Новинка сывороточных биомаркеров с 7 кортикальными повреждениями Тесла МРТ, лептоминированием и глубоким объемом серого вещества в ранних множественных склерозах.

Introduction

Multiple sclerosis (MS) is an autoimmune disease of the central nervous system with a clinical course traditionally characterized by relapses of neurological symptoms followed by remission, and risk for progressive disability accrual. Though relapsing–remitting (RR) MS is the most common disease course, there is substantial variability in clinical course among patients 1 . The disease also carries a risk of disease worsening that occurs gradually without relapses and at rates higher than relapse-associated worsening, so called PIRA (progression independent of relapses) 2 . Thus, there is an unmet clinical need to establish better disease surrogates and biomarkers for progression, both in the blood and by neuroimaging, particularly in early disease stages, where the clinical course is unclear and treatment choice may impact future disability.

Visualization of white matter inflammatory lesions by magnetic resonance imaging (MRI) is the well-established standard by which disease diagnosis and monitoring is performed in clinical care, and in clinical trials that have led to FDA-approval of numerous disease-modifying therapies for MS. However, with the advance of ultra-high-resolution 7 Tesla (7 T) brain MRI and increasing interest in its clinical application to assess gray matter and meningeal involvement, there is new capacity to link neuroimaging, clinical status and immune signatures to better define the pathophysiology of MS. Histopathological studies of MS have shown relevance of cortical gray matter lesions, deep gray matter lesions/atrophy and meningeal inflammation in MS 3 – 8 , though only recently, due to the increased sensitivity of 7 T imaging, has MRI afforded adequate detection of cortical lesions and leptomeningeal enhancement (LME, a proposed marker of meningeal inflammation in MS) in vivo 9 – 14 . Gray matter lesions and meningeal inflammation have been shown at higher rates in progressive MS by histology 3 , 4 , 7 though are evident early in the disease course as well 5 , 6 , 8 , 9 . These observations suggest new avenues to understand processes driving disease progression. The hypothesis asserting the importance of these mechanisms of damage is supported by histological studies showing correlation between meningeal inflammation and the extent of cortical demyelination 7 . Furthermore, cortical lesions have been shown to predict disability progression independent of white matter lesions, and have a higher correlation with physical disability and cognitive impairment than lesions in the white matter 14 , 15 . In addition to lesions in the gray matter, selective deep gray matter (DGM) volume loss has been demonstrated early in the disease course, and shows strong relationships to disability progression and cognitive impairment in MS 16 , 17 . The capacity to link such MRI outcomes to biofluid markers could lead to new therapeutic targets not reflected by traditional white matter lesion volume and relapses.

The relationship between cortical gray matter lesions, atrophy and meningeal involvement is not fully understood though several mechanisms have been hypothesized. Defects in the blood-meningeal barrier or choroid plexus may allow infiltration of T cells, B cells, and macrophages into the subarachnoid space or ventricles, after which these cells have capacity to attach to the meninges or ependyma and infiltrate the cortex or deep gray matter from the CSF interface 18 . Alternatively, inflammation of the gray matter may lead to migration of proinflammatory cells into the ventricles/subarachnoid space through the breakdown of the blood-brain barrier (BBB) 18 . Of note, organized accumulations of these cells can be found in the subarachnoid space (ectopic lymphoid follicles or B cell follicles) and are detected by histology in 40% of patients with secondary progressive MS; such follicles and are thought to be mediators of chronic inflammation in MS 4 .

There remains a substantial unmet need to identify serum biomarkers with prognostic clinical utility in MS. In MS patients, the serum concentrations of prospective biomarkers (sNFL, sGFAP) have shown meaningful relationships to clinical outcomes and 3 Tesla MRI-defined measures of disease activity 19 , 20 . Due to substantially higher sensitivity of 7 T MRI (LME prevalence rates ~ 20–50% at 3 T vs. 60–90% at 7 T 9 , 10 , 12 and cortical lesion detection increased by two-fold at 7 T vs. 3 T MRI 21 ) there is new technical capability to non-invasively identify biomarker-MRI patterns relevant to clinical status and prognosis. In the present study, we therefore leverage clinical 7 T MRI and blood biomarker analysis from a large cohort study to characterize patterns of 7 T MRI-proteomic association in early-stage MS.

Methods

Patient population and clinical outcomes

We used prospectively collected patient data from the Comprehensive Longitudinal Investigation of MS at the Brigham and Women’s Hospital (CLIMB), which is a longitudinal study comprised of MS patients 22 . Patients in the CLIMB study have biannual clinical visits and annual standardized MRI scans. A large set of variables are included for each patient in a relational Oracle database. Blood is sampled at annual CLIMB visits and stored according to standardized procedures. 7 T MRI is performed in a subset of patients, whose primary neurologist deemed would benefit from an MRI with a higher field strength. All patients provided informed consent. This study was approved by the Mass General Brigham Human Research Committee. All methods were performed in accordance with the relevant guidelines and regulations.

