Research in contextEvidence before this studyThe McDonald criteria, from their 2001 introduction through the 2017 revision, have enabled earlier diagnosis of multiple sclerosis (MS) by incorporating magnetic resonance imaging (MRI) to demonstrate dissemination in space and time following a first clinical event suggestive of MS. However, some individuals with suggestive MRI findings or limited clinical involvement of the central nervous system still do not meet these criteria and remain difficult to diagnose. Emerging evidence indicates that neuroaxonal injury and inflammation can be detected at early stages using blood-based biomarkers and advanced MRI measures, even before clear symptoms appear.
These findings suggest that MS-related pathological processes may begin earlier than captured by the 2017 McDonald criteria.Added value of this studyThe revised 2024 McDonald criteria broaden the diagnostic definition of MS to include individuals with asymptomatic MRI findings or limited clinical events not meeting the 2017 criteria. While several biomarkers have previously been associated with subclinical disease activity and future disease evolution, it remains unclear whether individuals newly captured by the 2024 criteria already harbour biological and radiological features consistent with established MS and whether such features are associated with subsequent disease activity. We found that individuals meeting the 2024 criteria (but not the previous 2017 MS diagnostic criteria) show clear evidence of MS-related pathological changes, including higher levels of a blood marker of neuro-axonal damage, reduced thalamic volume, and a greater burden of MS chronic inflammatory MRI lesions.
We also found that this blood marker of neuro-axonal damage was associated with future disease activity.Implications of all the available evidenceThese findings suggest that the latest revision of the McDonald diagnostic criteria for MS may identify individuals with early but clinically meaningful signs of MS. A simple blood test could help clinicians to better estimate the risk of disease activity and support earlier treatment decisions. Longer-term studies are needed to confirm whether earlier diagnosis under the 2024 criteria translates into improved long-term outcomes.
Introduction
For decades, the McDonald criteria have served as the cornerstone for diagnosing multiple sclerosis (MS), providing clinicians with a structured framework to identify the disease accurately and in a timely manner.1, 2, 3, 4 Originally formulated in 20011 and revised multiple times since,2, 3, 4 these criteria have evolved over time to incorporate advancements in both soluble and imaging biomarkers,4 with the goal of improving diagnostic sensitivity and specificity. A key objective of these updates has been the earlier identification of individuals with MS, enabling timely therapeutic intervention and the potential for better long-term outcomes.1, 2, 3, 4
Continued advances in pathophysiology biomarkers have prompted a substantial revision of the 2017 McDonald criteria4 for MS diagnosis, resulting in the 2024 McDonald criteria.5 These criteria provide a unified approach to diagnosing MS allowing asymptomatic radiological findings or clinical events suggestive of typical MS attacks that do not meet 2017 McDonald criteria to fulfil the diagnostic requirements. These new criteria introduce pivotal refinements designed to expedite MS diagnosis while maintaining specificity, including the incorporation of advanced MRI markers—such as the central vein sign (CVS),6, 7 and paramagnetic rim lesions (PRL)6, 7—the recognition of optic nerve (ON) as a fifth anatomical location,5, 8 and the adoption of cerebrospinal fluid (CSF) kappa free light chain (kFLC) analysis alongside the traditional oligoclonal bands (OCB).5, 9
Given the increased sensitivity of the 2024 McDonald criteria,5 it remains critical for future therapeutic decision-making to determine whether the expanded diagnostic boundaries identify individuals who already exhibit biological evidence of MS pathology and are at increased risk of subsequent disease activity. To address this, we investigated whether individuals fulfilling the new 2024 criteria,5 but not the previous 2017 criteria,4 feature increased MS-related pathological changes, as depicted by independent advanced laboratory and imaging biomarkers. In addition, we evaluated the longitudinal predictive value of these biomarkers in identifying individuals at risk of subsequent disease activity.
Methods
Study population
We retrospectively selected a cohort of patients participating in a longitudinal MS natural history study at Cliniques universitaires Saint-Luc (Belgium) between February 2021 and March 2026. This cohort consecutively enrols patients referred to the Neuroimmunology unit who present with central nervous system (CNS) white matter (WM) lesions suggestive of demyelinating disease. In accordance with the prospective protocol, patient inclusion is based on the initial radiological findings at presentation, irrespective of clinical symptoms or final formal diagnosis.
All participants undergo standardised and temporally matched peripheral blood sampling and 3T brain MRI on the same day as part of the study protocol.
Inclusion criteria for the primary analysis of this study encompassed individuals presenting either (i) incidental brain MRI focal features highly suggestive of MS in the absence of typical neurological symptoms, or (ii) clinical and MRI features suggestive of MS following a first clinical event (first attack), yet not fulfilling the 2017 McDonald diagnostic criteria4 and lacking a better alternative explanation.
For MRI features suggestive of MS, inclusion required at least one WM T2-hyperintense brain lesion measuring ≥3 mm in its long axis, and located in the infratentorial, intracortical/juxtacortical, or periventricular area.
Advanced blood and MRI biomarkers assessment
After collection, blood samples were immediately processed and stored at −80°C. Serum samples were subsequently thawed in a controlled manner, and levels of neurofilament light chain (sNfL) were quantified using the Atellica Immunoassay System (Siemens Healthineers Atellica® IM), according to the manufacturer's standardised protocols. Raw sNfL values were converted into age-corrected Z-scores according to Fitzgerald et al.,10 using age at blood sample collection.
These values are hereafter referred to as age-corrected sNfL Z-scores. Throughout the manuscript, the term age-corrected sNfL Z-score refers to the Fitzgerald-based calculation unless otherwise specified. As no consensus currently exists regarding the most appropriate approach for age adjustment of sNfL levels measured using the Atellica Immunoassay System, an additional Z-score calculation based on the approach proposed by Evliyaoglu et al.11 was also performed and reported.
