Introduction

Adolescence is a critical period of brain development. Structural and functional interconnectivity of brain regions form a reliably and accurately functioning network to efficiently process information and behavioural output. The development of the brain during adolescence has predominantly been studied in relation to age (Andersen 2003; Blakemore 2012; Giedd and Rapoport 2010; Lenroot and Giedd 2006; Mills and Tamnes 2014). During normative gray matter development, a reduction in general cortical volume with increasing age has been consistently reported (Ducharme et al. 2016; Mills et al. 2016; Vijayakumar et al. 2016; Wierenga et al. 2014). However, more recently, studies have highlighted that the influence of pubertal maturation—not only age—on brain growth is crucial in disentangling the underlying physiological mechanisms (Kaczkurkin et al. 2018).

When considering the effect of age and pubertal status on brain development, it is important to note that the age in which biological changes occur in association with puberty varies across individuals. Therefore, age and pubertal status need to be considered as partially distinctive measures. For example, variation in puberty onset can differ by up to 4–5 years (Parent et al. 2003). More recent studies have hence examined the effects of pubertal status on brain development, which has also been recently reviewed (Mills and Tamnes 2014; Vijayakumar et al. 2018). Existing findings suggest that although pubertal status is associated with overall brain development and maturation, particularly in the frontal cortex, findings are inconsistent potentially due to disparities in study designs and measures used. Moreover, previous studies investigating white matter tract development in association with pubertal status are scarce. Consequently, it is still unclear how both age and/or pubertal development are associated with individual trajectories of specific regional brain development.

The present study sought to investigate individual trajectories in brain development comparing the influence of age and pubertal status with a longitudinal design of repeated within-subject measurements during adolescence. More specifically, we aimed to describe longitudinal changes in regional gray matter volume and white matter tract measures modelled by normative physical growth, age, or pubertal status in healthy adolescents tracked over 3 years.

Methods

Study design and recruitment

The study was approved by the human research ethics committee of the Medical Faculty at Heidelberg University, Germany (study ID: S-604/2011). All participants and at least one legal guardian signed written informed consent prior to inclusion in the study. Recruitment took place in Heidelberg, Germany in 2014 and 2015. Children were recruited from the general public using letters sent out to households (n = 2398) with children within the age range of interest, identified by contacting the citizens registration office at Heidelberg City Council, Germany.

Participants with a history of psychiatric diagnosis or treatment, developmental disorders, premature birth (birth weight below 2000 g and/or birth before 36th week of pregnancy), intellectual impairment (i.e., IQ < 80), poor knowledge of the German language, or those reporting endocrine disorders were excluded. Adolescents were also excluded if either of their parents had a history of psychiatric diagnosis or treatment. Exclusion criteria also included MRI safety regulations, in which participants reporting claustrophobia, those with metal implants, a history of brain injury, or vascular/neurological diseases (i.e., pathology that may influence brain function) were excluded.

After a screening interview, participants subsequently underwent a structured assessment (detailed below). Participants who agreed to undergo MRI scanning were invited to the Division of Radiology at the German Cancer Research Center, Germany, for a second appointment (detailed below). To capture 6 years of normative brain development, two cohorts of healthy children and adolescents were recruited: a) cohort 1, which consisted of 9–10 years old and b) cohort 2, which consisted of 12–13 years old at baseline. These children and adolescents were subsequently followed-up every year until 2 years after the initial scan, making a total of three time-points. Further details on participant flow and reasons for the exclusion of subjects are provided elsewhere (Mürner-Lavanchy et al. 2020).

Structured assessments

The first assessment consisted of interviews to collect data on demographics, pubertal status using the Pubertal Development Scale (Petersen et al. 1988). The PDS is a self-report measure composed of 5 items. Three general items (for boys and girls) assess growth spurts, changes in body hair, and skin. Each of these is rated on a 5-point scale: “not yet started”, “barely started”, “definitely started”, “seems complete”, and “I don’t know” (treated as missing). Girls are further asked to indicate breast growth and onset of menstruation. Boys are asked to indicate changes in voice and facial hair growth. Puberty status is indicated on a 5-point scale (prepubertal, early pubertal, mid pubertal, late pubertal, and post pubertal). Intelligence level was assessed using the General Ability Index (GAI; (Raiford et al. 2005), which is a compressed measure of general intellectual ability measured by the German version of the Wechsler Intelligence Scale for Children, fourth edition (Petermann and Petermann 2008; Wechsler 2003). Psychiatric health of the participants was confirmed using the German version of the Mini-International Neuropsychiatric Interview for Children and Adolescents (M.I.N.I- KID 6.0) (Sheehan et al. 2010). All interviews were performed by trained clinicians in the field of child and adolescent psychiatry.

