Introduction
Multiple sclerosis (MS) is a chronic autoimmune disease of the central nervous system (CNS), characterized by inflammation, demyelination, gliosis, and neuronal loss [ 1 ]. The mean age of onset ranges from 20 to 30 years and commonly affects women (female to male sex ratio is approximately 3:1) [ 2 ]. Worldwide, MS active cases are roughly 3 million, and the incidence rate is 2.1 per 100,000 persons/year; according to this rate, it is possible to estimate that someone in the world is diagnosed with MS every 5 min [ 3 ]. Neurological signs and symptoms are closely related to the localization of lesions within the CNS, with autonomic, visual, motor, and sensory deficits representing the most common clinical features. The onset and progression of these manifestations are primarily driven by inflammatory and neurodegenerative processes, alongside genetic, environmental, and immunological factors, which together contribute to the heterogeneity of the disease. Recent research suggests that inflammation and neurodegeneration are not distinct phases but rather constitute a single disease continuum that coexists from disease onset [ 4 ]. This evolving perspective on MS pathophysiology has significant implications for its clinical characterization, particularly regarding cognitive dysfunction. Cognitive impairment has gained recognition as an integral and early manifestation of the disease, reflecting the widespread and early involvement of both inflammatory and degenerative mechanisms within the CNS. Cognitive disorders are highly prevalent in MS, even from the early stages of the disease, with estimates ranging from 40 to 65% [ 5 ]. Neuropsychological disorders are closely associated with functional status, often leading to reduced participation in social life, increased difficulty with daily activities, and a higher risk of unemployment [ 6 ].
Cognitive impairment can also be associated with depressive symptoms, the most rapid disease progression, and increased disability in general, thus being responsible for poor Quality of Life (QoL) in subjects with MS [ 5 ]. Cognitive domains typically affected in MS include episodic and working memory, attention, conceptual reasoning, executive function (EF), information processing speed (IPS), orientation, and verbal fluency [ 7 ]. However, the neurobiological basis of cognitive functioning in MS is still not fully understood with both neuroinflammation and neurodegeneration possibly playing a crucial role in the onset of cognitive dysfunction [ 8 , 9 ]. In recent years, advances in molecular biomarker research have improved diagnostic accuracy and contributed to the early identification of patients at risk of developing cognitive deficits across the disease course [ 10 ].
This research field has already achieved robust results in the context of other neurodegenerative diseases. For example, cerebrospinal fluid (CSF) Tau and Beta-amyloid (Aβ42) are regularly used in Alzheimer’s disease (AD) [ 11 ], and the use of blood biomarkers in Parkinson’s disease is currently being explored [ 12 ]. Nevertheless, there is still a paucity of literature in MS, and this is possibly due to the lack of harmonization in the methodological choices adopted by the available studies [ 10 ]. This systematic review aims to evaluate longitudinal studies examining the relationship between serum and CSF biomarkers and cognitive performance in MS, to clarify their temporal dynamics and potential for predicting cognitive decline and tracking disease progression.
Materials and methods
A systematic review was conducted using the PRISMA (Preferred Reporting Items for Systematic Review and Meta-Analyses) guidelines [ 13 ] to investigate the existing literature about the longitudinal relationship between serum and CSF biomarkers and cognitive performance in MS. A protocol for this systematic review was registered on the International Prospective Register of Systematic Reviews (PROSPERO) with ID: CRD420251143923.
Search strategy
According to the aim of this systematic review, a search strategy was developed. The studies were identified by searching in PubMed, Scopus and Web of Science databases from July to August 2025. The search combined the following terms in PubMed: (“multiple sclerosis”[All Fields] OR “MS”[All Fields]) AND (“biological biomarker”[All Fields] OR “molecular biomarkers”[All Fields] OR “BDNF”[All Fields] OR “NfL”[All Fields]) AND (“cognition”[MeSH Terms] OR “cognition”[All Fields] OR “cognitions”[All Fields] OR “cognitive”[All Fields] OR “cognitively”[All Fields] OR “cognitives”[All Fields] OR “cognitive impairment”[All Fields]). No specific search filters were applied.