The 3T brain MRI protocol included a submillimetre isotropic 3D T2∗-echo-planar imaging (EPI) sequence for the detection of the CVS and PRL imaging biomarkers, in addition to 3D double inversion recovery (DIR) and 3D magnetisation-prepared rapid acquisition gradient-echo (MPRAGE) to assess cortical lesion (CL) burden and volumetric changes, as previously described.6, 12, 13
Specifically, given that raters were partially involved in the recruitment of cases, all images were de-identified before analysis. For the CVS, patients were dichotomised as perivenular positive vs. perivenular negative based on the “Select6∗” algorithms,5, 6, 14 where a scan was considered as perivenular positive if there were ≥6 eligible lesions that met the CVS-positive North American Imaging in Multiple Sclerosis (NAIMS) criteria,15 or, when there were <10 lesions, if CVS-positive lesions outnumbered CVS-negative lesions.6, 15 PRL were identified on unwrapped phase images as chronic lesions showing a hypointense rim involving at least two-thirds of the lesion perimeter, without enhancement on post-contrast T1-weighted sequences, according to the NAIMS guidelines.16
In each individual, CVS, PRL, and CL analysis was independently assessed by 2 trained investigators (S.B. and P.M. for CVS, S.B. and A.S. for CL and PRL), each unaware of the other's analysis and blinded to participant's diagnosis. In case of discrepancies, cases were jointly reviewed for final adjudication.
Brain volumes, normalised to intracranial volume, were computed using FreeSurfer v7.2.0 (surfer.nmr.mgh.harvard.edu)12 after lesion filling on MPRAGE images. WM T2-hyperintense lesion volume (hereafter referred to as T2 lesion load [T2LL]) was derived from automated lesion segmentation on 3D-FLAIR images using FLAMeS.17
CSF, ON, and spinal cord assessments
While serum samples matched to research MRI for advanced biomarker assessment were prospectively collected in all participants as part of the research protocol, the CSF, ON, and spinal cord assessments had been retrospectively obtained as part of routine clinical work-up at the discretion of the treating neurologist.
CSF data were used to determine the presence of intrathecal inflammation through the detection of OCB and/or kFLC, in accordance with the recently published literature.9, 18 Isoelectric focussing of IgG was routinely performed using a semi-automated system (Hydragel 9 CSF, Hydrasys 2 Scan, Sebia GmbH). According to established guidelines, a positive OCB test is defined by the presence of at least two CSF-specific bands.19 κFLC were assessed using a blot/migration-based qualitative assay in cases of high MS suspicion but negative or borderline OCB results, and have been used in our centre as a complementary marker to OCB in neuroimmunological practice since the early 1990s.20
To evaluate ON involvement, a comprehensive retrospective review of clinical and paraclinical data was conducted. This included: (i) a documented history of acute clinical optic neuritis, based on characteristic clinical features and supported, when available, by paraclinical findings obtained during the acute episode—such as orbital MRI (typically including fat-suppressed T2-weighted or STIR and post-contrast T1-weighted sequences), visual evoked potentials (VEP), and/or optical coherence tomography (OCT)8, 21—with no better aetiological explanation; or, in the absence of a known clinical optic neuritis episode, (ii) structural evidence of ON damage on OCT, such as relative thinning of the peripapillary retinal nerve fibre layer (6 μm or greater) and/or the macular ganglion cell–inner plexiform layer (4 μm or greater) according to published guidelines8; and/or (iii) functional abnormalities consistent with ON involvement detected by VEP, such as prolonged P100 latency or significant inter-eye latency asymmetry, according to local laboratory cut-offs.8, 22
Spinal cord MRI data were reviewed, when available, to determine the presence of spinal cord lesions. The acquisitions were heterogeneous across participants but typically included sagittal T2-weighted and axial T2/T1-weighted sequences, with or without gadolinium contrast, according to local clinical protocols.
Applying the 2024 McDonald criteria
Based on whether participants fulfilled the 2024 McDonald criteria,5 they were stratified into two groups: McDonald 2024-positive (McDonald2024+) and McDonald 2024-negative (McDonald2024−).
In cases with incidental brain MRI findings ≥50 years of age and/or where the brain MRI was performed due to headaches (including migraine or tension-type) and/or in the presence of vascular comorbidities, additional supportive features were required due to the presence of clinical red flags.5 These features included one or more of the following: (i) CVS Select6∗ positivity, (ii) ≥1 spinal cord MRI lesion, or (iii) positive CSF findings (OCB and/or kFLC).5
According to the 2024 McDonald framework,5 dissemination in space (DIS) was defined as the involvement of at least two of the five predefined CNS topographies (infratentorial, intracortical/juxtacortical, periventricular, spinal cord, and/or ON). Among cases with incidental MRI findings, participants showing involvement of a single CNS topography were classified as McDonald2024–.
Although PRL are part of the 2024 McDonald framework, they are used only in the specific scenario of a clinical presentation suggestive of MS with involvement of a single CNS topography (supported by positive CSF or DIT findings).5 We therefore assessed PRL count—together with sNfL, advanced CL assessment, and brain volumetrics—as independent biomarkers to investigate whether the 2024 criteria identify individuals with greater MS-related pathological changes. While the 2024 criteria require the presence of at least one PRL in this specific diagnostic scenario, PRL count is not itself a diagnostic criterion. PRL count was therefore considered a quantitative measure of pathological burden, providing information beyond the binary presence of a PRL required for diagnostic classification.
