Skeletonized mean diffusivity and neuropsychological performance in relapsing‐remitting multiple sclerosis

Abstract Background Peak width of Skeletonized Mean Diffusivity (PSMD), as a novel marker of white matter (WM) microstructure damage, is associated with cognitive decline in several WM pathologies (i.e., small vessel disorders). We hypothesized that markers combining alterations in whole WM could be associated with cognitive dysfunction in relapsing‐remitting multiple sclerosis (RRMS) patients. Methods We used PSMD based on tract‐based spatial statistics (TBSS) of diffusion tensor imaging (DTI) magnetic resonance (MR) scans. We investigated RRMS patients (n = 73) undergoing interferon beta (IFN‐β) therapy. In this cross‐sectional study, we investigated the association between neuropsychological data and clinical and MRI variables: PSMD, WM hypointensities, and normalized brain volume (NBV). Results In our cohort, 37 (50.7%) patients were recognized as cognitively impaired (CI) and 36 (49.3%) patients were cognitively normal (CN). In regression analysis, PSMD was a statistically significant contributor in the California Verbal Learning Test (CVLT) list A (p = 0.04) and semantic fluency (p = 0.036). PSMD (p < 0.001, r 2 = 0.35), NBV (p = 0.002, r 2 = 2.6) and WM hypointensities (p < 0.001, r 2 = 0.40) were major contributors to upper extremity disability (9HPT) in the CN subgroup. A significant contributor in the majority of neuropsychological measures was education attainment. Conclusion We investigated PSMD as a new parameter of WM microstructure damage that is a contributor in complex cognitive tasks, CVLT performance, and semantic fluency. PSMD was a statistically significant contributor to upper extremity disability (9HPT) together with WM hypointensities and NBV. Education attainment proved to be relevant in the majority of cognitive domains. Further studies are needed to estimate PSMD relevance as a marker of CI in MS.


INTRODUCTION
Multiple sclerosis (MS) is a neuroinflammatory and neurodegenerative condition of the central nervous system (CNS), in which inflammation, demyelination, and axonal loss lead to the progression of disability. There are known biomarkers of white matter (WM) pathology in MS such as lesion volume (LV), lesion number, and persistent T1-hypointense lesions (black holes) (Sahraian et al., 2009), which correlate with functional disability (Popescu et al., 2013) and CI (Rovaris et al., 2006). However, WM pathology, detectable by MRI, only partially contributes to disability and cognitive deficits in MS patients (Rocca et al., 2015). The altered microstructure of normal appearing white matter (NAWM) and normal appearing gray matter (NAGM) is beyond the scope of conventional MRI techniques but is presumed to be associated with greater physical disability and CI (Rocca et al., 2015). MR diffusion tensor imaging (DTI)-based studies are a sensitive method for the detection of abnormalities within NAWM in MS patients (Werring et al., 1999). Numerous studies have shown increases in mean diffusivity (MD) and a reduction in fractional anisotropy (FA) associated with focal WM lesions and NAWM among patients with MS (Miller et al., 2014). There is also evidence that an alteration of the diffusion parameters may precede the formation of WM lesions (Werring et al., 2000). Furthermore, diffusion tensor MRI studies in MS exhibited additional lesion-independent substrates of CI (Roosendaal et al., 2009). Previous DTI MR studies in the MS population revealed inconsistent results. The authors used different methods of DTI analysis in cross-sectional and longitudinal studies.
In our study, we use a novel marker of the integrity of WM microstructure based on DTI and the skeletonization of WM tracts-PSMD (Baykara et al., 2016). PSMD showed clinical relevance as a biomarker of WM pathology and cognitive performance in a population of patients with small vessel disease (SVD) such as cerebral autosomal-dominant arteriopathy with subcortical infarcts and leukoencephalopathy (CADASIL), cerebral amyloid angiopathy (CAA), or Alzheimer's disease (AD) (Baykara et al., 2016;McCreary, 2020).
CI is widespread in the MS patient population (43-70%), affecting mainly a few predominant cognitive domains: Information processing efficiency, attention, episodic and long-term memory, processing speed, and executive functions including verbal fluency and word list generation (Chiaravalloti & DeLuca, 2008;Rocca et al., 2015). Previous studies have already suggested that PSMD might be a reliable marker of altered cognition in MS, especially in relation to the dysfunction of information processing speed (Vinciguerra et al., 2020). This is a cross-sectional study revealing that PSMD might be used to assess microstructural abnormalities of WM as a background of cognitive impairment in MS. In order to avoid the influence of drugs (various mechanisms of action) on the MRI diffusion parameters, unlike in other studies, herein one of the inclusion criteria was the same immunomodulatory therapy (IMT) across the whole patient sample (interferon beta [IFN-β] therapy). Similar to previous data (Vinciguerra et al., 2020), we used a combined method of TBSS "skeletonization" of WM tracts with an analysis of diffusion histograms (Baykara et al., 2016). The combination of these techniques has two main advantages: First, the reduction of DTI data regarding contamination with cerebrospinal fluid (CSF) and other nonbrain structures (Smith et al., 2006); second, the analysis of the whole WM microstructure including NAWM. The process of skeletonization is directed at the analysis of the MD of the main fiber tracts of hemispheres that overcome the contamination of whole brain MD data through CSF (Baykara et al., 2016), comprising both affected and NAWM tracts. Histogram analysis of skeletonized MD is an appropriate method for WM diffuse pathology in MS, quantifying the total disease burden, because it captures the distribution of water molecule diffusivity obtained from the center of the main WM tracts (Deary et al., 2019). Contrary to DTI measures consisting of FA and MD metrics of WM in the selected brain region of interest, PSMD reveals the integrity of the whole WM skeleton. Moreover, PSMD as an automatic method omits many postprocessing steps and subjective operation errors (Wei et al., 2019). This quantitative, easy-to-implement method reflects the disease burden that could be applied to a large sample (Baykara et al., 2016). The aforementioned advantages of PSMD prompt us to assess its utility as a radiological marker of cognitive dysfunctions in major domains, in an RRMS population, and with each patient treated with the same IMT.

