Serum GFAP and Age Accurately Distinguish AQP4‐IgG Positive and Double Seronegative NMOSD From MOGAD After a Recent Attack

ABSTRACT

Background

Serum neurofilament‐light chain (sNfL) and glial fibrillary acidic protein (sGFAP) are associated with multiple sclerosis (MS) activity. This study aimed to identify patient‐ and disease‐level determinants of sNfL and sGFAP and to evaluate their ability to support differential diagnosis in neuromyelitis optica (NMO).

Methods

sNFL and sGFAP levels were measured in 640 patients from the OFSEP cohort (88 NMO spectrum disorder (NMOSD) with anti‐aquaporin‐4 antibodies [AQP4‐NMOSD], 28 double‐seronegative NMOSD [DN‐NMOSD], 61 myelin oligodendrocyte glycoprotein antibody–associated disease [MOGAD], 64 clinically isolated syndrome [CIS], 237 relapsing–remitting MS [RRMS], and 162 primary progressive MS [PPMS]). The association between log[sNFL] and log[sGFAP] and age, sex, time since relapse, EDSS, oligoclonal bands, and MRI characteristics was analyzed by multiple linear regression. The diagnostic performance of sNfL and sGFAP in discriminating between NMOSD and MS subgroups was evaluated using multivariable logistic regression and area under the curve (AUC).

Results

For each group, specific associations of sNFL and sGFAP levels with patient characteristics were observed. Clearest associations were observed in AQP4‐NMOSD patients. After a recent attack (≤ 90 days), multivariate analysis revealed that sGFAP and age discriminated AQP4‐NMOSD from MOGAD (AUC = 0.939 [0.910;0.982]), AQP4‐NMOSD and DN‐NMOSD from MOGAD (AUC = 0.863 [0.776;0.977]), and AQP4‐NMOSD from all other groups (AUC = 0.950 [0.918;0.987]). Diagnostic probability heatmaps based on sGFAP and age illustrate the probability of each diagnosis relative to others for bedside interpretation.

Conclusion

After an acute attack of NMOSD, risk score based on sGFAP and age may guide the management of patients before autoantibody status is available.

After a recent NMOSD attack, serum GFAP and age together provide a simple and powerful way to distinguish AQP4‐IgG NMOSD from MOGAD and related disorders. In 240 patients evaluated within 90 days of an attack from a 640‐patient cohort, this combination achieved high diagnostic accuracy (AUC up to 0.95). Probability heatmaps now enable a rapid, bedside approach to guide clinical decisions before antibody status is confirmed.

Abbreviations

AQP4‐NMOSD

Aquaporin‐4 IgG‐positive Neuromyelitis Optica Spectrum Disorder

CIS

clinically isolated syndrome

CSF

cerebrospinal fluid

DN‐NMOSD

Neuromyelitis Optica Spectrum Disorder (NMOSD) without antibodies against aquaporin‐4 IgG or myelin oligodendrocyte glycoprotein IgG

EDSS

expanded disability status scale

MOGAD

myelin oligodendrocyte glycoprotein antibody‐associated disease

NMO

neuromyelitis optica

OBCs

oligoclonal bands

OFSEP

Observatoire Français de la Sclérose En Plaques

PPMS

primary progressive multiple sclerosis

RRMS

relapsing–remitting multiple sclerosis

sGFAP

serum glial fibrillary acidic protein

sNfL

serum neurofilament‐light chain

1. Introduction

Despite recent advances in diagnostic criteria, differential diagnosis between neuromyelitis optica (NMO) subtypes and multiple sclerosis (MS) remains challenging due to overlapping clinical and imaging features.

Combinations of severe optic neuritis (ON) and longitudinally extensive transverse myelitis (LETM) were originally described as Neuromyelitis Optica Spectrum Disorders (NMOSD) [1], with NMO subgroups later defined by the presence of specific antibodies. Associated with typical clinical features, anti‐aquaporin‐4 (AQP4) and anti‐myelin oligodendrocyte glycoprotein (MOG) antibodies have become key diagnostic criteria for AQP4‐IgG positive NMOSD (AQP4‐NMOSD) and MOG‐antibody‐associated disease (MOGAD), respectively, while patients without antibodies are referred to as double seronegative NMOSD (DN‐NMOSD) [2, 3, 4, 5]. During severe first attacks, rapid probabilistic therapy improves recovery, especially for AQP4‐NMOSD and DN‐NMOSD patients who show poorer prognosis than MOGAD patients [6, 7, 8, 9, 10].

Recently, cerebrospinal fluid (CSF) granulocyte activation markers (GAM) and astrocyte damage markers (ADM) were demonstrated to differentiate NMOSD versus MS and AQP4‐NMOSD versus MOGAD and MS [11]. However, there is a need for less invasive, rapid, and accessible biomarkers to triage acute AQP4‐NMOSD, MOGAD, DN‐NMOSD and MS while awaiting results of anti‐AQP4 or anti‐MOG antibodies [12, 13]. In NMOSD, levels of serum neurofilament‐light chain (sNfL) and glial fibrillary acidic protein (sGFAP), reflecting neuroaxonal and astrocyte injury, respectively, appear higher than in MS and may serve as tools for rapid, accurate differential diagnosis of NMO subtypes and MS [14]. sNfL and sGFAP have been extensively studied in MS and correlated with disease activity and progression [15, 16]. However, contradictory results were observed concerning their accuracy to discriminate AQP4‐NMOSD from MOGAD [17, 18, 19] and data for AQP4‐NMOSD versus DN‐NMOSD are sparse [20]. Integrating age, body mass index (BMI), and kidney function in z‐scores improved the accuracy of sNfL and sGFAP in MS clinical studies [21]. Similarly, demographic, clinical, biological and MRI features should be taken into consideration when analyzing their diagnostic value in NMO.

In this study, we aimed to identify patient and disease characteristics associated with sNfL and sGFAP in NMOSD and MS subgroups using data from the OFSEP cohort, and to evaluate their accuracy for differential diagnosis in acute demyelinating syndromes.

2. Patients and Methods

2.1. Study Design and Participants

This is a retrospective exploratory analysis of the OFSEP biological collection including all patients enrolled between January 2014 and December 2018 with serum samples available and prospectively followed for at least 24 months [NCT02889965] [22]. The OFSEP registry has been authorized by the French data protection agency and an ethical committee [23]. This project uses data and samples collected as part of the OFSEP registry and was approved by the OFSEP scientific board and steering committee.

Patients enrolled in the NOMADMUS (OFSEP) registry provided written consent to participate. In accordance with the French legislation, OFSEP was approved by both the French data protection agency (Commission Nationale de l'Informatique et des Libertés [CNIL]; authorization request 914066v3) and a French ethical committee (Comité de Protection des Personnes [CPP]: reference 2019‐A03066‐51). The CPP TOURS—Centre‐West Region 1 approved the study (2019T2‐05_RIPH2 HPS) on February 26, 2019. The data processing associated with this study is registered in the CNIL registry of the Nîmes University Hospital under the reference methodology MR004 (registration no. 2226188 v0) and registered in the Health Data Hub under the identifier 22280922.

Prospectively acquired clinical, biological, and neuroimaging data were obtained from the OFSEP national cohort, where all information is systematically recorded using the EDMUS electronic medical record system [24]. Data included age, sex, time since last relapse (≤ 90 days or > 90 days), presence of more than 9 brain lesions on FLAIR MRI scans, presence of CSF oligoclonal bands (OCBs) assessed up to one year from the diagnosis, presence of at least one active lesion (contrast enhancement after gadolinium injection), presence of a longitudinally extensive transverse myelitis (spinal cord lesion extending over three or more vertebral segments, LETM), and EDSS within 3 months of sampling.

