Serum Glial Fibrillary Acidic Protein and Retinal Neuronal Loss as Additive Prognostic Markers of Disability in Multiple Sclerosis

Abstract

Background and Objectives

In people with multiple sclerosis (pwMS), optical coherence tomography (OCT) quantifies loss of neurons (macular ganglion cell-inner plexiform layer [mGCIPL]) and axons (peripapillary retinal nerve fiber layer [pRNFL]) in the retina. Serum glial fibrillary acidic protein (sGFAP) is a promising astrocytic biomarker to capture disease progression in pwMS. We aimed to investigate the relationship between OCT markers and sGFAP in pwMS and explore their additive value in predicting disability progression.

Methods

PwMS and healthy controls underwent OCT at baseline (BL), excluding eyes with inter-eye asymmetry. Age, sex, and body mass index–adjusted Z scores of sGFAP were calculated. Cross-sectional and longitudinal associations between sGFAP and retinal layers were estimated using linear regression- and mixed-effects models (LMM). The additive effect of BL-OCT and BL-sGFAP on the trajectory of the Expanded Disability Status Scale (EDSS) was estimated using LMM, whereby pwMS were stratified into: group (1): low sGFAP Z score (<3rd quartile, <Q3) and thick mGCIPL (>Q1); group (2): high sGFAP Z score (≥Q3) and thick mGCIPL or low sGFAP Z Score and thin mGCIPL (≤Q1); and group (3): high sGFAP and thin mGCIPL.

Results

Two hundred and sixty-one pwMS (mean age: 48 years (y), female: 63%, on disease-modifying treatment: 80%, mean thickness of pRNFL: 94 μm and mGCIPL: 66 μm) and 52 controls (age: 52 years, female: 65%, pRNFL: 101 μm, mGCIPL: 72 μm) were included. At BL, pRNFL (β = −0.01, p = 0.042) and mGCIPL (β = −0.02, p = 0.013) were negatively associated with sGFAP Z scores in pwMS, but not in controls (p = 0.950, p = 0.386). BL-mGCIPL was also associated with sGFAP trajectories (β = −0.003, p = 0.044), over a median follow-up of 2.9 years. Compared with pwMS with good results in both markers (group 1), those with either high sGFAP or thin mGCIPL had a steeper EDSS increase (β = 0.030, p = 0.048), while pwMS with both high sGFAP and thin mGCIPL (group 3) showed the steepest trajectory of the EDSS (β = 0.101, p < 0.001).

Discussion

Our findings show a close relationship between astrocytic activation/injury and neurodegeneration in the CNS, measured at the retinal level. Moreover, they highlight an additive role of mGCIPL and sGFAP for identification of pwMS at higher risk of disability worsening.

Introduction

Neurodegeneration within the CNS is recognized as the primary pathological substrate of disability progression in people with multiple sclerosis (pwMS).^1-3 Thus, measures sensitive to neuronal and axonal loss in the CNS can be used to identify pwMS at higher risk of progression and to quantify potential neuroprotective treatment effects. In this regard, optical coherence tomography (OCT), a noninvasive technique measuring retinal layer thicknesses, can play an important role. Thinning of the peripapillary retinal nerve fiber (pRNFL) and the macular ganglion cell-inner plexiform layer (mGCIPL), containing CNS-axons and -neurons, respectively, is associated with gray matter integrity in the brain,^2,4 physical and cognitive disability,^5-10 as well as progression independent of relapse activity (PIRA)^11-13 in MS.

In addition, serum biomarkers can be valuable tools in individual stratification in pwMS. Serum neurofilament light chain (sNfL), a cytoskeletal protein of neurons, is a well-established marker for treatment response in active MS.^14-16 However, its capacity to capture progression, especially when acute inflammatory disease is suppressed by high efficacy therapies, is still under debate.¹⁶ In this regard, serum glial fibrillary acid protein (sGFAP), another cytoskeletal protein specific for astrocytes, was recently associated with gray matter atrophy^16,17 and higher risk of PIRA in pwMS.^16,18,19

Our aim was to investigate the association between retinal atrophy and sGFAP in pwMS, at cross-sectional and longitudinal level and to examine if it is independent of focal inflammation in the visual pathway. Moreover, as data on the added value of OCT metrics and sGFAP are lacking, we aimed at investigating if their combination is associated with disability worsening in pwMS.

Methods

Study Participants and Design

We included 261 pwMS from the Swiss MS Cohort Study (SMSC²⁰) in the University Hospital of Basel (UHB) with ≥1 OCT and blood samples every 6 or 12 months including complete neurological examination assessing the Expanded Disability Status Scale (EDSS) according to the Neurostatus definitions²¹ by certified neurologists, who were blinded to the results of OCT and blood samples (Figure 1). The inclusion criteria were: (1) 18 years or older, (2) MS diagnosis according to the 2017 revised McDonald Criteria,²² and (3) eligibility for OCT (i.e., ability to fixate with each eye). The exclusion criteria included: (1) presence of any ocular pathology that may interfere with the validity of the OCT (e.g., glaucoma, macular oedema), (2) refractive error >6 diopters, and (3) history of bilateral optic neuritis (ON), according to the participants' clinical files (Figure 2).

