The Breast Cancer and Environment Research Centers (BCERC) are a consortium established by the National Institute of Environmental Health Sciences and The National Cancer Institute to elucidate influences of environmental factors on timing of pubertal development in girls. For this purpose, we evaluated exposure using concurrent urinary biomarkers representing three classes of environmental agents in relation to breast and pubic hair development among the girls in this cohort. Biomarkers were selected based on a pilot study that revealed a wide range of values and high detectability in our cohort (Wolff et al. 2007). We hypothesized that phenols would be associated with earlier puberty, that phthalate biomarkers would be related to later pubertal timing, that phytoestrogens would be associated with later breast development, and that associations could be modified by obesity.

As expected, the environmental biomarkers were detected in almost all of the urine samples [Supplemental Material, Table 2 (doi:10.1289/ehp.0901690)]. Overall, phenol median concentrations were < 100 μg/L, and urinary phytoestrogen and phthalate biomarker medians were > 400 μg/L. Geometric mean concentrations (micrograms per liter) of urinary exposure biomarkers are presented in Table 2 mutually adjusted for covariates. Benzophenone-3 (BP3), a sunscreen, was higher in samples collected in the summer and among white children, compared with blacks and Hispanics. 2,5-Dichlorophenol (2,5-DCP) was lower in whites and higher in MSSM participants; it is the metabolite of 1,4-dichlorobenzene used in mothballs and room deodorizers. Parabens are found as preservatives in many personal care products; their levels were higher among blacks and in samples collected in summer. Parabens often occur together with BP3 and low MWP in consumer products. Low-MWP biomarkers were higher in blacks and Hispanics than whites and Asians. The isoflavones daidzein and genistein, which are found in soy products, were somewhat higher among Asians. Enterolactone, a gut metabolite of lignans found in flax, beans, grains, and berries, tended to be higher among blacks and whites.

One year after urine samples were collected for biomarker measurements, breast development was present in 30% and pubic hair in 22% of girls (Table 1). As shown in Table 3, the frequency of any breast and pubic hair development increased or decreased across quintiles of most biomarkers, but the adjusted PRs were close to null. Low MWP had weak positive associations with both breast and pubic hair stages. For breast development, the adjusted PR was 1.06 (95% CI, 0.99–1.14; p-trend = 0.087) in the fifth versus first quintile. The isoflavones daidzein and genistein had weak inverse associations with breast stage; for daidzein, the adjusted PR was 0.96 for the fifth versus first quintile (95% CI, 0.90–1.02; p-trend = 0.061). Triclosan had a suggestive inverse association with pubic hair development, but the trend was not monotonic, and the CIs were similar for every quintile (Table 3). An inverse relationship with pubic hair development was seen for high MWP.

Enterolactone modified the association of BMI with breast development. The proportion of low-BMI girls at pubertal stages B2+ increased from 15 to 20% across enterolactone quintiles (see note to Table 4). Among girls with high BMI, the proportion in the first quintile (48% B2+) decreased to 25% B2+ in the fifth quintile. This inverse association among high-BMI girls supported our hypothesis, and it remained after further adjustment for covariates (Table 4). The difference in the enterolactone PRs between low- and high-BMI groups suggests that higher enterolactone exposures may attenuate the BMI association. Specifically, in the first quintile of enterolactone in Table 4, the PR for pubertal stage B2+ versus B1 was 1.34 (95% CI, 1.23–1.45) for high BMI compared with low-BMI girls, the referent. This difference, the “BMI association,” diminished steadily so that in the fifth quintile of enterolactone, the PRs for B2+ versus B1 were comparable for low BMI and high BMI (PR = 1.14 and 1.20, respectively, with similar CIs) (Table 4). Thus, breast development differed among low- and high-BMI girls with low enterolactone exposure, but development was similar at the highest exposure levels regardless of BMI. These enterolactone associations with breast stage were similar in models stratified by BMI and further adjusted for BMI%; results were also consistent by race/ethnicity and by site.

Associations of concurrent exposure biomarkers with breast and pubic hair development in these girls were not strong, but those observed were among the chemicals with greatest exposure levels. The strongest finding was attenuation by enterolactone exposure of the BMI association with breast development. Along with the inverse relationships of daidzein and genistein with breast development and high MWP with pubic hair stage, the results were consistent with our a priori hypotheses and the experimental literature. Comparable associations of phytoestrogens with breast stage were seen in an earlier small study (Wolff et al. 2008). Phytoestrogens including enterolactone are known to possess hormonal activity (Adlercreutz 2002). A protective effect for puberty might be consistent with counteracting the influence of obesity (Horn-Ross 1995) or by reducing adiposity (Cederroth et al. 2007). In contrast, associations were positive, albeit very weak, for low MWP with both breast and hair development. It is not clear why low- and high-MWP metabolites could have opposite associations with developmental stages, yet various reports of such exposures in humans and animals show divergent hormonal associations, depending on timing and intensity of exposure or treatment and on rodent strain (Adlercreutz 2002; Rasier et al. 2006; Schoeters et al. 2008; Shen et al. 2009). In addition, patterns and density of ambient exposures no doubt differ for the low MWPs and high MWPs (Adibi et al. 2008).