2.2. Patients' Groups

Patients were classified as NMOSD, MOGAD, clinically isolated syndrome (CIS), relapsing–remitting MS (RRMS) or primary progressive MS (PPMS) as described in Brocard et al. [22]. DN‐NMOSD patients were defined as individuals fulfilling the 2015 IPND [5]. Each DN‐NMOSD has been evaluated and validated by the NOMADMUS scientific committee based on clinical, imaging and biological data and a minimum follow‐up [25]. Classical mimics including MS and sarcoidosis have been ruled out. All patients have been tested at least once by live CBA for MOG and AQP4‐IgG. Sampling, processing and storage were performed as previously described [22]. A recent attack was defined as the onset of clinical symptoms within 90 days before sampling [20, 26]. Patients without an attack within the previous 90 days were considered to be “stable”. Clinicians who determined the attacks had no access to biomarker results.

2.3. Biomarker Measure

Serum NfL and GFAP concentrations were determined with the Simoa Human Neurology 4‐Plex “A” kit (Quanterix Corp, Boston, MA) using the Simoa HD‐1 analyzer, as described in Gauthier et al. [27]. All analyses were performed in a single analytical run using one lot of reagents, and investigators remained blinded to clinical data throughout.

The lower limits of quantification (LLOQ) were 0.8 pg/mL for sNfL and 16.6 pg/mL for sGFAP. Biomarker concentrations were analyzed as continuous variables. No diagnostic positivity cut‐offs were applied. Owing to a high proportion of values falling below the LOQ for Tau and UCH‐L1, these biomarkers were excluded from the analyses.

Assays were performed in simplicate and internal quality‐control (QC) samples supplied in the kit were included on every plate to monitor assay performance. Low‐ and high‐concentration sNfL controls (QC1: mean 4.94 pg/mL; QC2: mean 490.2 pg/mL) showed low inter‐assay variability, with coefficients of variation (CV) of 11.5% and 11.7%, respectively. Likewise, low‐ and high‐concentration sGFAP controls (QC3: mean 21.0 pg/mL; QC4: mean 1122 pg/mL) demonstrated inter‐assay CVs of 5.20% and 6.28%, respectively.

2.4. Statistical Analysis

Median sNfL and sGFAP levels and follow‐up intervals were compared across diagnostic subgroups (CIS, RRMS, PPMS, AQP4 NMOSD, MOGAD, DN NMOSD) using pairwise Wilcoxon–Mann–Whitney tests, displayed as boxplots. Proportion of treated patients was compared using Fisher's exact test. No formal sample size calculation was performed and no correction for multiple testing was applied because all comparisons were exploratory in nature [28].

Because of right skewed distributions and extreme values, sNfL and sGFAP were log transformed prior to analysis. Within MS and NMOSD subgroups, associations between log transformed biomarkers and covariates (age, sex, time since last relapse, number of brain lesions, CSF OCBs, presence of active MRI lesions, presence of LETM, EDSS) were examined using univariate linear regression models. Exponentiated regression coefficients (e ^β ) were interpreted as relative effects. For continuous predictors, e ^β represents the proportional biomarker change per unit increase; for categorical predictors, the proportional difference in biomarker concentration between groups. Associations with continuous variables (age, EDSS) were also summarized using Spearman correlation coefficients. Multivariable linear regression models were constructed by including variables with p < 0.20 from univariate analyses. These analyses were performed separately for each biomarker and each disease subgroup and represented with their associated exponentiated regression coefficients (e ^β ) and confidence intervals [95% CI]. Missing EDSS values (7%–10% depending on subgroup) resulted in case wise exclusion from multivariable models (complete case analysis). No imputation was performed. No missing data occurred for sNfL or sGFAP index test measurements within the quantifiable range.

Diagnostic accuracy for distinguishing disease subgroups in patients with a recent attack was estimated using multivariable logistic regression. Backward stepwise elimination (p < 0.05 retention threshold) was used for model selection.

The discriminative ability of the final multivariable model was quantified using the area under the receiver operating characteristic curve (AUC). To reduce overfitting, internal validation was undertaken with 1000 bootstrap resamples, allowing estimation of optimism and calculation of an optimism‐corrected AUC. For each model, predicted probabilities were computed from the logistic transformation of the linear predictor (risk score) and visualized in probability heatmaps. Predicted probabilities were calculated from the logistic transformation of the linear predictor (risk‐score) as Probability = exp.(risk‐score)/[1 + exp. (risk‐score)].

Statistical analyses were performed using SAS software 9.4 and “R” 3.6.1. A p‐value < 0.05 was considered statistically significant.

3. Results

3.1. Patients' Characteristics

This retrospective study included 640 patients: 88 AQP4‐NMOSD, 61 MOGAD, 28 DN‐NMOSD and 463 MS patients (64 CIS, 237 RRMS, and 162 PPMS) from the OFSEP cohort.

Median age was higher in AQP4‐NMOSD (50 [33–61] years) and DN‐NMOSD (44 [31–60] years) compared with MOGAD (38 [26–47] years; p = 0.002 and p = 0.09) and in PPMS (51 [44, 55]) than in CIS and RRMS patients (37 [30, 47] and 33 [26, 41], respectively, both p < 0.001; Table 1).

TABLE 1.

Patient characteristics.

	AQP4‐NMOSD	DN‐NMOSD	MOGAD	CIS	RRMS	PPMS
	N = 88	N = 28	N = 61	N = 64	N = 237	N = 162
Age (years)	50 [33, 61]	44 [31, 60]	38 [26, 47]	37 [30, 47]	33 [26, 41]	51 [44, 55]
Female sex	80 (91%)	14 (52%)	32 (53%)	42 (66%)	169 (71%)	88 (54%)
Disease duration (years)	5.8 [1.3, 10.4]	6.6 [1.6, 13.6]	1.6 [0.5, 5.1]	0.2 [0.1, 0.4]	0.1 [0.1, 0.3]	0.6 [0.2, 3.7]
Time since last relapse (days)	402.0 [99.0, 1074.0]	409.0 [49.0, 1672.0]	179.0 [56.0, 517.5]	83.0 [33.5, 155.0]	51.0 [22.0, 109.0]	261.5 [88.0, 962.0]
EDSS score	4.0 [2.0, 6.5]	4.5 [3.5, 7.0]	1.5 [0, 3.0]	1.0 [0, 2.0]	1.5 [1.0, 2.0]	4.0 [3.0, 5.5]
Missing (N)	16	4	9	5	14	10
Presence of LETM	37 (49%)	8 (30%)	8 (16%)	1 (1.9%)	1 (0.6%)	2 (1.6%)
Missing	12	1	10	11	70	9
Number of brain lesions
< 9	54 (78%)	21 (84%)	43 (81%)	48 (81%)	86 (43%)	27 (20%)
≥ 9	15 (22%)	4 (16%)	10 (19%)	11 (19%)	116 (57%)	105 (80%)
Missing (N)	19	3	8	5	35	30
Presence of Gd+ lesion(s)	10 (63%)	3 (43%)	4 (31%)	16 (52%)	79 (55%)	8 (17%)
Missing (N)	72	21	48	33	93	116
Presence of CSF OCBs	18 (40%)	6 (38%)	5 (14%)	22 (42%)	170 (89%)	91 (89%)
Missing (N)	43	12	25	11	47	60
sNfL level (pg/mL)	17.1 [8.8, 32.8]	12.5 [7.9, 31.3]	10.1 [6.7, 21.5]	8.3 [6.0, 13.51]	11.5 [7.2, 18.5]	12.6 [9.6, 17.5]
sGFAP level (pg/mL)	145.5 [95.6, 234.3]	92.3 [78.5, 129.2]	88.8 [55.2, 135.3]	67.8 [48.8, 92.2]	76.8 [59.7, 105.9]	107.2 [77.9, 136.4]

Note: Data are shown as median [IQR] or number (%). no.: number; total: number of patients with available data.