Neuroimaging

MRI data were acquired using a Siemens 3 T Biograph mMR with a 16-channel head coil with a total acquisition time of 45 min. Anatomical T1-weighted images were acquired in the sagittal plane (192 slices, 1 mm slice thickness, 1 × 1 mm2 in-plane resolution, echo time (TE) = 2.98 ms, repetition time (TR) = 2300 ms, and flip angle = 9). A whole-brain single-shot spin-echo echo-planar imaging sequence was used to obtain the DTI images using the following parameters in axial plane: 50 slices, 64 directions, voxel size = 2.5 mm isotropic, TE = 112 ms, TR = 12,100 ms, FOV = 240 mm, matrix size = 96 × 96, and b values at 0, 1000 and 3000 s/mm). A GeneRalised Autocalibrating Partially Parallel Acquisition (GRAPPA) factor 2 was used.

Automated methods were used for calculating both gray matter volume and white matter tract measures. Each T1 and DTI image was visually checked for quality assurance by trained researchers before automated processing. Participants with abnormal scan reports or with movements of more than 2 mm were discarded from further analyses. Subsequent to analyses, outputs were overlapped onto native T1 image for additional quality assurance. For structural segmentation to calculate gray matter volumes, FreeSurfer version 6.0 was used to segment T1-weighted images (Fischl 2012). Details of FreeSurfer segmentation are described in previous papers by Fischl et al. (Fischl et al. 2002, 2004). Outputs from regions listed in the Desikan-Killiany-Tourville atlas were used (Klein and Tourville 2012). For calculating fractional anisotropy (FA) and mean diffusivity (MD) measures of the white matter tracts, DTI data were processed using TRACULA (TRActs Constrained by UnderLying Anatomy, FreeSurfer). This automated method includes pre-processing such as standard methods for image distortion correction of eddy currents and subject movement (Andersson and Sotiropoulos 2016) and B0 field inhomogeneity, intra-subject and inter-subject registration, and tensor fitting as detailed in previous papers (Yendiki 2011; Yendiki et al. 2014). As part of the FreeSurfer suite, TRACULA uses not only anatomical priors derived from an atlas, but also the cortical parcellation and subcortical segmentations derived from native space of each subject in FreeSurfer, allowing individual variations across subjects while still establishing the same tracts for comparison (Jbabdi et al. 2007).

Statistical analyses

A total of five multi-level linear regression models were calculated for each outcome of interest and compared to a null model (M0) using Bayesian model selection. The null model only included total segmented volume as fixed factor to predict regional brain volume or the white matter tract measure of interest (dependent variables). The four models of interest were composed as follows: Model 1 (M1): fixed factor age (exact age at the time of the scan including 6 decimals); Model 2 (M2): fixed factor puberty status; Model 3 (M3): both age and puberty status as fixed factors; Model 4 (M4): age and pubertal status main and interaction effects. All models included sex as a covariate and subject as random effect. Total segmented volume was added as additional covariate for all models of interest (M1 to M4) to control for global effects of brain size (e.g., O’Brien et al. 2011; Peelle et al. 2012 Each model (M1 to M4) was compared to the null model (M0) using Bayesian Information Criterion (BIC) differences as an estimation of the Bayes Factor (BF). Interpretation of the BF was based on the proposed convention by (Raftery et al. 2009), but with two studies showing sexual dimorphic patterns (Bramen et al. 2012; Hu et al. 2013). Interestingly, the amygdala and hippocampus did not show preference for pubertal status in the current study but rather the accumbens. In animal studies, the accumbens has been shown to be related to the dopaminergic pathways associated with pubertal hormonal changes, such that there is increased motivation for reward seeking behaviour including sexual behaviour (Sato et al. 2008). However, the distinct association of pubertal alteration in the accumbens in humans is still unclear and further longitudinal studies building on the present findings are needed.

