Links Between Hormonal and Pubertal Development, and Adolescent Females' Risk for Affective Symptoms

Abstract

Depression and anxiety symptoms surge during adolescence, particularly in females, yet the biological mechanisms underlying this vulnerability remain poorly understood. Although estrogen has traditionally been emphasized, androgens such as dehydroepiandrosterone (DHEA) and testosterone may represent distinct pathways to affective symptom risk. The present study examined whether increases in testosterone predict internalizing symptoms in adolescent females by examining hormonal and physical aspects of pubertal development and testing whether testosterone effects persist beyond estradiol and observable maturation. Using data from 5,476 females (ages 9–13) in the Adolescent Brain Cognitive Development Study, we analyzed annual salivary DHEA, testosterone, and estradiol alongside caregiver-reported pubertal development and youth-reported internalizing symptoms. Latent change score models captured developmental dynamics: univariate models characterized growth trajectories, bivariate coupling models tested directional hormonal influences, and the final model evaluated whether testosterone and pubertal development changes predicted internalizing symptoms, controlling for estradiol, age, and assay covariates. Testosterone increased gradually across early adolescence, whereas DHEA showed accelerating, nonlinear growth. Increases in testosterone predicted higher internalizing symptoms between ages 10 and 12 (b = 0.041, 95% CI [0.007, 0.076]; b = 0.075, 95% CI [0.009, 0.142]). Observable pubertal development was negatively associated with internalizing symptoms at one assessment. Estradiol predicted subsequent pubertal development and showed concurrent associations with internalizing symptoms at two assessments but did not explain the testosterone-internalizing link. Increases in testosterone predicted internalizing symptoms after accounting for observable maturation and estradiol, identifying a specific vulnerability window between ages 10–12 with implications for early identification and timing of preventive interventions.

Graphical Abstract

1. Introduction

Adolescence is a period of heightened vulnerability for internalizing symptoms especially in females (Demkowicz et al., 2025; Ge et al., 2001; Hankin et al., 2015). Symptoms often emerge at subclinical levels but nonetheless confer functional impairment and downstream risk for psychopathology (Beesdo et al., 2009; Fergusson et al., 2005; Lewinsohn et al., 2000). Symptoms often coincide with pubertal development, suggesting that maturational processes associated with both observable development and endocrine activity heighten risk (Oldehinkel et al., 2011; Patton et al., 2018). Ge and Natsuaki (2009) distinguished between the maturation disparity hypothesis and the hormonal influence hypothesis as complementary frameworks for understanding pubertal contributions to affective vulnerability.

The maturation disparity hypothesis emphasizes risk associated with visible physical development that outpaces psychosocial and cognitive readiness (Ge and Natsuaki, 2009). Observable pubertal maturation fundamentally alters adolescents' social experiences through multiple empirically documented pathways. Early-maturing females experience elevated rates of peer harassment, attention, and objectification social stressors that exceed their developmental capacity for adaptive coping (Ge et al., 2001; Mendle et al., 2007; Negriff and Susman, 2011). The timing of visible maturation relative to same-age peers activates social comparison processes, whereby adolescents evaluate their physical development against peer norms and internalized appearance ideals, often resulting in heightened body dissatisfaction and self-objectification (Carlson Jones, 2004; Kaltiala-Heino et al., 2003). Physical maturity also elicits differential treatment from adults and peers, who impose age-inappropriate social expectations and grant access to older peer contexts for which early maturers may lack emotional or cognitive preparation (Ge et al., 2001; Weichold and Silbereisen, 2003) This asynchrony between physical appearance and psychosocial readiness creates a developmental mismatch in which adolescents must navigate complex social demands without the regulatory capacities or life experience necessary to manage these challenges effectively (Ge and Natsuaki, 2009; Mendle et al., 2007).

In contrast, the hormonal influence hypothesis emphasizes endocrine activity itself as a direct mechanism of affective vulnerability, proposing that pubertal hormones influence brain development and emotional reactivity independent of visible physical changes (Ge and Natsuaki, 2009). Pubertal hormones, including both gonadal steroids and adrenal androgens, act on developing neural circuits involved in emotion regulation, reward processing, and stress reactivity, potentially increasing sensitivity to environmental stressors and negative affect (Peper and Dahl, 2013; Vijayakumar et al., 2018). Neurobiological evidence demonstrates that rising hormone levels during puberty coincide with structural and functional reorganization of corticolimbic circuitry, altering both bottom-up emotional reactivity and top-down regulatory capacity (Goddings et al., 2014; Herting et al., 2014). Importantly, these hormone-brain associations can occur before or independent of observable physical maturation, suggesting that endocrine processes may confer vulnerability through neurobiological pathways distinct from the psychosocial mechanisms emphasized by the maturation disparity hypothesis (Copeland et al., 2019; Marceau et al., 2015).

Pubertal development reflects activation of the hypothalamic–pituitary–gonadal (HPG) axis, which stimulates gonadal production of sex steroids such as testosterone and estradiol, and maturation of the hypothalamic–pituitary–adrenal (HPA) axis which produces adrenal androgens, including dehydroepiandrosterone (DHEA) (Biro et al., 2003; Dorn and Biro, 2011; Shirtcliff et al., 2009). These hormonal changes coincide with observable shifts in secondary sex characteristics and with neurobiological remodeling, including structural and functional changes in brain regions involved in emotion regulation, reward processing, and social cognition (Blakemore et al., 2010; Vijayakumar et al., 2018). Puberty reflects a dynamic, interactive regulatory system in which physiological and observable maturational processes occur in tandem (Kretzer et al., 2024; Mendle et al., 2019). Few empirical studies, however, have jointly modeled hormonal and physical development across time (Luo et al., 2024; Marceau et al., 2015; Shirtcliff et al., 2009).

