key: cord-0284821-7n8i5a46
authors: Grotheer, Mareike; Rosenke, Mona; Wu, Hua; Kular, Holly; Querdasi, Francesca R.; Natu, Vaidehi; Yeatman, Jason D.; Grill-Spector, Kalanit
title: White matter myelination during early infancy is explained by spatial gradients and myelin content at birth
date: 2021-08-23
journal: bioRxiv
DOI: 10.1101/2021.03.29.437583
sha: 3aa3531dacbe3fbbfd6b66dfebe2d00e158276f1
doc_id: 284821
cord_uid: 7n8i5a46

Development of myelin, a fatty sheath that insulates nerve fibers, is critical for brain function. Myelination during infancy has been studied with histology, but postmortem data cannot evaluate the longitudinal trajectory of white matter development. Here, we obtained longitudinal diffusion MRI and quantitative MRI measures of R1 in 0, 3 and 6 months-old human infants, and (ii) developed an automated method to identify white matter bundles and quantify their properties in each infant’s brain. We find that R1 increases from newborns to 6-months-olds in all bundles. R1 development is nonuniform: there is faster development in white matter that is less mature in newborns, and along inferior-to-superior as well as anterior-to-posterior spatial gradients. As R1 is linearly related to myelin fraction in white matter bundles, these findings open new avenues to elucidate typical and atypical white matter myelination in early infancy, which has important implications for early identification of neurodevelopmental disorders.

(http://vpnl.stanford.edu/babyAFQ/bb11_mri3_interactive.html) and 6 months of age 124 (http://vpnl.stanford.edu/babyAFQ/bb11_mri6_interactive.html) show the 3D structure of bundles 125 in an example infant. Data 3-5) . a. All bundles of an individual baby. Each row is a bundle, each column is a timepoint; left: newborn, middle: 3 months, right: 6 months. b. Comparison of AFQ and babyAFQ performances in identifying each bundle in newborns relative to manually defined (gold-standard) bundles. Error bars indicate standard error across participants. The dice coefficient quantifies the overlap between the automatically and manually defined bundles, revealing significantly higher performance for babyAFQ than AFQ. Abbreviations: ATR: anterior thalamic radiation, CS: cortico-spinal tract, pAF: posterior arcuate fasciculus, VOF: vertical occipital fasciculus, FcMa: forceps major; FcMi: forceps minor, AF: arcuate fasciculus, UCI: uncinate fasciculus, SLF: superior longitudinal fasciculus, CC: cingulum cingulate, ILF: inferior longitudinal fasciculus, IFOF: inferior frontal occipital fasciculus, MLF: middle longitudinal fasciculus. 139 We first measured the development of mean R1 in each bundle during the first 6 months of 140 life. Measurements of mean R1 of the 24 bundles identified by babyAFQ at 0, 3, and 6 months reveal 141 a substantial increase in R1 from 0 to 6 months of age (Fig. 2a) . Mean R1 across bundles±SE [range]: 142 0 months: 0.46s -1 ±0.007s -1 [0.42-0.55s -1 ], 3 months: 0.52s -1 ±0.008s -1 [0.46-0.63s -1 ], 6 months: 0.62s -1 143 ±0.009s -1 [0.54-0.73 s -1 ]. This is a profound change, as mean R1 increases on average by ~17% (0.16s -144 1 ) within just 6 months. We modeled mean R1 development in each bundle using linear mixed models 145 (LMMs) with age as predictor and a random intercept (estimated R1 at birth) for each participant. 146 Overall, LMMs explained ~90% of the R1 variance across development (adjusted Rs 2 >0.87, 147 ps<0.0001). As R1 in white matte is linearly related to myelin fraction, these data are consistent with 148 the idea that white matter bundles myelinate during early infancy.