Abbreviations: AQP4‐NMOSD, aquaporin‐4 antibody‐positive neuromyelitis optica spectrum disorder; CIS, clinically isolated syndrome; CSF OCBs, cerebrospinal fluid oligoclonal bands; DN‐NMOSD, Neuromyelitis Optica Spectrum Disorder (NMOSD) without antibodies against aquaporin‐4 (AQP4‐IgG) or myelin oligodendrocyte glycoprotein (MOG‐IgG); EDSS, expanded disability status scale; Gd+ lesions: gadolinium‐enhancing lesions; LETM, longitudinally extensive transverse myelitis; MOGAD, myelin oligodendrocyte glycoprotein antibody‐associated disease; PPMS, primary progressive multiple sclerosis; RRMS, relapsing–remitting multiple sclerosis; sGFAP, serum glial fibrillary acidic protein; sNfL, serum neurofilament‐light chain.

EDSS at sampling was higher in AQP4‐NMOSD (4.0 [2.0–6.5]), and DN‐NMOSD (4.0 [3.0–7.0]) than in MOGAD (1.5 [0–3.0], p < 0.001 and p < 0.001, respectively, Table 1). Disease duration was longer in AQP4‐NMOSD, DN‐NMOSD (5.8 and 7.1 years, respectively) than in MOGAD (1.6 years, p = 0.002 and p = 0.009, respectively) and RRMS (0.14 years, p < 0.001 and p < 0.001, respectively). LETM was observed in 49% of AQP4‐NMOSD, 30% of DN‐NMOSD, 16% of MOGAD and in < 2% of MS groups. Presence of CSF OCBs was above 90% in RRMS and PPMS, and below 50% in other groups (42.0% in CIS, 40.0% in AQP4‐NMOSD, 38.0% in DN‐NMOSD and 14.0% in MOGAD, Table 1).

3.2. sNfL and sGFAP Levels in NMOSD and MS Subgroups

sNfL and sGFAP levels showed a high variability among individuals in each group (Figure 1). The highest median sNfL level was observed in the AQP4‐NMOSD group (17.1 pg/mL [8.8–32.8]), and was significantly higher than in PPMS (12.6 pg/mL [9.6–17.5], p = 0.005), RRMS (11.5 pg/mL [7.2–18.5], p = 0.001), MOGAD (10.1 pg/mL [6.7–21.5], p < 0.05), and CIS (8.3 pg/mL [6.0–13.5], p < 0.0001) groups (Table 1, Figure 1A). sNfL levels in the DN‐NMOSD group (12.5 pg/mL [7.9–31.3]) were not different from those observed in the AQP4‐NMOSD and the MOGAD groups (p = 0.33 and p = 0.32, respectively). sNfL levels in the RRMS group were higher than in the CIS group (p = 0.003) and lower than in the PPMS group (p = 0.05).

FIGURE 1.

sNfL and sGFAP concentration among different populations of NMO and MS patients. Values of sNFL (A) and sGFAP (B) in pg/mL among different populations of NMO and MS patients are represented as box‐and‐whisker plots (the edges of the box are Q1 and Q3, with the central line denoting the median; the whiskers extend to 1.5 times the interquartile range (IQR) beyond Q1 and Q3). Values of sNFL (C) and sGFAP (D) in pg/mL among different populations of NMO and MS patients > 90 days (stable) or ≤ 90 days (attack) of a clinical attack. AQP4‐NMOSD, aquaporin‐4 antibody‐positive neuromyelitis optica spectrum disorder; CIS, clinically isolated syndrome; DN‐NMOSD, Neuromyelitis Optica Spectrum Disorder (NMOSD) without antibodies against aquaporin‐4 (AQP4‐IgG) or myelin oligodendrocyte glycoprotein (MOG‐IgG); MOGAD, myelin oligodendrocyte glycoprotein antibody‐associated disease; PPMS, primary progressive multiple sclerosis; RRMS, relapsing–remitting multiple sclerosis; sGFAP, serum glial fibrillary acidic protein; sNfL, serum neurofilament‐light chain. Wilcoxon‐Mann–Whitney test; *** and ^✧✧✧ p < 0.001; **p < 0.01; *p < 0.05.

Similarly, the highest median sGFAP level was also observed in the AQP4‐NMOSD group (145.5 pg/mL [95.6–234.3]), and was significantly higher than in the PPMS (107.2 pg/mL [77.9–136.4], p < 0.0001), DN‐NMOSD (92.3 pg/mL [78.5–129.2], p = 0.005), MOGAD (88.8 pg/mL [55.2–135.3], p < 0.0001), RRMS (76.8 pg/mL [59.7–105.9], p < 0.001), and CIS (67.8 pg/mL [48.8–92.2], p < 0.001) groups (Table 1, Figure 1B). sGFAP levels of the RRMS group were higher than in the CIS group (p = 0.03) and lower than in the PPMS group (p < 0.001).

3.3. Association of sNfL With Patients' Characteristics in Each Group

Linear regression models showed that log[sNFL] was strongly correlated with age in the DN‐NMOSD group (R = 0.62, p < 0.0001) and weakly correlated in other groups (see eFigure 1). For PPMS, we observed elevated levels in both younger and older patients, showing a U‐shaped distribution (eFigure 1). Similarly, log[sNFL] was moderately correlated with EDSS in AQP4‐NMOSD, DN‐NMOSD, and MOGAD while weakly correlated in other groups (eFigure 2).

Multivariate analysis revealed that log[sNfL] levels were associated with age in DN‐NMOSD, MOGAD, CIS, and PPMS groups, with EDSS in AQP4‐NMOSD, MOGAD, CIS, and RRMS groups, with sex in MOGAD patients, with the number of lesions in RRMS patients, and with the presence of active lesions in the AQP4‐NMOSD group, but not with the presence of CSF OCBs or LETM (eTable 1).

3.4. Association of sGFAP With Patient Characteristics in Each Group

In linear regression models, log[sGFAP] was strongly correlated with age and EDSS in the DN‐NMOSD group (R = 0.62, p = 0.0005, eFigure 1 and R = 0.60, p = 0.003, eFigure 2).

Multivariate analysis revealed that log[sGFAP] levels were associated with age in CIS and PPMS groups, with EDSS in AQP4‐NMOSD, DN‐NMOSD, and RRMS groups, with the presence of CSF OCBs in CIS and in RRMS patients, and with the number of brain lesions in RRMS, but not with the presence of active lesions or of LETM (eTable 2).

3.5. sNfL and sGFAP Levels After a Recent Attack

A recent attack was reported in 240 patients (37.5%), including 21 AQP4‐NMOSD (23.9%), 19 MOGAD (31.7%), 8 DN‐NMOSD (29.6%), 34 CIS (53.1%) and 158 RRMS (66.7%).

Following an attack, log[sNFL] and log[sGFAP] were strongly or very strongly correlated in NMO patients (R = 0.74, p < 0.001 for AQP4‐NMOSD, R = 0.85, p = 0.011 for DN‐NMOSD), strongly correlated in MOGAD patients (R = 0.68, p < 0.001), and moderately correlated in MS patients (R = 0.58, p < 0.001 for CIS and R = 0.56, p < 0.001 for RRMS) (eFigure 3).

Characteristics of patients with a recent attack and those with stable disease are compared in Table 2.

TABLE 2.