Estradiol, the primary estrogen in females, has been implicated in mood variability during puberty, although empirical findings remain inconsistent and context-dependent (Andersen et al., 2024a; Angold et al., 1999; Gunnar and Quevedo, 2007). In contrast, androgens and their precursors, historically emphasized in male neurodevelopment, are now increasingly linked to adolescent females’ emotional reactivity and sensitivity to social threat (Andersen et al., 2024b; Ho et al., 2022). Androgenic processes may heighten vulnerability by shaping neuroaffective development while also intensifying social challenges associated with visible, androgen-driven physical maturation. (Andersen et al., 2024b; Luo et al., 2024; Marceau et al., 2015)

DHEA, which rises early in puberty during adrenarche, modulates stress systems and serves as a precursor to testosterone (Black et al., 2022; Labrie, 2010; Maninger et al., 2009). It also contributes to observable markers of maturation in females, including pubic and underarm hair growth, the earliest visible signs of pubertal development (Auchus and Rainey, 2004; Rainey et al., 2002). Testosterone itself increases gradually across adolescence, facilitating physical maturation and influencing neural circuits involved in reward processing and emotion regulation (Apter, 1997; de Macks et al., 2011). Such changes in testosterone may heighten sensitivity to social evaluation and stress, providing a key pathway to internalizing risk in females (Peper and Dahl, 2013; Tyborowska et al., 2024).

The present study focuses on androgenic processes as predictors of internalizing symptoms in females across early to mid-adolescence. Although estradiol has long been considered central to pubertal emotional changes, we treat it as a covariate rather than a focal predictor, allowing us to advance knowledge by isolating androgenic contributions. By modeling temporal patterns of DHEA and testosterone alongside pubertal development, this study evaluates whether androgenic hormones, beyond the influence of estrogen, represent distinct biological contributors to internalizing risk. This targeted approach moves beyond traditional emphases on estradiol to illuminate underexplored mechanisms of pubertal emotional vulnerability. In Aim 1, we examine how DHEA, testosterone, and pubertal development influence one another over time, hypothesizing that these processes will show coordinated patterns of change with DHEA functioning as an upstream regulator of subsequent testosterone and pubertal changes. In Aim 2, we link this hormonal-physical system to internalizing symptoms, hypothesizing that increases in testosterone and accelerated pubertal development will predict elevated internalizing symptoms even after accounting for estradiol and chronological age.

2. Method

The Adolescent Brain Cognitive Development (ABCD) (Consortium, 2023) Study is a longitudinal, multisite investigation of brain, cognitive, and emotional development in U.S. youth. At baseline, 11,880 children aged 9–10 years were recruited from 21 sites. The analytic sample included 5,476 females with at least one valid hormone or pubertal measure across four annual waves (ages 9–13). The sample was defined by sex assigned at birth as recorded in ABCD; we use "female adolescents" to refer to biological sex throughout, acknowledging that sex and gender identity are distinct constructs with different implications for pubertal and mental health processes. Detailed information about ABCD Study recruitment, retention, and sample characteristics is available at https://abcdstudy.org.

Menarche was reported by caregivers at each annual assessment. At baseline (ages 9–10), 2.6% of participants had initiated menarche. Across the study period, 68% had initiated menarche by the final assessment. Among participants who experienced menarche during follow-up, approximately 35% showed minor inconsistencies in reported age across waves, most reflecting a one-year difference. Age at menarche was derived as the earliest non-missing report across waves. The mean age at menarche was 11.30 years (SD = 2.00; median = 11; range = 7–18). At Year 1 (ages 10–11), mean gynecological age (chronological age minus age at menarche) among post-menarcheal participants was 0.79 years.

2.1. Measures

2.1.1. Hormone Assays

Salivary DHEA, testosterone, and estradiol were assayed using commercially available enzyme immunoassay kits (Salimetrics, State College, PA) following standardized ABCD protocols (Uban et al., 2018). Samples were collected during morning research visits, refrigerated, shipped on dry ice, and stored at −80 °C. Assay sensitivities were 5 pg/mL for DHEA, 1 pg/mL for testosterone, and 0.1 pg/mL for estradiol; intra-assay and inter-assay coefficients of variation were <5% and <10%, respectively (Salimetrics technical specifications, https://www.salimetrics.com).

Participants reporting hormonal birth control use were excluded from analyses (Chaku and Barry, 2024). Each hormone was assayed in duplicate. Individual replicates were set to missing if they exhibited quality concerns (e.g., discoloration), fell outside assay sensitivity, or were below detection limits (Chaku and Barry, 2024). When both replicates were valid, they were averaged; when only one was valid, that value was used. To preserve the longitudinal structure of the data, hormone values were set to missing at specific waves when necessary, while retaining participants who contributed valid data at other time points. This approach maintains sample size and leverages all available information, which is appropriate for latent change score modeling estimated using full information maximum likelihood (FIML) (Newman, 2014). Values were winsorized at the 1st and 99th percentiles. ABCD standardizes collection and assay procedures across sites; no additional harmonization or batch corrections were applied. Fewer than 0.01% of samples were below detection limits.

