Introduction

Psychological development in adolescence is a prime example of the importance of gene-hormone interplay.Footnote 1 Pubertal hormones play a significant role in psychological change in adolescence: Extreme variations in the onset (timing) of pubertal development are linked to risks for depression and externalizing behavior problems (for reviews, see Ge and Natsuaki 2009; Graber 2013; Negriff and Susman 2011); attainment of midpubertal status appears to increase normative changes in risky behavior and to trigger psychopathology in vulnerable individuals (e.g., Gunnar et al. 2009; Trotman et al. 2013). These changes are hypothesized to reflect effects of social responses to youths’ changing bodies, and sex hormones and social experiences acting on the developing brain and stress systems.

The physical changes of puberty reflect an integration of processes influenced by hormones (Styne and Grumbach 2011). Hormone levels (measured in saliva or blood) provide an incomplete picture; they reflect genetic variation and environmental exposures as well as pubertal maturation. The psychological and physiological effects of hormones are not simply related to circulating levels, but also depend on tissue sensitivity and hormone modulators (genetic, physiological, and environmental). Thus, the role of gene-hormone interplay in adolescent psychological development may profitably be studied by examining genetic influences on the physical changes of puberty and their links to behavior at that time.

It is clear that pubertal development itself is influenced by both genes and the environment. The KISS1 gene and its GPR54 receptor appear to play a role in the normal initiation of puberty (Navarro et al. 2007). Variations in the timing of pubertal onset are highly heritable (Eaves et al. 2004; Mustanski et al. 2004; van den Berg et al. 2006), although pubertal timing may be altered by environmental factors, both physical (such as endocrine-disrupting chemicals, Lee and Styne 2013) and social (such as father absence, Belsky et al. 1991; Webster et al. 2014).

Behavior genetic studies have begun to identify how genes and environment transact with pubertal hormones to produce psychological changes in adolescence. Such studies have shown that (a) puberty affects expression of genes involved in some behavior problems, such as disordered eating (Culbert et al. 2009; Klump et al. 2007); (b) early puberty increases risk for behavior problems through a variety of mechanisms, including shared environmental influences on the association between internalizing problems and early puberty in girls (Marceau et al. 2012), genetic and environmental influences on the link between dieting and early puberty in girls (with the source of variation depending on the index of puberty, Harden et al. 2012), and gene-environment interactions accounting for the association between early puberty and delinquency in girls (Harden and Mendle 2012); (c) causes of behavior problems may depend on pubertal development; for example, risk for conduct disorder is mediated by environmental mechanisms for early maturers, but by genetic mechanisms for youth who develop on time (Burt et al. 2006).

But, many behavior genetic studies of puberty and behavior—and, indeed, many developmental studies in general—have not included clear conceptualization or measurement of puberty, or have conflated different aspects of puberty (Dorn and Biro 2010; Dorn et al. 2006; Mendle 2014). Puberty is not a single process but includes adrenarche (maturation of the adrenal glands), gonadarche (development of the hypothalamic–pituitary–gonadal axis), and growth. The three systems do not develop together, and show a different pattern of sex differences: adrenarche occurs earlier than gonadarche, with adrenarche occurring close to the same age in both sexes, but gonadarche occurring earlier in girls than in boys. The hormones involved have differential effects on physical development; for example, body hair is influenced by adrenal hormones in both sexes, testes and voice changes by testosterone in boys, breast development and menarche by estrogens in girls, and height by sex steroids and growth hormone (reviewed in Styne and Grumbach 2011).

These multiple processes are rarely considered in psychological studies. Instead, most studies include a total summary score of pubertal development (typically measured by a self-report questionnaire) or age at menarche in girls (a traditional measure often used to index a girl’s current pubertal development or the timing of her development in relation to others; see distinction below). But summary scores lose information about pubertal features that develop on different timetables (Tanner 1978), are differentially apparent to others (so have differential social signaling), and may have different psychological significance for the youth him/herself. For girls, breast development is typically the first sign of puberty and visible to others, whereas menarche occurs late in puberty and can be concealed; it is unclear which event has the most significance for the girl herself. For boys, testicular enlargement is typically the first sign of puberty and is generally not apparent to others, whereas the height spurt (visible to others) does not occur until midpuberty; again, it is unclear which event is most important. The focus on summary scores has begun to change, with some studies separating indices of adrenarche and gonadarche (e.g., Marceau et al. 2011), and calls to consider synchrony of development (relation among different pubertal features; Mendle 2014). In a related vein, there have been renewed calls to differentiate objective and subjective (youth self-perceived) indicators of pubertal development, and to consider their different associations with behavior (Dorn et al. 2006; Harden et al. 2012; Mendle 2014).

Another issue that has complicated understanding of the psychological significance of puberty is the confounding of two key aspects of development in children who are not yet mature: pubertal timing, which reflects an individual’s development relative to peers, and pubertal status, which reflects the current stage of an individual’s development. Timing has significance during puberty; for example, girls who have started to develop earlier than their peers are more likely to have problems because of involvement with older peers (usually boys) (Caspi and Moffitt 1991; Ge et al. 1996). Timing also has long-term significance for some aspects of psychological function; for example, adult women who were early maturers have higher rates of depression than women who matured on time (Copeland et al. 2010; Graber et al. 2004). Pubertal status is thought to have psychological significance by marking changing hormones and social roles. For example, the sex difference in depression that emerges in adolescence has been interpreted to reflect gender intensification (Hill and Lynch 1983; Wichstrøm, 1999); pubertal onset (whenever it occurs) is hypothesized to trigger diverging trajectories in boys and girls, because the onset of sexual maturity and anticipation of adult roles leads to reactions from social agents and youth themselves that exaggerate gender-stereotypical behavior.

Individuals who have completed puberty all have the same pubertal status (attained adult development), but they differed in timing (relative to peers, some started early, some started the same, and some started late). Individuals who are in the midst of puberty differ in both timing and status; for example, a 10-year-old White girl who has moderate breast development is both midpubertal and an early maturer, whereas a similar age girl who has not started development is prepubertal and may mature on time or late (her timing cannot be known until she starts development). This means that a typical study of youth will necessarily confound pubertal timing and status for at least some of the sample, and fail to obtain useful information on others.

