Serum neurofilament light chain levels associations with gray matter pathology: a 5‐year longitudinal study

Abstract Background Gray matter (GM) pathology is closely associated with physical and cognitive impairment in persons with multiple sclerosis (PwMS). Similarly, serum neurofilament light chain (sNfL) levels are related to MS disease activity and progression. Objectives To assess the cross–sectional and longitudinal associations between sNfL and MRI–derived lesion and brain volume outcomes in PwMS and age–matched healthy controls (HCs). Materials and Methods Forty‐seven HCs and 120 PwMS were followed over 5 years. All subjects underwent baseline and follow–up 3T MRI and sNfL examinations. Lesion volumes (LV) and global, tissue–specific and regional brain volumes were assessed. sNfL levels were analyzed using single molecule array (Simoa) assay and quantified in pg/mL. The associations between sNfL levels and MRI outcomes were investigated using regression analyses adjusted for age, sex, baseline disease modifying treatment (DMT) use and change in DMT over the follow‐up. False discovery rate (FDR)–adjusted q‐values <0.05 were considered significant. Results In PwMS, baseline sNfL was associated with baseline T1‐, T2‐ and gadolinium‐LV (q = 0.002, q = 0.001 and q < 0.001, respectively), but not with their longitudinal changes. Higher baseline sNfL levels were associated with lower baseline deep GM (β = −0.257, q = 0.017), thalamus (β = −0.216, q = 0.0017), caudate (β = −0.263, q = 0.014) and hippocampus (β = −0.267, q = 0.015) volumes. Baseline sNfL was associated with longitudinal decline of deep GM (β = −0.386, q < 0.001), putamen (β = −0.395, q < 0.001), whole brain (β = −0.356, q = 0.002), thalamus (β = −0.272, q = 0.049), globus pallidus (β = −0.284, q = 0.017), and GM (β = −0.264, q = 0.042) volumes. No associations between sNfL and MRI–derived measures were seen in the HCs. Conclusion Higher sNfL levels were associated with baseline LVs and greater development of GM atrophy in PwMS.


Introduction
Multiple sclerosis (MS) is a chronic, inflammatory, demyelinating disease of the central nervous system which presents with reoccurring and transient neurological deficits followed frequently by insidious accumulation of physical and cognitive disability. 1 Furthermore, emerging evidence identify neurodegenerative processes as one of the major contributors to long-term MS disability accumulation. 2 Therefore, there is an increasing need for establishing simple, easily accessible, and accurate biomarkers that are able to quantify neuro-axonal injury and neurodegeneration. As such, magnetic resonance imaging (MRI)-derived brain volume measures and serum-and cerebrospinal fluid (CSF)-derived neurodegenerative biomarkers have been extensively investigated. 3,4 Persistent inflammation, concurrent axonal transection, and increased oxidative stress lead to a common final pathway of neuronal apoptosis and initiation of Wallerian degeneration. 5 Thus neurodegenerative processes are substantially greater in highly connected structures (i.e. thalamus) and their reciprocal cortical areas. 6 Multiple studies have highlighted the importance of gray matter (GM) pathology as the main driver of MS disability. 7,8 A multicenter study demonstrated that atrophy of the highly connected deep GM (DGM) was the only GM-associated region associated with concurrent disability progression. 9 Moreover, thalamus atrophy, the largest structure within the DGM, has been recognized as an important biomarker of disease progression that can be utilized consistently throughout the disease and regardless of disease phenotype. 10,11 Contrarily, recent evidence suggests that an independent and pial-driven neurodegenerative pathology contributes to generalized neuronal loss and subsequent atrophy of the neighboring cortex. 12,13 Neurofilament light chain (NfL) is the main scaffolding component of the axonal cytoskeleton. Multiple neurodegenerative diseases are highlighted by damage and loss of the neuro-axonal unit, processes which result in abnormally high NfL levels in CSF and in blood. 14 Despite the relatively low serum concentrations of NfL, recent technological development of single molecule array (Simoa) assays allow for reliable quantification. 15 Such analyses were previously utilized in MS cohorts and showed good predictive ability for concurrent and future MS disability, and of global MRI changes. 16,17 Based on this background, we aimed to analyze associations between serum neurofilament (sNfL) levels with current and future neurodegenerative pathology in a heterogeneous group of persons with MS (PwMS) and age-matched healthy controls (HCs) over 5 years. In particular, we attempted to determine specific associations of sNfL with MS lesions, global brain volumes, and development of distinct pathology in the DGM, cortical and leptomeningeal structures.

