Serum Glial-Related Biomarkers Show Distinct Contributions to Radiologic and Clinical Outcomes in Early Multiple Sclerosis

Abstract

Background and Objectives

To investigate the association between serum levels of glial-related biomarkers, including glycoprotein nonmetastatic melanoma B (GPNMB), soluble triggering receptor expressed on myeloid cells 2 (sTREM2), galectin-3, serpin peptidase inhibitor clade A member 3 (SERPINA3), chitotriosidase-1 (CHIT1), and chitinase 3-like 1 (CHI3L1) and 2 (CHI3L2), and the presence of paramagnetic rim lesions (PRLs) and the risk of progression independent of relapse activity (PIRA) in patients with a first demyelinating event.

Methods

PRLs were identified using 3T susceptibility-based imaging. Univariable and multivariable linear regression models assessed the association between biomarker levels and the presence of PRLs at baseline. Univariable and multivariable Cox proportional hazards models investigated the association between biomarker levels and risk of PIRA during follow-up. sTREM2 and CHI3L1 levels were measured using single molecule array assays, whereas levels of the remaining biomarkers were determined using ELISA.

Results

We included 135 patients (median age, 34.4 [interquartile range (IQR) 29.2–41.7] years; 98 [72.6%] women; median follow-up, 5.4 [IQR 2.7–9.4] years). PRLs were detected in 59 (43.7%) patients. The presence of ≥1 PRLs at baseline was associated with increased levels of CHI3L1 in univariable (multiplicative effect [ME] = 1.32; 95% CI 1.10–1.58; p = 0.003) and multivariable analyses (ME = 1.31; 95% CI 1.05–1.63; p = 0.01). Thirty (22.2%) patients developed PIRA during follow-up. Doubling of serum galectin-3 levels was associated with an increased risk of PIRA in univariable (hazard ratio [HR] = 1.76; 95% CI 1.06–2.94; p = 0.02) and multivariable analyses (HR = 2.15; 95% CI 1.22–3.80; p = 0.008). Similarly, doubling of serum CHI3L2 levels was associated with an increased risk of PIRA in both univariable (HR = 1.36; 95% CI 1.05–1.76; p = 0.02) and multivariable analyses (HR = 1.36; 95% CI 1.02–1.82; p = 0.03). The combined elevation of galectin-3 and CHI3L2 was associated with an increased risk of PIRA in multivariable analysis (HR = 8.44; 95% CI 3.01–23.70; p < 0.0001).

Discussion

These findings suggest distinct contributions of glial-related biomarkers to radiologic and clinical outcomes in early multiple sclerosis. Validation in independent cohorts will be required before their value as blood prognostic biomarkers can be established.

Introduction

Multiple sclerosis (MS) is a chronic inflammatory disease of the CNS characterized by demyelination, neurodegeneration, and progressive neurologic disability.¹ Although the early stages of MS are predominantly driven by peripheral immune infiltration, increasing evidence indicates that central inflammatory responses mediated by CNS-resident cells, particularly microglia and astrocytes, play a key role in disease progression.^2,3 Microglia and astrocytes are present at sites of MS pathology and actively contribute to the inflammatory environment within the CNS. Microglia, the resident immune cells of the CNS, are involved in lesion activity, being frequently observed in active and chronic lesions.^2,3 Upon activation, microglia can contribute to tissue damage through the release of proinflammatory cytokines, oxidative stress, and other neurotoxic mediators.⁴ Astrocytes also play a central role in MS pathology by producing proinflammatory cytokines and chemokines, contributing to immune cell recruitment, and participating in lesion evolution, including the formation of gliotic scars that may limit tissue repair. In addition, astrocytes can adopt proinflammatory phenotypes that amplify neuroinflammation within the CNS.³ Importantly, bidirectional communication between microglia and astrocytes further modulates inflammatory responses and may influence disease progression.³

A subgroup of chronic active lesions is characterized by a paramagnetic rim of iron-laden, activated microglia/macrophages, detectable by susceptibility-based MRI sequences.^5,6 These paramagnetic rim lesions (PRLs) are considered imaging biomarkers of chronic inflammation and are associated with a more aggressive disease course.^7,8 Previously, in patients with a first demyelinating event,⁹ serum levels of neurofilament light chain (NfL) and glial fibrillary acidic protein (GFAP), reflecting neuronal damage and astrocyte injury, respectively, as well as cerebrospinal fluid (CSF) levels of chitinase 3-like 1 (CHI3L1), a marker of microglia/macrophage and astrocyte activation,¹⁰ were found to be associated with the presence of PRLs. Interestingly, in multivariable analysis, only CSF CHI3L1 remained significantly associated with PRLs. Given that blood-based biomarkers are more practical for clinical use, in this study, we investigated whether serum levels of CHI3L1 are also associated with the presence of PRLs in patients with a first demyelinating event.

Considering the role of glial cells in PRLs, we further expanded our analysis beyond CHI3L1 to include a panel of circulating biomarkers selected a priori to reflect complementary aspects of microglial and astrocytic activation, including glycoprotein nonmetastatic melanoma B (GPNMB),¹¹ and soluble triggering receptor expressed on myeloid cells 2 (sTREM2),¹² as markers related to microglial activation and phagocytic responses; galectin-3,^13,14 as a mediator of innate immune signaling expressed by activated microglia and astrocytes; serpin peptidase inhibitor clade A member 3 (SERPINA3),^15,16 as a marker linked to reactive astrocytes and inflammatory tissue responses; chitotriosidase-1 (CHIT1),¹⁷ and chitinase 3-like 2 (CHI3L2),¹⁸ together with CHI3L1 as members of the chitinase/chitinase-like family previously associated with glial activation. Because progression independent of relapse activity (PIRA) contributes to long-term disability in MS from the early disease stages¹⁹ and is thought to reflect compartmentalized inflammation within the CNS, and given that PRLs have been associated with PIRA,^20,21 we also examined the potential association between these glial-related biomarkers and the risk of PIRA.

