Diffusion tensor imaging tractography reveals altered fornix in all diagnostic subtypes of multiple sclerosis

Abstract Introduction Diffusion tensor imaging (DTI) has shown abnormalities of the fornix and other limbic white matter tracts in multiple sclerosis (MS), mainly focusing on relapsing‐remitting MS. Methods The goal here was to evaluate the fornix, cingulum, and uncinate fasciculus with DTI tractography at 1.7 mm isotropic resolution in three MS subgroups (11 relapsing‐remitting (RRMS), nine secondary progressive (SPMS), eight primary progressive (PPMS)) versus 11 controls, and assess correlations with cognitive and clinical scores. Results The MS group overall showed extensive diffusion abnormalities of the fornix with less volume, lower fractional anisotropy (FA), and higher mean and radial diffusivities, which were similarly affected in all three MS subgroups. The uncinate fasciculus had lower FA only in the secondary progressive subgroup, and the cingulum had no DTI differences in any MS subgroup. The FA and/or volumes of these tracts correlated negatively with larger total lesion volume. The only DTI‐cognitive correlation was lower right cingulum FA and greater depression over the entire MS cohort. Conclusions Diffusion tractography identified abnormalities in the fornix that appears to be affected early and consistently across all three primary MS phenotypes of RRMS, SPMS, and PPMS regardless of Expanded Disability Status Scale, time since diagnosis, or cognitive scores.

not just limited to the clinical-MRI visible lesions in MS. A wholebrain meta-analysis of the WM "skeleton" using tract-based spatial statistics (TBSS) has shown lower fractional anisotropy (FA) in the corpus callosum, which has been a major focus of prior work due to its preferential location for MS lesions and its large size, but also in smaller tracts such as the fornix and cingulum where lower FA has been associated with greater physical disability and impaired cognition (Welton, Kent, Constantinescu, Auer, & Dineen, 2015). Separate whole-brain analyses of mostly relapsing-remitting MS (RRMS) participants have reported lower FA throughout much of the WM with FA in some specific WM regions (e.g., again including cingulum and fornix) correlating to deficits such as processing speed and working memory (Dineen et al., 2009;Kern et al., 2012;Roosendaal et al., 2009;Yu et al., 2012).
Tracts of the limbic system, like the fornix, play crucial roles in aspects of cognition, memory, behavior, and reasoning (Catani & de Schotten, 2012), and they have been implicated in the above whole-brain analyses of MS. Given this relevance, limbic system tracts have been the focus in MS diffusion MRI studies assessed with either region of interest/atlas (Dineen, Bradshaw, Constantinescu, & Auer, 2012;Koenig et al., 2013;Mesaros et al., 2012;Van Geest et al., 2018) or tractography (Fink et al., 2010;Kern et al., 2015;Keser et al., 2017;Louapre et al., 2016;Syc et al., 2013), both targeted analyses that can have distinct advantages over whole-brain voxel-based methods (e.g., far fewer multiple comparisons and do not need intersubject registration for native space tractography). The majority of these DTI studies have been in patients with RRMS, but FA abnormalities in the limbic WM tracts, including the fornix, have also been demonstrated in primary progressive (PPMS; Bodini et al., 2009) and secondary progressive MS (SPMS) (Meijer et al., 2016).
One such voxel-wise whole-brain DTI study showed that PPMS had widespread diffusion abnormalities in WM and that SPMS had more extensive WM diffusion changes than either RRMS or PPMS (Preziosa, Rocca, & Caputo, 2011).
Three key limitations can be identified from these limbic white matter DTI studies of MS. First, with the exception of the latter three studies above (Bodini et al., 2009;Meijer et al., 2016;Preziosa et al., 2011), analyses were based on mainly RRMS participants or analyses that treated all MS patients as a single cohort.
Second, the fornix is a very small tract (~3, 3.5, and 6.5 mm mean cross-sectional diameters in the posterior and anterior columns and body, respectively, even in healthy participants; Pascalau, Stănilă, Sfrângeu, & Szabo, 2018) that passes through cerebrospinal fluid (CSF) and hence is not ideal for analysis when DTI data are acquired at typical low spatial resolutions (e.g., >2 mm isotropic, voxel size >8 mm 3 ). Third, the typically used TBSS voxel-wise analysis of the WM has been shown to have particular difficulty with the FA skeleton of the fornix (Bach et al., 2014).
The purpose of this study was to (a) use DTI deterministic tractography at high field strength (4.7T) and spatial resolution of 1.7 mm isotropic to investigate whether microstructural damage to specifically the small limbic tracts (fornix, cingulum, and uncinate fasciculus) is similar in all three diagnostic subgroups of MS (RRMS, SPMS, and PPMS), and (b) correlate the DTI-derived tract diffusion metrics and volumes of the entire MS cohort with both lesion numbers/volumes and cognitive impairments (visuospatial memory speed, depression, and fatigue), disability (EDSS), and time since diagnosis.
MS lesions were segmented on FLAIR by a probabilistic lesion prediction algorithm in the Lesion Segmentation Tool (LST) MATLAB toolbox (v2.0.15) for SPM (Schmidt et al., 2012) to yield total lesion number (TLN) and volume (TLV), and all the lesion map results were visually checked. DTI was corrected for subject motion and eddy current distortions with the Artefact correction in diffusion MRI (ACID) toolbox (SPM12), and none of the datasets were discarded for excessive movement (all were <2° rotations and 2 mm translations). The ACID rotation and translation corrections for the axial, sagittal, and coronal planes for each subgroup were not significantly different between the subgroups (data not shown).  (Concha, Gross, & Beaulieu, 2005;Malykhin, Concha, Seres, Beaulieu, & Coupland, 2008;Fink et al., 2010;Syc et al., 2013;Kern et al., 2012Kern et al., , 2015Keser et al., 2017). Along-the-tract analysis also was performed in ExploreDTI bilaterally in the body of the fornix, which was the fornix tract area consistently identified across the whole MS cohort. The body was segmented by using two "AND" ROIs placed on the columns and the commissural area of the fornix just at the start of the left and right crus to guarantee streamlines connecting both ROIs. FA was calculated along 35 tract location points from anterior to posterior.

