Regional growth trajectories of cortical myelination in adolescents and young adults: longitudinal validation and functional correlates

Abstract

Adolescence is a time of continued cognitive and emotional evolution occurring with continuing brain development involving synaptic pruning and cortical myelination. The hypothesis of this study is that heavy myelination occurs in cortical regions with relatively direct, predetermined circuitry supporting unimodal sensory or motor functions and shows a steep developmental slope during adolescence (12–21 years) until young adulthood (22–35 years) when further myelination decelerates. By contrast, light myelination occurs in regions with highly plastic circuitry supporting complex functions and follows a delayed developmental trajectory. In support of this hypothesis, cortical myelin content was estimated and harmonized across publicly available datasets provided by the National Consortium on Alcohol and NeuroDevelopment in Adolescence (NCANDA) and the Human Connectome Project (HCP). The cross-sectional analysis of 226 no-to-low alcohol drinking NCANDA adolescents revealed relatively steeper age-dependent trajectories of myelin growth in unimodal primary motor cortex and flatter age-dependent trajectories in multimodal mid/posterior cingulate cortices. This pattern of continued myelination showed smaller gains when the same analyses were performed on 686 young adults of the HCP cohort free of neuropsychiatric diagnoses. Critically, a predicted correlation between a motor task and myelin content in motor or cingulate cortices was found in the NCANDA adolescents, supporting the functional relevance of this imaging neurometric. Furthermore, the regional trajectory slopes were confirmed by performing longitudinally consistent analysis of cortical myelin. In conclusion, coordination of myelin content and circuit complexity continues to develop throughout adolescence, contributes to performance maturation, and may represent active cortical development climaxing in young adulthood.

Appendix

Myelin content with respect to the original HCP data

The data release S900 of the Human Connectome project (HCP) provided for each subject the output of the minimal processing pipeline (Glasser et al. 2013), which included the supratentorial volume and the myelin content scores along the midthickness surface (i.e., the surface with equal distance to pial and white matter) based on the raw MRI data (i.e., 0.7 mm isotropic resolution). For each voxel inside the cortical gray matter, the pipeline computed a myelin content score by the ratio between T1-weighted and T2-weighted MRI (Glasser and Van Essen 2011). These voxel-based myelin scores were then mapped to each vertex of the midthickness surface by computing their average with respect to a cylinder centered on that vertex. The myelin score at this step is referred to as the raw myelin content. To correct for residual bias field of the raw myelin content (Glasser et al. 2013), a 14.14 mm Gaussian filter kernel was applied to the vertex-wise values, the resulting low frequency patterns of the myelin content scores were then subtracted from that of the average myelin content across the Conte69 subjects (the reference myelin content) (Glasser and Van Essen 2011), and the difference was then added to the vertex-wise values. This score is called the normalized myelin content as the myelin distribution is normalized to the reference. Finally, the midthickness surface encoding the normalized myelin content was mapped onto the 2 mm HCP template (Glasser et al. 2013), which encoded each hemisphere as a sphere defined by a 2 mm standard CIFTI grayordinate space comprising 32000 vertices (Glasser et al. 2013).

Figures 8 and 9 reveal findings that are consistent with respect to the young adults at the lower resolution (see Figs. 4 and 5). Compared with the lower resolution, myelin content computed with respect to the original resolution generally varied less across age and resulted in significant age-dependent regions that were larger in size. Table 6 reaffirms the finding that age effects (β_age) were less steep for areas 4,23c, and p24’ compared with those of adolescents (see Table 2). Furthermore, the age effects were bilaterally significant for areas 4 and p24’.

Fig. 8

Cross-sectional myelin development patterns for young adults (HCP) measured on the original resolution (0.7 mm isotropic). (A) Age-related differences are computed from the normalized myelin content. The vertex-wise (B) and regional (C) P values of those differences are FWER corrected via permutation testing (iterations: 5000). They are grayed out if non-significant (i.e., P ≥ 0.01) and have a negative sign if the relative age-effects (β_age) were negative, i.e., myelin density showed smaller gains relative to the cortex as a whole

