Positive Cortisol–Testosterone Hormonal Coupling Among Adolescents in Argentina and Jordan

ABSTRACT

Objectives

Puberty is regulated by the hypothalamic–pituitary–adrenal (HPA) and gonadal (HPG) axes. It has been proposed that if HPA and HPG coactivate during pubertal development, the hormones cortisol and testosterone would be positively coupled during puberty and decoupled postpuberty. Our objective was to test for hormonal coupling in less‐studied, non‐Western populations.

Materials and Methods

We analyzed longitudinal and cross‐sectional data from marginalized populations: Indigenous Qom/Toba females in Argentina (n = 46, 777 urine samples) and Syrian/Jordanian youth (n = 768, dried blood spots). We used Bayesian hierarchical models to assess the hypothesis that cortisol and testosterone are positively coupled during puberty but decouple at later stages.

Results

We found positive, age‐specific cortisol–testosterone coupling among adolescents in both populations, with patterns varying by age and sex. Coupling increased across pubertal ages but did not decline at older ages, contradicting the expectation that there is hormonal de‐coupling.

Discussion

This is the first study to demonstrate positive cortisol–testosterone coupling across adolescence in two socio‐ecologically distinct, non‐Western populations. While hormonal decoupling was not observed, coupling patterns suggest population‐level differences in pubertal timing. These findings challenge assumptions derived from Western‐based research and underscore the need for global, context‐sensitive models of adolescent development.

1. Introduction

Adolescence, spanning approximately ages 10–24, is a life stage marked by sociocultural transitions into adulthood alongside significant pubertal and developmental changes (Dorn et al. 2006; Reiches 2019). Puberty unfolds across early to mid‐adolescence with gradual changes in hormones, secondary sex characteristics, and pubertal linear growth, all facilitated by the hypothalamic–pituitary–gonadal (HPG) axis (Ellison 1994; Reiches and Ellison 2022; Worthman 1999). The HPG reactivates during puberty following dormancy in childhood and regulates the production and release of gonadal hormones such as testosterone (Ellis 2004; Ellison 2017b). Testosterone is integral to changes in secondary sex characteristics such as voice pitch, pubic hair, and lean muscle mass and is responsive to social stressors (Bogin 1999; DuBois and Shattuck‐Heidorn 2021). Testosterone exhibits exponential increases throughout puberty, then decelerates to adult levels, with generally higher mean levels in males but measurable levels in females (Ellison 2017a; Handelsman et al. 2016). While the role of the HPG in puberty is well characterized, less attention has been paid to the HPA axis, which also undergoes developmental changes in adolescence. The HPA axis has a range of metabolic functions, regulating responses to challenges and stressors, metabolism, mood, and immune function. An HPA end product—cortisol—is released in response to changes in metabolic rate and current or anticipated threats to survival or well‐being (Jimeno and Verhulst 2023; Sapolsky 2021; Sapolsky et al. 2000) with consequences for immune activation and increased glucose and blood flow (Sapolsky et al. 2000).

In adult populations, diverse psychosocial and nutritional stress measured by HPA function may suppress or inhibit HPG function and reproduction (Ellison 1990; Karsch et al. 2002; Poitras et al. 2024; Shirtcliff et al. 2015; Son et al. 2022). However, in adolescence, cortisol and testosterone may increase in concert to promote pubertal processes. HPA upregulation may occur during puberty for many reasons, including heightened neurological development, sensitivity to social context, and upticks in psychoemotional maturation. It is plausible that these functional increases in cortisol occur simultaneously, but independently of HPG upregulation, and/or that increases are not costly enough to warrant suppression of the HPA by the HPG (Flinn et al. 2011; Naulé et al. 2021; Shirtcliff et al. 2015). However, some researchers have proposed that the HPA axis is developmentally upregulated to help achieve the demands of adolescence—responding to psychosocial stress and facilitating physical and reproductive growth in concert with the HPG—such that cortisol and testosterone increase together in a functional manner or that the HPA upregulates the HPG (Marceau et al. 2015; Shirtcliff et al. 2015). Hormonal coupling—defined as the correlated changes in two or more hormones within a person or between individuals over time (Marceau et al. 2014; Zakreski et al. 2018)—is a concept that can be used to account for changes in both of these axes in order to explore the neuroendocrine and reproductive transition to adulthood. Hormonal coupling functionally implies that the co‐upregulation of the HPA and HPG is needed in order to facilitate relatively long‐term energetic (primarily anabolic) demands of puberty and adolescent physiological development, alongside the possibility that contextualized psychosocial stress may also impact biological outcomes and trade‐offs throughout this process (Lock and Kaufert 2001).

Some evidence of hormonal coupling has included positive associations of cortisol and testosterone across individuals sampled across the day, which could reflect the diurnal rhythms of both hormones or shared sources of variability (Black et al. 2018; Crewther et al. 2021; Dismukes et al. 2015; Harden et al. 2016). However, if HPA and HPG activity are developmentally modulated during puberty as hypothesized, then cortisol and testosterone measures across a sample population should show little to no positive statistical association before puberty, should associate positively in early adolescence (“couple”), and become more weakly associated as an adult phenotype is attained (“decouple”) (Marceau et al. 2015; Shirtcliff et al. 2015). Within groups of adolescents comprising multiple ages it can be assumed underlying pubertal development is variable. Thus, we expect to be able to detect hormonal coupling patterns across age groups within the adolescent period. Less positive associations or no associations at later ages would represent hormonal “de‐coupling,” whereas “coupling” that is progressively positive across ages would suggest a developmental coupling relationship (Moody et al. 2023; Phan et al. 2019; Zakreski et al. 2018). Some studies support this general hypothesis, finding HPA‐HPG coupling in early adolescence as compared to later adolescence and associations between positive hormonal coupling and increasing pubertal tempo or pubertal stage (Joos et al. 2018; King et al. 2020; Ruttle et al. 2015). In contrast, others do not find this association (Ankarberg‐Lindgren et al. 2020).

