Graded Variation in T1w/T2w Ratio during Adolescence: Measurement, Caveats, and Implications for Development of Cortical Myelin

Abstract

Adolescence is characterized by the maturation of cortical microstructure and connectivity supporting complex cognition and behavior. Axonal myelination influences brain connectivity during development by enhancing neural signaling speed and inhibiting plasticity. However, the maturational timing of cortical myelination during human adolescence remains poorly understood. Here, we take advantage of recent advances in high-resolution cortical T1w/T2w mapping methods, including principled correction of B1+ transmit field effects, using data from the Human Connectome Project in Development (HCP-D; N = 628, ages 8-21). We characterize microstructural changes relevant to myelination by estimating age-related differences in T1w/T2w throughout the cerebral neocortex from childhood to early adulthood. We apply Bayesian spline models and clustering analysis to demonstrate graded variation in age-dependent cortical T1w/T2w differences that are correlated with the sensorimotor-association (S-A) axis of cortical organization reported by others. In sensorimotor areas, T1w/T2w ratio measures start at high levels at early ages, increase at a fast pace, and decelerate at later ages (18-21). In intermediate multimodal areas along the S-A axis, T1w/T2w starts at intermediate levels and increases linearly at an intermediate pace. In transmodal/paralimbic association areas, T1w/T2w starts at low levels and increases linearly at the slowest pace. These data provide evidence for graded variation of the T1w/T2w ratio along the S-A axis that may reflect cortical myelination changes during adolescence underlying the development of complex information processing and psychological functioning. We discuss the implications of these results as well as caveats in interpreting magnetic resonance imaging (MRI)-based estimates of myelination.SIGNIFICANCE STATEMENT Myelin is a lipid membrane that is essential to healthy brain function. Myelin wraps axons to increase neural signaling speed, enabling complex neuronal functioning underlying learning and cognition. Here, we characterize the developmental timing of myelination across the cerebral cortex during adolescence using a noninvasive proxy measure, T1w/T2w mapping. Our results provide new evidence demonstrating graded variation across the cortex in the timing of T1w/T2w changes during adolescence, with rapid T1w/T2w increases in lower-order sensory areas and gradual T1w/T2w increases in higher-order association areas. This spatial pattern of microstructural brain development closely parallels the sensorimotor-to-association axis of cortical organization and plasticity during ontogeny.

Keywords: MRI; T1w/T2w; adolescence; development; myelin.

Figure 1.

Charting T1w/T2w development during youth. A, Age/sex histogram of the current sample of 628 youth who completed structural neuroimaging as part of the HCP-D. B, T1w/T2w maps were parcellated using the HCP multimodal atlas (Glasser et al., 2016a) and averaged across participants. T1w/T2w units are arbitrary, representing relative estimates of intracortical myelin content that are comparable within a consistently acquired study. T1w/T2w units of 1.4 and 1.9 correspond to the second and 98th percentiles in this dataset. Data from the left primary motor cortex (highlighted with black border) are shown in panels C, D. C, We fit Bayesian generalized additive models with thin-plate splines to estimate different properties of age-related change in cortical T1w/T2w. Specifically, we estimated the posterior smooth function of T1w/T2w on age for each of the 360 cortical parcels (data shown for the left primary motor cortex, highlighted in panel B). The shaded area represents the 95% credible interval of the posterior smooth function. Navy blue segments of the posterior smooth function indicate the slope is credibly as steep as the maximum slope, while yellow segments of the posterior smooth function indicate the slope is credibly less steep than the maximum slope (see panel D). D, Bayesian generalized additive models were fitted to estimate the posterior derivative of the smooth function of T1w/T2w on age to characterize the rate of change, with higher values indicating a steeper slope of change per unit age and 0 representing a flat slope (i.e., no age-related change). The shaded area represents the 95% credible interval of posterior derivative estimates. The vertical line marks the age of the maximum median derivative (steepest slope). To identify windows where the rate of age-related T1w/T2w growth credibly slows down, we computed the difference between the posterior derivative of the smooth and the posterior derivative at the age of the maximum median derivative; regions of this 95% credible interval (data not shown) that did not include zero were used to mark regions of the smooth that were credibly less steep than the slope at the point of greatest maturation. Data are shown for an exemplar cortical parcel to illustrate our analytical approach.