2.1.2. Pubertal Development

Caregivers completed the Pubertal Development Scale (five items plus menarche) about their child at four annual time points beginning at ages 9–10. Items were averaged to create a composite score (higher = more advanced development) (Petersen et al., 1988). The PDS demonstrates acceptable reliability and validity in this cohort (Cheng et al., 2021; Herting et al., 2021; Palmer et al., 2021). We used parent-report rather than youth self-report based on measurement quality considerations. At baseline (ages 9–10), most females were in early pubertal stages when youth self-report shows poor reliability compared to parent report. Beltz et al. (2026) systematically demonstrated that youth self-report in the ABCD study has some limited utility in early puberty and documented high rates of missing data that may not be random, with youth and parent reports differing systematically. Parent-report PDS showed better data completeness and measurement properties during this early developmental window.

2.1.3. Internalizing symptoms

Youth completed the 31-item Internalizing subscale of the Youth Self-Report (range 0–62) (Achenbach, 2001) at three annual time points beginning at ages 10-11. Higher scores reflect greater symptom severity. Reliability in this cohort was α ≈ .85–.90 (Ebesutani et al., 2010; Song et al., 1994). Youth self-report was unavailable at baseline (ages 9-10) because the YSR is designed for ages 11-18. Youth-reported symptoms became available at Year 1 (ages 10-11) when participants aged into the YSR range. At baseline, internalizing symptoms were assessed exclusively via parent-report CBCL. Given our focus on youth-reported symptoms to capture adolescents' subjective affective experience, we could not control for baseline symptom levels.

2.1.4. Covariates

Chronological age, time since awakening and flow rate were included in all predictive models and at each time point; estradiol collected at four annual time points beginning at ages 9-10 was included as a time-varying covariate predicting internalizing symptoms and subsequent pubertal development (Angold et al., 1999; Mendle et al., 2010; Uban et al., 2018).

2.2. Analytic Plan

Longitudinal changes in DHEA, testosterone, and pubertal development were modeled using latent change score (LCS) models (Collins and Sayer, 2001; McArdle and Hamagami, 2001a) using Mplus (version 8.8). Missing data were handled with full information maximum likelihood (FIML). All models accounted for family clustering by adjusting standard errors for nested data structure (allowing correlated residuals within families) and applied propensity weights for the ABCD sampling design. Site-level variance was accounted for by including site as a stratification variable, consistent with recommendations for ABCD behavioral and biospecimen analyses (Dick et al., 2021). Model fit was evaluated using chi-square (with df and p), RMSEA (<.05 good, .05-.08 acceptable, >.10 poor), CFI, and TLI (>.90 acceptable, >.95 good). RMSEA is reported with 90% confidence intervals per convention; all parameter estimate confidence intervals are 95% (MacCallum et al., 1996). Given large sample size, we emphasized approximate fit indices over chi-square.

Univariate models were estimated without covariates to capture developmental trajectories without removing normative variation. We first estimated univariate dual change LCS models for DHEA, testosterone, and pubertal development. These models capture constant change (linear growth) and proportional change (dependence on prior levels), which together describe whether development accelerates, stabilizes, or decelerates(Barker et al., 2014; Cáncer et al., 2021). Means of these parameters summarize average trajectories across the sample.

Next, bivariate LCS models tested whether prior levels of one construct predicted subsequent change in another. Coupling parameters estimate directional, time-ordered effects while accounting for each construct's own growth parameters (Hilley and O'Rourke, 2022; McArdle, 2009). For each model, we retained both change components (constant and proportional) for each construct and added cross-lagged coupling paths to test all possible directional influences between constructs. We followed a stepwise model-building approach for bivariate models. First, we estimated models with coupling parameters constrained to be equal across time intervals, yielding a single estimate for each cross-construct effect. This parsimonious specification provides an initial test of whether stable cross-construct associations exist while reducing model complexity (Grimm et al., 2012; Kievit et al., 2018). These initial constrained models were estimated without covariates to characterize the raw bivariate associations. Second, for coupling parameters that showed significant effects in constrained models, we freed these parameters to vary across intervals and added covariates (chronological age, time since awakening, and saliva flow rate) to test whether associations differed by developmental stage and remained robust after accounting for potential confounds (Little, 2024; Selig and Little, 2012). Positive coupling indicates a developmental cascade, whereas negative coupling reflects a dampening or regulatory effect (Grimm et al., 2012; McArdle, 2009). Significant cross-domain effects identified in these time-varying covariate-adjusted bivariate models were retained in subsequent analyses (Grimm et al., 2016; McArdle and Grimm, 2010). This stepwise approach follows recommended practices in latent change score modeling, where parsimony and interpretability are balanced with thoroughness of specification (Little, 2024).

We then fit a trivariate LCS model jointly modeling DHEA, testosterone, and pubertal development. Coupling paths were selected based on bivariate model findings and theoretical considerations (DHEA → testosterone; DHEA → pubertal development) (McArdle and Hamagami, 2001b; Voelkle et al., 2012). This system allowed us to test unique prospective effects among constructs while accounting for shared variance (Hilley and O'Rourke, 2022). Finally, we extended this model to include internalizing symptoms. Predictive paths from testosterone and pubertal development change terms were specified based on prior evidence (Andersen et al., 2024b; Copeland et al., 2019; de Macks et al., 2011; Ge et al., 2001; Mendle et al., 2007). Estradiol was included as a time-varying covariate, predicting both internalizing symptoms and subsequent pubertal change (de Macks et al., 2011; Grumbach and Styne, 1998). The model also included autoregressive paths linking available internalizing assessments (Year 1 through Year 3), controlling for prior symptom levels when testing whether hormonal and pubertal changes predicted symptoms.