Many of these prior studies have employed limited sampling schemes compared to longitudinal data or limited age ranges to study the trajectories of cortisol–testosterone coupling at the population level. Most notably, this research has occurred mainly among adolescents within Western contexts with generally low levels of infectious disease and energetic stress, and relatively high rates of obesity and sedentism (Crewther et al. 2021; Henrich et al. 2010; King et al. 2020; Liebert et al. 2024; Marceau et al. 2014; Ruttle et al. 2015; Simmons et al. 2015; Turan et al. 2015; Zakreski et al. 2018). We may expect relatively higher testosterone and cortisol in Western contexts where there are observations of relatively higher gonadal and steroid hormonal profiles compared to Argentina and Jordan (Bribiescas 1998; Ellison et al. 2002; Gurven and Lieberman 2020; Núñez‐De La Mora et al. 2008; Rosetta 1996). There are likely different physiological profiles (Panter‐Brick et al. 2020) for adolescents outside of Western populations, though they are not well known due to the paucity of studies. Therefore, it may not be appropriate to generalize to other populations with socioecological conditions and hormonal patterning that may differ, underscoring the importance of understanding hormonal coupling across global contexts.

This study examines the age‐related hormonal coupling of cortisol and testosterone during adolescence. We opportunistically used data previously collected from two different study populations: Indigenous Qom/Toba girls aged 7–14 (n = 46 participants, n = 777 samples) from a single peri‐urban community, and Syrian refugee and Jordanian youth living in urban and rural communities in Northern Jordan, males and females aged 11–19 years (n = 768). The Qom/Toba are one of the most numerous remaining Indigenous peoples in the Gran Chaco region of South America, about thousands of whom live in the province of Formosa, Argentina (Lagranja et al. 2015; Olmedo et al. 2020; Valeggia et al. 2010). Colonialism and government policies in Argentina have destabilized and marginalized Qom/Toba and other Indigenous groups, forcing shifts from traditional subsistence practices to integration in the market economy (Biocca 2017; Olmedo et al. 2020; Pozzi and Nigra 2015). Most families in the community sampled for this research rely on government subsidies. Housing in the community ranges from brick homes with piped water to self‐constructed homes of available earthen and salvaged materials; there is no plumbing infrastructure. Rates of child and adult obesity are high due to food insecurity and dependency on limited, high‐carbohydrate, high‐fat staples like fried bread (Olmedo et al. 2020). In addition, prior research with the same study participants demonstrated a relatively high rate of adolescent obesity and earlier pubertal timing compared to other Indigenous Latin American girls (Martin and Valeggia 2018).

Arab adolescents in Jordan are ethnically and culturally diverse, coming from West Asia more broadly (El‐Abed 2021). The diversity of Arab youth in Jordan can be traced back to the origins of population groups in what is now Jordan before colonization, the establishment of Jordan as a nation following partition by Britain and France, the establishment of Transjordan, the Arab‐Israeli War, and refugee inflows from Palestine, Syria, and Iraq (Blanco‐Palencia 2021; El‐Abed 2021; Maggiolini and Ouahes 2021). In the current study, Arab youth in Jordan included Syrian refugees displaced into Jordan following the 2011 Civil War and Jordanian host teens from low‐income households living in nearby areas (Panter‐Brick et al. 2020). Both Syrian and Jordanian adolescents face socioeconomic marginalization, the effects of poverty, food insecurity, and water insecurity, and for Syrian adolescents, adversities related to displacement and armed conflict (Chen et al. 2019). There are varying estimates of obesity, with appreciable prevalence of thinness among adolescents living in Jordan (Ajlouni et al. 2020; Al‐Rahamneh 2020; Zayed et al. 2016). Pubertal timing among females in Jordan is estimated to be around 13 years of age (Bata 2012), and there are no studies to our knowledge on pubertal timing among males in Jordan.

The study populations in Argentina and Jordan differ significantly in their social and ecological characteristics and have distinct ancestral backgrounds. Syrian refugee and Jordanian youth, however, share close similarities in language, culture, ancestry, and environmental context. At the same time, both populations have undergone socioeconomic change over the last two decades, and each community has been impacted by social inequality and marginalization. For this particular study, we analyze Syrian refugee and Jordanian youth as one group to maximize the greatest number of participants for the current analysis. Although this analysis does not directly address the effects of psychosocial stress or marginalization on pubertal development in participants, the two study populations are relevant cases to consider hormonal and developmental variation in distinct contexts outside of the West. Our study assessed the developmental trajectories of cortisol–testosterone coupling using available data in Qom/Toba and Syrian/Jordanian groups in Argentina and Jordan.

1.1. Study Hypotheses

Consistent with the hypothesis that there is upregulated HPA and HPG activity earlier in puberty compared to later puberty, we predicted: (1) testosterone and cortisol would independently increase across age groups in all study participants and (2) for all participants, there would be positive interactions between age and cortisol on testosterone at earlier ages of adolescence, indicative of “hormonal coupling” and less positive, zero, or even negative associations indicating “decoupling” at relatively later ages. In addition, for Syrian/Jordanians, we expected (3) decoupling to occur at an earlier age among females compared to males due to earlier pubertal onset and progression, on average, compared to males. Although the age distributions were varied for Qom/Toba adolescents (ages 8–14) and Jordanian/Syrian adolescents (ages 10–19), underlying pubertal variability may be detectable via population‐level age trends in hormonal coupling with positive coupling at relatively younger ages (~9–14 or 15 years old) and decoupling at relatively older ages (~14–19 years old).

2. Methods

2.1. Study Design

2.1.1. Chaco Area Reproductive Ecology (CARE) Program Data Collection

CARE Program is a long‐running field research program aimed at studying the health and well‐being of Indigenous populations in Formosa, Argentina. Data for this study were collected between 2011 and 2015 from Qom/Toba participants in the peri‐urban community of Namqom as part of the Life History Transitions Study (NSF BCS‐0952264), aimed at documenting the hormonal, anthropometric, and behavioral correlates of crucial life transitions. All premenarcheal girls living in Namqom at the study onset were informed of the study and invited to participate. A total of 61 girls were recruited into the study between 2011 and 2013, when they were ages 7–10 years old, and remained in the study until reporting two to three menstrual cycles or otherwise leaving the study (Martin and Valeggia 2018). Age was calculated from dates of birth and field collection dates. The present analysis uses participant age, cortisol, and testosterone data collected from 46 female participants aged 7–14. Prior research calculated that this cohort's median age at menarche was 11.5 years old (Martin and Valeggia 2018).

