Quantitative Limbic System Mapping of Main Cognitive Domains in Multiple Sclerosis

Background and objective Cognitive impairment (CI) is common in multiple sclerosis (MS), but underlying mechanisms and their imaging correlates are not completely understood. The gray and white matter structures of the limbic system (LS) play crucial roles in different aspects of cognition. To investigate their role in MS related CI, and since a detailed evaluations are lacking in the literature, we used a comprehensive neuroimaging approach to evaluate CI’s correlations with the main components of the LS. Methods Ten non-cognitively impaired MS patients and 30 MS patients with diagnosed CI, who underwent a comprehensive neuropsychological evaluation were included in the analysis. Microstructural integrity, volumetry of main limbic gray and white matter structures and cortical thickness were assessed for associations with CI. Results Fornix and cingulum/cingulate cortices were found to be the strongest correlates of CI in MS. As expected, LS’ gray and white matter structures were involved in various cognitive functions. Uncinate fasciculi showed significant correlation with verbal and visuospatial learning and memory, phonemic and semantic fluency; hippocampi with visuospatial skills, phonemic and semantic fluency, executive functions, and processing speed; thalami with verbal learning, visuospatial skills, semantic fluency; and amygdala with verbal recognition discrimination. Conclusion This comprehensive neuroimaging approach elucidated the role of the main limbic structures in cognitive functions associated with MS-related CI.

The LS recently received more attention in the pursuit of understanding the neural correlates of abnormal cognitive symptomatology in MS (4,10). The LS's white (11,12) and gray matter structures (13,14) are found to be implicated in various cognitive symptoms in MS. Of note, compared with healthy controls, reduced cerebral blood flow was present in limbic regions of patients with CI (15). Although most studies focus on individual parts of LS, a comprehensive evaluation that included all components of the LS is lacking (see Table 1 for a summary of key MRI studies investigating limbic structures in MS).
Several studies have shown correlation between various MRI measure and histopathology (16,17,18). Diffusion tensor imaging (DTI) metrics of white matter tracts were found to correlate with a histologically derived measure of tract myelination (17,18). It is suggested that fractional anisotropy (FA) and mean diffusivity (MD) of white matter tracts are affected by axonal count in post mortem analyses in MS (18). Additionally, cortical thickness measures derived from MRI and histological measurements showed significant correlations in MS (17). While no specific postmortem DTI studies have been detailed for limbic structure damage in MS, cortical thickness (19) and diffusion measurements are a reliable in vivo technique to quantify gray and white matter injury in MS (18).
In this study, we attempted to provide a comprehensive analysis of the roles of the LS's gray and white matter structures on the main cognitive functions affected by MS. We used quantitative MRI (qMRI) methods such as cortical thickness and volumetric analysis of deep and cortical gray matter structures derived from T1-weighted (T1w) images. In addition, we used microstructural measures such as FA and MD derived from DTI. Cognitive scores were obtained from subsets of the Minimal Assessment of Cognitive Function in MS (MACFIMS), a comprehensive cognitive battery assessing various cognitive domains specifically related to MS (20).

MaTerials anD MeThODs
Ten non-cognitively impaired MS patients (MSNI) and 30 patients with diagnosed CI (MSCI) were included (36 relapsing-remitting, and 4 secondary-progressive) age 40.53 ± 11.53 (range 18-58) years, education 14.65 ± 2.33 years, disease duration 13.71 ± 8.82 years, Expanded disability status scale 3.21 ± 1.89 (0-7), left handedness 3/40 ( Table 2). Written informed consent was obtained from each subject following University of Texas Health Science Center Institutional Review Board approval of the protocol. All patients underwent the comprehensive cognitive testing as detailed below. Cognitive testing was performed in the morning to avoid fatigue, prior to and within 2 weeks of the imaging session. Inclusion criteria specified meeting 2010 McDonald Criteria for MS. Exclusion criteria included history of psychiatric disorders, recent history of drug or alcohol abuse, history of depression or relapse within 3 months of enrollment, history of allergy to gadolinium, history of other brain pathology, claustrophobia, or positive urine pregnancy test prior to MRI.