To summarize the LMM results we plotted each bundle's mean R1 measured in newborns 151 (Fig 2b) and as its rate of development (Fig 2c) with 3 notable findings: (i) Mean R1 measured in 152 newborns varies across bundles. At birth, projection bundles (CST and ATR) have the highest R1 and 153 the forceps minor (FMi) and inferior frontal occipital fasciculus (IFOF) have the lowest R1 (Fig 2b) . 154 (ii) The rate of R1 development during infancy varies between bundles. Across these 24 bundles, the 155 Forceps Major (FcMa) has the fastest rate of R1 development, while the Uncinate (UCI) and the 156 anterior thalamic radiation (ATR) have the slowest rate of R1 development between 0 to 6 months.

(iii) Relating the bundles' rate of R1 development to their R1 measured in newborns reveals no systematic relationship between mean R1 in newborns and rate of mean R1 development (Fig 2c) . 159 Indeed, there is no significant correlation between R1 in newborns and R1 slopes across bundles 160 (R 2 =0.003, p=0.81). For example, both the cortical spinal tract (CST) and the forceps major (FcMa) 161 have fast R1 development (steep slope) during early infancy, yet they have vastly different mean R1 in 162 newborns. Together, these analyses suggest that mean R1 in newborns does not seem to explain mean 163 R1 development rate during early infancy.

To relate our findings to previous work that evaluated diffusion metrics, we also measured the 165 development of mean diffusivity (MD) across bundles. Myelination of the white matter is expected to 166 result in decreases in MD. Consistent with this, we found that mean MD systematically decreases in 167 all 24 white matter bundles during the first 6 months of life (Supplementary Fig. 5a ). Like R1, mean 168 MD in newborns and the rate of mean MD development varied across bundles ( Supplementary Fig.   169 5b,c). Interestingly, while mean MD and R1 in newborns are correlated (R 2 =0.76, p<0.0001), the rates 170 of MD and R1 development during early infancy are not correlated (R 2 =0.08, p=0.17). That is, the 171 longitudinal developmental patterns observed using MD are different from those observed with R1.

For example, the uncinate (UCI) has slow R1 development (shallow slope) but rapid MD development 173 (steep slope). In contrast to R1, we find a negative correlation between the rate of MD development White matter bundles are large structures that span substantial distances across the brain and have 181 variable white matter properties along their length 32,34,44 . Thus, mean measurements across the entire bundle may not be representative and may even obscure differential development patterns along the 183 length of the bundles. Thus, we next evaluated R1 development along the length of 24 bundles. 184 We examined the development of R1 along each bundle using babyAFQ with two main 185 observations: (i) At each timepoint, R1 exhibits spatial variations along the length of these 24 bundles 186 (Fig 3) , with the range of variations differing across bundles. For example, the cortico-spinal tract 187 (CS , Fig 3a) , exhibits substantial variations in R1 along its length, whereas the vertical occipital 188 fasciculus (VOF, Fig 3d) shows only modest variations. (ii) Consistent with the analyses of mean R1, 189 along the length of each of these 24 bundles, R1 systematically increases from newborns (Fig 3-dotted   190 line), to 3-month-olds (Fig 3-dashed line) , to 6-months-olds (Fig 3-solid line) . 191 To quantify R1 development along white matter bundles during the first 6 months of life, we 192 used LMMs applied independently at 100 equidistant locations (nodes) along each bundle (LMM relating R1 to age; one LMM per node and bundle; random intercepts for individuals). The LMM 194 slopes estimate the rate of R1 development at each node (Fig 4-dashed lines) , and we compared the 195 slope to the measured R1 in newborns at each node (Fig 4-solid lines) . Results reveals two main 196 findings: (i) LMM slopes are positive throughout, indicating that R1 increases from birth to 6 months 197 of age. (ii) In all bundles, there is a nonuniform rate of R1 development along the length of the bundle.

For example, the posterior ends of the inferior longitudinal fasciculus (ILF) and middle longitudinal 199 fasciculus (MLF) show a larger change in R1 (more positive slope) than their anterior ends (Fig 4) . As 200 R1 is linearly related to myelin fraction, these data suggest that myelination occurs at different rates 201 along the length of these 24 bundles. relative to the measured R1 in newborns (Fig 4-solid) , we could also begin to assess the three Figure 7) . This analysis provides additional evidence that development of MD in 218 white matter bundles differs from R1 during early infancy.