Participants provided first‐morning void urine samples twice weekly over 1 month, every 3 months, for a total of eight samples per month every 3 months. The urine sampling design was optimized for cycling women (O'Connor et al. 2001), but implemented for both peri‐menopausal and pubertal participants in the Life History Transitions study to maximize measurement of hormone variation in puberty and standardize field data collection across participants. Participants' urine specimens were collected using sterile cups, transported to a research site in Formosa, and aliquoted in triplicate. Aliquots were frozen at −20°C until transport to the United States, where they were stored at −80°C in the Yale Reproductive Ecology Laboratory. A total of 777 urine samples from 46 participants used in this analysis were transferred on dry ice to The University of Washington (UW) Biodemography Lab in the Center for Studies in Demography and Ecology in August 2021 and stored at −80°C.

2.1.2. Advancing Adolescents (Nubader) Evaluation Data Collection

The Advancing Adolescents (Nubader) evaluation study was conducted in Jordan, close to the Syrian border, from 2015 to 2017 (Panter‐Brick 2022). In all, 817 Syrian refugee and Jordanian non‐refugee adolescents (aged 10–19) participated in a mixed‐method study with an intervention implemented by Mercy Corps (Panter‐Brick et al. 2020). This research aimed to understand how a community‐level psychosocial intervention influenced biological, cognitive, and psychosocial outcomes among Syrian refugee and Jordanian non‐refugee youth living side‐by‐side in relative socioeconomic disadvantage (Panter‐Brick et al. 2020). The study design included a randomized controlled trial of an 8‐week program to alleviate psychosocial stress. It featured biological, psychosocial, and cognitive measures at three time points (baseline, postintervention, and 1‐year follow‐up). During face‐to‐face interviews, eight fieldworkers collected biomarkers, such as DBS and hair cortisol samples, alongside other sociodemographic and mental health data (Panter‐Brick et al. 2019, 2020). The present analysis uses age, sex, free testosterone, and free cortisol data collected from 768 participants (ages 11–19) at the study baseline before experiencing the intervention.

DBS were taken in the morning to early afternoon with two to five finger‐prick blood spots on filter paper using disposable lancets, following standard protocols (McDade et al. 2007; Panter‐Brick et al. 2019). Cortisol and testosterone were analyzed from the same DBS samples reflecting the same temporal congruence and diurnal phase for each participant. Assaying cortisol again in DBS (as opposed to hair alone) allowed for an additional robustness check of hormonal coupling using both mediums (McDade et al. 2007; Shirtcliff et al. 2001). DBS samples were initially stored in Amman at −20°C until transport to the Yale Reproductive Ecology Lab, where they were stored at −80°C, then shipped on dry ice to the UW Biodemography Lab in 2021, where they are currently stored at −80°C. Hair cortisol values were also available from lab analyses of hair samples (~100 strands), showing that females, compared to males, had higher cortisol values (Dajani et al. 2018).

2.2. Laboratory Methods

2.2.1. Urinary Assays (CARE)

Testosterone assays were conducted by authors MM and MC using commercial EIA kits (Testosterone DetectX, Arbor Assay) at the Yale Reproductive Ecology Laboratory in 2017; inter‐ and intra‐assay CVs were 7.8% and 8.5%. Urine samples were analyzed using a validated in‐house urinary cortisol assay (Munro and Stabenfeldt 1985) by authors D.J.G. and E.B. in Fall 2021; intraassay CVs for cortisol assays were 4.9%, 5.9%, and 14.1%, and interassay CVs were 13.4%, 12.2%, and 21.93% for low, medium, and high controls. The highest interassay CV (21.93%) was elevated, which may suggest higher error for higher range samples across plates. Final concentrations for testosterone and cortisol were adjusted for specific gravity (Miller et al. 2004; Nokoff et al. 2019).

2.2.2. DBS and Hair Assays (Nubader Evaluation)

Free cortisol and testosterone—the unbound components of these hormones representing the bioavailable or biologically active components—were quantified in 1009 dB samples (n = 768 participants) by author J.K. using Multiple Reaction Monitoring with Waters Xevo‐TQS HPLC‐Mass Spectrometer at the University of Washington School of Pharmacy's Mass Spectrometry Center in 2024. Two 7 mm punches (approximately equivalent to 13 μL, though DBS punches do not scale linearly with volume) from the center of the dried blood spots (total 14 mm) were mixed with 396 μL H₂0 and 4 μL IS mix (100 ng/mL) and sonicated for 1 h. One milliliter of MtBE was added to this mixture, vortexed for 30 min, and centrifuged for 5 min at 18,000 rpm (room temperature). Next, the organic layer was dried under N2 gas. Samples were reconstituted in 50 μL H₂O/MeOH 50/50 and 20 μL were injected onto the column. The lower and upper limits of detection for testosterone were 0.1 and 0.50 ng/mL and for cortisol 1 and 500 ng/mL.

Hair samples (n = 741) were stored and analyzed at the Drug Safety Laboratory of the Robarts Research Institute at the University of Western Ontario (Etwel et al. 2014). Between 0 and 2 cm of undyed brown or black hair were measured, cut, weighed, and washed twice (3 min with 3 mL isopropanol), then dried for a maximum of 5 h. Dried samples were then ground and extracted in 2 mL of methanol in an incubator shaker for 16 h (56°C), then reconstituted in 0.25 mL of phosphate‐buffered saline with a pH of 8. The reconstituted hair samples were then assayed for cortisol using a modified commercially available salivary cortisol kit. The intra‐ and inter‐day CVs for the hair samples used to assay for cortisol were 3.6% and 8.6% (Dajani et al. 2018; Panter‐Brick et al. 2019).

Cortisol measured in hair reflects a longer timescale (~1–3 months from 1 to 3 cm from the scalp) and is not affected by diurnal variability. DBS may be influenced by acute stressors and diurnal variability, in general, which may be a limitation of the current design with sampling at one timepoint. However, unmeasured variability in cortisol concentrations assayed from hair resulting from exposure to UV light and hair washing may pose limitations to observed relationships (Hamel et al. 2011; Liu and Doan 2019). Thus, we assayed cortisol again in DBS to address some of these potential limitations of hair samples and to be temporally congruent with assays of testosterone in DBS. In addition, the assays of cortisol and testosterone in DBS allowed for an additional robustness check of hormonal coupling using both mediums (McDade et al. 2007; Shirtcliff et al. 2001).

