Overview of the rationale of the study.

(A) Context: the cerebellum exhibits protracted growth compared to other brain regions (e.g., thalamus), making it a particularly vulnerable region for neurodevelopmental disorders. Its specific microstructural organisation (i.e. layered with Purkinje cells and granular cells) changes between birth and adulthood. (B) The current “standard” is to probe regional growth with manganese-enhanced MRI (MEMRI). However, it is not sensitive to the underlying microstructural changes. (C) Pushing limits: diffusion-weighted MRS (dMRS) can potentially assess cell-specific microstructural changes in given regions during development. Different measures of metabolites diffusion properties (apparent diffusion coefficient, ADC, at long diffusion times, signal attenuation, S, at high diffusion weighting, b-values) can be interpreted with biophysical modelling and help characterise the nature of the microstructural changes.

Manganese-enhanced MRI confirms protracted cerebellar growth in rat pups.

(A) Representative slices around the thalamus and cerebellum of non-linearly averaged images acquired at P5, P15 and P30. (B) Absolute volumetric changes are shown for the segmented thalamus and cerebellum. Large dots correspond to the mean and small dots to individual data points. (C) Normalised volumetric changes. Volume at each age is normalised by the P30 volume. Large dots correspond to the mean and small dots to individual data points.

Metabolic profile changes postnatally.

(A) Spectra were binned per time point, region and b-value and averaged after phase and frequency correction. The SNR is clearly lower at P5/P10 in both regions: the voxel is about half of the P30 voxel size, and more animals were outliers. The SNR is also lower at P30: the rat size was slightly too big for the mouse cryoprobe at this age, so the skull is not at the top of the probe and the sensitivity is suboptimal (B) Spectra at b=0.035 ms/µm2 displayed in (A) were quantified with LCModel and absolute metabolite concentrations were normalised by the absolute macromolecular concentration and levelled to get Signal(tCr,P30)/Signal(MM,P30)=8 in the thalamus. See Figure S2 for a comparison with normalisation by water and individual estimations (rather than average spectra quantification).

Biophysical modelling issued from metabolites diffusion properties shows differential developmental trajectories in the cerebellum (A) and in the thalamus (B).

The first column (A,B) reports the signal attenuation as a function of b-value at TM=100ms, and the second column reports the ADC varying with diffusion time. Shadowed error bars represent the standard deviations. Modeling results from the spheres+”astrosticks” model is reported in the third column (A,B). P5 is shaded in the cerebellum due to possible motion artefact. The parameter space (sphere fraction, sphere radius) is displayed. Error bars represent the standard deviations. Llength, modeling output from the morphometric model is shown in the fourth column (A,B). P5 is shaded in the cerebellum due to possible motion artefact.

Mean values and standard deviations for parameters extracted from the spheres + “astrosticks” model (sphere radius and sphere fraction).

Statistics come from a linear model accounting for age and region as fixed effect. The interaction term was included only if it improved the fit significantly (c.f. Methods). p-values highlighted in yellow pass the α=0.05 threshold after multiple comparison correction and green p-values pass the α=0.01 threshold (Bonferroni correction, 12 hypotheses).

The biophysical parameters are interpreted based on the cerebellar microstructural development and main results are summarised.

Italic font corresponds to possible interpretations. (A) Schematic of cerebellar neuronal development and biophysical modelling parameters interpretation. Lsegment is interpreted as the distance between two branches of the dendritic tree, NBranch as the number of embranchments. The sphere radius Rsphere is interpreted as the average soma radius (i.e. granule and Purkinje cells). fsphere,LT represents the ratio of cerebellar layer thicknesses between the layers rich in sphere-like cell components (e.g., somas in the I/EGL, GL and PL) and the layers rich in fibrous cell components (e.g., cell processes or “neurite” in the ML). fsphere,LT can be seen as a rough approximation of fsphere and is inversely related to the dendritic tree growth, see Table S2. (B) Summary of the cerebellar results coming from the modelling. (1) and (2) refer to the models depicted in Figure 1. (C) Tau seems to change compartments in the thalamus with age, going from neuronal-like compartments (low sphere fraction) to glial-like compartments (high sphere fraction), also indicated by its important signal drop with age. It suggests that Tau is a marker of neuronal maturation. Its concentration remains high up to P30 in the cerebellum. Tau properties in the cerebellum (decreasing sphere fraction + relatively low sphere fractions) and contrast with the thalamus suggest that Tau could be a marker of cerebellar neuronal development.

Morphometric parameters extracted from mouse real cells 3D reconstructions on NeuroMorph shows a decrease of Lsegment up to P13 in developing cerebellar Purkinje cells, but not in developing thalamic interneurons.

(A) Examples of cells extracted from NeuroMorph. The total number of cells used per age and region is given in Supplementary Material. References: Cerebellum P6-P1055, P10-P1356, P35-P4357; Thalamus P9-P3058 (B) The values of Branch Length (Lsegment), Branching Order and total process length are directly estimated from the morphology using the function stat_tree in the Trees Matlab Toolbox.

dMRS acquisition parameters.