Patients in the present study met the following inclusion criteria: (1) diagnosis of CIS or MS according to the 2010 or 2017 McDonald criteria 23 , (2) 7 T brain MRI and available serum sample, both within five years of diagnosis and not > 1.5 years apart. Patients with a motion-degraded 7 T MRI or scans lacking contrast (unable to evaluate LME) were excluded from the study. The inclusion flow chart is shown in (Fig.  1 ). The group was depicted as two cohorts to create transparency for the reader that the proteomic analyses were run in two different batches, in cohort 1 (MRI performed between 2017–2019) and in a second cohort (MRI performed between 2019–2022). Fig. 1 Consort diagram. In total, 70 CIS/MS patients were identified who had a 7 T brain MRI and a blood sample available within 5y of CIS/MS diagnosis. From these 70 patients, nine patients had to be excluded because the MRI and blood sample were > 1.5y apart, one patient had a wrong MRI protocol, two patients had poor image quality due to movement (n = 1) or skull defect (n = 1), and in one patient the images were not retrievable from PACS. Therefore, 57 patients were included in the final analysis.

Consort diagram. In total, 70 CIS/MS patients were identified who had a 7 T brain MRI and a blood sample available within 5y of CIS/MS diagnosis. From these 70 patients, nine patients had to be excluded because the MRI and blood sample were > 1.5y apart, one patient had a wrong MRI protocol, two patients had poor image quality due to movement (n = 1) or skull defect (n = 1), and in one patient the images were not retrievable from PACS. Therefore, 57 patients were included in the final analysis.

The following clinical-demographic data were assessed, scored by a clinical neurologist at the timepoint of the 7 T MRI acquisition: age, sex, MS clinical phenotype, disease duration since diagnosis, disease-modifying therapy (DMT) use, and disability score on the expanded disability status scale (EDSS).

MRI methods

MRI acquisition

The 7 T MRI scanner used for the CLIMB study is a MAGNETOM Terra scanner (Siemens; Erlangen, Germany) with a single-transmit/32-channel receive head coil and is approved for routine patient care. MRI employed a uniform scan protocol among subjects, covering the whole brain with voxel sizes 0.7 × 0.7 × 0.7 mm 3 including T1-weighted (T1w) magnetization-prepared 2 rapid gradient-echo (MP2RAGE) and 3D T2-weighted FLAIR on a Siemens MAGNETOM Terra unit. Post-contrast imaging was performed after intravenous 0.1 mmol/kg of gadoterate meglumine (Dotarem®) which commenced with repeating both sequences (MP2RAGE followed by FLAIR, equating to a ~ 10 min delay between contrast administration and FLAIR initiation). Further acquisition details have been previously published 9 .

MRI analysis

MRI outcomes included LME status (presence/absence), LME foci number, cortical lesion number and volume, thalamic lesion number and volume, normalized total DGM (nDGM) structure volumes comprising the sum of the thalamus, caudate, globus pallidus, and putamen, total T2 white matter lesion volume (T2LV), normalized whole brain parenchymal volume (nBPV), and parenchymal gadolinium (gad)-enhancing lesion number.

LME detection was performed by manual comparison of pre- and post-contrast sagittal 3D FLAIR by an expert rater (JZ) 9 . Scans were evaluated for hyperintense signal on post-contrast FLAIR with no hyperintense signal on pre-contrast FLAIR. The analysis of cortical, DGM, and white matter (WM) lesions was performed by an expert rater (JZ) assisted by a semi-automated edge-finding tool in Jim software (v.8, Xinapse Systems Ltd, West Bergholt, UK; http://www.xinapse.com ). Cortical lesions were identified on non-contrast T1-weighted (T1w)-MP2RAGE sequence, required to be hypointense on T1w-MP2RAGE versus adjacent normal-appearing cortex. Corresponding FLAIR hyperintensity frequently provided confirmation, though it was not a strict requirement. The cortical lesion rater was blinded to LME status. In a small number of cases there was a secondary review by a senior neuroradiologist (RB) to adjudicate final determination of lesions used in final analysis. Due to the limited sensitivity of 0.7 mm MP2RAGE for subpial lesions, cortical lesions were not categorized by subtype. Axial T1w-MP2RAGE and FLAIR resampled sequences were used for the detection of DGM lesions. Lesions were required to be FLAIR hyperintense, traced on the axial resampled FLAIR pre-contrast sequence. Whole-brain T2 hyperintense lesion volume was traced on sagittal 7 T FLAIR. The SIENAX 5.0 automated pipeline was used to determine nBPV from MP2RAGE images. To segment DGM, FSL-FIRST pipeline was applied to MP2RAGE images. Thalamus, caudate, putamen, and globus pallidus automated segmentations were confirmed for quality. We used the SIENAX v5.0 whole brain/head normalization factor for intrasubject standardization of raw DGM volumes for head size. The raw volumes were multiplied by the normalization factor for each subject to generate normalized volumes 24 .

Protein analysis

For this study, the following three protein panels of 112 total proteins were analyzed in the serum samples: 1) sNFL and GFAP by Quanterix™ using the SIMOA assay, 2) an 18-Plex kit provided by Bio-Techne™ and analyzed using Luminex Flexmap® by the Dana Farber Cancer Institute core facility, and 3) a 92-Plex by Olink® using Proximity Extension Assay (PEA) technology. The individual proteins are listed in (Supplementary Table 1 ).

sNfL and GFAP were selected due to their specificity to the CNS 25 . The SIMOA assay has been validated by different groups for the serological measurement of NFL and GFAP 26 , 27 .

The proteins in panel 2 were individually selected by the authors due to their influence on B cell follicles and/or their prior association with meningeal inflammation and cortical lesions in MS patients 8 , 28 – 30 . These proteins have been previously validated by the manufacturer, Bio-Techne 31 , 32 .

Panel 3 (Olink Target 96 Neurology) contains a set of preselected proteins, which are related to different neurobiological processes and diseases 33 . PEA was shown to generate technically robust results for different proteins measured in Alzheimer’s disease 34 .