Finally, pubertal variations extend beyond current pubertal status and timing of pubertal onset (both of which may vary by feature) to include tempo of development (the rate at which youth proceed through the stages of puberty). Tempo has not been well-studied, in part because of the difficulty measuring it (see below). There is some suggestion that fast-maturing boys are at greatest risk for adjustment problems (Mendle et al. 2010), but findings are not consistent (e.g., Marceau et al. 2011), and significant questions remain concerning sex differences in tempo and the ways that tempo is linked to timing and to behavior (Beltz et al. 2014; Mendle 2014).

Recent advances in methods for describing puberty have helped to clarify in several ways what is measured, and the different processes captured by different indices. The work has also raised questions that can profitably be studied from a behavior genetic perspective. Thus, longitudinal data on puberty (measured by physical exam or self report) have been used to model trajectories of development using both a simple linear model and a logistic model (discussed in Beltz et al. 2014). A linear model represents constant development, and a logistic model represents S-shaped development that is symmetric at midpuberty. A logistic model fits better than a linear one (Beltz et al. 2014; Marceau et al. 2011), but estimates of timing derived from linear and logistic models correlate highly with each other and similarly with psychological outcome assessed a few years later (Beltz et al. 2014). A behavior genetic approach can address questions about the value of the different methods, by identifying sources of variation in the different measures, and in the covariations among them and with behavior.

Statistical modeling has the benefit of providing a direct estimate of tempo. Such estimates have enabled researchers to ask about the psychological significance of tempo as well as timing (Castellanos-Ryan et al. 2013; Marceau et al. 2011; Mendle et al. 2010). But tempo is conceptualized and estimated differently in linear and logistic models: A linear model represents uniform development across time, with tempo estimated as rate of change per year (in PDS or Tanner stages); a logistic model represents non-uniform development, starting slowly, increasing to maximum growth at the midpoint, and then decreasing in rate, with tempo estimated as instantaneous change at midpuberty, representing the maximum speed of development. Further, both linear and logistic estimates of tempo differ from traditional pediatric measures, which reflect the interval between two stages of puberty (usually time from the onset of breast development to menarche). Not surprisingly, linear, logistic, and traditional estimates of tempo do not correlate highly with each other, and are differently related to corresponding estimates of timing, and to psychological outcome; this is seen when studies using different methods are compared (Castellanos-Ryan et al. 2013; Marceau et al. 2011; Mendle et al. 2010), and when different methods are used with the same sample in a single study (Beltz et al. 2014). A behavior genetic approach can address questions about tempo, by identifying sources of variation in the different estimates, in the covariation among them, and in the links between tempo and timing, and tempo and behavior.

Statistical modeling has also drawn renewed attention to the multiple processes of puberty, particularly the relative importance of adrenal versus gonadal hormones; this work has been facilitated by procedures for using ratings of physical features to measure separately adrenarche and gonadarche (Shirtcliff et al. 2009). But, data using physical features to estimate timing of adrenarche and gonadarche are at odds with endocrine data: physical features produce estimates of adrenarche occurring later than gonadarche (Beltz et al. 2014; Marceau et al. 2011; Paus et al. 2010; Shirtcliff et al. 2009), but adrenal hormones rise earlier than gonadal hormones (reviewed in Styne and Grumbach 2011). This suggests that physical features alone are insufficient to separate adrenarche from gonadarche, perhaps because adrenal hormone levels are insufficient to produce physical changes until gonadarche (Dorn et al. 2006; Wan et al. 2012). A behavior genetic approach can address questions about the link between adrenarche and gonadarche as measured by physical features, by identifying sources of variation in each and in the covariation among them, and with behavior.

Throughout all the work on puberty, there has been recognition of the importance of sex differences. Girls achieve gonadarche earlier than boys, but there is little sex difference in timing of adrenarche. Pubertal timing has different consequences for the two sexes: early-maturing girls but both early- and late-maturing boys show increased depression or adjustment problems compared to on-time peers (Graber 2013; Mendle and Ferrero 2012). There is limited and inconsistent evidence regarding sex differences in pubertal tempo (e.g., Beltz et al. 2014; Marceau et al. 2011). Typical development is also likely influenced by different pubertal processes in the sexes. For example, risk taking is likely to be influenced by hormones that are higher in boys than in girls (e.g., testosterone) (Paus et al. 2010), whereas disordered eating and depression appear to depend on hormones that are higher in girls than boys (e.g., estradiol) (Angold et al. 1998; Klump et al. 2010); this means that midpubertal status should be associated with increased risk taking in boys and increased disordered eating and depression in girls. The lack of a simple measure of pubertal development in boys analogous to menarche in girls means that boys are studied less often than are girls. A behavior genetic approach combined with advanced modeling of pubertal data can address questions about sex differences in influences on pubertal development and links to behavior.

We applied genetic analyses to longitudinal self-report data on pubertal development to answer questions about the nature and significance of pubertal development. We considered (a) how genetic and environmental influences on puberty vary by the way that development is described (logistic vs. linear models versus traditional method) and by aspects of puberty (adrenarche vs. gonadarche), (b) how genes and environment contribute to the covariation between different pubertal indicators, and between pubertal development and behavior problems, (c) how puberty moderates the heritability of behavior problems, and (d) how the answers to all three sets of questions vary by sex. These questions are central to understanding the interplay of genes and sex hormones on adolescent psychological development. Secondary sex characteristics reflect an integration of hormonal processes, whereas direct hormone assays have limitations (see “Discussion” section).

Method

This study is an extension of work on modeling timing and tempo of pubertal development, and their links to behavior (Beltz et al. 2014). We focus here on genetic analyses to further address questions about the measurement and meaning of pubertal development.

Participants

Participants were members of the Colorado Longitudinal Twin Study (LTS), a project examining genetic and environmental contributors to variations in cognition, personality, and behavior problems (Rhea et al. 2006). Of the original sample of 966 individuals from 483 twin pairs recruited around age 1 year, 84 % provided sufficient data to enable calculation of trajectories of pubertal development (two assessments showing change); those who provided sufficient pubertal data were similar to those who did not on the outcome measures of interest. The current sample (total N = 808) included 222 monozygotic (MZ) and 181 dizygotic (DZ) girls, 199 MZ and 204 DZ boys, and 2 boys of unknown zygosity. Most participants were White (92 %) and not Hispanic (91 %).

