Temporal trajectories of normal myelination and axonal development assessed by quantitative macromolecular and diffusion MRI: Ultrastructural and immunochemical validation in a rabbit model

Abstract

Introduction:

Quantitative and non-invasive measures of brain myelination and maturation during development are of great importance to both clinical and translational research communities. While the metrics derived from diffusion tensor imaging, are sensitive to developmental changes and some pathologies, they remain difficult to relate to the actual microstructure of the brain tissue. The advent of advanced model-based microstructura metrics requires histological validation. The purpose of the study was to validate novel, model-based MRI techniques, such as macromolecular proton fraction mapping (MPF) and neurite orientation and dispersion indexing (NODDI), against histologically derived indexes of myelination and microstructural maturation at various stages of development.

Methods:

New Zealand White rabbit kits underwent serial in-vivo MRI examination at postnatal days 1, 5, 11, 18 and 25, and as adults. Multi-shell, diffusion-weighted experiments were processed to fit NODDI model to obtain estimates, intracellular volume fraction (ICVF) and orientation dispersion index (ODI). Macromolecular proton fraction (MPF) maps were obtained from three source (MT-, PD-, and T1-weighted) images. After MRI sessions a subset of animals was euthanized and regional samples of gray and white matter were taken for western blot analysis, to determine myelin basic protein (MBP), and electron microscopy, to estimate axonal, myelin fractions and g-ratio.

Results:

MPF of white matter regions showed a period of fast growth between P5 and P11 in the internal capsule with a later onset in the corpus callosum. This MPF trajectory was in agreement with levels of myelination in the corresponding brain region, as assessed by western blot and electron microscopy. In the cortex, the greatest increase of MPF occurred between P18 and P26. In contrast, myelin, according to MBP western blot, saw the largest hike between P5 and P11 in the sensorimotor cortex and between P11 and P18 in the frontal cortex which then seemingly plateaued after P11 and P18 respectively.

G-ratio by MRI markers decreased with age in the white matter. However, electron microscopy suggest a relatively stable g-ratio throughout development.

Conclusion:

Developmental trajectories of MPF accurately reflected regional differences of myelination rate in different cortical regions and white matter tracts. MRI-derived estimation of g-ratio was inaccurate during early development, likely due to the overestimation of axonal volume fraction by NODDI due to the presence of a large proportion of unmyelinated axons.

1. Introduction

Across its life span, the mammalian brain undergoes dramatic changes in microstructural organization in the cortex (Grydeland et al., 2019) and white matter tracts (Slater et al., 2019). This includes changes in neuron and glial cell numbers, density, dendritic arborization, myelination, and axon growth (Yakovlev PI, 1967). The developmental trajectories of these microstructures are non-uniform and vary between cortical brain regions and the connecting white matter tracts (Dubois et al., 2008a; Prastawa et al., 2010). While measures of brain structure at a single point in time can be useful, it is the trajectories of brain development that are better predictors of cognitive and behavioral outcomes (Giedd and Rapoport, 2010). Understanding normal region-and microstructure-specific developmental trajectories is essential to detect aberrant maturation patterns that are associated with various motor, cognitive and behavioral developmental disorders, including cerebral palsy, autism, schizophrenia, and others in individuals (Gilmore et al., 2020; Kar et al., 2022a).

Conventional MRI contrasts provide little information about changes in microstructural brain tissue between infancy and adulthood. Thus, emerging quantitative imaging tools that can be applied across imaging sites to assess cortical myelination and maturation between brain regions, while allowing for comparisons at different ages, are of great importance to both clinicians and researchers. Widely used quantitative metrics, derived from diffusion tensor imaging (DTI), including fractional anisotropy (FA) and directional diffusivities, are indeed sensitive to developmental and pathological changes (Dubois et al., 2008b; Imperati et al., 2011; Lebel and Beaulieu, 2011; Lebel et al., 2008) yet remain difficult to relate to the actual microstructure of the brain tissue. In contrast, metrics derived from multi-compartment biophysical models (Assaf and Basser, 2005; Palombo et al., 2020; Zhang et al., 2012) can provide insight to the relevant microstructure components, but they are typically based on numerous assumptions, which require correlative histological validation and calibration where necessary.

A number of studies have attempted to validate quantitative MRI indexes of myelination, axonal composition, and g-ratio (Jung et al., 2018; Khodanovich et al., 2019; Lehto et al., 2017; Thiessen et al., 2013; Underhill et al., 2011; West et al., 2018). In these studies, validation is typically performed by comparing MRI metrics with “gold standard ” histological or electron microscopy measurements from adult animal brains in either normal or abnormal settings. However, it is important to take into account that these MRI modalities and derived indexes are founded in the physical properties of water molecules. Therefore, they cannot only be affected by the anatomical structure of interest (i.e., the amount of myelin or axonal density) but also by multiple other aspects of tissue composition (e.g., cellular and axonal size, density, alignment, water contents, etc.). Given these same properties also change as a part of normal development, it is problematic that there is a paucity of histological validation of quantitative, MRI-derived measures in developmental settings.

The purpose of this study was to assess the validity of several recently developed MRI metrics of brain microstructure, by taking measurements at multiple stages of normal development and corroborating the results with concurrent histological analysis. The first modality assessed was Macromolecular Proton Fraction (MPF) mapping. MPF is one of the parameters that can characterize the magnetization transfer (MT) effect in tissues based on the two-pool model (Morrison et al., 1995; Sled, 2018). MPF values are defined as the relative molar concentration of macromolecular protons with restricted motion, which participate in magnetic cross-relaxation with free water protons. MPF has attracted remarkable attention as a quantitative metric of myelination due to the strong correlations between MPF and myelin content in neural tissues, as reported in a number of studies (a comprehensive review can be found in (Kisel et al., 2022). MPF can be measured by a variety of quantitative MT imaging methods that enable the reconstruction of a series of the two-pool model parameter maps. More detail about technical principles of these methods can be found in recent reviews (Kisel et al., 2022; Sled, 2018). One such technique, the single-point method (Yarnykh, 2012, 2016b) allows for the reconstruction of MPF maps in isolation from other two-pool model parameters, thus providing substantial improvements in time efficiency, spatial resolution, and quality of parametric maps due to a reduction of the parameter space. This method enables easy implementation and accurate MPF measurements using both clinical and preclinical MRI systems (Naumova et al., 2016; Yarnykh, 2012; Yarnykh et al., 2018). MPF measured by the single-point method has been histologically validated as a myelin marker in the normal rat brain (Underhill et al., 2011) and experimental models of toxic demyelination (Khodanovich et al., 2017), remyelination (Khodanovich et al., 2019), acute and chronic stroke (Khodanovich et al., 2018), and developmental perturbations (Lu et al., 2018). Sensitivity of single-point MPF maps to myelin development in the human brain has been demonstrated in recent studies of fetuses (Yarnykh et al., 2018), infants (Corrigan et al., 2022), and adolescents (Corrigan et al., 2021).