2.3. Statistical Methods

We first produced summary statistics for each respective study population (Table 1). We then examined whether log cortisol increased across age groups among Syrian/Jordanian participants (Table 4), as age–cortisol relationships were reported previously among the same Qom/Toba participants (Glass et al. 2024). Second, we modeled longitudinal hormonal coupling among Qom/Toba participants to make use of their longitudinal data (Table 2). This longitudinal model of hormonal coupling offers only a linear model of log testosterone as a function of continuous age and log cortisol (R package brms with MCMC inference). We chose log testosterone to be the outcome variable across hormonal coupling models, under the premise that the HPA may upregulate the HPG in puberty and that the relationship between cortisol and testosterone may change with age and development moving from positive coupling to decoupling.

TABLE 1.

Participant characteristics (median (SD) or n (%)).

Variable	n	Qom/Toba adolescents	n	Jordanian/Syrian adolescents
		Summary		Summary
Age (years)	46	10.29 (1.25)	700	14.36 (1.74)
Cortisol (ng/mL)	777	145.26 (141.37)	754	176.48 (96.36)
Males		—	323	189.6 (99.5)
Females	777	—	331	160.5 (89.11)
Testosterone (ng/mL)	1106	19.81 (15.75)	702	1.32 (3.64)
Males		—	397	3.5 (4.13)
Females	1106	19.81 (15.75)	303	0.82 (0.84)
Sex	46		760
Female		46 (100%)		332 (43.7%)
Male		—		428 (56.3%)
Height (cm)	46	138 (8.6)	183	160 (11.6)
Weight (kg)	46	34.5 (11.6)	183	53.2 (12.6)

TABLE 4.

Associations between age and log cortisol (from blood and hair) among female and male adolescents in Jordan (β estimate and 95% credible interval).

Coefficient	Model A n = 754 participants	Model B n = 741 participants
Intercept	5.18 (5.14, 5.21)	0.83 (0.80, 0.87)
Age group
11–12.9 years	−0.11 (−0.18, −0.05)	−0.08 (−0.12, −0.02)
13–13.9 years	−0.08 (−0.12, −0.04)	−0.05 (−0.08, −0.02)
14–14.9 years	−0.03 (−0.08, −0.01)	−0.02 (−0.05, 0.00)
15–15.9 years	0.02 (−0.02, 0.04)	0.01 (−0.02, 0.03)
16–16.9 years	0.08 (0.04, 0.12)	0.05 (0.02, 0.08)
17–19.9 years	0.13 (0.07, 0.20)	0.09 (0.04, 0.15)

Note: Bold font denotes a credible interval that does not cross zero.

TABLE 2.

Longitudinal hormonal coupling (log testosterone outcome) across age among Qom/Toba females (β estimate and 95% credible interval).

Coefficient	Model A—Age only n = 1106 observations, n = 46 females	Model B—Cortisol only n = 668 observations, n = 46 females	Model C—Age and cortisol n = 668 observations, n = 46 females	Model D—Age cortisol interaction n = 668 observations, n = 46 females
Intercept	0.19	0.68	−0.57	0.81
Age	0.11 (0.8, 0.12)	—	0.12 (0.10, 0.14)	−0.01 (−0.13, 0.10)
Log cortisol	—	0.24 (0.18, 0.30)	0.24 (0.18, 0.29)	−0.39 (−0.94, 0.15)
Age × Log cortisol Interaction	—	—	—	0.06 (0.01, 0.11)

Note: These models are brms models (with weakly informative priors) estimating the linear effect of age, cortisol, and their interaction on longitudinal log testosterone. Bold font denotes a credible interval does not cross zero.

We also implemented models with categorical age groups to allow for the estimation of progressively positive coupling or decoupling across age groups. Treating age categorically enabled us to explore the possibility of nonlinear trajectories in log testosterone overall and to quantify changes in coupling, and potentially decoupling, with different effect estimates for each age group. For age group models, we used slightly informative random walk 2 (RW2) priors that allowed us to share information across age groups, as opposed to making estimations within each age group alone (see Table S1 comparing an IID prior that uses data within age groups and a RW2 prior that leverages information across groups). A RW2 prior on our age group coefficients assumes that the estimated slope in the linear relationship between cortisol and testosterone for any given age group is related to the estimated slope of the two preceding and succeeding age groups. This smoothing prior offers more precision in the estimated relationship between age group and log cortisol on log testosterone, especially at the tails of the age distributions where there are relatively fewer observations (Figure S3). In addition, the INLA model specification included a before account for dependence among individual observations, similar to the use of random intercepts in frequentist linear mixed‐models.

The construction of age groups (half years vs. whole years) was chosen based on the granularity of chronological age, retaining within‐individual variability (two to three observations per girl, per half year) in the case of the Qom/Toba participants, and choosing a minimum of 5 age‐bin categories for the RW2 before be defined. These Bayesian models were estimated using the R package INLA (R package INLA version 23.05.30, R version 4.3.3), which is a computationally efficient alternative to MCMC inference that uses Integrated Nested Laplace Approximation (Martino and Riebler 2019; Ruiz‐Cárdenas et al. 2012) to approximate posterior distributions. All priors were specified as penalized complexity priors (Fuglstad et al. 2019; Simpson et al. 2017).

Tables 3, 4, 5, 6 (Tables S2 and S3) show log testosterone as the outcome with up to four submodels (A–D) per table: (A) effect estimates of age group only, (B) log cortisol only, (C) age group and log cortisol, and (D) an interaction between age group and log cortisol that represents age‐specific effect of log cortisol on log testosterone. We interpreted “hormonal coupling” (D models) by the patterning of the age group and cortisol interaction terms as positive or negative, their credible intervals, and their change or lack of change between age groups (e.g., progressively positive). “Decoupling” was indicated by patterning of the interaction terms becoming less positive, weaker, or showing no association. The equation and assumptions for the D models in Tables 3, 4, 5, 6 using INLA are described (code snippets in Figures S4 and S5).

TABLE 3.

Hormonal coupling (log testosterone outcome) across age groups among Qom/Toba females (β estimate and 95% credible interval).

