the study site reported 18 completed PET scans on 17 September 2026: a milestone, not an analysed sample.

Arisawa 2021 attributes the image mainly to labelled K-2OH rather than unchanged K-2 . In an in-vitro barrier model, permeability was 182.89 ± 3.28 ×10⁻⁶ cm/s for K-2 and 5.26 ± 0.09 ×10⁻⁶ cm/s for K-2OH, which showed no substrate behaviour in three efflux assays ; the rat work used a pharmacological 10 mg/kg dose and four male rats.

A brain-active metabolite does not invalidate a tracer; it obliges a specification of which labelled species binds, how it arrives and how much it contributes. Nothing here shows those relationships hold across endocrine states .

The validation chain has a reused link: Miyazaki 2022 reanalyses five of the eight patients of the original study , which itself (Miyazaki 2020) was not freshly read here . The repeatability registry (jRCTs031250631) plans two scans in ten men aged 20 to 39 and reports no number . Briz 2015 adds caution about direction: in male-rat hippocampal slices, G-1 or estradiol reduced membrane GluA1 in CA3–dentate but not CA1 at one hour, and G-1 alone did not change field responses but primed later long-term depression . Surface protein, total protein and plasticity moved separately; the sign of a human K-2 effect is not predicted .

SV2A: an emerging direction, signalled by programmes

Synaptic-density imaging in menopause is visible here only as a programme title and a grant announcement: a direction to watch, not a result.

The 15 May 2026 SLAM-DUNC programme (SLAM-DUNC 2026) lists Melissa Walsh’s presentation “Mapping Menopause-Related Synaptic Vulnerability: A Pilot SV2A-PET Study” . A 20 April 2026 announcement describes a $4.2 million NIMH grant for SV2A PET in premenopausal and postmenopausal women with and without depression (Esterlis and Pietrzak 2026) . Neither supplies results, data or recruitment status; they sit beside AMPA so that unpublished is not misread as unstudied.

GPER: a name that needs a qualifier

An effect called GPER-mediated is usually G-1-responsive or GPER-implicated; whether estradiol binds GPER directly is contested.

Liu 2024 reported structural, binding and signalling evidence that GPR30 (GPER) is not a direct estrogen receptor in the tested systems . That does not settle every GPER mechanism, and pathway or agonist evidence does not prove direct estradiol binding either; engineered-cell binding work continues (Ding 2022).

The paper’s 2025 author correction (Liu 2025) states that the G-1 structure shown in Fig. 1a and Supplementary Fig. S3b was incorrect, that the correct G-1 was used experimentally, and that results and conclusions are unaffected . It is not a retraction; figures should use the corrected structure. A separate 2024 correction fixed names in one reference. So when Briz 2015 calls its plasticity GPER1-dependent, read GPER-implicated through G-1 pharmacology, not proof that estradiol binds GPER .

What would discriminate

The decisive downstream study pairs a well-defined endocrine contrast with a tracer whose estimand, clock and quantification have been checked in the population studied.

Four ingredients, none exotic: a contrast that identifies something (randomized or within-person exogenous, hormones measured at the scan); a tracer whose sensitivity to endogenous competition is known and whose window fits the question; metabolite-corrected input or a tested reference, with repeatability in comparable women; and an orthogonal assay of the same biology.

One word, four claims

“Neuroinflammation” bundles at least four separable claims: how many immune cells a tissue holds, what state they are in, what their mitochondria are doing, and what disease surrounds them. A PET tracer can speak to some of these. None speaks to all of them at once.

When a paper says a tracer signal “reflects microglial activation”, it compresses several statements that each need their own evidence. Cell abundance is how many microglia, astrocytes or perivascular macrophages occupy a volume of tissue. Cell state is what those cells are doing (surveying, engulfing, signalling), read from morphology, transcripts and protein markers. Mitochondrial signalling matters because two of the best-known targets, TSPO and MAO-B, are mitochondrial proteins whose amount or modification can follow a cell's metabolic and steroid-handling business rather than its immune role. Disease context matters because a tracer can have a validated relationship with pathology in progressive supranuclear palsy or HIV that tells us nothing about a healthy brain passing through menopause.

These claims need not conflict, and the most informative TSPO literature shows exactly that: a tracer can be a faithful index of how many microglia a region contains while saying very little about whether each of those cells is “activated” . Holding the four apart is not pedantry. It decides which experiment can test an estrogen hypothesis, and which result would merely be compatible with one.

What a TSPO signal is a sum of

TSPO is a mitochondrial protein carried by microglia, astrocytes and the cells of vessel walls. In human tissue its PET signal tracks how many microglia there are, not how activated each one is.

Wijesinghe 2025 is the cleanest test available. Eight people with progressive supranuclear palsy, a tauopathy, had PK11195 PET in life and donated their brains. Double-labelling of the same donors' tissue found TSPO in microglia, astrocytes and endothelial cells . Regional binding potential before death correlated with the post-mortem burden of CD68-positive phagocytic microglia, and microglial TSPO was higher in the patients than in three control donors. TSPO per microglial cell, however, did not differ significantly . The rise came from more microglia, each carrying a similar amount. The authors conclude that disease-related PK11195 changes in tauopathy are largely driven by microglia . Two limits travel with that conclusion: scans preceded death by 6 to 49 months, a long interval for a moving process, and the study was not powered to confirm small age or sex effects.

Nutma 2023 reaches a compatible conclusion from the other direction. In mouse disease models, activated microglia increased TSPO expression; in a non-human primate disease model and in common human neurodegenerative and neuroinflammatory diseases they did not . The authors trace the difference to divergence in the TSPO gene's promoter, consistent with an activation-induced increase confined to a subset of rodent species, and argue that human TSPO-PET signals reflect the density of inflammatory cells rather than their activation state . Read here at abstract level, it is a species caution. It is not a verdict that human TSPO PET carries no inflammatory information.

The apparent contradiction, a human validation that “worked” set against a paper arguing that TSPO is not an activation marker, dissolves once the claims are separated. Both say that in human tissue TSPO PET follows cell number. Neither says it reports activation per cell. Applied to endocrinology, a TSPO increase after ovarian hormone loss would be a statement about how many TSPO-carrying cells are present, and possibly which ones, not about how inflamed each of them is .

Demography is part of this measurement, not background to it. Tuisku 2019 pooled PBR28 PET from 140 healthy adults (72 men, 68 women) scanned at three centres, with the tracer's genotype-dependent binding class entered in the model. Women had higher distribution volumes in every region examined, body-mass index was negatively related to binding everywhere, and age correlated positively with binding in men but not in women . Sex, body mass and age are all entangled with menopause, so a cross-sectional endocrine comparison inherits all three. The absent age effect in women does not refute the hypothalamic finding below, because tracer, regions and cohort differ. It does show how unstable an age-by-sex pattern can be from one sample to the next.

The endocrine bridges are indirect

Three studies connect TSPO to ovarian hormones, and each makes a different kind of claim: an age association in women, a delayed transcriptional response to ovariectomy in mice, and a mitochondrial signalling pathway in rats. None is an endocrine-annotated human TSPO PET contrast.

Butler 2022 comes closest to the human question. Forty-three healthy adults (19 women, aged 23 to 78), pooled from the control groups of several clinical studies, had dynamic PK11195 PET . Across everyone, binding correlated with age in the thalamus but not the hypothalamus. Analysed by sex, only the women showed an age-correlated hypothalamic signal, and the authors compared the two correlations directly rather than leaning on significance in one subgroup. They link this to menopause, the sex difference in reproductive ageing . That link is the gap. Reproductive stage, reproductive history and hormone exposure were not available, so age stands in for an endocrine variable nobody measured . The authors acknowledge that TSPO is expressed by several cell types and that the hypothalamus is small relative to PET resolution. In this chapter's terms, the title's “microglial activation” is a reading of the signal rather than a measured quantity.

Bloom 2026 supplies the time axis. Mice were examined 14 days and 4 months after ovariectomy, with bulk RNA sequencing of the posterior hypothalamus and preoptic area. Luteinizing hormone rose and then declined, and core temperature peaked early and then normalized. Inflammatory pathways, glial markers and gene networks of the KNDy neurons that drive reproductive hormone release changed with time, most strongly at four months in the posterior hypothalamus, and immunofluorescence showed heightened astrocytic reactivity in the arcuate nucleus . Age-related transcript patterns in 31 female human hypothalami correlated with the mouse model, particularly at pathway level.