Fig. 9

Combined cross-sectional normalized myelin development plots of adolescents (NCANDA) based on the original MRIs of NCANDA (resolution: 0.9375 × 0.9375 × 1.2 mm) and young adults (HCP) based on original MRIS of the HCP (resolution: 0.7 mm isotropic). Plots of the normalized myelin content for areas 4, 23c, and p24’ are shown for left [L] and right [R] hemispheres. In each plot, blue circles represent male and red represent female adolescents; green diamonds represent male and yellow represent female young adults; and the fits of the general liner model ± SD (standard deviation) are displayed in gray lines. The horizontal axes represent age in years and the vertical axes represent the myelin content (left for adolescents, and right for young adults). Left vertical axis is normalized myelin content for adolescents, and the right is for young adults. The offset of the vertical axis for young adults is adjusted to match with adolescents while the scale is the same. The age of the HCP samples was perturbed by a random number generator (range: [−0.5 0.5]) to improve visualization

Table 6 Cross-sectional age effects with respect to the normalized myelin content for young adults (HCP) estimated on the original resolution (0.7 mm isotropic)

Correlating average regional normalized myelin content and grooved pegboard test scores

For each region with significant age effects, an Independent Mediation Model tested whether the effect of age (independent variable) on the pegboard test scores (dependent variable) was mediated by the average regional normalized myelin content (intermediary variable). To measure the indirect effect, the regional and performance scores were first applied to a linear regression model with the confounding factors being sex, svol, site, ethnicity, and handedness (L/R). Separately for each hand, the residual scores and age were applied to the Independent Mediation Model. For each effect, the normalized β_M value was computed by dividing the β value of the effect divided with the β value of the direct (unmediated) effect (c_D) between age and the pegboard score. The P value of an effect was then inferred from Permutation testing (Boca et al. 2014) of the BRAVO toolbox V.2.0 (https://sites.google.com/site/bravotoolbox; iterations: 5000; one-sided t-test) applied to the normalized β_M value. The summary of that analysis is shown in Table 7.

Table 7 Individual Mediation tests for age, average normalized myelin content, and Grooved Pegboard Test scores for NCANDA adolescents

In the NCANDA group, the relation between age and pegboard scores was most strongly mediated by the myelin content of areas 4 [L], p24’ [R] and 4 [R] (where [L] denotes left and [R] denotes right) for both the dominant and non-dominant hand. All three right hemisphere regions showed greater indirect effects for the non-dominant than the dominant hand. Furthermore, the regions associated with significant direct effects of the myelin on the performance scores (Path b_X) agreed with those of the Pearson correlation (Table 3) with the exception of area 23c [R] on the Dominant hand scores, whose P value on the intermediate model was right below the significance threshold (P = 0.048).

Validating longitudinal myelin computation

The longitudinal approach improved the reliability of estimating normalized myelin content compared with the cross-sectional approach when applied to the 185 adolescents of the NCANDA data set who participated three times (Fig. 10). To verify this finding, the analysis was repeated with respect to travelling human phantoms.

Fig. 10

Longitudinal plots of myelin scores for adolescents (NCANDA) generated by cross-sectional and longitudinal approaches. Plots of areas 4, 23c, and p24’ are shown for the cross-sectional approach (left column) and the longitudinal approach (right column). In each plot, blue circles stand for male and red for female adolescents; visits are connected for each subject. The fits of the linear mixed effects model are drawn in gray with ± 1 and 2 SD. The horizontal axes represent age (in years) and vertical axes represent the normalized myelin content

Specifically, the difference in myelin content was measured between scan pairs, where each scan pair consisted of scans of the same subject that were either acquired at both sites (i.e., NCANDA collection site P and O) within 30 days (inter-site scan pair; average number of days between visits was 22.3 days) or acquired at the same site within a day (intra-site scan pair) according to the NCANDA scanning protocol. The data set contained a total of 11 inter-site scan pairs from 3 human phantoms (two women [age 30 and 64], one man [age 41]) and 3 intra-site scan pairs from a single human. The raw myelin content of areas 4, 23c and p24’ was computed using the cross-sectional approach (i.e., compute the scores independently for each visit) and the differences in the scores across the scans was computed. The procedure was then repeated for the normalized myelin content generated by the cross-sectional and longitudinal approach.