Coefficient	Model A—Age group only N = 1106 observations	Model B—cortisol only N = 696 observations	Model C—Age group and cortisol N = 696 observations	Model D—Age specific cortisol (interaction) N = 696 observations
Intercept	1.30 (1.23, 1.36)	0.56 (0.43, 0.69)	0.67 (0.54, 0.78)	—
Age group		—
8–8.49 years	−0.32 (−0.40, −0.25)		−0.32 (−0.39, −0.25)	0.44 (0.15, 0.69)
8.5–8.9 years	−0.30 (−0.35, −0.24)		−0.29 (−0.34, −0.24)	0.44 (0.25, 0.65)
9–9.49 years	−0.25 (−0.29, −0.20)		−0.25 (−0.30, −0.21)	0.47 (0.30, 0.66)
9.5–9.9 years	−0.21 (−0.26, −0.16)		−0.22 (−0.26, −0.18)	0.48 (0.34, 0.62)
10–10.49 years	−0.04 (−0.08, 0.00)		−0.05 (−0.09, −0.01)	0.49 (0.34, 0.62)
10–10.9 years	−0.05 (−0.10, −0.01)		−0.05 (−0.09, −0.01)	0.64 (0.51, 0.77)
11–11.49 years	−0.14 (−0.18, −0.09)		−0.12 (−0.16, −0.08)	0.61 (0.47, 0.74)
11.5–11.9 years	0.04 (−0.01, 0.09)		0.04 (−0.01, 0.08)	0.63 (0.43, 0.81)
12–12.49 years	0.19 (0.13, 0.27)		0.21 (0.15, 0.28)	0.73 (0.38, 1.04)
12.5–12.9 years	0.24 (0.17, 0.31)		0.24 (0.18, 0.31)	0.81 (0.55, 1.06)
13–13.49 years	0.34 (0.24, 0.44)		0.34 (0.25, 0.44)	0.81 (0.35, 1.19)
13.5–14 years	0.50 (0.37, 0.62)		0.49 (0.37, 0.60)	0.92 (0.31, 1.35)
Log cortisol	—	0.29 (0.24, 0.35)	0.29 (0.24, 0.34)	—
Age group × Log cortisol	—	—	—
8–8.49 years				0.24 (0.13,0.36)
8.5–8.9 years				0.25 (0.16,0.34)
9–9.49 years				0.26 (0.19,0.33)
9.5–9.9 years				0.27 (0.21,0.33)
10–10.49 years				0.28 (0.23,0.33)
10–10.9 years				0.29 (0.24,0.35)
11–11.49 years				0.31 (0.25,0.37)
11.5–11.9 years				0.33 (0.25,0.40)
12–12.49 years				0.35 (0.25, 0.46)
12.5–12.9 years				0.37 (0.24, 0.52)
13–13.49 years				0.39 (0.23, 0.61)
13.5–14 years				0.41 (0.22, 0.67)

Note: The estimates are drawn from the posterior joint distribution. Note that model D does not need an estimated main effect for cortisol because INLA is estimating slopes for every group, as opposed to a main effect plus an age‐specific slope. Bold font denotes a credible interval that does not cross zero.

TABLE 5.

Hormonal coupling (log testosterone outcome) across age groups among female adolescents in Jordan (β estimate and 95% credible interval).

Coefficient	Model A—Age group only N = 332 observations	Model B—Log cortisol only N = 332 observations	Model C—Age group and log cortisol N = 332 observations	Model D—Age specific cortisol (interaction) N = 332 observations
Intercept	−0.33 (−0.43, −0.23)	−2.25 (−3.26, −1.30)	−2.02 (−3.03, −1.03)	—
Age group		—
11–12.9 years	−0.27 (−0.43, −0.11)		−0.23 (−0.40, −0.08)	−2.08 (−3.82, −0.14)
13–13.9 years	−0.16 (−0.25, −0.07)		−0.14 (−0.23, −0.05)	−2.10 (−3.46, −0.65)
14–14.9 years	−0.05 (−0.09, 0.01)		−0.04 (−0.08, 0.01)	−2.01 (−3.13, −0.84)
15–15.9 years	0.06 (0.02, 0.11)		0.05 (0.01, 0.11)	−2.01 (−3.22, −0.85)
16–16.9 years	0.16 (0.07, 0.26)		0.14 (0.04, 0.23)	−1.99 (−3.45, −0.48)
17–19.9 years	0.26 (0.10, 0.41)		0.22 (0.06, 0.38)	−2.29 (−4.26, −0.29)
Log cortisol	—	0.38 (0.19, 0.58)	0.33 (0.14, 0.53)	—
Age group × Log cortisol	—	—	—
11–12.9 years				0.28 (−0.1, 0.63)
13–13.9 years				0.31 (0.02, 0.57)
14–14.9 years				0.33 (0.12, 0.55)
15–15.9 years				0.35 (0.14, 0.58)
16–16.9 years				0.38 (0.08, 0.65)
17–19.9 years				0.39 (0.01, 0.77)

Note: These models are INLA models estimating the nonlinear effect of age using age group, and the interaction between age group and log cortisol on log testosterone. The estimates are drawn from the posterior joint distribution. Bold font denotes a credible interval that does not cross zero.

TABLE 6.

Hormonal coupling (log testosterone outcome) across age groups among male adolescents in Jordan (β estimate and 95% credible interval).