Three limits govern reuse. The four-month group had no age-matched intact comparator, so ovariectomy and ageing are partly confounded. The human tissue carried no menopausal stage, hormone therapy or hormone measurements. And individual inflammation markers, including a discussed Tspo trend, did not survive correction for multiple testing . A pathway is not a target gene, and a transcript is not a binding site a tracer could find.

Wang 2021 points somewhere else entirely. In ovariectomized rats, an agonist of the G-protein-coupled estrogen receptor given into the cerebral ventricles activated protein kinase A, increased TSPO phosphorylation and eased depression- and anxiety-like behaviour, and a kinase inhibitor partly blocked the effect . Read at abstract level, it suggests estrogen signalling can modify TSPO through mitochondrial pathways that have nothing to do with how many immune cells are present. Phosphorylation is not binding-site abundance, and injection into the ventricles says nothing about exposure after systemic dosing. It is a useful counterweight to the default chain from estrogen loss to inflammation to a higher TSPO signal, a chain no study here has demonstrated end to end.

Five other targets, five other questions

MAO-B, P2X7, CSF1R, COX-1 and COX-2 are not interchangeable confirmations of neuroinflammation. Each is a different molecule with a different cellular home, a different level of human validation, and a different, mostly indirect, link to ovarian hormones.

The matrix below sets them beside TSPO. Its most important column is the fourth. For the targets other than TSPO, the endocrine evidence comes from animal tissue, cell culture or injury models, and none of the PET studies selected here tested a human hormone manipulation . That is the outcome of a bounded search, not proof that no such study exists. Several of the readings behind the matrix are at abstract level, and the right-hand column says so.

MAO-B: an astrocytic enzyme whose reference may itself respond

SMBT-1 is a fluorine-18 tracer for MAO-B. Lopresti 2024 modelled it in 14 older adults (8 controls, 5 with mild cognitive impairment, 1 with Alzheimer's disease) with arterial input. A two-tissue model with the K₁/k₂ ratio fixed to a whole-brain value described the kinetics best; cerebellar grey matter had the lowest distribution volume and was proposed as a pseudo-reference, and a 70–90 minute ratio to it tracked the modelled value with about 10% bias . The authors describe brain MAO-B as widespread and predominantly astrocytic, and associate reactive astrogliosis with its overexpression .

That pseudo-reference is where the endocrine question bites. In ovariectomized rats, acute or month-long estradiol increased MAO-B activity in the brainstem areas sampled and in the cerebellum (Chevillard 1981) . If anything similar happened in women, an endocrine contrast could move the denominator of an SMBT-1 ratio, the same structure chapter 1 found under the FES menopause finding . A later study found no brain MAO-B change after three weeks of high-dose estradiol in cortex, amygdala or hypothalamus, while MAO-B fell in liver, kidney and uterus (Holschneider 1998) . Different regions, doses and durations mean the two are not a failed replication. Together they say that peripheral and brain responses cannot stand in for each other, and that the cerebellum's stability is a hypothesis to test, not a default. MAO-A imaging across perimenopause and postpartum belongs to chapter 4's downstream map. It is a different enzyme and should not be folded into a claim about astrocytes.

P2X7: a bulk measure can hide opposite changes

P2X7 is an ATP-gated ion channel that its PET developers describe as predominantly microglial in the central nervous system (Koole 2019) . Their tracer, [¹⁸F]JNJ-64413739, was qualified in healthy men: 3 for dosimetry, 5 for arterial kinetic modelling, 4 of those rescanned 26 to 97 days apart, and 8 in an occupancy study with a P2X7 antagonist. These phases overlap, and the trial registry lists 16 participants in total, so the counts must not be added. Distribution volumes were about 3 and fairly uniform across regions, test–retest variability was about 11%, occupancy approached saturation above single doses of 50 mg, and no reference region could be identified. It is a working measurement and drug-engagement route, established in men only. That is an annotation gap for an endocrine study, not evidence that the tracer behaves differently in women.

The endocrine evidence is cellular, and it carries a warning about bulk measures. In ovariectomized mice with a cut hypoglossal nerve, a single estradiol injection enhanced two opposite changes on the injured side: P2X7 immunostaining in neurons fell further, while P2X7 in microglia rose further (Barabasi 2016) . Total P2X7 protein measured by Western blot did not differ between sides in either group, and estradiol did not change the area occupied by microglia. A tracer that sums P2X7 across a voxel could report nothing in a tissue where the receptor had moved substantially between cell types .

In a rat model of dry eye, estradiol-treated females showed larger increases in trigeminal brainstem P2X7, in the myeloid marker Iba1 and in the inflammasome component NLRP3 than untreated females, and a P2X7 antagonist injected into the brainstem reduced the pain-related response (Bereiter 2022). Both studies are injury or disease models, and a locally injected antagonist says nothing about systemic exposure. Neither tells us what P2X7 does in an uninjured brain losing ovarian hormones.

CSF1R: preferential expression is not tracer specificity

CSF1R is expressed on microglia and central nervous system macrophages, with little expression on other cell types, which is why it was proposed as a more cell-specific alternative to TSPO (Rubin 2023). In the first clinical pilot, 16 virally suppressed people with HIV and 15 uninfected participants had [¹¹C]CPPC PET with metabolite-corrected arterial input. After adjustment for age and sex, regional binding did not differ significantly between groups (p = 0.13); the effect size was moderate, and the authors read the study as underpowered . Nothing in the design involved ovarian state.

Specificity comes before any of that. A development study comparing CPPC with a higher-affinity fluorinated analogue found no CSF1R-specific binding at baseline for either radioligand in mouse or monkey, and found that both behaved as substrates of brain efflux transporters in vivo (Altomonte 2023) . Baseline specific binding in healthy animals may simply be too small to detect, and inflamed tissue may behave differently. Until specific binding is shown, a CPPC signal cannot be assumed to measure CSF1R, let alone microglia . The endocrine link is a mouse transcript: after early ovariectomy, with or without estradiol replacement, cortical Csf1r RNA responded differently in amyloid-bearing and wild-type mice (Saha 2024) . Bulk RNA cannot separate more microglia from more Csf1r per microglial cell, and the authors state that the model does not represent a menopausal brain.

COX-1 and COX-2: two enzymes, two hypotheses

Cyclooxygenases make prostaglandins, and the two isoforms are not two names for one target. In ovariectomized rats, COX-1 around preoptic GnRH neurons sat entirely within CD11b-positive cells with the small bodies and highly branched processes of ramified microglia, whether or not the animals received steroid replacement (Fujioka 2013) . The localization suggests a route from microglia to the neurons that drive the reproductive axis. Its constancy across steroid conditions makes it a constraint on direction, not a hormone-sensitive imaging target.

COX-2 differs in nearly every respect. It is sparse in healthy brain, preferentially neuronal, and can be markedly upregulated by excitatory input and inflammatory stimuli (Yan 2025) . In cultured rat microglia exposed to bacterial lipopolysaccharide, estradiol and both ERα- and ERβ-selective agonists attenuated the release of COX-2 and inflammatory cytokines (Smith 2011b) . That is a mechanism in a dish, not ovarian depletion and not imaging, and it does not make COX-2 a microglial marker.

The COX-2 imaging, by contrast, is unusually well characterised. Yan 2025 gave [¹¹C]MC1 to 27 healthy people with arterial sampling: 10 had baseline and celecoxib-blocked scans, and 17 had test–retest scans. Celecoxib reduced neocortical binding by about 25% with little effect in subcortical regions or cerebellum, a pattern that followed COX-2 messenger RNA. Test–retest reliability was moderate (intraclass correlation 0.65) with about 20% retest variability . The animal work is where an endocrine plan would stumble. In rats injected with lipopolysaccharide the tracer showed negligible specific binding despite strong COX-2 upregulation, because its affinity for rodent COX-2 is low; specificity was shown instead in mice carrying the human COX-2 gene, where 70–90% of brain uptake was blocked . That transgene is constitutively driven, so it establishes binding without reproducing normal inflammatory regulation .

A working human tracer does not make a rodent experiment informative

Species can decide an experiment before the hormone does. The standard ovariectomized-rat design is uninformative for a tracer that barely binds the rat target, and misleading for a target that rodents regulate differently.

Two of this chapter's clearest findings are species findings. Rodent microglia raise TSPO when activated and human microglia do not (Nutma 2023), and a COX-2 tracer able to measure sparse human COX-2 barely sees rodent COX-2 at all (Yan 2025). In the CSF1R development work, the analogue entered monkey brain but not mouse brain (Altomonte 2023). Together they give one practical rule: before an endocrine imaging experiment is designed in an animal, the tracer's affinity for that species' target and the target's regulation in that species both need to be established .