Figure 1. Overview of the Study Design.

OCT scans were acquired at baseline only, while serum samples and EDSS evaluation were obtained at baseline and during follow-up visits. The median follow-up duration for longitudinal sGFAP measurements was 2.9 years (IQR 2.2–6.5). EDSS = Expanded Disability Status Scale; OCT = optical coherence tomography.

Figure 2. Consort Chart of OCT Examinations.

PwMS with a bilateral ON were excluded. All OCT scans underwent quality control according to OSCAR-IB criteria. Finally, 261 participants were included with 444 peripapillary scans and 476 macular scans. pwMS = people with multiple sclerosis; OCT = optical coherence tomography; ON = optic neuritis.

In addition, 52 controls were recruited through online announcement to investigate potential (physiologic) associations. The controls underwent a single OCT examination and blood sampling. The inclusion criteria were (1) 18 years or older; (2) negative history of neurologic, ophthalmologic, or psychiatric diseases; and (3) no refractive error >6 diopters.

OCT of the Retina

All study participants underwent an OCT at baseline (BL), on a Heidelberg Engineering Spectralis device (Heidelberg, Germany), in a dark room, without pupil dilation. The following OCT markers were assessed on both eyes: (1) mean thickness of the pRNFL, (2) mean thickness of the mGCIPL, and (3) mean thickness of the macular inner nuclear layer (mINL). PwMS with history of bilateral ON (n = 7, 14 eyes) were excluded since the pronounced local damage would be a confounder to our analysis.

OCT quality control was performed applying the OSCAR-IB criteria²³ and a total of 37 peripapillary (n = 10 due to technical reasons and n = 27 due to incidental findings) and 37 macular scans (n = 17 due to technical reasons and n = 20 due to incidental findings) were excluded. The detailed OCT acquisition settings and scanning protocol according to the APOSTEL²⁴ recommendations were previously described by our group¹¹ and are also given as eTable 1.

We also excluded eyes with local damage due to previous unilateral ON or subclinical optic nerve demyelination, identified by increased inter-eye asymmetry, using previously described thresholds: for pRNFL ≥5 µm and for mGCIPL ≥4 µm.^25,26 In such cases, the eye with the thinner layer was excluded. For participants with intereye asymmetry in mGCIPL, the mINL of the worse eye was excluded as well from the analysis. For all OCT markers, we calculated the mean of both eyes when available. If only one eye measurement was present, this value was used.

Glial Fibrillary Acidic Protein and Neurofilament Light Chain in Serum

PwMS underwent blood sampling at BL and longitudinally. Five pwMS were excluded from the cross-sectional analysis, due to a prolonged interval (>90 days) between OCT and blood sampling. All blood samples were collected and stored at −80°C following standardized protocols.²⁷ sGFAP concentrations were measured using ultrasensitive single-molecule array (Simoa) technology (Quanterix, Billerica, MA) according to the manufacturer's instructions and as previously described.²⁸ Z scores were calculated using a reference database of 4′297 healthy control samples, adjusted for age, sex, and body mass index (BMI).²⁸ These Z scores provide continuous values, reflecting deviations from control measurements in terms of SDs.

From the same blood samples, the concentrations and Z scores of sNfL were also calculated, adjusted for age and BMI, as previously described. We used sNfL as a “control biomarker” throughout the analysis.¹⁴

MRI of the Brain

To adjust the analysis for disease activity and for local damage along the visual pathway, we included brain MRI data obtained within 90 days of each study visit. MRI was acquired on a 3-T scanner (Siemens Skyra). The protocol included a 3-dimensional (3D), T1-weighted, 1-mm isotropic magnetization prepared rapid gradient-echo (MPRAGE) after Gd administration and a 3D, 1-mm isotropic fluid-attenuated inversion recovery (FLAIR). FLAIR-hyperintense lesions were automatically segmented, and results were manually revised. Optic radiation lesion volume was obtained after coregistering the lesion masks with the Juelich brain segmentation atlas, as previously described.² MRI disease activity was defined as new or enlarging lesion on FLAIR (compared with the previous MRI) or Gd enhancing lesion on MPRAGE.

Statistical Analysis

The distribution of continuous variables was assessed visually using Q-Q plots; normally distributed variables are summarized as mean with SDs and non-normal variables as median with interquartile range (IQR).

To investigate the associations between Z scores of serum biomarkers (dependent variables in separate models) and OCT measures at BL, linear regression analysis was used, adjusted for age at disease onset, sex, disease duration, disease-modifying treatment (untreated, platform, orals, monoclonal), EDSS, and disease activity. The latter was defined as relapse and/or new/enlarging FLAIR lesions and/or gadolinium-enhancing lesions within the previous 90 days. Furthermore, a sensitivity analysis was conducted, correcting for lesions in the optic radiations.

For longitudinal analysis, linear mixed-effects models (LMM) with random effects per participant were employed, adjusting for the same covariates as above (DMT and disease activity were considered at each visit, not only at BL).

The interaction between BL OCT values and time since BL in years was included as a fixed effect to predict the trajectory of biomarker Z scores over time (dependent variable). This model type was chosen because it accounts for varying follow-up durations and irregular visit intervals, as they are suited for unbalanced longitudinal data and unequal sampling frequencies.^29,30

To avoid collinearity, separate models were fitted for each OCT marker. The same analyses conducted for sGFAP were repeated for sNfL Z scores.

