Industrialization increases the estrogen-recycling capacity of the gut microbiome.

Population ICC (intraclass correlation; equation and interpretation in Materials and Methods ), which measures how much variation occurs between versus within populations, ranged from 0.14 to 0.36 for estrobolome abundances across studies ( SI Appendix, Fig. S1 A – C ), indicating that approximately 14 to 36% of the variation is attributable to differences between populations, with the remaining 64 to 86% occurring within populations. Here estrobolome abundance refers to the proportion of amplicon sequence variants (ASVs) predicted to encode β-glucuronidase function (details provided in Materials and Methods ). Population ICC scores for estrobolome α-diversity ranged from 0.23 to 0.35, suggesting that 23 to 35% of the variation in α-diversity is due to differences between populations, while the remaining 65 to 77% of variation occurs within populations ( SI Appendix, Fig. S2 A – C ). These results show that although most estrobolome variation exists among individuals within populations, meaningful differences also exist between populations. This has important implications for both personalized and population-level understandings of estrogen metabolism. That is, the estrobolome is shaped by larger-scale ecological and lifestyle contexts in addition to individual differences. As subsistence and industrialization represent major axes of lifestyle divergence across the studied populations, we next test whether various subsistence types influence estrobolome characteristics. Pronounced differences in estrobolome characteristics were observed between industrial and nonindustrial populations, although there were no consistent differences among nonindustrial groups. Across all three datasets, industrial agriculture populations exhibited higher estrobolome abundances than nonindustrial subsistence groups ( Fig. 3 A – C and Dataset S1 A ). Estrobolome abundance estimates were 4 to 7 times higher ( Fig. 3 A ), 1.6 to 2.2 times higher ( Fig. 3 B ), and 3.6 times higher ( Fig. 3 C ) in industrialized populations compared to those from nonindustrial groups. By contrast, nonindustrialized groups did not differ significantly. Estrobolome α-diversity showed a similar trend, with industrialized populations exhibiting 1.3- to 2.3-fold higher α-diversity than nonindustrialized groups ( Fig. 3 E and F and Dataset S1 B ). Supplementary models confirmed these patterns using both industrial agriculture reference groups ( SI Appendix , Fig. S3 ). Only one within–nonindustrial group difference reached significance, with agropastoralists exhibiting 1.47-fold higher α-diversity than hunter-gatherers in a single dataset ( Fig. 3 D ). Subsistence-based comparisons of estrobolome relative abundances ( A – C ), α-diversity ( D – F ), and ASV composition ( G and H ) across three datasets ( 19 , 38 , 39 ). Error bars represent 95% CI. All P -values are Benjamini–Hochberg adjusted. In panels G and H , industrial agriculture subsistence was used as the reference group, and coefficient directions (yellow = positive, blue = negative) are relative to the reference group (e.g., blue means a decrease relative to the reference group). Estrobolome composition varied by subsistence type in two of the three datasets. No significant differences were detected in the Botswana/Tanzania/USA dataset. In the Malawi/USA/Venezuela dataset, industrialized estrobolomes were enriched in ASVs assigned to the Bacteroides genus ( Fig. 3 G and Dataset S1 C ), while nonindustrialized estrobolomes were enriched in Victivallis , Succinivibrio , one ASV in the Enterobacteriaceae family, and multiple ASVs from the Lachnospiraceae family ( Fig. 3 G ). In the Nepal/USA dataset, nonindustrialized groups exhibited elevated Escherichia–Shigella abundance ( Fig. 3 H ). These findings support our hypothesis that estrobolome function differs between industrialized and nonindustrialized populations, but not among nonindustrial subsistence types. Elevated estrobolome-recycling capacity may reduce fecal estrogen clearance and increase systemic exposure, potentially contributing to heightened risks of estrogen-linked conditions such as breast cancer, endometriosis, metabolic syndrome, and infertility ( 2 , 40 – 42 ). This may be especially consequential in industrialized populations, where women exhibit higher sex hormone levels compared to nonindustrialized groups ( 1 ), and men show similarly higher testosterone and adiposity ( 35 , 37 ). Amplified microbial estrogen recycling in these contexts