Serum Neurofilament Light Chain Levels are Associated with Lower Thalamic Perfusion in Multiple Sclerosis

Both perfusion-weighted imaging (PWI) measures and serum neurofilament light (sNfL) chain levels have been independently associated with disability in multiple sclerosis (MS) patients. This study aimed to determine whether these measures are correlated to each other or independently describe different MS processes. For this purpose, 3T MRI dynamic susceptibility contrast (DSC)–PWI and single-molecule assay (Simoa)-based sNfL methods were utilized when investigating 86 MS patients. The perfusion measures of mean transit time (MTT), cerebral blood volume (CBV), and cerebral blood flow (CBF) were derived for the normal-appearing whole brain (NAWB), the normal-appearing white matter (NAWM), the gray matter (GM), the deep GM (DGM), and the thalamus. The normalized CBV and CBF (nCBV and nCBV) were calculated by dividing by the corresponding NAWM measure. Age- and sex-adjusted linear regression models were used to determine associations between the DSC–PWI and sNfL results. False discovery rate (FDR)-adjusted p-values < 0.05 were considered statistically significant. A greater age and thalamic MTT were independently associated with higher sNfL levels (p < 0.001 and p = 0.011) and explained 36.9% of sNfL level variance. NAWM MTT association with sNfL levels did not survive the FDR correction. In similar models, a lower thalamic nCBF and nCBV were both associated with greater sNfL levels (p < 0.001 and p = 0.022), explaining 37.8% and 44.7% of the variance, respectively. In conclusion, higher sNfL levels were associated with lower thalamic perfusion.


Introduction
Multiple sclerosis (MS) is an inflammatory disease of the central nervous system (CNS) that presents as variable levels of demyelination, neurodegeneration, and axonal loss [1]. Moreover, MS is linked The MS patients were classified into relapsing-remitting MS (RRMS) and progressive MS (PMS). Due to sample size limitations, primary-progressive MS (PPMS) and secondary-progressive MS (SPMS) patients were merged into a single PMS group. A standardized questionnaire was used to collect additional data regarding the patient's body mass index (BMI) and disease-modifying treatment (DMT). The study was approved by the University at Buffalo Institutional Review Board (IRB ID 030-603069, last update 2/26/2020) and all patients signed a written informed consent form.

MRI Acquisition and Analysis
All study subjects underwent a 3T MRI protocol with a Signa Excite HD 12 Twin-Speed scanner (GE, Milwaukee, WI, USA) equipped with an eight-channel head and neck coil. For this analysis, a high-resolution 3-dimensional (3D) T1-weighted imaging (WI) sequence, a 2D fluid-attenuated inverse recovery (FLAIR)-T2 sequence, and a DSC-PWI sequence were utilized. In particular, the 3D T1-WI used a spoiled-gradient recalled (SPGR) protocol with an echo time (TE), an inversion time (TI), and a repetition time (TR) of 2.8 ms, 900 ms, and 5.9 ms, respectively; isotropic voxel of 1 × 1 × 1 mm 3 ; field of view (FOV) of 25.6 × 25.6 cm 2 ; flip angle (FLIP) of 10 • . A 2D T2-FLAIR sequence was constructed using a TE, a TI, and a TR of 120 ms, 2100 ms, and 8500 ms; the same isotropic voxel size of 1 × 1 × 3 mm 3 ; an FOV of 25.6 × 25.6 cm 2 ; an FLIP of 90 • . A 2D T1-WI spin echo (SE) was used with a TE/TR of 16 ms/600 ms, an FLIP of 90 degrees, an FOV of 25.6 × 25.6 cm 2 , and a voxel size of 1 × 1 × 3 mm 3 . Lastly, the DSC-PWI sequence was acquired during and after 15 mL of 0.1 mM/kg gadolinium-diethylenetriamine penta-acetic acid was injected with a power injector at a speed of 5 mL/s. It utilized single-shot echo-planar imaging with a TR/TE of 2275 ms/45 ms, an FLIP of 90 • , an FOV of 25.6 × 25.6 cm 2 , an echo train length of 1, a bandwidth of 250 kHz, and a voxel size of 2 × 2 × 4 mm 3 . A total of 40 volumes were acquired.
A semi-automated thresholding/contouring method was used to determine the T1 and T2 lesion volumes (LV) on the T1-WI and T2-FLAIR sequences, respectively. Brain volumes for the whole brain (WB), white matter (WM), gray matter (GM), deep GM (DGM), and thalamus were determined with 3D T1-WI using cross-sectional Structural Image Evaluation, using Normalisation of Atrophy (SIENAX) and FMRIB's Integrated Registration and Segmentation Tool (FIRST) protocols [17]. To prevent tissue misclassification, T1 hypointensities were inpainted before segmentation [18]. By excluding the regions of interest derived from the T2 LV map, volumes of the normal-appearing (NA) WB and NAWM were also produced.
The Java Image Manipulation (JIM) Perfusion Toolkit (Xinapse Systems, version 6.0, Essex, U.K.) was utilized for calculating the PWI-derived cerebral blood volume (CBV), cerebral blood flow (CBF), and mean transit time (MTT). This procedure included a motion correction and automatic identification of the arterial input function (AIF) before deriving the CBV, CBF, and MTT maps. The selected AIF voxels were manually inspected to ensure accurate identification. The structural and perfusion-based segmentations were co-registered in the same MRI space. The mean values of the CBV, CBF, and MTT measures for the brain regions were calculated. The MTT was taken as an absolute measure in seconds (s), whereas the CBF and CBV were relative and unique for each patient as we did not have a method to quantify the T1 changes. Thus, the CBV and CBF measures were divided by the corresponding NAWM value to obtain the normalized CBV (nCBV) and CBF (nCBF) [19].

