Cortical and white matter myelination proceed in concert during early infancy

The infant brain undergoes rapid myelination that is critical for healthy brain function. This development has been characterized for gray and white matter independently, but the link between gray and white matter myelination remains unexplored. To close this knowledge gap, we evaluated two complementary myelin-sensitive imaging metrics: Large-scale (N = 273) T1w/T2w and quantitative (N = 21) R1 data. Automated software was employed to identify 26 white matter bundles and map their cortical terminations, before evaluating T1w/T2w and R1 development shortly after birth. Here we show that for both metrics mean values as well as developmental slopes are correlated across tissues. The synchrony of brain T1w/T2w is impacted by postmenstrual age and prematurity, whereas inter-individual differences in this synchrony predict motor outcomes at 17 − 25 months of age. As T1w/T2w and R1 are associated with myelin content, our results reveal an intricate relationship between gray and white matter myelination.

The human brain shows rapid development during early life. At birth, total brain volume is about 35% of adult size, growing to nearly 80% by the age of two years old 1 (for review, see ref. 2 ). This remarkable expansion is accompanied by changes in tissue microstructure and organization, including growth of myelin in both gray and white matter 3 , 4 , 5 , 6 (for reviews, see refs. 7 , 8 , 9 ). Myelination is a critical process in brain development that involves wrapping neuronal axons in a fatty sheath, thereby enabling rapid and synchronized neural communication. This mechanism is essential for brain plasticity and learning, and disruptions in myelin development have been linked to various developmental and cognitive disorders 10 , 11 (for reviews, see refs. 12 , 13 , 14 , 15 ).

In the white matter, myelin insulates axons for rapid saltatory conduction, reduces the energetic cost of signaling by lowering membrane capacitance, supplies metabolic substrates, and preserves axonal structural integrity 16 , 17 , 18 (for reviews, see refs. 13 , 19 ). Although the majority of myelin is located in the white matter 20 , 21 , a considerable number of myelinated axons can also be found in the gray matter 22 , 23 (for review, see refs. 24 , 25 , 26 , 27 ). The precise role of cortical myelin remains debated, with several functions proposed, including (i) metabolic support 28 , (ii) network synchronization 29 , 30 (for review, see ref. 31 ), (iii) fine-tuning of conduction speed 32 , and (iv) growth inhibition to prevent aberrant axonal sprouting and synapse formation 33 , 34 , 35 , 36 (for review, see ref. 37 ).

In both white and gray matter, myelination during early infancy follows a complex spatiotemporal trajectory and, critically, similar mechanisms have been proposed to describe these developmental trajectories across tissues: (1) Spatial gradients: In white matter, several spatial gradients of myelination have been proposed including decreasing developmental rates from: central-to-peripheral 38 , 39 , 40 , 41 , posterior-to-anterior 5 , 38 , and superior-to-inferior 5 , 42 , 43 , 44 , 45 white matter locations. Similarly, in gray matter, studies revealed a systematic decrease in myelination toward parietal, temporal, and prefrontal cortices 4 , 22 , 46 , 47 , 48 , 49 , 50 . (2) Functional systems: Sensory and motor white matter pathways myelinate earlier than pathways associated with higher-level brain functions 38 , 41 (for review, see ref. 51 ). Similarly, sensory and motor gray matter regions myelinate early and heavyly, while higher-order areas such as the prefrontal and association cortices, myelinate later and remain more lightly myelinated even in adulthood 3 , 4 , 22 , 52 , 53 , 54 , 55 , 56 , 57 (for review, see ref. 58 ). 3) Myelin content at birth: Recent neuroimaging work suggests that white matter that is less myelinated at birth develops more rapidly postnatally 3 , 59 (for review, see ref. 60 ). Similarly, primary sensory and motor cortices, which exhibit higher myelin levels at term-equivalent age, myelinate more slowly than higher-order association and visual areas during early infancy 4 , 57 , 61 (for review, see ref. 62 ). Importantly, these mechanisms are likely not mutually-exclusive but rather shape myelin development in concert. Moreover, as there is overlap in the mechanisms that control the spatiotemporal trajectory of myelination in gray and white matter, there may be an as of yet unexplored link between gray and white matter myelination during infancy.

To test this link, it is critical to follow and compare the trajectories of myelin growth of gray and white matter in the living human brain. Recent advances in quantitative MRI 3 , 43 , 63 , 64 , 65 (qMRI, for review, see ref. 66 ), have made it feasible to quantify myelin levels and to compare them between individuals and across developmental timepoints. The validity of using qMRI measures such as the myelin-water-fraction (MWF) or the longitudinal relaxation rate (R1) to study myelination has been confirmed by comparisons to post-mortem histological data 67 , 68 , 69 . However, qMRI measures often require long acquisition times, and such measures are hence typically not included in large-scale open data. As a practical alternative, Glasser and Van Essen 22 proposed utilizing the ratio of T1-weighted (T1w) to T2-weighted (T2w) images to approximate myelin levels. Although the validity of T1w/T2w as a marker of myelin remains debated 70 , 71 , 72 , 73 , studies comparing T1w/T2w to R1 have validated its usability in adult cortex 74 and infant white matter 59 . As the T1w/T2w ratio is easily computed from commonly acquired images, it is particularly useful in large-scale datasets and/or early life samples where acquisition times are particularly restricted.

