What this is

Four podcasts have source packs prepared and are waiting to be built. This page is the reading list behind them: every paper, a working link, and an honest note on whether you can actually read it.

Copyright note: these are links, not copies. Where a paper is under publisher copyright, the link goes to the publisher or to PubMed Central, not to a reproduction. The two Mosconi papers are CC BY and could be redistributed, but linking is cleaner and always current.

The access finding worth knowing

While checking these, I found I had been calling several papers paywalled that are not.

"Open access" and "free to read" are different things. The flag that says a paper is not open access means it is not in the bulk text-mining subset. Many of those papers still have a full author manuscript sitting on PubMed Central that any human can read for free. That includes the two foundational hepcidin papers, the Dichtel depression trial, and the Zethraeus economics trial.

The practical rule: if there is a PMC number, click it before assuming you cannot read it.

---

Podcast 1: Brain estrogen receptor density after menopause

The strange finding, that receptor density is higher after menopause and higher density predicts worse memory.

What they contribute. The 2024 paper is the one with the counterintuitive result: brain estrogen receptor density is higher after menopause in estrogen-regulated networks, and within the postmenopausal group, higher density predicted worse memory and more self-reported symptoms. The 2021 paper establishes that menopause status rather than chronological age tracks the brain changes, shows partial gray matter recovery after menopause, and reports greater amyloid deposition specifically in APOE4-carrying peri- and postmenopausal women versus genotype-matched men.

What they do not contribute, and the episode says so explicitly. Neither measures dopamine. The tracers are estrogen receptor, glucose metabolism and amyloid. A great deal of popular menopause content borrows the authority of this imaging work and attaches it to a dopamine story it never tested.

Retrieval note: both publisher PDFs and the PMC PDFs refused automated download, and the NotebookLM URL importer was blocked by the publisher. Full text came through the Europe PMC API instead. Reading them in a browser works fine.

---

Podcast 2: Testosterone and reward anticipation in women

The one genuine human, female, placebo-controlled result, and its limits.

What it contributes. Twelve healthy women, single sublingual dose of 0.5 mg testosterone, placebo-controlled double-blind crossover, monetary incentive delay task. Ventral striatal response to reward-anticipation cues increased, and the effect was largest in the women who started with the least intrinsic appetitive motivation.

Why it needs careful handling. BOLD is a hemodynamic proxy for aggregate local activity. It cannot distinguish more dopamine released, greater receptor sensitivity, changed reuptake, altered cortical input, or a purely downstream change. The paper is routinely cited as evidence testosterone raises dopamine in women. It is not that. Twelve women, one acute dose, young healthy volunteers, never replicated at scale.

This is the only genuinely inaccessible paper in the set. The source pack was built from the freely available abstract plus context, which is the documented route for a paywalled source.

---

Podcast 3: Testosterone, hepcidin and the iron paradox

Correcting the claim that testosterone makes the body hoard iron. The opposite happens.

The three that matter most are free to read. The 2010 paper established hepcidin suppression in men. The 2014 paper is the one that reframes the mechanism as a reset of the erythropoietin-to-hemoglobin set point rather than a simple iron effect. The 2013 paper is where the downstream direction is established: ferritin falls, iron incorporation into red cells rises.

The correction the episode is built around. Testosterone does suppress hepcidin, and hepcidin does gate iron absorption. But testosterone simultaneously expands erythropoiesis, and red cell production is iron-hungry. The demand wins. Net flux is out of storage and into hemoglobin, so ferritin falls, soluble transferrin receptor rises, and functional iron stores are reduced. Iron is mobilized and consumed, not hoarded. In iron-deficient mice, testosterone made anemia worse.

Two nuances the episode keeps. Hepcidin is not the master switch: hepcidin-knockout mice and liver-specific androgen-receptor-knockout mice still raise hematocrit on testosterone. And direct bone marrow stimulation is not the mechanism either, since the erythroid effect requires androgen receptor signaling in non-hematopoietic cells.