Participants (born between 1984 and 1990) were assessed on multiple occasions from infancy through young adulthood; we focused on assessments of puberty throughout adolescence and behavioral outcomes in mid-to-late adolescence (conducted between 1994 and 2008). Puberty was assessed annually from the end of grade 3 [average age (SD in parentheses): 9.44 (.37) years, range 8.25–10.67) to the end of grade 9 (average age: 15.34 (.31) years, range 14.25–16.17], with an in-person visit after grade 6, and telephone interviews at other ages. We considered two types of behavioral outcomes: (a) behavior measured during puberty to examine whether pubertal status moderated heritability of behaviors that change in adolescence; this included measures of internalizing and externalizing behavior problems; (b) behaviors measured later in adolescence to examine longer-term effects of pubertal timing; this included measures of substance use and age at sexual initiation assessed between ages 16 and 18. Other psychological and health data not reported here were also collected during interviews. Parents provided informed consent for their child to participate in the study, and youth provided assent (under age 18) or informed consent (age 18 and older) for their participation in the study.

Measures

Pubertal development

Puberty was assessed by self report on the Pubertal Development Scale (PDS; Petersen et al. 1988). Each year, participants answered five questions about the development of secondary sexual characteristics: body hair, skin changes, and growth spurt in both sexes, facial hair and deepening voice in boys, and breast development and menarche in girls. All items except menarche were rated on a 4-point scale: 1 = “no development,” 2 = “yes, barely,” 3 = “yes, definitely,” 4 = “development completed.” Menarche was rated as absent (1) or completed (4); age at menarche was recorded for those who had reached it. Items were averaged to produce a summary PDS score at each age. PDS scores correlate about .70 with pubertal stage rated by health professionals (Schmitz et al. 2004; Shirtcliff et al. 2009), and from .20 to .60 with salivary hormone levels, similar to correlations between those hormones and physical exam ratings (Shirtcliff et al. 2009). The PDS is widely used and considered “most appropriate for broad estimates of development, or for use in longitudinal studies” (Coleman and Coleman, 2002), but still subject to debate (e.g., Dorn et al. 2006; Shirtcliff et al. 2009), a topic to which we return in the Discussion.

Linear and logistic estimates of overall pubertal timing and tempo in both sexes

Pubertal timing and tempo were estimated from group trajectories of development, calculated separately by sex from the longitudinal PDS data (average PDS score at each of seven waves of assessment), allowing individual deviations (for details, see Beltz et al. 2014). Development was represented and estimated separately by linear and logistic models. Pubertal timing was defined at the midpoint of puberty (PDS score of 2.5, corresponding to Tanner stage 3) for both models. Neither model permits independent estimation of pubertal onset (PDS 1.5) and midpuberty (PDS 2.5). Pubertal tempo was estimated as rate of change per year for the linear model, and as peak rate of change at midpuberty for the logistic model.

Trajectories were calculated previously, with participants from this LTS sample and from another sample, the Colorado Adoption Project (CAP) (for details, see Beltz et al. 2014). All analyses were performed with two separate replicates, with one member of a family in each; results were consistent across replicates. Results for LTS participants alone paralleled those reported for the full sample. Genetic analyses reported below are based only on the LTS sample.

Logistic estimates of adrenarche and gonadarche timing and tempo in girls

In addition to trajectories for “total” pubertal development (average PDS score at each age), separate trajectories were calculated for the processes of adrenarche and gonadarche, providing estimates of timing and tempo for both aspects of puberty. These analyses were restricted to logistic models, because a logistic model has been seen to fit both processes better than a linear model (Marceau et al. 2011), and to girls who were more likely than boys to have completed development over the course of assessment, and thus to show meaningful variability in both processes. Pubertal development is typically described in terms of Tanner (1978) stages, so the procedures for separating gonadarche and adrenarche also involved converting PDS scores into Tanner scores. Longitudinal PDS scores were converted into Tanner scores for adrenarche and gonadarche using an algorithm developed by others in another sample from PDS scores, circulating hormone levels, and Tanner staging from physical exams (Shirtcliff et al. 2009); adrenarche was estimated from PDS ratings of skin changes and body hair, and gonadarche from PDS ratings of menarche, breast development, and growth spurt.

Procedures for modeling trajectories of adrenarche and gonadarche development using Tanner scores paralleled those outlined for modeling development using PDS scores (Beltz et al. 2014). Briefly, a logistic growth curve model is represented as:

$${\text{Tanner stage = }}{{{\text{ ((}}\beta _{0} + (\beta _{1} - \beta _{0} ))} \mathord{\left/ {\vphantom {{{\text{ ((}}\beta _{0} + (\beta _{1} - \beta _{0} ))} {({\text{1 + e}}^{{ - \alpha _{{\text{i}}} (Age_{{it}} - \lambda _{i} )}} )){\text{ + r}}_{{{\text{it}}}} ,}}} \right. \kern-\nulldelimiterspace} {({\text{1 + e}}^{{ - \alpha _{{\text{i}}} (Age_{{it}} - \lambda _{i} )}} )){\text{ + r}}_{{{\text{it}}}} ,}}$$

where β0 is 1 (lower bound for Tanner scores); β1 is 5 (upper bound for Tanner scores); e is the exponential function; λ i is the age at midpuberty (Tanner 3); α i is the slope of the function at the midpubertal age; r it is the normally-distributed residual for an individual i at assessment t. Longitudinal Tanner scores were entered into logistic models to calculate group trajectories of development and individual deviations from the group, that is, person-specific estimates of pubertal timing (λ i ) and tempo (α i ).

Models were compared using the Akaike Information Criterion (AIC). The models for gonadarche (AICs of 5178 and 5070 for replicates 1 and 2, respectively) fit better than those for adrenarche (AICs of 5278 and 5141, respectively). Group mean trajectories are displayed in Fig. 1. Note that (a) results for replicates 1 (black lines) and 2 (gray lines) were nearly identical, (b) mean age at Tanner 3 (midpoint of the curve) was greater for adrenarche (dashed lines) than gonadarche (solid lines), and (c) mean tempo (slope of the curve at Tanner 3) was greater for gonadarche than adrenarche.