The second investigated modality, Neurite Orientation Dispersion and Density Imaging (NODDI) (Zhang et al., 2012) is one of several multi-compartment diffusion MRI and tissue modeling approaches that has been developed to increase the specificity of techniques such as DTI to key pathophysiological processes (Assaf and Basser, 2005; Fieremans et al., 2011; Jespersen et al., 2010; Novikov et al., 2018). NODDI fits diffusion data to a three-compartment tissue model: an intra-neurite compartment, in which diffusion is constrained except along the direction of neurites; an extra-neurite compartment, in which diffusion is Gaussian with hindered diffusion perpendicular to the direction of neurites and a CSF compartment, in which diffusion is Gaussian and isotropic. Within each voxel, the variability of neurite orientations is modelled by a Watson distribution. NODDI estimates specific metrics, such as the Orientation Dispersion Index (ODI) and intracellular volume fraction (ICVF) that are sensitive to the density and orientation distribution of dendrites and axons. Despite several simplifying assumptions, NODDI has demonstrated its ability to produce robust estimations of meaningful tissue model parameters in both clinical and developmental studies (Batalle et al., 2019; Chang et al., 2015).

NODDI has been validated by histological estimation of axonal orientation and density in translational models of the normal adult brain (Schilling et al., 2018) and spinal cord injury (Grussu et al., 2017). With a combination of myelin- and axon size-sensitive MRI measurements, an estimation of the ratio between the inner axon radius and outer, myelinated axon radius (g-ratio) it is possible (Stikov et al., 2015a). This is a fundamental parameter for predicting the speed and effectiveness of neural transmission. Although an MRI-based estimation of g-ratio in white matter, using NODDI and multi-echo T2 imaging, was shown to correspond to histological analysis (West et al., 2018), this was only observed in a fixed mouse brain model.

In this study, we compared the developmental trajectories of microstructure in the normal rabbit brain, described by recently developed, in vivo MRI indexes and well established, ex vivo histopathological analysis across various brain structures. To accomplish this, we quantitatively assessed several characteristic parameters of microstructural organization (myelination, g-ratio, axonal fractions, in the cortex and white matter, at several important developmental milestones. Recognizing that the establishment of exact voxel to voxel relationship of absolute values of MRI-derived and “gold-standard ” anatomical measures is unrealistic in practice, the comparison measures we used were the overall shapes of the developmental curves, the onsets of rapid growth, and the time to reach the half- and fully-developed values of the corresponding parameter. Based on the results of previous validation studies described above, it was hypothesized that the developmental trajectories of microstructural organization, described by MRI-based metrics, would correspond to the histologically acquired data.

2. Methods

All procedures reported in this study received approval by the NorthShore University HealthSystem Research Institutional Animal Care and Use Committee, and they were conducted in accordance with the United States Public Health Service’s Policy on Humane Care and Use of Laboratory Animals.

2.1. Study design

Timed pregnant New Zealand White rabbits (Charles Rivers, NJ) gave birth in a nest box at term (31.5 days). At postnatal (P) days P1, P5, P11, P18, and P26, rabbit kits of both genders underwent serial in- vivo MRI examination, including diffusion weighting imaging (DWI) and MPF measurements. Rabbit dams of age 6–8 months were scanned to obtain the adult MRI data. Several brain regions, specifically the corpus callosum and internal capsule white matter tracts, as well as the sensorimotor and prefrontal areas of the cortex were examined across imaging modalities using region of interests (ROIs) followed by tissue sampling and biochemical and histological examination in the same areas. To ensure anatomical correspondence of the imaging and tissue sampling areas we followed a strategy, as previously described (Drobyshevsky et al., 2014a) and shown on Supplementary Figure 1. Twelve oblique coronal slices for diffusion-weighted imaging were positioned on a sagittal localizer, with the fifth slice crossing the anterior commissure and the seventh slice crossing corpus callosum splenium at the thickest point. The number of slices was kept constant for all age groups to cover the same area of cerebrum. Slice thickness was, therefore, variable across individuals and age groups, such that they were about 1 mm for P1, P11, 1.2 mm for P18-P26 kits, and 1.5 mm for adults. In-plane resolution was 0.156 mm for P1 and 0.195 mm for P11-P26 kits. Regions of interests (ROIs) were placed on directionally encoded color FA maps, obtained from the diffusion weighted imaging experiment, as shown on Supplementary Figure 1. The ROIs were placed only on the central portion of the white matter tracts to facilitate tissue sampling and avoid the sampling of nearby areas. To ensure reproducible pitch angle on oblique coronal imaging sections, slices were oriented orthogonal to the frontal cortex pole–pons plane as determined with the aid of a multi-slice sagittal localizer scan. The ROIs were used to extract values for NODDI indexes for individual animals. Scull-stripped MT –weighted and B0 vol were registered for each animal using affine transform with FSL FLIRT. Accuracy of registration was confirmed by manual inspection with checkreg tool in SPM. The transformation matrix was applied to the ROI masks to extract MPF values.

After each MRI session, a subset of randomly designated rabbits was euthanized by overdose of a Ketamine/Xylazine mixture. For histological analysis, the brains were prepared by extirpation, chilled on ice, oriented using anatomical landmarks and were then sliced using a brain matrix (1 mm thick) along planes corresponding to the MRI slice orientation. Sectioned slices were separated into the two hemispheres. One hemisphere was fixed for histology. Blocks of tissue encompassing the corpus callosum and internal capsule white matter tracts were fixed for electron microscopy (EM). The other hemisphere was used to obtain samples of gray and white matter (about 2 mm ³), corresponding to the selected ROIs on MRI, with the aid of stereomicroscope and a tissue puncher. The samples were then flash frozen in liquid nitrogen for myelin determination using western blot of myelin basic protein (MBP). The number of animals with MRI and MBP data was 10 for P1, 5 for P11, 8 for P18 and 5 for P26, 6 for adults. EM samples were processed for 5 animals from each age group at P1, P5, P18 and adult group.

The time points of the study were chosen to capture major developmental milestones in rabbit cortex and white matter. At P1, rabbit kits exhibit poor motor abilities and myelination is yet to begin. Myelination begins by P5 in certain brain tracts (Drobyshevsky et al., 2014b, 2005). At P11, rabbit kits open their eyes and display more motor abilities, including hopping. At P18 and P26, the kits possess adult-like motor abilities, and their myelination is at a mature stage (Drobyshevsky et al., 2014b).

2.2. In vivo MRI methods

Rabbit kits were sedated with an intramuscular injection, containing a mixture of Ketamine (35 mg/kg), Xylazine (5 mg/kg), and Acepromazine (1.0 mg/kg). Animals were placed prone in a cradle with a heated water blanket at 35 °C and imaged in a 9.4 T Bruker Biospec system (Bruker, Billerica, MA). The receiver coil was a linear Bruker rat surface coil, allowing for full brain coverage in P1-P26 rabbits. The transmitter was a 70 mm quadrature volume coil. Adult rabbit dams were imaged in the same magnet using a surface transceiver coil 50 mm diameter.