Study selection
After removing the duplicates using Zotero, a screening for title and abstract was performed independently by two researchers (G.R. and C.S.). At this stage, reviews, systematic reviews, meta-analyses, case reports, letters to the editor, short communications, and non-English articles were excluded. G.R and C.S. read the full-text articles and made sure they met the following criteria: (a) participants were subjects with a diagnosis of MS; (b) the study assessed serum and/or CSF biomarkers; (c) the study design was longitudinal; and (d) papers included a standardized cognitive assessment. On the other hand, studies were excluded if they met the following criteria: (a) cross-sectional study design; (b) studies focusing exclusively on genetic polymorphisms; (c) studies without a standardized cognitive assessment; and (d) articles not available as full text. After an accurate evaluation, 8 articles fulfilled the chosen criteria. For the qualitative synthesis, the included studies were grouped according to the investigated biomarker in order to enhance the comparability of findings and to facilitate the identification of consistent associations between specific biomarkers and cognitive outcomes.
Data extraction
Following the full-text selection, data were extracted independently by two reviewers (G.R. and C.S) and reported in a table using Microsoft Excel. For each eligible study, they collected information on: (i) general study characteristics (authors, year of publication, sample size, follow-up duration); (ii) population characteristics (multiple sclerosis phenotype when available); (iii) biomarkers assessed (type, biological source, serum or cerebrospinal fluid, detection methods); (iv) cognitive outcomes (neuropsychological tests or batteries employed, definition of cognitive impairment when available); and (v) main findings regarding the association between biomarkers and cognitive performance or decline. Any doubt throughout the data extraction process between the two reviewers was resolved through discussion and, when required, consultation with a third reviewer (V.L.B.).
Assessing the quality of included studies—risk of bias
G.R. and C.S. independently evaluated the risk of bias in selected studies. Any disagreements throughout this process, as well as in earlier stages, were settled by consulting V.L.B., who provided the final decision. All eligible studies were examined using the Cochrane tool for non-randomized controlled studies— of exposures (ROBINS-E) tool [ 14 ], which encompasses seven domains: (i) bias due to confounding, (ii) bias arising from measurement of the exposure, (iii) bias in selection of participants into the study (or into the analysis), (iv) bias due to post-exposure interventions, (v) bias due to missing data, (vi) bias arising from measurement of the outcome, and (vii) bias in selection of reported result.
Results
Study selection
A total number of 827 search entries were generated by searching on PubMed, Scopus, and Web of Science databases (Fig. 1 ). After the removal of duplicates, 579 studies were screened by title and abstract. Following the screening, 34 full-text articles were assessed for eligibility. Finally, 8 studies were included in the systematic review.
Study characteristics
A total of 457 subjects with MS were included in this systematic review. All the included studies were longitudinal and explored the relationship between serum and CSF biomarkers and cognitive performance in MS. The most frequently investigated biomarker was neurofilament light chain (NfL; n = 7), followed by glial fibrillary acidic protein (GFAP; n = 2). Among the 7 studies evaluating NfL concentrations, 6 measured serum NfL ( n = 1 by enzyme-linked immunosorbent assay, ELISA; n = 5 by single-molecule array, SIMOA), while 1 study assessed CSF NfL using an ELISA kit. Regarding cognitive assessment, 4 studies exclusively administered the Symbol Digit Modalities Test (SDMT), whereas 1 study used the processing speed test (PST), both evaluating IPS. Table 1 summarizes the key characteristics of the included studies [ 15 – 22 ]. In the following, the results of the systematic review are disclosed.