Behavioral laboratory Measures
In order to determine CI in our cohort, an impairment index methodology was applied to the behavioral scores. Based on previously validated methodology (20), 20 MACFIMS parameters were identified as most pertinent in the measurement of MS-related cognitive deficits and a CI index was derived. The processing speed and working memory was evaluated by using Paced Auditory Serial Addition Test (PASAT) (21) and symbol digit modality test (SDMT) (22), memory and learning evaluated by using California Verbal Learning Test Second Edition (CVLT-II) (23,24) and Brief Visuospatial Memory Test-Revised (25), executive function using Wisconsin Card Sorting (26), visual perception/spatial processing using Judgment of Line Orientation test (JLO) (27), and VF measured by the controlled oral word association test (COWAT) (27). Patients were classified as cognitive impaired if their perform ance was more than one standard deviation below the mean on at least 40% of the pre-identified parameters and classified as cognitively intact if performance was less than 40% impaired.   Magnetic resonance imaging Data acquisition Whole brain MRI data were acquired on a Philips 3.0T Intera scanner using a SENSE receive head coil. The MRI protocol included conventional and non-conventional MRI sequences [dual echo turbo spin echo, fluid attenuation by inversion recovery (FLAIR) and 3D T1-weighted magnetization prepared rapid acquisition with gradient echo (MPRAGE)]. The T1-weighted sequence spatial resolution was 1 mm × 1 mm × 1 mm and FigUre 2 | Illustration of fiber tractography in color-coded diffusion tensor imaging map. For fornix, we have seeded first ROI in subcallosal green fibers seen in axial slice where body and crus of fornix can be observed (a) and second and third ROIs are seeded in the mid and distal parts of crus (B) (35). For uncinate fasciculus (UF), we have seeded first ROI in the coronal section of temporal lobe and second ROI in the superomedial green projections fibers at the anterior commissure level (c) (34). For cingulum, we have seeded first ROI and second ROI in supracallosal green projection fibers anteriorly (D) and posteriorly (e). Third ROI was seeded in the green hippocampal cingulum fibers seen in the medial temporal lobe (F) (4). Once a fiber tract was reconstructed, its entire trajectory was verified on a slice-by-slice basis to compare with established anatomical landmarks described in the human brain neuroanatomy atlases (33). field-of-view was 256 mm × 256 mm. Diffusion-weighted image (DWI) data were acquired axially from the same graphically prescribed conventional MRI volumes using a single-shot multislice 2-D spin-echo diffusion sensitized and fat-suppressed echo planar imaging (EPI) sequence, with the balanced Icosa21 tensor encoding scheme (28,29). The b-factor = 1,000 s mm −2 , TR/ TE = 7,100/65 ms, FOV = 256 mm × 256 mm, and slice thickness/gap/#slices = 3 mm/0 mm/44. The EPI phase encoding used a SENSE k-space undersampling factor of two, with an effective k-space matrix of 128 × 128, and an image matrix after zero-filling of 256 × 256. The constructed image spatial resolution for the DWI data was = 1 mm × 1 mm × 3 mm.
lesion load (ll) segmentation using conventional Mri Whole brain LL was quantified in all patients using the co-registered multispectral dual FSE and the FLAIR volumes. The lesion probability masks were computed in MRIcron (http://www.nitrc. org/projects/mricron/) (30,31). The lesion volumes were saved as binary masks to enable fusion with other multimodal volumes acquired from the same subject. We obtained both T1 and T2 LL to adjust for in the correlation analyses.