Furthermore, the EDSS linear trajectory (dependent variable) in association with BL OCT or BL serum biomarkers was estimated using LMM, adjusting for the same covariates as above with additional adjustment for BL EDSS.

To investigate whether the combination of OCT and serum biomarkers are associated with the EDSS course, we first identified the OCT layer and serum biomarker with the strongest associations to the EDSS trajectory, based on standardized β-coefficients. Then, we used the strongest predictors, to stratify participants into 3 groups: group 1: low serum biomarker Z score (<Q3) and thick OCT layer (>Q1); group 2: high Z score (≥Q3) and thick OCT layer or low Z score and thin OCT layer (≤Q1); and group 3: high Z score and thin OCT layer. In group 2, we included pwMS with only one “abnormal” measure, allowing us to assess whether participants with both biomarkers being “abnormal” exhibit a more pronounced change in EDSS over time. In addition, nonlinear spline-based models were fitted and compared with the corresponding LMM using analysis of variance (ANOVA).

Furthermore, a sensitivity analysis was conducted using alternative, predefined thresholds for sGFAP and mGCIPL, defining “abnormal” sGFAP as Z score ≥1.5 and reduced mGCIPL thickness as ≤ 77 µm as previously described.^10,16 Estimated coefficients, 95% CIs, and p values are reported for each model, with statistical significance defined as p < 0.05.

The statistical analysis was performed using R Version 4.4.2 with packages Ime4, ImerTest, MuMIn, pastecs, ellipsis, and ggplot2.

Standard Protocol Approvals, Registrations, and Consents

The study was approved by the local ethical standards committee on human experimentation (Ethikkommission Nordwest-und Zentralschweiz, EKNZ, Project ID: 2023-02144) and conducted in accordance with the declaration of Helsinki. All participants gave written informed consent before study participation.

Data Availability

The data that support the findings of this study could become available from the corresponding author upon reasonable request.

Results

Group Comparisons: pwMS vs Controls

Overall, 261 pwMS and 52 HC were included in the study. There were no group differences in age and sex. PwMS showed thinner pRNFL and mGCIPL as well as higher sGFAP and sNfL Z scores compared with controls (Table 1).

Table 1.

Participant Characteristics at Baseline

	PwMS (n = 261)	Controls (n = 52)	Comparison
Age, y	48.0 ± 12.8	51.5 ± 13.6	p = 0.09
Sex: female %	63.2	65.4	p = 0.82
Disease course: RMS %	86.6	—
DMT, %		—
Untreated	19.5
Platform	6.5
Orals	46.7
Monoclonals	27.3
Disease duration, y	13.1 (5.8–21.5)	—
EDSS	2.5 (1.5–3.5)	—
Disease activity %	15.7	---
pRNFL thickness, µm	94.1 ± 13.4	100.9 ± 7.3	p < 0.01
mGCIPL thickness, µm	66.4 ± 8.1	72.3 ± 4.1	p < 0.01
mINL thickness, µm	34.7 ± 2.4	34.4 ± 2.1	p = 0.45
sGFAP, pg/mL	100.85 ± 72.79	88.27 ± 54.13	p = 0.15
sGFAP Z scores	0.11 ± 1.12	−0.43 ± 1.0	p < 0.001
sNfL, pg/mL	10.42 ± 9.59	9.78 ± 5.03	p = 0.49
sNfL Z scores	0.16 ± 1.12	−0.22 ± 0.77	p = 0.004

Abbreviations: DMT = disease-modifying therapies; EDSS = Expanded Disability Status Scale; IQR = interquartile range; mGCIPL = macular ganglion cell inner plexiform layer; mINL = macular inner nuclear layer; pRNFL = peripapillary retinal nerve fiber layer; RMS = Relapsing Multiple Sclerosis; sGFAP = serum glial fibrillary acidic protein; sNfL = serum neurofilament light-chain; y = years.

Baseline demographic and clinical characteristics of the study population. Continuous variables are presented as mean ± SD or median (interquartile range, IQR), as appropriate. Categorical variables are as percentages (%). Platform includes interferon-betas and glatiramer acetates; orals include S1P-receptor modulators, teriflunomide, fumarate derivates and purine analogue; monoclonals include B-cell depleting therapies and natalizumab.

Retinal Layers and Serum Biomarkers: Associations at BL

Thinner pRNFL (per 10 µm) was associated with 0.12 Z score units higher sGFAP levels (β = −0.012, 95% CI [−0.022 to −0.0004], p = 0.042). Similarly, thinner mGCIPL (per 10 µm) was related to 0.24 Z score units higher sGFAP (β = −0.024, 95% CI [–0.043 to −0.005], p = 0.013). No cross-sectional associations were observed between mINL and sGFAP (Table 2, Figure 3).

Table 2.