Table 1
Overview of the key study characteristics
BICAMS: Brief International Cognitive Assessment for MS; BRB-N: Brief Repeatable Battery of Neuropsychological Tests for MS; BREMSO: Bayesian Risk Estimate for MS at Onset; BVMT-R: Brief Visuospatial Memory Test-Revised; cAb42: CSF Beta-Amyloid 42; cBDNF: CSF brain-derived neurotrophic factor; CHI3L1: Chitinase-3-like-protein 1; CIS: clinically isolated syndrome; cNfL: CSF NfL; CN: cognitively normal; CSF: cerebrospinal fluid; CVLT-II: California Verbal Learning Test-II; CXCL13: CXC motif chemokine ligand 13; EDSS: Expanded Disability Status Scale; ELISA: enzyme-linked immunosorbent assay; GFAP: glial fibrillary acidic protein; IL-6: interleukin-6; MFIS: Modified Fatigue Index Scale; miR: Micro RNA; MoCA: Montreal Cognitive Assessment; NfL: neurofilament light chain; OPN: osteopontin; PASAT-3: paced auditory serial addition test; PPMS: primary progressive MS; PST: processing speed test; PVALB: Parvalbumin; PVF: phonemic verbal fluency; RCFT: Rey Complex Figure Test; RIS: radiologically isolated syndrome; RNA: ribonucleic acid; RRMS: relapsing–remitting MS; sBDNF: serum brain-derived neurotrophic factor; sCI: severe cognitively impaired; SDMT: Symbol Digit Modalities Test; SIMOA: single-molecule array; sNfL: serum NfL; SPMS: secondary progressive MS; SRT-CLTR: selective reminding test consistent long-term retrieval; SRT-D: selective reminding test delayed; SRT-LTS: selective reminding test long-term storage; VLMT: Verbal Learning Memory Test
| Author, Refs. | Population | Biomarker | Matrix | Assay | Cognitive assessment | Follow-up duration | Biomarker assessment timing | Cognitive assessment timing | Outcome |
| Barro et al. [ 15 ] | 94 MS subjects | GFAP, NfL | Serum | SIMOA | SDMT | 3.1 years | At baseline | Annually throughout follow-up | Significant association between baseline sNfL and the change over time in SDMT after adjusting for the age, sex, EDSS, treatment group, and relapse in the previous 90 days, the change in the slope for a one-log increase in sNfL was −1.14 (95% CI −1.83, −0.44; p = 0.001). However, there was no significant association between sGFAP and concurrent or future SDMT (estimated change in slope = −0.47; 95% CI −1.37, 0.44; p = 0.31) |
| Bhan et al. [ 16 ] | At baseline, 41 out of 108 MS subjects invited to the study agreed to participate with lumbar puncture and blood work At 10-year follow-up, only 26 patients were left for the analyses | cNFl, sNfl | Serum/CSF | SIMOA for sNFl; ELISA for cNfL | SDMT | 10 years | At baseline | At baseline, year 5 and year 10 | Baseline sNfL was associated with SDMT during the 10-year observation period ( p < 0.001). For each one-unit higher log sNfL, expected baseline SDMT decreased by 9.5 points ( β −9.5, 95% CI −15.5 to −3.4, p = 0.002), after adjustment for age, sex, higher education, and disease duration. Higher levels of sNfL were also associated with an increased rate of reduction in SDMT from baseline to 5 years and from baseline to 10 years, but none of these associations were statistically significant ( p = 0.249). Baseline cNfL was also associated with longitudinal SDMT ( p = 0.005) although the model fit was slightly inferior compared to serum measures |