Statistical analysis

MRI outcomes were assessed for correlation to the serum proteins from the three different biomarker panels in an adjusted analysis (sex, age, and disease duration) using partial Spearman’s correlation coefficient and in an unadjusted analysis using Spearman’s correlation coefficient. For LME status (presence vs. absence), the proteins were compared using a Wilcoxon rank sum test. All statistical analysis was completed in the statistical package R version 4.3.0 ( www.r-project.org ), and the partial correlations were calculated using the ppcor library 35 . Pathway analysis was performed using Ma’ayan laboratory’s gene list enrichment analysis tool ( https://maayanlab.cloud/Enrichr/ ). Proteins and MRI correlations where there were > 3 correlations with p < 0.05 in the adjusted analysis were considered in this analysis. The pathway analysis was adjusted for multiple comparisons.

Results

Cohort characteristics

Fifty-seven patients were included in the present analysis (Fig.  1 ). The baseline characteristics of the two cohorts are shown in (Table 1 ). Over 90% of patients had a relapsing–remitting disease course. Sex distribution showed female predominance (~ 3–4:1), with a similar number of anti-B-cell treated patients (~ 50%) in both cohorts. Comparatively, Cohort 1 had more untreated patients (20 vs. 7.7%) and was slightly younger (mean ± SD age: 37.8 ± 10 vs. 45 ± 10.7 years) than Cohort 2. Table 1 Baseline cohort characteristics. Baseline characteristics Cohort 1 (n = 31) § Cohort 2 (n = 26) † Age at MRI (y) 37.8 ± 10 [23.6–57] 45 ± 10.7 [21.9–65.4] Disease duration since first diagnosis (y) 1.9 ± 1.1 [0–4.1] 2.6 ± 1.4 [0–5] EDSS 1.2 ± 1.3 [0–6] 1.7 ± 1.2 [0–3.5] Time difference MRI and blood sample (d) 103 ± 87 [0–322] 166 ± 172 [0–539] LME foci number 1.6 ± 1.7 [0–6] 2.0 ± 2.7 [0–12] T2 white matter lesion volume (mL) 4.7 ± 7.4 [0.27–32.6] 2.1 ± 1.8 [0.09–8.4] Cortical lesion number 5.4 ± 9.6 [0–44] 5.3 ± 4.3 [0–15] Cortical lesion volume (mL) 0.21 ± 0.35 [0–1.5] 0.13 ± 0.14 [0–0.49] Thalamic lesion number 1.2 ± 1.6 [0–5] 0.81 ± 0.94 [0–4] Thalamic lesion volume (mL) 0.034 ± 0.064 [0–0.34] 0.015 ± 0.019 [0–0.063] Normalized brain parenchymal volume (mL) 1563 ± 89 [1339–1718] 1575 ± 86 [1401–1797] Normalized total deep gray matter volume (mL) 46.1 ± 3.6 [35.3–53.7] 46.7 ± 3.2 [41.2–54] Sex  Female, n (%) 23 (74.2) 21 (80.8) Treatment  B cell-depleting therapy, n (%) 16 (48.4) 12 (46.1)  Fingolimod, n (%) 5 (16.1) 4 (15.4)  Dimethyl fumarate, n (%) 1 (3.2) 4 (15.4)  Other, n (%) 3 (9.7) 4 (15.4)  Untreated, n (%) 6 (19.4) 2 (6.5) MS phenotype  RRMS, n (%) 28 (90.3) 25 (96.2)  PPMS, n (%) 1 (3.2) 1 (3.8)  CIS, n (%) 2 (6.5) 0 (0) MRI status  LME-positive, n (%) 18 (58.1) 15 (57.7)  Parenchymal gadolinium lesion-positive, n (%) 3 (9.7) 1 (3.9) Data shown are mean ± standard deviation [range] unless otherwise indicated. EDSS expanded disability status scale, LME leptomeningeal enhancement, RRMS relapsing–remitting MS, PPMS primary progressive MS, CIS clinically isolated syndrome. § Disease modifying treatment for Cohort 1: Ocrelizumab (15), Rituximab (1), Other DMTs: Glatiramer acetate (2), Natalizumab (1). Number of Gd + lesions per patient in Cohort 1: 1, 1, and 5. † Disease modifying treatment for Cohort 2: Ocrelizumab (11), Rituximab (1), Other DMTs: Teriflunomide (3), Secukinumab (1). The Gd + patient in Cohort 2 had 4 Gd + lesions.