Characteristics of patients with a recent attack (symptom onset ≤ 90 days before sampling).

n	AQP4‐NMOSD (N = 88)			DN‐NMOSD (N = 27)			MOGAD (N = 60)			CIS (N = 64)			RRMS (N = 237)
	Attack	Stable	p	Attack	Stable	p	Attack	Stable	p	Attack	Stable	p	Attack	Stable	p
	N = 21	N = 67		N = 8	N = 19		N = 19	N = 41		N = 34	N = 30		N = 158	N = 79
Age (years)	52 [47, 60]	45 [32, 61]	0.3	36 [26, 53]	46 [32, 63]	0.3	33 [24, 41]	41 [31, 51]	0.029	34 [26, 41]	37 [31, 52]	0.053	32 [25, 40]	35 [27, 43]	0.15
Female sex	19 (90%)	61 (91%)	> 0.9	5 (63%)	9 (47%)	0.7	11 (58%)	21 (51%)	0.6	22 (65%)	20 (67%)	0.9	111 (70%)	58 (73%)	0.6
Disease duration (years)	1.4 [0.2, 4.9]	6.5 [2.7, 11.3]	< 0.001	1.35 [0.03, 6.35]	7.9 [4.9, 14.7]	0.009	0.5 [0.2, 6.3]	1.94 [0.62, 4.91]	0.11	0.10 [0.03, 0.17]	0.44 [0.34, 0.51]	< 0.001	0.09 [0.04, 0.14]	0.40 [0.30, 0.50]	< 0.001
Time since last relapse (days)	33 [25, 55]	822 [256, 1372]	< 0.001	22 [11, 48]	1033 [263, 3118]	< 0.001	41 [23, 55]	310 [174, 661]	< 0.001	37 [11, 63]	161 [124, 185]	< 0.001	32 [14, 51]	145 [109, 184]	< 0.001
EDSS score	6.5 [2.0, 8.0]	4.0 [2.0, 6.0]	0.08	6.5 [3.0, 8.0]	4.0 [3.5, 5.5]	0.3	2.0 [1.0, 3.0]	1.0 [0, 3.0]	0.3	1.0 [1.0, 2.0]	1.0 [0 1.5]	0.7	1.5 [1.0, 2.0]	1.5 [0, 2.0]	0.073
Missing (N)	3	13		1	2		1	8		3	2		10	4
Presence of LETM	9 (45%)	28 (50%)	0.4	1 (14%)	7 (37%)	0.5	3 (6.7%)	5 (14%)	0.2	1 (3.8%)	0 (0%)	0.5	1 (1%)	0 (0%)	> 0.9
Missing (N)	1	11		1	0		4	6		8	3		55	15
Number of brain lesions			0.5			0.6			0.5			0.2			0.6
< 9	14 (88%)	40 (75%)		5 (71%)	15 (88%)		12 (75%)	30 (83%)		22 (73%)	26 (90%)		54 (41%)	32 (46%)
≥ 9	2 (13%)	13 (25%)		2 (29%)	2 (12%)		4 (25%)	6 (17%)		8 (27%)	3 (10%)		78 (59%)	38 (54%)
Missing (N)	5	14		1	2		3	5		4	1		26	9
Presence of Gd+ lesion(s)	6 (86%)	4 (44%)	0.15	3 (75%)	0 (0%)	0.14	2 (40%)	2 (25%)	> 0.9	14 (64%)	2 (22%)	0.054	67 (63%)	12 (32%)	0.002
Missing (N)	14	58		4	16		14	33		12	21		51	42
Presence of CSF OCBs	7 (64%)	11 (32%)	0.086	3 (60%)	3 (30%)	0.3	4 (33%)	1 (4.3%)	0.038	12 (43%)	10 (40%)	> 0.9	108 (87%)	62 (94%)	0.2
Missing (N)	10	33		3	9		7	18		6	5		34	13
sNfL levels (pg/mL)	34.1 [21.9, 103.2]	13.1 [6.9, 22.2]	< 0.001	57.7 [8.5, 312.7]	12.0 [7.7, 15.5]	0.2	13.6 [7.9, 36.6]	9.8 [6.7, 15.5]	0.2	9.3 [6.3, 14.2]	7.8 [5.5, 13.2]	0.4	11.5 [7.3, 20.5]	10.0 [6.6, 16.6]	0.2
sGFAP levels (pg/mL)	215.9 [162.4, 635.3]	127.2 [91.7, 177.9]	< 0.001	175.8 [66.4, 484.9]	84.2 [77.7, 112.7]	0.3	71.1 [42.6, 153.3]	91.9 [59.9, 135.3]	0.2	64.3 [44.0, 85.9]	71.0 [56.2, 94.0]	0.1	74.6 [56.3, 105.9]	82.1 [64.7, 106.6]	0.2

Note: Data are shown as median [IQR] or number (%). no.: number; total: number of patients with available data.

Abbreviations: AQP4‐NMOSD, aquaporin‐4 antibody‐positive neuromyelitis optica spectrum disorder; CIS, clinically isolated syndrome; CSF OCBs, cerebrospinal fluid oligoclonal bands; DN‐NMOSD, Neuromyelitis Optica Spectrum Disorder (NMOSD) without antibodies against aquaporin‐4 (AQP4‐IgG) or myelin oligodendrocyte glycoprotein (MOG‐IgG); EDSS, expanded disability status scale; Gd+ lesions, gadolinium‐enhancing lesions; LETM, longitudinally extensive transverse myelitis; MOGAD, myelin oligodendrocyte glycoprotein antibody‐associated disease; RRMS, relapsing–remitting multiple sclerosis; sGFAP, serum glial fibrillary acidic protein; sNfL, serum neurofilament‐light chain. Significant p‐values are indicated in bold.

Disease duration was significantly shorter in the attack versus stable groups except for MOGAD. The median interval between attack onset and sample collection was not statistically different among groups (33 days [21–57] in AQP4‐NMOSD, 22 days [11–48] in DN‐NMOSD, 41 days [23–55] in MOGAD, 37 days [11–64] in CIS and 32 days [14–51] in RRMS). In the MOGAD group, CSF OCBs were more frequent in the attack subgroup than in the stable subgroup (33% vs. 4%, respectively, p = 0.038). Presence of gadolinium‐enhancing lesions was more frequent in the attack subgroup than in the stable subgroup in RRMS (63% vs. 32%, respectively, p = 0.002) and in CIS (64% vs. 22%, respectively, p = 0.054). Proportion of treated patients in AQP4‐NMOSD group (52.4%) was not significantly different as compared to DN‐NMOSD (37.5%, p = 0.6817) and MOGAD (21.1%, p = 0.0550) but higher than in CIS and RRMS (0.0%, p < 0.0001). sNfL and sGFAP were not different between treated and untreated patients within each group. The highest median sNfL level after a recent attack was observed in the DN‐NMOSD and AQP4‐NMOSD groups (57.7 pg/mL [8.5–313.0] and 34.1 pg/mL [21.9–103.2], respectively). It was significantly higher in AQP4‐NMOSD than in MOGAD (13.6 pg/mL [7.9–36.6], p = 0.003), RRMS (11.5 pg/mL [7.3–20.5], p < 0.001), and CIS (9.3 pg/mL [6.3–14.2], p < 0.001) (Figure 1C).

The highest median sGFAP level after a recent attack was observed in AQP4‐NMOSD (215.9 pg/mL [162.4–635.3]), and was significantly higher than in MOGAD (71.1 pg/mL [42.6–153.3], p < 0.0001), RRMS (74.6 pg/mL [56.3–105.9], p < 0.0001), and CIS (64.3 pg/mL [44.0–85.9], p < 0.0001) (Figure 1D).

In the AQP4‐NMOSD, DN‐NMOSD, and MOGAD groups, sNfL and sGFAP levels determined after the first attack (around one third of patients) or after subsequent attacks were not different, with similar time from attack to sampling in each group (eTable 3).

As compared to stable patients, sNfL and sGFAP levels significantly increased after a recent attack in AQP4‐NMOSD patients (34.1 pg/mL vs. 13.1 pg/mL, p < 0.001, and 215.9 pg/mL vs. 127.2 pg/mL, p < 0.001, respectively, Figure 1C,D, Table 2), but not in other groups, although a large increase of sNfL and sGFAP levels was observed after a recent attack in DN‐NMOSD patients (57.7 pg/mL vs. 12.0 pg/mL, p = 0.2, and 175.8 pg/mL vs. 84.2 pg/mL, p = 0.3, respectively, Figure 1C,D, Table 2). After a recent attack, AQP4‐NMOSD and DN‐NMOSD patients had high median EDSS scores (6.5), although LETM was observed in 45% of AQP4‐NMOSD patients and 14% of DN‐NMOSD patients (Table 2).