| Domínguez-Mozo et al. [ 17 ] | 73 MS subjects (48 females): 6 CIS, 49 RRMS, 7 SPMS, 6 PPMS, and 5 RIS | miR-9.5p, miR-29b.3p, miR-34a.5p, miR-126.3p, miR-138.5p, miR-146a.5p, miR-200c.3p, miR-223.3p, sNfL | Serum | Cell-free total RNA was extracted using the miRNeasy Serum/PlasmaAdvanced Kit. The total RNA was reverse transcribed using the miRCURY LNA RNA kit. MiRNA-specific quantification was performed using the miRCURY LNA SYBR green kit; SIMOA for NfL | PST | 1 year | At baseline | At baseline and year 1 | In univariate analysis, miR-126.3p was associated with EDSS ( r = 0.3; p = 0.019), PST ( r = −0.28; p = 0.028), and PST change ( r = −0.41; p = 0.04) at 1 year, while miR-9.5p was associated with PST change ( r = 0.57; p = 0.041) In multivariate analysis, miR-9.5p maintained a statistical association with PST change (beta = 0.04; p = 0.007). miR-126.3p kept a statistical association with EDSS (beta = 1.07; p = 0.008) and PST change (beta = −0.131; p = 0.031). The sNfL level did not achieve statistical significance for any clinical or radiological variable, but it was correlated with miRNA-9.5p ( r = 0.53; p = 0.043) |
| Friedova et al. [ 18 ] | 58 MS subjects | cNfL, sNfl | Serum/CSF | ELISA for cNfL; SIMOA for sNfL | BICAMS and the Rao adaptation of the PASAT-3 | 9 years | At baseline, year 1, and year 2 | At baseline, year 1, year 2 and year 9 | No reported correlations between sNfL levels and cross-sectional or longitudinal cognitive measures, except for an association between higher sNfL levels at baseline ( n = 49) and lower CVLT-II scores at year 1 (rho = −0.31, p = 0.028; q = 0.112). sNfL levels above the 90th percentile at year 1 were associated with a greater risk of CVLT-II decrease over 9 years compared with patients with lower sNfL levels [odds ratio 15.8; 95% CI 1.7–147.0; p = 0.015] |
| Oset et al. [ 19 ] | 50 RRMS subjects | CXCL13, CHI3L1, OPN, IL-6, GFAP, NfL | Serum | ELISA | SDMT | ≈28 months | At baseline | At baseline and year 2 | A tendency toward a positive correlation of IL-6 level ( R = 0.3496, p = 0.0128) and a negative correlation of OPN level ( R = −0.2998, p = 0.0344) with the SDMT score at baseline assessment has been observed. However, after correcting for multiple analyses, none of the biomarkers correlated significantly with SDMT In addition, subjects with higher NfL quantities at baseline experienced a more severe disease course based on EDSS score ( U = 107, Z = −2,74, p = 0.006) and presence of relapse ( U = 188, Z = −2.01, p = 0.044) |
| Tiu et al. [ 20 ] | 50 RRMS subjects | cNFl, sNfl, cAb42 | Serum/CSF | SIMOA for sNfL | MoCA, SDMT, BVMT-R | 1 year | sNfL at baseline, 3 and 6 months; cNfL and cAb42 at baseline | At baseline and year 1 | Statistically significant differences were found for subjects that experienced a cognitive decline for sNfL z -scores at baseline and 3 months, cNfL baseline values, cAb42 and the BREMSO score as well. The logistic regression model based on these 5 variables was statistically significant, χ 2 (4) = 22.335, p < 0.0001, R 2 = 0.671, with a sensitivity of 57.1%, specificity of 97.4%, a positive predictive value of 80% and a negative predictive value of 92.6% |
| Yalachkov et al. [ 21 ] | 50 MS subjects with RRMS, PPMS or CIS for the baseline measurement. 16 patients were lost to follow-up. Thus, 34 for cognitive assessment at both baseline and follow-up | sBDNF, cBDNF | Serum/CSF | SIMOA HD-1 and BDNF Discovery Kit | SDMT, RCFT, VLMT, and additionally the PASAT 3 | 1 year | At baseline | At baseline and year 1 | Higher cBDNF for subjects with cognitive improvement, independently from the level of education ( p = 0.004). Secondary analysis revealed significant correlations between sBDNF and EDSS change ( q = 0.036), cBDNF and average z -score change ( q = 0.04) and cBDNF and the number of cognitive tests with improvement ( q = 0.04) |