Neurological tests
Neurological assessments: The Expanded Disability Status Scale (EDSS), and the Multiple Sclerosis Functional Composite (MSFC) were performed on each patient by the same neurologist experienced in MS and included: Timed-25 foot walk (T25FW)-The ambulation time (in seconds) needed to walk the distance of 25 feet. The result is the average score of two 25-foot timed walk trials (Fischer et al., 2001).
Paced auditory serial additive test (PASAT)-A measure of auditory information processing and attention: The task involves the consecutive adding of 60 pairs of digits presented in a series with 3-s intervals (each digit is added to the preceding one); the final score is the number of correct responses (Fischer et al., 2001).
9 hole peg test (9HPT)-A quantitative measure of upper extremity (arm and hand) function. The time (in seconds) necessary to insert 9 pegs into holes and remove them using one hand is measured. The result is the average score from four trials on the 9-HPT (two trials for each hand are averaged, converted to reciprocals of the mean time for each hand and then the two reciprocals are averaged) (Fischer et al., 2001).
A detailed description of these neuropsychological tests is provided in Supplement 1.
In the study population, two groups: Cognitively impaired (CI) and cognitively normal (CN) were distinguished. Patients were classified as CI when the results of at least two cognitive tests were abnormal.

MRI data acquisition
The patients underwent MRI on the same day as a clinical examination. MRI examinations were performed with a 1.5 T scanner (Siemens Magnetom Aera) using a 20-channel head/neck coil. The MRI proto-col included the standard brain protocol for MS patients (sequences: T1-weighted 3D, FLAIR 3D, DIR 3D, SWI, DWI, T2-weighted sagittal and axial orientation, T1-weighted 3D-10 min after the injection of the contrast medium) and DTI. The analysis in this project was performed with T1-weighted 3D MPRAGE sequence and DTI; the parameters were as follows: T1-weighted MPRAGE sequence (