Comparison cohort based on the 2017 McDonald criteria
To contextualise the findings in the McDonald2024+ group, an independent, age- and sex-matched cohort of patients with a diagnosis of MS as defined by the McDonald 2017 criteria (McDonald2017+),4 was derived from the same prospective natural history study. This cohort underwent the same advanced blood and MRI biomarker assessment as the primary cohort. ON involvement was not considered for McDonald2017+ classification, as the 2017 DIS criteria are based on four CNS topographies (infratentorial, juxtacortical/cortical, periventricular, and spinal cord) and do not include the ON.4 Similarly, CVS and CSF findings (OCB and/or kFLC) were not included in this comparative analysis, as they were used for diagnostic classification according to the respective McDonald criteria and were not considered independent biological measures.
Longitudinal clinical and radiological assessment
Participants with available follow-up data were longitudinally evaluated for evidence of disease activity. Disease activity was defined either as clinical activity (occurrence of relapses during follow-up), or as radiological activity (presence of new T2 lesions and/or gadolinium-enhancing lesions on follow-up MRI scans). Time to disease activity was defined as the interval between baseline assessment and the first occurrence of either clinical or radiological activity, or last available follow-up in the absence of events, at which point observations were censored.
Ethics
The study was conducted in accordance with the principles of the 2024 Declaration of Helsinki. All participants provided written informed consent prior to inclusion in the study. The research protocol was reviewed and approved by the Ethics Committee of the Cliniques universitaires Saint-Luc, Brussels, Belgium (IRB B4032020000104).
Statistics
Comparisons between diagnostic groups were performed to assess biomarker differences. Normality was evaluated using the Shapiro–Wilk test and Q–Q plots. Continuous variables are reported as mean ± SD or median[IQR] as appropriate.
Group comparisons were performed using two-sample t-tests or Mann–Whitney U tests for continuous variables and Fisher's exact or chi-square tests for categorical variables.
For the comparison between the McDonald2024+ and McDonald2017+ groups, 1:1 nearest-neighbour propensity score matching based on age and sex was performed, with the propensity score estimated using logistic regression.
The study included all eligible participants available within the predefined study period; no formal sample-size calculation was performed.
To control for multiple comparisons, false discovery rate (FDR) correction was applied using the Benjamini-Hochberg procedure. Multivariable analyses were conducted using separate models for each biomarker outcome, adjusting for potential confounders, including age at inclusion, age at onset or disease duration (entered in alternative models), and biological sex obtained from participants’ medical records. T2LL was additionally included as a covariate where appropriate to assess whether associations with McDonald2024+ status were independent of lesion burden.
Because age was already accounted for in the derivation of the age-adjusted sNfL Z-scores, age was not additionally included as a covariate in models in which the sNfL Z-score was the outcome. For models evaluating other outcomes, including brain volumetrics, T2LL, CL and PRL counts, age was retained as a covariate. Linear regression models were used for continuous outcomes (sNfL Z-scores, brain volumetrics, and T2LL).
Model assumptions were assessed based on residual normality, independence (Durbin–Watson statistic ≈2.0), and multicollinearity (variance inflation factor <1.1). Robust estimators were applied where appropriate to account for skewness and heteroscedasticity. For highly skewed count variables (CL and PRL), generalised linear models with a negative binomial distribution and log link were used.
For linear regression models, associations are reported as regression coefficients (β) with 95% confidence intervals (CI). For negative binomial models, effect estimates are reported as incidence rate ratios (IRR) with 95%CI.
Missing data mechanisms were evaluated using a three-step approach: Little's test for Missing Completely at Random (MCAR), group comparisons of complete vs. incomplete cases and assessment of missingness distribution across McDonald2024+ and McDonald2024− groups to explore Missing at Random patterns, and a sensitivity analysis excluding McDonald2024− cases in which missing ancillary investigations could have altered diagnostic classification (to address potential Missing Not at Random mechanisms).
To address potential detection bias due to heterogeneity in follow-up, the duration of clinical and radiological observation prior to study inclusion was compared between groups using a Mann–Whitney U test.
Time-to-event analyses were performed using elastic net penalised Cox regression to identify predictors of disease activity. Candidate predictors were selected from biomarkers significantly associated with McDonald2024 status in the primary cross-sectional analysis. Model performance was assessed using the cross-validated concordance index (C-index), while stability of selection and uncertainty of effect estimates were evaluated using bootstrap resampling.
Bootstrap percentile intervals were calculated from the empirical distribution of the penalised Cox coefficients across 500 bootstrap samples and exponentiated to obtain hazard ratios (HR).
A two-tailed p < 0.05 was considered statistically significant. FDR-corrected q < 0.05 was used for multiple comparisons. Analyses were performed using SPSS (v29.0.1.0) and R (v4.3.2).
Role of funders
The funders had no role in the study design, data collection, data analysis, data interpretation, or writing of the manuscript.
Results
Demographics and cohort characteristics
A total of 95 cases not meeting the 2017 McDonald criteria were included in the study. Of those, 64 participants presented with incidental brain MRI abnormalities highly suggestive of MS but without typical MS symptoms (group diagnosis 1), and 31 exhibited clinical and MRI features suggestive of MS, yet not fulfilling the 2017 McDonald diagnostic criteria and lacking a better alternative explanation (group diagnosis 2).
In group 1, MRI indications included migraine or other primary headache disorders in 27/64 (42%), and nonspecific or paroxysmal neurological symptoms in 37/64 (58%) participants (including vertigo, sensory disturbances, hearing loss, Bell's facial palsy, epileptic seizures, post-infectious neurological symptoms, and retinal vasculitis). In group 2, the reasons for not fulfilling the 2017 McDonald criteria were absence of DIS in 18/31 (58%) and/or absence of DIT or CSF-restricted OCB in 25/31 (81%) participants.
No significant differences were observed between group 1 and 2 with respect to age, sex, spinal cord lesion presence, ON involvement, or advanced serum and MRI biomarker assessments, except for a higher frequency of DIS criteria fulfilment and CSF positivity (OCB and/or kFLC) in group 1 compared to group 2 (for detailed clinical/paraclinical characteristics in group 1 vs. 2, see Supplementary Table S1).
