Learning-related contraction of gray matter in rodent sensorimotor cortex is associated with adaptive myelination

Abstract

From observations in rodents, it has been suggested that the cellular basis of learning-dependent changes, detected using structural MRI, may be increased dendritic spine density, alterations in astrocyte volume, and adaptations within intracortical myelin. Myelin plasticity is crucial for neurological function, and active myelination is required for learning and memory. However, the dynamics of myelin plasticity and how it relates to morphometric-based measurements of structural plasticity remains unknown. We used a motor skill learning paradigm in male mice to evaluate experience-dependent brain plasticity by voxel-based morphometry (VBM) in longitudinal MRI, combined with a cross-sectional immunohistochemical investigation. Whole-brain VBM revealed nonlinear decreases in gray matter volume (GMV) juxtaposed to nonlinear increases in white matter volume (WMV) within GM that were best modeled by an asymptotic time course. Using an atlas-based cortical mask, we found nonlinear changes with learning in primary and secondary motor areas and in somatosensory cortex. Analysis of cross-sectional myelin immunoreactivity in forelimb somatosensory cortex confirmed an increase in myelin immunoreactivity followed by a return towards baseline levels. Further investigations using quantitative confocal microscopy confirmed these changes specifically to the length density of myelinated axons. The absence of significant histological changes in cortical thickness suggests that nonlinear morphometric changes are likely due to changes in intracortical myelin for which morphometric WMV in somatosensory cortex significantly correlated with myelin immunoreactivity. Together, these observations indicate a nonlinear increase of intracortical myelin during learning and support the hypothesis that myelin is a component of structural changes observed by VBM during learning.

Keywords: MRI; VBM; motor skill learning; mouse; myelin; neuroscience.

Figure 1.. Experimental setup and forelimb reach-and-grasp skill learning.

(A, B) Illustration (A) and MRI timeline (B) during a motor skill behavioral paradigm (SRT: skill reaching trained; NTC: nontrained controls). (C) Example of an individual in vivo T1-weighted MRI at 9.4 T at native resolution (0.1 mm isotropic, radiological display). (D) Mean performance scores during training of a skilled, single-pellet forelimb reach task, calculated as percentage (47 ± 1) of successful reaches (D) and percent (18 ± 1) accuracy (E) or percentage of successful reaches on the first attempted reach (19 ± 1) (F), during the 12-day training paradigm. Plots (D–F) represent the mean and error (SEM) for each performance score of trained subjects; n=37 for experimental days 1-6, n=35 for days 7-8, n=33 for days 9-12, n=31 for days 13-14. Non-trained controls (n=16) were evaluated only at experimental day 14 and the mean and error (SEM) for each performance score is indicated.

Figure 1—figure supplement 1.. Mean performance scores of animals used for a cross-sectional myelin immunoreactivity during learning of a skilled, single-pellet forelimb reach task.

Performance calculated as percentage (54 ± 5) of successful reaches (A) and percent (22 ± 3) accuracy (B) or percentage of successful reaches on the first attempted reach (C), during the 12-day training paradigm. Plots represent the mean and error (SEM) for each performance score of trained subjects; n=37 for experimental days 1-6, n=35 for days 7-8, n=33 for days 9-12, n=31 for days 13-14. Non-trained controls (n=16) were evaluated only at experimental day 14 and the mean and error (SEM) for each performance score is indicated.

Figure 2.. Whole-brain structural analysis revealed nonlinear decreases in gray matter volume (GMV) juxtaposed to nonlinear increases in white matter volume (WMV) with learning.

Significant changes were observed in GMV (A) and WMV (B) for which volumetric changes were modeled by three different time courses (linear model in red, asymptotic model in green, and quadratic model in blue) overlayed on the in vivo MRI template created from all subjects in this study. Decreases in volume (cold blue scale) and the increases in volume (warm red scale), in coronal sections ranging from A/P Bregma –0.1 to –0.7 mm, defined using the asymptotic model (p_{FDR corr}<0.01) and thresholded at D_AIC > 10 for asymptotic versus linear and/or quadratic models (C). Cortical and subcortical areas following an asymptotic model show decreases in GMV and increases in WMV and include primary motor cortex (MOp), primary somatosensory cortex for the forelimb (SSp-ul), and globus pallidus (GPe) contralateral to the trained limb, among others. Discrete clusters for which Akaike information criterion (AIC) values indicated a preferred quadratic model (D). Preferred quadratic clusters are observed in paramedian lobule of cerebellum (PRM), superior colliculus (SC), and main olfactory bulb (MOB) ipsilateral to the trained limb, medial septum and vermian lobule VI (DEC), and nucleus accumbens (ACb) contralateral to the trained limb.

Figure 2—figure supplement 1.. Forelimb reach-and-grasp training dynamically modulates macrostructural brain plasticity.