219 220 Spatial gradients and R1 at birth together explain R1 development 221 The prior visualizations of R1 along white matter bundles suggest that both R1 at birth and 222 the spatial location in the brain may contribute to the rate of R1 development during early infancy. To 223 gain a global understanding of the spatial nature of R1 development across the white matter of the 224 human brain, next, we visualized R1 measured in newborns and the rate of R1 development of white 225 matter bundles in the 3D brain space of newborns (plotting every 10 th node, Fig 5) , rather than along 226 each individual bundle (as in Figs 3,4) . These 3D visualizations yield the following observations: (i) 227 R1 in newborns varies spatially across the brain with overall highest values in central white matter and 228 lowest values in frontal white matter (Fig 5b) , and (ii) the rate of R1 development varies spatially 229 across the brain with faster increases in occipital and parietal white matter (yellow in Fig 5c) and 230 slower development in the temporal and frontal white matter (black in Fig 5c) . Overall, these 231 visualizations suggest that both R1 at birth and spatial gradients across the brain appear to contribute 232 to the rate of R1 development during early infancy. Thus, we next quantitatively tested the significance of each of these two factors separately, and then tested the viability of a model incorporating both 234 factors. We applied a similar approach to MD (Supplementary Fig. 8) . 235 First, we tested if the rate of R1 development is related to R1 measured in newborns (LMM 236 relating R1 slope to R1 measured in newborns at every 10 th node, with a random intercept per bundle).

The speed up hypothesis predicts a significant negative relationship but the starts-first/finishes-first 238 hypothesis predicts a significant positive relationship. LMM results reveal a significant negative 239 relationship between the rate of R1 development and R1 measured in newborns across the white 240 matter (=-0.003, p<0.0001), that accounts for 40% of the variance in R1 slopes (R 2 =0.40). That is, 241 nodes that have higher R1 in newborns develop more slowly than nodes that have lower R1 in 242 newborns, which is consistent with the speed-up hypothesis.

Second, we tested the spatial gradient hypothesis and evaluated if the rate of R1 development 244 at each node is related to its spatial location in the brain (LMM relating R1 slope at every 10 th node to 245 the nodes average coordinates in newborns |x|, y, z, and their interactions |x|*y, |x|*z, and z*y; 246 random intercept per bundle). Results show that there is a significant relationship between the rate of 247 Figure 5 . Spatial gradients and R1 at birth together explain R1 development. In all panels each point is a node. In all plots only every 10 th node of a bundle is plotted to ensure spatial independence of tested nodes. The coordinate of each node is the average |x|,y,z coordinate across newborns. As all data was acpc-ed, the 0,0,0 coordinate is the anterior commissure; |x|-axis is medial to lateral; y-axis is posterior to anterior; z-axis is inferior to superior. The axes are identical across panels. (a) 3D spatial layout of the 24 bundles in the average newborn brain volume. Nodes are color coded by bundle (see legend); approximate lobe annotations are included to clarify the spatial layout. Crucially, as quantitative R1 measures are comparable across MRI scanners of the same field 296 strength 9,15,26 , we can compare our R1 measurements in infants to those of other populations. For 297 example, we find that R1 in white matter bundles of full-term newborns ranges between 0.42-0.55[s -1 ], which is higher than R1 in the white matter of preterm newborns, which ranges between 0.29-299 0.36[s -1 ] 48 . This observation suggests that at birth there is some level of myelin in all 24 bundles 300 investigated here, contrasting with classic histological studies which reported myelin only in a handful 301 of white matter bundles in newborns (e.g., the cortical-spinal tract) 2-5 . As these classic studies used 302 qualitative visual inspection of myelin stains, rather than quantitative metrics, our data underscore the 303 utility of quantitative R1 measurements. Our measurements also reveal that R1 in 6-months-olds' 304 bundles ranges between 0.54-0.73[s -1 ], which is lower than the average R1 measured in adults' bundles, The finding that less mature white matter at birth myelinates faster during infancy is important 313 for several reasons. First, our data not only provides empirical evidence against the classic view that 314 white matter develops in a strictly hierarchically manner from early sensory to higher-level cognitive 315 regions 2,3 , but also offers new insights regarding the nature of white matter development in infancy.