Serum Neurofilament Light Chain Analysis
Blood samples were collected and appropriately stored until the batch analysis. Anticoagulant ethylenediaminetetraacetic acid (EDTA) vials were used for the collection of blood samples (100 µL of volume) and stored at −80 • C without a prior thaw cycle. The sNfL levels were derived with a validated single-molecule array assay (Simoa, Quanterix Corporation, Lexington, MA, USA) and quantified in picograms per milliliter. The assay analytical sensitivity was at 0.32 pg/mL, with the coefficient of variation below 8%. All assay analyses were conducted at University Hospital, Basel, Switzerland as part of a larger collaborative project [20]. A description and the validation of the Simoa assay can be found elsewhere [21].

Statistical Analyses
Statistical analyses were performed using SPSS version 26.0 (IBM, Armonk, NY, USA). The data distributions of the demographic, clinical, MRI, and serum-based variables were assessed using the Kolmogorov-Smirnov test for normality. Data with normal and non-normal distributions were described as either mean (standard deviation (SD)) or median (interquartile range (IQR)), respectively. Age-and sex-adjusted multivariable linear regression models assessed the association between sNfL and PWI-based measures (MTT, nCBV, and nCBF). The models were built with two blocks; first, a enter-based step was used, which adjusted for the main effects of age and sex, and second, a stepwise-based step was used, which included perfusion measures if they significantly explained the greater sNfL level variance (significant R 2 change with the stepping method criteria with an entry F probability of 0.05 and removal of 0.1). Regression model variables of R 2 , standard error of estimate, t-statistics, standardized β, and p-values are reported. This type of analysis allowed for the identification of PWI predictors that would independently explain the additional variance in sNfL levels while correcting for the known effect of age. A false discovery rate (FDR) correction for multiple comparisons was performed using a Benjamini-Hochberg (BH) procedure. Post-hoc analyses included measures of the whole brain volume (WBV) and thalamic volume as adjusting variables in either the stepwise block or as variables in the initial adjusting block. Models with additional corrections for BMI and DMT were performed. Corrected p-values lower than 0.05 were considered statistically significant.
All DSC-PWI measures are shown in Table 1. The MTT ranged from the longest 3.52 s in the NAWM to the shortest with 3.18 s in the thalamus. Similarly, the nCBV and nCBF ratios ranged from 1.8 and 1.9 in the GM to 1.5 and 1.7 in the thalamus. The thalamic PWI measures were relatively lower in comparison to the total DGM region. Similar descriptive data regarding the CIS/RRMS and PMS subtypes are shown in Table 2. PMS patients had a significantly longer thalamic MTT when compared to the CIS/RRMS group (3.04 vs. 3.42 s, p = 0.049).  normalized cerebral blood volume, nCBF: normalized cerebral blood flow, SD: standard deviation, IQR: interquartile range. χ 2 , Student's t-test, and the Mann-Whitney U-test were used for categorical, normally distributed numerical, and not-normally distributed numerical variables, respectively. p-values lower than 0.05 were considered statistically significant and shown as bold. The MTT is shown in seconds (s). nCBF and nCBV are shown as a ratio relative to the corresponding region of the NAWM.