Fig. 1
figure 1

Mean development trajectories for girls for different aspects of puberty. Dashed lines are adrenarche trajectories. Solid lines are gonadarche trajectories. Black lines are results for the first half of the sample (replicate 1). Gray lines are results from the independent second half of the sample (replicate 2)

Full size image

Linear and logistic estimates of pubertal status in both sexes

Pubertal status was estimated at each age as prepubertal or pubertal based on the logistic pubertal timing parameter (given previous results cited above that a logistic model describes development better than a linear model). For example, a child whose pubertal timing (age at midpuberty) was estimated to be 13.2 years would be prepubertal at all assessments before age 13.2, and pubertal at all subsequent assessments.

Traditional measures of pubertal development in girls

Pubertal timing was measured by age at menarche (assessed close in time to the event). Pubertal tempo was measured by the difference between age at pubertal onset (PDS 1.5 estimated from linear models) and menarche. Note that tempo is conceptualized and represented differently by the models versus the traditional approach: model parameters reflect rate of change, whereas the traditional measure reflects time between pubertal events, making the measures inversely related. Pubertal status was measured by whether the girl had achieved menarche.

Behavior problems

The Child Behavior Checklist (CBCL; Achenbach 1991) was used to obtain parent reports of child behavior problems at all ages. We focused on unstandardized scores for higher-order scales of internalizing and externalizing problems; these are reliable and have been shown to relate to a variety of clinical conditions. We considered whether pubertal status moderated the heritability of these behavior problems, indexing pubertal status by menarche in girls and by estimates from logistic trajectory analyses in both sexes. We used CBCL assessments at age 13 in girls and age 15 in boys because these ages show the most variability in pubertal status and are close to the average ages of midpuberty in this sample (with boys later than girls), and thus are the best ages for examining the question.

Substance use

The Composite International Diagnostic Interview-Substance Abuse Module (CIDI-SAM; Cottler and Keating 1990) was used at ages 16–18 to assess participants’ involvement with substances, including alcohol, cannabis, amphetamines, opiates, cocaine, sedatives, inhalants, PCP, and hallucinogens. Participants retrospectively recalled whether they had any of seven dependence symptoms for each substance. Psychometric properties are good, with discriminative and convergent validity (Crowley et al. 2001). We used the average lifetime number of symptoms experienced through adolescence across all substances, corrected for age at assessment (Button et al. 2010; Stallings et al. 2003). This measure was related to pubertal timing in the combined LTS/CAP sample (Beltz et al. 2014).

Age at sexual initiation

At age 17, all participants provided information on the age of their first sexual experience if it had occurred (Bricker et al. 2006). Participants were first asked “Have you ever had sex (‘gone all the way’) with someone?” If they indicated they had, they were asked “How old were you the first time you had sex?” Repeated assessment of a subsample of participants in CAP showed very high test–retest reliability for reported age at first sexual experience (Bricker et al. 2006). Some participants had not yet had sex by this assessment, and the primary value of this measure is to capture early sexual activity, so we used a categorical measure to maximize the number of participants with data. The categorical score was based on actual age for participants who reported this information (<age 15, age 15 or 16) or the last age of assessment for those who did not (≥age 17). This measure was related to pubertal timing in the combined LTS/CAP sample (Beltz et al. 2014).

Data analyses

We applied traditional behavior genetic analyses to address our questions (footnote 1). First, we conducted univariate genetic analyses to examine genetic (additive, A, and nonadditive, e.g., dominance, D) and environmental (common, C, and nonshared, E) sources of variation in pubertal timing and tempo, using estimates from linear and logistic modeling of the average PDS score at each age for both sexes, the traditional methods in girls, and logistic estimates of adrenarche and gonadarche in girls. Second, we conducted bivariate genetic analyses to examine sources of covariation between different estimates of the same aspect of puberty (e.g., linear and logistic timing) and between timing and tempo estimated from the same method. Third, we conducted bivariate genetic analyses to examine sources of covariation between pubertal timing and behavior; we focused on the logistic measure of timing given that the logistic model best fit the data, and two behaviors that showed the strongest correlations with pubertal timing in the sample, substance use and age at sexual initiation (Beltz et al. 2014). We did not conduct genetic analyses of links between pubertal tempo and behavior because of difficulties in interpreting logistic tempo (see below and Beltz et al. 2014) and the high correlations between linear timing and tempo (Beltz et al. 2014). Finally, we examined how puberty moderated the heritability of Internalizing and Externalizing Behavior Problems at age 13 for girls and age 15 for boys, defining pubertal status by estimates from logistic trajectory analyses in both sexes and by menarche in girls.

Genetic models

Univariate and bivariate models were tested in classic Mx (version 1.63 for Linux, Neale et al. 2002) with four groups: MZ and DZ male and female pairs. Mean sex differences were estimated, and tests for equality of means and variances across twins and groups were conducted prior to genetic modeling. We considered submodels to determine which parameters could be dropped. Alternate acceptable models were compared using the AIC to arrive at a final model, from which parameter estimates and confidence intervals were derived.

We examined sex differences by determining whether the acceptable and most parsimonious models were the same for the two sexes and whether the standardized model parameters (e.g., genetic and environmental correlations) had overlapping confidence intervals. We also conducted a more stringent test for sex differences: equating the A, C/D, and E covariance matrices across sex. If differences are trivial, moderation by sex can be ignored. Note that testing sex differences is complex because of sex differences in pubertal development.

Missing data

Calculation of trajectories of pubertal development required that a participant have at least two different PDS scores; as noted above, 84 % of the original sample met this criterion. Approximately 20 % of girls did not have data on traditional measures, primarily because they had not yet reached menarche. (For additional information, see Beltz et al. 2014) The maximum-likelihood (ML) approach used in Mx with raw data provides unbiased estimates of parameters when the usual assumptions of ML are met and data are either missing at random conditional on an observed variable (e.g., cotwin scores or other nonmissing variables), or completely at random (Little and Rubin, 1987). This approach is thus superior to estimation procedures which require complete twin pairs and complete data for individuals.

