Cortical maturation and myelination in healthy toddlers and young children

1Cortical maturation is an important feature of neurodevelopment, characterized by both morphometric and microstructural changes. While alterations in cortical morphological quantities, including volume, thickness, surface area, and curvature, have been associated with neuropsychiatric, neurological, and developmental disorders (for example, ), differences in the rate of development of these quantities across childhood have also been associated with cognitive ability and intelligence. Beyond these gross morphological measures, interest has also focused on exploring the underlying cortical cyto- and myeloarchitecture. Using non-invasive imaging, in particular magnetic resonance imaging (MRI), advances have been made in investigating these attributes in vivo through quantitative imaging (voxel-wise mapping of the longitudinal or transverse relaxation times, T1 and T2, respectively) or, more recently, the semi-quantitative ratio of T1 weighted (T1w) to T2 weighted (T2w) signal intensity. These methods provide information reflective of the cortical myeloarchitecture, and have been used to identify prominent myeloarchitectural features (i.e., the Stria of Gennari), as well as detect boundaries of differentiated cortical regions.

2Though myelin is most commonly associated with the brain's white matter, the cortical gray matter also contains significant numbers of myelinated fibers. Like the maturation of the myelinated white matter, myelination of the cortex likely plays a fundamental role in neurodevelopment, and the evolution of cognitive and behavioral functions. Indeed, changes in the T1w/T2w ratio have been demonstrated across the lifespan from childhood to late adulthood, as well as being associated with cognitive performance.

3However, while these prior studies have provided new insight into the evolution of cortical myeloarchitecture throughout older childhood and adulthood, little is known about how this process advances throughout early childhood (from birth through 8 years of age) in vivo beyond prior post-mortem investigations. Further, while the T1w/T2w ratio is an easily accessible measure, it is sensitive to both non-myelin tissue changes (i.e., iron and water content) and extrinsic (e.g., imaging hardware, pulse sequence, patient positioning, and post-processing) effects. One of the most established magnetic resonance imaging approaches for investigating myelin content is myelin water fraction (MWF) imaging. Though MWF imaging has previously been used to examine white matter development throughout infancy and childhood, its application to studying cortical development remains unapproached. Therefore, we sought to examine the use of quantitative MWF imaging to investigate cortical myelination in a large cohort of 215 healthy and typically developing toddlers and young children, 1 to 6 years of age. We further compared the MWF of these results to quantitative T1 values, which has formed the basis of many prior studies of cortical myeloarchitecture.

4A total of 215 healthy and typically developing children (105 female), 363 to 2198 days (approximately 1 to 6 years) of age, corrected to a 40 week gestation, were recruited from the local surrounding community for this study. Inclusion criteria included: 1. healthy singleton birth between 37 and 42 weeks gestation; 2. uncomplicated pregnancy and delivery; 3. APGAR scores > 8; 4. No reported abnormalities on fetal ultrasound; 5. no reported neurological history in the child; 6. no reported sibling or parental psychiatric history. Full demographic information is provided in Table 1. Inclusion criteria were confirmed during phone interview prior to enrollment, and again through infant and family medical histories taken as part of the study. Maternal SES was determined using the Hollingshead four factor index. Written informed consent was obtained from each child's parents or legal guardian, and the study was performed with approval from the host institution's internal review board.

5Voxel-wise T1 and mcDESPOT MWF data were acquired on each child using age optimized DESPOT1 and mcDESPOT imaging protocols (Table 2). A constant voxel dimension of (1.8 × 1.8 × 1.8) mm3 was used for all children, with the field of view and imaging matrix adjusted depending on child age and head size. The theoretical basis for DESPOT1 and mcDESPOT has been presented in detail previously and involves the acquisition of multi-flip angle T1-weighted SPGR images (DESPOT1), as well as T1/T2-weighted bSSFP images at two differing radio-frequency (RF) phase-cycling patterns. To these data, conventional single-component or multicomponent relexometry may be performed. For multicomponent analysis, a 3-pool tissue model comprising of myelin-associated, intra and extra cellular, and CSF water pools is fit to the combined SPGR and bSSFP data using an iterative stochastic region approach. Correction for flip angle (transmit magnetic field) inhomogeneities is performed through additional acquisition of inversion-prepared (IR-)SPGR data, and the two phase-cycled bSSFP data allows correction from main magnetic field variations.

