Longitudinal Changes in T1w/T2w Estimates of Cortical Myelin with Age and Pubertal Timing

Abstract

Puberty may regulate changes in sensitive period plasticity during adolescence. Experience-dependent myelination is a mechanism that may underlie such changes in plasticity. Intracortical myelin can be indirectly indexed by the ratio of T1-weighted to T2-weighted MRI images (T1w/T2w). While age-related T1w/T2w changes have been documented, less is known about the contributions of pubertal timing (being earlier/later relative to peers). Using Bayesian hierarchical generalized additive models with longitudinal data from 9- to 18-year-olds in the Human Connectome Project in Development, we examined how age and pubertal timing relate to T1w/T2w. Results confirmed that age-related change is patterned along the sensorimotor-association axis, though longitudinal effects were smaller than prior cross-sectional estimates. Pubertal timing accounted for up to 1.8% of variance in T1w/T2w across parcels, with negligible effects in most parcels. Modest sex and regionally specific effects were identified: Early pubertal timing was linked to greater T1w/T2w in sensorimotor regions and lower T1w/T2w in association areas (especially dorsolateral and frontopolar cortices), but only in females. In contrast, late puberty was linked to higher T1w/T2w in association areas in both sexes and reduced T1w/T2w in several mid-ranking parcels. Future work should replicate these effects and investigate associations with cognitive and psychosocial development.

Windows of heightened plasticity known as sensitive periods are crucial for neurodevelopment (Hensch 2004). While the earliest sensitive periods facilitate sensory and motor milestones, adolescence may herald sensitive periods for higher-order cognitive development (Blakemore and Mills 2014; Cheng et al. 2024). Myelination is an experience-dependent mechanism contributing to the stability and maturity of neural systems (Mount and Monje 2017), serving to close sensitive periods (Sydnor and Satterthwaite 2023). Recent advances in non-invasive neuroimaging have expanded opportunities to investigate cortical myelin development.

The ratio of T1-weighted to T2-weighted MRI images (T1w/T2w ratio) may serve as an indirect proxy of intracortical myelination (Glasser and van Essen 2011; Glasser et al. 2022). Emerging evidence suggests that age-related changes in T1w/T2w are patterned along the sensorimotor-association (S-A) axis (Baum et al. 2022) — a cortical hierarchy reflected in the human brain’s structure, function, and evolutionary history (Sydnor et al. 2021). From late childhood to early adulthood, sensorimotor areas tend to show more rapid, nonlinear T1w/T2w increases that plateau with age. In contrast, association areas show slower, more linear age-related changes and lower overall T1w/T2w values, potentially reflecting more protracted plasticity (Baum et al. 2022). These findings point toward adolescence as a period of substantial age-related T1w/T2w change that may reflect the closure of childhood sensitive periods.

Puberty is a major maturational process that has been proposed to regulate sensitive periods (Piekarski et al. 2017). Undergoing puberty earlier relative to one’s peers, also known as early pubertal timing, is an established risk factor for negative psychosocial outcomes (Mendle 2014; Ullsperger and Nichols 2017; MacSweeney et al., 2025), with less clear effects on learning and cognitive development (Laube and Fuhrmann 2020). For neurodevelopment, puberty and its associated hormone changes have been linked to changes in the cortex, especially gray matter reductions in the frontal and temporal lobes (Vijayakumar et al. 2018) and reduced surface area (Curtis et al. 2024). In some subcortical regions, early pubertal timing has been associated with faster changes in gray matter volumes (Goddings et al. 2014; Wierenga et al. 2018). Diffusion imaging studies suggest that pubertal hormones (Barendse et al. 2018) and early pubertal timing (e.g., Beck et al. 2023, Chahal et al. 2018) correlate with white matter microstructure in sex-specific ways, beyond the effects of age (Herting et al. 2017; Curtis et al. 2025). More advanced puberty often corresponds to greater fractional anisotropy and lower mean diffusivity (Beck et al. 2023). Overall, existing findings support the notion that earlier puberty accelerates some aspects of brain maturation, potentially facilitating a premature closing of sensitive periods occurring during childhood. If this were the case, early pubertal timing would be associated with higher levels of T1w/T2w, especially in lower S-A rank regions.

The Present Study

The most rapid, non-linear T1w/T2w changes occur in sensorimotor regions and coincide with the onset and progression of puberty (Baum et al. 2022). Existing research suggests that early pubertal timing is a facet of puberty that is associated with neurodevelopment changes (Vijayakumar et al., 2018) and adolescent-emergent psychopathology (Mendle, 2014). Yet, direct links between pubertal timing and T1w/T2w development are unknown. Prior studies investigating T1w/T2w changes in child and adolescent samples have been largely cross-sectional (e.g., Boroshok et al. 2022; Norbom et al. 2022; Weissman et al. 2023; Dipnall et al. 2024 is longitudinal), and none to our knowledge have investigated the role of puberty. Characterizing puberty-myelin associations may aid in our understanding of how pubertal timing impacts sensitive period closures, as well as psychosocial and cognitive outcomes.