Training mice in the single-pellet forelimb reach task produces nonlinear decreases in gray matter volume (GMV) (p_{FDR corr}<0.001) and nonlinear increases in white matter volume (WMV) (p_{FDR corr}<0.01) (A), whereas a linear increase in GMV was observed in nontrained control animals with time (B), whole-brain statistical maps (p_{FDR corr}<0.01) are represented on a study-specific in vivo template (AP –0.1 mm, DV –3.0 mm from Bregma; 0.08 mm isotropic, radiological display).

Figure 2—figure supplement 2.. Three different regression models representing three different time courses were used to test for different patterns of change in gray matter (GM) and white matter (WM).

(A) Linear, (B) increase followed by a stabilization (inverse-quadratic-asymptotic), and (C) increase followed by a renormalization (inverse-quadratic).

Figure 2—figure supplement 3.. Whole-brain structural analysis of nonlinear decreases in gray matter volume (GMV) juxtaposed to nonlinear increases in white matter volume (WMV) with learning.

Changes in GMV and WMV modeled using the asymptotic model and overlayed on the in vivo MRI template created from this study. Whole-brain decreases in GMV (cold blue scale) and the increases in WMV (warm red scale), in coronal sections ranging from A/P Bregma –4.3 to –7.5 mm.

Figure 2—figure supplement 4.. Nonlinear increases in white matter volume (WMV) in subcortical white matter regions of the brain (p_{FDR corr}<0.05) follow an asymptotic model.

Figure 3.. Sensorimotor-restricted voxel-based morphometry (VBM) analysis in cortex identified nonlinear decreases in gray matter volume (GMV) together with nonlinear increases in white matter volume (WMV) during motor learning.

(A) VBM analysis using an atlas-based bilateral mask for primary motor cortex (MOp), secondary motor cortex (MOs), and primary sensory areas (SSp) revealed significant decreases in GMV (cold blue scale) and increases in WMV (warm red scale) following an asymptotic model (p_{FDR corr}<0.001 and <0.01, respectively). Non-linear decreases in GMV are observed together, and overlap (green), with non-linear increases in WMV. (B) 3D representations of the atlas-based mask for MOp + MOs + SSp together with the overlap of significant GMV and WMV changes (green). (C) GMV changes, relative to nontrained controls, in sensorimotor cortex contralateral to the trained forelimb extracted using a volume of interest (VOI) based on fMRI mapping of forepaw stimulation (VOI represented in green). (D) WMV changes, relative to nontrained controls, from the same VOI in (C). Plots in (C) and (D) represent the average extracted value, normalized to non-trained controls (n=16), of each VOI for every trained subject ± SEM; n=31 at 0, n=31 at 2, n=33 at 6, n=35 at 8, n=37 at 10 and n=39, at experimental day 14.

Figure 3—figure supplement 1.. Caudal forelimb area (CFA)- and rostral forelimb area (RFA)-restricted voxel-based morphometry (VBM) analysis identified nonlinear decreases in gray matter volume (GMV) together with nonlinear increases in white matter volume (WMV) during motor learning.

(A) GMV, relative to nontrained controls, in CFA contralateral to the trained forelimb (volume of interest [VOI] represented in green). (B) WMV extracted and plotted from CFA VOI in (A). (C) GMV, relative to nontrained controls, in RFA contralateral to the trained forelimb (VOI represented in red). (D) WMV extracted and plotted from the same RFA VOI in (C) Plots represent the average extracted value, normalized to non-trained controls (n=16), of each VOI for every trained subject ± SEM; n=31 at 0, n=31 at 2, n=33 at 6, n=35 at 8, n=37 at 10 and n=39, at experimental day 14.

Figure 4.. Nonlinear decreases in gray matter volume (GMV) together with nonlinear increases in white matter volume (WMV) observed during motor learning in somatosensory cortex span from layer IV through layer VIa.

(A) Normalization of the Allen Mouse Brain Atlas to the in vivo MRI template from this study allowed for an accurate location of the nonlinear changes in primary somatosensory cortex (SSp) within the different cortical layers. (B) The cortical layers disposition from the Allen Mouse Brain Atlas was confirmed using a combined staining of cells and fibers in coronal sections of mouse brain tissue. (C) Nonlinear changes in SSp were coregistered to ex vivo histological sections and confirmed observations obtained using the cortical layers from the Allen Mouse Brain Atlas. Nonlinear decreases in GMV (cold blue scale) together with nonlinear increases in WMV (warm red scale) and the overlap of significant GMV and WMV changes (green).

Figure 5.. Motor learning evokes nonlinear plasticity of cortical white matter components that are associated with structural changes.