As myelination is experience-dependent 10-13 , and we find that the rate of myelination after birth is 317 negatively related to its initial (birth) level, one conjecture from our data is that the postnatal 318 environment and experiences may produce a flurry of myelination during the first 6 months of life, 319 overtaking earlier prenatal gradients. Second, as previous data has shown a link between cognitive 320 development, processing speed and myelin development during infancy and early childhood 51,52 , we 321 further hypothesize that the observed negative relationship between myelination at birth and the rate 322 of myelin development is functionally relevant. For example, one consequence of this developmental trajectory is that it generates a more uniform distribution of myelin across the white matter, which 324 may allow more coordinated and efficient communication across the human brain.

The rate of R1 development also varies spatially, with faster development occurring 326 prominently in the inferior-to-superior and anterior-to-posterior directions. As a result of these spatial 327 gradients, white matter that falls within the parietal and occipital lobes develops faster than central, In conclusion, we find that during early infancy myelin content at birth and spatial gradients 16 full-term and healthy infants (7 female) were recruited to participate in this study. Three 360 infants provided no usable data because they could not stay asleep once the MRI sequences started 361 and hence, we report data from 13 infants (6 female) across three timepoints: newborn (N=9; age: 8-362 37 days), 3 months (N=10; age: 79-106 days), and 6 months (N=10; age: 167-195 days). Two 363 participants were re-invited to complete scans for their 6-months session that could not be completed 364 during the first try. Both rescans were performed within 7 days and participants were still within age 365 range for the 6-months timepoint. The participant population was racially and ethnically diverse 366 reflecting the population of the Bay Area, including two Hispanic, nine Caucasian, two Asian, and 367 three multiracial participants. Six out of the 13 infants participated in MRI in all three timepoints (0, 368 3, 6 months). Due to the Covid-19 pandemic and restricted research guidelines, data acquisition was 369 halted. Consequently, the remaining infants participated in either 1 or 2 sessions.

Expectant mothers and their infants in our study were recruited from the San Francisco Bay 371 Area using social media platforms. We performed a two-step screening process for expectant mothers.

First, mothers were screened over the phone for eligibility based on exclusionary criteria designed to 373 recruit a sample of typically developing infants and second, eligible expectant mothers were screened once again after giving birth. Exclusionary criteria for expectant mothers were as follows: recreational matter. Each connectome consisted of 2 million streamlines. MRtrix3 software was also used to fit 501 tensor kurtosis models from which we estimated mean diffusivity (MD) maps for each individual. 

BabyAFQ uses anatomical ROIs as waypoints for each bundle. That is, a given tract is 513 considered a candidate for belonging to a bundle only if it passes through all waypoints associated 514 with that bundle. The waypoint ROIs were adjusted from those commonly used in adults 36 to better 515 match the head size and white matter organization of infants (Supplementary Fig 3) . Specifically, 516 we: (i) spatially restricted some of the waypoint ROIs to account for the more compact infant brain,

(ii) introduced a third waypoint for curvy bundles, (iii) as the VOF was the only bundle that used 518 cortical-surface waypoint ROIs, we generated new volumetric waypoint ROIs for the VOF 519 (Supplementary Figure 4) , so that all waypoints for all bundles are volumetric, and (iv) added new 520 waypoint ROIs for identifying the MLF, as the MLF was not included in prior AFQ versions.

Critically, these waypoints were defined in a neonate infant template brain (UNC Neonatal template 35 ) 522 and are transformed from this template space to each individual infant's brain space before bundle 523 delineation. The use of an infant template brain is critical as commonly used adult templates, such as 524 the MNI brain, are substantially larger and difficult to align to infants' brains. In cases where a given 525 tract is a candidate for multiple bundles, a probabilistic atlas, which is also transformed from the infant template space to the individual infant brain space, is used to determine which bundle is the better 527 match for the tract. Bundles are then cleaned by removing tracts that exceed a gaussian distance of 4 528 standard deviations from the core of the bundle. Critically, babyAFQ was designed to seamlessly 529 integrate with AFQ, so that additional tools for plotting, tract profile evaluation and statistical analysis 530 can be applied after bundle delineation.