2.3. Diffusion weighted imaging

The experiments consisted of 64 non-collinear directions diffusion weighted images, 4 b-values per direction, b = 0, 0.840, 1.500, 2.5 ms/μm², single shot EPI readout, with TR/TE/NEX 3000/35/1 matrix 128 × 128, δ = 5 ms, Δ = 15 ms. Twelve oblique coronal slices were oriented as described above, covering most of cerebrum. The scan duration was 9 min 54 s. Each voxel of the multi-shell data was fitted using an NODDI tissue model (Zhang et al., 2012). NODDI models diffusion in each voxel as three independent compartments: intra-neurite, extra-neurite and free water compartment (Zhang et al., 2012). Estimates of volume fraction of the restricted (intra-cellular) diffusion compartment (ICVF), an index of neurite density, the volume fraction of an isotropic diffusion compartment (V_ISO ), and dispersion orientation (ODI) were made using the NODDI toolbox version 1.01 (http://mig.cs.ucl.ac.uk/index.php?n=Tutorial.NODDImatlab). For fitting, the intra-neurite, intrinsic diffusivity value was set to 1.7 μm² /ms and 3.0 μm² /ms for the isotropically diffusing compartment.

2.4. Macromolecular proton fraction (MPF) mapping

3D MPF maps were obtained from three source images (Magnetization transfer (MT)-, Proton density (PD) -, and T1-weighted) using single-point methodology with synthetic reference images (Yarnykh, 2016a). PD- and T1-weighted GRE images were acquired with TR/TE = 16/2.6 ms and α = 3° and 16°, respectively. MT-weighted images were acquired with TR/TE = 25/2.6 ms and α = 9° Off- resonance saturation pulse was applied at the offset frequency 6 kHz with the effective saturation flip angle of 500° All images were acquired in the axial plane with whole-brain coverage and a resolution of 0.23 × 0.23 × 1.0 mm³. All images were obtained with four signal averages. In all 3D imaging experiments, linear phase-encoding order with 100 dummy scans, slab-selective excitation, and fractional (75%) k-space acquisition in the slab selection direction were used. To correct for field heterogeneities, 3D B0 and B1 maps were acquired using the dual-TE (TR/TE1/TE2 = 20/2.9/5.8 ms, α = 8°) and actual flip-angle imaging (AFI) (TR1/TR2/TE = 13/65/4 ms, α = 60°) methods, respectively (Yarnykh, 2007). Imaging time was 14 min 38 s. All reconstruction procedures were performed using custom-written C-language software available at https://www.macromolecularmri.org/.

2.5. Myelin Basic Protein (MBP) measurement by western blot

Contents MBP, as a quantitative measure of myelination, was determined by western blot using rat polyclonal anti-MBP antibody (MAB 386, Millipore, USA, dilution 1:500). Fresh frozen tissue samples were homogenized in an ice-cold RIPA lysis buffer. An equal amount of protein lysate was subjected to SDS-PAGE electrophoresis using Bio-Rad Criterion^™ XT 4012% Bis-Tris precast gels (Bio-Rad Laboratories, Inc., Hercules, CA) and transferred to PVDF membranes using a semi-dry transfer system (Bio-Rad). The membranes were blocked with 5% nonfat milk with 0.1% Tween-20 for an hour on a shaker at room temperature, probed with primary antibodies in 5% NFM in TBST overnight at 4 °C, and then incubated with secondary antibodies for 1 h at room temperature. The optical density of each band on the blot was quantified with ImageJ (NIH, Bethesda, MD), normalized to GAPDH. To compare MBP concentration between samples and ages, each blot included a reference tissue sample, obtained from the cervical spinal cord region of one of the adult animals. The data are presented as a fold change relative to the reference tissue.

2.6. Electron microscopy

For electron microscopy (EM), brain slices were post-fixed with 2.5% glutaraldehyde and 4% paraformaldehyde in 0.1 M sodium phosphate buffer, at a pH of 7.4 for 24 h at 4 °C. For each studied fiber tract (five animals per age), a 1 × 5-mm block was cut perpendicularly to the fiber course (sagittal orientation for the corpus callosum, coronal plane for the internal capsule) and processed for electron microscopy. The tissue blocks were embedded between Aclar sheets (Ted Pella Inc., Redding, CA) to preserve the cross-section fiber orientation. Slices were osmicated, dehydrated and embedded in Spurr’s resin. Sections across the fibers, 0.7 nm thick, were cut on an ultramicrotome and imaged with a JEOL 1230 Transmission Electron Microscope.

Stereological Area Fraction Fractionator method was used to obtain unbiased estimates of area fractions occupied by myelin sheaths and myelinated axons using custom Matlab (Natick, MA) software. The method utilized a Cavalieri point-counting estimate. At least four sections were counted for each animal at a 20,000-magnification field of view, taken in random locations across the fiber tract section. A grid of counting points, spaced by 1 μm in all directions, was placed on EM microphotographs (Supplementary Figure 2A). Fractions of intra-axonal myelin sheath occupied areas were calculated as a ratio of points hitting the object of interest to the total number of counting points. Since the sections were obtained randomly in uniform tracts (within sectioned areas less 0.2 mm thick) with the known orientation (perpendicular fiber course), estimated area fractions were assumed to approximate volume fractions of myelin (MVF_EM) and myelinated axons (AVF_EM).

The g-ratio for each axon, defined as the ratio of the inner axonal diameter relative to that of the axon plus the myelin sheath, was estimated by a fractionator counting frames 4 × 4 μm in size that was randomly placed on EM sections 6.5 μm apart (Supplementary Figure 2B). This approach allows unbiased sampling of lager areas of white matter tracts (Riise and Pakkenberg 2011), while mitigating issues with potential variations in the number of axons per ROI. Myelinated axon inner and outer diameters were manually measured in myelinated axons within counting frames. A minimum of 50 axons per white matter tract were counted for each animal on at least four sections. The rationale for the number of axons sampled in our study was that the coefficient of error (CE) due to counting noise (which results from the placement of the counting grid) should be approximately half or less than the observed coefficient of variation (OCV) to obtain an acceptable level of precision. With this requirement fulfilled, the observed variance reflects the inherent variance across brains (Gundersen, 1986). OCV of axonal sizes between animals in the corpus callosum and internal capsule across studied ages was between 0.28 and 0.45. CE in two-dimensional uniform random sampling can be approximated as 1/sqrt(number of sampled objects) (Riise and Pakkenberg, 2011), resulting in 0.14 for 50 axons sampled per ROI per animal in our study. The number of axons measured per ROI is listed in Supplementary Table 1 and a set of representative EM microphotographs across the studied regions and ages can be found in Supplementary Figure 4. If the axonal cross-section was not circular, the small axis of the ellipsoid was taken as an estimate of axonal diameter. The histological measures of g-ratio was defined as a simple mean (g-ratio_EM) or a root area-weighted mean square (g-ratio_EM_w) value of per-axon g-ratio measurement according to West et al. (West et al., 2016). A calibrated g-ratio index (g_ratio_MRI_cal) was obtained, using the linear regression equation between the g-ratio estimated from MRI (g-ratio_MRI) and the EM-derived g-ratio (g-ratio_EM), separately for corpus callosum and internal capsule.