Associations of OCT Measures With sGFAP Levels at Baseline

	sGFAP Z Score (n = 256)
	β	95% CI	p Value	β	95% CI	p Value	β	95% CI	p Value
pRNFL	−0.012	−0.022 to −0.0004	0.042*	—	—	—	—	—	—
mGCIPL	—	—	—	−0.024	−0.043 to −0.005	0.013*	—	—	—
mINL	—	—	—	—	—	—	−0.012	−0.033 to 0.009	0.253
Age at onset	−0.014	−0.028 to −0.0004	0.043*	−0.017	−0.032 to −0.003	0.018*	−0.015	−0.029 to −0.0004	0.044*
Sex: female	0.382	−0.077 to 0.467	0.171	0.126	−0.153 to 0.405	0.376	0.151	−0.126 to 0.428	0.283
Disease duration at BL	−0.001	−0.019 to 0.016	0.899	−0.005	−0.023 to 0.013	0.592	0.002	−0.016 to 0.019	0.832
Disease course: RMS	0.382	−0.166 to 0.931	0.171	0.569	−0.002 to 1.141	0.051	0.383	−0.171 to 0.936	0.175
DMT: Monoclonal vs untreated	−0.706	−1.136 to −0.277	0.001*	−0.573	−1.018 to −0.128	0.012*	−0.679	−1.111 to 0.246	0.002*
DMT: Monoclonal vs platform	−0.030	−0.565 to 0.625	0.921	0.066	−0.534 to 0.666	0.828	−0.018	−0.602 to −0.565	0.951
DMT: Monoclonal vs orals	−0.580	−0.913 to −0.248	<0.001*	−0.498	−0.833 to −0.163	0.004*	−0.520	−0.853 to −0.186	0.002*
EDSS at BL	0.160	0.040 to 0.281	0.009*	0.170	0.048 to 0.292	0.006*	0.178	0.056 to 0.301	0.005*
Disease activity within 90d	1.156	−0.221 to 0.534	0.416	1.113	−0.274 to 0.501	0.566	0.143	−0.238 to 0.524	0.460

Abbreviations: BL = baseline; DMT = disease-modifying therapies; EDSS = Expanded Disability Status Scale; mGCIPL = macular ganglion cell inner plexiform layers; mINL = macular inner nuclear layer; pRNFL = peripapillary retinal nerve fiber layer; RMS = relapsing multiple sclerosis; sGFAP = serum glial fibrillary acidic protein.

Cross-sectional baseline associations between the 3 OCT measures and serum GFAP Z scores. Results derive from multivariable linear regression models (1 model for each OCT measure) adjusted for age at onset, sex, disease duration and EDSS at baseline, DMT, and disease activity within 90 d. Regression coefficients (β) are reported with 95% CIs and p values. * Indicates significant results (<0.05).

Figure 3. Associations of OCT Measures With Serum Biomarkers at Baseline.

Cross-sectional associations between serum biomarker Z scores and retinal layer thickness at baseline. Scatter plots with linear regression lines (blue) and 95% CIs (shaded) are shown for (top row) serum GFAP Z scores and (bottom row) serum NfL Z scores in relation to baseline peripapillary retinal nerve fiber layer (pRNFL: left), macular ganglion cell inner plexiform layers (mGCIPL: middle), and macular inner nuclear layer (mINL: right). Note that the beta and p values derive from the adjusted linear regression models. BL = Baseline; pRNFL = peripapillary retinal nerve fiber layer; GCIPL = ganglion cell inner plexiform layer; INL = Inner nuclear layer, sGFAP = serum glial fibrillary acid protein, sNfL = serum neurofilament light chain.

There was a negative association between mINL and sNfL at BL (β = −0.025, 95% CI [−0.046 to −0.004], p = 0.022), but no associations were found between sNfL Z scores and pRNFL or mGCIPL (pRNFL: β = 0.002, p = 0.787; mGCIPL: β = −0.003, p = 0.702); eTable 2, Figure 3). Moreover, in controls, there were no associations between sGFAP or sNfL Z scores and retinal layers thicknesses (data not shown; p values 0.386–0.950).

Retinal Layers and Serum Biomarkers: Longitudinal Associations

The median follow-up for the longitudinal sGFAP measurements was 2.9 years (IQR [2.2–6.5] years). BL mGCIPL thickness showed a significant association with the trajectory of GFAP Z scores. Thinner mGCIPL (per 10 µm) at BL was associated with a steeper increase in sGFAP by 0.03 Z score units more per year (β = −0.003, 95% CI [−0.005 to −0.000], p = 0.044; Table 3, Figure 4). BL pRNFL and mINL were not associated with the course of sGFAP Z scores (Table 3, Figure 4).

Table 3.