Associations between sNfL and DSC-PWI-Based Measures
The regression models showing the age-and sex-adjusted associations between sNfL and PWI-based measures are presented in Table 3. In addition to the sNfL variance explained by age and sex (R 2 = 0.329), both the thalamic and NAWM MTT measures provided a significant increase in R 2 to 0.369 (p = 0.005) and 0.402 (p = 0.037), respectively. In particular, a longer thalamic MTT time (indicative of lower perfusion) was associated with greater sNfL levels (standardized β = 0.648, t-statistics = 2.868, adjusted p-value = 0.011). (Figure 1) The effect of the NAWM MTT did not survive the multiple comparison correction (BH-adjusted p-value = 0.053). In the post-hoc analysis, the thalamic MTT remained a significant predictor of the sNfL level variance (standardized β = 0.679, t-statistics = 2.906, p-value = 0.005), whereas the WBV and thalamic volume were excluded. Moreover, when the thalamic volume was adjusted in the first block, the thalamic MTT still remained a significant predictor of the sNfL level variance (standardized β = 0.207, t-statistics = 2.202, adjusted p-value = 0.031).
Similar findings were seen in the regression models for nCBV and nCBF with increases in R 2 to 0.378 and 0.447, respectively (Table 3). In addition to age and sex effects, a lower nCBV of the thalamus was associated with greater sNfL levels (standardized β = −0.221, t-statistics = −2.529, p = 0.013, adjusted p-value = 0.022). Correspondingly, a lower nCBF of the thalamus was also associated with greater sNfL levels (standardized β = −0.346, t-statistics = −4.188, p < 0.001, adjusted p-value = 0.001).
After further adjusting for the BMI, the DSC-PWI variables did not remain as significant factors in the analysis. The potential confounding effects of BMI on the PWI-based and sNfL measures are discussed further below. After adjusting only for the effects of age, sex, and DMT, the thalamic PWI-based measures remained statistically significant (r = 0.248, p = 0.025 for the MTT; r = −0.423, p < 0.001 for the nCBF; and r = −0.27, p = 0.014 for the nCBV). sNfL: serum neurofilament light chain, DSC-PWI: dynamic susceptibility contrast-perfusion-weighted imaging, MTT: mean transit time, NAWM: normal-appearing white matter, nCBV: normalized cerebral blood volume, nCBF: normalized cerebral blood flow, SE: standard error, BH: Benjamini-Hochberg. Each sNfL level linear regression model was built with a first block that force-entered and corrected for the effects of age and sex. The second stepwise block determined which PWI-based measure provided additional and significant explanatory power. The sNfL level was normalized using a natural logarithmic transformation. The false discovery rate (FDR) for multiple comparisons utilized the Benjamini-Hochberg procedure. Variables with a significant effect on the sNfL level are shown in bold. entered and corrected for the effects of age and sex. The second stepwise block determined which PWI-based measure provided additional and significant explanatory power. The sNfL level was normalized using a natural logarithmic transformation. The false discovery rate (FDR) for multiple comparisons utilized the Benjamini-Hochberg procedure. Variables with a significant effect on the sNfL level are shown in bold. The scatterplot demonstrates the association between the sNfL levels with the standardized residuals derived from the three regression predictors (age, sex, and thalamic mean transit time). This scatter plot is equivalent to an age-and sex-adjusted partial correlation between the sNfL level and thalamic MTT. The scatterplot demonstrates the association between the sNfL levels with the standardized residuals derived from the three regression predictors (age, sex, and thalamic mean transit time). This scatter plot is equivalent to an age-and sex-adjusted partial correlation between the sNfL level and thalamic MTT.