Therefore, this research uses longitudinal data to address two interrelated aims: First, we characterize age-related T1/T2w trajectories to assess the degree to which key findings from cross-sectional studies replicate longitudinally. Second, we establish how pubertal timing is related to cortical T1w/T2w development and whether these relationships vary by sex. We hypothesize that early pubertal timing will be linked to a higher T1w/T2w ratio in earlier-developing cortical regions (lower on the S-A axis) across sex, reflecting earlier closure of childhood sensitive periods.

Materials and Methods

Participants

The Human Connectome Project in Development (HCP-D) is a cross-sectional and longitudinal study in a diverse sample of youth ages 5–21, recruited across four sites: Harvard University, University of California-Los Angeles, University of Minnesota, and Washington University in St. Louis. Exclusion criteria included premature birth (<37 weeks gestation), serious neurological or endocrine conditions, long-term immunosuppressant or steroid use; history of serious head injury, hospitalizations >2 days due to certain physical or psychiatric conditions or substance use, psychiatric treatment of >12-month duration, claustrophobia, and pregnancy or other MRI contraindications. Participants provided informed consent and assent. For participants under 18 years of age, parents provided written informed consent. Procedures followed legal and institutional guidelines (Washington University in St. Louis IRB #201603135).

The present study focused on the three-wave longitudinal subsample (ages 9–18) because it followed participants during active phases of pubertal onset and change. HCP-D followed an accelerated longitudinal design with four cohorts collected in parallel (Somerville et al. 2018; Omary et al. 2025). Females were recruited younger to capture their earlier pubertal onset. The younger female cohort was recruited at ages 9–10 years and followed to 11–14; the older female cohort was recruited at 13–14 and followed to 15–17; the younger male cohort was recruited at 10–11 and followed to 12–15; and the older male cohort was recruited at 14–15 and followed to 16–18.

Of the full longitudinal subsample, a behavior-only protocol sans neuroimaging was completed at 54 observations, and one observation was excluded for poor resting-state data quality (impacting the surface registration pipeline). The final analytic sample included 254 participants (128 or 50% female) with 648 observations.

Measures

Pubertal Timing

A pubertal stage composite score was derived from two self-report questionnaires. On the Sexual Maturation Scale (SMS), participants selected among five line drawings to indicate which most closely resembled their physical development (Morris and Udry 1980). Line drawings corresponded to each of the five Tanner Stages (Stage I: pubertal development has not begun; Stage V: completion of puberty). Males reported on genital and pubic hair development; females reported on breast and pubic hair development.

The Pubertal Development Scale (PDS) inquires about the degree to which participants have progressed through height, body hair, and skin changes (Petersen et al. 1988). Males additionally reported on facial hair and voice deepening; females reported on breast development and menstruation. Scores were converted to a five-point scale to approximate Tanner stages (Shirtcliff, Dahl, and Pollak 2009).

Scores on both questionnaires were averaged to create a composite score (Byrne et al. 2023; Omary et al. 2025). When only one questionnaire was completed, that score was used (n = 11). As questionnaires evaluated self-perceived puberty (Cheng et al. 2021), participants sometimes had lower scores at later time points (13.6% of observations; 8.4% in females, 18.9% in males). Two participants were missing data on both puberty questionnaires and were excluded from puberty models.

Neuroimaging

Image Acquisition.

Data acquisition and processing pipelines were adapted for developmental populations (Harms et al. 2018; Baum et al. 2022). As described in Baum et al. 2022, high-resolution T1w MRI images were acquired on a 3T Siemens Prisma with a 32 channel head coil using a 3D multiecho MPRAGE sequence (Mugler and Brookeman 1990; van der Kouwe et al. 2008; 0.8-mm isotropic voxels, TR/TI = 2500/1000 ms, TE = 1.8/3.6/5.4/7.2 ms, flip angle = 8°, in-plane (iPAT) acceleration factor of 2, TA = 8:22, up to 30 reacquired TRs). Structural T2w images were acquired at 0.8 mm isotropic using the variable-flip-angle turbo-spin-echo 3D SPACE sequence (Mugler et al. 2000; TR/TE = 3200/564 ms; same in-plane acceleration, TA = 6:35, up to 25 reacquired TRs).

As in the HCP Young Adult pipeline, non-normalized reconstructions were used as processing inputs (Van Essen et al. 2013). To correct for subject motion between the T1w and T2w images, both versions of the T1w image were used to estimate the B1– receive field. Only the first two echoes of the T1w image were used, as artifacts in later echoes can impact surface reconstructions and T1w/T2w maps (Elam et al. 2021).

Real-time motion correction can reduce bias in metrics of brain morphometry (Reuter et al. 2015; Tisdall et al. 2016). Therefore, volumetric navigators (vNavs) were embedded to support prospective motion correction and selective reacquisition of heavily motion-corrupted lines in k-space (Tisdall et al. 2012). If the quality of T1w or T2w scans was deemed inadequate during acquisition, they were reacquired—typically within the same session but occasionally in a different one. The highest-quality pair from a single session was selected for processing. Additionally, 2-mm isotropic gradient echo (GRE) and spin echo (SE) images were collected and employed to compute the pseudo-transmit field.

Image Processing.