| MS subgroup characteristics, lesion numbers, and volumetric load
The control group and full MS cohort did not differ in sex (p = .74) and age (p = .38). There were differences between the three MS subgroups, namely RRMS were younger, had lower EDSS, and scored higher in PASAT, SPMS had longer time since diagnosis, and PPMS had more males than females and lower scores in PASAT. Lesion number and volumetric load were quite variable between MS participants and were not different between MS subgroups (Table 1) (Table 2).

| Qualitative assessment of the tracts
When using the spatial resolution of 1.7 mm isotropic at high magnetic field of 4.7T (Figure 1d-l), the full extent of the fornix, with high FA, was achievable by DTI tractography in 10/11 healthy controls over ages of 21-62 years. The only exception was a 75-yearold female control whose fornix tractography was "transected" at the bilateral crus, but she is 9 years older than the oldest MS volunteer. In the case of the MS cohort, only 36% of the participants   4/13 participants with <3 cm 3 total lesion volume (TLV) and 14/15 participants with more than 3 cm 3 total lesion volume showed transected fornices. There is a progressive deterioration of the tracts in the MS cohort for all the phenotypes, which is coupled with FA decrease and it could be associated with greater total lesion volume in the whole brain. In the case of the controls, tracts are not deteriorated at the same degree but slightly lower volumes and FA could be qualitatively appreciated as the controls get older showing bilateral discontinuities in the oldest control due to aging (Figure 4a).
If the FA threshold for tracking is modified from 0.2 to 0.13, qualitative tract differences could be appreciated in the MS cohort primarily, due to a greater number of streamlines arising from the lower FA threshold. In some participants, this lower FA threshold enables streamlines to connect across the previously

| Tract diffusion metrics and volumes
The Even if a lower FA threshold of 0.13 is applied in the entire healthy and MS cohort, which yields fewer "disconnections," larger volumes, higher MD, and lower FA due to the inclusion of more vox-  Figure S2).  Table 2). There are four participants (two SPMS and two PPMS) with total lesion volumes each >15 cm 3 that appear to be driving the negative correlations. Lower FA values in the right cingulum correlated with more severe depression in the entire MS cohort (Figure 8, Table 2). There are seven participants (one RRMS, three SPMS, and three PPMS) that appear to be driving the correlations with depression. They all similarly showed high fatigue scores and five of them (two SPMS and three PPMS) also showed lesion volumes >7 cm 3 .