2.7. Data analysis

2.7.1. Developmental curve fitting

For quantitative comparison of the rate of changes in myelination between different regions and between different modalities of myelin quantification, each myelination index was normalized by the range of changes during development between zero and one and presented as a fraction of adult value. MRI and histology data were fitted with models that best described the data distribution across development. Myelination indexes in MRI, western blot and EM histology clearly exhibited three developmental phases: pre/early and slow myelination, fast growth, and plateau. To compare the relative onset and rate of change, the data were modeled with 3-parameter logistic growth equation $\mathrm{y}(\mathrm{t})=L/(1+\mathrm{e}\mathrm{x}\mathrm{p}({-\text{k}}^{\mathrm{*}}(\text{t-m})))$ where $\mathrm{L}$ represents the curve’s maximum value, $\mathrm{m}$ is the time value of the sigmoid midpoint, $\mathrm{k}$ represents logistic growth rate or steepness of the curve. Models for non-linear fits were chosen based on Akaike’s corrected Information Criterion (AICc), calculated in GraphPad Prism 9.5 (GraphPad Software, San Diego) and summarized in Supplementary Table 2. In the developmental curves where the pattern was not apparent or close to being linear, growth rate was modeled with a simple linear regression.

2.7.2. G-ratio estimation on MRI

G-ratio in the regions of interest on MRI volume (g-ratio_MRI) was calculated using volume fractions according to the model of Stikov and colleagues (Stikov et al., 2015a) and expanded by (West et al., 2018): $\text{g-ratio\_MRI}=(1+\text{MVF}/\text{AVF}{)}^{-0.5}$, where $\text{MVF}$ and $\mathrm{A}\mathrm{V}\mathrm{F}$ represent myelin and axonal volume fractions, respectively. $\text{MVF}$ was calculated using linear regression between MPF and myelin volume fraction on EM. $\text{AVF}$ was calculated as $(1-\mathrm{M}\mathrm{V}\mathrm{F}{)}^{\mathrm{*}}{\left(1-{\mathrm{V}}_{\text{ISO}}\right)}^{\mathrm{*}}\mathrm{I}\mathrm{C}\mathrm{V}\mathrm{F}$ according to ( Stikov et al., 2015a ).

3. Results

3.1. Development of myelination in gray and white matter on MPF maps and MBP western blot

Representative MPF maps of several developmental stages (P1, P5, P11, P18, P26, and adult) are shown in Fig. 1. MPF values in selected gray and white matter regions are presented in Supplementary Table 3. During the pre-myelination stage at P1 ( Fig. 1A), MPF was between 2 and 3%, both in gray and white matter. MPF was relatively higher in the cortical plate relative to subcortical regions and fiber tracts, apparently reflecting higher density of cellular components in the cortex at this age, including lipids and other macromolecules. As the brain developed, MPF rapidly increased in white matter tracts as well as in the striatum and thalamus. The increase of MPF in the cortex was less apparent, shown at P26 on Fig. 1B. Motor abilities in rabbits are already fully developed at P26, but MPF in white matter continued to increase after P26 for several months until adulthood (Fig. 1C). Notably, myelination already began at P5 in projection fibers, such as in the internal capsule, as demonstrated on MBP immunostaining (Drobyshevsky et al., 2005), but it was not readily apparent on MPF maps until P11 (Fig. 1C).

Fig. 1.

Parametric maps of macromolecular proton fraction (expressed in%) across postnatal development in rabbits. A. MPF map at P1 with color mapping MPF scale 1–10% to visualize gray/white matter contrast at pre-myelination stage. B. MPF map at P26. C. MPF maps of a slice through anterior thalamus across development from P1 to adulthood. Scale bar in bottom row is 3 mm. pCx–prefrontal cortex, Ac–anterior commissure, Co–optic chiasm, IC–internal capsule, pWm–periventricular white matter, Fi-fimbria, mCx–sensorimotor cortex, Cc–corpus callosum, Cg–cingulum, Th–thalamus, Ot–optical tract, Hc–hippocampus.

Developmental changes in myelin contents of brain regions, measured by MPF on MRI and MBP western blot, are shown in Fig. 2 and Supplementary Table 3. MBP values closely corresponded to MPF across the developmental span in white matter and the cortex, with a linear fit shown in Fig. 2A, B and the corresponding MBP immunoblots in Fig. 2C,D. Pearson correlations between MPF and MBP across developmental span was 0.90 for white matter and 0.79 for cerebral cortex (Supplementary Table 4). Myelination of projection fibers in the internal capsule and cerebral peduncle was accelerated relative to the corpus callosum, both in MPF and MBP, indicated by earlier appearance of myelin in western blot and earlier increase in MPF. Similarly, the sensorimotor cortex myelinated faster than prefrontal cortex. Hence, regional differences in development were captured by both MPF and histological indexes.

Fig. 2.

Validation of myelin contents changes in postnatal rabbit development using MRI and immuno-histological indexes. Myelin contents changes were estimated by MPF on MRI and by MBP western blot. A. B. Scatter plot of MPF vs. normalized MBP values and linear regression fit for selected white (A) and gray matter (B) structures and corresponding representative MBP western blots (C, D). Densities of bands in western blot in each structure were normalized by GAPDH and by the reference tissue (adult spinal cord, present in each blot sample). Whiter matter was sampled in internal capsule (IC) at the level of anterior thalamus and in corpus callosum splenium (CC). Gray matter was sampled in sensory-motor cortex (mCx) and prefrontal cortex (pCx). E, F. Scatter plots of MPF and MBP western blot values and logistic fit curves in selected white matter (E) and gray matter (F) regions vs. postnatal age.

The MPF and MBP changes with age were fitted with a 3-parameter logistic function, representing a typical S-shaped growth curve. This type of fitted curve captured the initial periods of slow growth, which transitioned to fast growth, followed by an asymptotic line in adulthood. Normalized MPF and MBP data and logistic curve fits are shown in Fig. 2E–F, with curve parameters listed in Table 1.

Table 1.

Parameters of the fitted logistic model of myelination indexes vs. postnatal age.