In this work, we combine these approaches to gain both precision and robustness by evaluating a small-scale ( N = 21) R1 dataset collected locally at Stanford University (Stanford VPNL Baby Project (SVBP) data) as well as a large-scale ( N = 273) T1w/T2w dataset shared openly by the Developing Human Connectome Project (dHCP). In both datasets, we test for a link between gray and white matter myelination in early infancy by relating R1 or T1w/T2w of white matter bundles to R1 or T1w/T2w measured at their corresponding cortical targets.

We find that white matter and cortical myelination is tightly coupled: for both R1 and T1w/T2w values within each bundle are positively correlated with those measured in the associated cortical regions. Similarly, for both R1 and T1w/T2w, the rate of change during early infancy is coupled between white and gray matter. We also observe large inter-individual variability in the strengths of white and gray matter coupling, which we explored further in the large-scale dHCP data set. We find that this inter-individual variablity is i) related to postmenstrual age, ii) weaker in preterm than full-term infants, and iii) correlated with motor performance later in life. By highlighting the tight coupling of gray and white matter myelination, our findings provide a comprehensive understanding of early life brain myelin growth and suggest that white and gray matter myelination should be considered conjointly.

We used two complementary myelin-sensitivity imaging metrics, T1w/T2w and R1, to assess if myelin development of white matter bundles and their corresponding gray matter targets is linked in early infancy. For this, we leveraged two infant datasets collected shortly after birth: (i) large-scale dHCP data, containing dMRI and T1w/T2w measurements from 273 infants, including both preterm and full-term infants (gestational age (GA) at birth: mean ± SD: 38.05 ± 3.67 weeks; time between birth and scan: mean ± SD: 1.47 ± 2.02 weeks) and (ii) locally-collected quantitative data (SVBP data), containing dMRI and R1 measurements from 21 infants (GA at birth: mean ± SD: 39.09 ± 1.63 weeks; time between birth and scan: mean ± SD: 4.30 ± 1.35). In both datasets, we used pyBabyAFQ to identify 26 white matter bundles in individual infants’ brains (Supplementary Figs. 1 and 2 ), mapped their endpoints to cortex (Fig. 1 ), and then assessed the development of myelin-sensitive imaging metrics across tissues.

Bundles were identified with pyBabyAFQ and are shown in the native brain space of one example full-term infant from the dHCP data scanned at 40 weeks postmenstrual age. The yellow dots represent the cortical endpoints of the respective bundles. For bilateral bundles only the left hemisphere is shown.

To investigate the relationship between white and gray matter myelin levels in early infancy, we correlated the myelin-sensitive imaging metrics of white matter bundles with their gray matter targets. In the dHCP data, our analyses of T1w/T2w revealed a positive correlation (Fig. 2a , r ² = 0.55, r (24) = 0.74, P value = 1.45e −5 , 95% CI [0.50, 0.88]), indicating that if a white matter bundle has high T1w/T2w, its gray matter targets also have high T1w/T2w. Similarly, in the SVBP data, our analyses of R1 showed a positive correlation (Fig. 2b , r ²=0.32, r (24) = 0.57, P value = 0.002, 95% CI [0.23, 0.78]) indicating that if a white matter bundle has high R1, its gray matter targets also have high R1. For both metrics, these observed correlations for the true bundle, target pairings were above the 95% confidence interval (CI) of chance-level correlations (95% CI of chance-level: r ² = 0.14 for T1w/T2w and r ² = 0.12 for R1) obtained by 1000 iterations of correlating shuffled bundle, target pairings, omitting the true pairs (Supplementary Fig. S3 ).

Each dot is a bundle, lines indicate least-squares regression fit, and shaded regions indicate 95% confidence interval. a Mean correlation with T1w/T2w values averaged across subjects ( N = 273; r ² = 0.55, P value = 1.45e −5 , two-sided). b Mean correlation with R1 values averaged across subjects ( N = 21; r ² = 0.32, P value = 0.002, two-sided). c Example individual subjects exhibiting high (left; PMA at scan: 41.43 weeks; r ² = 0.68, P value = 2.28e −7 , two-sided) and low (right; PMA at scan: 31.43 weeks; r ² = 0.017, P value = 0.52, two-sided) correlation between T1w/T2w of white and gray matter. d Example individual subjects exhibiting high (left; PMA at scan: 42.5 weeks; r ² = 0.51, P value = 4.21e −5 , two-sided) and low (right; PMA at scan: 40.4 weeks; r ² = 0.03, P value = 0.44, two-sided) correlation between R1 of white and gray matter. Source data are provided as a Source Data file. WM white matter, GM gray matter, AF Arcuate Fasciculus, ATR Anterior Thalamic Radiation, CC Cingulum Cingulate, CS Cortico-Spinal Tract, FcMa Forceps Major, FcMi Forceps Minor, IFOF Inferior Frontal Occipital Fasciculus, ILF Inferior Longitudinal Fasciculus, MLF Middle Longitudinal Fasciculus, OR Optic Radiation, SLF Superior Longitudinal Fasciculus, UNC Uncinate Fasciculus, VOF Ventral Occipital Fasciculus, pAF Posterior Arcuate Fasciculus, L left, R right, PMA postmenstrual age.