Associations Between Baseline OCT Measures and Time Course of sGFAP Over Follow-Up

	sGFAP Z Score (n = 261)
	β	95% CI	p Value	β	95% CI	p Value	β	95% CI	p Value
BL pRNFL* time	−0.001	−0.003 to 0	0.084	—	—	—	—	—	—
BL mGCIPL* time	—	—	—	−0.003	−0.005 to 0	0.044*	—	—	—
BL mINL* time	—	—	—	—	—	—	<0.001	−0.002 to 0.003	0.760
Age at onset	−0.017	−0.030 to −0.005	0.008*	−0.018	−0.031 to −0.005	0.006*	−0.018	−0.030 to −0.005	0.007*
Sex: female	0.193	−0.061 to 0.446	0.142	0.160	−0.098 to 0.418	0.231	0.192	−0.063 to 0.446	0.146
Disease duration at BL	−0.001	−0.017 to 0.015	0.861	−0.004	−0.021 to 0.012	0.601	0	−0.015 to 0.016	0.958
Disease course: RMS	0.306	−0.178 to 0.790	0.222	0.480	−0.025 to 0.985	0.067	0.319	−0.169 to 0.807	0.206
DMT: Monoclonal vs untreated	−0.27	−1.209 to 0.675	0.575	−0.313	−1.245 to 0.625	0.513	−0.359	−1.295 to 0.581	0.454
DMT: Monoclonal vs platform	−0.017	−1.026 to 0.997	0.974	0.007	−0.998 to 1.015	0.99	−0.134	−1.140 to 0.873	0.794
DMT: Monoclonal vs orals	−0.433	−1.381 to 0.517	0.373	−0.443	−1.386 to 0.500	0.358	−0.486	−1.432 to 0.461	0.316
EDSS at BL	0.161	0.053 to 0.270	0.004*	0.165	0.055 to 0.275	0.004*	0.175	0.066 to 0.285	0.002*
Disease activity within 90 d	0.045	−0.100 to 0.190	0.549	0.011	−0.137 to 0.161	0.881	0.037	−0.108 to 0.182	0.622

Abbreviations: BL = baseline; DMT = disease-modifying therapies; EDSS = expanded disability status scale; mGCIPL = macular ganglion cell inner plexiform layers; pRNFL = peripapillary retinal nerve fiber layer; RMS = relapsing multiple sclerosis; sGFAP = serum glial fibrillary acidic protein.

Results derive from linear mixed-effects models (1 model for each OCT measure) with random intercepts for participant ID and using each time the interaction of time since BL and OCT measures, adjusted for age at onset, sex, disease duration and EDSS at baseline, DMT, and disease activity within 90 d. Regression coefficients (β) are shown with 95% CIs and p values. * Indicates significant results.

Figure 4. Associations of OCT Measures at Baseline With sGFAP and sNfL Trajectories.

Predicted longitudinal trajectories of serum GFAP and NfL Z scores in relation to baseline OCT measures. Marginal effect plots with predictions (lines) with 95% CIs (shaded areas) are shown for serum GFAP Z scores (top row) and NfL Z scores (bottom row) over time since baseline (in years). Trajectories are stratified by different baseline retinal layer thicknesses (pRNFL, mGCIPL, and mINL). BL = baseline; GCIPL = ganglion cell inner plexiform layer; INL = inner nuclear layer; pRNFL = peripapillary retinal nerve fiber layer; sGFAP = serum glial fibrillary acid protein; sNfL = serum neurofilament light chain; y = years.

BL pRNFL and mGCIPL showed also a negative association with the course of sNfL Z scores over time (pRNFL: β = −0.003, 95% CI [−0.004 to −0.001], p = 0.004; mGCIPL: β = −0.004, 95% CI [–0.007 to 0.001], p = 0.009; eTable 3, Figure 4).

Retinal Layers, Serum Biomarkers, and EDSS Trajectory

A thinner pRNFL (per 10 µm) at BL was associated with a steeper increase in EDSS by 0.02 more per year (β = −0.0024, 95% CI [–0.003 to −0.001], p < 0.001) and a thinner mGCIPL by 0.06 (β = −0.006, 95% CI [–0.007 to –0.004], p <0.001). mINL showed no association (β 0.001, p = 0.428). When looking at the standardized estimates of these relationships, we found a stronger association of mGCIPL (standardized β = −0.06) compared with pRNFL (standardized β = −0.04) with the EDSS trajectory.

Regarding serum biomarkers, sGFAP Z scores showed a trend (β = 0.011, p = 0.052) for association with the EDSS trajectories, while sNfL Z scores were not prognostic (β = −0.001, p = 0.904).

Based on these results, mGCIPL was combined with sGFAP for stratification. Group 1 (low sGFAP Z score (<0.84) and thick mGCIPL (>61.45 µm) included 141 pwMS, group 2 (either low sGFAP Z score with thin mGCIPL (≤61.45 µm), or high sGFAP Z score (≥0.84) with thick mGCIPL included 89 participants, while 19 pwMS were in group 3 (both high sGFAP Z score and thin mGCIPL). When using group 1 as a reference, pwMS in group 2, showed a steeper increase in EDSS over time (β = 0.030, 95% CI [0.0003 to 0.059], p = 0.048), while group 3 showed the steepest EDSS trajectory (β = 0.101, 95% CI [0.054 to 0.157], p < 0.001). These results are depicted in Figure 5. Notably, the nonlinear spline-based models showed comparable results to those obtained with the LMM (p value of ANOVA for model comparison: p = 0.241).

Figure 5. EDSS Trajectories Depending on Groups When Combining mGCIPL and sGFAP.

EDSS trajectories by combined mGCIPL/sGFAP groups. Group 1 (n = 141): low sGFAP (Z score <0.84) and thick mGCIPL (>61.45 µm); Group 2 (n = 89): high sGFAP (Z score ≥0.84) with thick mGCIPL or low sGFAP (Z score <0.84) with thin mGCIPL (≤61.45 µm); Group 3 (n = 19): high sGFAP and thin mGCIPL. BL = baseline; EDSS = Expanded Disability Status Scale; y = year.