For bihemispheric areas 4, 23c and p24’, the longitudinal approach showed lower mean, standard deviation, and maximum differences in the normalized myelin content within the inter- and intra-site scan pairs than estimating the normalized myelin content independently for each visit (i.e., cross-sectional approach; Fig. 11 and Table 8). Visual comparison of myelin maps associated with the same human phantom confirmed the previous findings as the normalized myelin content across scans was very similar when computed by the longitudinal approach but showed inconsistencies when produced by the cross-sectional approach (see arrows in Fig. 12). The least reliable score was the raw myelin content generated by the cross-sectional approach.

Fig. 11

Boxplots for intra-site and inter-site human phantom validations. The inter-site validation shown in (A) measured differences of myelin content between scan pairs (total 11 pairs from 3 human phantoms used) collected one at each NCANDA site (sites P and O) within 30 days. The intra-site validation shown in (B) measured differences of myelin content between scan and same day rescan pairs (total 3 pairs from a human phantom used) at NCANDA sites. For (A) and (B), cross-sectional approach using the raw myelin content (red), cross-sectional approach using the normalized myelin content (green) and the longitudinal approach using the normalized myelin content (blue) are compared for bihemispheric areas 4, 23c and p24’. Each boxplot shows the median, 25th and 75th percentiles. The whiskers extend to the most extreme data points except for outliers, which are plotted as ‘+’. In (B), individual data points are shown as black dots

Table 8 Statistics of myelin content differences for inter- and Inter-site human phantom validations

Fig. 12

Single human phantom examples for intra-site and inter-site human phantom validations. For both (A) inter- and (B) intra-site validations, the normalized myelin content for the first scan is shown in the left column, the middle column shows the myelin content for the second scan estimated by cross-sectional approach, and in the right column by the longitudinal approach, respectively. Arrowheads point to regions showing visually noticeable differences. The range of color bar is the same across all figures

Effects of normalization

As shown in the Fig. 13, all trajectories of the normalized myelin content of the entire cortex and area 4 and p24, had smaller variances than those of the raw myelin content, which indicates that much of the between-subject variance in the raw myelin content was due to low frequency bias most likely caused by scanner or coil loading variability (Glasser et al. 2013). Furthermore, the plots supported the cortical myelin development hypothesis visualized in Fig. 3.

Fig. 13

Observations from adolescents (NCANDA) and young adults (HCP) data for the cortical myelin development hypothesis. For each raw (left column) and normalized (right column) myelin content, example observations for (A) primary motor cortex (bihemispheric mean of area 4), (B) mean over the entire cortex, and (C) midcingulate cortex (bihemispheric mean of area p24’) are displayed. In each plot, blue circles represent male and red represent female adolescents; green diamonds represent male and yellow represent female young adults; and the fits of the general liner model with ± SD (standard deviation) are displayed in gray lines. The horizontal axes represent age in years and the vertical axes represent the myelin content (left for adolescents, and right for young adults). The offset of the vertical axis for young adults is adjusted to match with adolescents while the scale is the same. The age of the HCP samples was perturbed by a random number generator (range: [−0.5 0.5]) to improve visualization. Between left (the raw myelin content) and right (the normalized myelin content) columns, the same axis scale was used for comparisons

Effects of cortical folding pattern removal

To study the effect of cortical folding pattern on our findings, the mean curvature along the cortical surface was regressed out from the normalized myelin content of the NCANDA cohort and the analysis for identifying cohort-specific developmental patterns was applied to residual myelin content (Fig. 14). The entire process was repeated regressing out convexity instead of curvature. In both experiments, regression had little effect on the findings with the bihemispheric areas 4, 23c and p24’, revealing again significant age effects.

Fig. 14

Cross-sectional myelin development patterns for adolescents (NCANDA) after curvature or convexity (sulc) corrected. (A) Age-related differences are computed from the normalized myelin content (i.e., the change in myelin content relative to the mean over the entire cortex) by applying a general linear model that views age as a predictor and sex, ethnicity, svol as confounding factors. Each value is displayed on the averaged (inflated) brain surface of the corresponding cohort with the boundaries of the HCP multimodal parcellation atlas (MMP 1.0) overlaid. The vertex-wise (B) and regional (C) P values of those differences are FWER corrected via permutation testing (iterations: 5000). They are grayed out if non-significant (i.e., P ≥ 0.01) and have a negative sign if the relative age-effects (β_age) were negative, i.e., myelin density showed smaller gains relative to the entire cortex