Interestingly, R1 and T1w/T2w were correlated with each other in each tissue across the datasets (Supplementary Fig. S4 ; Gray matter: r ² = 0.66, r (24) = 0.81, P value = 5.32e −7 , 95% CI [0.62, 0.91]; White matter: r ² = 0.85, r (24) = 0.92, P value = 1.88e −11 , 95% CI [0.83, 0.97]) and revealed similar patterns among different bundles: The corticospinal tract (CS) along with its cortical targets, showed particularly high T1w/T2w and R1. In contrast, the posterior arcuate fasciculus (pAF) and its cortical targets exhibited comparatively low values for both metrics, with other bundles showing intermediate values of T1w/T2w and R1. In the anterior thalamic radiation (ATR) the link between gray and white matter was less pronounced than in the other bundles, with the cortical targets of the ATR having lower values than expected based on their white matter values. Interestingly, our analyses of T1w/T2w and R1 also revealed similar values across hemispheres in gray and white matter of bilateral bundles as indicated by the close proximity of data points representing corresponding bundles in each hemisphere. A detailed overview of differences in T1w/T2w and R1 values across bundles for each tissue is presented in Supplementary Fig. S4 .

In addition to evaluating the relationship of the mean T1w/T2w and R1 values of gray and white matter at the group-level, we also evaluated this relationship within each individual participant. These individual subject analyses corroborated the group-level findings, as white and gray matter values were correlated within most individuals for both T1w/T2w (correlation significant in 265 out of 273 subjects (97%), mean r ² across subjects: 0.42, mean r (24) across subjects: 0.64, mean P value across subjects: P = 0.008) and R1 (correlation significant in 17 out of 21 subjects (81%), mean r ² across subjects: 0.26, mean r(24) across subjects: 0.49, mean P value across subjects: P = 0.04). Nonetheless, there were large inter-individual differences in the degree of white and gray matter coupling in both T1w/T2w (range of r ²: 0.01-0.68, standard deviation of r ²: 0.13) and R1 (range of r ²: 0.03-0.51, standard deviation of r ²: 0.12). Figure 2 c, d show two example individuals for high and low T1w/T2w and R1 correlations across tissues. Similar to the group level data, in both T1w/T2w and R1, the individual subject analyses showed particularly high gray and white matter values for the CS and comparatively low values for the pAF, with other bundles falling in between. Further, again similar to the group-level data, the cortical targets of the ATR showed lower T1w/T2w and R1 values than expected based on their white matter T1w/T2w and R1. Finally, in both metrics, bilateral bundles showed similar values in gray and white matter across hemispheres in the individual subject analyses as well.

We further investigated the relationship between myelin-sensitive imaging metrics of white matter bundles and their cortical targets by evaluating the slopes of T1w/T2w and R1 growth during early infancy (Fig. 3 ). The slopes represent the rate at which each metric changes with infants’ age at scan in the respective age ranges covered by each dataset. Figure 3a, b show example slopes for the corticospinal tract (CS, top) and the cingulum cingulate (CC, bottom); visualizations of all bundles are provided in Supplementary Figs. S6 and S7 .

a , b Two example bundles (corticospinal tract (CS, top) and cingulum cingulate (CC, bottom)) illustrate T1w/T2w ( N = 273) ( a ) and R1 ( N = 21) ( b ) slopes in gray and white matter (for all bundles see Supplementary Figs. S6 and S7 ). Gray circles indicate gray matter values in each subject, colored circles represent white matter values in each subject, lines indicate least-squares regression fits, the steepness of the lines indicate the slopes (rates of change relative to the infant’s postmenstrual age at the time of measurement in weeks), the two hemispheres are presented in different hues, and shaded regions indicate 95% confidence interval. c , d Relationship between the slopes of T1w/T2w ( c ) and R1 ( d ) development along white matter (WM) bundles and at their gray matter (GM) targets (T1w/T2w: N = 273; r ² = 0.49, P value = 6.28e −5 , two-sided; R1: N = 21; r ² = 0.45, P value = 0.0002, two-sided). Each circle indicates a bundle, the two hemispheres are presented in different hues. Shaded regions indicate 95% confidence interval. Source data are provided as a Source Data file. AF Arcuate Fasciculus, ATR Anterior Thalamic Radiation, CC Cingulum Cingulate, CS Cortico-Spinal Tract, FcMa Forceps Major, FcMi Forceps Minor, IFOF Inferior Frontal Occipital Fasciculus, ILF Inferior Longitudinal Fasciculus, MLF Middle Longitudinal Fasciculus, OR Optic Radiation, SLF Superior Longitudinal Fasciculus, UNC Uncinate Fasciculus, VOF Ventral Occipital Fasciculus, pAF Posterior Arcuate Fasciculus, L left, R right.