Coefficient	Model A—Age group only N = 428 observations	Model B—Log cortisol only N = 428 observations	Model C—Age group and log cortisol N = 428 observations	Model D—Age specific cortisol (interaction) N = 428 observations
Intercept	1.11 (1.00, 1.20)	−2.14 (−3.55, −0.708)	−0.25 (−1.38, 0.84)	—
Age group		—
11–12.9 years	−1.31 (−1.49, −1.13)		−1.27 (−1.45, −1.12)	−0.95 (−2.99, 1.09)
13–13.9 years	−0.78 (−0.88, −0.68)		−0.75 (−0.85, −0.66)	−0.78 (−2.23, 0.63)
14–14.9 years	−0.25 (−0.30, −0.16)		−0.24 (−0.28, −0.10)	−0.37 (−1.57, 0.68)
15–15.9 years	0.27 (0.22, 0.38)		0.26 (0.22, 0.42)	−0.26 (−1.56, 0.96)
16–16.9 years	0.78 (0.68, 0.88)		0.76 (0.66, 0.87)	−0.17 (−1.95, 1.68)
17–19.9 years	1.29 (1.09, 1.46)		1.24 (0.97, 1.42)	−0.56 (−3.00, 2.04)
Log cortisol	—	0.58 (0.31, 0.86)	0.26 (0.05, 0.47)	—
Age group × Log cortisol	—	—	—
11–12.9 years				0.13 (−0.26, 0.52)
13–13.9 years				0.20 (−0.07, 0.48)
14–14.9 years				0.26 (0.06, 0.49)
15–15.9 years				0.33 (0.1, 0.58)
16–16.9 years				0.40 (0.08, 0.73)
17–19.9 years				0.48 (0.03, 0.92)

Note: These models are INLA models estimating the nonlinear effect of age using age group, and the interaction between age group and log cortisol on log testosterone. The estimates are drawn from the posterior joint distribution. Bold font denotes a credible interval that does not cross zero.

Gaussian likelihood:

$$\log \left({\text{testosterone}}_i\right)\mid a,{\text{cortisol}}_{\mathit{ai}}\sim \mathcal{N}\left({\mu }_{\mathit{ai}},{\sigma }^2\right)$$

Mean model equation:

$$\begin{array}{c}\log \left({\text{testosterone}}_{\mathit{ai}}\right)={\mu }_{\mathit{ai}}={\beta }_a·{\mathit{\text{AgeGroup}}}_{\mathit{ai}}+u_a·\log \left({\text{cortisol}}_{\mathit{ai}}\right)\hfill \end{array}$$

where ${\beta }_a$ is the intercept for age group a and $u_a$ is the random effect serving as the linear slope between log cortisol and log testosterone for age group a. The RW2 prior on $u_a$ can be written as a function of second differences and the precision parameter, or inverse of the variance, $\tau$:

RW2 priors:

$$u_a\sim \text{RW2}\left(\tau \right)$$

$$\left(u_a-u_{a-1}\right)-\left(u_{a-1}-u_{a-2}\right)~\mathcal{N}\left(0,{\tau }^{-1}\right).$$

With the CARE study, we used longitudinal data (n = 46 participants and n = 668 samples). There were a total of 1106 testosterone and 777 cortisol samples (median 24 samples per girl ±1), and we down‐sampled to 668 samples of cortisol and testosterone to retain the same number of cortisol and testosterone samples per age half‐year and to retain as much within‐person precision as possible for the age groupings (Tables 2 and 3, Supporting Information Text and Figures 1 and 2). Associations between age and cortisol were published previously among the same sample of Qom/Toba participants, with no strong prior relationship between adiposity and cortisol levels or joint effects of age, cortisol, and adiposity on age at menarche (Glass et al. 2024).

FIGURE 1.

Age‐specific estimated effect of log cortisol on log testosterone for each age group (Table 3, Model D) among Qom/Toba participants.

FIGURE 2.

(a) Age‐specific estimated effect of log cortisol on log testosterone for each age group (Tables 5 and 6, D Models) among Syrian/Jordanian male and female participants. (b) Repeats (a) with the addition of interaction and adjusted estimates. These refer to the effect of log cortisol on log testosterone after adjusting for age group, representing an averaged effect and the cortisol and age interaction. The coefficients for age groups refer to the age‐specific effect of cortisol on log testosterone (Tables 5 and 6, C and D Models).

With the Nubader evaluation study, we restricted our analysis to samples provided by 768 Syrian and Jordanian participants at the point of study entry (preintervention) to maximize the greatest number of participants across adolescence (ages 11–19). We presented basic models assessing associations between age and log cortisol in blood and hair (Table 4). We ran the blood‐free cortisol–testosterone coupling models with 1–2‐year age groups (Tables 5 and 6), stratified by sex to account for sex‐specific differences in absolute hormone levels and sex‐specific differences in the timing of puberty. To explore whether there were differences in coupling based on the sampling in blood versus hair, we ran sensitivity analyses assessing blood‐free testosterone and hair cortisol coupling (n = 745) from the same participants (Tables S2 and S3).

Among the 768 individuals from the Nubader evaluation study that had usable DBS samples at enrollment, there were 4 participants with a missing age value and no recorded date of birth. There were 10 participants for whom there was not enough sample to assay cortisol, and there were 52 participants for whom testosterone was below the limit of detection. We were not able to rerun these samples due to limited sample availability, and current best practices suggest caution with extrapolation (Duggan 2019). Participants for whom testosterone was below the limit of detection were 11–13‐year‐olds, 54% female and 46% male. There were no participants for whom free cortisol fell below the limit of detection. Therefore, for analyses only with cortisol and age, there were 754 observations; for those involving age, testosterone, and cortisol, there were 702 observations.

2.4. Ethics

Adolescents and/or their adult caregivers provided verbal informed consent in Spanish (CARE project) or written informed consent in Arabic (Nubader evaluation project). Formal approval was also obtained from the Prime Minister's Office of Jordan for the Nubader evaluation project. The research studies were approved by the internal review boards of the University of Pennsylvania (Protocol #811200) and Yale University (HSC Protocols #1406014104, #1502015359) for original data collection and ongoing analysis (HSC #2000026021). Additional approvals from The UW for ongoing data analysis using de‐identified data from the Nubader evaluation project were applied under two protocols as required by two funding mechanisms (# STUDY00017769 and # STUDY00018065).

3. Results

3.1. Participant Characteristics

Participants in Argentina were female, 7–14 years old, with a median age of 10.29 (Table 1). Their median cortisol level was 145.26 ng/mL, and their testosterone level was 19.81 ng/mL. Participants in Jordan (332 females and 428 males) were 11–19 years old, with a median age of 14.36. The median free cortisol level was 176.48 ng/mL, and free testosterone was 1.32 ng/mL. The average height and weight were 138 cm and 34.5 kg for Qom/Toba participants and 160 cm and 53.2 kg for a small subset of participants in Jordan who provided height and weight data (Table 1).