This also bears on how existing animal results are read. A rise in rodent TSPO PET after a hormone manipulation would combine cell number with an activation response that human microglia do not appear to share; a human study could find the cell-number component and nothing else, and both results would be correct.

No tracer sees only one cell type

Every lens in this chapter weights several cell types, or carries an unresolved specificity question, or changes its meaning with species and cell state. The figure below makes that concrete on one field of real cells.

Choose a tracer, then switch between human and rodent and between surveying and reactive cells. The field does not change; what each lens would emphasise does. The weights are deliberately qualitative. They summarise the cellular statements quoted above and are not measured proportions, and where the sources say nothing about a combination, the panel says so instead of guessing.

Two patterns stand out. The CSF1R lens is the only one close to a single cell type, and it is also the one whose radioligand has not shown specific binding: the cleanest cellular target sits behind the least resolved tracer gate. And the COX-2 lens goes dark in rodents while its target is present, the clearest case in this report of an available human tracer that would turn a standard animal design into a guaranteed null.

Design the bridge from the cell outwards

The opportunity is to choose an endocrine contrast, a time point and a context first, establish in tissue which cell quantity actually changes, and only then ask which tracer could see that change.

The usual order runs the other way: take an available tracer, scan an endocrine contrast, and interpret whatever moves as inflammation. The cases above show how that misleads. A TSPO change can be cell number or a different cell type; a P2X7 change can cancel inside a voxel; a CSF1R signal may not be CSF1R; a COX-2 null in rats is a property of the tracer. Starting from the cell makes each step testable on its own, and lets a negative result mean something.

The three threads are worked examples, not proposals. The first follows the delayed hypothalamic changes after ovariectomy and asks whether astrocyte number or per-cell target protein is what moves, before choosing between TSPO and MAO-B and before trusting a reference region. The second takes the P2X7 redistribution seriously: if the receptor moves between cell types after injury, a bulk read-out is the wrong first measurement. The third is the species gate. Each ends with a tracer only once the tissue has said what to look for.

Some resources lower the cost of starting. The COX-2 study deposited its data openly: a human dataset of 27 participants and a separate animal dataset whose 18 subject folders are labelled 12 mouse and 6 rat, both released into the public domain in the brain-imaging data standard . Only descriptors and folder listings were read. The human participant schema defines sex, age and weight and the animal schema sex and age; neither has a reproductive-stage field, which is an annotation finding rather than proof that hormone information exists nowhere. These data could support a methods exercise, for example how arterial and reference-tissue estimates diverge on a known cohort. They cannot estimate an endocrine effect. The ovariectomy transcript studies declare deposited RNA data and code (Bloom 2026; Saha 2024), not yet inventoried, and the human PET and pathology studies share data on request.

The next chapter turns to the vessels every tracer travels through, where the same discipline applies: a change in delivery can look exactly like a change in target.

Five quantities that share one word

Total arterial inflow, regional tissue perfusion, tracer delivery, barrier permeability and transporter function are different physical quantities with different units. A study that reports "blood flow" has usually measured one of them, and rarely the same one as the study it is compared with.

Start with what each instrument actually reads. Total arterial inflow is the volume of blood crossing a plane through the arteries that feed the brain, per minute: one number for the whole organ, in millilitres per minute. Tissue perfusion is the rate at which arterial blood reaches a given mass of tissue through its capillaries, conventionally in millilitres per 100 g per minute, and it can differ between regions while the total stays fixed. Relative regional flow is perfusion in one region divided by the mean over the whole brain. It is a dimensionless pattern, and by construction it cannot register a change that affects every region equally.

Arrival time is a fourth quantity, a delay rather than a rate: how long blood labelled in the neck takes to reach a tissue voxel. Delivery, in the sense a PET tracer cares about, is the influx constant K₁, the product of perfusion and the fraction of tracer that leaves the capillary in a single pass. That fraction depends on the permeability and surface area of the capillary wall. Transporter function, finally, is active: efflux pumps in the capillary wall move some molecules back towards the blood, against passive diffusion, so a well-perfused region can still keep a substrate out.

These are not five estimates of one underlying flow, and they can move independently. A hormone could redistribute perfusion between regions with no change in total inflow; lower total inflow while local regulation holds tissue perfusion steady; shorten arrival time without changing perfusion; or alter an efflux pump without touching blood flow at all . A result reported as cerebral blood flow becomes interpretable once you know which of these was measured.

A relative map, two years apart

The oldest of the three human studies measured flow with oxygen-15 water PET, but what it reports is a pattern of relative change between two groups who chose their own treatment. It is evidence about a pattern, not about whether estrogen raises brain perfusion.

Maki and Resnick 2000 followed 12 women using estrogen replacement and 16 non-users from the Baltimore Longitudinal Study of Aging over two years, scanning each at rest and during verbal and figural recognition tasks . Every measurement used a 2.8 GBq bolus of oxygen-15 water and a 60-second acquisition; the isotope's half-life of about two minutes is what makes repeated injections in one session practical. The users were not randomised. Ten took conjugated equine estrogens, one estradiol and one a triple-estrogen preparation; six also received progesterone; and they had been treated for a mean of 15 years. The authors name the healthy-user bias themselves.

The 28 women are a follow-up subset of an earlier cross-sectional comparison of 15 users and 17 non-users, so the first-year images overlap with a published analysis and are not an independent replication . Across the three conditions combined, users showed larger longitudinal increases in relative flow in right hippocampus, entorhinal cortex and middle temporal gyrus, among other regions; non-users showed larger relative increases in pons, superior frontal gyrus and elsewhere. Performance on the in-scanner memory tasks did not differ between the groups.

The phrase that matters is in the methods: proportional normalisation. Every image was divided by its own global mean before the groups were compared, so "increase" means increase relative to the rest of that brain on that day . That is a sensible choice for detecting a regional pattern, and it leaves the study silent on the question a reader most wants answered, whether estrogen use changed how much blood the brain received.

The statistical threshold was a voxel-level p < 0.01 combined with a cluster extent above 37 voxels, a common convention at the time that is not modern family-wise error control. With 12 and 16 women and many regions, some reported clusters would probably not survive a stricter test . None of this makes the study careless. It makes it a historical bridge between endocrine exposure and functional imaging, and a worked lesson in what normalisation removes.

One randomised trial, one neutral average, one dip in the middle

The only randomised design among the three measured total inflow through the neck arteries, and found a dip at one visit that recovered at the next: an overall neutral result whose shape carries more information than its average.

Sorensen 2001 enrolled 16 naturally postmenopausal women, aged 55 ± 3 years, in a double-blind crossover: twelve weeks of a sequential hormone regimen and twelve weeks of matching placebo, in random order, three months apart . Each 28-day cycle gave 4 mg estradiol daily for 12 days, 4 mg estradiol with 1 mg norethindrone acetate for 10 days, and 1 mg estradiol for the last 6. Flow was measured with velocity-encoded phase-contrast MR in both internal carotid and both vertebral arteries and summed, at baseline and in weeks 2, 9 and 11 of each period, up to eight examinations per woman. Two women withdrew, one because of venous thrombosis; a trial of this size cannot speak to safety.

Total flow was lower than baseline at week 9, reported as −37 ± 15 mL/min (P = .01), and rose by 45 ± 19 mL/min between weeks 9 and 11, once the progestogen phase had resumed . The authors summarise this as an overall neutral effect. Their own reading of the dip is more interesting than the summary: no reduction appeared while estradiol was given alone at the start, and the week-9 visit came after two cycles of progestogen exposure and withdrawal, so they raise progestogen withdrawal, rather than estradiol, as the trigger . That makes this a study of an endocrine sequence, not of estradiol in isolation. The systemic circulation moved too: vascular resistance and blood pressure fell within about ten days of starting estradiol, which is a reminder that a hormone can act upstream of the brain.

One detail explains why this chapter draws no effect-size graphic from the paper. Flow was corrected for body surface area, and the legend gives the background mean as 461 ± 24 mL per minute per square metre, yet the changes are reported in mL/min. The reported +9.8% for the rise from week 9 to week 11 equals 45 divided by 461, which suggests either that the changes are in the corrected units and labelled otherwise, or that the percentage mixes two units. The text does not let a reader decide, and the figure images were not available to check . Until the original figures are recovered, the size of the dip stays unconverted.