6In addition to the quantitative T1, T2 and MWF data, a higher-resolution (1.2 × 1.2 × 1.2) mm3 T1-weighted anatomical IR-SPGR image was also acquired (Table 2). As above, the imaging FOV and matrix size were adjusted to maintain the same voxel dimensions over the full head volume.

7All children were imaged on a Siemens Tim Trio 3 T scanner. Scanning was performed during natural, non-sedated sleep, or, if tolerated, while watching a favorite movie. To minimize acoustic noise, the maximum imaging gradient slew rates and peak values were reduced. Passive measures, including a sound-insulating bore liner, MiniMuff ear pads, and sound-attenuating ear protectors were also employed.

8To ensure that children included in this study were typically developing and had average cognitive ability for age, each child was assessed following MRI using the Mullen Scales of Early Learning. These scales provide age-normalized scores for fine and gross motor control, visual reception, and expressive and receptive language for children up to 5 years, 9 months of age, as well as an overall Early Learning Composite, ELC, expressed as a standard score with mean 100 and standard deviation of 15. In addition, assessments including the Modified Checklist for Autism in Toddlers (M-CHAT) and the Communication and Symbolic Behavior Scales Developmental Profile (CSBS-DP) were used to identify children with abnormal behaviors or language delay. No children included in this study screened as at-risk using these broad assessment tools or had an ELC < 85 (i.e., less than one standard deviation below the mean).

9Following image acquisition, the mcDESPOT data from each child were linear co-registered to account for subtle intra-scan motion, non-brain signal was removed, B0 and B1 field calibration maps were calculated, and voxel-wise T1 and MWF maps were calculated.

10The higher resolution anatomical image for each child was intensity normalized using the N3 algorithm and the Freesurfer analysis pipeline was used to segment the cortical ribbon and delineate the cortex into 34 regions per hemisphere. At each stage throughout the processing pipeline, images were visually inspected and, if needed, manually edited and corrected. This included inspecting data for poor skull-stripping, which required re-running this processing step, the additional use of gcut (http://freesurfer.net/fswiki/FsTutorial/SkullStripFix_freeview) and, in extreme cases, manual removal of the remaining dura and eye signal. Additional editing was required on each of the participants under 3 years of age.

11The calculated T1 and MWF maps from each participant were then linearly co-registered to the corresponding high resolution anatomical image by first aligning the high flip angle T1-weighted SPGR image acquired as part of the mcDESPOT protocol, and then applying the calculated transformation matrix to the quantitative T1 and MWF images. At each point along the cortical surface, the T1 and MWF value at the halfway depth along the surface normal was obtained and projected onto the surface. Sampling at the midpoint of the cortical surface was chosen to ensure values were from within the cortical ribbon and not the underlying adjacent white matter. This process was repeated for each of the 215 children. Age-averaged T1 and MWF surfaces were then calculated by registering all children's surfaces to a custom Freesurfer template, and averaging children between: 363–712 days (mean = 547 ± 112, n = 78); 718–1099 days (mean = 931 ± 116, n = 55); 1102–1442 days (mean = 1278 ± 98, n = 32); 1469–1814 days (mean = 1700 ± 110, n = 29); and 1836–2198 days (mean = 2022 ± 118, n = 21). The custom template was created using the make_average_subject command within Freesurfer (http://surfer.nmr.mgh.harvard.edu/fswiki/make_average_subject) and including 5 male and 5 age-matched females from each age-range (for a total of 50 participants). This was done so as to not bias the template to any particular gender or age range. Age ranges were chosen so as to end up with group mean ages of 1.5, 2.5, 3.5, 4,5, and 5.5 years.