MRI data analysis
T1-weighted images and DTI images were converted to nii format by MRI Convert (https://lcni.uoregon.edu/downloads/mriconvert). Volumes of brain structures and the cortical thickness were measured by the freely available software FreeSurfer, version 6.0 (http://surfer.nmr. mgh.harvard.edu) (Fischl et al. 1999). The standard FreeSurfer processing stream recon-all was used. Data were visually inspected.
The volumes obtained from analyses were normalized to the estimated total intracranial volume (eTIV).
The quality of diffusion tensor images was checked. The number of outliers in the model and the motion were inspected in Explore DTI software. Patients whose rotation and movement were more than 1 degree and 1 mm, respectively, were excluded from the analysis. Diffusion images were preprocessed using the FreeSurfer script TRAC-ULA (Yendiki, 2011). The processing steps included: Correction for distortions due to head motion and eddy currents, and the calculation of diffusion parameter maps (FA, MD, RD, AD). In the next step, PSMD was calculated with the PSMD tool provided at http://www.psmd-marker.com9 based on FA and MD maps. First, DTI data were skeletonized using the TBSS, part of the functional magnetic resonance imaging of the brain (FMRIB) software library (FSL) (http://www.fmrib.ox.ac.uk/fsl) and the

Statistical analysis
The mean and standard deviations or quartiles were presented for the demographic, clinical, and radiological data. Categorical  We also performed an additional univariate linear regression analysis of clinical and cognitive data and radiological measures in the two subgroups of patients: CN and CI.

Demographic and clinical characteristics
Totally, 73 patients with RRMS were studied. The group of patients with RRMS was heterogeneous, both in terms of disability (EDSS) and the duration of the disease.
We observed a statistically significant difference in education duration and cognitive test results (Supplement 3, Tables 1 and 2).

Correlation between PSMD and clinical variables
In the studied RRMS population, the mean values (Table 2)

Multivariate linear regression analysis
We performed a multivariate linear regression analysis, testing 11 models to explain the main neuropsychological data.
In model 1: The SDMT score was over 36% explained by the clinical and radiological variables used in the model but only education duration was a statistically significant variable (p < 0.001).
In models 2 and 3: The clinical and radiological variables explained only about 2% of the archived PASAT score and VFT phonological fluency score. None of the used variables were statistically significant.

In model 4: The clinical and MRI variables explained about 13%
of the archived score in VFT and semantic fluency, and education duration was a statistically significant variable (p < 0.001). In