Considering the previously reported associations of blood levels of NfL and GFAP with PRLs and PIRA,^22,23 these biomarkers were also included as reference measures to contextualize the associations observed for the glial-related biomarkers investigated in this study.

Therefore, the primary aim of this study was to investigate the association between serum levels of glial-related biomarkers and the presence of PRLs in patients presenting with a first demyelinating event. Secondary exploratory analyses evaluated the relationship between these biomarkers and PIRA.

Methods

Standard Protocol Approvals, Registrations, and Patient Consents

Ethical approval was obtained from the local hospital ethics committee, and all participants provided written informed consent.

Patients

The study included patients (1) experiencing a first demyelinating event consistent with a clinically isolated syndrome (CIS)²⁴; (2) who underwent a baseline MRI on a 3T scanner within 3–5 months of CIS onset, with a standardized imaging protocol including at least 3D T1-weighted, 3D susceptibility-weighted imaging (SWI), 2D or 3D T2-FLAIR, and contrast-enhanced T1-weighted sequences to ensure homogeneity of imaging assessment; and (3) with an available blood sample collected within 3 months of the baseline MRI.

Systematic testing for serum MOG-IgG and AQP4-IgG antibodies was not performed.

MRI Acquisition and Analysis at Baseline

Brain MRI scans were obtained at baseline. All MR images were acquired on a 3T Siemens MRI system (Siemens, Erlangen, Germany) with either a 12- or 64-channel phased array head coil and a whole-body transmit coil. The following sequences were acquired in all patients: (1) 3D sagittal T1-weighted magnetization-prepared rapid acquisition gradient echo (MPRAGE) (repetition time [TR] = 2,300 ms; echo time [TE] = 2.98 ms; inversion time [TI] = 900 ms; voxel size = 1.0 × 1.0 × 1.0 mm); (2) 3D axial gradient-echo SWI without contrast (TR = 33 ms; TE1 = 6.08 ms; TE2 = 24.6 ms; flip angle = 15°; matrix size = 288 × 384 × 104; voxel size = 0.65 × 0.65 × 3.0 mm); (3) 2D transverse T2-FLAIR (TR = 9,000 ms; TE = 87 ms; TI = 2,500 ms; flip angle = 119°; matrix size = 412 × 512 × 46; voxel size = 0.49 × 0.49 × 3.0 mm) or sagittal 3D FLAIR (TR = 5,000 ms; TE = 394 ms; TI = 1,800 ms; flip angle = 120°; matrix size = 240 × 256 × 176; voxel size = 1.0 × 1.0 × 1.0 mm); and (4) a 2D transverse T1-weighted gradient-echo sequence after contrast administration (baseline only), using a gadolinium-based hydrophilic and neutral macrocyclic contrast agent (gadobutrol, 0.1 mmol/kg [Gadovist, Bayer Healthcare Pharmaceuticals]) acquired with a 5–10 minutes delay (TR = 297 ms; TE = 2.46 ms; flip angle = 66°; voxel size = 0.78 × 0.78 × 3.0 mm).

Focal brain lesions were identified on 3D T2-FLAIR images and analyzed for the presence of PRLs on SWI-filtered phase images, which were visually evaluated by an experienced neuroradiologist according to the 2024 NAIMS Consensus Statement on Imaging Chronic Active Lesions.²⁵ In summary, PRLs were defined as T2-hyperintense lesions surrounded by a paramagnetic rim involving all or part of the lesion border, with no corresponding gadolinium enhancement on postcontrast T1-weighted images.²⁵

As a global measure of brain atrophy, the brain parenchymal fraction (BPF) was calculated from the 3D MPRAGE images using the Statistical Parametric Mapping suite. After image segmentation, a threshold of 0.7 was applied, and the corresponding volumes of gray matter, white matter, and CSF compartments were obtained. The BPF was calculated as BPF = (gray + white)/(gray + white + CSF). T2 lesion volumes (T2LV) were estimated using the Lesion Segmentation Toolbox. In each baseline contrast-enhanced T1-weighted scan, the number of enhancing lesions was visually recorded.

We analyzed the T2LV, number of PRLs, number of contrast-enhancing lesions (CELs), and BPF at baseline.

Progression Independent of Relapse Activity (PIRA)

Disability accumulation was assessed using the Expanded Disability Status Scale (EDSS), with evaluations performed every 6 months throughout follow-up. Confirmed disability accumulation (CDA) was defined as an increase in EDSS of ≥1.5 points if the reference EDSS score (baseline or re-baseline) was 0, ≥1.0 point if the reference EDSS was 1.0–5.0, or ≥0.5 points if the reference EDSS was >5.0. Confirmation required persistence of the EDSS increase at least 6 months later, in line with standard definitions of confirmed progression.²⁶

PIRA events were defined as 6-month CDA occurring in the absence of clinical relapses. Relapses were not allowed between the reference, event, and confirmation visits. Periods free of relapse were defined as the interval between 2 consecutive relapses, starting 3 months after a relapse (or 6 months after the first demyelinating event). The first EDSS assessment obtained ≥6 months after the initial attack, or ≥3 months after any subsequent relapse, was considered the baseline or rebaseline, respectively. No rebaseline EDSS value could be lower than the initial baseline EDSS score. The date of PIRA was assigned to the date of confirmation of CDA.²⁷