12To visualize and investigate how cortical T1 and MWF change with age, we next superimposed each of the 68 delineated cortical regions (34 per hemisphere) for each child onto their T1 and MWF maps, and calculated the mean and standard deviation regional T1 and MWF. These mean values were then plotted against the child's age, corrected to a 40-week gestation. To these plots, we then fit continuous linear, quadratic, and logarithmic functions and compared Bayesian information criterion (BIC) values for each to determine which provided the most parsimonious fit to the data.

13To examine potential hemispheric differences or asymmetries in regional cortical T1 and MWF development, we first plotted each cortical region's cross-sectional T1 and MWF values, separated by hemisphere. Logarithmic curves were then fit to each hemisphere's data, as well as to the combined left + right hemisphere data. An F-test was then used to determine if the data justified the fitting of two hemisphere-specific curves. If a region passed this F-test step, a bootstrap with residual sampling approach was used to reconstruct the logarithmic slope and intercept distributions for each hemisphere, and these distributions were then compared using an unpaired t-test. Significance was defined as p < 0.0014 (p < 0.05 corrected for the 34 region comparisons).

14In addition, we also calculated the mean index of laterality (left − right / left + right) for each region and for each individual. These values were then plotted against the individual's age. A single sample t-test was used to determine if the laterality index differed significantly from 0 (fully symmetric) across the entire cohort age range. We examined the effect of age on laterality more specifically by using a sliding window approach. Here, the mean laterality index was calculated, and a single population t-test was performed, using a constant window of 50 subjects. That is, ordering subjects by increasing age, the first 50 children (#1–#50) were grouped and the laterality index calculated; then subjects #2–#51 were grouped and laterality index calculated, and so forth. Significance was defined as p < 0.00035 (p < 0.05 corrected for the 154 repeated measures per region).

15As prior studies of cortical thickness development across older childhood and adolescence have observed gender related differences, we sought to determine if similar differences could be seen in the cross-sectional trajectories of T1 or MWF development in our younger cohort. As with the hemisphere analysis above, we first plotted each cortical region's cross-sectional T1 and MWF values for each gender independently. Logarithmic curves were then fit to each gender's data, as well as to the combined male + female data. An F-test was then used to determine if the data justified the fitting of gender-specific curves. For regions that passed this step, a bootstrap approach reconstructed the logarithmic slope and intercept distributions for each gender, and these distributions were then compared using an unpaired t-test. Significance was defined as p < 0.0014 (p < 0.05 corrected for the 34 region comparisons).

16Figs. 1 and 2 show the average cortical surface projections of MWF and T1 for each of the five age-groups: 1.5 years (363–712 days); 2.5 years (718–1099 days); 3.5 years (1102–1442 days); 4.5 years (1469–1814 days); and 5.5 years (1836–2198 days). Examining these maps we see the expected trend of increasing MWF, and decreasing T1, across age. Examining the earliest map (1.5 years), we note a strong qualitative similarity between regions of higher and lower MWF and regions depicted by Flechsig and others as myelinating earlier or later during development. Specifically, the primary somatosensory and motor, primary and secondary visual, and auditory cortices; and superior temporal gyrus are seen to have high MWF (and low T1) bilaterally at 1.5 years of age. In contrast, areas such as the temporal pole, prefrontal and association cortex have low MWF (and high T1) at this age. These regions map onto those areas histologically shown to have rapid or protracted myelination, respectively.

17As the T1 relaxation time is related to lipid myelin and macromolecule content (in addition to water content and iron content), we examined the relationship between T1 and MWF in each cortical region. In Fig. 3, we show a subset of the 34 bilateral regions, composed of early, middle, and late myelinating regions as illustrated in Figs. 1 and 2. In all cases, T1 and MWF were found to be significantly correlated, with correlation coefficients R > 0.75 in all investigated regions. This result in cortex differs from similar analysis performed in white matter, in which the relationship between MWF and T1 was found to be weak, and varies depending on age.

18To investigate how cortical T1 and MWF change with age, a series of growth models, including linear, logarithmic, and quadratic functions were fit to the mean values, and the models were compared using the BIC. Results of this analysis (Table 3) show that both T1 and MWF across the investigated age range are best characterized by logarithmic functions. As a result of this analysis, for all following analysis and results, logarithmic functions were used to model the growth trajectories. In Fig. 4, a comparison of T1 and MWF vs. age for a selection of early and fast (precentral gyrus), moderate (inferior temporal gyrus), and late and slowly (inferior frontal gyrus) maturing brain regions is shown.

