Whole-brain ex-vivo quantitative MRI of the cuprizone mouse (#12310)

11 Myelin is a critical component of the nervous system and a major contributor to contrast in Magnetic Resonance (MR) images. However the precise contribution of myelination to multiple MR modalities is still under debate. The cuprizone mouse is a well established model of demyelination that has been used in several MR studies, but these have often imaged only a single slice and analysed a small region of interest in the corpus callosum. We imaged and analyzed the whole brain of the cuprizone mouse ex-vivo using high-resolution quantitative MR methods (multi-component DESPOT, Diffusion Tensor Imaging and Tensor Based Morphometry) and found changes in multiple regions, including the corpus callosum, cerebellum, thalamus and hippocampus. However the presence of inflammation, confirmed with histology, presents difficulties in isolating the sensitivity and specificity of these MR methods to demyelination using this model. 12


Review guidelines
2 Please in full read before you begin How to review When ready submit your review online. The review form is divided into 5 sections. Please consider these when composing your review: Speculation is welcome, but should be identified as such.
The Disease. Non-invasive methods to quantify the myelination state of the nervous system are hence highly 28 useful in order to better track the progression of these diseases, and any protective or regenerative 29 treatments that become available Dubessy et al. (2014). 30 Myelin is also a uniquely useful structure for MRI as it contributes to almost every known contrast  comparison of the sensitivity and specificity of DTI to relaxometry across the whole brain.

72
To summarise, the aims of this experiment were: The bSSFP images were acquired with TE/TR=3/6ms, a band-width of 62.5kHz and also 12 flip-

114
The MR Images were first converted to NIFTI format from the manufacturer's proprietary format, and   was then calculated from the SPGR data using the B1 map to correct the flip-angles. Because we 124 acquired multiple flip-angles (required for mcDESPOT), we used a non-linear Levenberg-Marquadt 125 algorithm instead of the common linearization method to fit the data. An initial value of 1s was 126 chosen for T 1 and no issues were observed with convergence to local minima.

154
The mcDESPOT processing produces ten separate parameter maps (ignoring the B 0 and B 1 parameter 155 maps that correct for field inhomogeneities). However the MWF, IEWF and FWF are defined as fractions 156 that must sum to one, and so are not independent parameters. Hence of these only the MWF was used for 157 statistical analysis. Of the remaining parameters the myelin water residence time τ m could potentially 158 be an indicator of myelin sheath integrity. However, as will be shown below the current mcDESPOT 159 methodology cannot reliably fit this parameter, so we did not analyse it further.

160
The following procedure was then used to split the images into individual subjects and register them   3. All subjects were non-linearly registered to the study templates using their FSE and FA images.

175
Logarithmic Jacobian determinants were calculated from the inverse warp fields in standard space 176 to estimate apparent volume change. The transforms from native to study template, and from study 177 to standard space were concatenated and applied to all relaxometry and DTI parameter maps to 178 align them to the template. These images were resampled to match the voxel size of the template 179 using a Gaussian interpolator. The FWHM of the interpolator was set to 100µm for the relaxometry 180 data and 125µm for the DTI, due to their differing acquisition voxel sizes. inverse transforms from the atlas to the study template and from the study template to each subject 183 were applied to calculate the brain volume of each subject.

237
At the end of treatment the mean weight of the control and cuprizone groups were 27.5 ± 2.5g and 238 21.8 ± 1.2g respectively, which were significantly different when assessed with a two-tailed T-test 239 (p = 0.0004). However the mean brain volumes were 387.8 ± 10.1mm 3 and 381.5 ± 9.0mm 3 , which was 240 not a significant difference (p = 0.26).  The myelin T1&2 maps are fairly flat across the brain, indicating that the fitting routine finds fairly 249 consistent values for these parameters. The exception to this is that the T2 of myelin water in the internal 250 capsule appears to be lower than that found in the corpus callosum. The T2 of the IE-water shows 251 some differences between white and grey matter, but less than is found in the single-component T2 map. Manuscript to be reviewed  Similarly, figure 3 shows the DTI parameters from a single subject. These appear more blurred than 256 the relaxometry maps due to the larger acquisition voxel size and interpolation FWHM, however there is 257 still contrast between white and grey matter. AD is visibly higher than RD particularly in the hippocampus.