Myelination index	Region	k (fastest slope), units/day	m (midpoint), days	L, asymptotic value	R2 _adj
MPF, normalized	Corpus callosum	0.1681 (0.1317, 0.2045)	27.39 (25.84, 28.94)	1.004 (0.9557, 1.053)	0.9756
MPF, normalized	Internal capsule	0.1416 (0.1093, 0.174)	19.15 (17.23, 21.08)	0.9873 (0.9107, 1.064)	0.9338
MBP western blot	Corpus callosum	0.1847 (0.1387, 0.2307)	19.51 (17.66, 21.35)	0.9897 (0.8988, 1.081)	0.9425
MBP western blot	Internal capsule	0.2807 (0.1923, 0.3691)	11.64 (10.25, 13.04)	0.9433 (0.8781, 1.008)	0.9318
MPF, normalized	Prefrontal cortex	0.1864 (0.1171, 0.2558)	20.29 (17.69, 22.88)	0.996 (0.8786, 1.113)	0.8981
MPF, normalized	Sensorimotor cortex	0.1158 (0.07276, 0.1588)	19.32 (15.55, 23.09)	0.9966 (0.8725, 1.121)	0.8490
MBP western blot	Prefrontal cortex	0.3586 (0.1158, 0.6014)	19.37 (17.26, 21.49)	1.007 (0.8758, 1.138)	0.9000
MBP western blot	Sensorimotor cortex	0.228 (0.1089, 0.347)	12.19 (9.281, 15.1)	0.9815 (0.8485, 1.115)	0.7943
Myelin volume fraction (MPF)	Corpus callosum	0.05343 (0.03928, 0.06758)	21.38 (19.65, 23.1)	0.2703 (0.2552, 0.2854)	0.9580
Myelin volume fraction (MPF)	Internal capsule	0.1328 (0.1089, 0.1567)	12.11 (10.76, 13.45)	0.2433 (0.2308, 0.2559)	0.9472
Myelin volume fraction (EM)	Corpus callosum	0.1758 (0.05817, 0.2934)	18.49 (15.41, 21.58)	0.2653 (0.2351, 0.2955)	0.8930
Myelin volume fraction (EM)	Internal capsule	0.1907 (0.1063, 0.2751)	10.29 (7.926, 12.66)	0.2619 (0.2308, 0.2931)	0.8243

Both MRI and histological indexes reflected differences in timing of myelination between the different brain regions. There was an apparent left shit in both the MBP and MPF fitted curves of the internal capsule in relation to the corpus callosum (Fig. 2C), indicating earlier and faster myelination in projection fibers relative to commissural fibers. Similarly, earlier development of the sensorimotor cortex was captured by a corresponding left shift in both the MBP and initial portion of the MPF fitted curves (Fig. 2D). MPF in the prefrontal cortex experienced a sharper rise between P18 and P26, unlike the steady increase observed in the somatosensory cortex. There was also a notable difference between MPF and MBP developmental curve within the same regions. Fitted MBP curves, both in the white matter tracts and cortex, were shifted left relative to MPF fitted curves, resulting in a significantly earlier estimation of the timing of 50% of asymptotic maximum (Table 1 ).

3.2. Development of white matter tissue compartments, assessed on electron microscopy

Representative electron microphotographs of the corpus callosum and internal capsule at several stages of development from P5 to adulthood, obtained from sections corresponding to MRI measurements, are shown on Fig. 3 A. At P5, the rabbit corpus callosum represented a largely unmyelinated tract, packed with unmyelinated axons around 20 nm in diameter (Fig. 3A top row), as reported previously (Drobyshevsky et al., 2014a). In contrast, groups of small myelinated fibers were apparent at P5 in the internal capsule (Fig. 3A bottom row). By P11, both tracts had multiple small axons undergoing myelination. Through development, axon caliber, as well as myelin thickness increased in both structures, as indicated by a right shift in corresponding histograms for the corpus callosum (Fig. 3B, C) and internal capsule (Fig. 3E,G). Even at the adult stage, both structures retained a large proportion of unmyelinated fibers (Drobyshevsky et al., 2014a). Across development and into adulthood, both structures demonstrated high non-uniformity of axonal sizes. While the majority of myelinated fibers had diameters of 1–2 μm, a large proportion of axons (10–30%) reached 8–10 μm in diameter (Fig. 3B, E). Myelin sheath thickness increased with axonal diameter across developmental time points and white matter tracts (Supplementary Figure 3). Lager diameter axons tended to have a larger g-ratio, despite a thicker myelin sheath, as was apparent on scatterplots of g-ratio vs. axon diameter, shown in Figs. 3D,G. An increased g-ratio with axonal diameter is a common finding and is typically approximated with a log-linear relationship in adult animals (West, Kelm et al. 2015). Through development, axons with the same diameter tended to have smaller g-ratios, due to the increased myelin sheath thickness, as indicated by a right shift of logarithmic curve fit of g-ratio vs. axonal diameter on Figs. 3D,G.

Fig. 3.

Measurements of myelination indexes on electron microscopy samples. A. Representative sections of corpus callosum (top row) and internal capsule (bottom row) at several postnatal ages. Both structures retain a large portion of unmyelinated axons until adult age. Scale bar is 2 μm. Asterisk indicates an oligodendrocyte cell body. Age-dependent right shift oh histograms indicate increase of myelinated axon diameter (B) and myelin sheath thickness in the corpus callosum (C) and internal capsule (E,F). D, G - Scatter plot of g-ratio vs. axon diameter in corpus callosum (D) and internal capsule (G). Data were fitted with straight or semi-log line depending on AICc criterion. Larger axons tend to have smaller g-ratio in all studied ages across developmental span.

3.3. Calibration of myelin volume fraction by EM

MVF_EM and MPF were strongly correlated in corpus callosum (F_1,18 = 48.65, p< 0.001) and in internal capsule (F_1,14 = 1481, p<0.001) (Fig. 4A) with intercepts and slopes of the linear regression being nearly identical in the corpus callosum and internal capsule. Using region specific linear regression parameters shown in Fig. 4A, myelin volume fractions in the corpus callosum and internal capsule were derived from MPF values on MRI (MVF_MRI), and fitted using a logistic function against subject age (Fig. 4B). Regarding the developmental curves of myelin volume fractions, MVF_EM and MVF_MRI demonstrated an earlier myelination onset (timing of rapid rise) in the internal capsule relative to the corpus callosum and similar slopes in the fast growth phases (Table 1 ). There was also a right shift in MVF_MRI curves relative to the corresponding EM estimates, suggesting imperfect correspondence between MRI and EM myelin indexes.

Fig. 4.

Calibration of myelin water fraction, MRI index by histological myelin fraction.

A. Volume fraction of myelin sheaths on electron microscopy sections (MVF_EM), estimated by stereological method in corpus callosum and internal capsule vs. MPF values on MRI. B. Developmental curves of myelin volume fraction obtained from MRI and EM measurements. MPF was scaled to the myelin volume fraction (MVF_MRI) using region regression equations shown in panel A. Data were fitted with a three-parameter logistical model. Left shifts in both MRI and histological myelin estimates indicate accelerated myelination of internal capsule relative to corpus callosum.

3.4. Tissue microstructure parameters from NODDI modeling

Multi-shell diffusion-weighted images were acquired in vivo at different time points across the rabbit’s lifespan and fitted to NODDI model. Representative maps of intracellular volume fraction (ICVF) and neurite orientation dispersion (ODI) are shown in Fig. 5 of axial brain sections at approximately the same anatomical location (anterior thalamus) and at different developmental time points. There was an apparent increase of intracellular volume fraction (ICVF) both in gray and white matter. Neurite orientation dispersion, however, experienced a rapid increase of in the cortex and a slower increase in the white matter. Our longitudinal data support the following distinctions in regional, diffusion-parameter trends across development in the rabbit brain, which has previously been described in animals and humans (Batalle et al., 2019; Jelescu et al., 2015).

Fig. 5.