Study population
The study utilized PwMS and HCs that were initially enrolled in a larger prospective, case-controlled study which examined the cardiovascular, environmental, and genetic factors in PwMS (CEG-MS). 13,18 The PwMS inclusion criteria for this substudy were: (1) being diagnosed as clinically definite MS according to the 2010-revised McDonald criteria, 19 (2) baseline age of 18-75 years, (3) both baseline and follow-up MRI examination within 30 days of the clinical visit, (4) use of the same imaging 3T protocol at baseline and follow-up, and (5) baseline and follow-up availability of serum samples within 30 days from MRI. PwMS were excluded from the study if they had (1) a clinically documented relapse or 2) steroid use within the 30 days of the study date, and 3) were pregnant or nursing mothers. On the other hand, the study inclusion criteria for the HCs were: (1) baseline age of 18-75 years, (2) not being diagnosed with current nor past neurological disease, (3) both baseline and follow-up MRI examination within 30 days of the clinical visit, (4) use of the same imaging 3T protocol at baseline and follow-up, and (5) baseline and follow-up availability of serum samples within 30 days from MRI.
Both at baseline and follow-up, the PwMS underwent a full neurological examination and Expanded Disability Status Scale (EDSS) scores were derived. 20 The heterogeneous group of PwMS was further classified as either relapsing-remitting MS (RRMS) or progressive MS (PMS) patients. The use of MS-specific disease modifying treatment (DMT) at baseline and follow-up time points was determined during the clinical visit. Due to the large number of DMT combinations, the longitudinal change was coded as either PwMS DMT switchers (different DMT at baseline and follow-up time points) or as PwMS with stable DMT (remained on baseline DMT during the entire study period). The study was approved by the University at Buffalo Institutional Review Board (IRB) and participants signed a written consent form.

MRI acquisition and analysis
The MRI examinations were performed on a 3T GE Signa Excite HD 12 Twin Speed 8-channel scanner (General Electric, Milwaukee, WI, USA) with an 8-channel head and neck (HDNV) coil. There were no major MRI hardware or software changes over the follow-up period. The specific MRI acquisition parameters are provided in the Data S1. T 1 , T 2 , and gadolinium (Gd)-enhancing lesion masks were obtained using semiautomated contouring/thresholding technique, as described elsewhere. 21 Lesion volumes (LV) were further quantified in milliliters (mL). Accrual of new/enlarging T2-LV over the follow-up period was additionally calculated. 22 Prior to the brain volume segmentation, the T 1 hypointensities were filled to avoid tissue misclassification. 23 Whole brain volume (WBV), white matter volume (WMV), gray matter volume (GMV) and cortical volume (CV) were obtained with SIENAX software (version 2.6, FMRIB, Oxford, UK). 24 The longitudinal percentage change of the WBV and the WMV, GMV, CV were calculated using SIENA 24 and SIENAX multitime point algorithms, respectively. 25 Similarly, the crosssectional and longitudinal changes of regional tissuespecific volumes of the total DGM and the specific nuclei (thalamus, caudate, globus pallidus, putamen and hippocampus) were derived using the FMRIB's Integrated Registration and Segmentation Tool (FIRST) software (FMRIB, Oxford, UK). 26 Both global and DGM volumes were normalized with the SIENAX-derived scaling factor and quantified in milliliters. The presence of leptomeningeal contrast enhancement (LMCE) was defined as post-contrast signal intensity within the subarachnoid space which was substantially higher when compared to the brain parenchyma. 27 The 3D-FLAIR images were examined with Java Image Manipulation Tool (JIM) (Version 6.0, http://www.xinapse.com) and LMCEs were evaluated, as previously reported. 13 sNfL levels analysis The blood samples were collected at the time of the MRI examination and properly stored. Later on, the samples were sent to the University of Basel where the baseline and follow-up levels of the sNfL were derived using a validated single molecule array (Simoa) assay and quantified in pg/mL. The full description of Simoa assay is published elsewhere. 15