Structural MRI data were analyzed using the HCP Pipelines (Glasser et al. 2013) version 4.0.0, instantiated into the QuNex container environment (qunex.yale.edu). The PreFreeSurfer pipeline for processing T1w and T2w volumes included gradient nonlinearity distortion correction, initial brain-extraction, rigid registration into an anterior/posterior-commissure aligned “native” space, registration of the T2w volume to the T1w volume using boundary-based registration (Greve and Fischl 2009), correction for the receiver coil bias field based on the smoothed square root of the product of the T1w and T2w images, and registration of the structural images to MNI space. Next, the FreeSurfer pipeline (v6.0.0; Dale et al. 1999; Fischl 2012) was used to compute “white” and “pial” surfaces. Finally, the PostFreeSurfer pipeline produced cortical surface data in GIFTI and CIFTI formats, registering each subject’s cortical surface to a common 32k_FS_LR mesh using areal-feature-based cortical surface registration (multimodal registration constrained by cortical T1w/T2w and resting-state maps; Robinson et al. 2018; Glasser et al. 2016b).

An experienced individual reviewed the quality of the white and gray matter surface placement, informed by the T1w/T2w maps (Glasser and Van Essen 2011; Elam et al. 2021). Participants with more than minor (focal) issues underwent manual editing (n=7, 12 observations). Previous analyses revealed less accurate surface segmentation in HCP-D compared to the HCP Young Adult study. As this issue was traced to artifacts in the longer echos, subsequent analyses used the mean of the shortest two echoes (i.e., excluded the longest two of four echoes) as the T1w input (Elam et al. 2021).

T1w/T2w Processing.

Following methods developed by Glasser and colleagues (2011, 2013, 2014, 2022), T1w/T2w maps were created both as cortical ribbon volumes and as surface maps. In short, the T1w/T2w ratio was calculated for putatively gray matter voxels (located between white matter and pial surfaces) and then projected to the surface mesh. To reduce partial volume effects, voxels near tissue boundaries (pial and gray/white surface boundaries) were de-emphasized (Glasser et al. 2013). Taking the ratio of T1w to T2w images enhances the contrast related to myelin (Glasser and Van Essen 2011) and cancels out intensity biases from the radio frequency receiver coils (since both images share the same sensitivity profile, not accounting for head motion). Subsequent maps were parcellated into 360 parcels (vertex values averaged within-parcel) according to the HCP-multimodal atlas (Glasser et al. 2016a) via the wb_command -cifti-parcellate function in Connectome Workbench v1.4.2 (Marcus et al. 2011). Resulting T1w/T2w values are in arbitrary units comparable within a consistently acquired dataset, and higher values may indirectly index greater myelination.

B1+ Transmit Field Correction of T1w/T2w Myelin Maps.

This study employed a recent, empirically validated “pseudo-transmit field” correction to mitigate B1+ bias in T1w/T2w maps to reduce spurious body-size (and therefore age-related) T1w/T2w differences (Glasser et al. 2022). Pseudo-transmit field maps were computed by averaging GRE/SE image ratios across phase encoding directions. A reference T1w/T2w map was generated at the group level by finding the scaling between the group pseudo-transmit field and group T1w/T2w map that minimized spurious left-right asymmetries. For individual correction, the pseudo-transmit map was scaled to minimize correlated differences between individual T1w/T2w maps, the reference T1w/T2w map, and the pseudo-transmit map. For details, see Glasser et al. (2022).

Statistical Analyses

Analyses were performed in R version 3.5.1 (R Core Team 2018). Models accounted for the scanner used and for B1+ transmit field covariates. As detailed in Glasser et al. (2022), these include scanner transmit voltage, the mean of the pseudotransmit map, and four regularization parameters (T2* dropout threshold, full-width at half maximum (FWHM), a correction factor adjusting for the impact of smoothing on the pseudotransmit field’s intensities, and the slope parameter of the correction); additionally, there was a T1w/T2w lateral ventricular CSF regressor adjusted to exclude partial volume voxels and CSF flow effects.

To assess the longitudinal stability of T1w/T2w, intra-class correlation coefficients (ICCs) were calculated for each parcel using linear mixed effects models with a random subjects intercept (lmer; lme4 package; Bates et al. 2015). ICCs reflect the proportion of total variance attributed to between-subject variability.

To model developmental effects, we fit separate Bayesian generalized additive models (GAMs) using brms (Bürkner 2018) with the Stan backend (Stan Development Team 2021). This approach estimated linear and non-linear effects without specifying a functional form (e.g., polynomial; Wood 2017). Thin plate regression splines (default k=10 basis dimensions) formed the smoothing basis. Default priors were used, including flat priors for regression coefficients, Student’s t distributions for the intercept (df=3, μ=outcome mean, σ=2.5), and half-t distributions with a lower bound of zero for the spline/error term standard deviations (df=3, μ=0, σ=2.5). The model ran 4 chains of 4,500 iterations (2,000 warmup), yielding 10,000 posterior draws. Gaussian distributional assumptions and model fit were evaluated using posterior predictive checks (pp_check from brms). Models generally adequately reproduced the overall shape of the observed distribution, with a peak at the mean. Moderate mismatches (overly wide predicted distributions) were observed in 6 parcels (1.7%), none of which exhibited credible effects.