Table 1 Baseline cohort characteristics. Baseline characteristics Cohort 1 (n = 31) § Cohort 2 (n = 26) † Age at MRI (y) 37.8 ± 10 [23.6–57] 45 ± 10.7 [21.9–65.4] Disease duration since first diagnosis (y) 1.9 ± 1.1 [0–4.1] 2.6 ± 1.4 [0–5] EDSS 1.2 ± 1.3 [0–6] 1.7 ± 1.2 [0–3.5] Time difference MRI and blood sample (d) 103 ± 87 [0–322] 166 ± 172 [0–539] LME foci number 1.6 ± 1.7 [0–6] 2.0 ± 2.7 [0–12] T2 white matter lesion volume (mL) 4.7 ± 7.4 [0.27–32.6] 2.1 ± 1.8 [0.09–8.4] Cortical lesion number 5.4 ± 9.6 [0–44] 5.3 ± 4.3 [0–15] Cortical lesion volume (mL) 0.21 ± 0.35 [0–1.5] 0.13 ± 0.14 [0–0.49] Thalamic lesion number 1.2 ± 1.6 [0–5] 0.81 ± 0.94 [0–4] Thalamic lesion volume (mL) 0.034 ± 0.064 [0–0.34] 0.015 ± 0.019 [0–0.063] Normalized brain parenchymal volume (mL) 1563 ± 89 [1339–1718] 1575 ± 86 [1401–1797] Normalized total deep gray matter volume (mL) 46.1 ± 3.6 [35.3–53.7] 46.7 ± 3.2 [41.2–54] Sex  Female, n (%) 23 (74.2) 21 (80.8) Treatment  B cell-depleting therapy, n (%) 16 (48.4) 12 (46.1)  Fingolimod, n (%) 5 (16.1) 4 (15.4)  Dimethyl fumarate, n (%) 1 (3.2) 4 (15.4)  Other, n (%) 3 (9.7) 4 (15.4)  Untreated, n (%) 6 (19.4) 2 (6.5) MS phenotype  RRMS, n (%) 28 (90.3) 25 (96.2)  PPMS, n (%) 1 (3.2) 1 (3.8)  CIS, n (%) 2 (6.5) 0 (0) MRI status  LME-positive, n (%) 18 (58.1) 15 (57.7)  Parenchymal gadolinium lesion-positive, n (%) 3 (9.7) 1 (3.9) Data shown are mean ± standard deviation [range] unless otherwise indicated. EDSS expanded disability status scale, LME leptomeningeal enhancement, RRMS relapsing–remitting MS, PPMS primary progressive MS, CIS clinically isolated syndrome. § Disease modifying treatment for Cohort 1: Ocrelizumab (15), Rituximab (1), Other DMTs: Glatiramer acetate (2), Natalizumab (1). Number of Gd + lesions per patient in Cohort 1: 1, 1, and 5. † Disease modifying treatment for Cohort 2: Ocrelizumab (11), Rituximab (1), Other DMTs: Teriflunomide (3), Secukinumab (1). The Gd + patient in Cohort 2 had 4 Gd + lesions.

Baseline cohort characteristics.

Data shown are mean ± standard deviation [range] unless otherwise indicated.

EDSS expanded disability status scale, LME leptomeningeal enhancement, RRMS relapsing–remitting MS, PPMS primary progressive MS, CIS clinically isolated syndrome.

§ Disease modifying treatment for Cohort 1: Ocrelizumab (15), Rituximab (1), Other DMTs: Glatiramer acetate (2), Natalizumab (1). Number of Gd + lesions per patient in Cohort 1: 1, 1, and 5.

† Disease modifying treatment for Cohort 2: Ocrelizumab (11), Rituximab (1), Other DMTs: Teriflunomide (3), Secukinumab (1). The Gd + patient in Cohort 2 had 4 Gd + lesions.

Proteomics analysis

Proteins that correlate (p ≤ 0.05) in the adjusted analysis with any MRI measure can be found in Fig.  2 a,b (bi-directional correlational plot). The results of the entire analysis (including all p- and rs-values) can be found in (Supplementary Table 2 ). Fig. 2 ( a ) Correlation between MRI measures and SIMOA/Luminex proteins. Shown are the proteins analyzed in the SIMOA/Luminex assay that correlated (p < 0.05) in the adjusted analysis with any MRI measure (bi-directional correlational plot). The analysis was adjusted for sex, age, and disease duration. Partial Spearman’s correlation coefficient was used. Positive Spearman’s correlations (r s  > 0) are shown in red, while negative correlations (r s  < 0) are shown in blue with the intensity of color depicting the strength of association. The absolute number in the respective fields indicates the p-value (not the rs-value). GFAP was analyzed using SIMOA. All other proteins were analyzed in the Luminex assay. ( b ) Correlation between MRI measures and Olink proteins. Shown are the proteins analyzed in the OLink assay that correlated (p < 0.05) in the adjusted analysis with any MRI measure (bi-directional correlational plot). The analysis was adjusted for sex, age, and disease duration. Partial Spearman’s correlation coefficient was used. Positive Spearman’s correlations (rs > 0) are shown in red, while negative correlations (rs < 0) are shown in blue with the intensity of color depicting the strength of association. The absolute number in the respective fields indicates the p-value (not the rs-value).

( a ) Correlation between MRI measures and SIMOA/Luminex proteins. Shown are the proteins analyzed in the SIMOA/Luminex assay that correlated (p < 0.05) in the adjusted analysis with any MRI measure (bi-directional correlational plot). The analysis was adjusted for sex, age, and disease duration. Partial Spearman’s correlation coefficient was used. Positive Spearman’s correlations (r s  > 0) are shown in red, while negative correlations (r s  < 0) are shown in blue with the intensity of color depicting the strength of association. The absolute number in the respective fields indicates the p-value (not the rs-value). GFAP was analyzed using SIMOA. All other proteins were analyzed in the Luminex assay. ( b ) Correlation between MRI measures and Olink proteins. Shown are the proteins analyzed in the OLink assay that correlated (p < 0.05) in the adjusted analysis with any MRI measure (bi-directional correlational plot). The analysis was adjusted for sex, age, and disease duration. Partial Spearman’s correlation coefficient was used. Positive Spearman’s correlations (rs > 0) are shown in red, while negative correlations (rs < 0) are shown in blue with the intensity of color depicting the strength of association. The absolute number in the respective fields indicates the p-value (not the rs-value).