Figure 2.

Waves of cortical T1w/T2w maturation from childhood through early adulthood. Parcellated T1w/T2w maps were averaged across participants within three age groups to illustrate the spatial patterning of age-related increases in T1w/T2w during youth. T1w/T2w units of 1.4 and 1.9 correspond to the second and 98th percentiles in this dataset. Age groups include 8–10 year olds (left; n = 145), 14–16 year olds (middle; n = 154), and 19–21 year olds (right; n = 120). Age-related increases in cortical T1w/T2w were observed across the cortical sheet, reinforcing a S-A hierarchy in cortical microstructure. The archetypal S-A axis (inset right) was adapted with permission from Sydnor et al. (2021).

Figure 3.

Effect size estimates for regional Bayesian models of T1w/T2w development. A, Partial R² values for age splines were estimated for each cortical parcel, representing the proportion of variance in T1w/T2w differences explained by age. B, Age-related increases in T1w/T2w were relatively strong in sensorimotor areas such as the right V2 (outlined and labeled in panel A), where age explained 25% of variance in T1w/T2w. C, By contrast, age-related increases in T1w/T2w were relatively weak in heteromodal and paralimbic association areas such as the left anterior cingulate cortex (area p32; outlined and labeled in panel A), where age explained 2.8% of variance in T1w/T2w. D, The variance in T1w/T2w explained by age (R²) was correlated with the cortical area's position along the archetypal S-A axis of cortical organization. In panel D, the blue data point corresponds to the exemplar sensorimotor parcel (V2) and the yellow data point corresponds to the exemplar association parcel (ACC) highlighted in panels A–C. ACC = anterior cingulate cortex; S-A = sensorimotor-association. P_spin is the permutation-based p-value calculated from a conservative parcel-based spatial permutation (“spin”) test. Data are shown for exemplar sensorimotor and association parcels to illustrate differences in T1w/T2w development. Models were estimated independently for all 180 cortical parcels in each hemisphere (360 in total).

Figure 4.

Annualized rate of change in T1w/T2w during youth. A, Annualized rate of change (AROC) in T1w/T2w was estimated for each cortical parcel. B, The AROC was relatively high in sensorimotor areas such as the right primary motor cortex (area 4; highlighted in panel A, right). C, In contrast, the AROC in T1w/T2w was relatively low in prefrontal and paralimbic association areas such as the left medial prefrontal cortex (area 9m; highlighted in panel A, left). D, The AROC in T1w/T2w was correlated with the cortical parcel's position along the archetypal S-A axis of cortical organization. The blue data point corresponds to the exemplar sensorimotor parcel (primary motor cortex) and the yellow data point corresponds to the exemplar association parcel (medial PFC) highlighted in panels A–C. Shaded areas in the plots of panels B, C represent the 95% credible interval of the posterior smooth function on age. Segments of the age curve where the slope is credibly less steep than the maximum slope (indicating a credibly reduced rate of change) are highlighted in yellow. PFC = prefrontal cortex; S-A = sensorimotor-association. P_spin is the permutation-based p-value calculated from a conservative parcel-based spatial permutation (“spin”) test. Data are shown for exemplar sensorimotor and association parcels to illustrate differences in T1w/T2w development. Models were estimated independently for all 180 cortical parcels in each hemisphere (360 in total).

Figure 5.