Diffusion Tensor White Matter Tractography
We used a brute force and multiple regions-of-interest (ROI) tracking method and the fiber assignment with continuous tractography (FACT) algorithm (32, 33) (DTI Studio, Johns Hopkins University, Baltimore, MD, USA) to reconstruct fornix, cingulum, and uncinated fasciculus with a FA threshold of 0.15 and an angle threshold of 70°. Reproducibility of the fiber construction in both hemispheres was tested on all subjects by two experienced raters (Zafer Keser and Khader M. Hasan). We used color-coded principal eigenvector red-green-blue (RGB) map derived from DTI to seed ROIs. Multiple ROI-based deterministic tractography was used for fornix, cingulum, and uncinate fasciculus as described in Figure 2 (4, 34-37) (Figure 2).

Tissue segmentation and Parcelation
Using T1-Weighted and DTi Data Using FreeSurfer software library (version 5.3) (38), the T1-weighted brain data were automatically segmented into cerebellar, brainstem, brain, and cerebrospinal fluid, which also included thalamus, amygdale, and hippocampus (39). DTI-derived data volumes (FA, mean, axial, and radial diffusivities) were coregistered to the T1-weighted volume to obtain diffusivity values for gray matter limbic structures. In brief, all the T1-weighted data were visually inspected to rule out artifacts and input to FreeSurfer's "recon-all" routine for segmentation and extraction of morphometric measurements. FreeSurfer provided average cortical thickness using the cortical atlas labels described elsewhere (40).

statistical analyses
We used raw scores for cognitive scores. We tested normality using the Kolmogorov-Smirnov test for all reported measures. As our data couldn't assure normal distribution for all the variables likely due to small sample size, we used non-parametric Spearman correlation analyses. For the whole cohort (n = 40), individual

california Verbal learning Test ii (cVlT ii)
California Verbal Learning Test II evaluates episodic verbal learning and memory (24). In our study, the fornix, bilateral uncinated fasciculus (UF) bilateral thalami, left parahippocampus, and right posterior cingulate cortex were found to be critical regions for encoding function. For short term recall, our analyses showed significant correlations for right anterior and posterior cingulate cortices and thalamus whereas for long-term recall; bilateral thalami, and bilateral anterior and posterior cingulate cortices. Fornix was found to be crucial for both short and long term recall. Recall discriminability measures, which we recently included in CVLT version II for better recall accuracy, were shown to be related with most of the limbic gray and white matter structures bilaterally; frontal limbic cortices, bilateral anterior and posterior cingulate cortices, bilateral thalami, fornix and right uncinate and cingulum tracts. Please see Table 3 for the numeric results and Figures 3A-E for the prominent scatter plots.

The Brief Visuospatial Memory Test-revised
Brief Visuospatial Memory Test-Revised has been commonly used to evaluate visuospatial memory abilities in neuropsychological populations (25). In our MS population, visuospatial recall is found to be related to fornix, right UF, right anterior, and posterior cingulate cortices whereas learning to right cingulum, left UF, left EC. Delayed recall scores showed correlations with right UF, right anterior, and posterior cingulate cortices.
Although verbal recall discrimination is found to be carried out by more structures in LS, visuospatial recall discrimination is found to be related to bilateral anterior and posterior cingulate

controlled Oral Word association Test
Controlled oral word association test evaluates phonemic fluency with FAS and semantic fluency with animal's test (27). Previously (41), phonemic and semantic fluency are related to temporal regions. In our study, main limbic pathways connecting frontal and temporal lobes are shown to correlate with verbal phonemic and semantic fluency as well as right anterior and posterior cingulate cortices. Interestingly, bilateral thalami showed more association to semantic VF than phonemic VF ( Table 3; Figure 3F).

symbol Digit Modality Test
Symbol Digit Modality Test is a useful screening test for CI in MS and assesses attention, visual scanning, and motor speed (22), in our cohort; fornix, right cingulum, left OFC, and right posterior cingulate cortex, we found to be associated with the SDMT score ( Table 3; Figures 4A,B).

Paced auditory serial addition Test
Paced Auditory Serial Addition Test measures auditory information processing speed and flexibility, as well as calculation ability (21). All the main limbic white matter pathways, bilateral hippocampi, left PH cortex, and right thalamus were found to be associated with PASAT (Table 3; Figures 4C,D).