Sensitivity Analyses

Cross-sectionally, after adjustment for lesions in the optic radiations, the association between sGFAP and pRNFL was no longer present (β = −0.010, 95% CI [–0.022 to 0.003], p = 0.135), whereas the association with mGCIPL remained (β = –0.023, 95% CI [–0.045 to −0.001], p = 0.037). No associations of OCT metrics with sNfL were observed.

Moreover, using different thresholds for sGFAP and mGCIPL for patient stratification, similar findings were observed as in the primary analysis: Compared with group 1 (n = 14), group 2 (n = 206) showed a steeper increase in EDSS (β = 0.089, p = 0.002), while group 3 (n = 28) showed the steepest EDSS trajectory (β = 0.116, p < 0.001).

Discussion

In this study, we investigated the relationship between OCT markers and sGFAP as well as their additive value to predict disability worsening, in a large, well-characterized MS cohort of 261 pwMS. We found that retinal axonal (thinner pRNFL) and neuronal loss (thinner mGCIPL) are associated with higher sGFAP Z scores. Notably, these associations were observed only in pwMS (not in controls) and persisted after adjusting for multiple patient characteristics, including disease duration, DMT, and signs of disease activity including relapses and MRI activity, to minimize the impact of inflammatory processes on our results. Moreover, the retina-sGFAP relationship was not only evident at cross-sectional level but also longitudinally, whereby lower mGCIPL thickness at BL was associated with a steeper increase of sGFAP over time.

These results align with prior findings suggesting that sGFAP—as biomarker of astrocytic activation and damage—may reflect neurodegeneration independent of inflammatory activity in MS.^16,18,19 It was previously shown that sGFAP is related to gray matter and spinal cord atrophy¹⁶; our results now indicate also a relationship to loss of neurons and axons in the retina, as part of the CNS. In line with this, a recent prospective study with 116 participants showed higher sGFAP levels in pwMS with faster thinning of pRNFL and mGCIPL.³¹

It has to be noted that GFAP is also expressed locally in the retina, primarily in Müller cells in the mINL during reactive gliosis.^32,33 However, we did not find an association between mINL thickness and sGFAP in our cohort. Owing to the much smaller size of the retina compared with the entire brain and the exclusion of eyes with prior ON or subclinical optic nerve involvement from our analysis, we do not expect a significant impact of local, retinal GFAP expression on our results. As an explanation of our findings, we rather hypothesize, that astrocytic damage/activation in the brain is linked to more severe neurodegeneration in the entire CNS, which is also reflected by thinner pRNFL and mGCIPL at the level of the retina. The contribution of focal lesions in the visual pathway (optic radiations) to the observed associations was evaluated in a sensitivity analysis. After adjustment for lesion volume in the optic radiations, the relationship between sGFAP and pRNFL did not remain, whereas the association with mGCIPL persisted. This suggests that lesion-related astrocytic activation and retrograde degeneration may partly drive the relationship between retinal axonal loss (pRNFL) and sGFAP, while the persistent mGCIPL-sGFAP association likely reflects more diffuse, chronic neurodegenerative processes affecting both brain and retina.

sGFAP was previously linked to white matter lesion load on MRI.³⁴ However, associations between sGFAP and progression remained after adjustment for T2-weighted lesion volume in a large previously published cohort.¹⁶ Future studies using a multimodal approach integrating serum biomarkers, OCT and MRI, could provide deeper insights into the respective roles of focal vs diffuse damage in the sGFAP-OCT associations.

Despite very limited prior data on the relationship of OCT metrics with sGFAP in pwMS,³¹ there are some studies in other neuroinflammatory diseases, such as neuromyelitis optica spectrum disorders (NMOSDs). Lin et al.³⁵ included 33 pwMS with aquaporin-4 antibody-positive NMOSD and found that lower mGCIPL thickness in non-ON eyes was associated with higher sGFAP levels, whereas this was not the case for participants with myelin oligodendrocyte glycoprotein antibody–associated disease. Aly et al.³⁶ also found an ON-independent relationship between foveal thinning and higher sGFAP levels in participants with NMOSD. Interestingly, this latter study included also 21 people with relapsing-remitting MS, in which foveal thinning was not correlated with sGFAP, but the generalizability of these results is limited due to the small sample size.

A main finding of our study is the prediction of EDSS trajectories by the combination of mGCIPL and sGFAP in our pwMS. We found that the combination of sGFAP with mGCIPL at BL was superior to each marker alone in predicting the trajectory of EDSS over a median follow-up of 2.9 years. PwMS with both high sGFAP levels and thin mGCIPL showed a higher EDSS increase of 0.11 points per year compared with participants with low sGFAP levels and thick mGCIPL. Importantly this corresponds to a difference of approximately 0.5 EDSS points over 5 years, which is considered clinically meaningful, particularly at higher EDSS levels.³⁷ The predictive value of the combined model remained consistent when applying different cutoffs, highlighting its robustness.