BabyAFQ quality assurance 533 To evaluate the quality of the bundle delineation by babyAFQ, we compared the automatically 534 identified bundles to manually delineated "gold-standard" bundles. Manual bundle delineation was 535 performed for the newborns in DSI Studio (http://dsi-studio.labsolver.org/) by 2 anatomical experts 536 who were blind to the results of babyAFQ. As a benchmark, we also delineated bundles with AFQ, 537 which was developed using adult data, and compared these bundles to the "gold-standard" bundles. is the intersection between these two sets of voxels (Fig 1b) . We compared dice coefficients between 542 babyAFQ and AFQ in two repeated measures analyses of variance (rmANOVAs Fig 4) . 548 In addition to the quantitative evaluation, we examined all bundles delineated using babyAFQ 549 and AFQ qualitatively at all time-points (Supplementary Fig 9) to evaluate how well they match the typical spatial extent and trajectory across the brain. We also created with pyAFQ 34 an interactive 3D 551 visualization of an example infant's bundles at each time point: 0 months, 3 months, and 6 months.

Modeling R1 development 554 After identifying all bundles with babyAFQ, we modeled their R1 development using linear 555 mixed models (LMMs). First, we modeled mean R1 development within each bundle using LMMs 556 with age as predictor and a random intercept (estimated R1 at birth) for each individual (Fig 2a) . We individuals. To evaluate differences in developmental trajectories between bundles, we plotted the 560 mean R1 measured in newborns (Fig 2b) and well as the mean R1 development rate (slopes of LMMs) 561 for each bundle (Fig 2c) . 562 Next, we evaluated the development of R1 along the length of each bundle. For this, we 563 divided each bundle into 100 equidistant nodes and evaluated R1 at each time-point in each node (Fig   564   3) . We then determined the rate of R1 development at each node (one LMM per node; random 565 intercepts for each individual as above). For each bundle, we then plotted R1 measured in newborns 566 and the rate of R1 development across nodes to visualize their relationship along each bundle (Fig 4) . 567 Finally, we evaluated the relationship between the rate of R1 development (LMM slope) and 568 both the measured R1 in newborns as well as the spatial location in the brain (Fig 5) . This analysis 569 was done for every 10th node along each bundle to ensure independence across nodes within a bundle.

All subplots in Fig 5 show the data at each node plotted at their average location in the newborn's 571 brain (average|x|, y and z coordinates in the newborn sample). For the x axis we used the |x| 572 coordinates, as previous work suggests a medial to lateral spatial gradient of development across both 573 hemispheres of the infant brain 5 . As all newborn data was acpc-ed, the (0,0,0) coordinate corresponds 574 to the average coordinate of the anterior commissure across newborns . Fig 5a is included to orient the reader to the spatial layout in these plots. Fig. 5b shows the spatial layout of measured R1 in 576 newborns across the white matter, and Fig. 5c shows the spatial layout of R1 development rate across 577 the white matter. 578 We quantified the relationship between R1 development rate and initial R1 as well as spatial 579 location via a series of LMMs. In these models we used every 10 th node of each bundle to ensure 580 independence. In the first LMM, we related R1 development rate to R1 measured in newborns, with 581 a random intercept for each bundle:

(1) R1Slope~ 1+ R1 in Newborns + (1|Bundle).

In the second LMM, we related R1 development rate to location in the brain (|x|, y, z, |x|*y, 584 y*z, and z*|x| coordinates, all coordinates were z-scored before including interaction terms), with a 585 random intercept per bundle:

(2) R1Slope~ 1 + |x| + y + z + |x|*y + |x|*z + y*z + (1|Bundle).