In the dHCP data, in which scan age ranged from 29.29 to 44.7 weeks PMA (total period: 15.41 weeks), all bundles showed a positive correlation between infants’ age at measurement and T1w/T2w (Supplementary Table S1 and Supplementary Fig. S6 ) in both the white matter bundles (all r ²>0.40, all r (271) > 0.63, all P < 5.44e −32 ) and at their cortical targets (all r ²>0.38, all r (271) > 0.62, all P < 7.72e −30 ). In the SVBP data, in which scan age ranged from 39.1 to 48.1 weeks PMA (total period: 9 weeks), R1 also increased with age, although the effect was not uniformly significant across bundles (Supplementary Table S2 and Supplementary Fig. S7 ) in this more restricted age range (white matter: all r ²s between 3e⁻⁵ and 0.55, all r (19)s between 0.005 and 0.74, all ps between 0.0001 and 0.94, 19 out of 26 bundles were significant; gray matter: all r ²s between 0.04 and 0.48, all r (19)s between 0.20 and 0.69, all ps between 0.0005 and 0.37, 20 out of 26 bundles were significant). For both T1w/T2w and R1 slopes were steeper in the white matter than the gray matter (T1w/T2w: t = 3.82, P value = 0.0008; R1: t = 3.41, P = 0.002). Interestingly, in a few bundles these differences in developmental slopes between tissues lead to a crossing over where initially lower white matter values began to exceed gray matter values during development. For example, in the corticospinal tract (CS), we observed a cross-over in T1w/T2w at around 37 weeks PMA and in the superior longitudinal fasciculus (SLF) we observed a cross-over in R1 at around 43 weeks PMA (Supplementary Figs. S6 and S7 ).

Similar to the mean values reported above, the slopes of T1w/T2w and R1 change in gray and white matter also varied across bundles. For T1w/T2w, the CS, for instance, stood out with the highest rate of change in both the white matter and at its cortical target. In contrast, the ATR, pAF, and the FcMi exhibit comparatively low rates of change, with the other bundles falling in between. For R1, the CS did not stand out as clearly, but the FcMi and ATR again exhibit comparatively low rates of change, with the other bundles falling in between. Further, again similar to the mean values, bilateral bundles showed similar gray and white matter slopes between the left and right hemisphere in both T1w/T2w and R1. As the slopes of T1w/T2w and R1 are derived from datasets with different scan age ranges, they are not directly comparable. However, even when we matched the scan age ranges (by including only those subjects from the dHCP that are within the same range as the SVBP, N = 183), we find that T1w/T2w and R1 slopes are not significantly correlated with each other (Supplementary Fig. S8 ) in either the gray matter ( r ² = 0.07, r (24) = 0.26, P value = 0.20, 95% CI [0.14, 0.59]) or the white matter ( r ² = 0.06, r(24) = 0.24, P value = 0.24, 95% CI [0.16, 0.57]).

To determine if there is a relationship between the slope of white and gray matter myelin development, we correlated T1w/T2w as well as R1 slopes of both tissues across bundles (Fig. 3c, d ). Our analyses revealed a positive correlation between the slopes of white matter bundles and their cortical targets in both metrics (T1w/T2w: r ² = 0.49, r (24) = 0.70, P = 7.57e −5 , 95% CI [0.42, 0.85]; R1: r ² = 0.45, r (24) = 0.67, P value = 0.0002, 95% CI [0.39, 0.84]), suggesting coupled white and gray matter development during early infancy. For both metrics, these observed correlations for the true bundle, target pairings were above the 95% confidence intervals (CI) of chance-level correlations (95% CI of chance-level: r ² = 0.15 for T1w/T2w, r ² = 0.14 for R1) obtained by 1000 iterations of correlating shuffled bundle, target pairings, omitting the true pairs (Supplementary Fig. S9 ).

As described earlier, we found considerable inter-individual differences in the strength of T1w/T2w coupling across tissues and as such we aimed to further characterize these inter-individual differences in the large-scale dHCP data. We find that the strength of the T1w/T2w coupling is dependent on gestational age at birth ( r ² = 0.18, r (271) = 0.42, P value = 5.15e −13 , 95% CI [0.32, 0.51], Fig. 4a ) and on postmenstrual age at scan ( r ² = 0.18, r (271) = 0.42, P value = 2.23e −13 , 95% CI [0.32, 0.51], Fig. 4b ), but not on sex ( t = −1.88, P value = 0.06, Fig. 4c ). We also tested whether these inter-individual differences in T1w/T2w coupling correlate with later behavioral outcomes. To do so, we analyzed a subset of infants from the dHCP data ( N = 215) that also completed the Bayley-III questionnaire, and hence allowed us to link T1w/T2w measurements taken at birth to behavioral outcomes assessed between 17 and 25 months of age. For the three Bayley-III subscales (motor, language, and cognition) we investigated the relationship between each individual’s age-standardized test scores and their T1w/T2w coupling across tissues. For cognition and language, we found no significant link between the T1w/T2w coupling across tissues and performance (all r ²<0.007, all P s>0.11, Fig. 4d, e ), however, we found that the T1w/T2w coupling across tissues correlates with later-life motor performance ( r ² = 0.02, r (203) = 0.17, P = 0.015, 95% CI [0.03, 0.30], significant with a Bonferroni corrected threshold, Fig. 4f ). We also tested if T1w/T2w measured in gray matter or in white matter alone is linked to later-life behavioral outcomes (Supplementary Fig. S10 ), but found no significant relationship between T1w/T2w in either tissue and any of the three subscales (all r ²<0.007, all P s>0.23). Inclusion of demographic covariates (birth age, scan age, or sex) did not significantly improve the linear model relating T1w/T2w coupling and motor performance (scan age: F (1,212) = 2.58, P = 0.11, birth age: F (1,212) = 1.09, P = 0.30, sex: F (1,212) = 1.18, P = 0.28).