19Building on the results presented in Table 3 and Fig. 4, we investigated potential hemispheric differences and laterality in the cross-sectional developmental trajectories of each cortical region. In Figs. 5 and 6, MWF and T1 vs. age trajectories for 11 representative regions are shown, color-coded by hemisphere. An F-test performed on each region revealed that no region justified fitting each hemisphere independently for either MWF or T1. Thus, there were no differences in the rate of development between the left and right hemisphere any of the 34 regions.

20To determine if there was a significant left–right hemisphere difference in MWF or T1 magnitude, we quantified and examined the laterality index, LI (Figs. 7 and 8). Performing a single sample t-test to determine if the mean LI differed significantly from 0, we found that the occipital pole, lingual gyrus, and superior frontal gyrus were significantly left lateralized (LI > 0) with respect to MWF after corrected for the 34 individual comparisons. With respect to T1, lingual gyrus and superior frontal gyrus, were found to be significantly right lateralized (LI < 0, p < 0.05 corrected), in agreement with the relationships shown in Fig. 3, with increased T1 predicting decreased MWF.

21Finally, using a sliding window approach, we examined age-related changes in LI by testing the mean LI against 0 in blocks of 50 children. Results of this analysis (Figs. 9 and 10) show differential age-related trends depending on cortical region. Regions including the temporal pole, inferior temporal gyrus, inferior and superior frontal gyrus, precuneus, occipital pole, and lingual gyrus show significant leftwards laterality (except for inferior frontal, which had significant rightward laterality) in early childhood, prior to 2 years of age. Regions including the inferior temporal gyrus, occipital pole, and lingual gyrus exhibit significant leftwards asymmetry between 3 and 4 years of age. And regions, including precentral gyrus and lingual gyrus, show significant right or left asymmetry, respectively, in late childhood (i.e., 4.5 years of age and older).

22Finally, we examined differences in the development if MWF and T1 vs. age in children stratified by gender (Figs. 11 and 12). As above, we first used an F-test to determine if the data from each cortical region justified the fitting of each gender separately, and found no region in which this was supported. Further, an unpaired t-test was used to determine if there was a significant male–female difference in mean MWF or T1 across the age-range. No significant differences were found, even without correcting for multiple comparisons.

23Alongside white matter maturation, cortical development plays an important role in the development of cognitive and behavioral functioning. In this work, we have sought to extend our prior investigations of white matter maturation by examining cortical myelination throughout early childhood for the first time using quantitative T1 and MWF imaging. Results of this analysis reveal a pattern of cortical myelination consistent with that previously established histologically using postmortem brain specimens. That is, myelination appears earliest in primary somatosensory, motor, visual, and auditory cortices, whereas areas including the temporal pole, prefrontal and association cortices are among the last to mature. Consideration of prior reports of cortical gray matter volume growth in early childhood suggests a qualitative relationship between myelination and the rate of gray matter volume growth. Specifically, areas with early myelination (high myelin content by 1 year of age), e.g., motor and visual cortices, appear to have fast volume growth between birth and 1 year of age, but slower growth between 1 and 2 years of age. In contrast, cortical regions with later myelination (e.g., temporal and angular gyri), exhibit slow growth between birth and 1 year of age, but rapid gray matter growth between 1 and 2 years of age. As myelin constitutes a significant portion of brain volume, it is likely that the increase of cortical myelin is a primary driver of cortical gray matter volume growth. However, this hypothesized relationship may only be valid during the earliest stages of neurodevelopment. In prior investigations of cortical thinning during later childhood and adolescence, the authors proposed that observed cortical thinning may be the result of increased myelination of neural fibers in the lower cortical layers.