Cortical lesions (Fig.  3 )

Proteomic correlates of cortical lesion volume and number. Only proteins correlating (p < 0.05) with both cortical lesion volume and number in the adjusted analysis are shown. The rs- and p-values of the adjusted analysis is shown in the plot. IL-15 and TNF-alpha correlated positively with both cortical lesion volume and number in the adjusted and unadjusted analysis. BAFF correlated with cortical lesion volume and number in the adjusted but not in the unadjusted analysis. FcRL2 correlated negatively with cortical lesion volume and number in the adjusted and unadjusted analysis. For the adjusted analysis partial Spearman’s correlation coefficient was used. The analysis was adjusted for sex, age, and disease duration.

Cortical lesions were found readily in both early MS cohorts (mean cortical lesion number ± SD [range]: 5.4 ± 9.6 [0–44] and 5.3 ± 4.3 [0–15]), with similar lesion volume (0.21 ± 0.35 and 0.13 ± 0.14 mL) and low lesion burden compared to prior studies of RRMS subjects 9 .

IL-15 and TNF-alpha correlated positively with cortical lesion volume and number in the adjusted and unadjusted analysis, while FcRL2 correlated negatively with these measures in the adjusted and unadjusted analysis. BAFF correlated with cortical lesion volume and number in the adjusted but not in the unadjusted analysis. GFAP, IFNg, and IL-1B did not correlate with cortical lesion number but with cortical lesion volume in the adjusted analysis but not in the unadjusted analysis. nCDase did not correlate with cortical lesion volume but was inversely correlated with cortical lesion number in the adjusted and unadjusted analysis.

LME (Fig.  4 a,b)

( a ) Proteomic correlates of LME foci number. Only proteins correlating (p < 0.05) with LME foci number in the adjusted analysis are shown. PLXNB3 is positively and nCDase negatively correlated to LME foci number in the adjusted und unadjusted analysis. CNTN5 is negatively correlated to LME foci number in the adjusted but not in the unadjusted analysis. For the adjusted analysis partial Spearman’s correlation coefficient was used. The analysis was adjusted for sex, age, and disease duration. PLXNB3 Plexin B3, CNTN5 Contactin-5, nCDase neutral ceramidase. ( b ) Proteomic correlates of LME presence. Only proteins correlating (p < 0.05) with LME foci number in the adjusted analysis are shown (proteins from Fig. 4a). Higher PLXNB3 levels are associated with LME presence and higher nCDase levels as well as higher CNTN5 levels with LME absence. The protein levels were compared using a Wilcoxon rank sum test. PLXNB3 Plexin B3, CNTN5 Contactin-5, nCDase neutral ceramidase.

LME prevalence for the full cohort was n = 33/57 (58%) with a similar burden of LME in the two cohorts—mean ± SD [range] LME foci number was 1.6 ± 1.7 [0–6] and 2.0 ± 2.7 [0–12] for Cohorts 1 and 2, respectively.

PLXNB3 correlated positively with LME foci number and was associated with LME presence, while nCDase inversely correlated with LME foci number and was associated with the absence of LME. CNTN5 inversely correlated with LME foci number in the adjusted but not in the unadjusted analysis and was associated with the absence of LME. G.CSF was inversely correlated with LME foci number in the unadjusted but not in the adjusted analysis, while being associated with LME absence. CDH3 was associated with the absence of LME but did not correlate with LME foci number. Higher sNfL was associated with the presence of LME but did not correlate with LME foci number.

DGM structural volumes (Fig.  2 a,b)

IL-1B was inversely correlated with normalized thalamic, putaminal, and total DGM volume in the adjusted and unadjusted analysis. CDH6 was positively correlated with normalized thalamic, globus pallidus, and total DGM volume in the adjusted and unadjusted analysis. SIGLEC9 was positively correlated with normalized caudate and total DGM volume in the adjusted and unadjusted analysis. HAGH positively correlated with normalized putaminal and globus pallidus volume in the adjusted and unadjusted analysis. PLXNB3 correlated positively with globus pallidus volume in the adjusted but not in the unadjusted analysis. IL21 was inversely correlated with total DGM volume in the adjusted but not in the unadjusted analysis. Normalized putaminal volume was inversely correlated with the following biomarkers in the adjusted and unadjusted analysis: BAFF, IL10, MAPT, and MSR1. GDNF correlated positively with the normalized putaminal volume in the adjusted and unadjusted analysis.

Thalamic lesions

Approximately 54% of subjects had at least 1 thalamic lesion though lesion burden was low, with mean TL number ± SD: 1.2 ± 1.6 [0–5] and 0.81 ± 0.94 [0–4] and mean TL volumes 0.034 ± 0.064 and 0.015 ± 0.019 mL for Cohorts 1 and 2, respectively. The mean thalamic lesion volume among subjects from either cohort having at least 1 lesion was 0.047 ± 0.060 mL. There were no correlations in the adjusted analysis between TL number or volume and any of the tested proteins.

Other MRI outcomes (Fig.  2 a,b)

Normalized brain parenchymal volume correlated with NCAN and GZMA in the adjusted and unadjusted analysis.

Only 4 subjects in the cohort were found with parenchymal gadolinium-enhancing lesions. sFRP.3 correlated inversely with gadolinium-enhancing lesions in the adjusted and unadjusted analysis, while MATN3 correlated inversely only in the adjusted but not the unadjusted analysis.