Parametric maps of intra-cellular volume fraction (ICVF) and orientation dispersion index (ODI), calculated from the NODDI model. Axial section through the corpus callosum splenium, at the level of the anterior thalamus are shown across postnatal rabbit development. Scale bar in ODI maps (white) is 5 mm. Cc-corpus callosum and Ic- internal capsule ROIs.

3.5. Developmental trajectory of axonal volume fraction on MRI and EM

The scatterplots of the estimates of axon volume fractions on MRI (AVF_MRI) and myelinated axon volume fractions on EM (AVF_EM), obtained from the same animals in the corpus callosum and the internal capsule, and regional linear fits are shown on Fig. 6A. The MRI axon volume fraction was a significant predictor of AVF_EM on regression analysis in corpus callosum (F_1,12= 7.36, p< 0.019) and in internal capsule (F_1,17= 24.12, p< 0.001), although there was a large proportion of unexplained variance, indicated by small R². The term “axon volume fraction” is used for the MRI index, since NODDI does not model myelinated and unmyelinated axons separately, but estimates intra-axonal (“neurite ”) fraction. There was a notable discrepancy, such that NODDI estimated an AVF of around 0.3 when myelinated axons are not present on EM.

Fig. 6.

Calibration of axonal fraction MRI index, derived from NODDI modeling and by histological myelinated axon fraction. Model fit between from EM- and MRI-derived axonal fractions vs. age in corpus callosum (B) and internal capsule (C).

Both axon fractions estimates AVF_MRI and AVF_EM increased with age, and this growth was best fitted as a linear function for the corpus callosum (Fig. 6B), and a logarithmic function for internal capsule (Fig. 6C). The biggest difference between AVF_MRI and AVF_EM estimates was observed at the earlier ages, while the estimates at the adult stage were similar for both modalities and across white matter tracts.

3.6. Validation of developmental changes of g-ratio in corpus callosum and internal capsule

G-ratio indexes were estimated on MRI (g_ratio_MRI) from MPF and NODDI derived myelin and axon volume fractions. The average g-ratio was also directly sampled using a stereological method on corresponding EM sections (g-ratio_EM). G_ratio_MRI decreased with age in both the corpus callosum (Fig. 7A) and internal capsule (Fig. 7B), reflecting an estimated larger increase in myelin volume, relative to axon volume. In contrast, g-ratio_EM essentially did not change with age (ANOVA with age factor was F_3,21 = 2.17, p = 0.127 for CC, F_4,26 = 0.65, p = 0.505 for IC).

Fig. 7.

Developmental trajectories of g-ratio in the corpus callosum (A) and internal capsule (B) by MRI- and histology-derived measurements. G-ratios curves from MRI are approximated by a second-degree polynomial. If MRI-derived axonal volume fraction is scaled using the regression equation from Fig 6A, the resulting developmental trajectory (g-ratio_MRI_cal) more closely approximates histologically derived developmental g-ratio curves. No apparent relationship evaluated on the full studied developmental range (red) was observed between MRI- and histologically derived g-ratio indexes in either the corpus callosum (C) or internal capsule (D). Black dotted line is the identity line.

Developmental curves were nearly identical for g-ratio_EM and g-ratio_EM_w (blue and green curves, respectively, in Fig. 7A, B). The largest discrepancy between MRI and EM derived g-ratios was observed at the earlier developmental stages. However, estimates from both modalities converged at the adult age, thereby suggesting g-ratio, derived from MPF and NODDI, is more accurate in adulthood. Considering that a likely source of discrepancy at earlier ages is an overestimation of myelin volume, we applied the established fitting parameters between g_ratio_MRI and g-ratio_EM, shown above in Fig. 6A, to correct the axon volume fraction in calculation of g-ratio from MRI indexes. The corrected developmental trajectory of the g-ratio (g_ratio_MRI_cal), plotted in Fig. 7A, B in purple, was closer to the developmental trajectory of the histologically measured g-ratio_EM than the uncorrected g_ratio_MRI.

We did not observe a significant correlation between the g-ratio estimates by EM and MRI (Fig. 7C, D), either in the corpus callosum or internal capsule when the correlation is examined across the full range of examined ages. When examined by age group on the other hand, g-ratio_EM and g_ratio_MRI were proportional only in adults (Fig. 7C, D, regression shown for adult age). In both of the examined anatomical regions, g_ratio_MRI underestimates the g-ratio_EM values before adulthood.

4. Discussion

In this study, we tested the ability of advanced quantitative MRI measures, MPF and NODDI indexes, to accurately depict developmental trajectories of axonal and myelin development in gray and white matter. Validation of these techniques was gauged by comparing the MRI-derived indexes to “gold standard” histological measures across the development of rabbit brain. We found a general agreement in the developmental trajectories of myelination between MPF and histological measures (i.e., MBP western blot and EM). Both MRI and histological measures support the established faster myelination of projection fibers compared to commissural white matter tracts (Yakovlev and Lecours, 1967). Moreover, myelination began earlier in the rabbit sensorimotor cortex than in the prefrontal cortex, as indicated by both MPF and by MBP western blot. This is in agreement with the human findings, which have shown relatively steeper myelin growth in the primary motor cortex and flatter trajectories in the mid/posterior cingulate cortices (Kwon et al., 2020).

4.1. Histological correlative validation of quantitative myelin assessment in the developing brain

Non-invasive brain myelin assessment has been studied intensely due to the critical importance of myelin in healthy brain function and its involvement in a number of developmental and neurodegenerative disorders. Several MRI techniques have been developed to assess myelin in the brain and spinal cord, including multicomponent T2-relaxometry (MacKay et al., 1994), magnetization transfer (Gareau et al., 2000; Kucharczyk et al., 1994), quantitative susceptibility (Lodygensky et al., 2012a) and other modalities that are influenced by, but cannot directly measure, myelin.

Metrics derived from diffusion tensor imaging are sensitive to developmental and pathological changes but are difficult to relate to actual brain tissue microstructure. The advent of advanced model-based microstructural metrics requires correlative histological validation. Many myelin-sensitive MRI techniques already have some histological validation (Dula et al., 2010; Lodygensky et al., 2012b; Sepehrband et al., 2015). A recent systematic review (Lazari and Lipp, 2021) of 71 studies, found meta-analytic correlation effect sizes ranging between R² = 0.26 and R² = 0.82 between MRI metrics and histological myelin estimations in animal and human tissue. Typically, MRI-histology correlations are performed in adults when myelination is complete. While these timepoint correlations are important, it is the development of these brain regions and microstructures that are better predictors of cognitive and behavioral outcomes.

To the best of our knowledge, this is the first study focused on longitudinal histological validation of a quantitative MRI-based myelin assessment in the normal development. MPF was chosen as a fast and clinically feasible quantitative index, which is based on a two-pool model (Yarnykh, 2016b) and has been validated in normal (Underhill et al., 2011) and pathological brains (Goussakov et al., 2019; Kisel et al., 2022). In the current study, we found that MPF indexes were in strong agreement with the myelin content estimated histologically by MBP western blot and EM, to describe regional developmental trajectories of gray and white matter myelination. Both EM and MBP histological markers explained approximately the same amount of variance in MPF with the regression R² ranging between 0.8–0.9. The advantage of EM marker is that it can be measured in absolute value, while MBP has to be normalized to a standard. In turn, MBP allowed us to assess myelination in cortex where EM measurement of myelination is difficult.