may exacerbate reproductive and metabolic hormonal health problems. From an evolutionary and life history perspective, such shifts in systemic hormone exposure may alter growth trajectories, age at reproductive maturation, fertility patterns, and patterns of reproductive aging, with downstream consequences for population-level health and disease risk. Contrary to predictions based on total gut microbiome diversity being lower in industrialized populations ( 17 ), industrialized populations exhibited higher estrobolome diversity. This may reflect modern dietary shifts toward Western diets, which experimental models suggest can drive multigenerational, diet-induced extinctions of microbial taxa due to reduced intake of microbiota-accessible carbohydrates (MACs) found in dietary fiber ( 43 ). Industrialized microbiomes are thought to be especially vulnerable to taxonomic and functional losses ( 17 ), which may open ecological niches or reduce competition, allowing surviving microbes to diversify and increase in abundance. The estrobolome might represent one such opportunistic expansion, possibly contributing to the elevated prevalence of estrogen-linked diseases in industrialized societies. Moreover, these shifts may also exemplify an evolutionary mismatch, wherein modern lifestyles select for microbial configurations that are poorly aligned with current human biology ( 18 ). Several studies have linked diet to fecal β-glucuronidase activity, which may help explain the elevated estrobolome capacity observed in industrialized populations. In humans, higher β-glucuronidase activity is associated with diets high in protein and fat, while fiber intake is linked to reduced activity ( 44 – 47 ). Animal studies support these findings. Rats fed fiber-free diets have significantly higher β-glucuronidase activity than those consuming high-fiber diets ( 48 ), and supplementation with chitosan or its oligosaccharides reduces β-glucuronidase activity in rats on high-fat diets ( 49 ). In a controlled human trial, only a high-fiber, high-fat diet lowered fecal β-glucuronidase activity, suggesting that fiber may play a more important role than fat reduction alone ( 50 ). Probiotic interventions such as Lactobacilli or yogurt have been reported to decrease activity in some studies ( 51 , 52 ), although others found no effect ( 53 ). These findings are consistent with our results, given that industrialized agriculturalist populations typically consume Western diets that are low in fiber and high in fat and sugar. We highlight that “industrialization” is a broad term encompassing multiple interrelated aspects of a society. Other potential contributing factors common in industrialized populations include reduced physical activity, increased exposure to environmental toxins, improved sanitation, access to healthcare systems, and lifestyle features linked to urbanization and economic development. A combination of these factors may underlie the elevated estrobolome abundance and diversity observed in industrialized populations. Antibiotic use may also contribute, although prior work suggests it suppresses estrobolome function ( 54 ), whereas we observed enhanced function in industrial groups. Further research is needed to determine which features of industrialization contribute to estrobolome variation and what their relative contributions are. In the only dataset that included both breastfed and formula-fed infants (Malawi/USA/Venezuela), estrobolome-associated ASV relative abundances were markedly higher in formula-fed infants, approximately two- to threefold greater than in breastfed infants and higher than those observed in breastfed infants even within industrialized populations ( Fig. 4 A and Dataset S2 A ), and exceeded those of breastfed infants even within industrialized populations. Estrobolome α-diversity was also substantially greater, with formula-fed infants exhibiting diversity 2.5 to 11 times higher than those of breastfed infants ( Fig. 4 B and Dataset S2 B ). Additionally, three estrobolome-associated ASVs differed significantly by feeding mode ( Fig. 4 C and Dataset S2 C ). Bacteroides was more abundant in formula-fed infants compared to breastfed infants from rural farming and market agriculture communities. The Ruminococcus gnavus group was also more abundant in formula-fed infants relative to all three breastfed groups. In contrast, an Enterobacteriaceae family ASV was more abundant in all breastfed groups. Comparisons of estrobolome relative abundances ( A ), α-diversity ( B ), and ASV composition ( C ) between breastfed and formula-fed infants in one dataset ( 19 ). Error bars represent 95% CI. All P -values are Benjamini–Hochberg adjusted. In panel C , industrial agriculture formula-fed infants served as the reference group and coefficient directions indicate differences relative to this group (yellow = increase, blue = decrease). Formula-fed infants exhibited estrobolome profiles with higher predicted capacity for estrogen recycling and significantly greater estrobolome diversity than their breastfed counterparts, even within the industrialized population. These findings parallel original work from the same dataset showing that formula-fed infants had greater overall gut microbiome diversity ( 19 ). The observed differences also aligned with broader divergence patterns reported in previous studies comparing overall gut microbiota by feeding mode ( 21 – 25 ) and reinforced the idea that the perinatal period represents a sensitive window for microbiome development with long-term health implications ( 26 , 27 ). Our findings extend this framework by suggesting that early-life diet also affects the estrogen-recycling capacities of infants. That formula-fed infants exhibited higher predicted estrogen-recycling capacity raises the possibility that diet-driven microbial differences could alter systemic hormone exposure during critical developmental windows. These effects may have downstream consequences for growth, maturation, and susceptibility to estrogen-related conditions later in life. Although we did not measure estrogens, this is consistent with studies showing that formula-fed infants have higher height- and weight-for-age z-scores (implying faster growth, although this may also reflect nutritional differences between formula and breast milk) ( 21 ) and higher fecal estradiol levels ( 28 ) than breastfed infants. Longitudinal studies linking feeding mode, microbial function, and hormonal outcomes will be essential to determine whether early estrobolome differences persist into adulthood and/or carry physiological consequences. Consistent with prior literature on the overall infant gut microbiome, Bacteroides was more abundant in formula-fed infants relative to breastfed infants in rural farming/market agriculture communities. Bacteroides are common human gut mutualists that typically colonize the infant gut shortly after birth ( 55 ) but tend to reach higher abundances only after weaning in breastfed infants ( 56 ). Elevated Bacteroides abundance in formula-fed infants has been previously documented ( 56 – 58 ), reinforcing that infant diet shapes the timing and extent of colonization by this genus. R. gnavus was also more abundant in formula-fed infants relative to all breastfed groups. This taxon is known to appear earlier in infants who are weaned from breast milk before 3 mo of age relative to infants that are exclusively breastfed ( 59 ) and is capable of degrading nondigestible oligosaccharides abundant in both host mucin and non-human-milk dietary sources, such as infant formula ( 59 ). Contrary to our finding that an Enterobacteriaceae family ASV was more abundant in all breastfed groups compared to the formula-fed group, previous studies have reported higher overall Enterobacteriaceae abundance in formula-fed infants ( 23 ). This discrepancy may reflect methodological differences, as our analysis focused specifically on ASVs with predicted estrobolome function, rather than broader family-level comparisons commonly used in the literature. We found no consistent associations between sex, age, or BMI and estrobolome characteristics across the three datasets analyzed. Estrobolome relative abundances were similar between males and females in all datasets when partially pooled across populations ( Fig. 5 A – C and Dataset S3 A ). This pattern held when comparisons were restricted to individuals within the same subsistence type ( SI Appendix, Fig. S4 A – C ) and when limiting the sample to premenopausal women (age ≤ 40) and men ( SI Appendix, Fig. S5 A – C ). Estrobolome α-diversity was also comparable between sexes across datasets ( Fig. 5 D – F and Dataset S3 B ), with no differences detected when stratifying by subsistence type ( SI Appendix, Fig. S4 D – F ) or restricting comparisons to premenopausal women and men ( SI Appendix, Fig. S5 A – C ). Only one ASV, identified in the Malawi/USA/Venezuela dataset and belonging to the Lachnospiraceae family, differed significantly by sex (approximately four times higher in females than males, β = 1.39; P BH < 0.05; Dataset S3 C ). Sex-based comparisons