In the third model, we related R1 development to both R1 measured in newborns as well as 588 spatial location with a random intercept per bundle:

(3) R1Slope~ 1 + R1 in Newborns + |x| + y + z + |x|*y + |x|*z + y*z + (1|Bundle). 590 We used a likelihood ratio test to assess whether this third model outperforms the second 591 model. Similar LMMs were also performed on mean diffusivity (MD) data, to relate our findings to 592 previous work. MD results are presented in Supplementary Figs 5- 

A structural MRI study of human brain development from birth to

Plasticity in gray and white: Neuroimaging 641 changes in brain structure during learning

Neuronal activity promotes oligodendrogenesis and adaptive myelination 643 in the mammalian brain. Science (80-. )

Myelin remodeling 645 through experience-dependent oligodendrogenesis in the adult somatosensory cortex

A critical period for social experience-648 dependent oligodendrocyte maturation and myelination

Evaluation of neonatal brain myelination 650 using the T1-and T2-weighted MRI ratio

Investigating 652 white matter development in infancy and early childhood using myelin water faction and 653 relaxation time mapping

Observer-independent 655 method for microstructural parcellation of cerebral cortex: A quantitative approach to 656 cytoarchitectonics

The early development of brain white matter: A review of imaging studies in 658 fetuses, newborns and infants

Imaging structural and functional brain 28

Using high-resolution quantitative mapping 683 of R1 as an index of cortical myelination

Myelin and iron concentration in the human brain: A quantitative study of 685 MRI contrast

Quantitative 687 magnetic resonance imaging of brain anatomy and in vivo histology

Superficial white matter imaging: Contrast mechanisms and whole-brain in 690 vivo mapping

Tract 692 Profiles of White Matter Properties: Automating Fiber-Tract Quantification

Recognition of white matter bundles using local and global streamline-695 based registration and clustering

Evaluating the reliability of human brain white matter tractometry

Infant Brain Atlases from Neonates to 1-and 2-Year-Olds

Reproducibility of quantitative tractography methods applied to cerebral 701 white matter

The posterior arcuate fasciculus and the Visual Cortex

The vertical occipital fasciculus: A century of controversy resolved by in 47

Differential White Matter Maturation from Birth to 8 Years of Age

Evolution of T1 relaxation, ADC, and fractional anisotropy during early 728 brain maturation: A serial imaging study on preterm infants

Separate lanes for adding 731 and reading in the white matter highways of the human brain

Rapid high resolution T1 734 mapping as a marker of brain development: Normative ranges in key regions of interest

Myelination is associated with processing speed in early childhood: 737 Preliminary insights

White matter maturation profiles through early childhood predict general 739 cognitive ability

The UNC/UMN Baby Connectome Project (BCP): An overview of the study design 741 and protocol development

Modelling brain development to detect white matter injury in term 743 and preterm born neonates

Neonatal white matter 745 tract microstructure and 2-year language outcomes after preterm birth

Early Detection of Cerebral Palsy Using 64

Volume-based analysis of 6-month-old infant brain MRI for autism biomarker 767 identification and early diagnosis

Notes in Artificial Intelligence and Lecture Notes in Bioinformatics) 11072 LNCS

Fast T1 mapping using slice-771 shuffled Simultaneous Multi-Slice inversion recovery EPI

A framework for multi-component analysis of diffusion MRI data over the 776 neonatal period

Automated processing pipeline for neonatal diffusion MRI in the 778 developing Human Connectome Project

MRtrix3: A fast, flexible and open software framework for medical image 780 processing and visualisation

Diffusion tractography in crossing 782 fiber regions

Denoising of diffusion MRI using random matrix theory

Incorporating outlier 786 detection and replacement into a non-parametric framework for movement and distortion 787 correction of diffusion MR images

Unsupervised 3-tissue response function 791 estimation from single-shell or multi-shell diffusion MR data without a co-registered T1 792 image

Multi-tissue 794 constrained spherical deconvolution for improved analysis of multi-shell diffusion MRI data

Measures of the Amount of Ecologic Association Between Species