(A) Representative image of myelin immunohistochemistry. (B) Positive correlation between cortical white matter volume (WMV) values and myelin immunoreactivity at the individual level in trained animals (Pearson’s r = 0.75, p=.03) and nonsignificant negative correlation between gray matter volume (GMV) and myelin immunoreactivity (Pearson’s r = –0.38, p=–0.35). (C) Myelin immunoreactivity increases until experimental day 6, after which it decreases toward baseline levels. The changes detected in myelin immunoreactivity follow a quadratic model rather than a linear one (AICc > 2). Data are represented as mean ± SEM (n=12 for each experimental day) (D) Illustrative representative of the three measurements acquired at the sensorimotor cortex in the same area where myelin was quantified. (E) No significant changes were observed in sensorimotor cortical thickness between trained animals and nontrained controls, (unpaired t-test; t(98)=0.5561) and no differences in the cumulative distribution of cortical thickness were observed between groups in the primary somatosensory cortex for the forelimb (SSp-ul) (F). (G) Correlation between learning rate and the asymptotic fit (measured as AICc) for WMV on an individual level (Pearson’s r = –0.378, p=0.0360) in which animals with WMV that best fit an asymptotic model (lower AICc values) exhibited higher learning rates.

Figure 5—figure supplement 1.. Experimental design for the cross-sectional immunohistochemistry analysis of myelin immunoreactivity (A) and representative image of myelin basic protein immunohistochemistry (B) highlighting the specificity of the immunodetection of myelin in cortex and striatum.

Figure 5—figure supplement 2.. Comparison of myelin immunoreactivity between trained and non-trained controls or among non-trained controls.

No differences in myelin immunoreactivity were observed between trained animals (n=12) and non-trained control animals (n=12) (A) at the last experimental day (day 14); unpaired t-test, t(22)=0.4096, alpha level of 0.05. In addition, no differences in myelin immunoreactivity were observed between non-trained control animals at experimental day 0 and at training day 12 (experimental day 14) (B); n=12, paired t-test, t(11)=0.5865, alpha level of 0.05.

Figure 5—figure supplement 3.. Higher magnification image of a representative myelin basic protein (MBP)-immunolabeled coronal section.

(A) MBP-immunolabeled coronal section at Bregma A/P –0.34. (B) Magnification of the cortical areas of the MBP-immunolabeled section in (A). (C) Enlargement of the area for which MBP immunoreactivity was quantified by densitometry. Individual myelinated axons can be appreciated.

Figure 6.. Confocal microscopy and 3D reconstruction of myelin basic protein (MBP)-labeled axons revealed nonlinear changes in myelinated axons with motor skill learning.

(A) Representative confocal images and skeleton reconstructions of myelinated axons at experimental days 0, 6, and 14; pseudo-colorized to reflect thickness. (B) Length density of myelinated axons follows a quadratic model rather than a linear one (Akaike information criterion [AIC] > 2) in which there is a 175% increase from baseline to experimental day 6, followed by a 60% decrease from experimental day 6 to experimental day 14 (one-way ANOVA, F_2,7 = 8.249, P < .05). (C) There are no significant changes in diameter of myelinated axons with learning (one-way ANOVA, F_2,7 = 1.196, alpha level of 0.05) nor in the volumetric fraction of myelinated axons in (D) (one-way ANOVA, F_2,7 = 1.748, alpha level of 0.05). Plotted values represent the mean and error (SEM), n=3–4.

Figure 6—figure supplement 1.. Confocal microscopy evaluation of myelinated axons.

(A) Confocal images (nine probes/animal) were acquired in myelin basic protein (MBP)-immunolabeled coronal sections of trained animals in the primary somatosensory cortex for the forelimb (SSp-ul) contralateral to the trained forelimb. (B) Skeleton reconstruction of myelinated axons from (A); pseudo-colorized to reflect thickness. (C) Magnification of a representative probe and its skeleton reconstruction in (D).

Figure 6—figure supplement 2.. Confocal microscopy evaluation of myelinated axons and optical zoom of generated fiber skeletons.

(A) Confocal images were acquired in myelin basic protein (MBP)-immunolabeled coronal sections and fiber skeletons were generated using Amira software and the Auto Skeleton package. (B) Optical zoom magnification of MBP immunofluorescence and the generated fiber skeletons pseudo-colorized to reflect thickness. The skeleton structure consisted of myelinated axon segments with cylinders (called nodes) spaced at 0.5 µm along the fibers. The myelinated axonal radius is calculated within the ‘Auto Skeleton’ function in Amira at each cylinder (or node) between the center of the node and the myelin boundary, taking into account the magnification factor and dimensions of the input data. See ‘Materials and methods’ for additional details.

Appendix 1—figure 1.. Figure depicting sagittal, coronal, and horizontal sections of an in vivo T1-weighted image from one individual scan during the longitudinal study (A) and the study-specific template created for mouse brain (B).

Study- and MR sequence-specific brain tissue probability maps (C). Sagittal, coronal, and horizontal sections of tissue probability maps (TPMs) of gray matter (GM), white matter (WM), and cerebrospinal fluid (CSF).