of estrobolome relative abundances ( A – C ) and α-diversity ( D – F ) across three datasets ( 19 , 38 , 39 ). Error bars represent 95% CI. Across all datasets, estrobolome relative abundance and composition were not associated with age ( SI Appendix, Fig. S6 A – C and Dataset S4 A ). A weak positive association between α-diversity and age was observed in the Malawi/USA/Venezuela study (0.29% increase per year; β = 0.0029; P BH 0.05). BMI varied widely across datasets, with the highest values occurring in nonindustrial populations in the Botswana/Tanzania/USA dataset and in the United States in the Malawi/USA/Venezuela dataset, yet there were no associations with overall estrobolome abundance or diversity despite considerable variation ( SI Appendix, Fig. S7 A – F and Dataset S5 A and B ). However, two estrobolome ASVs were significantly associated with BMI: one Lachnospiraceae family ASV decreased by 35% per SD increase in BMI (β = –0.437; P BH < 0.05), while a Lachnoclostridium ASV increased 2.2-fold per SD increase (β = 0.786; P BH < 0.05; Dataset S5 C ). These findings do not support the hypothesis that systemic estrogen levels strongly influence estrobolome characteristics. If host estrogens were major drivers of estrobolome characteristics, we would expect consistent and clear sex differences across varied environments. Instead, males and females shared similar estrogen-recycling capacities across populations. These results suggest that ecological and lifestyle factors exert stronger effects on the estrobolome than endogenous hormone levels. Similarly, there were no sex differences in estrobolome α-diversity. Although some studies suggest that overall gut microbial diversity may differ by sex (but findings are inconsistent, see refs. 60 and 61 ), our results indicate that the subset of the microbiome associated with estrogen metabolism does not differ between sexes. Only one Lachnospiraceae ASV showed sex-specific abundance. Lachnospiraceae are a core component of the human gut microbiota, comprising over 58 genera and many unclassified strains ( 62 ). Some evidence suggests that female laboratory mice have higher abundances of Lachnospiraceae family members than males ( 63 ), which is consistent with our result. But without finer taxonomic resolution for this specific ASV, further speculation is difficult. Age was not a major determinant of estrobolome composition, diversity, or abundance. Although one dataset showed a slight age-related increase in α-diversity, this was not replicated elsewhere, and the biological relevance of its effect size (0.31%) is questionable. While total gut microbiome diversity often declines with age ( 64 ), estrobolome diversity may be more consistent throughout the life course. Longitudinal studies will be essential to investigate more detailed potential age-related dynamics in microbial estrogen metabolism. BMI was not associated with estrobolome abundance or diversity. However, two ASVs were BMI-associated in one dataset. One, a Lachnospiraceae ASV distinct from the female-associated variant, decreased with increasing BMI. Prior studies report mixed findings for Lachnospiraceae and body composition: one study found higher abundance in overweight individuals ( 65 ), while another reported negative associations with body fat and cardiovascular lipid risk factors ( 66 ). Our focus on estrobolome-associated taxa and inability to resolve this ASV beyond the family level limits our ability to make meaningful taxonomic comparisons with these findings related to cardiovascular lipid risk factors. Overall, we propose that if host estrogens modulate estrobolome composition, such effects are modest relative to the stronger signals associated with industrialization. That males and females exhibit similar estrobolome profiles across diverse settings suggests estrogen exposure is more strongly shaped by estrobolome function than vice versa. This raises important questions about the sex-specific physiological implications of estrobolome function. Given that males and females appear to have similar estrogen-recycling capacities, one possibility is that the estrobolome plays a greater biological role in females due to their higher circulating estrogen levels (i.e., more estrogen available to recycle), with limited physiological relevance for males. Alternatively, perhaps even small fluctuations in estrogen may have meaningful effects in males, such that estrobolome-driven changes could elicit biologically significant and even distinct responses, thus being important for both sexes.