Sample Collection and Quantification of Biomarkers

Peripheral blood was collected at the time of the first demyelinating event and allowed to clot at room temperature for 30 minutes. Serum was then obtained by centrifugation and stored at −80°C until used. Serum levels of GPNMB, galectin-3, CHIT1, CHI3L2, and SERPINA3 were measured using the following commercially available ELISA kits: GPNMB and galectin-3 (R&D Systems, Minneapolis, MN; #DY2550 and #DY1154, respectively), CHIT1 and CHI3L2 (#CY-8074 and #CY-8087, CircuLex, MBL, UK), and SERPINA3 (#KA2128, Abnova, Taipei, Taiwan). Serum sTREM2 levels were quantified using a commercially available immunoassay kit (#104542, Quanterix, Billerica, MA), and serum CHI3L1 levels were determined using a homemade Simoa-based assay, as previously described,²⁸ using a fully automated ultrasensitive Simoa HD-X Analyzer (Quanterix). Serum NfL and GFAP were measured using the Neuro 2-Plex B Advantage kit (#103520, Quanterix) on a fully automated ultrasensitive Simoa HD-X Analyzer (Quanterix). All samples for both ELISA and Simoa analyses were run in duplicate according to manufacturer instructions, including appropriate standards and internal controls. For ELISA-based measurements, serum samples were diluted before analysis using the following empirical dilution factors: CHI3L2, 1:5; CHIT1, 1:20; galectin-3, 1:2; GPNMB, 1:10; and SERPINA3, 1:5,000. Most sample concentrations fell within the readable/linear range of the corresponding standard curves; only a small number of samples required adjustment of the dilution factor, including further dilution or analysis at lower dilution or undiluted, to ensure reliable interpolation. Samples with a coefficient of variation >20% between duplicates were reanalyzed. The intraassay and interassay coefficients of variation are shown in eTable 1.

Statistical Methods

Statistical analyses were performed using RStudio (Posit Software, PBC) running R version 4.4.2. Correlations between serum biomarker levels were assessed using the Spearman rank correlation coefficient (rho).

Univariable and multivariable linear regression models were used to analyze the association between serum biomarker levels and the number of PRLs at baseline, considering PRLs as a categorical variable using 2 stratifications: a binary classification (0 vs ≥ 1 PRLs) and a three-level classification (0, 1–3, and >3 PRLs), with 0 PRLs as the reference category, as well as a continuous variable. Biomarker levels were log-transformed to meet the normality assumption, and estimates were back-transformed to represent multiplicative effects (ME). Multivariable models were adjusted for age, sex, T2LV, number of CELs, and BPF. Multicollinearity was assessed using variance inflation factors (VIF).

Univariable and multivariable Cox proportional hazards models were used to investigate the association between serum biomarker levels and risk of PIRA. Biomarker levels were log2-transformed and estimates represent the effects on the risk of PIRA per biomarker doubling. Multivariable models were adjusted for age, sex, baseline T2LV, baseline number of PRLs, and whether patients received disease-modifying therapies (DMTs) before the occurrence of the end point. Results are presented in tables and forest plots showing hazard ratios (HR) and 95% CI.

Optimal cutoff points for serum CHI3L2 and galectin-3 levels were determined using the surv_cutpoint function from the survminer R package, which identifies the threshold that maximizes the difference in survival between groups based on the log-rank test. Patients were categorized into 4 groups according to whether CHI3L2 and galectin-3 levels were above or below these cutoffs (“high/high,” “high/low,” “low/high,” “low/low”). Kaplan–Meier curves were generated to estimate the time to PIRA for each group, and differences were assessed using the log-rank test. In addition, a Cox proportional hazards model adjusted for age, sex, baseline T2LV, baseline number of PRLs, and treatment was used to assess the risk of PIRA in each group compared with the low/low reference category.

All analyses in this study were exploratory and hypothesis-generating in nature. In this context, nominal p values are presented to indicate the magnitude of the observed effects rather than to imply definitive statistical significance, as the analyses were conducted using available convenience samples rather than based on a predefined sample size calculation. To account for multiple testing, the Benjamini-Hochberg false discovery rate (FDR) procedure was applied to the primary biomarker analyses, and the corresponding FDR-adjusted p values are reported in the supplementary tables.

Data Availability

The data sets used and/or analyzed during the current study are available from the corresponding author on reasonable request.

Results

Demographic, Clinical, and Radiologic Characteristics of Patients

A total of 135 patients fulfilling the inclusion criteria were enrolled between October 2010 and September 2024. Of them, 115 (85.2%) were diagnosed with MS by the time of the last follow-up visit.²⁹ CSF IgG oligoclonal bands were present in 92 of the 129 patients (71.3%) with available CSF analysis. Table 1 summarizes the main demographic, clinical, and radiologic information of patients. The median age of patients was 34.4 years (interquartile range [IQR] 29.2–41.7), and 98 (72.6%) were women.

Table 1.