3. Results

3.1. Univariate Dual Change LCS Models

We estimated univariate LCS for DHEA, testosterone, and pubertal development across ages 9-13. For DHEA, model fit was good (χ²(7) = 84.458, p < .001, RMSEA = 0.045, 90% CI [0.037, 0.054], CFI = 0.966, TLI = 0.971, SRMR = 0.057). The constant change component (α = −57.25, 95% CI [−72.54, −41.95]) was negative, while the proportional change parameter (π = 0.921, 95% CI [0.724, 1.117]) was strongly positive, indicating that females with higher DHEA levels at each timepoint showed greater subsequent increases, reflecting an accelerating trajectory.

For testosterone, model fit was good (χ²(7) = 68.226, p < .001, RMSEA = 0.040, 90% CI [0.032, 0.049], CFI = 0.970, TLI = 0.974, SRMR = 0.057). The constant change component (α = 5.35, 95% CI [1.84, 8.85]) was positive, reflecting a modest overall increase in testosterone across early adolescence. The proportional change parameter (π = 0.041, 95% CI [−0.047, 0.129]) was not statistically significant, indicating that these increases were relatively uniform across females regardless of starting levels.

For pubertal development, model fit indices were mixed (χ²(7) = 503.549, p < .001, RMSEA = 0.114, 90% CI [0.106, 0.123], CFI = 0.908, TLI = 0.921, SRMR = 0.221), with incremental fit indices (CFI, TLI) approaching acceptable levels while absolute fit indices (RMSEA, SRMR) indicated substantial misfit. Given this misfit, parameters from the univariate pubertal development model should be interpreted with caution. The constant change component (α = 0.624, 95% CI [0.539, 0.708]) indicated a general trend of increasing development. The proportional change parameter (π = −0.038, 95% CI [−0.068, −0.008]) was negative, indicating that females with more advanced pubertal status at each timepoint showed smaller subsequent increases, reflecting a decelerating pattern.

3.2. Bivariate Dual Change LCS models

We next examined pairwise associations among DHEA, testosterone, and pubertal development using: (1) unidirectional, (2) reverse unidirectional, and (3) bidirectional models. At this stage, coupling parameters were constrained to be equal across intervals, yielding a single estimate for each cross-construct effect, enabling evaluation of whether stable differences in one construct predict systematic change in the other before examining interval-specific effects (Grimm et al., 2012; Kievit et al., 2018). Table 1 summarizes model fit statistics and coupling parameter estimates for these constrained models without covariates. Several coupling parameters were negative, with higher DHEA predicting less change in testosterone (in the bidirectional model) and pubertal development, and more advanced pubertal development predicting less change in testosterone, while other models did not converge or produced inadmissible model solutions. Model fit was generally acceptable across pairs (see Table 1).

Table 1. Bivariate DCSM Results.

Bivariate Pair	Model Direction	Coupling Path	Estimate (γ)	95% CI	X2(df)	RMSEA	CFI/TLI	SRMR
DHEA & testosterone	DHEA → testosterone	DHEA → Δtestosterone	Model non-convergence
	testosterone → DHEA	testosterone → ΔDHEA	−0.325	[−.940,.291]	238.2(24)	0.04	.981/.978	0.052
	Bidirectional	DHEA → Δtestosterone	−0.177	[−.322, −.032]	234.0(23)	0.041	.981/.977	0.049
		testosterone → ΔDHEA	0.434	[−0.400, 1.27]
DHEA & Pubertal dev.	DHEA → Pubertal dev.	DHEA → ΔPubertal dev.	−0.013	[−.021, −.005]	639.3(24)	0.068	.932/.920	0.093
	Pubertal dev. → DHEA	Pubertal dev. → ΔDHEA	−5.021	[−10.9,.841]	614.2(24)	0.067	.935/.924	0.131
	Bidirectional	DHEA → ΔPubertal dev.	Inadmissible Model Solution
		Pubertal dev. → ΔDHEA
testosterone & Pubertal dev.	testosterone → Pubertal dev.	testosterone → ΔPubertal dev.	Model non-convergence
	Pubertal dev. → testosterone	Pubertal dev. → Δtestosterone	−4.369	[−7.21, −1.53]	625.0(24)	0.067	.931/.919	0.13
	Bidirectional	testosterone → ΔPubertal dev.	Inadmissible Model Solution
		Pubertal dev. → Δtestosterone

Abbreviations: CI, confidence interval; RMSEA, root mean square error of approximation; CFI, comparative fit index; TLI, Tucker-Lewis index.

Note: Boldface indicates statistically significant estimates (P < .05).

In the next set of models, significant coupling parameters were freed to vary across time intervals to test whether associations differed by developmental stage (Selig and Little, 2012; Usami et al., 2019). Covariates (age, time since awakening, and saliva flow rate) were also added to these models to account for potential confounding influences (Little, 2024). Table 2 presents these interval-specific unstandardized estimates with covariates included. The model testing DHEA → testosterone change produced an inadmissible solution (non-positive definite covariance matrix) when parameters were freed and covariates were added. Across all three intervals, higher DHEA levels predicted less subsequent change in pubertal development, and higher pubertal development levels predicted less subsequent change in testosterone, with the magnitude of this negative effect increasing over time (Table 2).