This study reveals substantial global variation in microbial estrogen-recycling capacity that is shaped primarily by lifestyle and industrialization rather than by host sex, age, or BMI. Industrialized populations consistently showed higher estrobolome-recycling capacities and, contrary to our predictions, greater diversity than nonindustrialized groups, with no consistent differences among the nonindustrial populations. Formula-fed infants exhibited greater estrobolome diversity and higher predicted recycling capacity than breastfed infants, demonstrating that divergence begins in early life and may alter systemic hormone exposure during critical developmental windows. Importantly, these findings do not support the hypothesis that host estrogen levels are major drivers of estrobolome characteristics. Instead, they suggest a predominantly unidirectional influence of microbial estrogen recycling on systemic hormone concentrations, with potential consequences for reproductive and metabolic health, growth, and life history trajectories. Although a “healthy” estrobolome has not been formally defined, dysbiosis of the estrobolome could plausibly inhibit estrogen-recycling capacity or, conversely, promote excessive reabsorption. In extreme cases, these disruptions may hypothetically lead to deficient systemic estrogen levels or hyperestrogenism. More severe disruptions may drive clinically apparent conditions by acutely altering circulating estrogen levels, while subtler shifts in estrobolome function may mostly influence cumulative lifetime estrogen exposure—both of which are key factors in the development of hormone-dependent diseases. The estrobolome has garnered considerable clinical interest, particularly for its potential role in estrogen-dependent cancers such as breast cancer ( 2 , 40 ). However, empirical findings are inconsistent, and no specific gut microbial profiles have been conclusively linked to breast cancer outcomes, for example ( 42 ). Beyond cancer, the estrobolome may have broader and equally important implications for other estrogen-related health conditions, including obesity, infertility, metabolic syndrome, endometriosis, polycystic ovary syndrome, cardiovascular disease, and both cognitive and reproductive dysfunction ( 41 ). Several limitations of this study should be considered. Estrogen-recycling capacity was inferred from the presence of β-glucuronidase-producing taxa and ASVs, but enzymatic activity was not directly measured. Similarly, estrogen was not directly measured, but correlates that capture group differences such as sex were used. Moreover, all industrial populations were from various regions of the United States. Future research should test whether this relationship is observed in other industrialized contexts outside of the United States. Infant feeding studies are especially needed to determine whether early microbial differences persist into adulthood and whether they influence lifelong estrogen exposure. Because glucuronidation and subsequent microbial deconjugation by beta glucuronidases apply to a wide range of substrates, the implications of these findings extend beyond endogenous estrogens to toxins, xenobiotics, and other compounds whose clearance depends on this pathway. This work more specifically highlights the need to consider microbial contributions to endocrine regulation and suggests that estrobolome-targeted interventions may offer avenues for addressing hormone-related conditions in industrialized societies. These findings also raise important questions about how microbial functions have coevolved with human reproductive biology, how ecological shifts in microbiome composition may disrupt long-standing host–microbe interactions, and how industrialization may have introduced novel selective pressures on life history traits by altering systemic hormone exposure. The estrobolome may represent a critical, previously overlooked mechanism linking environmental change to human life history, physiology, and health.

We used publicly available 16S rRNA cross-sectional gut microbiome data to extract estrobolome-associated ASVs and taxa. Because studies differ in the hypervariable regions sequenced and in upstream processing methods, their results are not always directly comparable due to region-specific differences that can lead to discrepancies in taxonomic coverage or resolution ( 67 , 68 ). To address this, we focused on studies with broad geographic sampling sufficient for within-study comparisons, particularly those including both industrialized and nonindustrialized populations with diverse subsistence strategies. Given the limited sample metadata available for public microbiome sequences, we prioritized datasets that reported participant sex and age. While no single study encompassed the full global diversity of human microbiomes across continents, we identified three studies that cumulatively offered relatively broad geographic representation. The first dataset from Hansen et al. ( 38 ) included Bantu, Herero, and San populations from Botswana; Burunge, Hadza, Maasai, and Sandawe populations from Tanzania; and a U.S. population from Philadelphia recruited through an adjacent study using the same laboratory methods ( 69 ). The second dataset from Yatsunenko et al. ( 19 ) comprised Chamba, Makwhira, Mayaka, and Mbiza populations from