a, c Upper row relates T1w/T2w coupling across tissues to gestational age at birth ( a N = 273; r ² = 0.18, P value = 5.15e −13 ), postmenstrual age at scan ( b N = 273; r ² = 0.18, P value = 2.23e −13 ), and sex ( c female N = 117, male N = 156; t = −1.88, P value = 0.06). d, f Bottom row relates the correlation of T1w/T2w in gray and white matter to the cognition ( d N = 209; r ² = 0.01, P value = 0.14), language ( e N = 210; r ² = 0.01, P value = 0.11), and motor ( f N = 205; r ² = 0.03, P value = 0.02) subscale of the Bayley-III. We found a relationship between the T1w/T2w coupling and motor skills, but not the other subscales. Each dot is a subject, red lines indicate linear regression lines, shaded regions indicate 95% confidence interval, boxes in ( c ) indicate the interquartile range (IQR, 25th, 75th percentile), the line inside each box denotes the median and the whiskers extend to 1.5×IQR. Source data are provided as a Source Data file.

To examine if the coupling of T1w/T2w across white and gray matter is impacted by prematurity, we examined this relationship in a small longitudinal sub-sample from the dHCP that focuses on infants born preterm. This sample contained dMRI and T1w/T2w data from 26 preterm infants (GA at birth: mean ± SD: 32.04 ± 2.97 weeks) scanned once shortly after birth (PMA at first scan: mean ± SD: 34.33 ± 1.76 weeks) and once after they reached term-equivalent age (PMA at second scan: mean ± SD: 40.70 ± 1.08 weeks), as well as a group of full-term infants with scan ages and sex matched to the preterm infants’ second scan (PMA at scan: mean ± SD: 40.69 ± 1.07 weeks). T1w/T2w of white matter bundles correlated with T1w/T2w of their cortical targets in all three groups: We observed positive correlations in the preterm infants scanned shortly after birth (Fig. 5a ; r ² = 0.43, r (24) = 0.65, P value = 0.0002, 95% CI [0.36, 0.83]), in the same preterm infants scanned at term-equivalent age (Fig. 5b ; r ² = 0.54, r (24) = 0.73, P value = 1.98e −5 , 95% CI [0.48, 0.87]), and in full-term infants matched on age and sex to the preterm infants’ second scan (Fig. 5c ; r ² = 0.58, r (24) = 0.75, P value = 9.03e −6 , 95% CI [0.52, 0.88]). To determine whether the observed differences in correlation coefficients between groups were statistically significant, we employed a nonparametric bootstrap resampling procedure (1000 iterations) to estimate the sampling distribution of each pairwise difference. The resulting 95% confidence intervals (CIs) were then compared against 0 to assess whether group differences were significant. This analysis revealed weaker correlation between T1w/T2w of white matter bundles and their cortical targets in the preterm infants at their first scan, compared to their full-term peers (Fig. 5d , mean difference in r ² = −0.10, 0 fell outside 95% CI [−0.101, −0.097]). Similarly, we also found weaker correlations within the preterm infants at their first compared to their second scan (Fig. 5e , mean difference in r ² = −0.08, 0 fell outside 95% CI [−0.081, −0.072]). When we compared preterm infants at their second scan to their age-matched full-term peers, we found a higher correlation in the full-term infants (Fig. 5f , mean difference in r ² = −0.02, 0 fell outside 95% CI [−0.020, −0.016]).

a, c Correlation of T1w/T2w across tissues in three groups: a infants born preterm scanned shortly after birth ( N = 26; r ² = 0.43, P value = 0.0002, two-sided); b infants born preterm scanned at term-equivalent age (i.e. at ≥37 weeks postmenstrual age, N = 26; r ² = 0.54, P = 1.98e −5 , two-sided); c age-matched infants born full-term ( N = 26; r ² = 0.58; P = 9.03e −6 , two-sided). Each dot is a bundle, gray lines indicate least-squares regression fit, shaded regions indicate 95% confidence interval. d, f Bootstrap distributions of correlation differences between groups: d preterm infants scanned shortly after birth vs. full-term infants; e preterm infants scanned shortly after birth vs. preterm infants scanned at term-equivalent age; f preterm infants scanned at term-equivalent age vs. full-term infants. The vertical red and green lines indicate the upper and lower bound of the 95% confidence intervals (CIs), blue line indicates mean. Zero fell outside the 95% CIs in all comparisons. Source data are provided as a Source Data file. AF Arcuate Fasciculus, ATR Anterior Thalamic Radiation, CC Cingulum Cingulate, CS Cortico-Spinal Tract, FcMa Forceps Major, FcMi Forceps Minor, IFOF Inferior Frontal Occipital Fasciculus, ILF Inferior Longitudinal Fasciculus, MLF Middle Longitudinal Fasciculus, OR Optic Radiation, SLF Superior Longitudinal Fasciculus, UNC Uncinate Fasciculus, VOF Ventral Occipital Fasciculus, pAF Posterior Arcuate Fasciculus, L left, R right.

