The maturation of the white and gray matter myeloarchitecture is a critical process of neurodevelopment52 that underpins cognitive development as evidenced by the functional impairments associated with dys- and de-myelinating disorders such as leukodystrophies and multiple sclerosis. While comparative studies of myelination spatiotemporal patterns overlap with the emergence and evolution of cognitive and behavioral skills, direct evidence of the relationships between these processes in healthy human infants and children has been scarce. While prior MRI studies using diffusion tensor and white matter volumetric imaging have characterized developmental differences across different white matter regions and pathways, and provided important evidence supporting myelin-function relationships8,18,22,53,54, direct associations between individual changes in white and gray matter myeloarchitecture and individual improvements in cognitive function or performance remain lacking.
Extending the structural covariance framework to delineate brain areas based on their spatiotemporal myelination pattern, this work provides new insight into the rate, sequence, and functional specificity of developing brain regions and their extension to networks. Underlying more domain-specific regions, we identified a series of early and rapidly maturing regions that were broadly associated with overall development. These core brain regions included central brain systems and functional relay centers such as the cerebellum, brainstem, thalamus, hippocampus, amygdala, basal ganglia, corpus callosum, and internal capsule, that have also been identified as central network nodes within the human functional and structural connectomes55–58. Given this central role, and their preserved presence across species, it is not surprising they contribute to a wide range of cognitive and behavioral processes, including general cognitive ability and intelligence, executive functioning, memory, motor control, coordination, language, amongst other cognitive, behavioral, and socioemotional functions59–63.
Building on top of this core network, we also identified more functionally specific regions. With respect to fine motor function, we find contributions somatosensory and motor cortices, the pons, cerebellum, corpus callosum, caudate nucleus, putamen, thalamus, orbital frontal cortex, and occipital and visual cortex. These regions have well known connections with each other, for example cortical-ponto-cerebellar and cerebello-thalamic-cortical pathways linking the cerebellum, pons, and thalamus to visual and motor areas 64,65; striatal-cortical connections including the caudate nucleus, putamen, and motor areas; and orbital frontal connections to secondary motor areas 66,67. Given these networks, these areas have previously been independently and collectivity associated with motor control and sensory processing.
While many of these fine motor areas overlap with those associated with gross motor function, we also find more specific and unique associations with frontal lobe regions, including supplementary motor and premotor cortices, and dorsolateral and ventrolateral prefrontal cortices, and supramarginal gyrus. Though frontal lobe regions are amongst the last to myelinate in the human brain12, following many of the regions associated with fine motor function, this may more directly relate to these regions involvement in walking, balance, coordination, and spatial processing68, skills that do not evolve until 9–12 months of age and following touching, grasping, and manipulation of small objects, which begin within the first few months of life 69 (Fig. 4).
The importance of visual processing to fine motor skill70 likely underlies the overlap and close developmental trajectories of brain regions we found contribute to both functions (Figs. 3 and 4). The main difference between the two being with greater involvement of occipital white matter and visual cortex areas for visual processing, and anatomical specificity to lateral thalamic nuclei.
While the cerebellum is traditionally associated with balance, visuospatial processing, and motor control, its role in almost all skills and functions has been increasingly characterized via advanced neuroimaging studies71. This range of function is also observed across our results, with aspects of the cerebellum associated with motor, vision, and language skills. In addition to cerebellum, we also note that myelination of temporal lobe regions (predominately left hemisphere), prefrontal cortex, superior and inferior longitudinal fasciculus, arcuate fasciculus, and striata regions is associated with expressive and receptive language skills. Many of these regions are expected and align with prior functional neuroimaging studies of language development in children72. Examining the relative rates of myelination within these regions, we find that those more associated with expressive language are slower to development, approximately in time with gross motor regions, whilst receptive language regions are amongst the earliest to develop. This aligns with both developmental milestones, as well as functional neuroimaging studies showing the ability to react, orient to, understand, and discriminate sounds and voices is in place within the first weeks of life73,74, whilst the development of vocalizations and words is far more protracted throughout the first 6 to 18 months.
As a developmental process, myelination is driven in part by neural activity75 and, thus, may be altered or delayed by differences in environmental stimuli, nutrition, or other aspects that might modulate neuronal activity76. Here we examined the additional aspects of gender, birth weight, and maternal education (as a metric of family socioeconomic status) on measures of cognitive development. While we noted relationships between these environmental measures on cognitive performance, a direct relationship was not observed between them and myelination rate, and mediation analysis (not shown) did not support and indirect association between gender, birth weight or maternal education on cognitive measures via myelination rate. This result is surprising with respect to the demonstrated influence of socioeconomic status on child development77. However, it may be that this influence is exerted on more specific cognitive measures, such as executive functions or academic skills78.
Differences in the rate of myelination, or its absence, are hypothesized to contribute to a range of intellectual, psychiatric, and neurological disorders, including language delay, ASD, schizophrenia, ADHD, and MS. The work here points the potential for identifying the sequence and timing of potential deficits that may contribute to these and other disorders, as well as potential corrective and mitigation approaches. For example, Fig. 4 provides support for the importance of early interventional strategies within the first 2 years of life given the rapid maturation and myelination of the core network areas over this period and, therefore, potentially the period when it is most sensitive to change. While functionally specific regions develop more slowly, latter interventions targeting these regions may constitute a ‘changing of the window coverings’ rather than a more complete renovation. Indeed, we can examine this more directly by modeling the relative contributions of the core and domain-specific region myelination to each functional score (Table 6). We see that while inclusion of both the core and domain-specific network myelination measures independently are significant predictors of overall functional score, and both improve the model fit compared to those without the myelin data, the inclusion of domain-specific region information does not significantly impact the model fit beyond the core network alone.
In this work we have examined correlations over the full profile of development, examining overall rate of myelination vs. rate of cognitive function change. We may also expect that the rate of development and, therefore, the relationship between regional myelination and functional performance, may vary with age, with potentially different regions contributing significantly over differing developmental periods. While we have not examined this directly here, novel and advanced statistical methods, such as functional concurrent regression79, could be used to investigate these time dynamic changes, which may further inform on optimal interventional periods.
While this work has focused on myelin content, additional aspects related to myelin and white matter microstructure may also be of importance. For example, measures of neuronal density, orientation dispersion and axon diameter, such as that provided by non-tensor diffusion models (e.g., AxCaliber 80 and NODDI81) have also been related to functional processing and connectivity82. These measures may also be combined with MWF estimates to derive the myelin thickness or g-ratio83, which is related to axonal conduction and network efficiency84 and also likely linked to processing speed and cognitive ability.