Benton Judgment of line Orientation (JlO)
Judgment of Line Orientation is a standardized test of visuospatial skills (27) commonly associated with functioning right parietal and occipital activation, as well as bilateral frontal activation (42). In our MS population, although it was found to be lateralized to right cingulum and UF, and to left OFC, overall it showed bilateral associations to gray and white matter limbic structures. Fornix, bilateral cingulate cortices, bilateral hippocampi, and thalami were the most prominent structures related to the scores ( Table 3).

Wisconsin card sorting Test
Wisconsin Card Sorting Test scores test attention, working memory, and visual processing. These scores measure frontal lobe functions; strategic planning, organized searching, utilizing environmental feedback to shift cognitive sets, directing behavior toward achieving a goal, and modulating impulsive responding (26). As expected, OFC showed correlation but lateralized to left. Temporal lobe structures were also found to be involved; bilateral hippocampi and entorrhinal cortices as well as left PH cortex. Bilateral cinguli and fornix as well as right thalamus showed significant correlations with the functions tested ( Table 3).

Fornix
Our data indicate forniceal white matter to be essential for multimodal cognitive functions. It might be due to its central location connecting important limbic centers.  Left cingulum was found to be associated with visual spatial memory, phonemic fluency, executive functions, and processing speed whereas right cingulum similar to fornix was associated with broader cognitive functions such as executive functions, attention motor and processing speed, phonemic and semantic fluency, visuospatial learning and skills, and verbal learning.
When it comes to parts of cingulate cortices, the isthmus of right CC showed correlation with all cognitive modalities; right anterior caudal CC with visuospatial memory and skills, verbal learning, semantic fluency; isthmus of left CC with visuospatial skills and verbal memory, and left anterior caudal CC with executive functions, visuospatial skills, and verbal learning. Posterior CC was associated with verbal learning bilaterally and executive functions on the right side.

Uncinate Fasciculus
Left UF showed significant correlation with verbal and visuospatial learning, phonemic fluency, and processing speed whereas right UF with visuospatial memory and skills, verbal learning, phonemic and semantic fluency, and processing speed.

Orbitofrontal cortex
Left lateral OFC correlated with written and oral attention, executive functions, whereas right lateral OFC with verbal learning and right medial OFC with visuospatial learning and memory.

hippocampus/Parahippocampus
Right hippocampus was microstructurally related to visuospatial skills, executive functions and processing speed, left hippocampus related to phonemic and semantic fluency, executive functions and processing speed. Bilateral parahipppocampi related to visual and verbal learning, and left parahippocampus with processing speed and executive functions.

Thalamus
Left thalamus was found to be correlated with verbal learning, visuospatial skills, semantic fluency whereas right thalamus with verbal learning, visuospatial skills, processing speed and executive functions, phonemic and semantic fluency.
amygdala Bilateral amygdala were associated with verbal recognition dis cri mi nation.

entorhinal cortex
Bilateral entorhinal cortices were correlated with executive functions and left entorhinal with verbal recognition discrimination.