3.6. Accuracy of Biomarkers for Differential Diagnosis After a Recent Attack

After a recent attack, univariate logistic analysis showed that log[sNFL], log[sGFAP], EDSS and age discriminated AQP4‐NMOSD from all other groups, and distinguished AQP4‐NMOSD from MOGAD, and AQP4‐NMOSD and DN‐NMOSD from MOGAD (Figure 2, Table 3). Sex discriminated AQP4‐NMOSD from all other groups and AQP4‐NMOSD from MOGAD (Figure 2, Table 3). None of these factors discriminated AQP4‐NMOSD from DN‐NMOSD.

FIGURE 2.

Univariate associations between serum biomarkers, clinical factors, and diagnostic categories in AQP4‐NMOSD, DN‐NMOSD, MOGAD, and MS. Forest plots display odds ratios (OR) with 95% confidence intervals (CI) from univariate logistic regression analyses assessing the association of clinical variables (age, sex, EDSS) and serum biomarkers (log‐transformed NfL and GFAP) with diagnostic groups. Comparisons include: AQP4‐NMOSD versus all other diagnoses, AQP4‐NMOSD versus MOGAD, AQP4‐NMOSD versus DN‐NMOSD, and AQP4‐NMOSD/DN‐NMOSD versus MOGAD. OR values > 1 indicate higher odds of belonging to the AQP4‐NMOSD group. The vertical dashed line represents the null value (OR = 1). p‐values correspond to the significance of each univariate model.

TABLE 3.

Accuracy of biomarkers for differential diagnosis after a recent attack.

	Univariate analysis				Multivariate analysis
	OR [95% CI]	p	AUC [95% CI]		OR [95% CI]	p	AUC [95% CI]
AQP4‐NMOSD versus MOGAD/DN‐NMOSD/RRMS/CIS
Log[sNfL]	2.27 [1.60;3.22]	< 0.001	0.850 [0.793;0.902]
Log[sGFAP]	7.82 [3.76;16.27]	< 0.001	0.933 [0.891;0.974]		8.984 [3.70;21.8]	< 0.001	0.950 [0.918;0.987]
Age	0.912 [0.880;0.945]	< 0.001	0.787 [0.644;0.911]		1.11 [1.06;1.17]	< 0.001
Sex (M)	4.46 [1.01;19.7]	0.048	0.612 [0.532;0.674]
EDSS	0.613 [0.509;0.740]	< 0.001	0.798 [0.661;0.916]
AQP4‐NMOSD versus MOGAD
Log[sNfL]	2.46 [1.20;5.01]	0.013	0.772 [0.596;0.917]
Log[sGFAP]	8.44 [2.23;32.0]	0.002	0.880 [0.747;0.972]		8.95 [1.72;46.8]	0.009	0.939 [0.910;0.982]
Age	0.905 [0.850;0.965]	0.002	0.827 [0.680;0.952]		1.20 [1.04;1.40]	0.020
Sex (M)	6.91 [1.24;38.5]	0.027	0.663 [0.536;0.789]
EDSS	0.737 [0.573;0.948]	0.018	0.745 [0.562;0.894]
AQP4‐NMOSD versus DN‐NMOSD
Log[sNfL]	1.02 [0.602;1.72]	0.948	0.470 [0.185;0.738]
Log[sGFAP]	1.95 [0.803;4.74]	0.140	0.655 [0.381;0.899]
Age	0.964 [0.918;1.01]	0.135	0.682 [0.440;0.893]
Sex (M)	5.70 [0.739;43.9]	0.095	0.640 [0.452;0.812]
EDSS	1.10 [0.809;1.51]	0.533	0.599 [0.381;0.821]
AQP4‐NMOSD/DN‐NMOSD versus MOGAD
Log[sNfL]	2.02 [1.15;3.53]	0.014	0.733 [0.572;0.868]
Log[sGFAP]	4.17 [1.70;10.2]	0.002	0.826 [0.693;0.938]		3.16 [1.34;7.41]	0.008	0.863 [0.776;0.977]
Age	1.08 [1.03;1.13]	0.003	0.77 [0.628;0.891]		1.09 [1.02;1.17]	0.018
Sex (M)	0.286 [0.076;1.08]	0.065	0.624	[0.493;0.764]
EDSS	1.42 [1.10;1.82]	0.006	0.776	[0.619;0.901]

Note: Univariate and multivariate analysis are shown in columns. Results are shown in p‐value and odd ratios (OR) with [95% CI]. Area under the curve (AUC) with [95% CI] are also depicted for univariate analysis and corrected AUC with [95% CI] for multivariate analysis (see Methods). Risk‐score equations allowing estimation of the probability of first diagnosis versus second diagnosis: AQP4‐NMOSD versus MOGAD/DN‐NMOSD/RRMS/CIS score = −17.7080 +2.1954 * Log[sGFAP] +0.1066 * age; AQP4‐NMOSD versus MOGAD score = −18.3978 +2.19203 * Log[sGFAP] +0.1828 * age; AQP4‐NMOSD/DN‐NMOSD versus MOGAD score = −8.602 +1.1496 Log[sGFAP] +0.0866 * age. Significant p‐values are indicated in bold.

Abbreviations: AQP4‐NMOSD, aquaporin‐4 antibody‐positive neuromyelitis optica spectrum disorder; DN‐NMOSD, Neuromyelitis Optica Spectrum Disorder (NMOSD) without antibodies against aquaporin‐4 (AQP4‐IgG) or myelin oligodendrocyte glycoprotein (MOG‐IgG); EDSS, expanded disability status scale; MOGAD, myelin oligodendrocyte glycoprotein antibody‐associated disease.

In multivariate analysis, a model including log[sGFAP] and age discriminated AQP4‐NMOSD from all other groups (AUC 0.950 [0.918;0.987]), AQP4‐NMOSD from MOGAD (AUC 0.939 [0.910;0.982]), and AQP4‐NMOSD and DN‐NMOSD from MOGAD (AUC 0.863 [0.776;0.977]) (Table 3).

3.7. Risk‐Score of AQP4‐NMOSD in Acute NMO

In the subgroup of patients with a recent attack (n = 240), risk‐score equations allowing estimation of the probability of AQP4‐NMOSD versus all other groups, of AQP4‐NMOSD versus MOGAD and of AQP4‐NMOSD/DN‐NMOSD versus MOGAD are depicted in Table 3 footnotes. Heatmaps representing the probability of being AQP4‐NMOSD versus all other groups, AQP4‐NMOSD versus MOGAD and of being AQP4‐NMOSD/DN‐NMOSD versus MOGAD according to sGFAP levels and age illustrate the influence of these factors (Figure 3). For instance, a 60‐year‐old patient with sGFAP levels of 5, 10 and 40 pg/mL has a probability of being AQP4‐NMOSD versus MOGAD of 9%, 46%, and 94%, respectively, while the same probability is obtained for sGFAP levels of 100, 200, and 400 pg/mL, respectively, in a 20‐year‐old patient (Figure 3).

FIGURE 3.

Probability heatmaps of belonging to one diagnostic group, against other groups, depending on sGFAP concentration and age. Probability of being AQP4‐NMOSD versus MOGAD/DN‐NMOSD/RRMS/CIS (top), AQP4‐NMOSD versus MOGAD (middle) or AQP4‐NMOSD/DN‐NMOSD versus MOGAD (bottom) calculated from the respective risk‐scores depicted in Table 3 according to the formula: Probability = exp. (score)/[1 + exp. (score)]. AQP4‐NMOSD, aquaporin‐4 antibody‐positive neuromyelitis optica spectrum disorder; DN‐NMOSD, Neuromyelitis Optica Spectrum Disorder (NMOSD) without antibodies against aquaporin‐4 (AQP4‐IgG) or myelin oligodendrocyte glycoprotein (MOG‐IgG); MOGAD, myelin oligodendrocyte glycoprotein antibody‐associated disease.