Nonlinearity of age-related changes in T1w/T2w during youth. A, The nonlinearity of age-related changes in T1w/T2w was estimated for each cortical parcel using the mean absolute posterior second derivative, where higher values indicate more nonlinear growth. B, Sensorimotor areas such as the left V2 (highlighted in panel A, left) had relatively nonlinear T1w/T2w growth, with the rate of change decreasing credibly by 17.5 years old (indicated by yellow segment). C, Cortical parcels in heteromodal and paralimbic association areas such as the right anterior ventral insula (highlighted in panel A, right) had relatively linear growth in T1w/T2w (constant slope). D, The nonlinearity of age-related changes in T1w/T2w was correlated with the cortical parcel's position along the archetypal S-A axis of cortical organization (Sydnor et al., 2021). Sensorimotor areas exhibited significantly higher nonlinearity in age-related increases in T1w/T2w compared with areas of association cortex. In panel D, the blue data point corresponds to the exemplar sensorimotor parcel (V2) and the yellow data point corresponds to the exemplar association parcel (anterior insula) highlighted in panels A–C. Shaded areas in the plots of panels B, C represent the 95% credible interval of the posterior smooth. Segments of the posterior smooth where the slope is credibly less steep than the maximum slope (indicating a credibly reduced rate of change) are highlighted in yellow. P_spin is the permutation-based p-value calculated from a conservative parcel-based spatial permutation (“spin”) test. Data are shown for exemplar sensorimotor and association parcels to illustrate differences in T1w/T2w development. Models were estimated independently for all 180 cortical parcels in each hemisphere (360 in total).

Figure 6.

Age of peak T1w/T2w growth during youth. A, The age of peak growth (i.e., steepest slope) in T1w/T2w myelin was estimated for each cortical parcel as the median posterior age where the slope is at its maximum. B, Sensorimotor areas such as left V1 (highlighted in panel A) had relatively early age of peak T1w/T2w myelin growth (median age = 9.6 years). C, Association areas such as the right prefrontal cortex (area 8c; highlighted in panel A) had relatively later age of peak T1w/T2w myelin growth (median age = 15.0 years). D, The age of peak T1w/T2w myelin growth was correlated with the cortical parcel's position along the archetypal S-A axis of cortical organization. Sensorimotor areas exhibited a significantly earlier age of peak growth compared with areas of association cortex. In panel D, the blue data point corresponds to the exemplar sensorimotor parcel (V1) and the yellow data point corresponds to the exemplar association parcel (PFC) highlighted in panels A–C. Segments of the age curve where the slope is credibly less steep than the maximum slope (indicating a credibly reduced rate of change) are highlighted in yellow. Insets at the top of panels B, C show the posterior density distribution of the age of maximum slope for exemplar areas. The shaded blue area represents the 95% credible interval of the posterior distribution; black vertical lines mark the median of the posterior distribution (values projected on the cortical surface in panel A). PFC = prefrontal cortex. P_spin is the permutation-based p-value calculated from a conservative parcel-based spatial permutation (“spin”) test. Data are shown for exemplar cortical parcels to illustrate differences in T1w/T2w development. Models were estimated independently for all 180 cortical parcels in each hemisphere (360 in total).

Figure 7.

Graded variation across data-driven clusters of cortical T1w/T2w development. A functional latent mixture model was used to identify groups of cortical parcels exhibiting similar age-related patterns of T1w/T2w growth during youth. A, The most parsimonious model identified three latent clusters or groups of cortical parcels with graded differences in the rate of age-related changes in T1w/T2w. The inset shows the penalized log-likelihood for each model solution from k = 2 to k = 7 clusters. The best fitting three-cluster solution is highlighted in red. B, Functional latent clusters closely aligned with the S-A axis of cortical organization. Sensorimotor areas, primarily in cluster 1 (magenta), had the highest rate of cortical T1w/T2w growth. We observed graded decreases in the rate of cortical T1w/T2w growth in cluster 2 (orange), which included heteromodal association parcels in frontoparietal and temporal cortex, and further in cluster 3 (yellow), which included paralimbic association parcels in the medial prefrontal and insular cortices. C, Regional cluster assignments based on T1w/T2w development reflect the S-A axis. L-L = log-likelihood.