Several recent studies highlighted the value of combining different biomarkers to increase accuracy of patient stratification in MS. For example, a study by Lin YT et al. investigated the combined effect of abnormal sNfL and thin mGCIPL in 78 pwMS and found that individuals with both markers abnormal had a significantly higher risk of NEDA-3 violation, new brain lesions, and new relapses compared with those with elevated sNfL alone.³⁸ Moreover, Krajnc et al.¹³ demonstrated that the combination of PRLs, indicative of chronic inflammation, with yearly thinning of the pRNFL or mGCIPL, reflecting neurodegeneration, was associated with a higher risk of PIRA. Considering also the easy implementation of both OCT and serum biomarkers, a combination of retinal neuronal loss (as reflected by mGCIPL thinning in eyes without ON) and sGFAP, may be particularly useful in identifying pwMS more prone to neurodegeneration as well as for quantifying effects of novel treatments in progressive MS.³⁹

Our analysis with sNfL as “control biomarker” also revealed some interesting results. We hypothesized that sNfL would be less associated with retinal neuronal and axonal loss than sGFAP, in line with prior results on gray matter atrophy.¹⁶ This was confirmed at BL, since sNfL did not show any significant relationship to pRNFL or mGCIPL. Interestingly, we observed a negative association between mINL and sNfL at BL. This might seem unexpected, since a thickening of this layer was previously described as possible sign of disease activity in pwMS.^40,41 However, our model was corrected for clinical and MRI-disease activity, which may explain this result. Moreover, BL pRNFL and mGCIPL were associated with the sNfL trajectory, indicating a possible relationship between OCT and ongoing disease activity,⁴² as described before. In a previous study by Krajnc et al. a single sNfL measurement did not predict future retinal thinning.⁴³ However, in our study, the analysis was performed “vice versa,” showing that OCT at BL may be predictive of sNfL increase over time. Sotirchos et al. also reported a relationship between elevated sNfL levels and pRNFL/mGCIPL thinning.⁴⁴ Overall, these findings confirm that inflammatory and neurodegenerative processes in MS are closely intertwined and cannot be fully disentangled.⁴⁵

Our study has limitations. First, all pwMS were recruited from a single center and were predominantly of European descent, which may limit the generalizability of our findings. Second, we only had longitudinal measurements for serum biomarkers, not for OCT measures; thus, a temporal or potentially causal relationship between astrocytic injury and retinal neurodegeneration cannot be inferred by our data. Longitudinal OCT studies are essential to determine whether sGFAP elevations precede, occur in parallel with, or follow retinal thinning.

Furthermore, adjusted Z scores were used only for serum and not OCT markers. The calculation of Z scores for the OCT metrics using large databases (similar to the approach for serum biomarkers¹⁴) is needed to minimize the impact of age-related retinal axonal and neuronal loss and other confounding factors in future studies.⁴⁶

As disability measures like the timed 25-foot walk (T25FW) or the 9-hole peg test (9-HPT) were not available in this cohort, the use of only EDSS as clinical outcome is a limitation. However, EDSS scoring in this study was performed according to the Neurostatus system by certified raters using standardized definitions and rigorous quality control to ensuring low inter-rater and intrarater variability.⁴⁷ Finally, disability progression was evaluated using the EDSS values at every visit, which may be less specific than PIRA events, to reflect progression driven by neurodegeneration. This was done due to the limited number of PIRA events in this cohort. Still, analysis with longer observational period and inclusion of PIRA/progression independent of relapse and MRI events is needed to confirm the additive role of OCT and sGFAP as predictors of progression in pwMS. While the combined stratification approach is clinically intuitive, it currently serves as a risk enrichment strategy rather than a tool for individual level prognostication. Further longitudinal validation is required before direct clinical applicability can be established.

To conclude, our findings demonstrate a close relationship between sGFAP levels and retinal neuronal and axonal loss in pwMS. The additive effects of mGCIPL thinning and elevated sGFAP on EDSS worsening over time highlight the potential of combining retinal and serum biomarkers for improved patient stratification in MS.

Ethical Considerations and Consent to Participate

The study was approved by the local ethics committee (Ethikkommission Nordwest-und Zentralschweiz (EKNZ); under the Project ID: 2023-02144) and conducted in accordance with the declaration of Helsinki. All participants gave written informed consent before inclusion in the study.