4. Discussion

In samples collected within 90 days of an acute NMO attack, sGFAP levels and age accurately discriminate AQP4‐NMOSD versus all other groups, AQP4‐NMOSD from MOGAD and AQP4‐NMOSD/DN‐NMOSD from MOGAD. In line with previous studies, sNfL and sGFAP increased significantly during AQP4‐NMOSD attacks (Figure 1C,D), suggesting severe astrocytopathy and associated neuronal injury in this condition [17, 18], while the absence of differences in sNfL and sGFAP between attack and stable phases in MOGAD is unexpected and may reflect lesion topography, timing of sampling, or sample size. Accordingly, high sNfL levels were associated with the time since last relapse and the presence of gadolinium‐enhancing lesions in AQP4‐NMOSD. However, in line with the massive anti‐AQP4‐IgG‐mediated astrocytic damage in this condition, sGFAP had higher accuracy than sNfL to decipher AQP4‐NMOSD from other groups. Although sNfL was accurate in univariate analysis, it was not retained in the final model (Table 3), potentially because sNfL and sGFAP levels were strongly correlated (eFigure 3). Moreover, despite high EDSS scores in AQP4‐NMOSD and DN‐NMOSD patients, sGFAP and age overpassed EDSS in multivariate models (Table 3), suggesting that pathophysiological markers are more specific than the location of the CNS injury.

We analyzed the distribution and factors influencing the values of sNfL and sGFAP levels in 173 NMO and 463 MS patients, supporting previous studies that showed elevated sNfL and sGFAP levels in AQP4‐NMOSD [17, 18, 19, 20] and PPMS [29], although these reflect distinct pathological features between conditions. Similarly, we confirmed the higher levels of sNfL and sGFAP in PPMS versus RRMS, potentially reflecting the compartmentalization of inflammation with diffuse astrocytic activation in PPMS, although this may partially reflect the older age of PPMS patients (Figure 1) [16]. Interestingly, sGFAP levels were also associated with the presence of CSF OCBs in CIS and in RRMS, suggesting the link between diffuse inflammation and intrathecal synthesis (eTable 2) [30].

Overall, the correlations between sNfL levels and age were markedly weaker than those reported in healthy controls, indicating that disease‐related factors modulate sNfL dynamics differently across pathological conditions [31]. Multivariate models including biomarkers and patient disease characteristics confirmed the association of age with sNfL and sGFAP in CIS and PPMS (eTable 1 and eTable 2) [16]. They also revealed an association between age and sNfL in MOGAD and DN‐NMOSD that was not observed in AQP4‐NMOSD, suggesting a negligible effect of age on sNfL levels compared with the severe neuroaxonal damage in this disease (eTable 1). However, as previously reported by Kim et al., regression analysis showed that sGFAP did not correlate with age in AQP4‐NMOSD or MOGAD (eFigure 1), suggesting the absence of chronic astrocytic activation and of smoldering disease in AQP4‐NMOSD and MOGAD, contrarily to MS [32]. This is further supported by the absence of correlation of sGFAP and EDSS in MOGAD (eFigure 2), with lower astrocytic damage in MOGAD than in AQP4‐NMOSD during the acute phase [17, 19], in line with the absence of a progressive stage in MOGAD [2]. This profile was also observed in CIS patients (eFigure 2), suggesting weaker astrocytic pathology at the early stages of MS compared with RRMS and PPMS [30].

Our results replicated and extended some from Carta et al., showing a moderate correlation between sNfL and EDSS in DN‐NMOSD (eFigure 2), although not confirmed in multivariate analysis (eTable 1) [20]. In contrast, sGFAP levels were very high in AQP4‐NMOSD and DN‐NMOSD in our study, unlike previous serum and CSF studies that reported significantly lower GFAP levels in DN‐NMOSD [20, 33]. Nevertheless, sGFAP levels correlated with EDSS in DN‐NMOSD, suggesting intense astrocytic damage contributing to disability in DN‐NMOSD (eTable 2 and eFigure 2), suggesting common pathological features with AQP4‐NMOSD (Figure 1) [20]. This is further supported by the poor prognosis of DN‐NMOSD attacks, comparable to that of AQP4‐NMOSD [8].

Given that index attack and subsequent attacks showed similar sGFAP levels (eTable 3), the risk score based on age and sGFAP levels could help distinguish AQP4‐NMOSD and DN‐NMOSD from MOGAD patients while awaiting results of anti‐AQP4 and anti‐MOG antibody testing. If validated in other cohorts, a high risk score could prompt clinicians to prescribe immediate PLEX with high‐dose methylprednisolone therapy, known to reduce the risk of disability in AQP4‐NMOSD patients and generally recommended as first‐line in DN‐NMOSD [6, 8, 9]. The risk score seems less accurate at differentiating AQP4‐NMOSD/DN‐NMOSD versus MOGAD than at differentiating AQP4‐NMOSD versus MOGAD (Figure 3), although this result remains exploratory due to heterogeneity and the small number of DN‐NMOSD patients.

This study has limitations. Although data were prospectively collected in the OFSEP cohort, this study was based on samples available in the OFSEP biocollection, potentially introducing selection bias. Moreover, samples from NMO patients were obtained during follow‐up in the OFSEP cohort, whereas those from MS patients were collected primarily at diagnosis. This explains that some NMO patients were treated by disease modifying drugs (DMDs) but not MS patients. However, DMDs were heterogeneous, precluding integration in our analyses. This remains an important question that should be explored in larger cohorts, especially integrating new therapies (anti‐IL6R, anti‐CD19 and anti‐complement) that were not used in our cohort. However, the patients' characteristics are largely representative of the expected features of adult MS and NMO patients. However, detailed topography and extent of inflammatory lesions were not analyzed, with a potential confounding effect.

Presence of anti‐GFAP antibodies was not tested in our cohort and we could not exclude the presence of patients with GFAP astrocytopathy (GFAP‐A) in the DN‐NMOSD group [25]. Indeed, GFAP‐A patients showed similar sGFAP concentrations as compared to AQP4‐NMOSD patients and can present with LETM suggestive of NMO, although associated with extraspinal symptoms simultaneously or preceding myelitis onset [34, 35]. Moreover, the efficacy of PLEX has not been evaluated in GFAP‐A.

sNfL and sGFAP were determined by Simoa and results must be confirmed using other accurate and more accessible techniques that are progressively implemented in clinical routine and known to be highly correlated to Simoa results (e.g., Ella, Fujirebio, Lumipulse, Elecsys, Alinity and MSD) [36, 37, 38]. Moreover, they were not corrected for BMI and renal function as these data were lacking, precluding z‐score calculation whose performance should be explored in prospective validations studies.

The proportion of missing data differed between groups, as different diagnostic work‐ups were performed in NMO and MS, reducing the integration of covariables in the multivariate models, especially for the presence of CSF OCBs (49% in AQP4‐NMOSD, 25% in MOGAD, and 45% in DN‐NMOSD), although kFLC index recently performed better than CSF OCBs to discriminate AQP4‐NMOSD from MOGAD [39]. Moreover, the presence of CSF OCBs may be transient in MOGAD, as suggested by the different prevalence of CSF OCBs after a recent attack versus stable MOGAD patients (Table 2) [40].

Finally, in the probability heatmaps presented in Figure 3, the validity of the probability estimates depends on the representativeness of our dataset, which could be reduced in future validation studies, especially given the increasing prevalence of MOGAD versus AQP4‐NMOSD and regional or ethnic disparities. Moreover, the scores were based on data from NMO attacks that generally occurred during follow‐up of NMO patients, which may differ from the first attack [41]. Altogether, the use of risk‐score equations should be restricted to the context of differential diagnosis between AQP4‐NMOSD, DN‐NMOSD, MOGAD, and CIS/MS after a recent NMO attack.