Moreover, our analyses showed that neither FA nor MD of any of the tracts that were tested showed significant preference for the pubertal status model. However, previous research investigating white matter tract development in association with pubertal status has been scarce. While some studies implicate a predominantly positive association between pubertal status and FA (Herting et al. 2012; Menzies et al. 2015), only a few studies have observed any association with pubertal status and MD (Menzies et al. 2015). The few studies that investigated such association found similarities to findings on gray matter volume development, where cortico–cortico and cortico-subcortical tracts that are associated with the frontal and temporal lobes showed the most consistent associations with pubertal status (Herting et al. 2012; Menzies et al. 2015). In the current study, however, cortico-spinal and hemispheric connections were associated with age and pubertal status, where FA of both of these tracts were negatively associated with age and pubertal status. Most interestingly, beta values indicated that when both age and pubertal status are added as fixed factor in a joint model, FA has a positive association with age and a negative association with pubertal status. This finding highlights a potential role of cortico-spinal and hemispheric connections in studying norm-variants in delayed or premature pubertal development. Additional studies in more diverse samples are needed to ascertain and disentangle the intertwining influences of age and pubertal status in these regions of interest.

Inconsistencies seen with previous studies could be influenced by the various measures of pubertal status used. For example, there are two prominent systems for measuring pubertal status: the Tanner Stage (Tanner, 1692), and the PDS (Petersen et al. 1988). The two measures are not correlated, and therefore, they are potentially capturing differential aspects of pubertal development. The current study used the PDS (Petersen et al. 1988), and due to its self-report format, it could be considered as one of the limitations of the study, potentially introducing bias (Shirtcliff et al. 2009). Although the study design could improve by the addition of physical examination, we opted out of this option due to its intrusiveness. Another reason for inconsistent associations seen with brain development and puberty could be due to some studies using hormone measures as an indicator of pubertal development. Although hormones are a direct measure of pubertal maturation, there is unfortunately a wide range of variability in hormone levels during adolescence and its associated pubertal stages of development (Dorn et al. 2004; Dorn and Biro 2011). For example, hormone levels vary according to the time of day the sample is obtained, the menstrual cycle in females, diet, stress, and other environmental factors. Furthermore, differential collection and analysis methods, such as saliva, blood, and urine samples, have also produced varying results (Vesper et al. 2014). Measuring hormone levels from hair could provide a more stable measure in this regard.

One major pitfall of the present study is the two-cohort design that was applied to facilitate the recruitment process. Although this allowed a faster collection of data with a larger age range, a complete within-subject trajectory over 6 years could not be investigated. Therefore, we were only able to investigate the trajectory from three time-points for two cohorts rather than six time-points for the entire sample. The age of data collection is a crucial aspect for a study on longitudinal developmental. Past neuroimaging studies investigating gray matter development have reported widespread alterations in the cortex during adolescence. For example, an inverted U-shaped developmental pattern, peaking at various ages in different cortices, has been reported from studies using the National Institute of Mental Health Child Psychiatry dataset in males and females aged between 4 and 25 years (Giedd et al. 1999; Lenroot and Giedd 2006; Raznahan et al. 2014). However, other studies report a linear decrease in total gray matter volume across late childhood-to-adulthood, where the greatest proportion and highest rate of decrease in gray matter volume occurred in participants in their teens and no significant change was seen in participants aged between 22 and 32 years (Lebel and Beaulieu 2011; Tamnes et al. 2013; Wierenga et al. 2014). Another study showed stable volumes of gray matter up to 10 years old, and then, a decrease was observed between ages 10 and 20 (Aubert-Broche et al. 2013). These heterogeneous findings indicate that although there may not be a clear age of “peak” of gray matter volume, it reaches its maximum during late childhood and decreases throughout adolescence to adulthood, where volume loss decelerates. The current three time-points from ages 9 and 12 may not have best captured the peak of development. One advantage of the present study is that an automated method has been used to calculate both volume and white matter tract information, allowing fast, reliable, and easily reproducible results, which also reduces the risk of manual error or bias depending on the researcher. Moreover, the outputs from volumetric comparisons were used to compute white matter tract information, allowing the tract calculation to be completed in native space. However, given the focus on gray matter volume and white matter tracts, we did not assess other brain structural variables of potential interest such as surface area and cortical thickness. Finally, our sample represents a high-function group of individuals, as further reflected in the high average IQ scores, way beyond what would be expected for the norm. While likely linked to specifics in the regional catchment area, it is not clear how the present findings generalize to samples from a diverse educational or socioeconomic background.

To summarize, the present study highlights some regions of interest, which—alongside the majority of brain regions that are best predicted by age in their development—are driven in their development across adolescence by pubertal status. Further longitudinal studies are needed to replicate the findings in more diverse and larger samples across the pubertal age span. These studies could be improved by incorporating gonadal hormone measures in combination with physical examinations of pubertal stats. Furthermore, the addition of functional MRI and measures of socioenvironmental factors would aid our understanding on the physical development and associated changes in functional activity and connectivity.