The gap the episode names out loud. All human hepcidin work is in men, at male doses. No human study has established hepcidin suppression in women at female-physiologic transdermal doses.

---

Podcast 4: Testosterone undecanoate, and the paradox that isn't one

Researched. The interesting finding is that the puzzle dissolves rather than resolving.

The pharmacology: why undecanoate is genuinely the stable one

The mechanism, and the part nobody mentions. Every injectable ester is a prodrug whose release rate is governed by how strongly it partitions into the oil depot rather than the surrounding water. Undecanoate is an eleven-carbon chain against cypionate's eight and enanthate's seven, so it stays in the depot far longer. Absorption, not metabolism, is rate-limiting.

But the vehicle does as much work as the ester. The same 1000 mg dose of the same drug, given in two different oils, behaved as two different drugs: terminal half-life 20.9 days in tea seed oil versus 33.9 days in castor oil. A 60 percent extension from the vehicle alone. Castor oil is more viscous, carries ricinoleic acid, and dissolves undecanoate at twice the concentration, which halves the injection volume and the depot surface-to-volume ratio.

The clean head-to-head is in monkeys: enanthate overshoots to 100 to 177 nmol/L and empties within a month; undecanoate peaks lower at 58 nmol/L and then plateaus at 40 to 68 for 45 days.

Oral versus injectable undecanoate: not the same drug

Oral undecanoate has about 3 percent bioavailability, and over 90 percent of what does get through arrives via intestinal lymphatics rather than the bloodstream, which makes absorption dependent on dietary fat. Roughly 97 percent of an oral dose never becomes systemic drug.

The number that makes this concrete comes from 200 postmenopausal women on 40 mg/day for four weeks: the achieved increase ranged from zero to 10.15 nmol/L. At the bottom, a woman taking the drug daily for a month absorbed nothing measurable. At the top, another gained a mid-normal male increment. Same drug, same dose, same trial. That is not a formulation anyone can titrate.

The puzzle, and why it dissolves

Three corrections to the premise, all from the source paper itself:

1. The comparator was not cypionate or enanthate. It was a European mixed-ester product injected every 2 to 4 weeks, whose components span three to ten carbons. Part of the dose is not short-acting at all.
2. The finding rests on 345 of 5528 measurements, and 38 percent of measurements had an unknown route, with that unknown category carrying its own significant odds ratio of 1.9.
3. Duration of therapy was listed as a determinant and then omitted from the adjustment model, while the same paper shows cumulative erythrocytosis risk climbing from 8 percent at one year to 38 percent at ten and 50 percent at fourteen. Undecanoate only reached market partway through the observation window, so anyone observed on it had already been on testosterone for years. The formulation coefficient is structurally positioned to absorb a duration effect.

And the "signal disappeared at the higher threshold" observation is not evidence of anything. At hematocrit above 0.52 the odds ratio was 1.0 with a confidence interval of 0.3 to 3.4, which contains the 3.1 seen at the lower threshold. That is a power failure being read as a null result.

The finding is contradicted on this exact comparison, three times over:

• A prospective study at the same center found trans men on undecanoate were less likely to reach hematocrit of 50 or 52 percent than those on esters.
• A Bayesian network meta-analysis of 29 randomized trials in 3,393 men, which separated long from short-acting injectables, put intramuscular undecanoate at 1.6 percent hematocrit rise versus 4.0 percent for enanthate and cypionate.
• The systematic review devoted to this question in trans men reached the same direction.

What does replicate is a different comparison: injectable undecanoate versus gel (odds ratio 2.9 in the cohort; 22.7 versus 5.0 percent in an 802-man registry). The simplest reading is not ester chemistry. A 1000 mg depot delivers considerably more androgen than a daily gel, and erythropoiesis integrates exposure over months.

The mechanism settles it. Testosterone raises hematocrit by increasing erythropoietin, resetting the erythropoietin-to-hemoglobin set point, and suppressing hepcidin transcription. Transdermal gel does this with no injection peak whatsoever. Peaks are not the mechanism, so a peak-based explanation was never required.