Results

Pubertal development: phenotypic correlations

For completeness, we present (Table 1) the phenotypic correlations among the measures of pubertal development. They confirm that the pattern of results seen in the combined LTS and CAP sample (Beltz et al. 2014) was apparent in the LTS sample alone, and help to frame the genetic analyses. Measures of pubertal timing were highly correlated with each other, measures of pubertal tempo were generally not strongly associated with each other, and within-method links between timing and tempo varied across methods.

Table 1 Correlations among measures of pubertal development
Full size table

Pubertal timing and tempo: univariate genetic analyses

We conducted univariate genetic analyses of the linear and logistic estimates of timing and tempo for both sexes. For girls, we also examined the traditional measures of timing and tempo, and separate logistic estimates of timing and tempo for adrenarche and gonadarche. The results of univariate analyses for pubertal timing and tempo are shown in Tables 2 and 3, respectively; each table includes MZ and DZ twin correlations, a baseline model that describes the pattern of twin correlation differences, a reduced model with the statistics that support model simplification, and the standardized estimates for sources of variation under this best-fitting model. The model-fit statistics for alternative reduced models are shown in supplementary Table S.1. Results are shown separately for girls and boys, and for a direct sex comparison.

Table 2 Univariate analyses for pubertal timing
Full size table
Table 3 Univariate analyses for pubertal tempo
Full size table

Results were consistent across the different measures of pubertal timing (as shown in Table 2), but not across the measures of pubertal tempo (as shown in Table 3), as indicated by the best-fitting (most parsimonious) models. Variations in pubertal timing (Table 2) were largely attributable to genes in both sexes, no matter how timing was measured, as shown by the standardized estimates for additive genetic variance (A). This was the case for both linear and logistic estimates based on the total PDS score, and for the logistic estimates for adrenarche and gonadarche, although there was more nonshared environmental influence (E) on adrenarche than on gonadarche, as shown (in the last two columns in panel 2A) by the non-overlapping 95 % confidence intervals (CIs) for E. Note that the standardized variance estimates in the reduced models sum to 100 % (i.e., in AE models, variance not attributed to genes is attributed to nonshared environment), while the 95 % CIs indicate how precisely each source of variance is estimated within this sample.

Sex comparisons for univariate genetic analyses of pubertal timing and tempo

We compared models for boys and girls, although the PDS trajectories are not derived from exactly the same indicators, and there are no opposite-sex twin pairs to test sex-limitation models. Sex differences were tested by determining whether: (a) a single, common reduced model adequately fits the data for both sexes (although the best-fitting model for each sex separately might differ from each other and from the common model, as shown in panel A for girls and in panel B for boys); (b) the standardized estimates for parameters have overlapping CIs; and (c) the unstandardized variance components can be equated.

Sex comparisons for univariate models of pubertal timing (estimated for PDS total scores using linear and logistic trajectories) are shown in panel 2C. For both linear and logistic estimates, a common model is acceptable for both boys and girls, and the standardized variance estimates have overlapping CIs. But, generally the raw variance components cannot be equated. In sum, results shown in Table 2 indicate that pubertal timing in both sexes, regardless of how it is measured, is consistently and primarily attributable to additive genetic effects.

Variations in pubertal tempo (Table 3), however, were attributable to different sources in boys and girls, and across methods of measurement. There were sex differences in the preferred reduced models. Different models were also preferred for the different estimates of tempo; for example, among girls, genetic influences predominated for the linear estimate of tempo, but common environmental influences (C) predominated for the logistic estimate of tempo. The logistic estimate of tempo in girls is notable for showing no genetic influence, although there were genetic influences on logistic estimates of tempo measures for adrenarche and gonadarche; the discrepancy appears to reflect differing DZ correlations for logistic estimates based on total PDS scores versus subscores of adrenarche and gonadarche. Among both girls (panel 3A) and boys (panel 3B), nonshared environment, which includes measurement error, appears to play a larger role in the logistic estimate of tempo than in the linear estimate; this is seen in non-overlapping CIs. With respect to direct sex comparisons (panel 3C), the best-fitting model common to pubertal tempo in both girls and boys differs for the linear and logistic parameters. There are not simple correspondences between the best-fitting common models and those for each sex separately. This is likely due to sources of variation in the logistic estimates of tempo in girls, i.e., large shared environmental influence on the total PDS score (due to high DZ correlations), but large genetic influences on other measures. For the linear tempo measures, CIs overlap, although complete equality is not achievable. In contrast, for the logistic tempo measures, the CIs for parameters in girls and boys do not overlap, with a much higher estimate of the effect of the shared environment in girls. In sum, results shown in Table 3 indicate that, in contrast with timing, no single parsimonious model can adequately describe influences on tempo across both sexes and alternative measures.

Covariations among pubertal indicators: bivariate genetic analyses

We next conducted bivariate genetic analyses to understand phenotypic correlations between descriptors of pubertal development. This included examining sources of phenotypic correlations across methods in estimates of timing (Table 4) and of tempo (Table S.3), and within method in estimates of timing and tempo (Table S.4) in both sexes; and of the correlations between adrenarche and gonadarche in girls. For all bivariate analyses, we show a baseline model, a best-fitting reduced model, the phenotypic correlation between measures from the baseline model, estimates for genetic (rA or rD), shared environmental (rC), and nonshared environmental (rE) correlations between measures based on the reduced model, and the proportion of the covariance between measures that can be attributed to genetic or shared environmental sources. Model fit statistics for alternative bivariate reduced models are shown in supplementary Table S.2 (Table S.6 for reduced models involving tempo in Tables S.3 and S.4).

Table 4 Bivariate analyses among measures of pubertal timing
Full size table

Pubertal timing across methods: bivariate genetic analyses

In accord with results of univariate analyses, results of bivariate analyses were more consistent for pubertal timing than for pubertal tempo. For pubertal timing (Table 4), phenotypic correlations among the different indicators were largely due to genes, as shown by the consistent AE reduced models, the high estimates for the genetic correlations between measures, and the proportion of the phenotypic covariance that is attributable to additive genetic influences acting on both measures. In addition, similar estimates of additive genetic correlations for linear and logistic timing measures are found in both sexes. The correlation between timing of adrenarche and gonadarche in girls was also attributable entirely to genetic covariation, although there was some unique genetic variance in each measure: rA  = 0.64, and 95 % CIs did not include 1.00. The nonshared environmental correlation between adrenarche and gonadarche timing could be set to zero, reflecting the different PDS features comprising the two measures. Moreover, pubertal timing in girls appears to reflect gonadarche more than adrenarche for both genetic and nonshared environmental reasons: timing of total PDS score has lower rE and rA with adrenarche than with gonadarche.