Glossary

ANOVA

analysis of variance

BL

baseline

BMI

body mass index

EDSS

expanded disability status scale

FLAIR

fluid-attenuated inversion recovery

IQR

interquartile range

mGCIPL

macular ganglion cell-inner plexiform layer

mINL

macular inner nuclear layer

MPRAGE

magnetization prepared rapid gradient-echo

NMOSD

neuromyelitis optica spectrum disorder

OCT

optical coherence tomography

ON

optic neuritis

PIRA

progression independent of relapse activity

pRNFL

peripapillary retinal nerve fiber layer

pwMS

people with multiple sclerosis

sGFAP

serum glial fibrillary acidic protein

sNfL

serum neurofilament light chain

UHB

University Hospital of Basel

Author Contributions

S. Sellathurai: 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. K. Schoenholzer: drafting/revision of the manuscript for content, including medical writing for content; major role in the acquisition of data. F. Burguet Villena: major role in the acquisition of data. N. Cerdá-Fuertes: major role in the acquisition of data. L. Hofer: analysis or interpretation of data. S.A. Schaedelin: analysis or interpretation of data. V. Kana: drafting/revision of the manuscript for content, including medical writing for content. P. Mahenthiran: major role in the acquisition of data. M. D'Souza: drafting/revision of the manuscript for content, including medical writing for content. A. Maleska Maceski: major role in the acquisition of data. T.J. Derfuss: drafting/revision of the manuscript for content, including medical writing for content. J. Oechtering: drafting/revision of the manuscript for content, including medical writing for content. K. Gugleta: drafting/revision of the manuscript for content, including medical writing for content. L. Kappos: drafting/revision of the manuscript for content, including medical writing for content. A. Petzold: drafting/revision of the manuscript for content, including medical writing for content. C. Granziera: drafting/revision of the manuscript for content, including medical writing for content. P. Benkert: analysis or interpretation of data. J. Kuhle: drafting/revision of the manuscript for content, including medical writing for content; major role in the acquisition of data. A. Papadopoulou: 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.

Study Funding

This study was supported by the Swiss National Science Foundation (Project number PZ00P3_216,468; “Ambizione” Grant to: A.P).

Disclosure

K. Schoenholzer and S. Sellathurai are supported by the Swiss National Science Foundation (PZ00P3_216,468). F. Burguet Villena received travel support by TEVA and a research grant (“Young talents in clinical research”) by the Swiss Academy of Medical Sciences and of the G. & J. Bangerter-Rhyner Foundation. NCF is an employee of Neurostatus UHB MD is CEO of Neurostatus AG and has received travel support from Bayer AG, Biogen, Teva Pharmaceuticals and Sanofi Genzyme and research support from the University Hospital of Basel. L. Hofer, S.A. Schaedelin have nothing to disclose. V. Kana received honoraria for advisory roles and/or lectures for Biogen, Novartis, Merck, Roche, Teva, and travel support from Biogen, Merck, Roche, none of it relevant to this work. P. Mahenthiran has nothing to disclose. M. D' Souza is an employee of the University Hospital Basel and the CEO of Neurostatus-UHB Ltd. He has received research support from the University Hospital Basel. AMM has nothing to disclose TD serves on scientific advi-sory boards for Actelion (Janssen/J&J), Bayer, Biogen, Celgene (Bristol Myers Squibb), GeNeuro, MedDay, Merck, Mitsubishi Pharma, Novartis, Roche, and Sanofi; has received funding for travel and/or speaker honoraria from Biogen, Merck, Novartis, Roche, and Sanofi; and receives research support from Actelion, the European Union, Novartis, Roche, the Swiss MS Society, and the Swiss National Foundation. J. Oechtering served on advisory boards for Roche and Merck. KG is supported by the Swiss National Science Foundation Grant No SNF212656. Advisory Boards: AbbVie, AZAD, Santen. Research support: Swiss National Science Foundation. L. Kappos received no personal compensation; his institutions (University Hospital Basel/Stiftung Neuroimmunology and Neuroscience Basel) received payments for steering committee, advisory and data safety monitoring board participation, consultancy services and educational activities from Bayer, Biogen, Bristol Myers Squibb, Celltrion Inc, Clene Nanomedicine Inc., EMD Serono Research and development, Galapagos NV, Genentech, Immunic AG, Janssen, Kiniksa Pharmaceuticals, Laboratoires Juvise Pharmaceuticals, Merck Healthcare AG, MSD Merck Sharp and Dohme AG, Minoryx Therapeutics S.L., Neurostatus UHB AG, Novartis, Roche, Sanofi, Shionogi BV, Wellmera AG, Zai Lab; research support from Novartis, Roche and Innosuisse. The University Hospital Basel (USB), as the employer of C. Granziera, has received the following fees which were used exclusively for research support: (1) advisory board and consultancy fees from Actelion, Genzyme-Sanofi, Novartis, GeNeuro and Roche; (2) speaker fees from Genzyme-Sanofi, Novartis, GeNeuro and Roche; (3) research support from Siemens, GeNeuro, Roche. J. Kuhle received speaker fees, research support, travel support, and/or served on advisory boards by Swiss MS Society, Swiss National Research Foundation (320030_189140/1 and 320030_212534/1), University of Basel, Progressive MS Alliance, Bayer, Biogen, Bristol Myers Squibb, Celgene, Merck, Novartis, Octave Bioscience, Roche, Sanofi. P. Benkert has nothing to disclose A. Papadopoulou received speaker-fees/fees for advisory boards/for consulting from Sanofi-Genzyme, Eli Lilly, AbbVie, Lundbeck, Pfizer, Organon and TEVA (all used for her research in the University of Basel) and travel support from Bayer AG, AbbVie, Teva and Hoffmann-LaRoche. Her research was supported by the University and University Hospital of Basel, the Swiss Multiple Sclerosis Society, the “Stiftung zur Förderung der gastroenterologischen und allgemeinen klinischen Forschung sowie der medizinischen Bildauswertung”, the “Freie Akademische Gesellschaft Basel” and the Swiss National Science Foundation (Project numbers: P300PB_174480 and currently: PZ00P3_216468). The other co-authors have nothing to disclose. Go to Neurology.org/NN for all disclosures.