Further comparisons between the three groups revealed a few additional interesting patterns. First, bundles show an overall similar order of development across the groups: the corticospinal tract (CS) exhibits highest T1w/T2w in both white and gray matter in each group, while the posterior arcuate fasciculus (pAF) shows consistently low T1w/T2w, with the other bundles falling in between. Second, across the groups, the cortical targets of the anterior thalamic radiation (ATR) have lower T1w/T2w and the cortical targets of the ventral occipital fasciculus (VOF) have higher T1w/T2w than expected based on their white matter T1w/T2w. Third, there are differences in the degree to which bilateral bundles show similar T1w/T2w values in gray matter across the age group: bilateral bundles are more similar at the preterm infants second scan compared to their first scan (group difference in delta of T1w/T2w across hemispheres: t = 3.54, P = 0.0002) and compared to full-term infants ( t = −2.79, P = 0.01), while the preterm infants first scan and the full-term infants were not significantly different ( t = −0.65, P = 0.52).

In this study, we investigated the association between myelin-sensitive imaging metrics of white matter bundles and their corresponding cortical targets in two complementary infant datasets: A large-scale dataset shared openly by the dHCP that contained dMRI and T1w/T2w data and a quantitative dataset collected locally at Stanford University that contained dMRI and R1 data. Our analyses revealed five key findings: (i) both T1w/T2w and R1 of white matter bundles are positively correlated with T1w/T2w and R1 at their respective cortical targets; (ii) in both T1w/T2w and R1 the rate of change (slope) is correlated between white and gray matter; (iii) there is substantial inter-individual variability in the degree of T1w/T2w and R1 coupling across tissues; (iv) inter-individual variability in T1w/T2w coupling is related to birth and scan age and correlates with motor performance at 17-25 months of age; (v) the T1w/T2w coupling is lower in preterm compared to full-term born infants, even at term-equivalent age. Overall, these results suggest that white and gray matter myelination proceed in concert during early infancy.

Here we find that T1w/T2w and R1 development is tightly linked between white matter bundles and their corresponding cortical targets. We hypothesize that this observation might be accounted for by the theory of activity-dependent myelination 25 , 75 : According to this theory, active neurons send trophic signals (e.g., glutamatergic release, ATP signaling) along their axons that promote oligodendrocyte differentiation both within the cortex and along white matter projections, such that increased neuronal firing can increase myelination in both compartments 76 , 77 , 78 , 79 (for review, see ref. 80 ). Alternative hypotheses for the observed coupling in myelination include: (1) Coordinated genetic regulation: As similar genetic programs direct the proliferation, migration, and differentiation of oligodendrocyte precursor cells (OPCs) in both cortex and white matter 81 , 82 , these genetic influences might drive the coordinated myelination of white and gray matter. (2) Common metabolic and microenvironmental factors: Both cortical and white matter myelination are energy-demanding processes that depend on a supply of lipids and other metabolic substrates 18 . As such, the local availability of these resources, coupled with support from glial cells like astrocytes could concurrently influence myelination in both tissues 83 , 84 .

Our findings relate to prior work linking other gray and white matter properties in different developmental populations: For example, during childhood and adolescence (ages 5-23), the mean diffusivity of white matter bundles was shown to be closely linked to mean diffusivity of their cortical targets 85 . White matter properties are also linked to cortical thickness and cortical expansion from childhood to adulthood 86 , 87 , 88 , 89 , 90 . Further, in adults, white matter connectivity, in combination with cortical microstructure, can be used to predict the functional organization of cortex 91 and training-induced changes in functional properties of cortex are also associated with white matter changes 92 , 93 . Studies linking gray and white matter properties during infancy are limited to date, however research showed that white matter connectivity is more strongly related to a cortical region’s cytoarchitecture than its functional properties across infancy, childhood, and adulthood 94 , 95 . While not specific to changes in myelin, these studies across developmental populations underscore the importance of conjointly evaluating white and gray matter tissue properties.

By simultaneously evaluating changes in myelin-sensitive imaging metrics in both white and gray matter, we had the opportunity to directly compare growth rates across tissues. In both T1w/T2w and R1 slopes were steeper in the white matter than the gray matter during early infancy. These findings can be corroborated when integrating previous research using R1, as R1 is directly comparable across studies. Across studies, R1 growth rates reported for white matter (approximate R1 increase of 0.026 s −1 /month) 5 , exceed R1 growth rates reported for cortex (approximate R1 increase of 0.015 s −1 /month) 4 during the first 6 months of life, suggesting that white matter myelin content increases faster during early infancy than cortical myelin content. In the present study, we expanded upon this prior work by directly comparing developmental growth rates across both tissues within the same individuals. Interestingly, we observed a crossing over of myelin-sensitive imaging metrics in several bundles, where values were higher in cortex in younger participants but higher in the white matter in older participants. This observation likely relates to the commonly observed inversion of imaging contrasts between gray and white matter during infancy 66 and provides an interesting direction for future research that could explore inter-individual differences in the timing of this crossing over.