Sex comparisons for bivariate genetic analyses of links among measures of pubertal timing

The overlap between the estimates of (total PDS) timing from linear and logistic models reflects effects of additive genes and nonshared environment for both sexes to a similar extent: a common AE model fits with high rA and rE. But, the CIs do not overlap for the standardized estimates, reflecting the higher phenotypic correlation in girls than in boys and, as might be expected from the inability to equate variation in the univariate models (panel 2C), it is not possible to equate the unstandardized covariance matrices. In sum, results shown in Table 4 again indicate the high degree of overlap of genetic and nonshared environmental influences across sexes and measures of pubertal timing.

Pubertal tempo across methods: bivariate genetic analyses

Results of bivariate analyses of pubertal tempo (Table S.3) are similar to those for univariate analyses in showing few clear patterns. For girls, all logistic estimates of overall tempo (PDS total score measure) required a nongenetic (CE) model, but a genetic source of variation (AE or ACE model) for the other tempo measure (panel S.3A); thus, overlap between measures can only be attributed to environmental (C or E) covariance. For example, 55 % of the covariation between logistic PDS and gonadarche tempo estimates was due to common environmental influences, but covariation between logistic PDS and traditional measure estimates was only due to nonshared environmental influences. There were genetic influences, however, on covariation between adrenarche and gonadarche tempo estimates in girls and between logistic and linear PDS tempo estimates in boys. The covariation in boys was also partly attributable to the nonshared environment (panel S.3B). It is apparent that no single reduced model explains the shared covariation among tempo measures across sex.

Sex comparisons for bivariate genetic analyses of links among measures of pubertal tempo

Direct comparison of boys and girls (panel S.3C) shows that the simple common models (AE, CE, DE) do not adequately describe the covariation between linear and logistic tempo parameters in either sex, but a complex model is acceptable: AE for the linear estimate of tempo, and CE for the logistic estimate of tempo. Thus, nonshared environment is the only common source of covariation between linear and logistic estimates of tempo, since familial influences are attributed to different sources for the different estimates (A for linear and C for logistic). The nonadditive genetic contribution to the covariation in tempo in boys (panel 5B) is not present in the common model across sex. In sum, results shown in Table S.3 again indicate the lack of consistency in influences on pubertal tempo across sex and alternative measures.

Pubertal timing–tempo links: bivariate genetic analyses

Bivariate analyses of links between timing and tempo within method (Table S.4) showed that genes were a major source of covariation for most indicators, as seen in the AE best-fitting reduced models. Exceptions concerned the logistic estimates: For girls, the positive covariation between timing and tempo was attributed primarily (77 %) to shared environmental influences; for boys, the modest negative correlation (−.26) between timing and tempo was partially attributed to nonadditive genes. Although phenotypic correlations between timing and tempo differed between related measures (e.g., in girls, for adrenarche, r = −.39, for gonadarche, r = .50), genetic and environmental influences acted similarly within measure (as indicated by the identical signs of rA and rE).

Sex comparisons for bivariate genetic analyses of pubertal timing–tempo links

Comparison of boys and girls (panel S.4C) showed a common AE model for the linear timing–tempo link (although covariances cannot be equated; see supplementary Table S.6), but no common model for the logistic timing–tempo link. The latter finding is consistent with the sex difference in the direction of the phenotypic correlations between logistic timing and tempo (positive in girls, negative in boys). Other aspects of the modeling confirm the distinctness of the logistic estimate of tempo in girls (e.g., contrasting signs for parameter estimates, lack of overlap of CIs). In sum, results shown in Table S.4 indicate that bivariate relationships involving tempo cannot be described simply; sources of covariation differ across sex and measures.

Puberty-behavior associations: bivariate genetic analyses

Phenotypic correlations between puberty and behavioral outcomes in this sample combined with CAP reported previously (Beltz et al. 2014) are consistent with others in showing small associations between early pubertal timing and behavior problems, especially early age at sexual initiation and substance use (as noted above, analyses were restricted to the logistic estimate of timing, because the logistic model fit the data better than a linear model). Correlations were similar in the LTS sample alone: Pubertal timing (logistic estimate of the trajectory for the total PDS score) correlated with age at sexual initiation (using the 3-category measure) .14 in girls and .15 in boys, and with substance use .18 in girls and .08 in boys; all correlations except the last were significantly greater than 0.

The results of bivariate genetic analyses examining sources of the phenotypic correlations between pubertal timing and behavior are shown in Table 5. Pubertal timing was linked for additive genetic reasons with age at sexual initiation and substance use in girls, and for nonadditive (dominance) genetic reasons with age at sexual initiation in boys; there was no evidence of shared or nonshared environmental influences (see results of alternative models in Table S.2). Comparisons of boys and girls (panel 5C) show that reduced AE models with rE set to zero are acceptable for both sexes for both behavioral measures. Results shown in Table 5 between pubertal timing and behavioral outcomes assessed several years later contrast with those across and within pubertal indicators (shown in Tables 4, S.3, and S.4) by their lack of a significant nonshared environmental contribution to the phenotypic correlations.

Table 5 Bivariate analyses of logistic measure of pubertal timing and behavior
Full size table

Moderated heritability

Finally, we considered the extent to which pubertal processes might change the expression of genes important for behavior, by examining whether pubertal status predicted the magnitude of genetic and environmental influences on outcomes. Pubertal status was defined as the age at midpuberty estimated from the logistic model in both sexes, and as the age at menarche in girls. The behaviors likely to be relevant here are those that increase in incidence or level during puberty. LTS was not designed to address this question, so the measures available to explore the question were limited to parent-reported CBCL Internalizing and Externalizing Behavior Problems collected at yearly intervals during adolescence.

We used a twin moderation model (Klump et al. 2007; Purcell 2002) to test whether the estimates for genetic and environmental influences on the CBCL measures varied linearly with age of midpuberty in both sexes or with age at menarche in girls (our sample size prevented testing for nonlinear effects). We examined CBCL scores collected at the age closest to the average midpoint of pubertal development, which was age 13 for girls and age 15 for boys, and used log-transformed standardized scores to reduce skewness.