Summary of Demographic, Clinical, and Radiologic Characteristics of Patients With a First Demyelinating Event, Stratified by PLR Status

Characteristics	Entire cohort	0 PRLs	≥1 PRLs	p Value
N	135	76 (56.3)	59 (43.7)
Female/male (% women)	98/37 (72.6)	56/20 (73.7)	42/17 (71.2)	0.898
Age at sample collection (y)	34.4 (29.2–41.7)	34.5 (29.3–40.7)	34.1 (28.7–41.7)	0.729
T2 lesion volume (mL)	1.3 (0.5–4.1)	0.5 (0.3–1.3)	3.4 (1.8–7.0)	<0.0001
N° CELs (%)a				<0.0001
0	69 (56.6)	49 (76.6)	20 (34.5)
≥1	53 (43.4)	15 (23.4)	38 (65.5)
BPF	0.82 (0.80–0.84)	0.83 (0.81–0.84)	0.80 (0.78–0.82)	<0.001
EDSS at sample collection	1.5 (1.0–2.0)	1.5 (1.0–2.0)	1.5 (1.0–2.5)	0.834
Time since symptoms onset (d)b	67.5 (43.9–106.5)	65.9 (43.1–108.5)	68.9 (45.6–102.4)	0.944
Time to MRI (d)c	49.0 (25.0–69.5)	49.0 (23.5–70.2)	48.0 (25.5–64.5)	0.788
Treatment with corticosteroids (N(%))	49 (36.3)	28 (36.8)	21 (35.6)	1.000
Follow-up time (y)d	5.4 (2.7–9.4)	4.5 (2.4–8.4)	6.4 (3–10.2)	0.053
Serum GPNMB (pg/mL)	16,555.0 (13,680.0–20,360.0)	15,940.0 (13,690.0–19,955.0)	17,400.0 (13,725.0–20,535.0)	0.275
Serum sTREM2 (pg/mL)	725.7 (538.2–961.9)	707.3 (534.0–960.0)	800.9 (559.4–958.2)	0.757
Serum CHI3L1 (ng/mL)	20.4 (15.6–27.6)	19.1 (14.5–23.2)	20.8 (18.0–30.9)	0.006
Serum galectin-3 (pg/mL)	1,725.0 (1,317.4–2,352.5)	1,743.0 (1,260.0–2,299.5)	1,704.1 (1,382.5–2,392.5)	0.668
Serum SERPINA3 (ng/mL)	159,099.9 (127,919.9–192,522.7)	157,782.6 (122,237.9–184,844.7)	161,850.6 (131,343.2–194,703.2)	0.544
Serum CHIT1 (pg/mL)	15,283.0 (11,142.5–23,675.0)	14,941.0 (11,165.0–22,715.0)	15,328.0 (11,114.0–24,370.0)	0.588
Serum CHI3L2 (pg/mL)	1,198.2 (645.1–2,013.9)	1,207.4 (631.4–1,839.4)	1,174.2 (661.2–2,135.0)	0.841
Serum NfL (pg/mL)	7.3 (5.6–13.78)	6.2 (5.0–9.4)	11.6 (6.8–21.0)	<0.0001
Serum GFAP (pg/mL)	75.3 (62.2–109.1)	74.3 (59.9–104.2)	76.8 (66.7–124.6)	0.126

Abbreviations: BPF = brain parenchymal fraction; CELs = contrast-enhancing lesions; CHI3L1/CHI3L2 = chitinase 3-like 1 and chitinase 3-like 2, respectively; CHIT1 = chitotriosidase-1; GFAP = glial fibrillary acidic protein; GPNMB = glycoprotein nonmetastatic melanoma B; NfL = neurofilament light chain; PRLs = paramagnetic rim lesions; SERPINA3 = serpin peptidase inhibitor clade A member 3; sTREM2 = soluble triggering receptor expressed on myeloid cells 2.

Data are expressed as median (interquartile range).

p Values refer to comparisons between patients with 0 and ≥1 PRLs. Continuous variables were compared using the Mann-Whitney U test, and categorical variables using the χ² test. Significant p values are shown in bold.

^aInformation is missing for 13 (9.6%) patients.

^bRefers to the time between symptom onset and sample collection for biomarker assessment.

^cRefers to the time between sample collection and the baseline MRI.

^dRefers to the time between sample collection and the last follow-up visit.

At baseline, the median T2LV was 1.3 mL (IQR 0.5–4.1), and the median BPF 0.82 (IQR 0.80–0.84). CELs were present in 53 (43.4%) patients. The median EDSS was 1.5 (IQR 1.0–2.0). The median time between symptom onset and sample collection was 67.5 days (IQR 43.9–106.5), and the median time between sample collection and baseline MRI was 49.0 days (IQR 25.0–69.5). Forty-nine (36.3%) patients received corticosteroids before sample collection. None of the patients were receiving DMTs at the time of sample collection, nor between sample collection and baseline MRI for the evaluation of PRLs. The median clinical follow-up time was 5.4 years (IQR 2.7–9.4), with a minimum follow-up of 0.1 years. A summary timeline of the study procedures is provided in eFigure 1.

Correlation Between Biomarker Levels

Table 1 shows the median (IQR) serum levels of the biomarkers, and eTable 2 summarizes the correlations between them. Only weak but statistically significant positive correlations were observed between serum CHIT1 levels and both CHI3L1 (rho = 0.28; p = 0.001) and sTREM2 (rho = 0.23; p = 0.007), as well as between CHI3L1 and sTREM2 (rho = 0.23; p = 0.008). A weak but statistically significant inverse correlation was also observed between SERPINA3 and galectin-3 (rho = −0.19; p = 0.03). The remaining pairwise correlations were not statistically significant (eTable 2).

Association Between Glial-Related Biomarkers and PRLs at Baseline

As shown in Table 1, PRLs were present in 59 (43.7%) patients. Patients with PRLs had significantly higher T2LV (p < 0.0001), a greater percentage of CELs on baseline MRI (p < 0.0001), and lower BPF (p < 0.001). No significant differences were observed in the time between symptom onset and sample collection, the interval between sample collection and MRI, the proportion of patients receiving corticosteroids before sample collection, or the follow-up time.

Patients with ≥1 PRLs had significantly higher serum CHI3L1 and NfL levels compared with those without PRLs (p = 0.006 and p < 0.0001, respectively), whereas no differences were observed for the remaining biomarkers (Table 1).