Application of the 2024 McDonald criteria
Applying the 2024 McDonald criteria, MS was diagnosed in 31/64 (48%) participants in group 1, and 15/31 (48%) in group 2, for a total of 46/95 (48%) participants who were McDonald2024+, and 49/95 (52%) participants McDonald2024–. A detailed diagnostic flowchart is shown in Fig. 1, and a summary of clinical and biomarker characteristics is provided in Table 1 and Supplementary Table S2.Fig.
1Diagnostic flow chart. This flowchart (adapted from Montalban et al. Lancet Neurol 2025)5 outlines the inclusion process.
Patients presenting with symptoms or MRI findings suggestive of MS underwent an initial diagnostic workup, including medical history, clinical examination, imaging, and laboratory testing. A total of 95 patients were identified. Reasons for exclusion of 91 patients primarily included the presence of alternative neurological diagnoses better explaining the clinical and/or MRI findings.
Of the included cohort, 31 presented with clinical and radiological features indicative of a demyelinating event suggestive of MS, yet not fulfilling the 2017 McDonald diagnostic criteria and lacking a better alternative explanation (left part of the diagram) and 64 with incidental brain MRI focal abnormalities highly suggestive of MS in the absence of typical neurological symptoms (central part of the diagram). These patients were subsequently classified as MS (McDonald2024+, n = 46) or not-MS (McDonald2024-, n = 49) according to the 2024 McDonald criteria.5 An age- and sex-matched cohort of 46 patients with a diagnosis of MS per McDonald 2017 criteria (McDonald2017+)4 was included for comparison (right part of the diagram).Alt text begins. A flowchart showing patient eligibility assessment, diagnostic classification pathways, and final McDonald 2024 and McDonald 2017 outcome groups.
Alt text provided by Elsevier.Alt text begins. The flowchart describes patient evaluation and classification. The top box is titled Eligibility criteria and lists age greater than or equal to 18, imaging findings suggestive of demyelinating disease, and matched 3T brain Magnetic Resonance Imaging including 3D-T2 star-EPI and sNfL.
A downward arrow leads to Assessed for eligibility between February 2021 and March 2026, N equal to 393 patients. A second arrow leads to Initial workup suggestive of Multiple Sclerosis, including medical history and clinical examination, imaging and laboratory tests, and differential diagnosis. A left branch labeled No leads to better explanations, N equal to 91.
The main pathway divides into three streams. The left stream is Clinical presentations suggestive of Multiple Sclerosis, but McDonald 2017 is not fulfilled, N equal to 31. This divides into Lesions present in greater than or equal to 2 out of 5 Central Nervous System topographies, N equal to 21, followed by one or more of the following: Cerebrospinal Fluid positive, Central Vein Sign positive, Dissemination in Time, or lesions present in greater than or equal to 4 out of 5 topographies.
This leads to Cerebrospinal Fluid positive N equal to 4, Central Vein Sign Positive N equal to 13, Dissemination in Time N equal to 2, and greater than or equal to 4 out of 5 topographies N equal to 2, which converge into Multiple Sclerosis N equal to 14 and Non-Multiple Sclerosis N equal to 7. A second branch contains Lesions present in 1 out of 5 Central Nervous System topography, N equal to 10, followed by one or more of the following: Cerebrospinal Fluid positive and Central Vein Sign positive, Cerebrospinal Fluid positive and Paramagnetic Rim Lesion positive, Dissemination in Time and Central Vein Sign positive, or Dissemination in Time and Paramagnetic Rim Lesion positive. This leads to Cerebrospinal Fluid positive or Dissemination in Time N equal to 2 and Central Vein Sign positive or Paramagnetic Rim Lesion positive N equal to 4, which converge into Multiple Sclerosis N equal to 1 and Non-Multiple Sclerosis N equal to 9.
The center stream contains Incidental imaging findings suggestive of Multiple Sclerosis in asymptomatic patients or with symptoms not specific to Multiple Sclerosis, N equal to 64. This leads to Lesions present in greater than or equal to 2 out of 5 Central Nervous System topographies, N equal to 56, followed by one or more of the following: Cerebrospinal Fluid positive, Central Vein Sign positive, or Dissemination in Time. The pathway then divides into Cerebrospinal Fluid positive N equal to 20, Central Vein Sign positive N equal to 27, and Dissemination in Time N equal to 15, which converge into Multiple Sclerosis N equal to 31 and Non-Multiple Sclerosis N equal to 33.
The right stream contains Multiple Sclerosis according to McDonald 2017, N equal to 207, which leads to Age- and sex-matching. The outcome boxes at the bottom from the left to right are McDonald 2024 plus, N equal to 46; McDonald 2024 minus, N equal to 49; and McDonald 2017 plus, N equal to 46. Alt text provided by Elsevier.Table 1Comparison of patients presenting with symptoms or MRI findings suggestive of MS, not meeting diagnostic criteria for MS under McDonald 2017,4 and subsequently classified as MS (McDonald2024+, N = 46) or not-MS (McDonald2024−, N = 49) according to the 2024 McDonald criteria.5McDonald2024+ (N = 46)aVs McDonald2024− (N = 49)ap-valuebFDR-adjusted p-valuecVs McDonald2017+ (N = 46)ap-valuebFDR-adjusted p-valuecBaseline characteristics Age [years]42 ± 1242 ± 110.98n.s.46 ± 140.12n.s.
Sex: Female32 (70%)43 (88%)0.0300.04133 (72%)0.82n.s. Male14 (30%)6 (12%)--13 (28%)-- Patients with CSF analysis39 (85%)30 (61%)0.0100.01542 (91%)0.34n.s. Patients with spinal cord MRI44 (96%)37 (76%)0.0060.01042 (91%)0.40n.s.