Table 2. Time-Varying Bivariate DCSM Results with Covariates.

Coupling Path	T1-T3 (γ)	95% CI	T3-T5 (γ)	95% CI	T5-T7 (γ)	95% CI	X2(df)	RMSEA	CFI/TLI	SRMR
DHEA → Δtestosterone	Inadmissible Model Solution
DHEA → ΔPubertal dev.	−0.045	[−.055, −.036]	−0.035	[−.043, −.028]	−0.032	[−.038, −.026]	904(159)	0.029	.904/.925	0.106
Pubertal dev. → Δtestosterone	−8.05	[−13.2, −2.94]	−12.2	[−19.1, −5.38]	−16.1	[−24.3, −8.0]	1190(159)	0.034	.897/.920	0.092

Abbreviations: CI, confidence interval; RMSEA, root mean square error of approximation; CFI, comparative fit index; TLI, Tucker-Lewis index.

Note: Boldface indicates statistically significant estimates (P < .05).

3.3. Trivariate Model (Modeling the Regulatory System)

We then jointly modeled DHEA, testosterone, and pubertal development as an integrated dynamic system. Based on bivariate findings and theoretical considerations regarding DHEA's role as an upstream regulator of pubertal processes, we retained coupling paths from DHEA to both testosterone change and pubertal development change (DHEA → testosterone, DHEA → pubertal development) (Grimm et al., 2012; McArdle, 2009). The path from pubertal development to testosterone was not included to maintain parsimony. Within this framework, higher DHEA levels predicted smaller subsequent increases in both testosterone and pubertal development across intervals.

3.4. Integrative Model (Linking Pubertal Dynamics to Internalizing Symptoms)

Building on the trivariate framework, the final model examined pathways to internalizing symptoms. Direct predictive paths from testosterone and pubertal development change terms to internalizing symptoms were specified. DHEA was retained to capture indirect influences via its effects on testosterone and pubertal development. Model fit was adequate (χ²(368) = 3057.695, p < .001, RMSEA = 0.037, 90% CI [0.035,0.037], CFI = 0.909, TLI = 0.915, SRMR = 0.116).

Estradiol was included as a time-varying covariate predicting both internalizing symptoms and pubertal development change. Estradiol was concurrently associated with internalizing symptoms at Year 2 (b = 0.205, 95% CI [0.011, 0.398], p = .038) and Year 3 (b = 0.314, 95% CI [0.008, 0.621], p = .044) but not at Year 1 (p = .199). Estradiol predicted subsequent pubertal development change between the final two time points (Year 2 → Year 3: b = 0.087, 95% CI [0.042, 0.132], p < .001) but not at earlier intervals (ps > .09).

Greater testosterone changes predicted higher internalizing symptoms at Year 1 (b = 0.041, 95% CI [0.0057, 0.076], p = .019) and Year 2 (b = 0.075, 95% CI [0.009, 0.142], p = .026) but not at Year 3 (p = .372). Pubertal development change showed a significant negative association with internalizing symptoms in Year 2 only (b = −0.637, 95% CI [−1.257, −0.017], p = .044), with higher rates of pubertal development predicting lower symptoms. These results indicate that testosterone changes were associated with internalizing symptoms even when accounting for pubertal development changes and estradiol. The final path model is presented in Figure 1; unstandardized estimates are in Table 3.

Figure 1. Model-Implied Trajectories for DHEA and Testosterone Across Early Adolescence.

(A) DHEA and (B) testosterone trajectories from ages 9–13 estimated using univariate latent change score models. Solid blue lines represent model-implied trajectories derived from estimated constant change (α) and proportional change (π) parameters. Shaded bands depict 95% confidence intervals around the model-implied trajectory, derived by propagating Mplus-estimated intercept and slope factor variances and their covariance through the latent change score recurrence. Gray lines depict observed trajectories from a random subsample of participants with complete data across all four waves (n = 100). DHEA exhibited significant accelerating change (π = 0.92, p < .001), with increasingly larger increments at each successive wave. Testosterone showed steady linear growth with no significant proportional change effect (π = 0.04, p = .358). Y-axes are clipped to the 2nd–98th percentile of the observed subsample to limit the influence of extreme values on axis scaling.

Table 3. Path estimates for the Final Integrated Model.