The delayed onset of changes in MPF in early development, observed in this study relative to MBP and myelin fraction on EM, was likely due to the different sensitivity of MPF and histological markers to changes in myelin contents. For example, we have previously observed beginning of myelination in the rabbit internal capsule as early as at the P5 stage on MBP immunostaining (Drobyshevsky et al., 2005), but the onset of myelination in this region was not readily apparent on MPF maps until P11. A possible source of these discrepancies may be the noise floor, which is typically more pronounced for lower MPF values (Yarnykh, 2012). Alternatively, MPF could have a lower sensitivity to structurally immature myelin compared to mature myelin, which is more compact and characterized by thicker sheaths with more layers. In line with this explanation, a dramatic increase in MBP expression precedes myelin compaction during early white matter development in the spinal cord (Grever et al., 1996). Theoretically, MPF measurements also could be affected by concomitant changes in the water content and non-myelin macromolecules. However, the expected magnitudes of these effects are in disagreement with our experimental observations. Particularly, it has been shown that water content decreases by 10–15% from the neonatal to adult brain (Dobbing and Sands, 1973). Alterations in water concentration may cause relative MPF changes to a comparable (Khodanovich et al., 2018). However, a two-to-four-fold increase in MPF over the course of development (Supplementary Table 3) as well as the 30–40% difference between MPF and MBP temporal trajectories at the midpoint (Fig. 2) substantially exceed the possible effect of water. The effect of non-myelin contributions to MPF manifests as the intercept in the linear regression model of MPF against histological measures of myelin (Khodanovich et al., 2019, 2017; Underhill et al., 2011). Previous studies indicate that this intercept is much smaller than MPF in adult white matter and has nearly identical values in normal and demyelinated white and gray matter (Khodanovich et al., 2019, 2017; Underhill et al., 2011). Similarly, MPF demonstrated a small and uniform offset relative to the MBP content across brain structures in this study. In the context of a temporal lag between myelin development trajectories depicted by MPF and MBP or EM, alterations in the non-myelin macromolecular content are unlikely to provide a biologically plausible explanation. As MBP surges earlier than MPF, there should be a molecular or cellular mechanism, which offsets an increase in MPF due to myelin by reducing non-myelin macromolecular content. The presence of such a mechanism does not seem to be physically possible due to both the small scale of the non-myelin portion of MPF (about 3% in the absolute scale, Fig. 2) and insensitivity of MPF to large variations in the content of non-myelin components of neural tissue, particularly microglia (Khodanovich et al., 2018) and astroglia (Khodanovich et al., 2021). As such, the observed differences between myelin development trajectories according to different methods are more likely to be related to either instrumental factors or biological asynchrony between myelin proliferation, compaction, and MBP expression.

4.2. Asynchronous regional myelination in brain gray and white matter

It has been established, based on histological observations, that brain development and myelination exhibits regional asynchrony with varying onsets and rates of change. Reports by Yakovlev and Lecours (Yakovlev PI, 1967) demonstrated that the primary motor and sensory cortices show robust myelination in the first decade of life. In contrast, associative cortical areas only showed substantial increases in myelin in the second decade. White matter, projection and limbic fibers had accelerated development relative to associative fibers. With the advent of non-invasive MRI techniques, regional developmental trajectories have been described based on T1 and T2 contrasts, magnetization transfer (Dubois et al., 2014; Nossin-Manor et al., 2013; Westlye et al., 2010), and diffusion tensor metrics in white matter tracts (Imperati et al., 2011; Lebel and Deoni, 2018; Lebel et al., 2008) and cortex (Kwon et al., 2020). The changes in these indexes have been typically attributed to myelination. Although, they may also be affected by other developmental changes in microstructure, such as axonal packing and streamlining, dendritic arborization, changes in water and lipid contents. MRI methods, specifically aimed at modeling myelin contents, have been used across human development (Dean et al., 2014; Deoni et al., 2011), but without histological validation. Here, for the first time, we characterized developmental trajectories in regional gray and white matter, assessed by myelin and compartment water diffusion sensitive MRI techniques along with histological validation.

We found that with MPF assessment, the relative differences in timing of brain myelination were preserved between projection and commissural tracts, as well as between somatosensory and frontal cortex, in agreement with the MBP western blot and EM assessment of myelination. Since MPF is a rapid and clinically feasible MRI technique that provides a sensitive measure of myelin in absolute units in white matter tracts and the cortex, it opens opportunities for longitudinal assessment of regional myelination trajectories in human developing brain. The region specific maturation and myelination rate is becoming important as an index of normal (Dubois et al., 2008b) and pathological (Kar et al., 2022b) brain development. The rabbit data in our study provide histological validation, relevant to human development, since rabbit brain developmental and microstructural changes occur during the perinatal period (Clancy et al., 2001; Workman et al., 2013), which is similar to humans.

4.3. Accuracy of G-ratio assessment by in vivo MRI in postnatal development

G-ratio is an important structural index of fiber tract development and has been shown to affect signal conduction effectiveness and velocity (Rushton, 1951). The ability to measure region specific g-ratio non-invasively in relation to brain growth and functional development is invaluable, since the existence of an optimal g-ratio has been postulated (Chomiak and Hu, 2009) and might be sensitive to developmental disturbances. Recent advances in biophysical tissue modeling allow for the assessment of g-ratio using MRI, which requires the estimation of myelin and axonal volume fractions (Stikov et al., 2015b). Such modeling is not straightforward, requires several assumptions (Mohammadi and Callaghan, 2021), and is especially problematic with the presence of non-myelinated fibers and cell bodies. G-ratio calculations using a model by Stikov (Stikov et al., 2015b) was histologically validated using “gold standard” EM sections of fixed spinal cord (West et al., 2016) and brains of control and knockout mice with different degrees of myelination (West et al., 2018). The investigators reported a significant correlation between MRI-based and histological measures with a slight overestimation of MRI g-ratio estimation, which was attributed to the presence of non-myelinated axons. This model assumes a constant g-ratio of all axons within a voxel. However, West et al. later showed that MRI g-ratio estimates are equal to the square root of the axon-area weighted mean of squared g-ratio values from all axons in a voxel (West et al., 2016a).