5. Conclusion

sNfL and sGFAP levels showed different associations with patients' characteristics and disease activity and severity in NMO and MS subgroups, potentially reflecting differences in their underlying pathophysiology. Our results indicate factors to correct on when assessing the diagnostic value of sNfL and sGFAP.

After a recent NMO attack, our data point out sGFAP as a candidate triage biomarker to suspect AQP4‐NMOSD diagnosis while awaiting antibody confirmation via cell‐based assays. It also highlights clinical and pathophysiological similarities between AQP4‐NMOSD and DN‐NMOSD, thereby further strengthening the potential for immediate PLEX in NMO attacks with a high‐risk score.

Author Contributions

Manon Rival: investigation, writing – original draft, writing – review and editing, formal analysis, data curation, validation. Fabien Rollot: methodology, writing – review and editing, writing – original draft, formal analysis. Sabine Laurent‐Chabalier: writing – original draft, writing – review and editing, methodology, formal analysis. Romain Casey: methodology, writing – review and editing. Illiasse El Bahi: formal analysis. Guillaume Brocard: writing – review and editing, resources. Romain Marignier: conceptualization, writing – review and editing, validation, investigation. Sophie Trouillet‐Assant: formal analysis, writing – review and editing. Sandra Vukusic: writing – review and editing, investigation. Agnès Fromont: investigation, writing – review and editing. David Axel Laplaud: investigation, writing – review and editing. Guillaume Mathey: investigation, writing – review and editing. Elisabeth Maillart: investigation, writing – review and editing. Laure Michel: investigation, writing – review and editing. Aurélie Ruet: investigation, writing – review and editing. Pierre Labauge: investigation, writing – review and editing. Xavier Ayrignac: investigation, writing – review and editing. Hélène Zephir: investigation, writing – review and editing. Jonathan Ciron: investigation, writing – review and editing. Jérôme de Sèze: investigation, writing – review and editing. Arnaud Kwiatkowski: investigation, writing – review and editing. Caroline Papeix: investigation, writing – review and editing. Pierre Branger: investigation, writing – review and editing. Christine Lebrun‐Frenay: investigation, writing – review and editing. Eric Berger: investigation, writing – review and editing. Pierre Clavelou: investigation, writing – review and editing. Olivier Heinzlef: investigation, writing – review and editing. Jean Pelletier: investigation, writing – review and editing. Abdullatif Al‐Khedr: investigation, writing – review and editing. Olivier Casez: investigation, writing – review and editing. Bertrand Bourre: investigation, writing – review and editing. Alain Créange: investigation, writing – review and editing. Thomas David: investigation, writing – review and editing. Laurent Magy: investigation, writing – review and editing. Jean‐Philippe Camdessanche: investigation, writing – review and editing. Inès Doghri: investigation, writing – review and editing. Amélie Dos Santos: investigation, writing – review and editing. Thibault Mura: conceptualization, writing – original draft, writing – review and editing, methodology. Hanane Agherbi: investigation, data curation, writing – review and editing, writing – original draft. Eric Thouvenot: conceptualization, investigation, funding acquisition, writing – original draft, writing – review and editing, validation, formal analysis, supervision.

Funding

This work was supported by Novartis. Data collection has been supported by a grant provided by the French State and handled by the “Agence Nationale de la Recherche,” within the framework of the “France 2030” programme, under the reference ANR‐10‐COHO‐002, Observatoire Français de la Sclérose en Plaques (OFSEP) and by the Eugène Devic EDMUS Foundation against multiple sclerosis.

Conflicts of Interest

Manon Rival reports no disclosures relevant to the manuscript; Fabien Rollot reports no disclosures relevant to the manuscript; Sabine Laurent‐Chabalier reports no disclosures relevant to the manuscript; Romain Casey reports no disclosures relevant to the manuscript; Illiasse El Bahi reports no disclosures relevant to the manuscript; Guillaume Brocard reports no disclosures relevant to the manuscript; Romain Marignier reports no disclosures relevant to the manuscript; Sophie Trouillet‐Assant reports no disclosures relevant to the manuscript; Sandra Vukusic reports consulting and lecturing fees, travel grants and unconditional research support from Biogen, Coloplast, Janssen, Merck, Novartis, Roche, Sandoz and Sanofi; Agnès Fromont reports no disclosures relevant to the manuscript; David Axel Laplaud reports served on scientific advisory boards for Alexion, BMS, Roche, Sanofi, Novartis, Merck, Janssen and Biogen, received conference travel support and/or speaker honoraria from Alexion, Novartis, Biogen, Roche, Sanofi, BMS and Merck and received research support from Fondation ARSEP, Fondation EDMUS and Agence Nationale de la Recherche; Guillaume Mathey reports Travel grants from Biogen, Merck, Novartis, Roche, Sanofi‐Genzyme; Elisabeth Maillart reports research support from Biogen and ARSEP Foundation and personal fees for lectures and advisory boards from Alexion, Amgen, Biogen, Janssen, Lilly, Merck, Novartis, Roche, Sandoz, Sanofi; Laure Michel reports no disclosures relevant to the manuscript; Aurélie Ruet reports Consulting and lecturing fees, travel grants and/or unconditional research support from Alexion, Biogen, BMS, Horizon Th, Merck, Novartis, Sanofi Genzyme, and Roche; Pierre Labauge reports fees and grants from Biogen, Sanofi Genzyme, Novartis, Alexion, Merck; Xavier Ayrignac reports consulting and lecturing fees, travel grants, and unconditional research support from Alexion, Biogen, Genzyme, Janssen, Juvise, Novartis, Merck Serono, Roche and Sanofi; Hélène Zephir reports consulting or lectures, and invitations for national and international congresses from Biogen, Merck, Teva, Sanofi‐Genzyme, Novartis and Bayer, as well as research support from Teva and Roche, and academic research grants from Académie de Médecine, LFSEP, FHU Imminent and ARSEP Foundation; Jonathan Ciron reports consulting, serving on a scientific advisory board, speaking, or other activities with Biogen, Novartis, Merck, Sanofi, Roche, Alexion and Amgen; Jérôme de Sèze reports consulting and lecturing fees, travel grants and unconditional research support from Biogen, Genzyme, Novartis, Roche, Sanofi Aventis and Teva Pharma; Arnaud Kwiatkowski reports Consulting, lecturing fees and travel grants from Biogen, Jannsen, Merck, Novartis, Sandoz, Sanofi and Roche; Caroline Papeix reports Consulting, lecturing fees and travel grants from Amgen, Alexion, Biogen, Juvise, Novartis, and Roche; Pierre Branger reports consulting, serving on a scientific advisory board, speaking, or other activities with Alexion, Expression santé, Merck, Novartis and Sanofi‐Genzyme; Christine Lebrun‐Frenay reports no disclosures relevant to the manuscript; Eric Berger reports honoraria and consulting fees from Novartis, Sanofi Aventis, Biogen, Genzyme, Roche and Teva Pharma; Pierre Clavelou reports consulting and lecturing fees, travel grants and unconditional research support from Biogen, Janssen, Medday, Merck, Novartis, Roche, Sanofi‐Genzyme and Teva Pharma; Olivier Heinzlef reports consulting and lecturing fees from Bayer Schering, Merck, Teva, Genzyme, Novartis, Almirall and BiogenIdec, travel grants from Novartis, Teva, Genzyme, Merck Serono and Biogen Idec and research support from Roche, Merck and Novartis; Jean Pelletier reports no disclosures relevant to the manuscript; Abdullatif Al‐Khedr reports Consulting and lecturing fees, travel grants from Biogen, Novartis, Roche, Merck, Alexion and Sandoz; Olivier Casez reports Scientific advisory board for Roche, Juvise, Merck, Novartis and received funding for travel and/or honoraria from Merck, Novartis, Sanofi, Roche, Janssen, Biogen, Coloplast; Bertrand Bourrereports served on scientific advisory board for Alexion, BMS, Biogen, Sanofi, Janssen, Merck, Horizon, Novartis, Roche, Sandoz, and received funding for travel and honoraria from Alexion, Merck, Novartis, Sanofi, Roche and Janssen; Alain Créange reports Grants and personal fees from Biogen, Novartis, Roche, grants from Alexion, BMS, Horizon and Merck outside the submitted work, and travel grants Novartis and Roche; Thomas David reports No personal fees, hospitality support (travel, accommodation, and congress registration) from Alnylam, Amgen, Novartis, Merck; Laurent Magyreports Consulting and lecturing fees from Akcea, Alexion, Alnylam, Argenx, Biogen, CSL‐Behring, Genzyme, Grifols, Laboratoire Français des Biotechnologies, Merck‐Serono, Novartis, Pfizer, travel grants from, Argenx, Biogen, CSL‐Behring, Grifols, Novartis, Pfizer; Jean‐Philippe Camdessanche reports Consulting and lecturing fees from Akcea, Alexion, Alnylam, Argenx, Biogen, Bristol Myers Squibb, CSL‐Behring, Genzyme, Grifols, Johnson & Johnson, Laboratoire Français des Biotechnologies, Merck‐Serono, Natus, Novartis, Pfizer, Pharmalliance, UCB Pharma, Teva, SNF‐Floerger; travel grants from Akcea, Alexion, Alnylam, Argenx, Biogen, Bristol Myers Squibb, CSL‐Behring, Genzyme, Grifols, Johnson & Johnson, Laboratoire Français des Biotechnologies, Merck‐Serono, Natus, Novartis, Pfizer, UCB Pharma, Teva, SNF‐Floerger; Inès Doghri reports Consulting and lecturing fees, travel grants or unconditional research support from Merck, Sanofi, Sandoz, Novartis, Alexion, Horizon, Juvisé; Amélie Dos Santos reports no disclosures relevant to the manuscript; Thibault Mura reports no disclosures relevant to the manuscript; Hanane Agherbi reports no disclosures relevant to the manuscript; Eric Thouvenot reports consulting and lecturing fees, travel grants or unconditional research support from the following pharmaceutical companies: Actelion, Biogen, Janssen, Merck Serono, Novartis, Roche, Sanofi.