In univariable analysis, the presence of ≥1 PRLs was associated with a 32% increase in serum CHI3L1 levels compared with having 0 PRLs (ME: 1.32; 95% CI 1.10–1.58; p = 0.003). This association remained after multivariable analysis, with 31% higher CHI3L1 levels in patients with PRLs (ME: 1.31; 95% CI 1.05–1.63; p = 0.01) (Figure 1; eTable 3). None of the other glial-related biomarkers were associated with the presence of PRLs. Regarding the reference biomarkers, serum NfL levels were significantly associated with the presence of PRLs in univariable analysis (ME: 1.83; 95% CI 1.41–2.39; p < 0.0001), although this association did not remain after multivariable adjustment (ME: 1.28; 95% CI 0.97–1.68; p = 0.08). Serum GFAP levels showed some evidence of an association with PRLs in univariable analysis (ME: 1.15; 95% CI 0.99–1.33; p = 0.06), but this association was not observed after multivariable analysis (Figure 1; eTable 3). No evidence of multicollinearity was observed among the variables included in the multivariable models (all VIF <2; data not shown). A sensitivity analysis incorporating treatment with corticosteroids before sample collection into the multivariable model yielded similar results (data not shown).

Figure 1. Effect of PRLs as Categorized Variable on Glial-Related Biomarker Levels.

Forest plots showing the effects of PRLs categorized as 0 vs ≥1 on serum biomarker levels. The plots display mean effects (solid squares) with 95% confidence intervals for having ≥1 PRLs compared with having none. Two models are shown: a univariable (“simple”) model and a multivariable (“adjusted”) model that includes age, sex, baseline number of CELs, baseline T2 lesion volume, and baseline BPF as covariates. CHI3L1/CHI3L2 = chitinase 3-like 1 and chitinase 3-like 2, respectively; CHIT1 = chitotriosidase-1; GPNMB = glycoprotein non-metastatic melanoma B; PRLs = paramagnetic rim lesions; SERPINA3 = serpin peptidase inhibitor clade A member 3; sTREM2 = soluble triggering receptor expressed on myeloid cells 2.

When patients were stratified into 3 categories according to PRL burden (0, 1–3, and >3 lesions), higher serum CHI3L1 levels were associated with the presence of 1–3 PRLs in both univariable (ME: 1.32; 95% CI 1.08–1.62; p = 0.007) and multivariable analyses (ME: 1.31; 95% CI 1.04–1.65; p = 0.02). Higher CHI3L1 levels were also observed in patients with >3 PRLs in univariable analysis, although the evidence for this association was weaker (ME: 1.31; 95% CI 0.98–1.74; p = 0.06), and it was not retained after multivariable adjustment (Figure 2; eTable 4). No associations were observed for the other glial-related biomarkers. Serum NfL levels were significantly associated with both 1–3 PRLs (ME: 1.54; 95% CI 1.15–2.05; p = 0.004) and >3 PRLs (ME: 2.91; 95% CI 1.93–4.38; p < 0.0001) in univariable analysis. After multivariable adjustment, the association with 1–3 PRLs was not observed, whereas the association with >3 PRLs remained (ME: 1.59; 95% CI 1.03–2.47; p = 0.040). Serum GFAP levels were associated with >3 PRLs in univariable analysis (ME: 1.55; 95% CI 1.24–1.93; p = 0.0002), but this association was not observed after multivariable adjustment (Figure 2; eTable 4).

Figure 2. Effect of PRL Burden on Glial-Related Biomarker Levels.

Forest plots showing the effects of PRL burden categorized as 0, 1–3, and >3 lesions on serum biomarker levels. The plots display mean effects with 95% confidence intervals for patients with 1–3 PRLs (solid circles) and >3 PRLs (solid squares), each compared with patients without PRLs. Two models are shown: a univariable (“simple”) model and a multivariable (“adjusted”) model that includes age, sex, baseline number of CELs, baseline T2 lesion volume, and baseline BPF as covariates. CHI3L1/CHI3L2 = chitinase 3-like 1 and chitinase 3-like 2, respectively; CHIT1 = chitotriosidase-1; GPNMB = glycoprotein nonmetastatic melanoma B; PRLs = paramagnetic rim lesions; SERPINA3 = serpin peptidase inhibitor clade A member 3; sTREM2 = soluble triggering receptor expressed on myeloid cells 2.

When the number of PRLs was analyzed as a continuous variable, serum CHI3L1 levels showed limited evidence of an association in univariable analysis (ME: 1.05; 95% CI 0.99–1.10; p = 0.07), although this association was not observed after multivariable adjustment (ME: 1.04; 95% CI 0.97–1.11; p = 0.25) (eTable 5). No associations were observed for the other glial-related biomarkers in either univariable or multivariable analyses. With respect to the reference biomarkers, serum NfL (ME: 1.21; 95% CI 1.13–1.30; p < 0.0001) and GFAP (ME: 1.09; 95% CI 1.05–1.13; p < 0.0001) levels were significantly associated with the number of PRLs in univariable analysis, although these associations were not observed after multivariable adjustment (eTable 5).

Association Between Glial-Related Biomarkers and Risk of PIRA

The predefined relapse-free windows and EDSS assessments used for PIRA ascertainment are illustrated in eFigure 2. As shown in Table 2, 30 (22.2%) patients developed PIRA during follow-up after a median time of 5.2 years (IQR 3.4–7.3). Patients with PIRA had significantly higher baseline T2LV (p = 0.005) and longer follow-up (p < 0.0001). There was also a higher proportion of treated patients during follow-up in this group (p = 0.04). eTable 6 summarizes DMTs received during follow-up. Although a higher proportion of patients with PRLs at baseline was observed among those who developed PIRA (57% vs 40%), the difference was not statistically significant (p = 0.15). Similarly, no significant baseline differences were observed between patients with and without PIRA in the BPF, EDSS, proportion of patients with CELs, or the time between symptom onset and sample collection and between sample collection and MRI (Table 2).