Age Models

In brms syntax, the age model for each parcel was:

$$\text{brms}(\text{T1w}/\text{T2w}~\text{s}(\text{age})+(1|\text{scanner/subject})+\text{covariates}$$

Age was mean-centered and modeled as a smooth to capture nonlinear effects. Using bayes_r2 from brms, partial R² was calculated by comparing models with and without the age term. Note that sex was not included as a covariate because it explained little T1w/T2w variance (up to .41% across parcels) and partial R² attributed to sex was not correlated with S-A rank (r=−.02, p_spin=.374). Excluding sex as a covariate also maximized comparability of our findings with prior cross-sectional analyses, which did not include sex (Baum et al., 2022).

Associations with the S-A axis.

To investigate whether partial R² was patterned along the S-A axis, the Spearman correlation between partial R² and S-A axis rank was calculated across parcels. S-A axis rankings were based on ten key brain characteristics that varied systematically across sensorimotor and association regions (Sydnor et al. 2021). Parcel-based spatial permutation tests determined statistical significance of the correlation (p_spin) in a manner that provided family-wise error control and accounted for spatial dependencies in cortical data (Alexander-Bloch et al. 2018). Spatially randomized maps were created through 1,000 rotations on the cortical surface, maintaining spatial contiguity and hemispheric symmetry. The null distribution of Spearman correlation coefficients was then calculated between the S-A axis map and the randomly rotated map, and p-values were determined by comparing the observed correlation to the null distribution.

Pubertal Timing Models

In brms syntax, the puberty model for each parcel was:

$$\text{brms}(\text{T1wT2w}~\text{s}(\text{pubertal\_stage},\text{by}=\text{sex})+\text{s}(\text{age})+(1|\text{scanner/subject})+\text{covariates}$$

This model was designed to evaluate the non-linear effect of pubertal timing across all time points for each sex. When pubertal stage is entered into models that also control for age, the resulting puberty effect estimates (“stage for age”) are commonly interpreted as reflecting pubertal timing (Cheng et al., 2021; Vijayakumar et al., 2018). Puberty was centered at stage 3, inverted (such that earlier puberty is on the left in visualizations), and treated as non-linear and time-varying. Factor-smooth interactions with sex were included given theory and evidence for sex-specific impacts of puberty on neurodevelopment (e.g., Herting et al. 2017). This specification estimated separate non-linear effects of pubertal timing for each sex, which follows best practices because pubertal stages are not necessarily biologically aligned across sex (e.g., despite being similarly named, stage 3 in males and stage 3 in females mark points along distinct developmental paths). Partial R² for pubertal timing (across sex) was calculated by comparing models with and without the factor-smooth term.

Additional parameters of interest.

For greater specificity and interpretability (given hundreds of parcels with smooth effects), we investigated early and late puberty for each sex. For each parcel and sex, the difference between the expected T1w/T2w associated with (1) early or late puberty (±1 sd from mean puberty) and (2) on-time (mean) puberty was calculated. The mean and standard deviation of puberty was calculated across all time points and separately for each sex. When calculating expected T1w/T2w, age was fixed to its mean for each sex (12.7 for females, 13.6 for males) and all other covariates were fixed to their median values. Differences (early versus on-time, late versus on-time) were calculated for each of 10,000 posterior draws using posterior_epred from brms; mean and 90% uncertainty intervals were calculated from the distribution of differences and identified as credible based on certain criteria (see Thresholding).

Associations with the S-A axis.

Following the same procedure as the age models, spatial permutation tests evaluated the correlation between mean T1w/T2w differences and S-A rank. Bonferroni-correction (ɑ=0.025) was applied when males and females were tested separately. Correction was not applied across early and late puberty, as there were separate hypotheses for these conditions.

Transparent Reporting of Analysis Decisions

Planned Model Tuning

A pre-registration and addendum were posted prior to examining full results (https://osf.io/5vmja/). Exploration within pre-registered parameters is described in the Supplementary Materials (S1 and S2), including choices (1) to highlight longitudinal analyses, (2) to model puberty at each time point, and (3) to use smooths for puberty and age.

Additional Adjustments

Divergent transitions.

Divergent transitions occur when the sampling algorithm (Hamiltonian Monte Carlo) has difficulty exploring the full range of the posterior distribution; their presence can indicate reduced reliability. Post-hoc exploration adjusting the random effects structure found that removing the random slope of age, but not scanner, substantially reduced the frequency and magnitude of divergent transitions across parcels. Thus, reported results are from models that included random intercepts for subject and scanner only (Supplementary Materials S3).

Thresholding.

Differences were considered credible if their uncertainty interval excluded zero (as pre-registered) and if their mean difference exceeded 0.01—an additional criterion to prevent overinterpreting minimal deviations from zero. T1w/T2w units are relative and only appropriate for within-sample comparisons. In our sample, mean whole brain T1w/T2w approximated a normal distribution (M=1.53, SD=.08); .01 from the mean covers ~5% of observations. In the results, effect sizes are further contextualized relative to each parcel’s unique variance.

Post-Hoc Analyses

Pre-registered models estimated separate puberty smooths by sex. To directly test puberty-related sex differences, we examined the overlap in 90% uncertainty intervals for early vs. on-time and late vs. on-time puberty across sex, focusing only on parcels credibly linked to early or late puberty.