Malawi; U.S. populations labeled by the original study as Boulder (Colorado), Missouri born, St. Louis (Missouri), St. Louis area (Missouri), and Philadelphia (Pennsylvania); and Coromoto and Platanillal populations from Venezuela. The third dataset from Jha et al. ( 39 ) sampled Chepang, Raji, Raute, and Tharu populations from Nepal, along with Americans of European descent from the United States. We refer to these datasets as “Botswana/Tanzania/USA,” “Malawi/USA/Venezuela,” and “Nepal/USA,” corresponding to the Hansen et al. ( 38 ), Yatsunenko et al. ( 19 ), and Jha et al. ( 39 ) studies, respectively. Collectively, these data represent estrobolome profiles from 24 populations across four continents, encompassing a wide range of human lifestyles and subsistence strategies. Raw 16S rRNA sequences (.fastq files) were downloaded from the National Center for Biotechnology Information (NCBI). Details on the sequencing region, layout, platform, and BioProject accession numbers for each study are referenced in Dataset S6 A . Sequences from each dataset were imported into QIIME 2 (amplicon version 2023.9.1) ( 70 ), where they underwent pair joining (for paired-end reads only), denoising, quality filtering, and taxonomic assignment. Taxonomic classification of ASVs was performed separately for each dataset using a 99% identity threshold and the latest release of the SILVA database (v138.1) ( 71 ). Taxonomy classifiers for 16S V1-V2 (27f/338r) and V4 (515f/806r) were trained using the RESCRIPt QIIME 2 plugin ( 72 ). The number of ASVs that were unable to be assigned taxonomy at each taxonomic level is available in Dataset S6 B . Processed and filtered reads were moved to R (4.2.1) ( 73 ) for downstream processing. All ASVs that matched with chloroplasts or mitochondria were removed. Based on rarefaction curves, we excluded samples with fewer than 3,000 reads from analysis, resulting in the removal of one sample from the Botswana/Tanzania/USA dataset ( Dataset S6 C ). Sequencing depth summary statistics are provided in Dataset S6 C . We highlight that the mean sequencing depth of the Malawi/USA/Venezuela samples was significantly higher than in the other two datasets, which were comparable in depth. Final sample sizes per country, population, subsistence strategy, and sex are available for each study in Table 1 . To predict estrobolome functional capacity, PICRUSt2 was used to identify β-glucuronidase producing ASVs and their respective abundances (KEGG Enzyme EC 3.2.1.31) ( 74 ). These predictions represent the functional potential of estrobolomes, as β-glucuronidase concentrations were not measured directly in any of these studies. We defined “recycling capacity” as the proportion of ASVs predicted to contain KEGG Enzyme EC 3.2.1.31 relative to ASVs without this function, and refer to this metric as estrobolome abundance for brevity. The proportion of ASVs that were unable to be placed into the reference metagenome for PICRUSt2 functional predictions ranged from 0 to 1.5% ( Dataset S6 D ). The number of unique ASVs predicted to contain the β-glucuronidase function are provided at the genus level in Dataset S6 E . All relative abundance and α-diversity analyses were conducted using multilevel models implemented in the glmmTMB package (version 1.1.9) ( 75 ), with population included as a random effect to account for intrapopulation similarity via partial pooling, and family id as a random effect for formula-fed vs breastfed analyses. Relative abundance models employed beta family distributions or ordered beta distributions when zeros were present. α-diversity was modeled using Tweedie distributions with a log link function. The Shannon Index was used to quantify α-diversity and was calculated on rarefied datasets (3,000 reads) to account for variation in sequencing depth between samples, which is known to bias α-diversity estimates ( 76 ). Microbial composition analyses were performed using the MaAsLin2 package (version 1.10.0) ( 77 ), applying linear models with log transformations and total sum scaling (TSS). For microbial composition analyses involving multiple pairwise comparisons, P -values were adjusted for false discovery rates using the Benjamini–Hochberg procedure, and estimates are reported on the log scale. Analyses were conducted at the ASV level, with the lowest taxonomic rank available reported for significant associations. Minimum thresholds of 45% sample prevalence and 0.25% relative abundance were applied to ASVs to focus analyses on robust, reproducible associations and avoid unstable estimates from rare features. In the Botswana/Tanzania/USA dataset, no ASVs met these criteria, so the prevalence threshold was lowered to 20%. Model diagnostics were assessed using the DHARMa (version 0.4.6) ( 78 ) and performance (version 0.12.0) ( 79 ) packages. Estimated marginal means for abundance and α-diversity analyses, including BH-adjusted pairwise analyses, were generated using the marginaleffects package (version 0.17.0) ( 80 ) and are reported on the response scale. All statistical analyses were conducted in R (version 4.2.1) ( 73 ). P -values, coefficients, SE, test statistics, and 95% CI not reported in the text are provided in the corresponding figures and tables. For models exhibiting overdispersion, we specified a group-level dispersion structure in glmmTMB , modeling dispersion as a function of