There is a paucity of direct experimental validation of myelin fraction and g-ratio measurements by MRI in brain development, in either humans or animals. This is especially important due to the inconsistent data on MRI-derived g-ratio changes during development. A decrease of g-ratio in all white matter tracts was found in children from 3 months old to 7.5 years of age (Dean et al., 2016). G-ratio in this study was estimated from myelin content information (obtained with T1 and T2 relaxation-based technique mcDESPOT) and neurite density information (obtained through NODDI diffusion imaging). Using similar techniques of g-ratio estimation, an inverse U-shaped relationship between aggregate g-ratio and age was reported in most human cerebral regions from 21- to 84-years-old (Bouhrara et al., 2021). This suggests myelin continues to mature until middle age, followed by a decrease at older ages. On the other hand, g-ratio in major human fiber tracts was found to be essentially constant between 6 and 81 years old (Berman et al., 2018). The tract-specific analysis showed this relationship to follow a nearly linear increase of g-ratio starting 20 years of age (Cercignani et al., 2017). Consistent with the human data, the trajectory of MRI-estimated g-ratio of gray matter in our study could also be described as an inverse U-shape curve (Fig. 7A) with a rapid decline of g-ration in early development. However, this trajectory was in disagreement with the histologically measured g-ratio on EM, where it essentially did not change during neonatal and adolescent period, and the U-shaped approximation was much flatter.). The phenomenon of relatively constant g-ratio with maturation may be explained by concomitant increases of myelin sheath thickness (Fig. 3C, F) and myelinated axon sizes (Fig. 3B, E), Supplementary Figure 3, with age. Our EM data were consistent with a study of myelin and g-ratio estimation on EM in the corpus callosum of non-human primates from 2 weeks old to 35 years of age. In the primates, myelin fraction increased, plateaued, and decreased, the g-ratio followed a stable trajectory over the lifespan and then slightly decreased with subject age (Watson et al., 2022).

The discrepancy between g-ratio estimation by MRI relative to the EM data in our study was the largest in the early stages of development, while the adult values agreed. The over-estimation of g-ratio at early development was likely caused by the overestimation of axonal myelinated axons volume fraction by NODDI modeling (Fig. 6), which does not separate the contribution of non-myelinated axons and, possibly, by violations of other assumptions of the NODDI model. This discrepancy in AVF and g-ratio MRI estimation could be corrected if the EM data are available. Since such corrections are likely to species- and brain region-specific, as suggested by our data for the corpus callosum and internal capsule. Therefore, the corrections identified in this study cannot be readily applied for human g-ratio data in developmental settings. A systematic overestimation of intra-axonal fraction was reported in comparison to NODDI with the White Matter Tract Integrity (WMTI) model using a dataset of human children of 0–3 years old (Jelescu et al., 2015). One source of the discrepancy was attributed to a fixed value of axial diffusivity in NODDI, while the WMTI model estimates the value. This value was observed to be dependent on brain region and age. The accuracy of WMTI to estimate ICVF in development remains to be histologically validated. While the WMTI model has its own limitations, such as, marginal applicability in a tract with larger orientation dispersion (e.g., the internal capsule), both NODDI and WMTI are based on a similar biophysical model and do not separate myelinated from unmyelinated compartments. Thus, both NODDI and the WMTI model do not provide an accurate estimation of myelinated axon volume fraction. The development of novel techniques, which can model non-myelinated axons and cell soma compartments (Palombo et al., 2020), are necessary to improve estimations of axonal fraction and g-ratio in white matter tracts and the cortex across development.

4.4. Limitations

In the current study, we attempted to validate MPF and NODDI metrics against histological markers of axon and myelin development. One of the limitations was a relatively small number of developmental time points that were studied, particularly in older rabbits. As a result, the model fit to the change in myelin measure vs. age may not accurately reflect what was happening between day 26 and adulthood.

The first methodological limitation was inaccuracies that resulted from tissue sampling and processing. EM is commonly considered to be the “gold standard” for white matter structural assessment. Problems with tissue shrinking and distortion during chemical fixation are well-known, having been extensively described elsewhere, and affect the estimates of compartment volume fractions. This in one of the reasons why MBP determination of myelin content was added to the study as an additional confirmation technique. Another limitation is that estimated area fractions on EM sections were assumed to approximate volume fractions of myelin (MVF_EM) and myelinated axons (AVF_EM). This is a common assumption for tracts like mid-corpus callosum of spinal cord, but may deviate in tracts where EM sections may be not strictly perpendicular to the fiber course.

There was an apparent deviation from linearity in the relationship between MPF and MBP at the larger values of MBP, which can likely be attributed to the saturation effect of western blot (Pillai-Kastoori et al., 2020). We also found that there was a left shift in the developmental curve of histological indexes relative to MPF, suggesting different sensitivities between the markers to myelin contents.

The choice of values for fixed parameters to estimate the NODDI model may have affected the estimated compartment fractions between ages and regions. In the original NODDI model paper (Alexander et al., 2010), the variation in intrinsic diffusivity across regions and subjects was not considered to be significant enough to remove trends in the estimated parameters. The chosen parameter of intrinsic diffusivity set to 1.7 μm² /ms for the NODDI model is based on human parallel diffusivity in the corpus callosum and is very similar to that of the adult rabbit. In our rabbit data, axial diffusivity had differing developmental trajectories across white matter tracts: the corpus callosum increased from 1.85 to 1.9 μm² /ms from P1 to P11, but decreased from 1.7 to 1.45 in the internal capsule during the same age interval. In a study, exploring optimality of the parallel diffusivity value using a minimal NODDI model residuals as a criterion (Guerrero et al., 2019), the value of 1.7 μm² was found appropriate for adult brain white matter, but suboptimal in infant brains. In human infants it is sometimes set to a higher value 2.0 μm² /ms (Kunz et al., 2014). Increasing parallel diffusivity was shown to increase estimates of intracellular compartment volume fraction. The optimum value of parallel diffusivity, using histologically validated value of ICVF as a criterion, remains unexplored for younger brains and specific white matter tracts. Therefore, for simplicity of comparison between ages and regions, intrinsic diffusivity was set to fixed intermediate value of 1.7 μm² /ms. Isotropic diffusivity values should not be dependent on species and age, and was confirmed to be close to 3.0 μm² /ms by measuring ADC in rabbit lateral ventricles.

In general, the results of the study are limited by the correlative nature of the analysis between MRI and histology metrics, as well as the need for species- and brain region-specific calibrations for certain MR measurements. While being able to show developmental changes, MRI indexes are not able to directly measure myelin and are, therefore, not immune to other epiphenomenal changes in tissue microstructure. The next step to validate the sensitivity and specificity of the metrics would be to construct regional developmental curves with experimental manipulation of myelination and axonal growth in animal models. The consequences of clinically relevant developmental injuries to neurons and/or oligodendrocyte cell lineages are complex in terms of axon and myelin development and their interactions may challenge the validity of current biophysical models and derived MRI-metrics. Specific mechanistic manipulations or disease models affecting axonal and myelin integrity, similar to the excellent experimental work by Jelescu et al. (Jelescu et al., 2016), are needed to extend the scope beyond normal development and selected white matter tracts.

4.5. Conclusion

Overall, our findings suggest quantitative and non-invasive MRI metrics can elucidate developmental trajectories of normal myelination and axonal development which have important implications for motor, cognitive and behavioral outcomes in patients. While MPF metrics of myelination were in agreement with histological counterparts, for both white and gray matter structures in normal development, NODDI-derived estimation of axonal volume fraction and, consequently, g-ratio deviated from the corresponding histological measurements.

Data Availability

Data will be made available on request.