We also explored whether threatening childhood experiences confounded early puberty effects. However, preliminary checks found no significant association between such experiences and puberty in our sample (t = −1.056, p = .292; Supplementary Materials S4).

Data Access and Availability

Minimally preprocessed HCP-D neuroimaging data (https://nda.nih.gov/) and parcellated summary maps of this study (https://balsa.wustl.edu/study/jX652) are available for download.

Results

Participants

We examined effects of age and pubertal timing on T1w/T2w using longitudinal data from adolescents (254 participants, 648 observations, 9–18 years of age). Of the analytic sample, 5.5% participants were Asian, 11% were Black/African American, 65.8% were White, 16.9% were more than one race, and 0.8% had a racial background that was unknown or not reported. Further, 13.1% were Hispanic or Latino, 85.5% were not, and 1.39% were unknown or not reported. For age and pubertal stage distributions, see Figures 1–2 and Table 1.

Figure 1.

Three-wave longitudinal sample by age and sex. Each line represents a participant and each dot represents a datapoint. The Human Connectome Project in Development sample followed an accelerated longitudinal design with males and females recruited into younger and older cohorts. Data collection began a year earlier in females to capture their earlier onset of puberty. The timing of the third time point was more variable due to the Covid-19 pandemic. Analyses included 646 observations from 254 participants.

Figure 2.

Distribution of the pubertal composite score by age and sex in the analytic sample. Splines were used to fit major trend lines for males and females separately. Faint lines connect longitudinal data from the same participant.

Table 1.

Longitudinal participants’ average maturation per time point, per cohort. All analyses were collapsed across cohorts. Pubertal stage is an approximate Tanner Stage ranging from 1–5 based on a composite score averaged across two self-report measures.

Time point	Cohort	Sex	N	Age (Mean)	Age (SD)	Pubertal Stage (Mean)	Pubertal Stage (SD)
1	Younger	Female	66	9.51	0.32	1.52	0.54
		Male	67	10.5	0.32	1.69	0.52
	Older	Female	62	13.5	0.33	3.79	0.74
		Male	59	14.5	0.32	3.38	0.71
2	Younger	Female	57	10.7	0.36	2.12	0.72
		Male	55	11.7	0.35	1.95	0.7
	Older	Female	52	14.8	0.32	4.25	0.51
		Male	52	15.7	0.31	3.83	0.64
3	Younger	Female	44	12.3	0.47	3.23	0.81
		Male	39	13.3	0.53	2.95	0.98
	Older	Female	48	16.3	0.43	4.51	0.49
		Male	47	17.2	0.39	4.21	0.60

Intra-Class Correlations

Across parcels, ICCs ranged from .17 to .65 and were negatively correlated with S-A axis rank (r=−.61, p_spin<.001; Figure 3). As the highest ICC was .65, there was substantial within-person variance (>35%) across parcels that could be explained by time-varying factors.

Figure 3.

Intra-class correlations (ICCs) were patterned along the S-A axis. A. ICCs visualized across parcels. V2 is outlined as a representative region of sensorimotor cortex (low SA axis rank), while ACC is outlined as a representative region of association cortex (high SA axis rank). B. Strong negative correlation between ICC and S-A ranking. Lower ICCs seen in association areas suggest greater within-person variability, indicating that stable individual differences have yet to emerge. Conversely, higher ICCs seen in sensorimotor areas suggest greater between-person stability (i.e., more stable individual differences). V2 is indicated by the blue dot, and ACC is indicated by the yellow dot.

Age

We sought to replicate age-related changes observed in prior cross-sectional research with a partly overlapping sample (Figure 4; Baum et al. 2022). Age explained up to 7.37% of additional variance across parcels (M=3.05%, SD=1.82%), and age partial R² was negatively correlated with S-A axis rank (r=−.58, p_spin<.001). As in cross-sectional work, age explained a greater proportion of variance in sensorimotor areas (e.g., right V2, R²=.040) compared to association areas (e.g., left anterior cingulate cortex, R²=.005; Figure 4B).

Figure 4.

Longitudinal analyses replicate topographic patterns of age-related change in T1w/T2 seen in cross-sectional analyses (Baum et al. 2022). T1w/T2w values reflect properties of cortical microstructure associated with myelin such that higher values may index greater myelination. A. Mean T1w/T2w per parcel across different age bins suggest higher values with increasing age, especially in sensorimotor regions. B. Partial R2 associated with age across parcels. V2 is outlined as a representative region of sensorimotor cortex (low SA axis rank), while ACC is outlined as a representative region of association cortex (high SA axis rank). C. Up to ~7% of variance is explained by age, which was lower than prior cross-sectional estimates of up to ~30%. Age partial R2 is negatively correlated with S-A axis rank (r=−.58, pspin<.001), which is notably similar to prior cross-sectional estimates (r=−.65, pspin<.001). V2 is indicated by the blue dot, and ACC is indicated by the yellow dot.