Statistical analysis
Statistical analyses were performed using SPSS version 25.0 (IBM, Armonk, NY). The normal distribution of the data was determined using the Kolmogorov-Smirnov test. The differences between PwMS and HCs in demographic, clinical, sNfL levels, and MRI-derived outcomes were compared using v 2 test, Student's t-test, one-way analysis of variance (for parametric variables), Mann-Whitney Utest (for nonparametric variables), age-adjusted analysis of covariance (ANCOVA), and negative binomial regression accordingly. The associations between the sNfL levels and the MRI-derived lesion and brain volumes were analyzed using linear regression models where the MRI variables were set as dependent variables and the sNfL levels, age, sex, baseline DMT and change in DMT over the follow-up were set as independent variables. Regression model-derived R 2 , beta (B), standard error (SE), standardized b, and P-values were derived. All regressionmodel metrics are fully reported in the Data S1. Due to the data skewness of the sNfL and lesion volumes, the variables were logarithmically transformed before being used in the models. Due to the nature of the % change in Gd-enhancing LV variable, the data were initially transformed with zero-inflated model and Poisson loglinear statistical modeling was performed. The associations between sNfL levels and the presence of LMCE was determined using logistic regression models. False discovery rate (FDR) correction was performed using Benjamini-Hochberg procedure and adjusted P-values (hereafter presented as q-values) were calculated. Q-values <0.05 were considered statistically significant.
A secondary, post hoc analysis employed a baseline sNfL cut-off of 30 pg/mL. 28 The PwMS were divided based on sNfL levels of <30 or ≥30 pg/mL and compared with previously described methods. The MRI-derived variables were corrected by baseline demographic and clinical differences. Alternatively, we performed two additional cut-off analyses where the PwMS were labeled with "elevated sNfL levels" based on population averages and were determined as either the 97.5th or 99th percentiles for the age-specific 45year-old HC population. 16 The second analysis individually classified each subject with "elevated sNfL levels" as determined by the age-specific 95th and 97.5th sNfL reference. 16 Lastly, associations between sNfL levels and MRI-derived volumetric measures were further analyzed by additional correction for baseline Gd-LV, T1-LV accrual, and accrual of new/enlarging T2-LV. Post hoc analyses were considered significant at P < 0.05.

Demographic and clinical characteristics
Detailed demographic and clinical characteristics of the study populations are shown in Table 1. There were no significant differences between the PwMS patients and HCs in their age (48.1 vs. 44.5 years, t-test P = 0.148) and in sex ratio (F/M, 34/13 vs. 85/35, v 2 P = 1.000). Within the MS phenotype, the PMS (five primary-progressive and 31 secondary-progressive MS patients) group was older and had longer disease duration when compared to the RRMS population (56.5 vs. 44.6 years, t-test P < 0.001 and 22.6 vs. 13.4 years, t-test P < 0.001). Furthermore, there were significant differences in the annualized relapse rate (ARR) but not in the absolute change of EDSS over the follow-up period (RRMS ARR 0.219 vs. PMS ARR 0.09, negative binomial regression P = 0.008 and RRMS EDSS change of 0.4 vs. PMS EDSS change of 0.3, t-test P = 0.663). There were no differences in the follow-up period between the groups (5.5 years for HCs, and 5.5 years for PwMS, t-test P = 0.821). The particular DMT use at baseline and change in DMT over the follow-up period are also shown in Table 1.

sNfL levels in the study groups
The differences in baseline, follow-up, and longitudinal changes in sNfL levels between the HCs and the PwMS are shown in Table 1 and Figure 1. PwMS had higher median sNfL levels when compared to the HCs at both baseline and at 5-year follow-up (median 20.6 vs. 15.3 pg/mL, age-adjusted ANCOVA q = 0.002; and median 23.8 vs. 16.7 pg/mL, age-adjusted ANCOVA q = 0.002, respectively). Overall, there was no difference between the PwMS and HCs in the absolute sNfL change over the follow-up (median 2.1 vs. 2.3 pg/mL, age-adjusted ANCOVA q = 0.976).
The baseline sNfL levels were not significantly higher in the PMS group when compared to the RRMS group (median 25.8 vs. 18.1 pg/mL, age-adjusted ANCOVA q = 0.172). On the other hand, the PMS group had higher sNfL at follow-up (median 32.2 vs. 20.6 pg/mL, age-adjusted ANCOVA q = 0.028). The PMS group had numerically greater absolute longitudinal increase in sNfL levels, compared to the RRMS group (median 6.7 vs. 0.9, age-adjusted ANCOVA q = 0.612).
3D-FLAIR postcontrast imaging required for LMCE analysis was available for 88 MS patients (58 RRMS, 30 PMS). Seventeen out of 58 RRMS (29.3%) and 14 out of 30 PMS (46.7%) had at least one definite LMCE.  The differences in sNfL levels at baseline and follow-up were calculated with analysis of covariance (ANCOVA) adjusted for baseline age and utilized logarithmically transformed sNfL data. In bold are displayed significant P-values. *The differences in relapse rate was calculated using the exact count data and with negative binomial regression modeling.

