Mapping microglia and astrocyte activation in vivo using diffusion MRI

Abstract

While glia are increasingly implicated in the pathophysiology of psychiatric and neurodegenerative disorders, available methods for imaging these cells in vivo involve either invasive procedures or positron emission tomography radiotracers, which afford low resolution and specificity. Here, we present a noninvasive diffusion-weighted magnetic resonance imaging (MRI) method to image changes in glia morphology. Using rat models of neuroinflammation, degeneration, and demyelination, we demonstrate that diffusion-weighted MRI carries a fingerprint of microglia and astrocyte activation and that specific signatures from each population can be quantified noninvasively. The method is sensitive to changes in glia morphology and proliferation, providing a quantitative account of neuroinflammation, regardless of the existence of a concomitant neuronal loss or demyelinating injury. We prove the translational value of the approach showing significant associations between MRI and histological microglia markers in humans. This framework holds the potential to transform basic and clinical research by clarifying the role of inflammation in health and disease.

Fig. 1.. Histological characterization of microglia reaction and its associated MRI signature.

(A) Experimental scheme showing bilateral stereotaxic injection of LPS (left hemisphere)/saline (right hemisphere) and the composition of the four groups: Six animals were scanned 8 hours after injection, 11 animals were scanned 24 hours after injection, 7 animals were treated with PLX5622 for 7 days before the injection and then scanned 24 hours after injection, and 5 animals were scanned 15 days or more after injection. (B) Normalized change (P_injected − P_control)/P_control in process density, cell size, and process dispersion parameter for the injected versus control hippocampus, measured in Iba-1⁺–stained microglia for the different groups. Asterisks represent significant paired difference between injected and control (**P < 0.01 and ***P < 0.001). Error bars represent SD. IHC, immunohistochemistry. (C) Morphology reconstruction of representative microglia at the different times. (D) Geometrical model used for microglia. (E) Normalized change (P_injected − P_control)/P_control between MRI parameter calculated in the injected versus control hemisphere for the microglia compartment. Asterisks represent significant paired difference between injected and control (*P < 0.05 and ***P < 0.001). (F and G) Correlations between stick fraction from MRI and process density from Iba-1 at 8 (F) and 24 hours after injection (G). (H) Mean stick fraction maps at 24 hours after injection, normalized to a rat brain template and averaged over all rats.

Fig. 2.. Histological characterization of astrocyte reaction and its associated MRI signature.

(A) Morphology reconstruction of representative astrocytes at the different times in black and two-dimensional (2D) convex hull in orange. (B) Geometrical model used for astrocytes. (C) Normalized change (P_injected − P_control)/P_control in convex hull mean radius for the injected versus control hippocampus, measured in GFAP⁺-stained astrocytes for the different groups. Asterisks represent significant paired difference between injected and control (***P < 0.001). Error bars represent SD. (D) Normalized change (P_injected − P_control)/P_control between MRI-derived large sphere radius calculated in the injected versus control hemisphere for the astrocyte compartment (shown in the inset). Asterisks represent significant paired difference between injected and control (*P < 0.05). (E) Correlation between mean sphere radius from MRI and convex hull mean radius from GFAP. (F) Large sphere radius maps at 24 hours after injection, normalized to a rat brain template and averaged over all rats.

Fig. 3.. Characterization of inflammation in the presence of neuronal death.

(A) Normalized change (P_injected − P_control)/P_control in histological measures for the injected versus control hippocampus. Asterisks represent significant paired difference between injected and control (*P < 0.05, **P < 0.01, and ****P < 0.0001). (B) NeuN and GFAP–Iba-1 staining of a representative animal (left, control; right, injected). GL, granular layer. (C) Normalized change (P_injected − P_control)/P_control in MRI parameter calculated in the ibotenic-injected versus control hemisphere for microglia and neuron compartments (light gray). For comparison, the same parameters obtained in group 2 of the LPS-injected animals are reported in white. Asterisks represent significant paired difference between injected and control (*P < 0.05 and **P < 0.01). (D) Normalized change (P_injected – P_control)/P_control for MRI and histological markers of neuronal death calculated separately in the untreated animals and in those treated with minocycline. Asterisks represent significant unpaired difference between the two groups (*P < 0.05 and **P < 0.01).

Fig. 4.. Specificity of glia biomarkers in demyelinated tissue.

(A) Normalized change (P_injected − P_control)/P_control in histological Myelin basic protein measures for the injected versus control hippocampus. (B) MBP staining of a representative animal (left, control; right, injected). (C) Normalized change (P_injected − P_control)/P_control in histological staining calculated in the lysolecithin-injected versus control hemisphere for microglia and astrocyte compartments. (D) Same as (C) but for MRI parameters.

Fig. 5.. Feasibility of the framework translation to human and MR parameter reproducibility analysis.

(A) Boxplot of stick fraction as measured separately in the hippocampus of six subjects scanned five times (s1 to s6) and pooling all subjects together (red). The same is shown for the stick dispersion parameter (B), small sphere radius (C), large sphere radius (D), mean diffusivity (E), and fractional anisotropy (F). (G) Average coefficient of variation calculated within subject (light gray) and between subjects (striped). (H) Region of interest (ROI) in the hippocampus used for the reproducibility analysis, defined according to (56).

Fig. 6.. Correlation between the stick fraction and microglia density in human brain.

(A) Stick fraction according to the MCM normalized to the brain template defined in (56), masked for gray matter tissue, and averaged across subjects. (B) Multiple linear regression using stick fraction and dispersion to explain microglia density measured using histological staining of postmortem human tissue in eight gray matter regions (hippocampus, cerebellum, substantia nigra, basal ganglia, thalamus, motor, frontal, and occipital cortices) as reported in (26). Regression confidence intervals are calculated using bootstrap.