A novel application of the VERDICT diffusion imaging technique in a mouse brain glioma model for the estimation of cell density and cell size, and comparison with histological analysis

Animal model: C57BL/6 mice (n=5) were injected with 2x10⁴ GL261 mouse glioma cells into the right striatum. Imaging was initially carried out 12-14 days post injection (Day 0) with repeat scanning three days later (Day 3) to examine any microstructural changes with tumour growth.

MRI: Imaging was carried out on a 9.4T MRI scanner (Agilent, USA). Ear bars were used to secure the mice and image without respiration gating. For DWI, a pulsed-gradient spin echo echo planar imaging (EPI) sequence was used with the following parameters: TR = 2s, TE = min, DM = 64², FOV = 20mm², slice thickness = 0.5mm, slices = 5, shots = 3, averages = 2. In total, 46 diffusion weightings in 3 directions were acquired in addition to a 42 direction DTI acquisition (b = 1000 s/mm²), as detailed previously¹. Total imaging time for VERDICT was 60 minutes. After VERDICT, mice were injected with gadolinium contrast agent and slice-matched T1-weighted spin echo EPI images were acquired for drawing tumour ROIs (Fig 1c).

Image analysis: Image processing was carried out using MATLAB. DW images were normalised to their corresponding B0 image to eliminate T2 effects with variable echo times. For modelling, the signal across the tumour ROI was averaged and then fitted. The following models, based on the taxonomy of Panagiotaki et al.³, were fit to the tumour region delineated on T1-weighted post-Gd images: Ball (ADC), BallBall (IVIM), Tensor (T), ZeppelinZeppelin (ZZ), ZeppelinCylinder (ZC), BallSphere (BS), BallBallSphere (BBS), BallSphereAstrosticks (ABS), BallSphereStick (BIS). All parameters were fitted with constraints detailed by Bailey et al.⁴. To assess the best model fits to the data, mean AIC values were calculated across the five mice, including both Day 0 and Day 3 imaging sessions.

Histology: Immediately after imaging, mice (n=4) were perfuse-fixed with paraformaldehyde. Brains were sliced and stained with H&E and NeuN, a neuronal antigen (Fig 2). For histological analysis, intracellular volume fraction (fic) was estimated by segmenting the tumour in an H&E slice and applying a k-means clustering threshold: effectively, fic = stained volume/total tumour volume. Cell radius was estimated by manual segmentation of cells at 40x magnification (Fig 2a).

In DWI images, contrast between the tumour and normal brain was subtle (Fig 1b), so post-Gd T1-weighted images were used for tumour ROI segmentation (Fig 1c). Histology showed a large tumour region within the mouse brains (Fig 2b). NeuN staining showed very little neuronal presence within tumours, suggesting that VERDICT would be an appropriate model to fit to this region. Model comparison (Fig 3) showed that VERDICT (BIS: Ball = extracellular, Sphere = intracellular, Stick = vascular compartments) with all parameters unfixed had the lowest AIC value across the cohort. However, comparison (not shown) of the fitted parameters fic and cell radius with histological analysis resulted in large discrepancies. Fixing the diffusivities in the BallSphereStick model (to dBall = 3x10^-9, dSphere = 1.5x10^-9, dStick = 8x10^-9 m²/s) resulted in fic and cell radius values which more accurately reflected the histological analysis (Fig 5 – purple points). All VERDICT parameters were broadly similar (no significant differences, Fig 5 – black points) following tumour growth between Day 0 and Day 3.We present a novel application of the VERDICT MRI technique for the characterisation of the tumour microenvironment in a mouse model of glioma. The BallSphereStick model with fixed diffusivities – BIS(f) – best captured the cellular density and cellular radius of the tumour when compared with histological analysis, although fixing the extracellular diffusivity (dBall) to 3x10^-9 m²/s seems unusually high. Future experiments using a glioma therapy to induce cellular apoptosis and alter the tumour microenvironment will provide further validation of this model.