Why this still deserves an episode

Not as a paradox, but as a case study in how a fragile subgroup finding becomes received wisdom. The claim traveled because it was surprising. The correction requires holding six things at once: what the comparator actually was, how many measurements supported it, which covariate was missing, what a confidence interval containing the rival estimate means, what the prospective and randomized evidence says, and what the mechanism predicts.

And for women specifically, the guidance holds but not for the reason people assume. The smallest injectable undecanoate unit is 750 mg, roughly 7 mg of testosterone per day released over three months, which is one to two orders of magnitude above a female physiologic dose, with no fractionable product. Stability does not help if the plateau itself is male-range. And the 33.9-day half-life makes an overdose worse, not better: the consensus mandates a repeat level at 3 to 6 weeks precisely so dosing can be corrected, and on undecanoate that measurement arrives when nothing can be done about it.

---

Podcast 5: The implant you cannot switch off

Why hormone pellets are a different clinical object from a gel or a patch, and how that turns a side effect into an operation.

The pharmacokinetics: exposure stacks

What they establish. The audit found a median 223.5 days between implants against an estimated median 311 days to return to baseline, so each implant is placed while the previous one is still releasing and exposure compounds. The 1982 study found estradiol still above pretreatment values more than 68 weeks after estradiol-plus-testosterone implants. The tachyphylaxis paper found 3 percent of 1,388 women on implants had estradiol above 477 pg/mL. And the dosing study, from a pellet proponent's own series, found a 41.9 percent coefficient of variation in achieved testosterone under identical dosing, which is the reproducibility problem stated by an advocate rather than a critic.

The consequence: bleeding nobody can stop

The 1990 BMJ paper is the one to read, and it is free. It documents endometrial stimulation continuing a mean of 35 months after the final implant, with a range extending to 43 months, and reports a patient who required hysterectomy 26 months after her last implantation because of persistent irregular bleeding despite high-dose progestogen. That single case is the whole mechanism in one patient: bleeding that cannot be stopped because the hormone source cannot be withdrawn.

The 2003 randomized trial is the controlled corroboration: of 40 women randomized, nine in the estradiol arm had vaginal bleeding and five had a hysterectomy during or after the study, and the authors concluded the implant had a high complication rate.

The 2007 study is the combined estradiol-plus-testosterone implant data: of 258 women on two years of continuous therapy, 17.1 percent had endometrial thickening above 5 mm, and of those biopsied, 20.4 percent showed simple hyperplasia. Note the denominator carefully, since 20.4 percent is 9 of 44 biopsied, which is 3.5 percent of the cohort rather than 20 percent of women on implants.

The other half: progestogen that may not be protecting anything

The systematic review states it as a formal recommendation: transdermal micronized progesterone does not provide endometrial protection. Only oral micronized progesterone at 200 mg daily for 12 to 14 days per month is endorsed. The randomized histologic study behind that is the 2000 paper, which found transdermal progesterone at 16, 32 or 64 mg daily produced levels insufficient to induce any detectable endometrial change.

The trap inside that finding, and it is the most important sentence in this section: only one patient bled during that trial. Absence of bleeding on a progesterone cream is not evidence that the endometrium is protected.

The compounding paper is free to read and quantifies the potency problem. Identical prescriptions filled by 13 compounding pharmacies yielded progesterone capsules ranging from 90.8 to 135 mg against a 100 mg label, and estradiol capsules from 0.365 to 0.551 mg against a 0.5 mg label, against a pharmacopeia tolerance of plus or minus 10 percent.

Why this is an episode rather than a footnote

The chain is fully documented and each link has a citation, but no single study contains the whole thing. That is exactly what a synthesis episode is for.