Table 2.

Summary of Demographic, Clinical, and Radiologic Characteristics of Patients With and Without PIRA

Characteristics	No PIRA	PIRA	p Value
N	105 (77.8)	30 (22.2)
Female/male (% women)	77/28 (73.3)	21/9 (70.0)	0.897
Age at sample collection (y)	33.7 (29.0–41.2)	36.1 (31.2–42.8)	0.276
T2 lesion volume (mL)a	1.1 (0.4–3.1)	2.8 (1.0–6.6)	0.005
N° CELs (%)a,b			0.141
0	56 (60.9)	13 (43.3)
≥1	36 (39.1)	17 (56.7)
N° PRLs (%)a			0.157
0	63 (60.0)	13 (43.3)
≥1	42 (40.0)	17 (56.7)
BPFa	0.82 (0.80–0.84)	0.81 (0.80–0.83)	0.176
EDSS at sample collection	1.5 (1.0–2.0)	1.5 (1.0–2.5)	0.622
Time since symptoms onset (d)c	67.4 (43.3–110.3)	63.4 (47.1–95.8)	0.650
Time to MRI (d)d	50.0 (25.0–70.0)	46.5 (25.2–68.2)	0.913
Time to PIRA (y)	—	5.2 (3.4–7.3)	—
Follow-up time (y)e	4.4 (2.0–7.9)	10.6 (6.5–12.2)	<0.0001
Treatment (N (%))	73 (69.5)	27 (90.0)	0.043
Serum GPNMB (pg/mL)	16,530.0 (13,170.0–19,527.5)	17,335.0 (14,140.0–21,687.5)	0.322
Serum sTREM2 (pg/mL)	711.9 (522.2–943.5)	848.8 (582.6–999.7)	0.210
Serum CHI3L1 (ng/mL)	19.8 (15.1–25.1)	22.3 (18.5–33.6)	0.029
Serum galectin-3 (pg/mL)	1,694.9 (1,265.8–2,409.8)	1,898.0 (1,441.0–2,329.0)	0.316
Serum SERPINA3 (ng/mL)	159,814.1 (122,331.9–194,517.0)	157,391.5 (138,824.3–174,466.5)	0.768
Serum CHIT1 (pg/mL)	14,784.0 (10,644.0–22,620.0)	18,938.0 (12,492.0–27,720.0)	0.168
Serum CHI3L2 (pg/mL)	1,148.0 (608.5–1,907.5)	1,498.0 (828.5–2,276.5)	0.085
Serum NfL (pg/mL)	6.5 (5.3–12.4)	9.6 (6.8–15.7)	0.019
Serum GFAP (pg/mL)	74.7 (61.6–105.5)	75.5 (68.5–110.7)	0.328

Abbreviations: BPF = brain parenchymal fraction; CELs = contrast-enhancing lesions; CHI3L1/CHI3L2 = chitinase 3-like 1 and chitinase 3-like 2, respectively; CHIT1 = chitotriosidase-1; GFAP = glial fibrillary acidic protein; GPNMB = glycoprotein nonmetastatic melanoma B; NfL = neurofilament light chain; PIRA = progression independent of relapse activity; PRLs = paramagnetic rim lesions; SERPINA3 = serpin peptidase inhibitor clade A member 3; sTREM2 = soluble triggering receptor expressed on myeloid cells 2.

Data are expressed as median (interquartile range).

p Values refer to comparisons between patients with and without PIRA. Continuous variables were compared using the Mann-Whitney U test, and categorical variables using the χ² test. Significant p values are shown in bold.

^aRefers to information at baseline MRI.

^bInformation is missing for 13 (9.6%) patients.

^cRefers to the time between symptom onset and sample collection for biomarker assessment.

^dRefers to the time between sample collection and the baseline MRI.

^eRefers to the time between sample collection and the last follow-up visit.

At baseline, patients with PIRA had significantly higher serum CHI3L1 and NfL levels compared with those without PIRA (p = 0.02 and p = 0.01, respectively) (Table 2).

Doubling of serum galectin-3 levels was associated with a higher risk of PIRA in both univariable analysis (HR = 1.76; 95% CI 1.06–2.94; p = 0.02) and multivariable analysis (HR = 2.15; 95% CI 1.22–3.80; p = 0.008) (Figure 3; eTable 7). Similarly, doubling of serum CHI3L2 levels was associated with a higher risk of PIRA in both univariable (HR = 1.36; 95% CI 1.05–1.76; p = 0.02) and multivariable analyses (HR = 1.36; 95% CI 1.02–1.82; p = 0.03). As shown in eTable 7, the number of PRLs at baseline and treatment during follow-up were independently associated with the risk of PIRA in the multivariable models.

Figure 3. Association Between Glial-Related Biomarkers and Risk of PIRA.

Forest plot shows the effects per doubling of serum biomarker levels on the risk of PIRA. The graphs display hazard ratios (solid squares) with 95% confidence intervals from univariable (“simple”) and multivariable (“adjusted”) models after adjusting for age, sex, baseline T2 lesion volume, baseline number of PRLs, and treatment during follow-up. The x-axis is on a log10 scale. CHI3L1/CHI3L2 = chitinase 3-like 1 and chitinase 3-like 2, respectively; CHIT1 = chitotriosidase-1; GPNMB = glycoprotein nonmetastatic melanoma B; PIRA = progression independent of relapse activity; SERPINA3 = serpin peptidase inhibitor clade A member 3; sTREM2 = soluble triggering receptor expressed on myeloid cells 2.