3.2. Cortisol–Testosterone Coupling in Argentina

We hypothesized that longitudinal testosterone and cortisol would independently increase across age. As expected, in longitudinal, linear models with weakly informative priors using R package brms, we found an independent effect of age on log testosterone (estimate = 0.11, credible interval (CI) = (0.08, 0.14), Table 2, Model A). However, we did not find an effect of age on log cortisol, as reported elsewhere in the same study cohort (Glass et al. 2024). We found mean log cortisol was positively associated with mean log testosterone after adjusting for age (0.24, CI = (0.18, 0.29)) (Table 2, Model C). The interaction between continuous age and log cortisol had a small positive effect on log testosterone (0.06, CI = (0.01, 0.11)) (Table 2, Model D).

In age‐group models among the Qom/Toba sample, we found progressively positive associations between age group and log testosterone, with age group estimates becoming positive at ages 11.5–11.9 (Table 3, Model A), a positive effect of cortisol alone on log testosterone (0.29, CI = (0.24, 0.35), Model B), and relatively unchanged associations with the inclusion of both age group and log cortisol in the model (Table 3, Model C). We found a weak but increasingly positive coupling trend among Qom/Toba adolescents, as demonstrated by the age‐specific effect of cortisol on testosterone per age group (Table 3, Model D and Figure 1). The magnitude of the hormonal coupling relationship in this sample increased on average across ages 10–14. As expected, the positive association between cortisol and testosterone was weaker at earlier ages (ages 8–9) compared to later ages (Figure 1). We did not find evidence of decoupling at later, postmenarcheal ages (ages 13–14).

3.3. Cortisol–Testosterone Coupling in Jordan

Similar to our prediction among the Qom/Toba, we hypothesized testosterone and cortisol would independently increase across age among Syrian/Jordanian adolescents. In models examining the effects of age on log cortisol, we found a progressively positive relationship between age group and log cortisol, with the strongest coefficient occurring among 17–19.9‐year‐olds (Table 4, Model A). We additionally modeled this relationship with previously assayed hair cortisol and found a less robust relationship between age and hair cortisol. However, we did find for the earliest and latest age groups, there were positive associations between age and log cortisol with relatively smaller coefficients than that of log free cortisol in blood (Table 4, Model B). We found progressively positive, robust effects of age on free log testosterone among both male and female individuals in the Nubader evaluation study, with relatively larger coefficients for age groups among males compared to females (Tables 5 and 6 Model A, Figure 2a,b). However, there was an overlap in the distributions of age coefficients for males and females. As the age group increased, the effect size of age on log testosterone grew larger and became fully positive at the 15–15.9 age groups (Table 5, Model A and Table 6, Model A).

There were positive associations between log cortisol and log testosterone among females (Table 5, Model B) and among males (Table 6, Model B) with larger coefficient estimates for cortisol in males compared to females. In models examining the effects of log cortisol on log testosterone adjusting for age group, there were significant associations for both age group and log testosterone (Tables 5 and 6, Model C). In models of hormonal coupling, we found that age‐specific cortisol coefficients increased across the adolescent period for both males and females. While we did not find the hypothesized pattern of de‐coupling, the age‐specific cortisol coefficients were progressively positive, but there was less change in coefficients at later ages for females (Table 5, Model D). The age‐specific effect of cortisol became stronger with larger increases in coefficients (Table 6, Model D). These coupling relationships are shown in Figure 2a, with a comparison of the coefficients and 95% CIs for the age‐specific cortisol coefficients (interactions, derived from Tables 5 and 6, Model D) to the overall effect of cortisol on log testosterone (adjusted for age, derived from Tables 5 and 6, Model C) in Figure 2b.

Finally, we tested the hormonal coupling hypothesis using log‐free testosterone assayed from dried blood as an outcome and log cortisol assayed from hair to explore whether there were differences in using hair versus dried blood samples. The interpretation of hair cortisol may reflect “chronic” HPA activity representing weeks to months, and free cortisol assayed in dried blood is likely to represent bioavailable cortisol on a shorter timescale (DeCaro and Helfrecht 2022; McDade et al. 2007). We did not find associations between log hair cortisol or the age‐specific coefficients of log hair cortisol on testosterone in males or females (Tables S2 and S3).

4. Discussion

Our study supports the general premise of hormonal coupling reflected in the evidence of positive hormonal coupling among Qom/Toba adolescents and Syrian/Jordanian, but finds no evidence of “decoupling” as hypothesized. Although we were not able to directly assess pubertal status, patterns of hormonal coupling across age groups in both populations may further indicate differences in relative pubertal timing, which will be discussed. The present study is one of the few studies that have used both longitudinal and cross‐sectional data to model the hormonal coupling of cortisol and testosterone across mixed‐age adolescents. Our study may be among the first to examine cortisol–testosterone coupling among adolescents outside Western contexts, where physiological and hormonal profiles may differ based on numerous latent social and ecological factors, warranting further investigation in future studies.

Overall, we found that among the Qom/Toba and Syrian/Jordanian participants, mean testosterone increased with age, as expected. In contrast, cortisol only increased across ages among Syrian/Jordanian participants. It is possible that the population differences in age‐related cortisol increases may be attributable to body size and growth rates, but we were unable to formally assess this due to the paucity of anthropometric data within the Nubader evaluation study. Furthermore, as previously reported, we found no associations between cortisol, adiposity, and most measures of linear growth among the Qom/Toba participants (Glass et al. 2024). However, adolescent‐aged increases in cortisol among Syrian/Jordanian participants are consistent with other population‐based studies showing that morning cortisol increases in adolescence or is associated with age, which may reflect underlying changes in pubertal development (Liebert et al. 2024; Sabbi et al. 2020; Shenk et al. 2022). A strength of our modeling approach for this analysis was our ability to maximize mixed‐age groupings from longitudinal and cross‐sectional data, covering a wide span of the pubertal transition within adolescence. Among the Qom/Toba, there was repeated sampling over at least 1 year per individual. Despite there being a small overall sample size, we were able to account for variability within an individual over a longer period of time in relation to the population‐level inferences. Among the Syrian/Jordanian adolescents, we did not have repeated sampling, but there was the benefit of a large overall sample size with greater information power within each age group. While neither approach could fully account for daily, diurnal variability, drawing from both of these study designs and populations enables greater population‐level assessment of hormonal patterning.