Path	Figure 2. notation	Estimate	95% CI
DHEA baseline → ΔTestosterone from Baseline to Year 1	Ƴ 1	−0.635	[−0.788, −0.482]
DHEA Year 1 → ΔTestosterone from Year 1 to Year 2	Ƴ 2	−0.571	[−0.694, −0.448]
DHEA Year 2 → ΔTestosterone from Year 2 to Year 3	Ƴ 3	−0.484	[−0.604, −0.364]
DHEA baseline → ΔPubertal dev. from Baseline to Year 1	Ƴ 4	−0.040	[−0.051, −0.034]
DHEA Year 1 → ΔPubertal dev. from Year 1 to Year 2	Ƴ 5	−0.032	[−0.037, −0.027]
DHEA Year 2 → ΔPubertal dev. from Year 2 to Year 3	Ƴ 6	−0.030	[−0.034, −0.026]
Estradiol baseline → ΔPubertal dev. from Baseline to Year 1	b1	0.017	[−0.040, 0.075]
Estradiol Year 1 → ΔPubertal dev. from Year 1 to Year 2	b2	0.043	[−0.009, 0.095]
Estradiol Year 2 → ΔPubertal dev. from Year 2 to Year 3	b 3	0.084	[0.038, 0.131]
Estradiol Year 1 → Internalizing Year 1	b4	0.109	[−0.057, 0.275]
Estradiol Year 2 → Internalizing Year 2	b 5	0.205	[0.011, 0.398]
Estradiol Year 3 → Internalizing Year 3	b 6	0.314	[0.008, 0.621]
ΔTestosterone from Baseline to Year 1 → Internalizing Year 1	b 7	0.041	[0.007, 0.076]
ΔTestosterone from Year 1 to Year 2 → Internalizing Year 2	b 8	0.075	[0.009, 0.142]
ΔTestosterone from Year 2 to Year 3 → Internalizing Year 3	b9	−0.017	[−0.055, 0.021]
ΔPhenotype from Baseline to Year 1 → Internalizing Year 1	b10	−0.031	[−0.352, 0.290]
ΔPhenotype from Year 1 to Year 2 → Internalizing Year 2	b 11	−0.637	[−1.257, −0.017]
ΔPhenotype from Year 2 to Year 3 → Internalizing Year 3	b12	0.301	[−0.182, 0.784]

Abbreviations: CI, confidence interval.

Note: Boldface indicates statistically significant estimates (P < .05).

4. Discussion

The univariate models revealed distinct developmental patterns across pubertal processes. DHEA demonstrated an accelerating, nonlinear trajectory in which females with higher levels at each timepoint showed greater subsequent increases. This pattern aligns with normative adrenarche, during which DHEA rises gradually in early puberty and then accelerates sharply as the adrenal cortex undergoes functional maturation (Auchus and Rainey, 2004; Havelock et al., 2004; Kamin and Kertes, 2017). Testosterone showed modest, uniform increases across early adolescence regardless of baseline levels, reflecting the gradual and consistent rise characteristic of female gonadarche during this developmental window (Apter, 1997; Mendle et al., 2019). Observable pubertal development followed a decelerating pattern: females with more advanced maturation showed smaller subsequent increases, consistent with normative pubertal progression that slows as adolescents approach maturational limits (Marshall and Tanner, 1969; Shirtcliff et al., 2009). Model fit for the univariate pubertal development model (RMSEA = 0.114, SRMR = 0.221) was notably weaker than for DHEA and testosterone, likely reflecting the inherent difficulty of capturing a continuous developmental process with discrete ordinal items, and these trajectory parameters should be treated as approximate characterizations rather than precise estimates. Model fit was good for DHEA and testosterone (RMSEA < .05, CFI/TLI > .96). The divergent trajectories underscore that puberty reflects the coordinated but partially independent activation of multiple endocrine systems, each following distinct temporal dynamics (Dorn et al., 2006).

Consistent with emerging research implicating androgens in socioemotional sensitivity during adolescence (Andersen et al., 2024b; Tyborowska et al., 2024), we found that testosterone predicted internalizing symptoms even when accounting for the effects of estradiol and pubertal development. These findings suggest that testosterone confers a distinct risk pathway for emotional vulnerability.

Several mechanisms may underlie this association. Neuroimaging evidence links testosterone to heightened amygdala reactivity and altered prefrontal-limbic connectivity, potentially increasing emotional sensitivity to social evaluation and stress (Spielberg et al., 2015; Tyborowska et al., 2024; Vijayakumar et al., 2018). Pubertal testosterone increases coincide with reorganization of corticolimbic circuitry involved in threat processing and emotion regulation concurrently with maturing regulatory systems (Vijayakumar et al., 2023). Animal models demonstrate that pubertal androgens exert organizational effects on stress-responsive neural circuits, including lasting alterations in prefrontal inhibitory signaling and limbic dendritic morphology (Schulz and Sisk, 2016; Sisk and Zehr, 2005). Testosterone also modulates HPA axis functioning, potentially amplifying physiological stress responses (Handa and Weiser, 2014; Romeo, 2013). Finally, androgen-driven physical changes may increase exposure to social comparison and evaluative pressures, providing an indirect psychosocial pathway to symptoms (Ge and Natsuaki, 2009; Reynolds and Juvonen, 2011).

Observable pubertal development showed a complex pattern of associations with internalizing symptoms. At one assessment (Year 2, ages 11-12), higher rates of pubertal development change predicted lower internalizing symptoms, a pattern not observed at earlier or later timepoints. This suggests that the psychosocial consequences of observable maturation may vary across developmental stages and social contexts. During some developmental periods, more advanced maturation may confer social advantages that buffer against internalizing symptoms, whereas the same developmental changes may carry different implications during earlier or later pubertal stages (Natsuaki et al., 2009; Reynolds and Juvonen, 2011). This pattern diverges from a substantial literature linking accelerated pubertal development to elevated internalizing symptoms, which has primarily characterized associations with pubertal timing relative to peers rather than ongoing rates of change during mid-puberty. The distinction between timing and tempo may be consequential here, as early timing confers risk through social comparison and developmental mismatch mechanisms, whereas ongoing pubertal progression during mid-adolescence may carry different implications depending on developmental stage and social context (Mendle et al., 2010).