When patients were classified into 4 groups according to the presence of high or low serum levels of galectin-3 and CHI3L2, those with both high galectin-3 and CHI3L2 levels had a shorter time to PIRA compared with the other groups (log-rank test p < 0.0001; Figure 4). In multivariable analysis, using patients with low levels of both galectin-3 and CHI3L2 as the reference group, the combined elevation of both proteins was associated with an increased risk of PIRA (HR = 8.44; 95% CI 3.01–23.70; p < 0.0001).

Figure 4. Kaplan-Meier Curves in Patients Classified According to the Presence of High/Low CHI3L2 and Galectin-3 Levels.

Kaplan-Meier curves in patients classified into 4 groups according to high or low CHI3L2 and galectin-3 serum levels, based on cutoff values of 1,493.5 pg/mL and 1,743.6 pg/mL, respectively (“low CHI3L2/low galectin-3,” “high CHI3L2/high galectin-3,” “high CHI3L2/low galectin-3,” “low CHI3L2/high galectin-3”). The graph shows the log-rank test p value. The dashed line indicates the median time to PIRA. The number of patients at risk for each group at different follow-up times is shown in the table. Results from the Cox proportional hazards model adjusted for age, sex, baseline T2LV, baseline number of PRLs, and treatment are also shown as HR and 95% confidence intervals, comparing each group with the low/low reference category. CHI3L2 = chitinase 3-like 2; HR = hazard ratio; PIRA = progression independent of relapse activity.

Neither serum NfL nor GFAP levels were associated with the risk of PIRA in the univariable or multivariable analyses (eTable 7).

Discussion

We investigated the contribution of a panel of glial-related biomarkers to radiologic and clinical MS outcomes associated with disease progression in patients with a first demyelinating event. Previously, we observed a significant association between CSF CHI3L1 levels and the presence of PRLs, analyzed both as a categorical and as a continuous variable.⁹ CHI3L1 is primarily secreted by astrocytes and microglial cells, although peripheral monocytes/macrophages may also contribute to circulating levels.¹⁰ In a previous study, serum and CSF CHI3L1 levels were significantly correlated,³⁰ and serum CHI3L1 levels were associated with disability progression in patients with progressive disease.²⁸ In this study, we found that serum CHI3L1 levels were also associated with having ≥1 PRLs; however, the effect appeared to be weaker compared with CSF levels, as only limited evidence of an association was observed when the number of PRLs was considered as a continuous variable.

We also evaluated serum NfL and GFAP as reference biomarkers. Although both biomarkers were associated with PRLs in the univariable analyses, these associations were not retained after multivariable adjustment, consistent with our previous findings.⁹ Interestingly, when PRL burden was categorized as 0, 1–3, and >3 lesions, serum NfL showed a clearer relationship with increasing PRL burden than CHI3L1, whereas GFAP displayed a more modest association. These findings may suggest that CHI3L1 is more closely related to the presence of chronic active lesions, whereas NfL may better reflect the cumulative neuroaxonal damage associated with increasing PRL burden.

Regarding other glial-related biomarkers, GPNMB and CHIT1 have previously been identified among the most highly upregulated genes within chronic active lesion rims.³¹ However, neither serum GPNMB nor CHIT1 levels were associated with the presence of PRLs in our cohort. These findings suggest that, although both proteins are strongly expressed within chronic active lesions at the tissue level, this upregulation may not be fully reflected in circulating serum levels.

Regarding PIRA, the combined elevation of serum levels of galectin-3 and CHI3L2 was associated with a substantially increased risk of PIRA in patients with a first demyelinating event. Although these findings should be interpreted cautiously given the limited number of PIRA events, the possibility of model overfitting, and the lack of external validation, they suggest that glial-related pathways may contribute to disability progression. In multivariable analysis, the number of PRLs at baseline was also independently associated with a higher risk of PIRA, a finding that has been previously reported.^20,21 In addition, treatment was associated with PIRA, which may reflect that patients who developed PIRA were more likely to receive therapy during follow-up.

Galectin-3 is a multifunctional β-galactoside–binding lectin involved in immune regulation and tissue repair.³² It is highly expressed in immune cells such as macrophages and microglia.³² Upon activation, microglia upregulate and secrete galectin-3, which amplifies innate immune responses through Toll-like receptor signaling and promotes the production of proinflammatory cytokines.³³ Serum galectin-3 levels may thus reflect the degree of microglial activation. Interestingly, hippocampal galectin-3 knockdown in mice attenuated LPS-induced neurotoxicity,³⁴ supporting a role for galectin-3 in promoting neuronal injury beyond its effects on microglial activation. Activated astrocytes may also contribute to circulating galectin-3 levels.¹⁴

CHI3L2, also known as YKL-39, belongs to family 18 glycosyl hydrolases and is closely related to CHI3L1, and similarly to CHI3L1, it is a biomarker of microglial and astrocytic activation.¹⁸ CHI3L2 has been implicated in MS: elevated CSF levels are associated with an increased risk of MS and predict cognitive impairment in patients with CIS.³⁵ Notably, in patients with primary progressive MS, CSF CHI3L2 levels were significantly higher in those who exhibited faster disease progression after long-term follow-up.³⁶ Whether galectin-3 and CHI3L2 are secreted by the same microglial and astrocytic populations is unknown; however, the lack of correlation in serum suggests distinct CNS cellular sources or potential contributions from peripheral macrophages.