| D ISCUSS I ON
In agreement with previous MS DTI studies primarily focused on RRMS that have implicated fornix injury (Dineen et al., 2012(Dineen et al., , 2009Fink et al., 2010;Kern et al., 2012;Keser et al., 2017;Roosendaal et al., 2009;Syc et al., 2013;Yu et al., 2012), tractography of the fornix acquired with 1.7 mm isotropic resolution showed markedly lower volume (by ~54%) and FA (by ~16%), as well as higher MD (by ~18%) and RD (by ~25%) to a similar extent in RRMS, SPMS, and PPMS relative to controls. The fornix DTI abnormalities were present in all three MS cohorts despite the disparate ranges of 21-66 years of age, 1-34 years since diagnosis, or 1.5-8.5 EDSS, implying its widespread occurrence early on in the disease course. The three following DTI studies on SPMS and/or PPMS have implicated the fornix, but this tract was not their focus nor was its 3D trajectory identified and measured using tractography (Bodini et al., 2009;Meijer et al., 2016;Preziosa et al., 2011). Whole-brain TBSS analysis yielded lower FA and higher RD of the fornix in SPMS with and without cognitive impairment, with the former showing greater changes (Meijer et al., 2016). Fornix FA was reduced in PPMS, but data were not specific for this region (Bodini et al., 2009) (Kuhlmann, Lingfeld, Bitsch, Schuchardt, & Brück, 2002). The clinical presentation does not always allow an accurate differentiation between relapsing and progressive phases in the disease spectrum (Kuhlmann, 2013). Over the entire MS group, however, the left/right fornix volume and left fornix FA were negatively correlated with total lesion volume suggesting a link between the degree of visible whole-brain lesion pathology and fornix integrity. Periventricular lesions are common in MS and are suggestive that inflammatory factors in the CSF may be involved (Matejčíková et al., 2015). The fornix would be greatly exposed to such factors since it is bathed in CSF as it passes through the lateral ventricles which could explain the abnormal fornix tractography findings even in those with brief time since MS diagnosis. White matter fornix FA has been linked to Alzheimer's disease-related CSF factors in normal adults (Gold et al., 2014). Active and chronic inac- VALDÉS CABRERA Et AL. Fink et al., 2010;Kern et al., 2012;Koenig et al., 2013;Syc et al., 2013;Yu et al., 2012).
Tract-based spatial statistics (TBSS) mentioned above in many MS DTI studies is a commonly used voxel-based analysis method that involves skeletonization of the white matter and projection of maximal FA values nearby to the skeleton, but it has been shown to have difficulties in the assessment of a small, curvy tract such as the fornix (Bach et al., 2014). Tractography has its advantages since it is performed in native space which allows anatomical variations in shape and location, and does not require spatial normalization of the images to a common template. However, a limitation of tractography is that it can lead to discontinuities of tract streamlines when the voxel FA or eigenvector angle thresholds are not met, thereby not identifying voxels that are the most severely affected. For the fornix, which is an isolated bundle without other crossing fibers, this tracking stoppage usually occurs when a voxel falls below the set threshold due to partial volume effects with isotropic CSF in the bilateral crus and leads to these apparent discontinuities even in healthy controls (Concha et al., 2005). None of the previous fornix DTI papers of MS used any CSF suppression methods-acquisition or postprocessing. Furthermore, the implementation of FLAIR-DTI was not possible at 4.7T and the single shell DTI data used here are not suitable for postprocessing "free water elimination (FWE)" CSF correction algorithms that require two shells (Hoy, Koay, Kecskemeti, & Alexander, 2014). These FWE algorithms are not still the optimal approach, compared to FLAIR-DTI, to further reduce the standard deviation of diffusion metrics for tracts with high partial volume effects with CSF, such as the fornix (Hoy, Kecskemeti, & Alexander, 2015).
Certainly, fornix tractography is susceptible to stopping its tracking for the above-mentioned reasons but this is not necessarily detrimental. Even if one assumes that the fewer tracts are just due to CSF contamination, then this would imply smaller actual fornix that would increase CSF partial volume and stop tracking. If the fornix volumes were the same, then the error ought to remain consistent in MS as in controls; but it is not the same. Thus, there will be a link between the actual volume and that which is underestimated with tractography.  (Kern et al., 2015;Yu et al., 2012) or mixed MS cohorts (Fink et al., 2010;Keser et al., 2017;Mesaros et al., 2012;Preziosa et al., 2011). The uncinate fasciculus tractography was quite variable in our MS participants; for example, in one RRMS, two SPMS, and one PPMS participant, it did not show frontal lobe projections, unlike all the healthy controls. These variations may be related either to lesions in frontal cortex areas that this tract connects (Bodini et al., 2009) or to other parts of its pathway that could affect the tracking. The uncinate fasciculus volume and FA also showed significant correlations with total lesion volume.
The cingulum did not show any group diffusion differences relative to controls in agreement with prior work (Kern et al., 2012;Roosendaal et al., 2009) Keser et al., 2017;Mesaros et al., 2012;Preziosa et al., 2011;Syc et al., 2013;van Geest et al., 2018;Yu et al., 2012). In our study, lower FA in the right cingulum correlated with higher depression scores, possibly due to interactions between the anterior cingulate and the amygdala (Bubb, Metzler-Baddeley, & Aggleton, 2018). This finding is consistent with lower cingulum FA in young women at risk of depression (Keedwell et al., 2012), and lower FA in the right parahippocampal cingulum in those with catechol-O-methyltransferase (COMT) gene polymorphisms in major depressive disorder (Seok et al., 2013). Lower FA in cingulum has correlated with deficits of episodic and working memory (Dineen et al., 2009;Mesaros et al., 2012;Syc et al., 2013;Yu et al., 2012) and subjective fatigue    (Baron & Beaulieu, 2015), but this was not used in the present study. Furthermore, the small sample size of this study, particularly when looking at each RRMS, SPMS, and PPMS subgroup separately, is another limitation to consider when making inferences about differences between these clinical groups and it should be addressed in future studies. The heterogeneity of the MS subgroups, the high participant variability of the cognitive, depression and fatigue test scores, and the burden of cognitive impairment altogether with the small sample size of the MS cohort may have contributed to the inability to find more cognitive and clinical associations with the DTI findings.

| CON CLUS IONS
In summary, high-resolution diffusion MRI tractography identified abnormalities in the fornix that were similar across all three primary MS phenotypes of RRMS, SPMS, and PPMS regardless of EDSS, time since diagnosis, or cognitive scores. The fornix FA and tract volumes were more affected with greater total whole-brain lesion volumes suggesting a link to other brain pathology that needs further investigation.