Introduction

Alternation of iron content in normal-appearing white matter (NAWM)¹ in multiple sclerosis (MS) makes the effective transverse relaxation rate $$$R_2^*$$$ particularly interesting candidate for non-invasive assessment of tissue iron content². In white matter, however, $$$R_2^*$$$ is not only affected by presence of iron but also by tissue's myelin content as well as its microstructural organization^3,4. A number of models to describe such dependency has been proposed in the literature^5,6. Adopting two models from Lee et al.⁶, our study attempts to relate $$$R_2^*$$$ with measures, sensitive to myelin concentration and orientation of myelinated axons, to more accurately assess changes in NAWM iron content.

Materials and Methods

Sixty MS patients (age 20y–60y, median age 38y; 33 female) and 60 healthy controls (age 25y–67y, median age 33; 33 female) were examined on two different sites with two identical 3T MRI systems (Siemens Prisma) using identical 20 channel head-coils.

The acquisition protocol included three 3D multi-echo GRE sequences (PDw, T1w, and MTw) used to obtain the $$$R_2^*$$$ and magnetization transfer saturation (MTS) following the multi-parameter mapping approach⁷ (FA=6°/6°/35° for PDw/MTw/T1w; TE 1–5=8.12/13.19/19.26/24.33/29.4 ms for PDw/T1w and TE 1–2=8.12/13.19 for MTw; TR=37ms; monopolar readout, bandwidth=280Hz/px, voxel size=1mm×1mm×1mm). MTS was corrected for transmit field inhomogeneities using a $$$B_1^+$$$ map (Turbo-FLASH, voxel size=3.5mm×3.5mm×3.5mm). Two diffusion-weighted scans (voxel size=1.5mm×1.5mm×1.5mm; TR=3300ms; TE=84ms) with reversed phase encoding polarities and multi-shell diffusion scheme (8, 16, 32, and 48 directions with b-values of 0, 835, 1665 and 2500s/mm2, respectively) were acquired to assess the orientation of WM fiber bundles with respect to the static magnetic field. FLAIR scan (voxel size=0.9mm×0.9mm×3.0mm) was acquired to assess subject's condition.
For each subject FLAIR, PDw, and MTw scans were rigidly registered to the T1w GRE data using Aladin from NiftyReg⁸. $$$R_2^*$$$ maps were calculated from T1w GRE acquisition via monoexponential linear fit. Diffusion acquisitions were combined, distortion-corrected using FSL^9,10. The DTI model was fitted to obtain fractional anisotropy (FA), and predominant fiber orientation was determined using constrained spherical deconvolution¹¹. Fiber orientation was computed as the angle $$$\theta$$$ between the first ODF peak and the $$$B_0$$$ direction. FA and $$$\theta$$$ were transformed to the multi-echo T1w space based. White matter segmentation was obtained with Freesurfer¹² based on the first echo of T1w GRE data. From it, NAWM mask was obtained by subtracting MS lesions, identified by a radiologist based on FLAIR images. NAWM mask was additionally eroded with a sphere of radius 2mm. To exclude crossing fibers, voxels with FA<0.5 and the second peak of the ODF grater than 50% of the first peak were removed from the mask (see Figure 1).

Two models of $$$R_2^*$$$ orientation and/or myelin concentration dependency from⁶ were adopted in the study. Model 1 related $$$R_2^*$$$ with orientation of WM fibers:
$$R_{2}^{*} = a_{0} + a_{1}\cos^{2}\theta + a_{2}\cos^{4}\theta,$$
whereas model 2 included fiber orientation and MTS ($$$\delta$$$):
$$R_{2}^{*} = b_{0} + \delta(b_{1} + b_{2}\cos^{2}\theta + b_{3}\cos^{4}\theta).$$
Data were binned and models were fitted to median $$$R_2^{*}$$$ (model 1: $$$\Delta\theta=1^\circ$$$, model 2: $$$\Delta\theta\times\Delta\delta\approx 2^{\circ}\times 0.002$$$).
Mann-Whitney U test was applied to resulted distributions across the two groups. The significance is reported along with Glass' effect size $$$\Delta$$$.

Results

Figures 2 and 3 illustrate the fit of models 1 and 2 to the data from a representative subject. Model 1 captures most of the data variance ($$$R^2=0.98$$$), and model 2 responds to weakly MTS relatively weakly.
Figure 4 shows fitted parameters alongside with median $$$R_{2}^{*}$$$ across the entire ROI. Of all fitting parameters, $$$a_0$$$ and $$$b_0$$$ can be interpreted as global $$$R_2^{*},$$$ corrected for orientation anisotropy and contribution from myelinated axon correspondingly. Similarly to global median $$$R_2^{*}$$$, they appear to be decreased in MS group (see Table 1) which is consistent with previously reported decrease in iron content of NAWM¹.

Discussion

As can be seen from effect size in Table 1, when model 1 factors orientation-dependency out of global $$$R_2^{*},$$$ residual orientation-independent transverse relaxivity $$$a_{0}$$$ shows bigger separation between the groups. One possible explanation can be better sensitivity of $$$a_0$$$ to decreased iron content in NAWM, which is otherwise overshadowed in $$$R_2^{*}$$$ by contributions from myelin.
As can be seen from Figure 3, inclusion of MTS does not substantially change the behavior of the corresponding model parameters, which agrees with the fact that the data in Figure 2 show only weak dependency on MTS. It might suggest that MTS may lack sensitivity to myelin concentration, which is necessary to explain $$$R_2^{*}$$$ variation in WM, as MWF did in⁶. Such lack of sensitivity may explain lower fit quality as reflected in lower $$$R^2$$$ values for model 2 (see Figure 4) and consequently lower effect size.

Conclusion

Combination of multiple modalities of MRI offers a possibility to untangle contributions to $$$R_2^{*}$$$ from such sources as iron and myelin, potentially increasing specificity to pathological changes in either of the parameters. Particularly, this work suggests that accounting for $$$R_2^{*}$$$ anisotropy in WM based on concomitant diffusion acquisition may increase sensitivity to the decrease in iron content in NAWM in MS.
Despite previously shown success in increasing homogeneity of $$$R_2^{*}$$$ maps relating it with myelin content⁶, current work suggests that MTS might not be as suitable for the purpose as previously reported MWF.

Acknowledgements

The study was supported by the German Research Foundation (DFG, DE2516/1-1, RE1123/21-1) and the Austrian FWF (I 3001-B27).