24To examine the relationship between cortical MWF and thickness, we performed an ad-hoc investigation comparing these values in the cortical region. A summary of results of this analysis, shown in Fig. 13, presents a more complex picture. While some regions, such as the superior frontal gyrus exhibit a significant (p < 0.05 FWE corrected) negative relationship between cortical thickness and MWF (with increased MWF predicting small cortical thickness), most regions showed a non-significant relationship. While more detailed analysis is required, particularly to examine how early MWF changes may predict later thickness changes (or vice versa), these preliminary results may suggest that cortical myelination and growth are independent processes.

25Within each of the cortical regions examined, MWF was found to increase, and T1 to decrease, logarithmically from 1 to 6 years of age. These developmental trends broadly parallel the trajectories noted in white matter across the same age range, and are predictive of later age trajectories. T1 has been shown to decrease from childhood until approximately 40 years of age, depending on brain region. Similarly, recent analysis using the T1w/T2w ratio from 8 to 83 years of age noted an ‘inverted U’ trajectory, with cortical myelin increasing until approximately 40 years of age, and then decreasing with advancing age.

26Across the investigated age range, and in each cortical region, we found that MWF and T1 were significantly correlated. While changes in T1 have previously been linked to changing myelin content in the human brain, evident from both high resolution investigations of cortical myeloarchitecture, as well as changes in the T1w signal during neurodevelopment, T1 is also influenced by a myriad of other factors, including iron and water content. Thus, the demonstrated relationship between T1 and MWF suggests that changes in myelin content may be the dominant influence on cortical T1 in early development. Further, quantitative T1 may be an acceptable surrogate marker of myelin content within the cortex during this period. However, caution is urged in assuming this relationship in atypical development or disease, in which altered tissue microstructure, edema, or lipid and protein content could also alter T1. Of note, this result differs from our own prior results in cerebral white matter, which showed no significant relationship between T1 and MWF. This seemingly contradictory result could be further evidence of the notoriously non-specific nature of T1, which is influenced by water content, iron content, fiber density, macromolecule and lipid content, etc. The maturation of white matter is substantively different than the maturation of cortical gray matter, and the development of the myelin sheath, while the most visibly striking change in MR images, may not be the dominant influence on T1.

27In addition to developmental trajectories, we also investigated hemispheric asymmetry in MWF and T1. In prior investigations of older children, as well as fetal brain, statistically significant left–right asymmetries have been noted in global and regional cortical gray matter volume. Averaged across our full cohort and age range, we found no cortical asymmetry in MWF or T1. However, investigating asymmetry as a function of age revealed differential trends, in which some regions showed significant asymmetry at 1 year of age (e.g., superior frontal gyrus, occipital pole) but not at later ages; and other regions displaying significant asymmetry only at older ages (e.g., precentral and lingual gyri). Age-related changes in asymmetry may reflect on-going changes in functional localization, and the consolidation of brain structure–function relationships. In this work, we did not specifically examine the influence of gender, or handedness on right–left asymmetry.

28Although gender-related differences have been previously noted in brain development, with boys reported to have increased cortical gray matter volume than girls by 5 years of age, we found no significant male–female difference in cortical MWF or T1 values. Though significant associations have previously been documented between handedness and regional thickness (asymmetry) measures within the parietal and frontal cortices, we did not specifically address this in our data. This stemmed simply from our inability to reliably assess handedness in the youngest participants (under 18 months of age), and the potential inaccuracy in children between 18 and 30 months, since only 60–70% of children maintain handedness between this age range. Following these children further, however, may allow us to not only investigate asymmetries associated with handedness, but from longitudinal data, determine if and when handedness may be predicted.

29While this work details the first in vivo description of cortical myelination throughout early childhood, there remain some methodological limitations. The primary of these is the relatively low spatial resolution of the acquired MWF and T1 images, (1.7 × 1.7 × 1.7) mm3, which may introduce partial volume effects that vary depending on child age and cortical region. This is particularly true in regions such as the primary somatosensory and visual cortex, which even in adults may be less than 2 mm thick. In these regions, the estimated MWF or T1 value may be under or over-estimated, depending on whether the region contains CSF of white matter voxel averaging. This effect may explain the low MWF (and high T1) values evident in the superior and medial portion of the somatosensory and motor cortices (particularly noticeable at 1.5 years of age) in Figs. 1 and 2. In an attempt to minimize this, we projected MWF and T1 values from the middle of the cortical ribbon, rather than choosing values closer to the pial surface or white/gray matter boundary. To fully remove partial volume effects, acquisitions with increased spatial resolutions (ideally less than 1 mm3 isotropic) are likely necessary. While the time duration for such acquisitions may be too long for non-sedated pediatric imaging, the use of modern multi-channel radio-frequency coils combined with modest parallel imaging acceleration could permit whole-brain 1 mm3 isotropic MWF and T1 maps in less than 10 min in applications where acoustic noise is less of a concern.