Supraphysiologic estradiol goes in, in a delivery system that cannot be dose-reduced or removed and whose exposure stacks because reinsertion outpaces clearance. Endometrial protection may be a compounded preparation of uncertain potency, or a cream that the randomized evidence says does not protect at all, with no bleeding to warn anyone. The endometrium proliferates. Bleeding starts. In a postmenopausal woman every bleed mandates evaluation for cancer, and roughly 9 percent of postmenopausal bleeding turns out to be endometrial cancer, so the workup is not optional. The bleeding recurs because the hormone source is still releasing and no clinician can switch it off. Eventually somebody removes the organ.

And the counterintuitive coda: testosterone is not the villain in this story. A randomized trial in 63 postmenopausal women found endometrial thickness unaltered by testosterone alone, and concluded testosterone appears to partially counteract estrogen-induced proliferation. The pellets in the cohort study that raised the alarm contained estradiol and testosterone together, and it is the estradiol that does this.

---

Podcast 6: What dose actually works, and how you would know

The dose-response thread. This is the one that lets you judge every other testosterone claim, because it establishes the benchmark exposure against which any null result has to be read.

Why APHRODITE is the anchor. It is the trial that established the dose that works. The 300 microgram per day patch produced 2.1 satisfying sexual episodes per four weeks versus 0.7 on placebo at week 24. The 150 microgram arm missed at 1.2 events, p = 0.11. Two doses, one worked and one did not, in the same trial, which is what makes it a dose-response result rather than a single data point. It also showed the effect holds in women not taking estrogen, which is why it matters more than the earlier surgical-menopause trials.

The workaround for the paywall, and it is a good one. The achieved hormone concentrations live in the Supplementary Appendix, which is not openly accessible. But the EMA Summary of Product Characteristics for the same 300 microgram patch is free and contains the pooled clinical-program levels, which is arguably the better source anyway because it pools the whole program rather than one trial:

What that benchmark does to the other trials, and this is the payoff of the episode. The Dichtel depression trial is routinely dismissed as underdosed. Set the numbers side by side and it is not:

Dichtel produced roughly 30 percent higher total testosterone than the dose that was efficacious for libido, from a nearly identical baseline, using LC-MS/MS for total and equilibrium dialysis for free. Anyone waving that null away as timid dosing has to explain 105 ng/dL.

One caution the episode must include. Do not compare the free testosterone columns across those two trials directly. The roughly threefold difference in apparent free fraction is an assay artifact between equilibrium dialysis at one laboratory and the methods used in the Intrinsa program, not a biological difference. Total testosterone is the only measure comparable across labs here, which is itself a useful lesson about reading hormone data.

And the regulatory epilogue. Intrinsa held a European marketing authorization and it was subsequently withdrawn. There is no FDA-approved testosterone product for women in the United States today, so the dose with the best trial evidence behind it is not a dose anyone can be prescribed as an approved product.

---

Related papers referenced across the fact-check work

Not podcast sources, but frequently cited in the surrounding research and worth having linked.

---

Access summary

Eleven of twenty-six are readable right now without paying anything, and in each podcast set the single most load-bearing paper is among them. For episode 5 that is the 1990 BMJ paper, which contains both the 43-month endometrial stimulation finding and the 26-month hysterectomy case.

Only the Hermans paper is genuinely unavailable in a set where nothing else can substitute, and its abstract carries enough for the episode.

If you want the paywalled ones

Three routes that are legitimate and usually work:

1. Email the corresponding author. Authors are almost always glad to send a reprint, and it is entirely permitted.
2. A local university library. Many offer community borrower access, and interlibrary loan covers single articles.
3. Check for a preprint or an author's institutional repository copy, which many publishers permit after an embargo.

Status of the podcasts themselves

All four source packs are prepared, each with the papers plus an editorial brief that sets the angle, the tone, and the claims the episode must not make. None of the four notebooks is built. The NotebookLM automation failed repeatedly in the session where these were prepared: the publisher blocked the URL importer, PubMed Central returned a bot challenge that got imported as a source, and the add-source dialog stalled. Building them is roughly ninety seconds of clicking each, in a fresh session.