The divergent effects of testosterone and observable maturation underscore that endocrine processes capture biologically mediated risk not fully reflected in phenotypic maturation (Ge et al., 2001; Mendle et al., 2007). Gonadal hormones drive both observable physical maturation and concurrent reorganization of social-affective neural circuitry, such that phenotypic markers and neural development unfold as parallel outputs of the same underlying endocrine process (Goddings et al., 2014; Vijayakumar et al., 2018). Testosterone and estradiol act directly on developing amygdala and prefrontal circuitry involved in threat processing and emotion regulation, with hormonal effects on corticolimbic connectivity documented before or independent of visible physical change (Spielberg et al., 2015; Tyborowska et al., 2024; Vijayakumar et al., 2023). Physical markers of puberty have also been associated with neural development in social-affective circuitry, though these associations are thought to reflect shared hormonal drivers rather than independent effects of phenotypic change on brain development (Goddings et al., 2014). It is also possible that the psychosocial consequences of visible maturation, such as altered peer dynamics and social evaluation, themselves shape neuroaffective development through experience-dependent plasticity, a pathway that was not examined in the present study. These findings underscore the importance of assessing both hormonal and phenotypic indicators when evaluating pubertal contributions to mental health.

Our results further reveal stage-specific hormonal influences on affective vulnerability. During early-to-mid adolescence (ages 10-12), testosterone increases uniquely predicted internalizing symptoms, identifying a transient vulnerability window. Estradiol showed concurrent associations with symptoms at mid-to-late puberty and predicted subsequent phenotypic development change at later assessments, consistent with prior work linking estrogenic activity to observable maturation and affective processes in adolescent females (Andersen et al., 2024a; Angold et al., 1999). The testosterone-internalizing association was most robust during pubertal onset and attenuated by later adolescence. These patterns align with the hormonal influence hypothesis while highlighting developmental shifts in endocrine contributions to risk: androgenic processes may be particularly salient during pubertal onset, whereas estrogenic processes may exert stronger influence as adolescents progress toward mid-puberty.

Consistent with recent recommendations to model puberty as a coordinated system (Dorn et al., 2006; Shirtcliff et al., 2009), our modeling approach incorporated regulatory feedback mechanisms among pubertal hormones and observable development. We retained DHEA → testosterone coupling in multivariate models despite model instability in the freed bivariate specification, recognizing that DHEA functions as an upstream regulator in steroidogenic pathways and that estimation issues in specific parameterizations do not necessarily indicate absence of true effects (Kolenikov and Bollen, 2012; Labrie, 2010). We did not retain pubertal development → testosterone coupling to prioritize parsimony and avoid multicollinearity, given that observable maturation is primarily a consequence rather than driver of hormonal processes (Dorn et al., 2006). Over time, DHEA contributed to the pacing of downstream pubertal processes, such that higher DHEA levels predicted smaller subsequent increases in both testosterone and observable pubertal development. This pattern aligns with the view of DHEA as an upstream regulator whose influence shifts as levels approach adult concentrations, potentially serving as a developmental "brake" on further androgenic and physical pubertal development (Black et al., 2022; Havelock et al., 2004).

More broadly, these findings highlight the value of modeling pubertal development as an integrated biological process (Dorn et al., 2006; Mendle et al., 2019). DHEA, testosterone, and estradiol arise from interconnected steroid biosynthesis pathways with shared precursors and bidirectional conversion processes. Our analytic strategy isolates androgenic contributions while statistically accounting for estradiol, testing whether androgens confer unique risk beyond estrogenic influences. Although this approach necessarily simplifies the biochemical complexity of steroid interconversion (DHEA serves as a precursor to both testosterone and estradiol (testosterone itself aromatizes to estradiol centrally and peripherally) it proposes a framework for examining distinct developmental pathways to affective vulnerability in adolescent females. Such an approach can inform future neurodevelopmental research by pinpointing windows of heightened vulnerability, clarifying mechanisms linking puberty to mental health, and guiding targeted interventions.

Limitations

Several limitations warrant consideration. First, hormonal data were collected at most once annually, precluding capture of within-person variability in hormone concentrations, including diurnal fluctuation and menstrual cyclicity. Single annual assessments provide only snapshots of endocrine functioning and may underestimate associations driven by shorter-term dynamics. Although we included time since awakening as a covariate and modeled estradiol concurrently with symptoms to partially capture state-level associations, more intensive longitudinal sampling would improve precision in characterizing hormone-symptom coupling.

Second, pubertal development was assessed via caregiver rather than youth self-report. We selected parent-report based on measurement quality in early adolescence, when youth self-report shows lower reliability and higher missingness (Beltz et al., 2026). Nevertheless, subjective perceptions of pubertal maturation may be particularly relevant for psychosocial vulnerability (Dorn et al., 1990). Youth-reported development could show different associations with symptoms than parent-reported maturation, especially during mid-adolescence when self-perceptions become more salient (Pfeifer et al., 2013). Future work examining reporter-specific effects across developmental stages would clarify these dynamics.