30A second concern lies in the application of Freesurfer in children under 4 years of age. Though few studies have utilized Freesurfer in this young age range, owing to the difference in T1w image contrast and potential biases in the cortical parcellation, success has been reported in toddlers as young as 12 months — similar to the youngest toddlers included in this study. We purposely excluded children under 1 year of age from this study due to the lack of sufficient gray/white matter contrast that led to inaccurate cortical segmentation. This effect was most pronounced in children less than 6 months of age, but was relatively resolved by 9 months of age. At each stage of analysis, images where checked manually to ensure accurate segmentation. While the spatial resolution of the T1w images used for segmentation was less than the ideal 1 mm3 isotropic recommended for adult studies, our resolution of (1.2 × 1.2 × 1.2) mm3 is in line with, if not higher than, other pediatric imaging studies.

31A final concern lies in the interpretation of mcDESPOT MWF values as specifically representing myelin content. There exists a known bias between mcDESPOT-derived values, and those calculated using more established multi-echo approaches. Though the reason for this bias remains unknown alternative signal sources, including magnetization transfer (MT), have been suggested as potential confounds, though recent work indicates that MT effects do not significantly affect mcDESPOT values. The large numbers of independent variables contained within the mcDESPOT signal model may also impart bias or sensitivity to initial estimates; however recent simulation work suggests mcDESPOT MWF values are reproducible across a variety of algorithm choices and experimental conditions. While histological verification of mcDESPOT has, to date, been limited to qualitative histological comparisons in the shaking pup model of dysmyelination, the specificity of mcDESPOT to myelin can also be garnered indirectly through comparison with the known histological time-course of myelination in human infants, and demyelination studies in MS. These animal model and in vivo results give confidence that if not specific to myelin, mcDESPOT provides novel information regarding white matter microstructure, and offers differing, perhaps enhanced, sensitivity to myelin changes.

32The quantitative imaging methods utilized herein provide potential advantages over T1w and T1w/T2w ratio studies. Within the MRI literature, the term “mapping” is traditionally reserved for techniques that provide a quantitative measure of a physical parameter (i.e., T1 or T2 mapping). In this respect, T1w/T2w is not a mapping technique as it is not quantitative (i.e., its value can be influenced by imaging sequence parameters, differences in head positioning during imaging, differences in signal scaling between the T1w and T2w acquisitions, and altered through the signal normalization step performed during processing); and it does not reflect an intrinsic tissue property. Without strict normalization between the T1w and T2w acquisitions, the resultant ratio may be arbitrarily scaled between individuals, limiting its utility in comparison or association studies. Finally, T1w/T2w may not be applicable to non-cortex investigations (i.e., white matter or deep gray matter), or in non-neurotypical investigations, in which non-myelin pathology may alter the T1 or T2 signal. In contrast quantitative T1 and MWF mapping have lengthy histories in MRI investigations of normal and atypical brain and CNS tissue, though this is the first investigation of cortical MWF imaging.

33In this work, we examined intracortical myelination across early childhood using quantitative T1 and MWF imaging for the first time. We found that cortical MWF increased, and T1 decreased, logarithmically between the ages of 1 and 5 years. We also found qualitative agreement between regions exhibiting early (somatosensory, motor, and visual cortices) and late (frontal and temporal cortices) myelination and prior established histological patterns. Results lay the foundation for future studies examining the relationship(s) between measures of cortical maturation and cognitive and behavioral functioning, deviations in atypical development or disease, or the relationship between cortical and white matter maturation.

34The authors declare no competing financial interests.