258
FA shows good contrast between grey and white matter, especially in the internal capsule. Manuscript to be reviewed 293 Figure 9 shows a single slice of the Co-efficient of Variation (CoV) for selected parameter maps.

294
The CoV for T1 is excellent, and is less than 5% throughout almost the entire parenchyma, while T2 is 295 marginally worse. The CoV of MWF is highly region-dependent. In GM it is consistently above 10% and 296 approaches 30% is some areas. This is perhaps expected given the low (< 5%) absolute value of MWF in 297 these regions. However, even in WM tracts the CoV is generally close to 10% and does not fall below 298 5%. The CoV of τ m shows that this parameter is difficult to fit. Counter-intuitively, in GM areas the CoV 299 appears low while in WM areas it is high. However, it must be remembered that in GM areas there is 300 close to 0% MWF, so here the fitting procedure simply converges to the center of the fitting range. In

301
WM areas, where there should be sufficient MWF to fit a valid τ m , the CoV map increases indicating that 302 there is insufficient information to fit this parameter correctly.

303
For the DTI parameters the diffusivity parameters have a mostly acceptable CoV (< 10%) that 304 increases slightly in WM regions. We attribute this to partial volume effects and residual mis-registrations 305 arising from the larg voxel size in the anterior-posterior direction for the diffusion acquisition. FA has a 306 high CoV that is above 10% in much of the parenchyma. Manuscript to be reviewed

308
In this experiment we aimed to demonstrate the use of mcDESPOT in a pre-clinical model, acquire 309 and analyze MR images across the whole cuprizone brain, and compare the sensitivity and specificity 310 of multiple quantitative MR methods to demyelination. However, the presence of inflammation is a 311 significant potential confound that has not been adequately discussed in previous MR literature.

312
Validation of mcDESPOT sensitivity to myelination 313 A major aim of this study was to provide a pre-clinical validation of the MWF as measured by mcDESPOT 314 as sensitive and specific to myelination state. In this regard the study can be regarded as success, principally  Moreover, because the MWF is defined as a fraction of total water in a voxel, it is obvious that a 326 change in the absolute amount of IE-water will by definition change the MWF, although there has been 327 no change in the absolute amount of myelin water. This means that by definition the MWF can only be 328 sensitive and not specific to myelination state. As currently formulated, due to the need to normalize 329 intensities between the SPGR and bSSFP acquisitions, mcDESPOT cannot be adapted to image absolute 330 myelin and IE-water content, so further work is clearly needed in this area.

331
The final caveat is that the mcDESPOT model is difficult to fit correctly, and this has been remarked Manuscript to be reviewed    values to be sensitive to the fitting ranges used, in particular for τ M , the residence time of water in myelin.

337
As described above, this parameter is meaningless in GM regions and simply converges to the center of 338 any chosen fitting range.

339
In white matter, where the residence time is well defined, we found values below 50 ms. This is 340 significantly shorter than that reported for human studies. This could be attributed to species differences, or 341 the process of paraformaldehyde fixation, which is known to disrupt biological membranes and introduce  from inflammation to changes in the MR parameters. 379 We believe this study is the first to estimate brain volumes and apparent volume change in the 380 cuprizone mouse. We found no difference in total brain volume, despite confirming that cuprizone causes

388
The lack of change in FA (discussed further below) was beneficial for this study as this parameter 389 could be used to improve the registration process. Initially, we followed standard practice and used only   Making sense of these conflicting reports is hampered by differences in experimental procedures.

419
Some caution should be applied to interpretations comparing diffusion imaging from in-vivo and ex-vivo the worst. They found that T2 measurements performed significantly worse than diffusivities, while we 450 found it performed slightly better. We attribute this to the increased number of phase increments in our 451 bSSFP acquisition, which we found markedly decreased banding artefacts.

453
This experiment demonstrates that T1&2, the Myelin Water Fraction, and diffusivities are sensitive to 454 demyelination but not necessarily specific, due to confounding co-localized inflammation in the cuprizone 455 model. Fractional Anisotropy appears insensitive to myelination state. In addition we found that cuprizone 456 causes localized volume changes in the mouse brain. Collectively these results show that whole brain 457 acquisition and analysis is crucial to full understanding of the cuprizone model. We propose that similar 458 methods would be beneficial when using MRI to study other preclinical models of neurodegeneration to 459 better understand and refine the knowledge of brain pathology.