The finding that ~7% of variance is explained by age in the present longitudinal analysis is smaller than prior cross-sectional estimates (up to ~30%). Additional sensitivity analyses found that the reduction in R² in the longitudinal model is unlikely to be due to true age effects being absorbed by subject-level random effects, as these random effects were generally uncorrelated with age (across parcels, r ranged from-.06 to .13; M=.04). Furthermore, while recent work suggests that failing to separate between- and within-person effects can attenuate age-related effect sizes (Kang et al., 2024), ensuring separation of these effects did not meaningfully boost age-related R². (For more details about these sensitivity analyses, see Supplementary Materials S5.)

Pubertal Timing

As shown in Figure 5, pubertal timing smooths for each sex explained up to 1.82% of additional variance across parcels (M=.40%, SD=.31%), and partial R² was not correlated with S-A rank (r=−.09, p_spin=.268). For greater interpretability and specificity, we further investigated early and late puberty for each sex. Note that sex alone explained little variance (up to .41%) and partial R² attributed to sex was not correlated with S-A rank (r=−.02, p_spin=.374).

Figure 5.

Pubertal timing has predominantly negligible effects on T1w/T2w, and the magnitude of pubertal timing effects is not patterned by the S-A axis. A. Partial R2 associated with pubertal timing across parcels. B. Up to ~2% of variance is explained by pubertal timing, and puberty partial R2 is not correlated with S-A axis rank (r=−.09, pspin=.268). Note that the inclusion of an additional uninformative variable can lead to a small decrement in R2 when using the bayes_r2 algorithm, as reflected in the small proportion of parcels with puberty R2 < 0.

Early Puberty

Differences in expected T1w/T2w for early versus on-time puberty were calculated and averaged for males and females across posterior draws. On-time puberty was defined as the mean stage (2.9 for males, 3.2 for females), and early puberty as one standard deviation above the mean (SD_male=1.2; SD_female=1.3).

After thresholding, four parcels within the prefrontal cortex (PFC), specifically in rostral dorsolateral and frontopolar regions (Brodmann areas 9 and 10), showed credible T1w/T2w differences with early puberty in females only (Figure 6; Table 2). Mean differences ranged from .02–.03, reflecting 17%−23% of a standard deviation across the four parcels. T1w/T2w values were highly correlated across these parcels (r>.79); the left rostral PFC was selected for visualization (Figure 7).

Figure 6.

Associations between early pubertal timing, T1w/T2w, and S-A rank for males and females. A. Mean differences and 90% uncertainty intervals across 10,000 posterior draws revealed parcels with credible T1w/T2w differences related to early versus on-time puberty. Credible differences appeared only in females, with early puberty linked to lower T1w/T2w in prefrontal parcels. B. A significant negative correlation between mean T1w/T2w differences and S-A rank was observed in females only. C. Whole-brain maps of mean T1w/T2w differences in females and males. Parcels with credibly lower T1w/T2w linked to early puberty are outlined and labeled in blue.

Table 2.

Early versus on-time puberty.

Sex	Contrast	Region	Glasser parcel (abbreviation)	Mean difference [90% UI]
Females	On time > Early	Left rostral PFC	L area 10p polar (L_10pp)	−.021 [−0.041, <−.001]
		Right rostral PFC	R area 10p polar (R_10pp)	−.030 [−.054, −.007]
		Right dorsolateral PFC	R area 9 anterior (R_9a)	−.028 [−.048, −.006]
			R area 9 posterior (R_9p)	−.026 [−.044, −.005]

L = left, R = right, PFC = prefrontal cortex. Effect sizes displayed are mean differences (T1w/T2w expected for early minus on-time puberty) along with 90% uncertainty intervals across posterior draws.

Figure 7.

Conditional effects of pubertal timing on T1w/T2w development in the left rostral prefrontal cortex. A. Region outlined in blue, B. Approximately linear puberty trend in females, but not males, illustrated by plotting expected T1w/T2w and its 90% uncertainty interval by pubertal timing and sex, controlling for age and other covariates. Individual unadjusted data points are in light gray. C. Lower T1w/T2w with early puberty in females only, illustrated by expected T1w/T2w ratios for early versus on-time puberty and their 90% uncertainty intervals. Credible sex differences between males and females were not found in this region, but rather in the right rostral PFC (not shown).

Across parcels, the correlation between mean T1w/T2w differences and S-A rank was negative and statistically significant in females (r=−.65, p_spin<.001) but not in males (r=.05, p_spin=.381). In females, early puberty was linked to greater T1w/T2w in sensorimotor areas and lower T1w/T2w in association areas (Figure 6B).

Late Puberty

Following the same procedures for early puberty, we examined late versus on time puberty (Figure 8; Table 3). In females, five parcels showed credible differences. Four (left auditory and posterior orbitofrontal cortex; areas of right dorsolateral PFC) exhibited greater T1w/T2 with late puberty. The fifth (left piriform cortex) exhibited greater T1w/T2w with on-time puberty. In males, three parcels showed credible differences. Only the subgenual anterior cingulate cortex exhibited greater T1w/T2w with late puberty; the left presubiculum and right perirhinal cortex exhibited greater T1w/T2w with on-time puberty. Absolute mean differences ranged from .01–.08; reflecting 14–40% of a standard deviation in T1w/T2w across the eight parcels.

Figure 8.