Patients with spinal cord lesion14/44 (32%)4/37 (11%)0.0320.04234/42 (81%)<0.001<0.001 Disease duration [months]12 [45]13 [47]0.71n.s.80 [177]<0.001<0.001McDonald 2024 work-up analysis Patients with CSF positivity for OCB and/or kFLC (% of cases with CSF analysis)26/39 (67%)3/30 (10%)<0.001<0.001Not applicable Patients with ON assessment37 (80%)23 (47%)<0.0010.003 Patients with ON involvement16/37 (43%)7/23 (30%)0.32n.s. DIS presence40 (87%)29 (59%)0.0020.005 DIT presence15 (33%)2 (4%)<0.0010.001 Select6∗ positivity40 (87%)7 (14%)<0.001<0.001 PRL presence23 (50%)2 (4%)<0.001<0.001Advanced biomarker assessment sNfL [pg/mL]11.6 [6.9]9.7 [2.9]0.0130.01912.4 [9.0]0.39n.s. Age-corrected sNfL Z-score (Fitzgeraldet al.)d0.28 [0.76]−0.12 [0.82]0.0030.0060.21 [1.01]0.72n.s.
Age-corrected sNfL Z-score (Evliyaogluet al.)e0.82 [1.27]0.15 [1.07]0.0050.0090.64 [2.07]0.66n.s. Normalised whole brain volumef0.76 ± 0.020.75 ± 0.030.16n.s.0.74 ± 0.030.0200.06 Normalised WM volumef0.30 ± 0.020.30 ± 0.020.64n.s.0.30 ± 0.020.14n.s. Normalised cortical volumef0.42 ± 0.020.43 ± 0.030.13n.s.0.41 ± 0.020.0260.07 Normalised thalamic volumef0.009 ± 0.00070.010 ± 0.00070.0030.0060.009 ± 0.00090.0380.09 T2 Lesion Load [mm3]2898 [3619]729 [2309]0.0030.0065032 [7392]0.0020.010 CL count1 [4]0 [0]<0.001<0.0015 [6]0.0020.010 PRL count1 [1]0 [0]<0.0010.0031 [3]0.41n.s.CSF: Cerebrospinal Fluid, MRI: Magnetic Resonance Imaging, OCB: Oligoclonal Bands, kFLC: kappa Free Light Chains, ON: Optic Nerve, DIS: MRI-based Dissemination in Space, DIT: MRI-based Dissemination in Time, sNfL: Serum Neurofilament Light Chain, PRL: Paramagnetic Rim Lesions, WM: White Matter; CL: Cortical Lesions.An age- and sex-matched cohort of 46 patients with a diagnosis of MS per McDonald 2017 criteria (McDonald2017+)4 was included for comparison.All p values are presented in italics; statistically significant p values (p < 0.05) are additionally shown in bold.a(Mean ± SD/Median [IQR]), n/N (%).bT-student test, Mann–Whitney U-test, Fisher's exact test or χ2 test.cBenjamini-Hochberg.dAge-adjusted sNfL reference values according to Fitzgerald et al.
Ann Neurol 2022.eAge-adjusted sNfL reference values according to Evliyaoglu et al. Clinical Chemistry 2026.fVolumes normalised to intracranial volume (ICV); values are expressed as volume/ICV ratios (unitless).
Regarding demographic characteristics, a higher proportion of male patients was observed in the McDonald2024+ group (Table 1). Compared with McDonald2024− cases, McDonald2024+ patients more frequently exhibited features directly related to the application of the diagnostic criteria, including a higher prevalence of spinal cord lesions, greater CSF positivity (OCB and/or kFLC), and more frequent fulfilment of DIS and DIT criteria, as well as Select6∗ positivity (Table 1). PRL were also more frequent in McDonald2024+ (vs.
McDonald2024−) group (Table 1) but rarely contributed to diagnostic classification. The presence of ≥1 PRL independently led to fulfilment of the 2024 criteria in only one case (Fig. 1).
The duration of clinical and radiological observation prior to study inclusion (disease duration) was comparable in McDonald2024+ (vs. McDonald2024−) group, indicating similar timing of assessment between groups (Table 1). Importantly, no participant was classified as McDonald2024+ solely based on DIT fulfilment, and all McDonald2024+ cases fulfilled the diagnostic criteria through other diagnostic features (e.g., CSF positivity and/or Select6∗ positivity).
Pathological features in McDonald 2024-positive cases
When comparing McDonald2024+ (n = 46) vs. McDonald2024– (n = 49) groups, the first group showed elevated age-corrected sNfL Z-scores (according to Fitzgerald et al. approach: median 0.28[0.76] vs. −0.12[0.82], p = 0.003/pFDR = 0.006; corresponding mean values: 0.40 ± 0.83 vs. −0.12 ± 0.68, respectively; according to Evliyaoglu et al. approach: median 0.82[1.27] vs. 0.15[1.07], p = 0.005/pFDR = 0.009; corresponding mean values: 1.81 ± 5.04 vs.
0.35 ± 0.94, respectively), reduced normalised thalamic volume (0.009 ± 0.0007 vs. 0.010 ± 0.0007, p = 0.003/pFDR = 0.006), greater T2LL (2898[3619] vs. 729[2309] mm3, p = 0.003/pFDR = 0.006), CL (1[4] vs.
0[0], p < 0.001/pFDR < 0.001) and PRL (1[1] vs. 0[0], p < 0.001/pFDR = 0.003) counts (Table 1, Fig. 2).Fig.
2Pathological features in McDonald 2024-positive cases. Panels A–C show MRI-derived pathological biomarkers: (A) brain volumetrics derived from 3D-MPRAGE, including normalised thalamic volume; (B) paramagnetic rim lesions (PRL) detected using 3D-EPI phase imaging; (C) cortical lesions (CL) visualised using 3D-DIR and 3D-MPRAGE. Scale bars in panels A–C represent 1 cm.