In the current study, we combined large-scale assessments of T1w/T2w with assessments of R1 derived from qMRI and found coupled gray and white matter development in both metrics. While there is no direct histological evidence confirming the link between T1w/T2w and myelin content, R1 has been shown to be directly linked to myelin in post-mortem brain tissue 67 . In fact, most studies aiming to validate T1w/T2w as a marker for myelin have used quantitative MRI metrics such as R1 as a gold-standard for comparison. These studies found that while T1w/T2w shows strong test-retest reliability 70 and aligns well with quantitative measures in adult cortex 74 , 96 and infant white matter 59 , it does not align well with quantitative myelin measures in adult white matter 71 , 72 . This inconsistency is underscored by a recent histological analysis of the corpus callosum 73 , which raised further doubts about the validity of using T1w/T2w as a marker of adult white matter myelination. Our findings add an additional perspective by showing that the mean values of T1w/T2w and R1 are correlated in early infancy, while their developmental slopes are not, which may point to a strong but non-linear relationship between the two metrics in early infancy (also see ref. 96 ). Nonetheless, further research is necessary to fully evaluate the validity of using T1w/T2w as a marker for myelin across different age groups and brain tissues. This validation is especially crucial considering that large-scale initiatives, such as the dHCP, often acquire T1w and T2w images but not quantitative myelin metrics - inducing a trade-off between sample size and measurement validation.

Interestingly, we find that inter-individual differences in T1w/T2w coupling across tissues correlates with later life motor outcomes. Multiple studies have linked increased myelination of either white or gray matter to motor performance in humans 80 , 97 and other species such as mice 77 , 81 , 98 , 99 , 100 , 101 . Our findings complement this literature by showing that it may not only be the absolute amount of myelin that impacts motor performance, but also the degree of synchrony in myelination between white matter bundles and their corresponding gray matter targets. However, as the observed impact of T1w/T2w coupling on motor performance is comparatively small and the coupling itself is impacted by age, future work is needed to assess the implications of this effect. These future studies could relate T1w/T2w coupling to behavioral assessments conducted concurrently with MRI measurements, as here we are predicting motor performance at 17-25 months of age from neuroimaging data collected at birth, which is inherently ambitious. In addition to motor performance, we also explored the link between inter-individual differences in T1w/T2w coupling and later life language and cognitive skills, but found no significant relationship. One possible explanation for our findings being restricted to motor performance may be the time at which the behavioral assessments took place (17-25 months of age), as early infancy is characterized by a particularly rapid attainment of key motor milestones. In fact, most fine and gross motor skills, such as grasping, reaching, and postural control, are typically acquired between 6 and 12 months of age, with movement beginning to serve adaptive functions from 3 to 4 months post-term 102 . Future studies could explore a link between T1w/T2w coupling and language and cognitive abilities later during development.

Our data suggest that the synchrony of gray and white matter myelination may be dependent on brain maturity. First, we find that both gestational age at birth and postmenstrual age at scan are positively related to inter-individual differences in the degree of T1w/T2w coupling across tissues. Further, in a longitudinal sub-sample, we find lower correlation of T1w/T2w across gray and white matter in preterm compared to full-term infants at birth. Interestingly, even at term-equivalent age, when the age-matched preterm and full-term born infants have had the same amount of time to mature, they still show different degrees of T1w/T2w coupling with coupling being stronger in the full-term born infants. These results are consistent with prior work indicating that preterm birth can alter both the speed and degree of brain myelination, potentially leading to developmental differences when compared to full-term infants 59 , 103 , 104 (for review, see ref. 105 ).

An important limitation of this study is its reliance on mean values across entire white matter bundles, both cortical target regions, and over all cortical endpoints. By reducing each bundle to a single weighted average value, we are not able to detect any spatial heterogeneity that may exist along its length. Likewise, computing a single weighted average T1w/T2w or R1 value across all cortical endpoints fails to capture localized differences at specific cortical subregions, thereby precluding the detection of fine-grained patterns in cortical myeloarchitecture. This approach may also obscure microstructural variations across different cortical layers, as prior research has shown that T1w/T2w values can vary across cortical depths 22 , 47 , 52 . Our analyses thus provide only a birds-eye view of the coupled myelin development across tissues. Future studies employing more granular, regionally, or laminar-resolved T1w/T2w or R1 analyses could provide a more nuanced understanding of this relationship.

Overall, our study leveraged two complementary myelin-sensitive imaging metrics to reveal a tight coupling of myelin development across white and gray matter during early infancy. Recognizing that gray and white matter myelination unfolds in concert compels a rethinking of how we evaluate early life brain development both in research and in clinical settings. Ultimately, acknowledging and further testing the gray, white matter myelination synchrony will refine our understanding of early life brain development and the mechanisms that drive the rapid establishment of myelin across the infant brain.