There was no compelling evidence that genetic influences on CBCL Internalizing or Externalizing Behavior Problems varied with pubertal status. As shown in Table S.5, there was only one outcome for which puberty (logistic timing) moderated estimates of genetic and environmental influences, and this effect is opposite to expectation. For Internalizing Problems in girls, early pubertal status (or delayed development relative to peers) was associated with increased genetic influences and decreased nonshared environmental influences (which includes measurement error). No significant moderation was detected for internalizing problems in boys, or externalizing problems in either sex. Furthermore, mean transformed CBCL scores did not differ significantly from 0, and the means did not vary with pubertal status. Analyses using age at menarche (not shown) also failed to show evidence of pubertal moderation of genetic and environmental influences on behavior problems.

Discussion

We combined behavior genetic methods with advanced modeling of puberty to clarify the measurement and psychological significance of pubertal development. Specifically, we considered: (a) genetic and environmental influences on pubertal development described in different ways; (b) genetic and environmental influences on the covariation between different pubertal indicators, and between pubertal development and behavior problems; (c) changing heritability of behavior problems with pubertal development; and (d) sex differences in each of those three topics.

Variations in pubertal development

Variations in pubertal timing were found to be largely genetic in both boys and girls. There was more nonshared environmental influence on pubertal timing in boys than in girls, and on timing of adrenarche than on gonadarche, likely due to measurement, that is, better measurement of gonadarche than adrenarche, and better measurement of gonadarche in girls than in boys (Dorn et al. 2006; Dorn and Biro 2010). Results were similar across methods for describing timing (logistic estimate, linear estimate, menarche in girls). Furthermore, bivariate analyses revealed large genetic overlap among the measures of pubertal timing. These findings confirm results concerning large genetic and some nonshared environmental contributions to puberty (Eaves et al. 2004; Mustanski et al. 2004; van den Berg et al. 2006) and extend them to other methods; they also extend other findings from this sample that different measures of timing correlate highly with each other and in similar ways with behavioral outcome (Beltz et al. 2014).

Variations in pubertal tempo, however, were not consistently attributed to genetic or environmental sources; estimates varied across methods. This is the first genetic analysis of variations in pubertal tempo. But, the results are not surprising in light of other reports about tempo, particularly inconsistent links with behavior, partly due to variations in definition and measurement across studies (Castellanos-Ryan et al. 2013; Marceau et al. 2011; Mendle et al. 2010).

How best to measure pubertal development?

Our results with self reports of pubertal development provide little evidence to prefer one index of pubertal timing over another, and confirm the difficulties of measuring tempo. All indexes of overall pubertal timing are highly heritable and reflect highly similar genetic processes, as shown in the large genetic contribution to the (high) phenotypic correlations among logistic and linear timing, and menarche. The different indexes of pubertal tempo, however, are differently influenced by genes and do not appear to share genetic variation; this is consistent with other information about the measures, that is, at the phenotypic level, tempo measures are not highly correlated with each other, and not consistently correlated with timing or with behavior (Beltz et al. 2014). It is notable that the logistic estimate of tempo is probably the least meaningful from both genetic and endocrine perspectives. It is conceptualized in an unusual way, as the instantaneous speed of development at the midpoint of puberty. This likely explains why it shows little genetic influence or genetic overlap with logistic timing, and why it is correlated with timing in a way that is inconsistent with other methods.

Our results also suggest little value in separately examining timing of adrenarche, at least when it is measured by self-reported physical features. Using the logistic model in girls, timing of adrenarche showed large (but not complete) genetic overlap with timing of gonadarche. The larger nonshared environmental contribution to adrenarche than to gonadarche likely reflects measurement error associated with the relatively greater difficulty in rating body hair than genital development. The difficulty of estimating adrenarche separately from gonadarche extends beyond self reports: In several studies measuring puberty with physical features (self report or exams by health professionals), adrenarche was seen to occur later than gonadarche (Beltz et al. 2014; Marceau et al. 2011; Paus et al. 2010; Shirtcliff et al. 2009), but adrenal hormones rise earlier than gonadal hormones (reviewed in Styne and Grumbach, 2011); adrenal hormone levels may be insufficient to produce physical changes until gonadarche (Dorn et al. 2006; Wan et al. 2012).

Puberty-behavior associations

We focused on the link of pubertal timing to substance use and age at sexual initiation, given that these are the strongest phenotypic correlations found in both sexes in our study and are consistently found in other studies (e.g., Ge and Natsuaki, 2009; Graber 2013). Results suggest that covariations reflect additive genetic factors in girls, but are not as clear in boys. There is relatively little other behavior genetic work examining sources of covariation between pubertal development and these behaviors, especially in boys. Findings regarding other characteristics have not produced a consistent picture on the ways that genes and environment mediate associations between pubertal timing and behavior, and there is a need for more work on the topic, with particular attention to the conceptualization and measurement of puberty (see also Harden et al. 2012). We did not examine sources of phenotypic correlations between tempo and behavioral outcome given the genetic results for tempo, and that links between tempo and behavior are largely overlapping with timing (Beltz et al. 2014).

Do genetic influences on behavior change at puberty?

In light of suggestions that pubertal hormones regulate the expression of genes underlying behavior, and thus account for psychological changes at puberty (e.g., Klump et al. 2003, 2007), we studied whether puberty moderates genetic influences on some behavior problems. We aimed to examine whether sex hormones contribute to the increased incidence of behavior problems by modulating gene expression. We were limited by the LTS design to using CBCL measures of internalizing and externalizing at age 13 for girls, and age 15 for boys, ages which show expected variability in pubertal status and represent midpuberty in this sample. We examined whether pubertal status affected estimates of genetic and environmental influences on outcomes, indexing pubertal status with age at midpuberty estimated from the logistic model in both sexes and with age at menarche in girls.

We found no evidence that advancing puberty was associated with increased genetic influences for either internalizing or externalizing problems in girls or boys, but our study was not designed to provide a strong test of such effects. The lack of significant moderation should be considered in light of the relatively small sample size and the limited set of behavioral outcomes available; the lack of significant change in mean CBCL scores with advancing puberty suggests that these measures are probably not strong candidates for moderation effects.