population or subsistence, as appropriate. All statistical models were two-sided. To address whether estrobolomes are more variable within or between populations, we calculated intraclass correlation coefficient (ICC) scores for estrobolome abundances and α-diversity. ICC was computed as the proportion of between-population variance to the total variance, ICC = σ B 2 σ B 2 + σ W 2 , where σ B 2 represents between-population variance and σ W 2 represents within-population variance. An ICC value of 0.5 signifies equal variance within and between populations. Values above 0.5 indicate greater variance between populations, while values below 0.5 suggest that more variation occurs within populations. All three datasets were included in the industrialization and subsistence analyses. Subsistence metadata were available in the Botswana/Tanzania/USA and Nepal/USA datasets. The Malawi/USA/Venezuela dataset did not provide subsistence strategy information; therefore, classifications from Rubel ( 81 ) were adopted for those populations. Subsistence analyses were restricted to individuals aged 18 and older. The Malawi/USA/Venezuela dataset contained an uneven distribution of observations across subsistence groups, with the industrial agriculture group containing approximately 3 to 6 times more participants than other categories. This group included individuals from the St. Louis Area (n = 91), as well as those labeled St. Louis, Missouri-born, Philadelphia, and Boulder (combined n = 38). To achieve better sampling balance with the rural farming (n = 34) and rural farming/market agriculture (n = 21) groups, and to meet homogeneity of variance assumptions, a combined subset of individuals from St. Louis, Missouri-born, Philadelphia, and Boulder was used as the industrial agriculture reference group, and the St. Louis Area observations were excluded. A supplementary model using the St. Louis Area observations as the reference group was run and yielded consistent trends for all analyses, supporting the robustness of the results. The Malawi/USA/Venezuela project provided metadata on infant feeding status (breastfed or formula-fed). Among U.S. populations, this information was available only for the St. Louis group, in which some infants were breastfed and others were formula-fed. In contrast, infants in the Chamba, Coromoto, Mayaka, Mbiza, and Platanillal populations were exclusively breastfed. Because the gut microbiome undergoes substantial maturation during the first few years of life ( 82 ), and among infants sampled the formula-fed group was, on average, slightly older than the breastfed group, we restricted analyses to infants aged 6 mo or younger and included age as a covariate to minimize the effects of age-related microbiome development in our comparisons. Sex comparisons were restricted to individuals aged 18 y and older. To ensure consistency across datasets, participants under 18 were excluded from the Malawi/USA/Venezuela study, as the Botswana/Tanzania/USA and Nepal/USA studies included only adults (≥18 y). Within the Malawi/USA/Venezuela dataset, only the Coromoto, Platanillal, and St. Louis Area populations included both adult males and females; all other populations were excluded to allow population to be modeled as a random effect. Given the observed effect of industrialization in this dataset, we repeated these sex analyses within subsistence types. Because menopause status was not recorded in the original studies, and postmenopausal women generally have lower estrogen levels than premenopausal women, we also conducted parallel analyses comparing adult men (≥18 y) to a subset of women aged 18 to 40 y, as a proxy for premenopausal status. Sample sizes were insufficient to define a postmenopausal comparison group (i.e., women over 55 y), without having actual menopausal data for each individual. For age comparisons, the Botswana/Tanzania/USA and Nepal/USA projects included only individuals aged 18 and older, limiting age comparisons in those datasets to adults. In contrast, the Malawi/USA/Venezuela project included participants ranging from infancy to advanced age. Because the gut microbiome is still developing in early life and is thought to reach an adult-like composition around age three, with some continued maturation throughout childhood ( 83 ), we restricted age-based analyses in the Malawi/USA/Venezuela dataset to individuals aged four and older. This approach enabled us to examine whether estrobolome variation is associated with age in predominantly mature gut microbiomes, rather than tracking early developmental changes during the process of gut microbiome maturation. We excluded one category from the Nepal/USA dataset, labeled “adult >18,” because the specific adult age range represented by these samples was unclear. BMI data were available in the metadata of the Botswana/Tanzania/USA and Malawi/USA/Venezuela datasets. To minimize the influence of growth-related changes in BMI and maintain consistency with the Botswana/Tanzania/USA dataset, which included only adults, individuals under age 18 were excluded from the Malawi/USA/Venezuela dataset. To facilitate model convergence, BMI was standardized within each dataset by mean-centering and scaling to 1 SD.

Appendix 01 (PDF) Dataset S01 (XLSX) Dataset S02 (XLSX) Dataset S03 (XLSX) Dataset S04 (XLSX) Dataset S05 (XLSX) Dataset S06 (XLSX)