Supporting information

eFigure 1: Correlation of sNfL and sGFAP with age. Spearman correlation of Log[sNfL] (blue) and Log[sGFAP] (red) with age in NMOSD (AQP4 and DN‐NMOSD), MOGAD and MS (CIS, RRMS) patients. Spearman coefficient (R) and p‐value are indicated in the graphs. Quadratic regression has been used for correlation analysis in PPMS. AQP4‐NMOSD, aquaporin‐4 antibody‐positive neuromyelitis optica spectrum disorder; CIS, clinically isolated syndrome; DN‐NMOSD, Neuromyelitis Optica Spectrum Disorder without antibodies against aquaporin‐4 or myelin oligodendrocyte glycoprotein; MOGAD, myelin oligodendrocyte glycoprotein antibody‐associated disease; PPMS, primary progressive multiple sclerosis; RRMS, relapsing–remitting multiple sclerosis.

eFigure 2: Correlation of sNfL and sGFAP with EDSS. Spearman correlation of Log[sNfL] (blue) and Log[sGFAP] (red) with EDSS in NMOSD (AQP4 and DN‐NMOSD), MOGAD and MS (CIS, RRMS and PPMS) patients. Spearman coefficient (R) and p‐value are indicated in the graphs. AQP4‐NMOSD, aquaporin‐4 antibody‐positive neuromyelitis optica spectrum disorder; CIS, clinically isolated syndrome; DN‐NMOSD, Neuromyelitis Optica Spectrum Disorder without antibodies against aquaporin‐4 or myelin oligodendrocyte glycoprotein; MOGAD, myelin oligodendrocyte glycoprotein antibody‐associated disease; PPMS, primary progressive multiple sclerosis; RRMS, relapsing–remitting multiple sclerosis.

eFigure 3: Correlation of sNfL with sGFAP after a recent attack in NMO and MS subgroups. Spearman correlation of log[sNfL] with log[sGFAP] in AQP4‐NMOSD (A), MOGAD (B), DN‐NMOSD (C), CIS (D) and MS (E) patients. Spearman coefficient (R) and p‐value are indicated in the graphs. AQP4‐NMOSD, aquaporin‐4 antibody‐positive neuromyelitis optica spectrum disorder; CIS, clinically isolated syndrome; DN‐NMOSD, Neuromyelitis Optica Spectrum Disorder without antibodies against aquaporin‐4 or myelin oligodendrocyte glycoprotein; MOGAD, myelin oligodendrocyte glycoprotein antibody‐associated disease; RRMS, relapsing–remitting multiple sclerosis.

eTable 1: Association of sNfL levels with clinical and MRI variables in NMO and MS subgroups.

eTable 2: Association of sGFAP level with clinical and MRI variables in NMO and MS subgroups. Univariate and multivariate analysis.

eTable 3: Comparison of biomarker levels and delay of sampling at first attack and during subsequent relapses.

Appendix A. List of OFSEP Investigators Participating in the Study

This appendix provides the names and affiliations of the investigators and collaborative groups who contributed to the study on behalf of the OFSEP network.

Biology Group: Nicolas Collongues, MD, Hôpitaux universitaire de Strasbourg, Hôpital de Hautepierre, Service des maladies inflammatoires du système nerveux—neurologie, Strasbourg, France. Céline Cognet, Centre de Ressources Biologiques Plurithématique (CRB‐P), Bordeaux Biothèques Santé (BBS), Pôle de Biologie et de Pathologie, CHU de Bordeaux, Bordeaux, France. Mireille Desille‐Dugast, CRB, Laboratoire de Cytogénétique et Biologie Cellulaire, CHU Pontchaillou, Rennes, France. Nathalie Dufay, NeuroBioTec, Hôpital Neurologique Pierre Wertheimer, Hospices Civils de Lyon, Lyon/Bron, France. Mathilde Berthome, CRB, UF 7296, CHU de Nantes, Hôtel Dieu, Institut de biologie, Nantes, France. Michael Hisbergues, CRB/CIC1403, Centre de Biologie Pathologie Génétique, Lille, France. Sylvain Lehmann, CHU Saint Eloi, IRMB, Biochimie Protéomique Clinique, Montpellier, Franc. Laurence Liang, CRB PSL—AP‐HP, Sorbonne Université, Paris, France. Céline Loiseau, CRB, Laboratoire de cytogénétique, CHU de Nîmes, Nîmes, France. Julie Marechal, Etablissement Français du Sang, Service Biothèque‐CRB, Dijon, France. Estelle Nisse, CRB Lorrain‐ CHRU Nancy, Vandoeuvre‐les‐Nancy, France.

NOMADMUS Group: Françoise Durand‐Dubief, MD, Service de neurologie, sclérose en plaques, pathologies de la myéline et neuro‐inflammation, Hôpital Neurologique Pierre Wertheimer Hospices Civils de Lyon, Lyon, F‐6977, France. François Cotton, MD, Service de Radiologie, Centre Hospitalier Lyon‐Sud, Hospices Civils de Lyon, Université Claude Bernard Lyon1, Université de Lyon, Pierre Bénite; Lyon, France. Bertrand Audoin, MD, Aix Marseille Univ, APHM, Hôpital de La Timone, Pôle de Neurosciences Cliniques, Service de Neurologie, Marseille, France. Mikael Cohen, MD, Service de Neurologie, CRCSEP, Unité de Recherche Clinique Cote d'Azur (UR2CA), Centre Hospitalier Universitaire Pasteur 2, Nice, France. Nicolas Collongues, MD, Hôpitaux universitaire de Strasbourg, Hôpital de Hautepierre, Service des maladies inflammatoires du système nerveux—neurologie, Strasbourg, France. Romain Deschamps, MD, Department of Neurology, Fondation A. De Rothschild, Paris, France.

Data Availability Statement

The full dataset is available upon reasonable request to the corresponding author.