Total inflow is also blind to where the blood goes. The authors note that their global measurement does not contradict earlier reports of regional vasodilation, because blood can be redistributed between regions while the total stays the same . That is precisely the gap between this instrument and the previous one.

Perfusion and arrival time across the menstrual cycle

The newest study separates how much blood reaches tissue from how long it takes to arrive, and from how much oxygen the tissue uses. Higher hormone levels went with more perfusion and earlier arrival, and not with a detectable change in oxygen metabolism.

Wright 2026 scanned naturally cycling women with a mean age of about 23 in the early follicular, late follicular and mid-luteal phases, drawing blood for estradiol and progesterone at each session . Of 26 recruited, 21 contributed MRI data, giving 17, 17 and 18 observations across the three phases. Perfusion and arterial arrival time came from multi-delay pseudo-continuous arterial spin labelling at 3 T, with eleven post-labelling delays from 250 to 3,000 ms. Oxygen extraction was measured from venous blood signal and the metabolic rate of oxygen derived from it, and retinal vessel density and a modelled flow resistance came from optical coherence tomography angiography.

Across sessions, higher estradiol was associated with higher global perfusion and shorter arrival time, and progesterone, entered as the part of its variation not explained by estradiol, showed associations in the same directions . Neither hormone explained variation in oxygen extraction or oxygen metabolism. That combination is the paper's most useful contribution here. Delivery varied with endocrine state while the tissue's measured oxygen use showed no detectable change, so a perfusion association cannot simply be read as a metabolic one .

The design is associational, and the authors are candid about its limits. Phases were scheduled by counting days from menses, without ovulation tests, and the late-follicular estradiol peak was missed in several participants; they recommend measuring the hormone rather than assuming the phase . Blood pressure was not monitored. Residualising progesterone against estradiol separates the two statistically, which is not the same as varying them independently, and natural-cycle associations in young women do not transfer automatically to menopause or hormone therapy .

One sentence in the paper needs care. It describes both hormones as associated with "decreased global blood velocity (AAT)". Arrival time is a delay: the interval between labelling blood in the neck and that blood reaching a tissue voxel. A shorter interval means the label got there sooner. That could reflect faster transit somewhere along the route, a shorter effective path, or a change in how the arterial tree distributes blood, but it cannot mean slower blood, and velocity itself was not measured . This chapter keeps the measured endpoint, arrival time, and sets the mechanistic wording aside.

The data are public. The paper's data statement gives a G-Node repository DOI, and on 2 October 2026 that DOI resolved to a landing page describing MRI, retinal and hormone data from 26 volunteers at the three phases, offered as a 6.7 GiB archive under a CC BY 4.0 licence . The archive was not downloaded, so its contents, completeness and structure are unverified . A 2024 preprint is an earlier version of the same study, not a second cohort.

Read side by side, the comparison suggests a working rule rather than a correction. A hormone contrast in any molecular PET study arrives with some vascular state attached. Which part of that state matters depends on the tracer and on the summary measure it is read with, which is the subject of the next section.

Why a distribution volume is not a flow measurement

In a compartment model, delivery enters through K₁ and binding through k₃ and k₄. The equilibrium distribution volume depends on their ratios, not on flow. Yet a finite scan, an imperfect input function or a misspecified model can still let delivery leak into the estimate.

A compartment model describes tracer moving from arterial plasma into tissue with rate constant K₁ and back with k₂, and, if it binds, onto its target with k₃ and off again with k₄. The quantity most kinetic studies report is the total distribution volume, V_T: the tissue-to-plasma concentration ratio the system would reach at equilibrium. It is K₁/k₂ in a one-tissue model and K₁/k₂ × (1 + k₃/k₄) in a two-tissue model, and the binding potential relative to nondisplaceable uptake is k₃/k₄.

The crucial property sits in the ratio. If perfusion changes, K₁ and k₂ change together, because the same passive exchange across the capillary governs entry and exit, so K₁/k₂, and with it V_T, stays where it was. A correctly specified distribution-volume model is therefore not a flow measurement. What changes with perfusion is how fast the tissue approaches equilibrium, and anything read before equilibrium, including a single late time window, can move with delivery even when binding has not changed .

Efflux is different in kind. An active pump in the capillary wall can speed exit, or reduce effective entry, without any change in perfusion, and that does change the ratio K₁/k₂. This is why transporter PET reads P-glycoprotein function from the distribution volume of a substrate tracer: for those tracers V_T is expected to move with transport . In the model, flow alone does not change V_T; transport can.

Reference-region methods sit in between. A global change in delivery reaches target and reference alike, but it cancels in their ratio only if the two regions share the same kinetics. Where binding slows the target and the reference equilibrates quickly, a global delivery change can move the ratio more than it moves either region, because the two are caught at different points on their way to equilibrium . Chapter 1 discusses the reference region of the human estrogen-receptor tracer; the explorer below lets you try the kinetics with schematic curves.

Two things the explorer cannot show deserve a sentence. It assumes one well-mixed plasma input, whereas real inputs differ by region in delay and dispersion. And its parameters are arbitrary: whether a hormone-scale change in perfusion would move a particular tracer's late uptake by a fraction of a per cent or by several per cent depends on that tracer's own kinetics. That is an empirical question with a desk-sized first step.

The barrier wall is a separate, hormone-sensitive question

Efflux pumps in brain capillaries respond to estradiol within minutes in isolated rodent tissue. That is delivery biology in its own right, and it says nothing yet about whether any estrogen-related tracer is a substrate.

Two isolated-capillary studies make the point, and disagree on a detail. Hartz 2010 found that nanomolar estradiol reduced the transport activity of breast cancer resistance protein (BCRP) in rat and mouse brain capillaries within minutes, without new transcription or translation, reversibly, and through either estrogen receptor subtype . Banks 2024, in freshly isolated mouse capillaries, found a rapid decrease in Bcrp transport driven by extranuclear ERα and AMPK signalling: an ERα-selective agonist reproduced it, while ERβ- and GPER-selective agonists did not . The response was specific to Bcrp rather than P-glycoprotein or Mrp2 and separable from changes in transporter protein, ATP depletion or tight-junction leakiness, and females had lower baseline Bcrp expression, which ovariectomy normalised.

The receptor-subtype disagreement is real. The later authors attribute it to differences in experimental design, species, strain or hormonal history rather than to an error in either study. Both results share a practical shape: estrogen signalling can change active efflux at the barrier within minutes, whether or not the barrier has become leaky.

Banks 2024 also exposed capillaries to endocrine disruptors and receptor modulators. Bisphenols A and S lowered Bcrp transport, and tamoxifen increased it. That detail matters for imaging. A receptor modulator that changes a transporter has changed the delivery of every substrate that transporter handles. In a PET study, lower signal after such a drug could mean less binding or less delivery, and only a design that measures both can tell them apart .

In living people, efflux can be imaged with substrate tracers. van Assema 2012 scanned 35 healthy adults in three age groups with (R)-[¹¹C]verapamil, using a 60-minute dynamic acquisition and a metabolite-corrected arterial input, and read P-glycoprotein function from V_T under a constrained two-tissue model . V_T rose with age in men; young and older women were comparable, and young women had higher V_T than young men. The sex comparison rests on small groups, with four young, five middle-aged and seven older women. Premenopausal women were scanned during menstruation to limit hormonal variation, and hormone levels were not measured.

Salvi de Souza 2025 asked the sex question directly with [¹⁸F]MC225 in 20 adults aged 55 or older, 11 men and 9 women. The input was image-derived from the carotid arteries but calibrated against arterial samples and corrected for plasma-to-blood ratio and parent fraction, and V_T came from a reversible two-tissue model . Regional V_T did not differ between the sexes, with effect sizes below 0.2, and neither did blood concentrations, parent fractions or rate constants. The authors call for studies in younger people across hormonal cycles .

None of the studies reviewed here tested whether FES, or any other estrogen-related PET tracer, is a substrate of BCRP or P-glycoprotein . Chapter 1 notes that FES enters rat brain readily, which is not what a strongly efflux-limited molecule looks like. That argues against efflux as the dominant limit on brain FES signal without settling whether hormone-sensitive transport could modulate it, and the substrate question can be answered directly with standard transport assays.

A genotype that changes the vascular answer

In one mouse study, the cerebrovascular response to ovarian-hormone loss and estradiol depended on APOE genotype. That motivates testing for effect modification, not genotype-based prescribing.