Residual confounding or misclassification of exposures and outcomes remain possible explanations for our results. Collinearity of covariates, such as that among BMI, race/ethnicity, urinary creatinine, and urinary biomarkers and their variation by study site, are potential difficulties. We used methods with robust variance handling in an effort to minimize such problems. A potential explanation for the lack of strong associations is overadjustment of the models due to the inclusion of certain covariates (Greenland et al. 1999; Weinberg 1993). For example, BMI may be both a confounder and on the pathway between exposure and pubertal development. Considering the interrelationships among our variables, the models presented are the most appropriate. For our main analyses, we used creatinine-corrected biomarker concentrations to create quintile cut points. Creatinine correction may be inadequate for some or all analytes. However, we did not measure specific gravity, an alternate measure of urine dilution (Hauser et al. 2004; Miller et al. 2004). Other methods, such as the regression normalization procedures (Heavner et al. 2006), may not be appropriate for all urinary metabolites. Besides exposure misclassification, there is potential error in the outcome measurement of pubertal stage because of inter-rater variability in pubertal stage assessment. Therefore, both exposure and outcome measurements may be subject to nondifferential misclassification bias, which would likely shift the estimates toward the null. Additional considerations, including genetic and racial differences in exposure and development, would require considerably larger sample size. We estimated that for the main effects we had adequate power (80%) to detect PRs of 1.1 in 479 girls, if B2+ or PH2+ were > 20% in the fifth quintile (alpha 0.05), and a PR of 0.94 with 948 girls; these values are similar to the strongest associations we observed. Our effect estimates also may be conservative, because we used Poisson models instead of logistic regression models. For example, for the PR of 0.94 (95% CI, 0.88–1.00, high MWP and PH2+) in Table 3, we computed an odds ratio (OR) of 0.60 (95% CI, 0.34–1.06). However, measures of association using ORs may be over- or underestimated (Zou 2004). Finally, some or all of our findings may be due to chance; > 100 comparisons were made for the models presented in Tables 3 and ​and4.4. Associations of hormonal exposures in this study were small, which may be consistent with their relatively weak biological activity compared with endogenous hormones (Fang et al. 2000; Shen et al. 2009). Small effect estimates may be more realistic than those in previous studies that had small sample sizes (Wolff et al. 2008). There will be greater power to detect associations in longitudinal analyses that can also better reflect causal relationships than cross-sectional analyses; we plan to undertake such analyses as the cohort matures. The reports of delayed pubertal development in relation to blood lead concentrations in the NHANES population are informative for our findings, as the lead effects appear stronger than those we observed. Selevan et al. (2003) observed among black girls an OR of 0.62 (95% CI, 0.41–0.96, multivariate adjusted) for PH2+ versus PH1 among girls with blood lead > 3 μg/dL, quartile of exposure, compared with those having blood lead < 1 μg/dL, approximately the upper versus first quartile of their exposures. For the same NHANES population, Wu et al. (2003) found ORs for PH2+ of 0.27 (95% CI, 0.08–0.93) in the top exposure group (≥ 5 μg/dL) compared with ≤ 2 μg/dL blood lead. The proportion of PH2+ in the low exposure stratum was 81% versus 44% at high exposure. By comparison, the prevalence of PH2+ was 28% in the first compared with 20% in the fifth quintile of high MWP in Table 3, and the adjusted PR was 0.94 (95% CI, 0.88–1.00). High-MWP medians were 7-fold different between these quintiles compared with 3-fold differences in the lead exposure categories.

An additional consideration is that the peripubertal period is likely not the only critical window of exposure for pubertal development. Both animal and epidemiology studies suggest that prenatal and perinatal exposures also exert effects on later development (Rasier et al. 2006; Schoeters et al. 2008). Exposures during different windows may affect different molecular targets, including prenatal imprinting, the hypothalamic-pituitary axis, gonadotropin-releasing hormone neurons, hormone receptors, and aromatase action (Schoeters et al. 2008). Effective exposure ranges for these mechanisms may also differ widely, that is, toxic equivalents. Environmental agents in our study are cleared rapidly; it is possible that a single biomarker measurement is inadequate to quantify exposure relevant to pubertal development. However, single measurements of urinary biomarkers of phenols and phthalates were fairly representative of 6–12 months of exposure in children this age (Teitelbaum et al. 2008), likely because of common use and continuous exposure to many chemicals. Time-integrated multiple childhood exposure measures prenatally and prepubertally may be possible in alternate study designs. An important additional direction is to evaluate multiple exposures, including the extremes of exposure, multiple high exposures, early life exposures, and/or extremes of development (very late vs. very early changes) (Chou et al. 2009).