The following proteins correlated positively with T2LV in the adjusted and unadjusted analysis: IL21, PRTG, RGMA, CD38, LAYN, GFR.alpha.1, SCARA5, NTRK2, PDGFR. alpha, JAM.B, and EDA2R. The following only correlated positively in the adjusted analysis but not the unadjusted analysis: GFAP, IL-1B, CADM3, VWC2, NBL1, GDNFR.alpha.3, and N2DL.2.

Pathway analysis (Fig.  5 and Supplementary Table 3 )

Pathway analysis of cortical lesion volume. The results of the pathway analysis for cortical lesion volume is depicted. Pathway analysis was performed using Ma’ayan laboratory’s gene list enrichment analysis tool ( https://maayanlab.cloud/Enrichr/ ). Proteins and MRI correlations where there were > 3 correlations with p < 0.05 in the adjusted analysis were considered in this analysis. The rheumatoid arthritis and the cytokine-cytokine receptor interaction pathway generated the highest number (n = 5) of proteins and adjusted p-values from the entire pathway analysis (same hits for both pathways: BAFF, IFNg, IL-1b, IL-15, TNF-alpha). The results of the pathway analysis are provided in (Supplementary Table 3 ).

Overall, cortical lesion volume demonstrated the highest number of associations with different pathways as well as the lowest p-values and highest amount of hits (proteins) per pathway. Specifically, the rheumatoid arthritis and the cytokine-cytokine receptor interaction pathway generated the highest number (n = 5) of proteins and lowest adjusted p-values from the entire pathway analysis (same hits for both pathways: BAFF, IFNg, IL-1b, IL-15, TNF-alpha). The rest of the results of the pathway analysis can be found in Supplementary Table 3 and are not further presented and discussed in the main manuscript.

Discussion

This exploratory analysis of the association between serum biomarkers and 7 T brain MRI outcomes demonstrates a number of interesting and clinically relevant findings. We confirmed previous findings that 7 T MRI is highly sensitive for detection of gray matter lesions and LME and can be used to segment DGM in early MS. The demonstration of 7 T MRI leptomeningeal and cortical abnormalities at relatively high rates in early MS (~ 55–60% prevalence rate of LME and mean ~ 5 cortical lesions per early RRMS subject) though lower than the rates observed in heterogeneous MS cohorts, or in progressive MS, supports the hypothesis that these findings evolve along a spectrum and are important to investigate in relation to potential serum biomarkers. Of all the biomarker-MRI associations studied, the strongest and most consistent correlations were between cortical lesions and higher IL-15, TNF-alpha, and BAFF levels, and lower levels of FcRL2. We report novel associations between LME foci number and higher levels of PLXNB3 and lower levels of nCDase and CNTN5. Higher IL-1B levels correlated with lower DGM nuclei volumes while higher levels of CDH6, SIGLEC9, and HAGH correlated with higher DGM nuclei volumes. Though cortical demyelination, meningeal inflammation, and DGM atrophy are associated with the progressive MS stage, the extent to which they correlate with each other may vary between patients, and biomarker signatures from this early MS cohort suggest unique immunology may underlie the development of each of these factors.

Recent work using 3 T MRI has shown novel prognostic associations between CSF OPN and CXCL13 (among other biomarkers) and risk of disability accumulation and accrual of cortical atrophy in early relapsing MS, within four years of diagnosis 36 . However, few studies to-date have attempted to characterize 7 T-biomarker association, and none that have specifically examined early MS. For purposes of this discussion, we aim to review the 7 T-biomarker associations found in the present work in the context of current understanding of immune mechanisms and known or speculated clinical relevance to MS.

Cortical lesions

We demonstrated that higher cortical lesion burden was associated with higher serum levels of IL-15, TNF-alpha, and BAFF. Higher FcRL2 (Fc receptor-like 2) levels were associated with lower cortical lesion burden.

Saikali et al. showed that IL-15 is secreted by astrocytes in MS lesional sites and only to a very small degree in normal brain parenchyma 37 . IL-15 promotes cytotoxic T cell migration through the BBB into the brain while also enhancing germinal B cell proliferation 29 , 37 . There are several hypothesized mechanisms by which B cells may influence cortical lesion formation, making a link biologically plausible.

TNF (tumor necrosis factor) alpha is a well-known, pro-inflammatory cytokine implicated in the pathogenesis of MS 38 . Interestingly, seminal work from Magliozzi et al. found increased expression of TNF subreceptors in the meninges and CSF of post-mortem MS cases with high levels of meningeal inflammation and GM demyelination, which was substantiated by in-vivo findings from the CSF of MS patients with severe GM damage 8 .

BAFF (B-cell activating factor) is also known as TNF-ligand superfamily member 13B of CD257. Ligands for BAFF include TNFRSF13B/TACI, TNFRSF17/BCMA, and TNFRSF13C/BAFF-R. BAFF is a fundamental B cell survival factor required for germinal center maintenance 39 . Magliozzi et al. similarly identified elevated levels of BAFF in the CSF of MS patients with higher levels of GM damage 8 . And of note, after intrathecal anti-B-cell (rituximab) treatment, BAFF levels were reported to fall in the CSF but increase in the serum suggesting an intercompartmental migration of BAFF in MS 40 .

FcRL2, which showed negative association to cortical lesion burden, is an immunoglobulin receptor that has an inhibitory role in B-cell activation via interaction with regulatory B cell subsets and has been shown to be downregulated in MS patients with high neurodegeneration 41 , 42 . Due to the central involvement of B cells in cortical lesion formation and FcRL2’s inhibitory effect, an inverse correlation is biologically plausible, and suggests a possible protective role of FcRL2.