Associations between late pubertal timing, T1w/T2w, and S-A rank for males and females. A. Mean differences and 90% uncertainty intervals across 10,000 draws of the posterior distribution reveal parcels with credible T1w/T2w differences related to late versus on-time puberty across sex. B. A significant positive correlation between mean T1w/T2w differences and S-A rank is observed across sex. C. Whole-brain maps of mean T1w/T2w differences in females and males. Parcels with credibly higher T1w/T2w linked to on-time puberty are outlined and labeled in blue; those with credibly higher T1w/T2w linked to late puberty are in red.

Table 3.

Late versus on-time puberty.

Sex	Contrast	Region	Glasser parcel (abbreviation)	Mean difference [90% UI]
Females	Late > On time	Left orbitofrontal cortex	L posterior orbitofrontal complex (L_pOFC)	.077 [.024, .134]
		Left auditory 4 complex	L auditory 4 complex (L_A4)	.043 [.012, .073]
		Right dorsolateral PFC	R area 9 anterior (R_9a)	.030 [.015, .046]
			R area 9 posterior (R_9p)	.028 [.014, .043]
	On time > Late	Left piriform cortex	L piriform cortex (L_Pir)	−.054 [−.098, −.011]
Males	Late > On time	Right subgenual anterior cingulate cortex	R area 25 (R_25)	.070 [.024, .120]
	On time > Late	Left presubiculum	L presubiculum (PreS)	−.035 [−.067, −.005]
		Right perirhinal cortex	R perirhinal ectorhinal cortex (R_PeEC)	−.013 [−.024, −.001]

L = left, R = right, PFC = prefrontal cortex. Effect sizes displayed are the mean differences (T1w/T2w expected for late minus on-time puberty) along with 90% uncertainty intervals across posterior draws.

The correlation between mean T1w/T2w differences and S-A rank was similarly positive across females (r=.40, p_spin=.005) and males (r=.33 p_spin=.061), but only significant in females. Late puberty was linked to greater T1w/T2w in association areas (Figure 8B).

Post-Hoc Analyses

Of the ten parcels showing credible differences with early or late puberty, only the right rostral prefrontal parcel exhibited a credible sex difference. For this parcel, a lower mean T1/T2w with early puberty was identified in females (−0.030, 90% UI [−0.054, −0.007]) compared to males (0.037, 90% UI [−0.004, 0.077]).

Discussion

This study examined age- and puberty-related changes in T1w/T2w using longitudinal data from 9–18 year olds in the Human Connectome Project in Development study. Compared to prior cross-sectional analyses (Baum et al. 2022), longitudinal models revealed a similar topographic pattern of age-related effects, albeit with attenuated effect sizes. At most, pubertal timing accounted for 1.8% of variance in T1w/T2w across parcels, and pubertal timing effects were negligible in most parcels. Modest sex and regionally-specific effects were identified: Puberty-related findings partly supported the hypothesis that early puberty is linked to greater T1w/T2w in sensorimotor areas, but only in females. In females only, early puberty was also associated with lower T1w/T2w in association areas, including credibly lower T1w/T2w in rostral aspects of the prefrontal cortex. In contrast, late puberty was linked to greater T1w/T2w in association areas across sex and a more spatially diffuse pattern of credible differences.

Age

Longitudinal results replicated the topographic pattern of age-related T1w/T2w change along the S-A axis seen in prior cross-sectional work, with earlier and more dramatic changes in sensorimotor areas (Baum et al. 2022). Intra-class correlations suggest greater T1w/T2w stability in sensorimotor areas than in association areas, consistent with their earlier maturation. However, a more modest proportion of variance (up to ~7%) was explained by age in longitudinal analyses, compared to ~30% in prior cross-sectional work (Baum et al. 2022). This may reflect unmodeled individual differences and/or the inclusion of younger participants (by 1–2 years) in prior analyses.

Pubertal Timing

Early Puberty

The hypothesis that early puberty would be associated with greater T1w/T2w in sensorimotor areas was partly supported in females only. Early pubertal timing was associated with greater T1w/T2w in sensorimotor regions and lower T1w/T2w in association regions in females, reflecting greater cortical asymmetry across the S-A axis. In cross-sectional analyses with a wider age range (Supplementary Materials S2), the primary hypothesis was supported in males, suggesting that effects may vary by age and study design. Greater T1w/T2w in sensorimotor areas is consistent with theories that puberty accelerates the closure of sensitive periods in middle childhood (Piekarski et al 2017). Despite this trend, no sensorimotor regions exhibited credibly greater T1w/T2w.

Instead, early puberty was associated with credibly lower T1w/T2w in right dorsolateral PFC (rostromedial Brodmann area 9) and bilateral frontopolar cortex (area 10) in females. Right prefrontal networks may play a central role in inhibitory control, with right dlPFC implicated in affective regulation (Depue et al. 2016; White et al. 2023) and ventral frontopolar cortex implicated in guiding goal-directed behavior (Hogeveen et al. 2022; Mansouri et al. 2017) and imbuing stimuli with value and emotion (Orr, Smolker, and Banich 2015). The right frontopolar parcel exhibited credibly reduced T1w/T2w with early puberty in females versus males. Parcels credibly linked to early puberty are among the highest S-A rank association areas, exhibiting very low levels of age-related T1w/T2w change. Within this age range, it is unclear whether slightly lower T1w/T2w reflects a slower progression toward sensitive period closure, or rather individual differences in microcircuit specialization (Demirtas et al. 2019). Future studies with additional measures of plasticity or a focus on individual differences may be able to parse these interpretations.