Panel D shows box plots of age-corrected serum neurofilament light chain (sNfL) Z-scores in McDonald 2024-positive (McDonald2024+; n = 46) and McDonald 2024-negative (McDonald2024-; n = 49) cases, and in McDonald 2017-positive cases (McDonald2017+; n = 46). Boxes represent the interquartile range (IQR), the horizontal line indicates the median, and whiskers extend to the most extreme data points within 1.5 × IQR. Individual observations are shown as dots.
Between-group comparisons were performed using the Mann–Whitney U test (∗p < 0.05, ∗∗p < 0.001, n.s. = not significant).Alt text begins. Four-panel figure showing brain Magnetic Resonance Imaging analyses, lesion identification, volumetric assessment, and statistical comparisons of serum neurofilament light chain levels. Alt text provided by Elsevier.Alt text begins.
Four panels labeled A, B, C, and D. Panel A, titled Brain volumetrics on 3D-M P R A G E, shows an axial brain Magnetic Resonance Imaging slice with multiple color-coded segmented anatomical regions distributed across both cerebral hemispheres. A scale bar in the lower-right corner indicates 1 centimeter.
Panel B, titled P R L count on 3D-Phase, shows an axial phase Magnetic Resonance Imaging brain slice. A yellow box highlights a periventricular region, and a magnified inset in the upper-left corner displays the selected area with a red arrow indicating a P R L lesion. A scale bar in the lower-right corner indicates 1 centimeter.
Panel C, titled CL count on 3D-D I R and 3D-M P R A G E, shows an axial brain Magnetic Resonance Imaging image with a yellow box highlighting a cortical region. Two enlarged views are displayed on the right and labeled D I R and M P R A G E. Red arrows indicate corresponding lesions in both enlarged views.
A scale bar in the lower-right corner indicates 1 centimeter. Panel D, titled Serum N f L analysis according to diagnostic criteria, contains a box plot of Age-corrected s N f L Z-scores with values ranging from negative 2 to 2 on the vertical axis. The horizontal axis contains 3 groups: McDonald2024 minus, McDonald2024 plus, and McDonald2017 plus.
Individual data points are plotted around each box. Significance comparisons are shown below the plot, with double asterisks between McDonald2024 minus and McDonald2024 plus, non-significant (n.s.) between McDonald2024 plus and McDonald2017 plus, and a single asterisk between McDonald2024 minus and McDonald2017 plus. A legend identifies Diagnosis groups, and a note indicates asterisk p less than 0.05 and double asterisk p less than 0.01.
Alt text provided by Elsevier.
After adjustment for age and sex, McDonald2024+ status remained significantly associated with lower thalamic volume (β = −4.96 × 10−4, 95%CI: −8.12 × 10−4 to −1.81 × 10−4; p = 0.002), greater T2LL (β = 1906, 95%CI: 108–3704; p = 0.038), and higher CL (IRR = 6.8, 95%CI: 2.8–16.8; p < 0.001) and PRL (IRR = 28.6, 95%CI: 6.7–121.6; p < 0.001) counts.
After additional adjustment for T2LL, McDonald2024+ status remained significantly associated with higher sNfL Z-scores (β = 0.38, 95%CI: 0.01–0.75; p = 0.042), lower thalamic volume (β = −4.11 × 10−4, 95%CI: −7.27 × 10−4 to −9.54 × 10−5; p = 0.011), and higher CL (IRR = 6.6, 95%CI: 3.0–14.9; p < 0.001) and PRL (IRR = 16.9, 95%CI: 4.1–69.5; p < 0.001) counts.
Furthermore, replacing age at inclusion with either age at onset or disease duration yielded consistent associations. McDonald2024+ status remained significantly associated with higher sNfL Z-scores (β = 0.46, 95%CI: 0.14–0.78; p = 0.005), lower thalamic volume (β = −5.13 × 10−4, 95%CI: −8.32 × 10−4 to −1.94 × 10−4, p = 0.002; and β = −4.70 × 10−4, 95%CI: −8.08 × 10−4 to −1.33 × 10−4, p = 0.007, respectively), and higher CL (IRR = 6.5, 95%CI: 2.4–17.6, p < 0.001; and IRR = 7.9, 95%CI: 3.5–18.1, p < 0.001, respectively) and PRL (IRR = 28.6, 95%CI: 6.7–122.5, p < 0.001; and IRR = 28.1, 95%CI: 6.8–117.0, p < 0.001, respectively) counts.
The above main findings remained consistent in the sensitivity analysis addressing potential classification uncertainty related to missing ancillary investigations (see below).
Sensitivity analysis
Consistent with the retrospective real-world design, complete baseline investigations (CSF analysis, spinal cord MRI and ON assessment) were available in 47/95 (49%) participants. Little's test did not provide evidence against an MCAR mechanism (p = 0.90). In addition, missingness of ancillary investigations was not associated with baseline characteristics, including age, sex, and DIS status (all p > 0.05).
However, missingness was significantly associated with McDonald 2024 classification (p = 0.002), reflecting differential availability of ancillary investigations across diagnostic groups (Table 1).
A sensitivity analysis was therefore performed to assess the robustness of the findings to potential diagnostic classification uncertainty. Accordingly, 19/49 (39%) McDonald2024– participants were subsequently excluded because positivity on one or more missing investigations could have resulted in reclassification as McDonald2024+. Specifically, fourteen participants presented with incidental MRI findings involving ≥2 CNS topographies and lacked CSF analysis; in these cases, the presence of OCB and/or kFLC would have resulted in a McDonald2024+ classification.