The pathway analysis identified an association between cortical lesion volume and both the rheumatoid arthritis and cytokine-cytokine receptor interaction pathways, with overlapping inflammatory proteins (BAFF, IFNg, IL-1b, IL-15, TNF-alpha). While these findings highlight the role of cytokine-driven inflammation in MS, they do not necessarily imply a direct mechanistic overlap with rheumatoid arthritis. Instead, they support the broader involvement of pro-inflammatory signaling in neuroinflammation.

Leptomeningeal enhancement

LME is postulated to be a surrogate marker for meningeal inflammation in MS, though its relevance in early MS is unclear 43 . Histological work has established that meningeal inflammation and meningeal B-cell follicles are evident in early disease 5 , 6 , though there is not an established literature exploring its relationship to biomarkers. In the present study, we found that LME presence and foci burden were linked to higher concentrations of PLXNB3 (Plexin-B3). The absence of LME and lower LME foci burden were associated with higher nCDase (neutral ceramidase) and CNTN5 (contactin-5) levels.

PLXNB3, while not having a known role in autoimmunity, functions as a receptor for semaphorin 5A, where it plays a role in axon guidance, invasive growth, and cell migration 44 . It is expressed in many tissues including the brain, heart and intestines, as well as monocytes in the immune system and hypoxia-induced signaling of PLXNB3 can promote cancer tumorigenesis 44 . Given this association, one plausible hypothesis would be that PLXNB3 may play a role immune cell migration into the subarachnoid space and leptomeninges in MS.

nCDase is an enzyme that hydrolyzes ceramide into sphingosine 45 . It is expressed ubiquitously, though at high rates in the intestine. nCDase knockout mice demonstrate defective IL-1B processing via the inflammasome/caspase-1 pathway, indicating macrophages use sphingosine to sense immune pathogens in the intestine 46 . Thus, nCDase may play a role in promoting monocyte/macrophage phenotypes that are protective. CNTN5 is an immunoglobulin cell adhesion molecule that is exclusively expressed in the brain, especially in the cortex, hippocampus, dorsal thalamus, and auditory pathway 47 . It has a pivotal role in synaptogenesis and the maintenance of neuronal circuitry in the cortex 47 . There is little known about CNTN5’s role outside of brain development and no known role in immune disease.

Deep gray matter volumes

Total DGM volume was associated with IL-1B, CDH6 (Cadherin 6), SIGLEC9 (Sialic Acid Binding Ig Like Lectin 9), and HAGH (hydroxyacylglutathione hydrolase).

IL-1B showed inverse correlation with DGM volume. In primary progressive MS patients, RNA sequencing in peripheral blood mononuclear cells has shown higher IL-1B gene expression correlated with faster disability progression 48 . IL-1B is activated via the NLRP3 (NLR Family Pyrin Domain Containing 3) inflammasome and exerts its proinflammatory, neurotoxic effects via glutamate signaling 48 .

In our study, CDH6 (Cadherin 6) showed positive association with higher DGM volume. CDH6 is a cell surface glycoprotein involved in synaptogenesis, synaptic function, and the preservation of BBB integrity 49 which we hypothesize may have a protective effect on limiting DGM atrophy.

SIGLEC9 (Sialic Acid Binding Ig Like Lectin 9) was also positively correlated with DGM volume. SIGLEC9 can mediate inhibitory immune responses in different autoimmune disease. Soluble SIGLEC9—an extracellular domain of SIGLEC9—has anti-inflammatory effects in models of rheumatoid arthritis via alteration of the polarization of macrophages from M1 to M2 and alleviated experimental colitis in mice via inhibition of NF-kB activation in macrophages 50 . In MS, the isotype switch of microglia to M2 as well as NF-kB pathway inhibition have been shown to have neuroprotective effects 51 . Therefore, it is worthwhile to further investigate the potentially protective effect of SIGLEC9 in MS.

Lastly, we found positive correlations between HAGH (hydroxyacylglutathione hydrolase) levels and DGM volumes. HAGH is an enzyme highly expressed in the brain that regulates oligodendrocyte and neuronal function 52 . It has been shown to be a hypermethylated region in MS patients with decreased transcription levels in comparison to controls 52 .

Putting the DGM findings into perspective, DGM atrophy is not a phenomenon specific to MS patients but also occurs in healthy elderly people 32 . Because we did not perform a comparative analysis in healthy subjects, we cannot rule out that the same proteins also correlate with DGM volumes in healthy people and therefore might not be MS-specific.

Thalamic lesions

Thalamic lesions were present in approximately 54% of subjects in this early MS cohort. Prior work from our group using 7 T MRI to evaluate thalamic lesions in a heterogeneous relapsing MS cohort of varied disease duration found TLs present a rate of 73% 9 . Thalamic lesion volume from that heterogeneous cohort [0.073 ± 0.087 ml] was approximately 2–3 times greater than that shown in this early cohort, likely limiting ability to find significant correlations with serum proteins.