Associations between sNfL and MRI-derived lesion and brain volumes
Associations between baseline and change of sNfL levels with baseline and change of MRI-derived volumes are shown in Tables 3 and 4 There was no association between baseline sNfL, longitudinal sNfL change, and any MRI-derived volumes within the HCs population (Tables 3 and 4).
The associations between sNfL levels and the MRIderived volumes were additionally examined within the RRMS and PMS groups separately and are shown in Tables S1 and S2. The aforementioned MS findings were driven mostly by associations seen in the RRMS. RRMS baseline sNfL levels were associated with baseline T 1 -, T 2 - No associations between sNfL levels and presence of LMCE were found in neither RRMS nor PMS (Nagelkerke R 2 = 0.042, q = 0.833; and Nagelkerke R 2 = 0.049, q = 0.692, respectively).

Discussion
This study demonstrates cross-sectional and mid-term longitudinal associations of sNfL levels and concurrent global, cortical, and DGM neurodegenerative pathology in a heterogeneous population of PwMS. More importantly, baseline sNfL levels were associated with the future longitudinal atrophy rate of WBV, GMV, total DGM, thalamus, putamen and globus pallidus. These findings remained significant despite correcting for the extent of Longitudinal change for MRI-lesion derived outcomes was used in mL, whereas for brain volumes, the percentage changes were used. Regression models using two blocks (block #1 correcting for age, sex, DMT use at baseline, and DMT change over the follow-up period as covariates and block #2 step-wise addition of sNfL measure) were constructed. The standardized b and P-value demonstrate the main effect of sNfL in the model. The P-value from the regression models were corrected for false discovery rate utilizing Benjamini-Hochberg procedure. Q-values <0.05 were considered significant and displayed in bold. baseline inflammatory activity, accrual of new/enlarging T2-LV and formation of new T1 lesions. Despite the significantly higher sNfL levels seen in the PMS group, the associations between sNfL and the neurodegenerative pathology were mostly driven by the RRMS patients. Lastly, PwMS with sNfL ≥30 pg/mL had lower cross-sectional volumes and greater longitudinal atrophy rate of the global and regional volumes compared to PwMS with sNfL <30 pg/mL. Greater future neurodegenerative pathology was seen within PwMS with baseline sNfLs larger than previously determined age-specific 97.5th and 99th percentile HC sNfL levels and in analysis derived on a patient-specific level. Axonal degeneration has been implicated as an important driver of disability accumulation. 29 More so, the initial demyelination of the axons promotes inefficient Regression models using two blocks (block #1 correcting for age, sex, DMT use at baseline, and DMT change over the follow-up period as covariates and block #2 step-wise addition of sNfL measure) were constructed. The standardized b and P-value demonstrate the main effect of sNfL in the model. The P-value from the regression models were corrected for false discovery rate using Benjamini-Hochberg procedure. Q-values <0.05 were considered significant and displayed in bold. energy usage, mitochondrial dysfunction, and accumulation of oxidized stress. 30 These changes promote axonal fragmentation and ultimately lead to neuronal damage. 30 The retrograde, transaxonal spread of pathology has been corroborated by the neurodegeneration seen in anatomically distinct tracts and their corresponding GM structures. 6,31 As the events of axonal retraction are not simultaneous and occur over a certain follow-up period, baseline sNfL levels are expected to be better correlated with later occurring brain atrophy 32 Our findings of higher baseline sNfL levels, higher concurrent inflammatory activity, and future WBV atrophy are in line with several previous reports. 17,33,34 Associations between baseline sNfL levels, T 2 -LV, and Gd-enhancing lesions have been demonstrated as early as at the first demyelinating event, and are able to significantly predict conversion to clinically definite MS. 34 Moreover, studies report up to 17.8% and 4.9% increase in sNfL levels for every Gd-enhancing lesion and new/enlarging T 2 lesion, respectively. 17 In a 10-year follow-up study, sNfL levels measured at the 5-year time point were associated with greater WBV loss till year 10. 33 Correspondingly, a 5-year longitudinal study showed similar sNfL effect size and associations with greater WBV atrophy measured both at 2 and 5-year follow-up. 17 While we confirm the sNfL and global brain volume findings, we further demonstrate that the aforementioned associations are specifically driven by changes within cortical and deep GM regions. In order to further delineate the potential relationship between the released sNfL and longitudinal atrophy rate, we conducted a post hoc analysis with adjustment for potential lesion-derived influence. Despite the attenuation of the relationship, baseline sNfL remained associated with both longitudinal whole brain and DGM atrophy. Albeit statistical in nature, this analysis potentially isolates the neurodegenerative axonal propagation from acute lesional sNfL release or depicts the process of primary and independent neuronal neurodegeneration. Both thalamus and putamen have been highlighted as DGM nuclei with the strongest associations between baseline sNfL levels and its longitudinal atrophy rate. Since the thalamus has broad network of afferent and efferent reciprocal connections with cortical and subcortical regions, the implicated axonal dying-back pathology would be the most apparent. Furthermore, these findings are in line with previous results which demonstrated that lesioned tract disruption is associated with specific increase in regional atrophy of the putamen. 35 Future cortical parcellation of the study population may provide greater insight regarding associations between the axonal transection, sNfL levels, and atrophy of tract-specific deep GM and cortical surface regions.