Discussion
In addition to the age and sex effects, this cross-sectional biomarker study demonstrated that lower thalamic perfusion explained the significantly greater sNfL level variance. Although marginal, the PMS patients had significantly longer MTTs when compared to the CIS/RRMS counterparts.
Regardless of the PWI technique utilized, a recent systematic review has described a total of twelve different studies that investigate associations between clinical MS disability and perfusion metrics [22]. Seven out of the twelve studies describe positive findings, where three analyses demonstrated significant associations between a longer MTT and clinical disability scores (either the EDSS or the Multiple Sclerosis Severity Score) [23][24][25]. Corroborating our findings, these studies implicate pathology within the thalamus or the overall DGM region. These GM structures are considered to be major structural and functional connectivity hubs through which axons traverse and connect cortical regions [26]. Therefore, thalamic injury (through direct axonal transection, hypoperfusion, or neurodegeneration) would result in the release of free NfL into the CSF and serum. As an alternative explanation for the sNfL-perfusion association, we should also consider that the hypoperfusion may be a result of the lower energy/blood flow demand of previously damaged axons. However, sNfL levels are generally considered to be biomarkers with a relatively short temporal window in which changes relate to the acute state of axonal damage [5]. Conversely to our hypoperfusion association, these acute MS changes, commonly seen as contrast-enhancing MS lesions, are accompanied by inflammation-induced hyperperfusion [7]. Regardless of the proposed directionality in the casual relationship between hypoperfusion and sNfL levels, we demonstrated that both changes in perfusion and neurodegeneration Diagnostics 2020, 10, 685 8 of 10 may describe overlapping MS pathophysiological mechanisms. That said, without a longitudinal study, both explanations remain equally plausible.
After adjusting for the BMI and DMT, our models did not include any PWI measures as factors explaining the additional sNfL level variance. These findings can be explained by the fact that the BMI can significantly modulate both perfusion and sNfL measures and act as a mediator between them. Since sNfL levels are relatively proportional to the number of damaged axons, its serum concentration is directly dependent on the compartment denominator [27]. Therefore, sNfL levels are highly dependent on the total body blood volume, and thus, on the patient's BMI (heavier patients with greater blood volume will have lower sNfL levels when compared to lighter patients with an equivalent pathology) [27]. On the other hand, the BMI significantly modulates the patient vessel size and the CBV/CBF [28]. A similar argument can be made for the model adjustment with DMTs. Traditionally, the type of medication is prescribed according to the disease activity and disability levels, which have been associated with both biomarkers utilized in this study. Therefore, adjusting for DMT would bridge and lower the significance of the correlation.
This study has several design limitations. First, our DSC-PWI acquisition did not allow for the calculation of absolute CBV and CBF measures. We attempted to mitigate this limitation by normalizing the cortical and thalamic perfusion data with the perfusion measure of the corresponding NAWM region. However, the degree of NAWM pathology can vary across individual MS patients and between the MS subtypes. Due to ethical considerations, the use of gadolinium-based contrast perfusion imaging additionally prevented the acquisition of such metrics in a healthy control population. Both limitations can be effectively bypassed via the future use of an ASL sequence. Lastly, the cross-sectional study design can only provide associations between the two identified disease biomarkers without any temporal connotation. Future longitudinal studies should determine whether the lower perfusion is due to already existing thalamic axonal pathology (lower need for perfusion) or whether the perfusion contributes to greater axonal damage (ischemia leading to neuronal loss).
In conclusion, decreased thalamic perfusion, measured as a longer MTT and a lower nCBV and nCBF, was associated with greater sNfL levels in a heterogeneous population of MS patients. In addition to age effects, perfusion measures explained the significantly greater sNfL level variance.