Kehmeier 2026 studied young female mice homozygous for human APOE3 or APOE4, about six months old and fed a high-fat diet, which were left intact, ovariectomised, or ovariectomised and given estradiol, with 19 to 20 animals per group . In APOE3 mice, ovariectomy impaired endothelium-dependent dilation of the posterior cerebral artery and estradiol restored it; carotid passive stiffness rose with ovariectomy and fell with estradiol; and estradiol raised mitochondrial complex I respiration in cerebral arteries and arterioles. In APOE4 mice none of these endpoints responded to ovariectomy or estradiol. Not every outcome followed that pattern: the preprint version of the same study reports lower cerebral artery stiffness with estradiol in APOE4 but not APOE3 mice.

The authors close by suggesting that genotype may be a consideration when weighing hormone therapy . That is the step this report does not take. The animals were young, homozygous, on a high-fat diet and surgically ovariectomised, an abrupt loss unlike the menopausal transition (chapter 3), and the endpoints were vessel function, not imaging. What transfers is a design principle: if the vascular response to an endocrine change differs by subgroup, any vascular contribution to an imaging contrast may differ too, and the right analysis tests for that interaction instead of applying one correction to everyone .

What each study can carry

The vascular and efflux evidence spans three instruments in humans, two ex-vivo capillary preparations and one mouse vessel study. Each row answers a different question, and none is a measurement of an estrogen-related tracer's delivery.

The table also shows where the evidence is thin. Only one human study randomised a hormone, and it measured the instrument least able to see regional effects. The best-phenotyped study is associational. The transporter studies did not manipulate or even measure hormones. These are reasons to design the next study well, not reasons to dismiss what has been seen.

Does the endocrine contrast survive its delivery physiology?

The opportunity is not a universal perfusion correction. It is to ask, for one specific contrast and one tracer, whether the effect survives measurement of the delivery physiology that could produce it.

The evidence here gives no warrant for a blanket correction that attributes every hormone-related PET difference to perfusion, and vascular change does not invalidate molecular PET . A correctly modelled distribution volume is insensitive to flow by construction, a reference ratio can absorb a global change when target and reference share kinetics, and the perfusion associations found so far come from a natural cycle in young women, one observational study and one small randomised trial of a sequential regimen. Equally, nothing reviewed here shows that delivery is irrelevant for any particular estrogen-related contrast. The productive move is to make the question specific.

Three routes differ in cost and in what they can settle. The desk exercise needs no new participants, and its perfusion side could later be grounded in the public cycle-phase dataset once that has been inventoried . The paired design is the only one that lets delivery enter the model as a measured covariate rather than an assumption .

That framing keeps vascular physiology in its first role as well, as biology. Estrogen-sensitive efflux, genotype-dependent endothelial responses and cycle-linked perfusion are endocrine effects in their own right, and a collaboration could study them for their own sake. The next chapter turns from the vessel to the molecule: what happens when chemistry, rather than physiology, is the gate.

Five gates, and they do not substitute for one another

A usable brain tracer needs an exact chemical identity, adequate affinity and selectivity, enough intact unbound compound at the target, a signal that is demonstrably the target, and a way to turn that signal into a number. Passing one of these says nothing at all about the next.

The way this field usually talks about chemistry collapses the five into a single adjective. A compound is promising, or it failed. Both words destroy the information. Each gate is tested by a different experiment, in a different system, with a different unit, and a compound can sail through four and stop dead at the fifth — which is in fact the normal outcome, because the gates get harder in order and the evidence gets more expensive.

Gate one is identity: which molecule, which isomer, which deposited record. Gate two is pharmacology: affinity and selectivity, which are two numbers and not one. Gate three is exposure: how much intact parent, unbound, is present where the receptors are. Gate four is specificity: whether the signal in the brain is the target rather than everything else the isotope is attached to. Gate five is quantification: whether the signal can be turned into a reliable number, repeatedly, with an input function and a reference region someone has actually tested.

The figure has a deliberate shape: it empties out to the right. That is the chapter's main finding and it is worth stating plainly before the evidence. For brain estrogen-receptor imaging, no compound in the inspected literature has documented evidence at gate five, and the two compounds with the most informative published failures stopped at gates two and four respectively . Everything further right is programme, prediction or plan.

Gate one is bookkeeping, and the public record is not always clean

Before any comparison is meaningful, each compound has to be pinned to an exact structure with its stereochemistry, the assay it was measured in, the comparator that assay used and the species. This sounds like clerical work. It is where several published comparisons already fail.

The six structures below are drawn from their deposited PubChem records, parsed and laid out by RDKit. Nothing here is sketched from memory, and for each one the molecular formula and the hash of the structure computed from the record match what the database holds. They are not decoration: a reader who wants to understand why a 16α fluorine is tolerated while an 11β methoxy group changes a molecule's fate in a human body needs to see where those substituents sit.

Three readings of that plate matter for what follows. The first is how little separates the top row. 16α-fluoroestradiol differs from estradiol by one fluorine, and moxestrol adds a methoxy group at the 11β position and an ethynyl group at 17α to the same four-ring skeleton. These are small changes to a common core, and they produce entirely different fates in a human body.

The second reading is the bottom row, which is not steroidal at all. DPN is two phenol rings joined by a short chain with a nitrile; the analogue beside it adds one fluorine; erteberel closes a ring system onto a phenol. They reach the same pocket by a different route, which is why they can prefer one receptor subtype so strongly — and also why their absolute grip is weaker.

The third reading is the one the plate cannot show, and it is the chapter's first concrete finding. The fluorinated moxestrol analogue that produced the human imaging result discussed below is not drawn here, because its public record does not unambiguously state which isomer it is. The deposited entry's own synonym list contains both "16β-[¹⁸F]fluoromoxestrol" and a systematic name specifying the 16α configuration, and its computed name gives a single stereodescriptor at that carbon without resolving which convention produced it . The primary article is unambiguous — it studies the 16β isomer. The database record is not.

A second bookkeeping point is visible in the plate itself. Both diarylpropionitriles have one asymmetric carbon and both deposited records leave it unassigned . The synthesis in the source article introduces that centre by reduction and reports no separation of the resulting forms. The binding numbers quoted in the next section are therefore properties of a mixture, which is the honest way to read them and not a criticism of the work.

A ratio can rise because its denominator fell

The diarylpropionitrile series is the field's standard cautionary tale about subtype selectivity. Read closely it says something sharper than "selectivity is not enough": it shows a selectivity ratio improving while both affinities get worse.

Moon 2009 made nine new analogues of DPN, each carrying a small substituent at one ring position, and measured their affinity for purified human ERα and ERβ in a competitive assay against tritiated estradiol . The results are reported as relative binding affinity: estradiol is set to 100 and everything else is a per cent of it. In that same assay system, estradiol binds ERα with a dissociation constant of 0.2 nM and ERβ with 0.5 nM, which is the anchor that makes the per-cent figures interpretable at all.

Follow the fourth and fifth rows. Compound 1e carries a methyl group at one ring position: ERβ affinity 60% of estradiol, ERα 0.87%, a 69-fold preference — a genuinely attractive starting point. Moving that same methyl group one position along the ring gives compound 6a, and ERβ affinity collapses 36-fold to 1.67% . The selectivity ratio, meanwhile, doubles, to 139-fold. Nothing improved. The ERα affinity simply fell further than the ERβ affinity did, and the ratio obediently rose.

The most selective compound in the series, the meta-fluoro analogue 6d, reaches 272-fold — and holds 6.25% of estradiol's ERβ affinity, which is to say it binds its preferred target roughly sixteen times more weakly than the endogenous hormone does. The authors' own conclusion is the right one and they state it without hedging: despite the selectivities, the absolute affinities "are not good enough for them to be considered useful as potential PET or SPECT imaging agents" . They also note that labelling the most favourable compound with fluorine-18 would itself be difficult, because the fluorine sits on an aromatic ring next to a phenol rather than on an aliphatic chain.

This is the single most misread result in the chapter's source material, so it is worth being explicit about what it is. It is an in vitro candidate-prioritisation decision, made by chemists, on purified receptor protein, published as a negative structure–activity finding. It is not a brain imaging failure, a barrier failure or a human result of any kind . No animal was scanned. The compound stopped at gate two and the paper says so in its title's register: these are "selective ligands", evaluated and set aside.

One detail of the series deserves to survive into any future candidate list. The two analogues designed to be easy to label with fluorine-18 — the fluoroethyl and fluoropropyl compounds, where a radiolabel could be introduced by a simple aliphatic substitution — came out at 1.38% and 0.896% ERβ affinity with 58- and 45-fold selectivity . The compound that was easiest to make was the weakest; the compound with the best numbers was the hardest to label. That trade-off is structural, not incidental, and it reappears below in a different molecule's defluorination problem.