Although in our analysis globus pallidus had almost twice the atrophy rate when compared to the remaining DGM nuclei, we detected weaker longitudinal associations with the baseline sNfL levels. These differences in atrophy rate and sNfL associations can be explained by the multifactorial pathophysiological mechanisms of MS neurodegeneration. For example, a recent prospective study showed that globus pallidus and its higher iron-based magnetic susceptibility signal is highly associated with greater physical disability. 36 Therefore, the differential atrophy of the globus pallidus may be also driven by the processes of iron accumulation and oxidative stress, whereas the atrophy rate of the putamen may be mainly driven by the lesioned neuro-axonal transection and captured by sNfL levels. [37][38][39] On the other hand, the differential sNfL associations between the RRMS and PMS groups can be interpreted by recent evidence of temporal-and spatial-specific GM atrophy sequences, by a meningeal-driven neurodegeneration and by the relative lack of active inflammation within the aging PMS patients. 40,41 Certain DGM structures like the thalamus and the caudate have been shown to atrophy early on in the disease, whereas the long-standing MS disease is associated with more cortical-oriented atrophy. 41 Furthermore, PMS patients commonly present with a higher proportion of meningeal infiltrates, the tertiary folliclelike structures which provide compartmentalization of the inflammatory process. In our subanalysis, we were not able to show any cross-sectional or longitudinal associations between sNfL levels and presence of leptomeningeal pathology, as detected using LMCE on MRI. The discrepancy can be explained by the differential disruption of the brain-blood barrier (BBB) seen in the RRMS and PMS patients. 42 In addition to the lower overall inflammatory activity, the BBB is relatively intact in PMS patients, which will limit the extravasation of free NfL into the serum.
Admittedly, the 5-year follow-up study design may have contributed to certain limitations in our interpretation of the biological processes. The two time-point sampling may have been too far apart to estimate the temporal relationship between the sNfL levels and the DGM and cortical atrophy. In addition to the volumetric imaging provided in this study, use of diffusion-tensor imaging or susceptibility-based imaging would provide further understanding of the multiple overlapping neurodegenerative processes. A larger PMS sample size with better characterization of both the inflammatory (detection of cortical lesions) and the neurodegenerative changes can provide better understanding of the sNfL utility in this particular phenotype. Furthermore, to determine the differential DMT effect on the sNfL levels and the progression of MRI brain volume outcomes is of particular clinical importance, and should be further explored. The presence of cardiovascular or other comorbid etiologies may further influence the associations between the sNfL levels and the long-term clinical and MRI-derived outcomes. Lastly, determining age-specific HC sNfL levels derived from large databases would allow better classification of abnormal sNfL levels within the increasingly older MS populations. 43 In conclusion, the sNfL levels measured in a heterogeneous MS population are associated with higher inflammatory and future neurodegenerative pathology. The sNfL measurement is a convenient and easily accessible tool in determining MS patients who are at higher neurodegenerative risk.
Leppert -Critical revision of the manuscript for important intellectual content. Zuzanna Michalak -Analysis and interpretation; critical revision of the manuscript for important intellectual content. Michael G. Dwyer -Analysis and interpretation; critical revision of the manuscript for important intellectual content. Ralph HB Benedict -Study concept and design; analysis and interpretation; critical revision of the manuscript for important intellectual content; study supervision. Bianca Weinstock-Guttman -Study concept and design; analysis and interpretation; critical revision of the manuscript for important intellectual content; study supervision. Robert Zivadinov -Study concept and design; analysis and interpretation; critical revision of the manuscript for important intellectual content; study supervision. Table S2. Associations between sNfL and MRI-derived DGM volumes in RRMS and PMS subpopulations. Table S3. Associations between sNfL levels and MRIderived brain volumes in MS patients after correcting for inflammatory activity including baseline gadolinium lesion volume, accrual of T1-lesion volume and new and enlarging T2-lesion volume. Data S1. MRI acquisition parameters.