While identified regions support diverse cognitive processes, their shared role in affective processing may be relevant for understanding puberty. Our findings add nuance to prior findings linking early puberty with accelerated neurodevelopment (Dehestani et al. 2023; Holm et al. 2023), finding this trend primarily in sensorimotor areas and in females only. Findings align with prior research showing stronger puberty-related structural neurodevelopmental changes in females (e.g., Beck et al. 2023; Wiglesworth et al. 2023). Early puberty is a risk factor for mental health problems in females (Mendle et al. 2014; but see Ullsperger and Nichols 2017 for a contrasting view), and sex- or gender-specific factors may moderate associations between pubertal-timing and psychopathology (Bu et al. 2025, Vijayakumar et al. 2023). Future studies might integrate neurodevelopmental T1w/T2w profiles of early puberty into developmental psychopathology research (as in Patel et al. 2022).

Although studies linking puberty, neurodevelopment, and psychopathology have primarily focused on early timing, pubertal tempo (the rate of pubertal progression) is also a promising avenue for future study. As puberty unfolds non-linearly, estimates of pubertal tempo derived from participant slopes in three-time point studies are suboptimal, especially across large age spans. Accordingly, the present analyses focused on pubertal timing. Recent work using pubertal measures collected at up to five time points has examined associations between pubertal tempo and aspects of brain structure (McCann et al., in press), and future analyses could extend such approaches to examining associations with intracortical myelin development.

Late Puberty

Comparisons of late versus on-time puberty revealed spatially diffuse effects in males and females. Late puberty was linked to greater T1w/T2w in association areas, though this was statistically significant in females only. In females, credibly greater T1w/T2w was observed in two prefrontal regions implicated in affective processing (right dorsolateral PFC and posterior OFC), and in a secondary auditory area (auditory complex 4/Te3). In males, late puberty was linked to credibly greater T1w/T2w in the subgenual ACC, a region implicated in processing negative emotion (Vogt et al. 2005) and targeted in depression treatment (Lozano et al. 2008; Weigand et al. 2018). Late puberty was also associated with lower T1w/T2w in mid-rank parcels implicated in cognitive and sensory functions: spatial processing (presubiculum; Simonnet and Fricker 2018) and recognition memory (perirhinal cortex; Brown and Aggleton 2001) in males and olfactory processing (piriform cortex; Bolding and Franks 2020) in females.

The association between later puberty and greater T1w/T2w in association areas may reflect extended windows of cortical plasticity (Piekarski et al. 2017), cognitive development, and sociocultural learning (Worthman and Trang 2020) with delayed development. Links between T1w/T2w and cognitive or learning outcomes are mixed and may vary by age (Boroshok et al. 2023; Grydeland et al. 2013; Langense et al. 2022; Norbom et al. 2020; Patel et al. 2022). The literature relating pubertal timing to cognitive development is also nascent, with a recent review highlighting that early puberty, rather than late puberty, may confer cognitive advantages in some contexts (Laube et al. 2020). Given this, we refrain from interpreting T1w/T2w differences associated with early or late puberty as cognitive advantages or deficits.

Strengths and Limitations

Strengths include the recruitment of diverse participants from a longitudinal multi-site study, conservative correction for B1+ field effects, Bayesian modeling (circumventing multiple comparisons issues), and non-linear modeling to detect distinct effects of early and late puberty.

Limitations include that the T1w/T2w ratio is an indirect proxy of intracortical myelination influenced by additional microstructural properties of the brain (Carey et al. 2018). Also, pubertal timing was assessed using self-reported measures that capture perceived pubertal development (Cheng et al. 2021). Though widely used, such measures are subject to desirability biases (Schlossberger et al. 1992; Shirtcliff et al. 2009). Finally, controlling for age, puberty timing effects were small overall and only one parcel exhibited a statistically distinct effect of puberty by sex. Cross-sectional registration during preprocessing and the use of B1+ field covariates known to be correlated with body size (Supplementary Materials S6) may have reduced sensitivity to detect effects. Sensitivity may have also been lower in males due to greater pubertal measurement error (i.e., greater proportion of reported pubertal stage regressions). Given these considerations, the present analyses provide conservative effect estimates.

Conclusions

This study used Bayesian generalized additive models to characterize longitudinal T1w/T2 development in relation to age and pubertal timing. Age explained up to ~7% of variance in T1w/T2w and variance explained was patterned by S-A rank.

Puberty additionally explained up to ~2% of variance, with no such pattern; negligible additional variance was explained in most parcels. Separate examinations of early and late puberty revealed that early puberty in females was linked to greater T1w/T2w in sensorimotor areas and lower T1w/T2w in association areas, especially in rostral prefrontal regions. Late puberty was linked to greater T1w/T2w in association areas across sex. Together, findings align with resting-state EEG findings linking both early and late puberty to altered hierarchical cortical organization (Szakács et al. 2024). Our findings suggest that pubertal timing contributes to aspects of cortical myelin development and point to future potential links with psychosocial and cognitive outcomes.