The tracer that worked in rats and not in women

A fluorinated moxestrol analogue beat every one of more than twenty candidate estrogens in the rat screen, went to people on the strength of that, and localised none of the three known receptor-positive tumours it was given the chance to find. The explanation turned out to be a protein that rats do not have.

Jonson 1999 reports both halves of the story . In the rat screen the fluorinated moxestrol analogue had high receptor affinity, low affinity for the plasma binding globulin, very high uterine uptake, excellent target-to-background selectivity and slow metabolism. Those are, item by item, the properties a tracer is supposed to have. In the human series, twelve women with suspicious breast findings were scanned. Seven had primary carcinoma, three of them with receptor-positive tumours by assay. Uptake above surrounding tissue was seen at one lesion, and the authors write that it may be an artefact of patient movement during the scan.

The second half of the paper is the part worth keeping. The authors took hepatocytes from immature rats, mature rats, baboons and a human donor, and measured how fast each cell population consumed each tracer. In immature rat cells — the exact screening system that had selected the compound — the established tracer was metabolised on average 31 times faster than its rival. In mature rat cells the advantage shrank to about threefold, and in baboon and human cells to about twofold . The compound's whole advantage was an artefact of the screening species.

Their proposed explanation is the sex hormone-binding globulin, which rats do not have. Relative to estradiol at 100, the established tracer's affinity for that globulin is 9.5 and the moxestrol analogue's is 0.037 — the low value it had been selected for . Adding purified human globulin to mature rat hepatocytes slowed the established tracer's consumption by 26% in a direct comparison, which is a measured effect rather than an argument . The free-hormone intuition says a heavily protein-bound ligand should reach its target less well. Here the opposite held: being bound protected the molecule from the liver long enough to be delivered.

Two boundaries keep this from being the barrier parable it is often used as. It is a breast-tumour imaging result: nothing in it tests entry into the brain, and the tumours in question sit outside the barrier entirely . And three known receptor-positive tumours is a very small denominator — the result is a clear signal that something was wrong, not a precise estimate of how wrong. What the paper establishes is the mechanism of a gate-four failure in one organ, plus a screening lesson of unusual generality: the species that selects your compound can be the species that disqualifies it.

A measured response is not a measured occupancy

Three drug programmes have taken receptor-β-preferring compounds into people. Their endpoints are a working-memory contrast on MRI, an enzyme activity in blood platelets, and the stiffness of an artery. Each is a real measurement. None of them is occupancy.

The most instructive is a phase 2 trial of erteberel in men with schizophrenia, registered as NCT01874756 . Ninety-five participants were randomised across placebo and three daily doses for eight weeks. The trial was terminated, and the registry states the reason plainly: funding ended. Results are posted, which is more than most terminated trials offer.

Among its six primary measures, one is listed as "Cortical Target Engagement". Its description defines it as a change in frontal-parietal activation during a working-memory task on MRI . The posted values are regression coefficients, and they rise slightly from baseline to week eight in every arm, placebo included, with standard deviations larger than the changes. Whatever that outcome measured, it is a downstream functional contrast several steps removed from a receptor, and calling it target engagement in a registry field makes a real measurement sound like a different and much stronger one .

Equol is the second case, and it has the clearest exposure data of any compound here — which is exactly why it is useful. Jackson 2011 reports that under 1% of the plasma total circulates as the unconjugated compound . That is a statement about chemical form: more than 99% has been conjugated by the body into something else carrying the same name in a total-concentration assay.

Those two words get used interchangeably and they are not the same measurement. The unbound fraction of the unconjugated 1% is the quantity that governs delivery, and nothing in the inspected sources reports it . Wilkins 2017 gave fifteen women with Alzheimer's disease two weeks of the compound and measured cytochrome oxidase activity in their blood platelets, with eleven of the fifteen meeting a response definition at a reported p below 0.06 . A single-arm peripheral enzyme response is a long way from a receptor in a brain, and the authors present it as a pilot.

The third programme is the largest and the most carefully built. Sekikawa 2025 reports the design of the ACE trial, which randomised 369 adults aged 65 to 85 to 10 mg a day of equol or placebo for two years across three sites. Its primary endpoint is the progression of arterial stiffness measured by pulse wave velocity; white-matter lesion volume on MRI and a cognitive composite are secondary. Checked on 2 October 2026, the registry record NCT05741060 read active but not recruiting, with no results posted and an estimated primary completion at the end of that month .

None of that is a criticism. It is a well-powered trial of a real clinical question. It is simply not evidence about receptor occupancy in the brain, and it was never designed to be. The design paper contains one discrepancy worth preserving rather than smoothing: its results text gives arm sizes of 186 and 185 while its abstract and its baseline table give 185 and 184, for the same total of 369 .

Read down the last column and the chapter's argument becomes a list rather than a mood. Four of these rows are missing the same measurement — intact, unbound compound where the receptors are — and two of them are missing it because nobody has looked rather than because anyone looked and failed. That is a considerably more encouraging situation than "the chemistry is hard", and it points at a specific experiment rather than at a feeling about difficulty .

Predictions are the cheapest part of the chain, and the easiest to over-read

A recent perspective from this literature lays out the candidate landscape in full and attaches a computed barrier-permeation call to each compound. The landscape is the useful part. The calls are predictions from structure alone, and the authors say so.

Arjmand 2023 is the most complete map of the problem available, and it begins by making a distinction the rest of this chapter has quietly assumed . Estrogen receptors are not only nuclear. The same proteins appear at the plasma membrane, trafficked there by lipid modification and association with membrane scaffolding, where they initiate signalling within seconds rather than the minutes-to-hours of the transcriptional route. A separate G protein-coupled receptor adds a third population. For an imaging problem this is awkward in a specific way: the membrane and nuclear forms are the same protein, so a ligand that distinguishes them cannot do it by recognising a different pocket.

Two numbers from that review set the difficulty. Membrane receptors are put at under 5% of the total receptor population . And across twenty studies of successful brain tracers for other targets, the available site concentration Bmax ran from 0.7 to 103 nM and the dissociation constant from 0.02 to 6.1 nM, with every one of those tracers holding a ratio between them above three . Set those side by side with a target expressed at low density across most of the brain, and the required affinity is not a matter of taste.

The review's candidate list is organised by what each compound could be labelled with. Four ERβ-preferring compounds are identified as directly labellable or labellable by halogen exchange; two ERα-selective pyrazoles would need a chelator and a metal; pathway-preferential estrogens are proposed for the membrane population; estrogen-dendrimer conjugates offer a size-based route to the same distinction, with the caveat that the conjugate changes shape with pH in a way that can hide the hormone entirely; and the two standard GPER tools are proposed as scaffolds, with existing technetium-99m complexes binding in the 10 to 30 nM range .

Four compounds survive both computed filters: DPN, ERB-041, WAY-200070 and one pathway-preferential estrogen. A newer non-steroidal scaffold not in that table, derived from a clinical degrader and labelled with fluorine-18, appears in Peng 2025; only an indexed abstract was read, so no structure, affinity or uptake value from it is carried into this chapter . It is worth pausing on the first of those. DPN is the same scaffold whose analogues were set aside at gate two for inadequate affinity — and here it passes a gate-three filter comfortably. That is not a contradiction. It is exactly the shape of the problem: the gates are independent, and a compound can be well-behaved at one and disqualified at another .

The review is careful about what its own table is, and that care should be carried forward rather than dropped in citation. The transporter in the second column is P-glycoprotein, and the authors write that experimental data are needed to corroborate whether these compounds are effluxed, and add a species warning: rodents show higher efflux transporter activity than pigs, primates and humans, so a disappointing rodent result may not transfer either . A predicted yes at gate three is a hypothesis about gate three. It is not evidence at gate three, and the distinction is the difference between a candidate list and a result.

One further warning in that review connects directly to the chemistry section above. A labelled analogue of ERB-041 has been made — Zhou 2023 — and it suffers moderate defluorination because its fluorine sits on an aliphatic carbon, with the freed label accumulating in bone . For most targets that is a tolerable nuisance. For a brain target it is close to disqualifying, because the bone in question is the skull, and it surrounds the tissue being measured. The easiest position to label is once again the least stable one.