Study protocols for diffusion-weighted and R1 infant data collected locally at Stanford University were approved by the Stanford University Internal Review Board on Human Subjects Research. Diffusion and anatomical data collection as part of the dHCP was approved by the UK National Research Ethics Authority (14/LO/1169). Data analysis was approved by the Ethics Committee of the Department of Psychology of the Philipps University of Marburg.

In addition, we also used anatomical, diffusion-weighted and R1 infant data collected locally at Stanford University. We refer to this data as the Stanford VPNL Baby Project (SVBP). Details on data collection parameters are described in refs. 4 , 5 , 95 . The data encompassed 27 sessions, acquired from 27 individuals, which included all necessary data for the current analyses (diffusion MRI, R1 maps, and anatomical data). After quality assurance (described below) the data contained 21 infants scanned shortly after birth (Race and ethnicity: 3 Asian, 3 Hispanic, 5 Multiracial, and 10 White participants). Among these participants, 7 were female. The gestational age at birth ranged from 36.5 to 42 weeks (mean ± SD: 39.09 ± 1.63 weeks). Imaging was conducted between 39.10 and 48.10 weeks post-conceptional age (mean ± SD: 43.39 ± 2.53 weeks), corresponding to an interval of 2.3 to 7.1 weeks after birth (mean ± SD: 4.30 ± 1.35 weeks). Participants were paid US$25 per hour for participation and parents of infant participants provided written informed consent prior to their scan session.

Anatomical images were optimized for gray, white matter contrast using a contrast-to-noise driven parameter optimization and nominal neonatal T1/T2 values. T2-weighted and inversion-recovery T1-weighted multi-slice fast spin-echo images were acquired in sagittal and axial stacks (in-plane resolution 0.8 × 0.8 mm², slice thickness 1.6 mm with 0.8 mm overlap; T1-weighted sagittal overlap 0.74 mm; T2-weighted TR/TE = 12000/156 ms; T1-weighted TR/TI/TE = 4795/1740/8.7 ms; SENSE ≈ 2-2.6). 3D MPRAGE images were additionally acquired with 0.8 mm isotropic resolution (TR/TI/TE = 11/1400/4.6 ms; SENSE 1.2 RL). Multi-slice FSE data were motion-corrected and fused across orientations into a single 3D volume per contrast using slice‑to‑volume reconstruction. For more details on anatomical sequence optimization and reconstruction, see the original protocol description 106 . Briefly, diffusion MRI was optimized for the developing brain using a multi-shell acquisition with four b-value shells (b = 0, 400, 1000, 2600 s/mm²; 20, 64, 88, and 128 directions, respectively), with directions distributed uniformly and acquired using four phase‑encoding directions (AP, PA, RL, LR). The acquisition ordering was designed to account for infant motion and gradient duty cycle and allowed pausing and restarting the scan with overlap in diffusion weightings if the infant woke up. Data were acquired with an EPI sequence (multiband factor 4, SENSE 1.2, partial Fourier 0.86, in‑plane resolution 1.5 × 1.5 mm², slice thickness 3 mm with 1.5 mm overlap, TE = 90 ms, TR = 3800 ms) and reconstructed using a dedicated SENSE‑based algorithm. For more information on diffusion protocol design and reconstruction, see the corresponding methodological work 106 .

All MRI data were acquired on a GE Discovery MR750 3 T scanner equipped with a 32‑channel head coil.

Multishell diffusion-weighted data were collected with 9, 30, and 64 directions at b-values of 0, 700, and 2000 s/mm² (TE = 75.7 ms; TR = 2,800 ms; voxel size = 2 × 2 × 2 mm³; 60 slices; FOV = 20 cm; in-plane/through-plane acceleration=1/3; total scan time≈5 min). An additional short scan with reverse phase encoding and six b = 0 images was acquired to enable susceptibility distortion correction. For detailed information, see ref. 95 .

Structural data included T1-weighted images acquired with GE’s BRAVO sequence (TE = 2.9 ms; TR = 6.9 ms; voxel size = 0.8 × 0.8 × 0.8 mm³; FOV = 20.5 cm; scan time≈3 min) and T2-weighted images obtained using GE’s CUBE sequence (TE = 124 ms; TR = 3650 ms; voxel size = 0.8 × 0.8 × 0.8 mm³; FOV = 20.5 cm; scan time≈4 min), which provide superior tissue contrast in young infants. For more information see ref. 95 .

Quantitative R1 maps were derived from an inversion‑recovery EPI (IR‑EPI) sequence combined with multiple SPGR acquisitions. Acquisition parameters followed the protocol described in ref. 5 .

The anatomical and diffusion-weighted data used in this project underwent several preprocessing steps as part of the dHCP preprocessing pipelines 107 , 108 , 109 and here we used the preprocessed data. As accurate alignment of dMRI and anatomical data was critical for this project, we applied an additional rigid-body alignment using mrregister (from MRrtrix3 110 ) and visually inspected all images for alignment quality. We also used the T1w/T2w maps provided by the dHCP 111 . Information on the generation of these maps are provided directly by the dHCP 22 ; very briefly, the T1w image was first rigidly registered to the T2w image and the T1w/T2w was then estimated from the original T2w image and the transformed T1w image (prior to bias correction). Following this, the T1/T2 ratio was projected onto the midthickness surface, using volume-to-surface mapping.