Research sweep · 25 July 2026
Genetic variation in estrogen, progesterone and testosterone
How individual genetic differences affect sex-hormone metabolism and response, in women and in female laboratory animals. Seven parallel research threads run against PubMed, scite and Consensus. 73 sources banked, every DOI CrossRef-verified and screened for retractions.
The finding: the measurement problems are bigger than the genetics, and the route of administration is bigger than both. This held across all seven threads independently. Almost no genotype test is justified today. But two measurement gaps are real, quantified and fixable this week, and the single variable that most changes how much testosterone reaches a woman's receptors is the route of her estrogen.
What is actionable now
1. Oral estrogen works against testosterone
In a randomized crossover, oral conjugated estrogen raised SHBG by 132% and cut free testosterone by 32.7%. Transdermal estradiol moved SHBG 12.0% and free testosterone by 1.0% (Shifren 2007). The SHBG split was confirmed in 727 randomized women (Harman 2014).
Free testosterone on oral estrogen. A woman taking testosterone alongside oral estrogen can hold a steady or rising total testosterone while her free testosterone falls by a third. Nothing in seven threads of genetics comes close to that effect size.
One honest qualifier: a small three-arm study (n = 15 per arm) found SHBG rising in all groups including transdermal (Nii 2016). "Transdermal has a far smaller effect" is defensible. "Transdermal has zero effect" is not.
2. The androgens nobody measures
Adrenal 11-oxygenated androgens (11-ketotestosterone and relatives) are quantitatively major in women, age-stable, and invisible to standard panels. Not measured badly. Never measured.
- Median 11-ketoandrostenedione 7.99 nmol/L against testosterone 0.34 nmol/L in 588 healthy women (Skiba 2019). In PCOS they are 53% of the entire androgen pool (O'Reilly 2017).
- Statistically flat across ages 18 to 97 in women while classic testosterone declines (Davio 2020).
- Adrenal, not ovarian, proven by both adrenal and ovarian vein sampling (Turcu 2016; Auer 2022). Menopause does not touch this pool.
The clinical proof: women with Cushing's disease and obvious androgen excess had testosterone, androstenedione and DHEA-S comparable to healthy controls while their 11-oxygenated androgens were markedly elevated (Nowotny 2022). "Your testosterone is normal" was true and irrelevant.
3. Assay quality outranks every gene here
Direct immunoassay read serum testosterone more than 100% higher than mass spectrometry in the same women (Handelsman 2016). The Endocrine Society concluded lab proficiency was being judged by agreement between labs rather than accuracy (Rosner 2007). And calculated free testosterone rests on binding models its own reviewers call faulty (Keevil 2019). Insist on mass spectrometry. Measure SHBG rather than genotyping it.
4. Route also defuses the one real gene-drug interaction
| Group | Odds ratio for venous thrombosis |
|---|---|
| No mutation, no estrogen | 1.0 (reference) |
| Factor V Leiden, no estrogen | 4.1 (2.3 to 7.4) |
| Factor V Leiden + oral estrogen | 25.0 (6.9 to 95.0) |
| Factor V Leiden + transdermal | 4.4 (2.0 to 9.9) |
Read the last two rows together (Straczek 2005). Transdermal adds essentially nothing on top of the mutation itself. In absolute terms, because relative risk misleads here: WHI found 3.5 events per 1,000 woman-years on oral hormone therapy versus 1.7 on placebo, roughly 1.8 extra events per 1,000 women per year (Cushman 2004).
Only Factor V Leiden modified the risk. Prothrombin G20210A, MTHFR, factor XIII and PAI-1 were all null, so a routine thrombophilia panel is not supported. The Dutch Pharmacogenetics Working Group went further and withdrew its Factor V Leiden recommendations entirely in 2023, citing insufficient consistency about what the advice should be.
What failed, and should not be repeated
- Aromatase inhibitor genotyping. A purpose-built prospective study of 970 women tested 10 previously reported SNPs. None predicted symptoms or discontinuation, in any subgroup (Stearns 2024). The mechanism was separately refuted (Lunardi 2013).
- CYP2D6 and tamoxifen, the field's cautionary tale. In 4,393 phenotyped women there was no association with recurrence, and poor metabolizers had more hot flushes, the opposite of the hypothesis (Regan 2012). NCCN and ASCO recommend against testing.
- ESR1 PvuII and XbaI. Standardized meta-analysis of 18,917 people: no effect on bone density (Ioannidis 2004). In the coronary meta-analysis, heterogeneity was "largely explained by a quality score" and "studies that reported significant associations were generally of poorer quality" (Lluís-Ganella 2009).