A programme is a capability, and a scanner is a measurement condition

Two kinds of good news circulate in this area: that brain-specific ligand development is actively funded, and that new scanners are far more sensitive. Both are true, both are useful, and neither is a tracer.

The DEEPER project listing records a funded programme, running from July 2025 to June 2028, for the discovery and preclinical development of fluorine-18 ligands aimed specifically at brain estrogen receptors . It sits inside a larger funding programme whose wider portfolio of imaging and endocrine projects belongs to the next chapter. The DEEPER barrier work package makes one of its tasks concrete: brain endothelial cells on permeable supports, co-cultures with astrocytes and pericytes, and three-dimensional barrier assembloids, used to test whether candidate and radiolabelled compounds cross. That is the gate-three experiment, named and staffed.

What those documents establish is activity and intent on a stated timetable. They do not establish a candidate, a permeability result, a validated model, availability of anything, or anyone's willingness to work with anyone else . The honest summary is the useful one: the specific experiment this chapter keeps identifying as missing is being set up by someone, which is a reason to watch rather than a reason to wait.

The instrumentation argument needs the same discipline. Omidvari 2025 shows that a current brain-dedicated scanner holds quantitative accuracy at substantially reduced counts and can measure blood activity directly from the image. For a low-signal tracer that is genuinely enabling: shorter frames, smaller structures held apart, less reliance on arterial lines. It changes what is feasible.

Walk a compound through the gates yourself

The argument of this chapter is a bookkeeping argument, and bookkeeping is easier to believe when you can turn it over in your hands. Pick a compound and step it forward; the evidence at each gate is quoted as it was found, and the empty gates stay empty.

Four of the compounds below have a published stopping point; one is a lead read only at abstract level; one is a programme with no compound at all. The point of putting them in the same frame is not to rank them. It is that the frame has the same five columns for all of them, so the shape of what is missing becomes comparable even when the evidence is not.

Which candidate has the shortest remaining chain

The useful question is not which molecule is best. It is which one has the fewest missing measurements between its published position and a defensible brain signal — and that is answerable from the literature, before anyone synthesises anything.

The chapter's material supports a specific and cheap first move: a chemical translation dossier for one candidate. Fix the exact structure and stereochemistry from a primary source rather than a database synonym. Record the affinity with its assay system, its comparator and its species, and refuse to place it on a scale with affinities from anywhere else. Then ask, in order, what is known about intact parent in plasma, about the unbound fraction, about entry, about displaceable binding in tissue, and about whether anyone has ever measured the same thing twice .

Three design constraints follow from the cases above, and they are unusually concrete for a chapter about things that did not work. First, the screening species has to be chosen for the property that will decide the outcome: a rodent with no sex hormone-binding globulin cannot rank compounds whose fate depends on it, and a rodent with higher efflux activity than a human will reject compounds a human might have tolerated . Second, selectivity and affinity have to be reported as two numbers, because a ratio alone cannot distinguish a better compound from a worse one. Third, any claim about exposure needs to name its denominator: total or parent, conjugated or not, bound or free, plasma or cerebrospinal fluid or tissue.

And one measurement would do more than any other. For a single candidate with adequate affinity, run a baseline and blocked pair in the same animal while measuring intact parent and unbound fraction at the same time . That design answers gates three and four together, which is the junction every compound in this chapter failed to cross or was never taken to. It is also the point at which the barrier work being set up elsewhere and the quantification machinery of the earlier chapters would meet — the chemistry and the measurement problem are the same problem, approached from two ends.

The next chapter turns from molecules to material: the datasets, registries and access states that would make such a question answerable without first building a tracer.

A description, a dataset and a value are different things

A resource can be described in public, listed file by file, downloaded, opened and analysed. Each step is a separate achievement, and in this review no resource got past the second.

Every source in this report carries a reading depth: full text, partial reading, registry record, lead. Data resources need the same discipline with one more distinction. A paper's data statement is a description. A repository listing that names files, sizes and a licence is an inventory. Files on a local disk are acquired; files opened and checked are read; numbers computed from them are analysed. Each state licenses a different sentence, and the commonest error in the resources section of a review is to write the last sentence on the strength of the first.

The ladder below holds every data deposit this review found that bears directly on its questions. Read it from the left. Most lines stop at the second rung. Two reach partway into the third, because three small descriptor files were saved from each, none of them containing participant data . The two right-hand rungs are empty for every line. That is a statement about this review, not about the resources: no image, blood curve, expression value or participant record has been read or analysed anywhere in the report.

Three practical lessons come out of building that ladder, and they apply to any reuse plan. First, a text field is not an inventory. The animal COX-2 deposit's README repeats the human paper's abstract, so it reads as if it held 27 people; its own subject list holds twelve mice and six rats . Second, versions move. The human descriptor was first captured from a mutable branch, and a cached page showed an older version than the live record , so any analysis has to start from a pinned snapshot. Third, a listing is not a route: the reporter deposit's listing and README were readable on the web while two of its smallest files refused an automated download.

The open PET data belong to one tracer study

The two most complete open PET deposits are the data behind a single COX-2 validation paper. They are a resource for measurement methods, not extra cohorts for an estrogen question.

Yan 2025 asked whether the carbon-11 radioligand MC1 is sensitive enough to measure the low density of COX-2 in healthy human brain . Twenty-seven participants were scanned with concurrent arterial sampling: ten before and after celecoxib, a COX-2-selective inhibitor, and seventeen in a test–retest design. Celecoxib reduced neocortical binding by about a quarter, with little effect in subcortical regions and cerebellum, in line with regional COX-2 transcript levels. Test–retest reliability was moderate, an intraclass correlation of 0.65 with 20% absolute retest variability, and reference-tissue methods cut the variability . In humanized mice, MC1 bound human but not rodent COX-2.

The deposits are that study and nothing more. OpenNeuro ds004869 holds the human data: its snapshot 1.4.1 lists 27 subjects, sessions named baseline, blocked, test and retest, 278 files and 8.37 GB, under a CC0 dedication . OpenNeuro ds005605 holds the animal validation: snapshot 1.1.0 lists three wild-type and nine humanized COX-2 mice and six rats, three of them controls and three given an inflammatory challenge, in 65 PET files and 376 MB . Treating them as a 27-person cohort plus an 18-animal cohort for estrogen work would count one paper twice and ask it a question it never posed.

What is missing is endocrine annotation. The human participant schema declares sex, age and weight; the animal schema declares sex and age . Sex is recorded, so a sex-stratified methods check is conceivable. Sex is not endocrine state, though: nothing declared records cycle phase, menopausal status, contraception or a hormone measurement. That is a statement about two schema files, not proof that the study holds no such information .

The species result matters as much as the annotation. A tracer that binds human but not rat COX-2 cannot carry a standard ovariectomised-rat experiment, however good its human validation, and a humanized mouse tests binding to the human enzyme, not how that enzyme is normally regulated. Chapter 5 takes up what COX-2 imaging can and cannot say about inflammatory state; here the point is narrower. An open dataset's value is set by the question its design can answer, not by its size or licence .

Reporter, expression and docking deposits: orthogonal, unopened

The deposits that could link a ligand signal to receptor protein, transcription or cell type are small and openly licensed. Each comes with one specific boundary that decides what it can carry.

The optical reporter deposit accompanies Cara 2026, which built a neuronal reporter for estrogen-response dynamics . One fluorescent channel marks cells that were transduced and recombined; the other reports estrogen-responsive transcription. Neither is a calibrated measure of receptor abundance. The public listing gives 169.8 MB, with small quantification archives kept apart from the microscopy, and its README sets out exclusion rules. Some excluded animals remain in the deposited measurements, so any reuse has to reapply those rules rather than trust row counts . The listing displays a CC0 dedication; the photometry, the article states, is available on request rather than deposited. Thousands of cell rows are not thousands of animals.

The engineered-receptor monkey deposit belongs to Li 2024. It is a different resource, though both concern estrogen receptors and both reached this review through reporter work. Its public listing, rechecked on 2 October, shows eight archives totalling 2.52 GB under CC BY 4.0: MRI and PET for three animals and histology for two . No archive is named for blood or plasma data, so whether the arterial input of the native blocking phase is included cannot be known until one is opened . Chapter 1 explains why that phase, two animals with estradiol pre-treatment and full kinetics, is the best-controlled primate blocking study of native FES binding.

The sulfotransferase deposit from Miccoli 2026 is a single Origin project file of 2.34 MB under CC BY 4.0, holding the authors' processed docking and simulation outputs . It is tractable and open, and it contains predictions, not experimental binding constants. Reusing it cannot add an in-vivo contribution that the paper itself does not claim, and its proprietary format may need an export route before anything can be read.