Additionally, the univariate pubertal development model showed suboptimal fit by some indices (RMSEA = 0.114, SRMR = 0.221), though incremental fit indices approached acceptable levels (CFI = 0.908, TLI = 0.921). This likely reflects known psychometric limitations of the PDS in early adolescence, including measurement error inherent in capturing a continuous developmental process with discrete ordinal items (Beltz et al., 2026; Cheng et al., 2021). In complex latent change score models with multiple equality constraints, univariate fit may not fully predict performance in multivariate systems where shared variance and cross-construct associations can improve overall model specification(Grimm et al., 2016; Kievit et al., 2018). We retained pubertal development in subsequent models to test our primary hypotheses regarding observable maturation, consistent with theory-driven model-building approaches in longitudinal SEM (Little, 2024; McArdle, 2009).

Third, we did not adjust for race/ethnicity or socioeconomic status as covariates. These constructs reflect complex social and structural processes that cannot be reduced to statistical controls without altering the research question's conceptual framing (Helms et al., 2005; Semanaz et al., 2025). Examining whether pubertal hormonal pathways to symptoms differ across racial/ethnic or socioeconomic groups represents an important but distinct inquiry requiring dedicated moderation or stratified analyses (Trent et al., 2019). Our findings may not generalize uniformly across social contexts; future research should explicitly test these potential differences. Similarly, hormone samples were not anchored to menstrual cycle phase. Most participants had not yet experienced menarche at baseline and early waves, and date of last period showed substantial missing data among post-menarcheal participants. This unmeasured heterogeneity may attenuate hormone-symptom associations, particularly for estradiol.

Fourth, we focused exclusively on internalizing symptoms without examining externalizing outcomes. Testosterone has been linked to both internalizing and externalizing features in adolescent females, potentially through distinct mechanisms (Martel et al., 2025; Negriff and Susman, 2011). Including externalizing symptoms would clarify whether androgenic changes reflect domain-specific vulnerability or broader dysregulation.

Fifth, analyses were restricted to females. The puberty-internalizing association is particularly pronounced in females, motivating this focus. Hormonal trajectories and psychosocial correlates differ by sex. Testosterone increases more steeply in males and may confer distinct neurobiological or behavioral effects (Braams et al., 2015; Herting et al., 2014). Future work should examine parallel models in males to determine whether androgenic pathways are sex-specific or more general.

Sixth, estradiol measurement was limited to single annual assessments, constraining interpretation of estrogen-related effects. Emerging evidence suggests that sensitivity to estradiol variability, rather than absolute levels, may be particularly relevant for affective risk (Andersen et al., 2024a). Annual sampling cannot capture within-cycle fluctuations or rapid hormonal shifts characteristic of peri-puberty. Consequently, our estradiol findings reflect between-person differences at discrete timepoints rather than dynamic variability.

Finally, salivary assays capture free (unbound) hormone concentrations, reflecting systemic exposure rather than localized neural activity (Vining et al., 1983). Although salivary measures are widely used and biologically meaningful, they represent one index of endocrine function (Granger et al., 2004). More frequent sampling and multimodal hormonal assessment would strengthen future investigations.

Clinical Implications

These findings suggest that hormonal changes may provide information about internalizing symptom risk not fully captured by observable pubertal maturation. From a clinical perspective, this underscores the importance of considering biological maturation alongside psychosocial context when evaluating emotional risk during early adolescence. Traditional assessments often rely on phenotypic pubertal markers; however, endocrine processes, particularly testosterone increases, may confer vulnerability independent of visible development.

Importantly, these findings do not suggest routine endocrine screening is warranted. Rather, they highlight ages 10-12 as a window of heightened sensitivity when biological changes may intersect with emerging psychosocial stressors. Increased clinical attention to mood symptoms during early puberty, especially among youth experiencing rapid pubertal progression, may support earlier identification of difficulties.

The developmental specificity observed also has prevention timing implications. Universal interventions delivered before pubertal onset may build foundational emotion regulation and coping skills before peak hormonal shifts. Selective interventions targeting youth undergoing rapid endocrine change during early adolescence may provide additional support during vulnerability periods. Although further work is needed to determine whether hormonal indicators improve prediction beyond established risk markers, integrating biological and psychosocial perspectives may ultimately enhance developmental risk stratification and intervention design.

5. Conclusion

The present findings underscore that hormonal changes may convey information about risk for internalizing symptoms that is not captured by observable pubertal maturation. By modeling pubertal development as an integrated dynamic system, this study identified developmentally specific pathways linking testosterone to early adolescent internalizing symptoms after accounting for estradiol and physical maturation. These results highlight the importance of examining androgenic processes alongside estrogenic and observable pubertal changes and suggest that the window of early puberty may represent a period of heightened sensitivity for socioemotional development. Future research should examine neural mechanisms mediating testosterone effects and test whether these patterns generalize across diverse populations.

Figure 2. Final Integrated Path Model from the Latent Change Score Analysis.

Path model illustrating longitudinal associations among DHEA, testosterone, pubertal development, estradiol, and internalizing symptoms across 4 annual assessments (ages 9–13). Circles indicate latent true scores, triangles indicate latent change scores, and squares indicate observed variables. Bold solid lines represent significant paths; gray lines, nonsignificant paths. Δ indicates estimated change since the previous year; α, constant change; π, proportional change; γ, coupling parameter.

Highlights.

• Testosterone increases predicted internalizing symptoms in females ages 10-12.

• Estradiol was concurrently associated with symptoms in females ages 11-13

• Testosterone effects remained after controlling for estradiol.

• Ages 10-12 represent a transient vulnerability window for mood risk.

• Hormonal changes may inform timing of early identification efforts.