In addition, three participants presented with incidental MRI findings limited to a single CNS topography and were Select6∗ positive; in these three cases, positivity on the missing ON assessment and/or spinal cord MRI would have resulted in DIS fulfilment and shifted their classification to McDonald2024+. Finally, two participants with a clinical presentation suggestive of MS and involvement of a single CNS topography were Select6∗ positive (but CSF analysis negative for OCB and/or kFLC); these patients lacked ON assessment, which—if positive—could have resulted in DIS fulfilment and led to a McDonald2024+ classification.
This sensitivity analysis (n = 76; 46 McDonald2024+ and 30 McDonald2024−), confirmed an elevated age-corrected sNfL Z-score (0.28[0.76] vs. −0.14[0.79], p = 0.011/pFDR = 0.011), reduced normalised thalamic volume (0.009 ± 0.0008 vs. 0.010 ± 0.0006, p < 0.001/pFDR < 0.001), greater T2LL (2898[3619] vs. 516[2375] mm3, p = 0.009/pFDR = 0.011), CL (1[4] vs.
0[0], p < 0.001/pFDR < 0.001) and PRL (1[1] vs. 0[0], p < 0.001/pFDR < 0.001) counts in the McDonald2024+ (vs. McDonald2024−) group.
Comparison with McDonald 2017-positive cases
The McDonald2017+ group was age- and sex-matched to the McDonald2024+ group (n = 46). It included patients with an established diagnosis of MS after a first clinical manifestation, with longer disease duration (80[177] months vs. 12[45] months in the McDonald2024+ group; p < 0.001/pFDR < 0.001), heterogeneous MS phenotypes (30 [65%] relapsing and 16 [35%] progressive), and a majority (28 [61%]) who had not yet started treatment at the time of inclusion.
Overall, no significant differences were observed in sNfL levels, brain volumetrics, or PRL count between McDonald2024+ and McDonald2017+ cases (all pFDR > 0.05; Table 1). Univariable analyses suggested higher CL count (5[6]vs. 1[4], p = 0.002/pFDR = 0.011) and T2LL (5032[7392]vs.
2898[3619] mm3, p = 0.002/pFDR = 0.011) in the McDonald2017+ (vs. McDonald2024+) group. After adjustment for age and sex, T2LL remained significantly higher in the McDonald2017+ group (β = 2578, 95%CI: 410–4746; p = 0.020), whereas the difference in CL count was no longer significant (p = 0.10).
However, after further adjustment for disease duration, neither T2LL nor CL count differed significantly between groups (all p > 0.05). This finding was consistent after excluding participants with incidental MRI findings from the McDonald2024+ group and when the analysis was restricted to patients with relapsing onset in the McDonald2017+ group (all p > 0.05).
Predictors of disease activity during follow-up
Among 52 patients with available follow-up data, the median follow-up duration was 25 months (range, 1–59 months). During follow-up, 8 patients developed disease activity, defined as clinical relapse and/or radiological activity. No evidence of subsequent clinical progression was observed during follow-up.
Of the 18 patients classified as McDonald 2024-negative at baseline with available follow-up data, one (6%) was subsequently diagnosed with MS after a clinical relapse occurring 5 months after the initial assessment.
In time-to-event analyses, penalised Cox regression identified age-corrected sNfL Z-score and PRL presence as the main predictors of disease activity (C-index = 0.71). Bootstrap analysis supported the selection of both variables, with higher selection stability for sNfL (0.84) than PRL (0.76). The median bootstrap-derived HRs were 1.18 (95%CI: 1.00–2.88) for sNfL and 1.34 (95%CI: 1.00–5.48) for PRL presence.
Among participants who developed disease activity, the median baseline age-corrected sNfL Z-score was 0.60[1.03] according to Fitzgerald et al., and 1.40[3.29] when calculated according to the Evliyaoglu et al. approach; PRL were present in 6/8 (75%) patients.
Discussion
Among patients not meeting the 2017 McDonald criteria,4 those fulfilling the 2024 McDonald criteria5 exhibit increased MS-related pathological burden, including elevated sNfL, reduced thalamic volume, and higher counts of CL and PRL. Of these baseline measures, sNfL showed the greatest stability as a predictor of subsequent disease activity, highlighting its potential as a readily measurable biomarker with clinical utility. These findings suggest that blood-based biomarkers such as sNfL may provide complementary information to existing diagnostic and prognostic frameworks and may be useful in supporting early clinical decision-making in MS.
Prompt diagnosis and management are essential to prevent the accumulation of neurological disability occurring since the earliest MS stages.23 Elevated sNfL levels in cases fulfilling the 2024 McDonald criteria are consistent with previous findings showing that neuroaxonal damage, as measured by sNfL, already occurs in the presymptomatic stage of MS and is associated with a higher risk of developing clinical MS.24 Importantly, large-scale longitudinal studies have demonstrated that sNfL rises right after key pathogenic events such as Epstein–Barr virus seroconversion, linking early neuroaxonal injury to upstream disease triggers and underscoring the role of sNfL as an early MS biomarker.25 In our longitudinal analysis, baseline sNfL showed the greatest stability among the identified predictors of subsequent disease activity, including clinical relapses and radiological activity. As a fully quantitative blood test measurable with emerging cost-effective platforms,26 sNfL does not require specialised imaging expertise, and may offer a readily implementable source of complementary information to MRI, potentially informing early clinical decision-making.
Building on this, our study also revealed additional pathological features in patients fulfilling 2024 McDonald criteria, including reduced thalamic volume, and higher counts of CL and PRL. Although these advanced imaging measures are more complex to interpret and often require specialised expertise, they provide complementary evidence that the revised criteria identify individuals with biologically meaningful MS pathology. Thalamic atrophy, like sNfL, occurs early in the disease course,27 and can help distinguish MS from its imaging mimics,28 while higher CL and PRL burden reflects early inflammatory changes,29, 30 and is known to be associated with poorer prognosis.6, 31, 32 Although our exploratory time-to-event analysis suggested sNfL as a sensitive predictor of future disease activity, imaging features such as thalamic atrophy, CL, and PRL burden remain useful for characterising disease severity and support the clinical utility of the 2024 criteria in identifying patients with more pronounced underlying pathology who may benefit from closer monitoring.