- COMT as a guide to estrogen and cognition. Tested directly in 118 postmenopausal women: no main effects, no performance difference (Dumas 2018). The largest test of the underlying link, roughly 2,000 to 2,600 people, was flatly null (Wardle 2013).
- Urinary 2-OH/16-OH ratios as a tissue proxy. No 4-hydroxy metabolites were detectable in breast tissue while measurable in urine (Zhu 2010). Only nine women, but a real reason to distrust the commercial panels built on this.
The androgen receptor, and why women are the hard case
Longer CAG repeats lower receptor activity (Chamberlain 1994). But the gene is X-linked. Men have one copy. Women have two, one randomly silenced per cell, making every woman a mosaic of short-receptor and long-receptor cells. The meaningful measure is the X-weighted mean, and most published studies never computed it.
Worse for a menopause audience, skewing increases with age: severe skewing in 7% of women under 25 versus 16% over 60, with agreement between blood, cheek and urinary cells falling markedly (Sharp 2000). Sixteen percent of women in their 70s and 80s newly developed skewing across a single decade (Mengel-From 2021). A blood-based receptor genotype in an older woman is partly measuring aging blood.
The largest female study (n = 529, validated instruments, properly X-weighted) found longer repeats went with better sexual function and found nothing at all for desire (Wåhlin-Jacobsen 2018), the opposite of what the simple model predicts.
APOE4 and hormone therapy
No randomized trial has ever been designed to test this. Every APOE4 result in the literature is a secondary, covariate or post hoc analysis inside a trial powered for something else. KEEPS-Cog pre-specified APOE4 as a moderator and found nothing (Gleason 2015). WHIMS, the trial that produced the harm signal, never published an APOE4-by-treatment interaction on dementia despite genotyping thousands of women. That hole has stood for 22 years.
Two corrections to widely repeated claims. First, "APOE4 is worse for women" is narrower than usually stated: in ~58,000 people, ε3/ε4 odds were essentially identical in men and women across ages 55 to 85, with the female excess appearing only at ages 65 to 75, and explicitly "no significant differences between men and women with ε4/ε4" (Neu 2017). Second, the timing hypothesis failed its own randomized test: verbal memory difference −0.06 SD, timing interaction p = 0.88 (Henderson 2016). The defensible reading is not "early is protective" but "early is neutral, late is harmful."
For ε4/ε4 specifically, nobody knows. Not one trial reports a homozygote subgroup count. KEEPS genotyped 568 women at 14.4% ε4 frequency, which predicts roughly 12 homozygotes across three arms, about four per arm. No cognitive endpoint is estimable from that. And testosterone crossed with APOE4 in women has zero studies: a 2025 systematic review screened 5,914 records covering 1,016,055 participants and states outright that no included study examined testosterone (Melville 2025).
Why the female animal data may not transfer
- Rodents lack circulating SHBG after birth. Every lab mouse is a permanent "SHBG equals zero" human. Adding human SHBG as a transgene raised total hormones, left free testosterone unchanged, and reduced bioactivity (Laurent 2016). A named failure: low SHBG strongly predicts human diabetes, but human SHBG did not protect mice from diet-induced dysglycemia (Saez-Lopez 2017).
- Ovariectomy is not menopause. ACI rats given estradiol with ovaries intact reached 100% mammary carcinoma incidence within 197 days. Ovariectomized rats at matched serum estradiol developed zero tumors (Shull 1997).
- Strain is a hidden dose modifier. Sensitivity to estradiol varies more than 16-fold between mouse strains (Spearow 1999). A null result may be a strain result.
- The reason given for excluding female animals was false. Across 293 mouse articles and 311 rat articles, female animals were not more variable than males, and males were substantially more variable on several endpoints (Prendergast 2014; Becker 2016).
Gaps, recorded so nothing gets overstated
No primary source found for the estrous-versus-menstrual-cycle translational comparison. The commonly cited aromatase-knockout adiposity phenotype was not retrieved and should not be asserted. Cyp1b1-null established carcinogen activation, not the estradiol 4-hydroxylation phenotype the estrogen-genotoxicity story needs. The Sult1e1-null published phenotype is male. No support was found for the widely repeated COMT-estrogen-mood link in women. SRD5A2 A49T has no female phenotype data, so claims tying 5-alpha reductase variants to female-pattern hair loss are unsupported. And no GWAS of the 11-oxygenated androgen pool exists in either sex.