GSE268557, a Gene Expression Omnibus series, deposits hippocampal arrays from 3xTg-AD mice after VCD-induced ovarian failure and from controls, at peri- and post-failure stages. Its 28 sample titles divide 8, 6, 6 and 8 across the four groups, not seven each . A 46.0 MB processed workbook and an 805.8 MB raw archive are listed and were not downloaded. Whether 28 arrays are 28 animals, and which publication they belong to, is unresolved; the record lists its citation as missing, which is not proof that none exists. Two further expression deposits are declared by their authors and unverified here: GSE288245, with analysis code, for the delayed-ovariectomy hypothalamus work of Bloom 2026, and GSE245831 with PRJNA1030000 for the microglial-gene study of Saha 2024 .

One declared deposit moved while this chapter was being written. Wright 2026 separated tissue perfusion from arrival time across the menstrual cycle, and its dataset identifier, which had failed to open earlier, resolved on 2 October to a public landing page: brain MRI and retinal angiography from 26 volunteers at three cycle phases, with oestrogen and progesterone samples, as a 6.7 GiB archive under CC BY 4.0 . The file tree was not inventoried, so the rung it reaches is a landing page, not a listing. The paper reports 21 MRI contributors among 26 recruits , exactly the kind of completeness question the archive would have to answer before anyone plans around it.

Controlled, conditional and withheld are three kinds of not open

Some of the most relevant human data are deliberately not public. The terms differ, and the differences decide what a first project can promise.

The GnRH-agonist trial of Frokjaer 2015 randomised 63 healthy women to a goserelin implant or placebo, with serotonin-transporter PET before and after; 60 completed . Its catalogue entry, PN000018, lists 63 participants, anatomical and PET data and 39.54 GB behind a Data User Agreement . It is controlled access to a known study, not a new cohort, and 63 listed participants are not 63 proven complete scan pairs. Its variables and visit completeness cannot be judged from the public metadata, and no access has been requested.

The Pill Project, NCT05212389, registers a randomised contraceptive-versus-placebo 5-HT4 PET design. Its data-sharing statement says the data will become available once the planned analyses are published, expected from December 2026, by application to the Cimbi database of molecular brain PET, and only with approval from the Danish data-protection authority and a signed agreement . That is a conditional future route, attached to a registry record last verified in February 2022.

Some data are not shareable at all. The participants in the human cetrozole series did not consent to public data sharing, as the methods paper states (Jonasson 2020) . A question about those scans would have to be asked of the investigators who hold them, as a methods collaboration rather than a request for a dataset. The reporter study's photometry sits in a fourth category: available on request.

Registries are designs with dates attached

Five registry records add designs, endpoints and denominators. Status fields go stale, and a plan is not a result.

MOSAIC deserves one more sentence, because it echoes another record in the table. Sponsored by the Université de Sherbrooke, the group that developed 4FMFES, it pairs that receptor tracer with carbon-11 acetoacetate and FDG, the same two fuels FEMME measured . Checked again on 2 October 2026, it is a plan with a passed start date. If it runs, it would put a receptor tracer and a fuel-choice measurement in the same women; until then it is a registry lead, not recruitment and not results.

Registry records are easy to misuse in both directions. FEMME's original composite outcome shows zero participants because its results were split into four other measures, not because no PET was done . Conversely, an enrolment target is not an analysed sample. Each record is useful for exactly what it fixes in advance: the contrast, the timing, the endpoint and the denominator.

What remains thin

This review was broad, bounded and uneven. Some branches were read against originals; others rest on abstracts or were never searched with precision. The map shows where reading went, not how much literature exists.

Depth followed the questions. Direct receptor imaging, aromatase binding, the rodent and human endocrine designs, AMPA imaging, the chemistry failures and the perfusion designs were read against original papers, often with their supplements. The 73-row human study map is a different kind of material: 47 of its rows rest on abstracts alone . The neuroimmune branches beyond TSPO rest largely on abstracts and indexed methods. Several categories were left thin on purpose, so that effort could go into reading originals rather than paging through search results.

Two facts about the searching belong beside that map. One precise search, checked against papers already known to be relevant, missed three of them; a broadened version recovered all three . And the broad database searches were not paged to the end: of 1,578 records captured, most were never screened, and those records are occurrences, not studies. Neither fact undermines what was read. Both mean the report cannot say that a study does not exist, only that one was not found or not read.

Four ways in

Four candidate starting points follow from the chapters and the resources. Each is a question with existing material, a missing requirement and a first deliverable. They are listed alphabetically, because ranking them would require knowing a collaborator's interests, methods and facilities.

None of the four needs a new scan to begin, and none presumes access to anyone's tracer, cohort or laboratory. They differ in the kind of first contribution they invite: an evidence synthesis that can be done at a desk, a reproducible analysis of data that are open or obtainable, or the design of an experiment that would come later. Each card names the chapters it draws on, so the spectrum the report took apart can be read back as material for a project.

An assay bridge starts from the distinction the first two chapters kept returning to: a ligand signal is not a protein amount, and neither is a transcriptional response. The reporter deposit, the engineered-receptor archives and the VCD arrays each measure a different one of those quantities, in different species and states. Laid side by side for one target, they would show where evidence pairs up and where it never has, and only then whether any deposit is worth opening .

A chemical translation dossier takes the opposite end. Chapter 7 showed that every documented failure stopped at an identifiable gate. The dossier walks one candidate through those gates on published material and ends not with a verdict on the molecule but with the single experiment that would move it furthest. Current programmes show that the capability is being built. Within the CARE programme, a departmental DEEPER project listing records fluorine-18 brain estrogen-receptor ligand development from July 2025 to June 2028 , and the DEEPER barrier work package specifies endothelial cultures, co-cultures with astrocytes and pericytes, and assembloids for permeability testing. Programme documents establish activity; they do not supply a candidate.

An endocrine design synthesis needs no data at all. Chapter 3 showed that natural cycles, abrupt surgical loss, gradual depletion, suppression and replacement are different experiments, and Chapter 4's map showed how few designs isolate estradiol. A table of what each study manipulated, when, and in whom, including which reports reuse the same participants, would turn that argument into an instrument others could use, and its output would be the two or three contrasts worth testing.

A measurement-method study is the one entry point with open data waiting at the other end. The human COX-2 deposit holds arterial input, blockade and test–retest in one place, which is rare. Used for what it is, it could put a number on how far a reference-region assumption moves a low-density target's outcome. The same question asked of existing FES or cetrozole scans would bear directly on the estrogen findings, but those data are held by their investigators and could only be approached as a partnership .

Build a first project

Choose an entry point to see which bands of the report and which resources it draws on, what state those resources are in, and what would have to happen before work could start.

The tool below is a reading aid, not a planner. It shows dependencies the chapters establish. It does not estimate cost, likelihood of success, or anyone's appetite for the work.

Whichever door a collaboration opens, the first conversation is the same, and it is about people rather than data: which of these questions interests the people involved, which methods and material they bring, and whether the first contribution should be a synthesis, a reproducible analysis or the design of a later experiment. None of that is assumed here, and no data holder or prospective collaborator has been contacted in preparing this report. The menu exists so that conversation can start from evidence rather than from a single attractive image.

The spectrum, reassembled

The field is promising because it contains several informative measurements, not because any one of them already captures estrogen's action in the brain.

Seven chapters decomposed one hormone-sensitive signal into its candidate explanations: receptor binding and the denominator it is divided by; synthesis and inactivation as separate targets; the endocrine trajectory that defines a contrast; downstream windows with their own clocks; neuroimmune measures that ask different questions; delivery as both biology and measurement condition; and chemistry, where translation breaks at identifiable gates. None of those bands is the whole signal. Each is a real measurement with a defined meaning, and that plurality is the field's strength.

The opportunity lies at the interfaces between them: endocrine timing and fuel use; local synthesis and receptor response; inflammatory cell state and tracer biology; hormone-sensitive delivery and target availability. A collaboration that connects a well-defined endocrine contrast to a validated molecular measurement, and checks it against an orthogonal assay, would be doing something this literature has not yet done .

Null results, failed tracers and unfinished source chains belong in that picture. The pituitary-only displacement, the peripheral blocking failure without target confirmation, the fluorinated moxestrol that worked in rats and not in women, the refused download from a reporter deposit and the registry that never updated are not obstacles to the story. Placed honestly, they mark where the next question is cheapest to ask.