We found positive associations between cortisol and testosterone in partial models adjusting for age among Qom/Toba and Syrian/Jordanian youth. This may be because they are both independently involved in pubertal processes but may not be fully integrated when adjusting for age (and, therefore, masking effects of puberty). Another plausible explanation is that they may be independently related to body composition and energy use. Though we cannot test this, it may be the case that more well‐nourished individuals have relatively higher levels of cortisol and testosterone (Brown et al. 2024; Cohen et al. 2020; Jimeno and Verhulst 2023; Magid et al. 2018; Moriarty‐Kelsey et al. 2010).

Our models of hormonal coupling, on the other hand, indicate some level of integration that is potentially specific to the pubertal period in adolescence. Overall, we found some evidence of positive testosterone‐cortisol hormonal coupling among Qom/Toba adolescent females and among Syrian/Jordanian males and females who were in the pubertal transition. This finding aligns with research showing positive cortisol–testosterone coupling across age (Harden et al. 2016). One of the strengths of the current analysis is the maximization of longitudinal data among Qom/Toba females and mixed age groups among Syrian/Jordanian males and females with larger sample sizes per age group. The present analysis contributes to understanding population‐level pubertal hormone trends that may follow similar trajectories to sex steroids and linear growth‐related hormones (Ellison 2017a).

In the complete models testing for hormonal coupling, we find some evidence among Qom/Toba females that there are weak but positive increases in the age‐specific cortisol relationship with testosterone. The credible intervals for these interaction coefficients overlap, indicating a weak interaction. Still, it may suggest that the association between cortisol and testosterone may be more relevant to later pubertal processes around mid‐puberty and postmenarche among Qom/Toba females, for whom the median age at menarche is 11.5 (Martin and Valeggia 2018). Among Syrian/Jordanian adolescents, we find more evidence of increasingly positive coupling across age groups from ages 11 to 19 among males and females. Although the age‐specific coefficients of cortisol and their credible intervals overlap with other age groups, there are still associations in every age group and across adolescence. While we did not have a formal assessment of pubertal timing, some prior research suggests in girls that the median age at menarche for Jordanian and Syrian girls in the region may be between 13 and 14 years old (Bata 2012; Ghandour et al. 2023). Future research on this topic could implement an age‐matched design between Qom/Toba or Jordanian/Syrian adolescents and participants from Western contexts (e.g., United States or United Kingdom) accounting for formally assessed pubertal status and other confounders not addressed here such as current psychosocial stress. Future planned research will directly explore the relationship between early adversity, refugee status, and hormonal coupling.

Interestingly, among Syrian/Jordanian participants, there was more significant variability in the age‐specific cortisol coefficient sizes for males compared to females, with more strongly positive trends across ages for males (Figure 2a). This differs from prior studies that showed more positive coupling among females than males (Black et al. 2018; Matchock et al. 2007; Ruttle et al. 2015). This may be reflecting average differences in pubertal timing between males and females. Although associations are positive for females at the latest adolescent ages, they are weaker between age groups at older ages, and males continue to have more robust increases between age groups at older ages. In contrast to our original hypothesis, we did not see strong evidence of decoupling in either study population. It is plausible that among the Qom/Toba, who were only observed through age 14 years, decoupling would be observed with further sampling beyond age 14. However the age range was much larger for Jordanian and Syrians (11–19), which should have been sufficient to observe decoupling. Future research should further investigate hormonal decoupling, its measurement, and the developmental programming of the HPA/HPG among adolescents in varied contexts. Future research on coupling and decoupling would also benefit from cross‐species comparisons (Brown et al. 2024).

Finally, we did not find consistent patterns concerning hormonal coupling with testosterone assayed in DBS and cortisol assayed in hair (see Supporting Information). However, there was an interesting pattern for both males and females. For males, there was an opposite trend in the age‐specific relationship between log hair cortisol and log testosterone, with decreases across adolescence and far less overlap with the coefficient distributions of the blood cortisol–testosterone coupling relationship (Supporting Information). There was a similar pattern but smaller age‐specific cortisol coefficient sizes among females. These trends should be interpreted cautiously but may indicate less correspondence between hormonal patterns based on assay type. It may also suggest that the free cortisol assayed in dried blood is more representative of a biologically active process in puberty corresponding with testosterone. Future research would benefit from testing similar hypotheses with different sample mediums and incorporating both within‐person variability and longitudinal sampling to continue characterizing the hormonal patterning of cortisol and testosterone in adolescence.

Author Contributions

Delaney J. Glass: conceptualization (lead), data curation (supporting), formal analysis (lead), funding acquisition (lead), investigation (equal), methodology (lead), project administration (lead), resources (supporting), software (equal), writing – original draft (lead), writing – review and editing (lead). Jessica Godwin: data curation (supporting), methodology (equal), resources (supporting), software (lead), visualization (supporting), writing – review and editing (equal). Josefin Koehn: data curation (supporting), investigation (equal), writing – review and editing (supporting). Eleanna Bez: data curation (supporting), writing – review and editing (supporting). Margaret Corley: data curation (equal), resources (supporting), writing – review and editing (supporting). Rana Dajani: supervision (equal), writing – review and editing (supporting). Kristin Hadfield: data curation (lead), investigation (equal), resources (equal), writing – review and editing (supporting). Catherine Panter‐Brick: conceptualization (equal), funding acquisition (lead), investigation (equal), methodology (equal), resources (equal), supervision (equal), writing – review and editing (supporting). Claudia Valeggia: conceptualization (equal), funding acquisition (lead), investigation (equal), methodology (equal), resources (equal), supervision (equal), writing – review and editing (supporting). Melanie Martin: data curation (supporting), funding acquisition (equal), investigation (supporting), methodology (supporting), resources (equal), supervision (lead), writing – review and editing (supporting).

Conflicts of Interest

The authors declare no conflicts of interest.

Supporting information

Data S1: ajpa70134‐sup‐0001‐Supporting_information.docx.

Glass, D. J. , Godwin J., Koehn J., et al. 2025. “Positive Cortisol–Testosterone Hormonal Coupling Among Adolescents in Argentina and Jordan.” American Journal of Biological Anthropology 188, no. 2: e70134. 10.1002/ajpa.70134.

Data Availability Statement

The deidentified data and code files that support the findings of this study are available from the corresponding author upon reasonable request.