Neither serum NfL nor GFAP was associated with the risk of PIRA in our cohort. Although previous studies have reported associations between these biomarkers and disability progression,^23,37-39 differences in patient populations, disease stage, outcome definitions, follow-up duration, adjustment strategies, and study design, including the evaluation of patients receiving DMTs to minimize the contribution of inflammatory activity to disability progression, may account for these discrepant findings.

Neither serum SERPINA3, sTREM2 nor CHIT1 was associated with radiologic or clinical outcomes. SERPINA3, an acute-phase serine protease inhibitor and biomarker of reactive astrocytes,¹⁶ regulates inflammation and antiviral responses.⁴⁰ TREM2 is a microglial receptor involved in myelin debris clearance and tissue repair.⁴¹ Its soluble form (sTREM2), detectable in CSF and blood, reflects microglial activation and may regulate TREM2 signaling.^41,42 Although CSF SERPINA3 levels are increased in MS, particularly in progressive forms, and correlate with CSF NfL,⁴³ and elevated CSF sTREM2 has been linked to greater disability,^44,45 neither serum SERPINA3 nor sTREM2 was associated with PIRA in our cohort.

CHIT1 also belongs to the glycosyl hydrolase family 18, like CHI3L1 and CHI3L2, and is mainly secreted peripherally by activated macrophages, whereas activated microglia are a major source of CHIT1 in the CSF.^17,46,47 In a recent study, CSF CHIT1 levels were associated with disability progression in a cohort of MS patients with both relapsing and progressive forms of the disease.¹⁷ In our study, serum CHIT1 levels were not associated with the risk of PIRA. A common polymorphism in the CHIT1 gene has been described and is associated with reduced CHIT1 expression and enzymatic activity.⁴⁸ However, in our study, analyses after excluding a few patients with very low serum CHIT1 levels yielded similar results (data not shown).

Finally, although PIRA was defined using a standardized and recently proposed operational definition, alternative definitions have been used across studies, differing for example in relapse-free windows or confirmation criteria. Therefore, whether the associations observed in our study remain consistent when applying alternative PIRA definitions warrants further investigation.

Several additional limitations should be acknowledged. First, this was an exploratory study including a relatively limited sample size for the number of biomarkers investigated, and missing data further reduced the number of patients available for some multivariable analyses, which reduced statistical power and resulted in relatively wide confidence intervals for some estimates. Second, although the overall follow-up was relatively long, follow-up duration varied across participants, potentially limiting the identification of late PIRA events in some patients. Third, paired CSF samples were not available, precluding direct comparisons between serum and CSF concentrations. This is particularly relevant because CSF-serum correlations remain poorly characterized for most of the biomarkers investigated. Consequently, the absence of an association in serum should not necessarily exclude a relationship within the CNS compartment. Finally, the combined galectin-3/CHI3L2 model was derived and evaluated within the same cohort and has not yet been externally validated. Accordingly, the findings should be regarded as hypothesis-generating.

In conclusion, our findings suggest that serum CHI3L1, a biomarker of astrocytic and microglial activation, is associated with the presence of PRLs, although the strength of this association appears lower than that previously reported for CSF levels. In addition, CHI3L2, a member of the same glycosyl hydrolase family as CHI3L1 and mainly secreted by astrocytes and microglia, and galectin-3, which is more closely linked to microglial activation, were associated with an increased risk of PIRA. Although circulating levels may be influenced by peripheral sources, including myeloid cells such as macrophages, these findings require validation in independent cohorts before the clinical utility of these blood-based glial biomarkers can be established.

Glossary

BPF

brain parenchymal fraction

CDA

confirmed disability accumulation

CEL

contrast-enhancing lesion

CHI3L1

chitinase 3-like 1

CHI3L2

chitinase 3-like 2

CHIT1

chitotriosidase-1

CIS

clinically isolated syndrome

DMT

disease-modifying therapy

EDSS

Expanded Disability Status Scale

GFAP

glial fibrillary acidic protein

GPNMB

glycoprotein nonmetastatic melanoma B

HR

hazard ratio

IQR

interquartile range

ME

multiplicative effect

MPRAGE

magnetization-prepared rapid acquisition gradient echo

MS

multiple sclerosis

NfL

neurofilament light chain

PIRA

progression independent of relapse activity

PRL

paramagnetic rim lesion

SERPINA3

serpin peptidase inhibitor clade A member 3

sTREM2

soluble triggering receptor expressed on myeloid cells 2

TE

echo time

TR

repetition time

VIF

variance inflation factors

Author Contributions

M. Comabella: drafting/revision of the manuscript for content, including medical writing for content; major role in the acquisition of data; study concept or design; analysis or interpretation of data. L. Bollo: major role in the acquisition of data; analysis or interpretation of data. N. Fissolo: major role in the acquisition of data; analysis or interpretation of data. P. Benkert: analysis or interpretation of data. I. Klass: major role in the acquisition of data; analysis or interpretation of data. C. Tur: major role in the acquisition of data; analysis or interpretation of data. E. Saez de Gordoa: major role in the acquisition of data; analysis or interpretation of data. L. Gutierrez: major role in the acquisition of data; analysis or interpretation of data. C. Auger: major role in the acquisition of data; analysis or interpretation of data. J. Rio: major role in the acquisition of data; analysis or interpretation of data. J. Sastre-Garriga: major role in the acquisition of data; analysis or interpretation of data. M. Tintore: major role in the acquisition of data; analysis or interpretation of data. J. Kuhle: analysis or interpretation of data. À. Rovira: major role in the acquisition of data; analysis or interpretation of data. X. Montalban: major role in the acquisition of data; analysis or interpretation of data.

Study Funding

The authors report no targeted funding.

Disclosure

The authors report no relevant disclosures. Go to Neurology.org/NN for full disclosures.