DISCUSSION
In this study, we have used a selection of neuropsychological tests to identify their associations with PSMD in RRMS patients. The results of our study show that PSMD and WM hypointensities were significant contributors to the CVLT score (word list A generation). This result is consistent with the findings of other authors (Vinciguerra et al., 2020) evaluating PSMD as a radiological marker of cognitive dysfunction in RRMS. CVLT assesses: Learning ability, information processing capacity, attention, and memory (Elwood, 1995), revealing abnormality in many functional networks including frontal-limbic connections (Rao et al., 1984), and the thalamic-hippocampal-prefrontal circuit (Kern et al., 2015). Given that PSMD reveals the multisite disconnectivity of WM tracts, we could explain the association with CVLT processing failure. PSMD, as a parameter of the average magnitude of water molecular diffusion in the center of common WM tracts, provides information on individual variations in WM microstructure, and could better characterize the pathological process in RRMS rather than simple quantifying measures of WM LV and brain volume (Deary et al., 2019). We currently reveal PSMD is a significant contributor to semantic fluency (Supplement 4) in the CI subgroup. Semantic fluency is associated with semantic stock and the property of grouping named objects into categories (Barois et al., 2021). Structural and functional studies showed that semantic fluency is related to frontal and temporal inferior parietal lobe (Whiteside et al., 2016). VFT performance is associated with both executive function and language functioning (Whiteside et al., 2016). VFT is a reliable tool also in severe forms of MS and was established as an effective tool in assessing cognitive impairment in the MS population, with sensitivity and specificity, respectively, of 80.6% and 97.2% (Barois et al., 2021;Negreiros et al., 2008). We found currently no studies using VFT and diffusion MRI metrics in the MS population to compare results. This finding prompts us to further explore the association between verbal deficit in MS and widespread WM alterations.
Contrary to the aforementioned Vinciguerra et al.'s, 2020 study, in our cohort the main contributor of cognitive performance was education duration. This significant factor explained results achieved in such cognitive domains as: Processing speed (SDMT), attention (CCT, SDMT), semantic verbal fluency (VFT), and language learning processes and memory (CVLT-list B). Previous studies on MS suggested that educational attainment (as assessed by years of education) reduces the negative effect of structural damage on cognition in MS (Pinter et al., 2014). The protective effect of education, as a way of developing a cognitive reserve in MS was also highlighted in recent studies (Sumowski et al., 2009). Our study extends this consideration to almost all cognitive domains. Prior studies presented a high relevance of PSMD for the SDMT score; however, we did not observe such a robust association, perhaps due to the heterogeneity of our RRMS population.
We also used WCST to evaluate executive function and in our cohort a significant contributor was age. With regard to our findings, a number of previous studies proved the association between WCST and age (Rhodes, 2004). Impaired performance on WCST (Kopp et al., 2021) has previously also been associated with focal lesions involving prefrontal lobe structures (Arnett et al., 1994). Therefore, the lack of a correlation between PSMD and WCST results could be explained by the fact that we used radiological markers of global WM deterioration (PSMD) and brain atrophy (normalized brain volume [NBV]).
Another observation is that in the tested models, PASAT, contrary to SDMT, was not associated with any of the applied clinical and MRI variables, perhaps due to the lower sensitivity of PASAT, compared to SDMT in the assessment of cognitive efficiency (Sumowski et al., 2018).
In summary, we chose, for the regression models, MR variables such as PSMD, NBV, and WM hypointensities, which seem to reflect the pathological processes present in RRMS demyelination and brain atrophy (Wang et al., 2015). In our study, we observed that higher PSMD correlated with poorer performance. All mentioned markers were assumed to contribute to CI in the MS population. However, in our MS population, heterogenic in terms of disability and disease duration, educational attainment elucidated CI in the majority of evaluated cognitive domains. The results corroborate the observed lack of clear association between T1 lesion load, brain atrophy, and disability in RRMS in previous studies (Kolasa et al., 2019). In line with these results is the important role of regional cortical and subcortical GM atrophy underpinning cognitive deficit in MS (Petracca et al., 2021), which was omitted in the current study. Moreover, focal disruption of crucial functional brain networks due to axonal transection or demyelination could be critical for particular cognitive dysfunction (Arnett et al. 1994;da Silva et al., 2020;Tewarie et al., 2014 There is still a demand for future research into novel MR markers identifying subjects at risk of cognitive deterioration.

CONCLUSIONS
We investigated PSMD as a new parameter of WM microstructure damage that contributes to complex cognitive tasks, CVLT, and semantic fluency. PSMD proved also to be a relevant marker of upper extremity dysfunction. In the studied population, in the majority of cognitive domains, educational attainment was a significant contributor of CI.
Currently, the applicability of diffusion tensor MRI markers in the routine evaluation of cognitive dysfunction remains questionable due to still insufficient empirical data.

Calculations were carried out at the Academic Computer Center in
Gdańsk. This study constitutes a portion of the PhD project of M.Ch.
carried out under the supervision of B.K.

CONFLICT OF INTEREST
The authors declare that there is no conflict of interest that could be perceived as prejudicing the impartiality of the research reported.

FUNDING INFORMATION
This study was funded from the R&D -dedicated internal resources of the Medical University of Gdansk.

AUTHOR CONTRIBUTIONS
The conception and design of the work, and the receipt of funding for the

DATA AVAILABILITY STATEMENT
The data that support the findings of this study are available from the corresponding author upon reasonable request.

PEER REVIEW
The peer review history for this article is available at https://publons.