Other MRI outcomes

From the proteins correlating with either cortical lesion number or volume in the adjusted analysis, only GFAP and IL-1b also correlated with T2LV in the adjusted analysis. However, in the unadjusted analysis GFAP’s and IL-1b’s correlation with cortical lesion number and volume as well as T2LV were not present. Therefore, we hypothesize that different proteins are involved in cortical vs. white matter lesion development, which is line with previous findings showing that cortical damage and white matter lesions are (if at all) only weakly or even inversely correlated 53 . A detailed review of serum protein associations with other 7 T MRI outcomes including DGM sub-structures, BPV, T2LV, and parenchymal gad-enhancing lesions was felt to be beyond the scope of this discussion. These outcomes were of limited and secondary importance to our study design, which focuses on unique MRI outcomes at 7 T. However, the results of the entire analysis can be found in Supplementary Table 2 .

Serological NFL and GFAP

In our study, NFL correlated with LME status (presence vs. absence) but not LME foci number or any other MRI outcomes. Because NFL did not correlate with LME foci number in the adjusted or unadjusted analysis but only with LME status, we deem the correlation to be too weak to merit further discussion. We are unaware of studies reporting a significant correlation between NFL and LME. Furthermore, we did not find significant correlations between NFL and cortical lesion load. However, Hogel et al. 54 found a negative correlation between NFL and cortical volume 54 , which was supported by Cruz-Gomez et al. 55 showing a correlation between NFL and cortical thinning 55 . These diverging results might be explained by different MRI correlates that were used between studies. The structural correlate that we used (cortical lesion volume) is more specific to MS than cortical volume/thinning, and therefore less likely to be influenced by confounders.

Unlike NFL, GFAP in our study correlated with cortical lesion volume in the adjusted analysis, which is in line with Hogel et al. 54 showing that GFAP correlated negatively with gray matter volume 54 . On a neurobiological level, GFAP’s correlation with cortical lesions can be explained by the pronounced occurrence of reactive astrocytosis in cortical lesions 56 .

Luminex vs Olink

In comparison to the Olink panel, the Luminex assay found relatively speaking substantially more proteins (33.3 vs. 2%) correlating with cortical lesions in the adjusted analysis. The Olink panel, on the other hand, found relatively speaking (16 vs 13%) more proteins correlating with T2LV (i.e. white matter lesions) in the adjusted analysis, even though the difference is minor. The difference regarding cortical lesions was expected because the Luminex proteins were specifically selected by the authors due to their likely involvement in cortical lesion / LME formation and maintenance, whereas the Olink proteins were preselected by the manufacturer due to their involvement in different, rather unspecific neurobiological processes. This again supports the notion that cortical and white matter lesions have different underlying pathophysiological processes.

Strengths and limitations, future directions

We leveraged ultra-high field strength MRI and a well-established prospective clinical data set (CLIMB Study) to investigate MRI-biomarker associations in a systematic way with a consistent MRI platform and analysis. The inclusion of subjects with relatively short disease course (< 5 years since diagnosis) means that unique biomarker signatures are more likely to be attributable to newly/recently formed lesions, rather than chronic lesion burden. Lastly, we studied a wide array of proteins, including CNS structural proteins which reflect localized cortical damage, as well as immune markers to broadly probe possible associations to MRI.

Limitations of this exploratory study are mainly driven by the small cohort sizes. We present an analysis of a relatively large number of proteins and brain measurements in 57 subjects, where adjustment for multiple comparisons was only feasible for the pathway analysis. Regarding the correlational analysis, none of the correlations had a p-value below the threshold for multiple comparisons (< 0.000035, Bonferroni correction) in the adjusted analysis. Therefore, in the correlational analysis, we considered the p-values without additional adjustment for multiple comparisons. However, in the pathway analysis the p-values for multiple comparisons were considered. Another limitation is that treatment status cannot be ruled out as a confounder because we were not able to adjust for this or run the same analysis only in untreated patients. Either would have yielded a very low statistical power as only 8/57 patients were untreated, the total sample size consisted of only 57 patients, eight different DMTs were used, and the analysis was already adjusted for sex, age, and disease duration.

Furthermore, the non-random selection of patients for 7 T MRI in routine clinical care likely introduced a selection bias. Given its limited availability, neurologists may have prioritized patients for whom the enhanced sensitivity of 7 T MRI could provide greater diagnostic value. On the one hand, these could be patients with a more severe disease course in whom cortical lesions were suspected. On the other hand, these could also be patients in which dissemination in space was not demonstrated and, therefore, were referred for 7 T MRI imaging for the potential establishment of MS diagnosis.

Lesion detection was performed by a single expert rater, though it is possible that rating by multiple raters and/or independent adjudication may improve the sensitivity/specificity of lesion detection. As our analysis focuses on early disease stage MS patients, this might have additionally contributed to a lower overall cortical lesion number and power to detect biomarker associations in our cohort 57 . Another limitation is that three different assays were used, which might introduce a bias in the pathway analysis as this can lead to variability in biomarker quantification due to differences in sensitivity across assays. This might alter the relationship between biomarkers and inferred pathway activity.

While this work highlights interesting associations, we are unable to discern whether changes in biomarker concentrations are causally linked to MRI pathologies vs. secondarily up/downregulated. We used serum samples due to availability; however, future studies may explore these biomarkers in CSF samples. The cohort was largely treated with disease-modifying medication, and future studies will be specifically powered to investigate impact of treatment. With the ongoing clinical use of 7 T MRI at our center, future work will be aimed at linking biomarker signatures to longitudinal change in MRI and clinical status.

Supplementary Information

Supplementary Information 1.
Supplementary Information 2.
Supplementary Information 3.

Supplementary Information 1.

Supplementary Information 2.

Supplementary Information 3.