The optimal approach to studying moderated heritability should involve measures chosen to test hypothesized mechanisms accounting for behavioral changes at puberty, e.g., the importance of estradiol for triggering disordered eating and depression in girls, and testosterone in facilitating risk taking (and associated externalizing problems) in boys. Unfortunately, LTS was conducted before such focus on adolescent change so we were limited to examining this issue with CBCL scores. Further limiting our ability to test our hypotheses, these behavior problems did not increase at midpuberty in this sample. Nevertheless, we presented these results to illustrate how trajectory analyses might be used to study gene-hormone interplay.

Study strengths and limitations

This study is an example of gene-hormone interplay on behavior, and how combining a behavior genetic perspective and advanced modeling of pubertal development can help to understand behavioral effects of hormones, using self-reported physical development as a proxy for hormonal changes. We extended work (ours and others) regarding methods for estimating pubertal timing and tempo, and their links to behavior.

There are several new findings that should guide studies of pubertal development. First, there is little evidence for the superiority of one estimate of pubertal timing over others. Although a logistic model may describe development better than a linear model (Beltz et al. 2014; Marceau et al. 2011), it is not clear that there are advantages of logistic estimates. The different estimates of pubertal timing—linear, logistic, and traditional—are correlated with each other and relate to behavior in similar ways—for genetic reasons. Second, pubertal tempo is difficult to capture and is estimated idiosyncratically across method (Beltz et al. 2014; Marceau et al. 2011; Mendle et al. 2010)—with genetic analyses confirming that different constructs are measured by each method. Thus, calls for studies of the psychological significance of pubertal tempo (Mendle 2014) may be premature given the need for more data on the reliability and validity of different measures. Third, there appears to be little value to studying adrenarche alone, at least with the PDS and perhaps with physical features generally. Fourth, consistency of findings across sex strengthens our conclusions, and encourages the use of trajectories to study the psychological significance of puberty in boys as well as girls. Fifth, our results highlight the importance of separating pubertal status and timing, and methods for doing so.

Several factors should be considered in interpreting the results. First, we used physical indicators of puberty as a proxy for hormones, but this is not necessarily a limitation. Physical development reflects an integrated picture of the effect of adrenal and gonadal hormones. Direct assays of hormones, particularly at a single point in time, may not provide more accurate assessment of pubertal processes than do measures of physical features (even by self report; Shirtcliff et al. 2009). Hormone levels reflect more than pubertal development, including variations due to genes (Harris et al. 1998), circadian, monthly, and seasonal rhythms, environmental factors (e.g., diet, exercise), and behavior itself (Carré 2009; Stanton et al. 2011); responses to hormones also depend on other hormones that are present, and sensitivity of hormone receptors (Styne and Grumbach 2011). Furthermore, hormone assays are not straightforward (Handelsman and Wartofsky 2013; van Anders 2010).

Second, there is concern about the value of self report, including imperfect correspondence with physical exam by a health professional (e.g., Huang et al. 2012). But, self report has been shown to correlate with hormone levels as well as physical exam does (Shirtcliff et al. 2009), and will continue to be the preferred method of many investigators because it is easy and inexpensive to use, and is nonintrusive, so will enable data collection from representative samples. And, as we showed, self reports of pubertal development can readily be used to model trajectories of development in boys and girls and produce meaningful results. Trajectories also overcome one of the limitations of repeated self-report assessments: some youth report lower development from one time to the next.

Third, although sample size was sufficient for most analyses, it was not large enough for robust analyses of moderated heritability. We retained these analyses, however, to demonstrate the value of the trajectories for studying the role of pubertal status as well as timing and tempo.

Fourth, formal testing of sex-limitation models was not possible because PDS total scores were not identical in girls and boys and there were no opposite-sex DZ twins in LTS. But several sex comparisons were done, allowing us to determine whether sex-moderation could be ignored. Future work should examine different ways in which genetic models can incorporate the sex difference in pubertal timing.

Finally, there are some limitations to our behavioral measures, although all are widely-used and psychometrically sound. Particular concerns relate to insensitivity to problem behaviors in this sample, potentially reducing the size of links between pubertal timing and later behavior, and to limited measurement of problems likely to increase in adolescence, constraining our ability to examine pubertal status as a moderator of heritability of problems.

Conclusions and future directions

Applying a behavior genetic perspective to advanced models, we showed the value and limitations of different approaches to measuring pubertal development, and provided evidence to guide future studies linking hormones to behavior. Our data confirm and extend others in showing the value of youth self report of pubertal development, the importance of genetic influences on pubertal timing no matter how it is measured, and the limitations of current measures of pubertal tempo. Use of trajectories should also facilitate research on the psychological significance of puberty in boys as well as girls.

We took a systems-level approach to understanding gene-hormone interplay at adolescence: the physical changes of puberty reflect the integration of the developing adrenal glands and hypothalamic–pituitary–gonadal axis. Thus, our study provides information on the interplay between genes and the integrated effects of hormones. It also offers a road map to other researchers in terms of methodology (how to maximize the value of data on physical indicators) and conceptual framework (how to investigate understudied aspects of gene-hormone interplay in adolescence, and recognize the value of studying physical development).

Future work could profitably focus on a number of interesting questions regarding gene-puberty interplay in psychological development. One set of questions, amenable to study in several existing behavior genetic projects, concerns genetic and environmental contributors to links between variations in pubertal development and behavior problems. The phenotypic association between early puberty and behavioral risk is well-established in girls, but the sources of the covariation have received surprisingly little attention, and both the nature and causes of phenotypic associations in boys have not been well studied, in part because of concerns about measurement of puberty (especially by self report) in boys. Our results suggest that these questions can be addressed with measures that are already available, or easily computed, in many typical developmental behavior genetic studies. Another set of questions requires improvements in conceptualization and measurement of puberty. This includes refinement of measures of tempo, differentiation of aspects and processes of puberty (e.g., importance of development of pubic hair versus genitalia versus menarche in girls) and the synchrony of their development, and ability to measure the different pubertal stages (e.g., onset versus midpuberty). Puberty is an important developmental period that exemplifies gene-hormone interplay and thus represents an important research opportunity for behavior geneticists.