Interestingly, when comparing McDonald2024+ cases to a cohort of age- and sex-matched individuals with an MS diagnosis according to the 2017 McDonald criteria,4 no significant differences were observed in sNfL levels, thalamic volume, or PRL count. Although T2LL was higher in the McDonald2017+ group after adjustment for age and sex, this difference was no longer significant after accounting for disease duration, as was also the case for the difference in CL count observed in univariable analyses. These results suggest that, despite the greater sensitivity of the 2024 McDonald criteria—which might be expected to identify individuals with milder disease—they identify individuals with a biological disease burden comparable to that of patients meeting the 2017 McDonald criteria, further supporting the biological relevance of the 2024 framework.
One potential limitation of this study is that some of the investigated biomarkers (PRL and CL) are already included in the 2024 diagnostic guidelines, potentially introducing a degree of circularity. However, PRL plays only a marginal role in the revised criteria, being restricted to rare clinical scenarios,5 as confirmed in our cohort where this applied to only one individual not fulfilling the 2017 criteria. Similarly, CL detection in routine clinical practice remains technically challenging and is often underestimated, as it requires advanced MRI sequences.33 By employing optimised protocols (i.e., 3D-DIR combined with 3D-MPRAGE) our study identified a higher CL burden, thereby highlighting their relevance as an early29 and specific6, 34 biomarker of MS pathology.
Importantly, the other biomarkers analysed are not included in the current diagnostic criteria and provide independent measures of MS pathology, supporting the biological relevance of the observed group differences beyond circular diagnostic logic.
Additional limitations of this study include the relatively small overall sample size and the retrospective and clinically driven availability of some ancillary diagnostic tests (CSF analysis, ON assessment, spinal cord MRI), which were not systematically performed according to a predefined protocol. This resulted in heterogeneous availability of these paraclinical assessments across participants, which may have introduced variability in group comparisons. However, the sensitivity analysis addressing these missing data yielded consistent results, supporting the robustness of the main findings.
Time-to-event analyses were further limited to 52 patients with follow-up data, of whom only 8 developed disease activity, which restricts statistical power and limits the evaluation of less potent predictors. Even though these analyses should be considered exploratory and require validation in larger independent cohorts, baseline sNfL consistently emerged as a robust predictor of subsequent disease activity, supporting its role as an early biomarker. In addition, differences in analytical platforms (Atellica vs.
Simoa) and in the reference populations used for age-correction of sNfL may limit direct comparability of the reported values across studies. Our findings further indicate that age-corrected sNfL Z-scores are influenced by the specific age-adjustment methodology applied, underscoring the need for standardised approaches to facilitate interpretation and comparison across studies. Finally, race and ethnicity were not recorded, precluding analyses of their potential impact on the study findings.
Nevertheless, a major strength of this study lies in the rigorous and standardised assessment of the CVS, which carries substantial weight in the 2024 McDonald diagnostic criteria.5 In our cohort, all included individuals underwent the same optimised susceptibility-based MRI protocol (i.e., submillimetric 3D T2∗-EPI sequence),15, 35 enabling a robust and standardised evaluation of this biomarker. This contributes to the overall reliability of our McDonald 2024 classification and enhances the interpretability of our results in the context of the new diagnostic framework.
In conclusion, among patients not meeting the 2017 McDonald criteria, those fulfilling the 2024 criteria feature enhanced neuroinflammatory/neurodegenerative MS-related pathological changes, including elevated sNfL, reduced thalamic volume, and higher CL and PRL counts. Serum NfL emerges as a readily measurable potential predictor of subsequent disease activity that may complement the current diagnostic criteria in guiding early clinical decision-making. These findings highlight the need for longitudinal studies to clarify the therapeutic implications of an earlier, biomarker-informed MS diagnosis.
Contributors
SB, EB, JLB, MD, CVB, AS, FG, VT, SE, AC, SES, VVP, and PM contributed to data acquisition. SB, EB, JLB, CVB, AS, and PM analysed the data. SB and PM accessed and verified the underlying data.
SB, CB and PM performed the statistical analysis. SB and PM drafted the manuscript. SB and PM drafted the figures.
All authors read and approved the final version of the manuscript.
Data sharing statement
The data supporting the findings of this study, including individual patient data and a data dictionary defining each field, will be made available upon reasonable request to the corresponding author after publication. Access to data will require a signed data use agreement to ensure compliance with ethical guidelines and patient confidentiality. Statistical code used for analysis can also be provided upon request.
Data will be shared via a secure institutional transfer system. If data sharing requests are received, they will be reviewed to ensure the purpose aligns with the original research scope and respects participant privacy. Restrictions may apply to protect sensitive patient information, in accordance with local regulations and consent agreements.
Declaration of interests
S. Borrelli received speaker/consulting honoraria and/or travel grants from Sanofi, Roche, Janssen, Merck, Novartis, Alexion and Amgen, and research grants from Roche, Sanofi, and Brugmann Foundation not related to this work. JL Bayart received speaker/consulting honoraria from Siemens Healthineers and The Binding Site company.
C. Vanden Bulcke received financial support by the Fédération Wallonie Bruxelles–Fonds Spéciaux de Recherche (F.S.R.). V. van Pesch received consulting honoraria from Biogen, Merck, Sanofi, BMS, Novartis, Janssen, Almirall, Roche, Alexion.
P. Maggi received consulting honoraria from Novartis, Sanofi, Biogen and Merck. E.
Barbuti, M. Deltombe, A. Stölting, F.
Guisset, V. Tota, S. Elands, A.
Ceccarelli, C. Bugli and S. El Sankari report no disclosures relevant to the manuscript.