DiscUssiOn
This study provided associations between LS structures including volumes and microstructure measures with various cognitive functions, both evaluated in a comprehensive manner for the purpose of a detailed cognitive mapping of limbic regions in MS patients with and without CI. Our findings are mostly in line with previous less comprehensive MS studies as can be seen in Table 1. The novelty of this study is the adoption of a comprehensive approach to the role of all major limbic structures in different cognitive functions rather than studying one or some limbic structures and their cognitive functions. Results show that many limbic gray and white matter structures are involved in various cognitive functions rather than one specific center being responsible of a specific function. We adjusted our correlations for age, LL and education, which play independent roles in cognitive functioning of patients. Overall, fornix and cingulum/cingulate cortices were found to be the strongest correlates of CI in MS. Highlights of our findings are that cingulate cortex and its main white matter pathway the cingulum, which have multiple complex functions that make its exact behavioral significance very elusive but are known to mainly serve as a modulator among different cognitive networks (7). In our cohort, similar to the literature, the cingulate cortex especially its anterior part, showed correlation with multi cognitive modalities evaluated by the MACFIMS. Interestingly, although posterior cingulate is known to be involved with visuospatial awareness and memory, our MS cohort showed association to verbal learning and executive functions. Like fornix, cingulum was associated with broader cognitive functions. A previous study revealed that cingulum DTI metrics were the best classifiers across numerous cognitive tests measuring CI in MS [(43); see also Table 1].
Although classically hippocampal formation is responsible for semantic memory formation and spatial mapping, it also plays a critical role in modulating attention, reactivity to external cues, behavioral flexibility. Previously, gray matter loss in hippocampus was shown to be related with overall CI in MS (44,45). Our analyses showed that injury to hippocampal formation led to more cognitive slowing more than verbal and visual learning. Also other studies have shown hippocampal associations with verbal memory on the left (46,47), and spatial memory on the right in MS (47). We have not observed any lateralization in our cohort except for verbal recognition discrimination lateralized to the left EC also shown in a previous study (48). Our findings for lateralization should be interpreted cautiously as we had several left handed (3) subjects in our cohort.
In contrast to limited functions of hippocampus in our MS cohort, its main output the fornix, was a predictor of multimodal CI. It is not surprising that injury to fornix leads to such widespread effect on cognition as it is one of the more prominent pathways in the CNS and serves as bridge between most limbic structures (7). Previous MS studies support our findings showing fornix to be related to various cognitive functions (4) such as visuospatial memory (10,49), verbal memory (11,49), and delayed recognition (50).
Thalamus showed widespread associations with multimodal CI in our cohort similar to previous studies (13,14). Some other studies in MS patients showed association between thalamus and attention/executive functions (12,14,45) verbal memory (12).
Lesions affecting amygdala (i.e., Kluver-Bucy syndrome) are known to cause loss of insight and inability of recognizing one's self socially inappropriate behaviors (51). In parallel, in our study, degeneration in bilateral amygdala were associated with failure verbal recognition discrimination. OFC is classically known as top-down execution modulator of the limbic networks (8). It was previously shown to be related with overall CI in MS (48). In our cohort, language-related attention and executive functions mainly lateralized to left whereas visuospatial learning and memory lateralized to right. This can be explained by the fact that the right hemisphere is more involved with visuospatial awareness, whereas the left with language related functions. As the main white matter pathway connecting OFC to LS, the correlation profile of uncinate fasciculus and CI was similar to OFC without lateralization though. Our findings are somewhat in parallel with previous studies showing the role of UF in visuospatial memory (10), processing speed (11) in MS patients. cOnclUsiOn A more detailed cognitive mapping of the LS is important in understanding functional neuroanatomy of common cognitive domains affected by MS. It also provides a non-invasive, in vivo marker for possibly assessing cognitive recovery and designing more targeted treatment approaches for CI. To the best of our knowledge, our study is the first to have comprehensive cognitive mapping of the LS in an MS cohort, previous studies have limited correlations between specific structures and cognitive domains. Our cross-sectional study is limited by a relatively small sample size, lack of healthy subjects, and heterogeneity of the MS groups. Also most of the significant correlations in our study did not survive multiple comparisons possibly due to small sample size; however, false positivity cannot be ruled out. Longitudinal prospective studies with MS patients enrolled soon after diagnosis and monitored by MRI and cognitive testing for progression would more clearly reveal the role of LS in the MS related CI. I Larger clinical studies with adequate sample size will provide further exploration in the role of LS and other brain structures in MS, such studies are in the planning phase.

eThics sTaTeMenT
Written informed consent was obtained from each subject following University of Texas Health Science Center Institutional Review Board approval of the protocol. acKnOWleDgMenTs We wish to thank Vipul Kumar Patel for helping in data acquisition.

FUnDing
This study was funded by a K-23 (the project # K23 NS072134) training award to FN. DUNN research foundation to KH.