| Ziccardi et al. [ 22 ] | 72 MS subjects (50 females) | PVALB, NfL | CSF | ELISA | BRB-N, Stroop Test, Phonemic Semantic Alternate Verbal Fluency, Modified Five Point Test | ≈4 years | At baseline | At baseline, year 2 and year 4 | PVALB levels were higher in sCI subjects than in CN ones ( p = 0.033). Higher PVALB levels significantly correlated with worse global cognitive ( p = 0.024) and memory functioning ( p = 0.044). Higher PVALB levels at diagnosis were associated with a worse performance in specific cognitive tests at the end of the follow-up period: SRT-LTS ( r = −0.27, p = 0.024), SRT-CLTR ( r = −0.27, p = 0.022), SRT-D ( r = −0.39, p = 0.001), and PVF ( r = −0.35, p = 0.004). In addition, patients with higher levels of PVALB at diagnosis showed higher cognitive ( p = 0.024) and global fatigue ( p = 0.043) at follow-up assessed with MFIS Analyses conducted for NfL showed no significant results |
Neurofilament light chain (NfL)
NfL was the most assessed biomarker in the studies’ selection ( n = 7); however, they showed heterogeneous results in terms of predicting cognitive decline. A study reported that higher serum NfL (sNfL) concentrations at baseline were associated with higher future cognitive decline measured with SDMT over a median of 3.1-year observation period [ 15 ]. A similar result has been found by Bhan et al. [ 16 ], who showed that baseline sNfL levels were associated with a decrease in SDMT score during a 10-year observation period. Despite this association, after performing longitudinal analyses, neither sNfL nor CSF NfL (cNfL) showed a significant association with longitudinal change in SDMT. Another study observed an association between higher sNfL levels at screening and lower California Verbal Learning Test—second edition (CVLT-II) scores at year 1 on a 9-year follow-up period [ 18 ]. Tiu et al. [ 20 ] built a prediction model based on subjects that showed a significant cognitive decline on Montreal Cognitive Assessment (MoCA) or SDMT scores throughout a 1-year follow-up. Specifically, they demonstrated that sNfL, cNfL and Aβ42, combined with a clinical score (Bayesian Risk Estimate for MS at Onset, BREMSO), can predict an early cognitive decline for RRMS subjects at the time of diagnosis. On the other hand, some studies found no correlations between NfL and SDMT performance [ 19 ] or PST score [ 17 ], as well as in a study in which subjects underwent a more comprehensive neuropsychological assessment with the Brief Repeatable Battery of Neuropsychological Tests for MS (BRB-N) [ 22 ]. Although no significant association with cognitive performance was detected, a study reported the presence of higher levels of sNfL at the baseline in subjects characterized by a more severe disease course based on an increase in the Expanded Disability Status Scale (EDSS) score, the presence of relapse, and self-reported judgment of decline in health [ 19 ]. A similar result was also reported by the aforementioned study [ 15 ] in which the predictive value of sNfL in detecting cognitive decline was more evident in subjects with acute inflammatory disease activity.
Glial fibrillary acidic protein (GFAP)
Only two studies in our systematic review assessed GFAP concentrations; however, no significant correlations with decline over time in PST [ 19 ] or SDMT performance [ 15 ] were reported.
Osteopontin (OPN) and interleukin-6 (IL-6)
A marginal negative correlation of OPN level with SDMT performance at the baseline assessment was observed in a single study [ 19 ]. IL-6 association with cognitive performance was explored in a recent 2-year follow-up study included in this work; however, after correcting for multiple comparisons, no significant correlation with SDMT performance was observed [ 19 ].
Chitinase-3-like 1-protein (CHI3L1), C–X–C motif chemokine ligand 13 (CXCL13)
Serum concentrations of Chitinase-3-like 1-protein (CHI3L1, also known as YKL-40) and C–X–C motif chemokine ligand 13 (CXCL13) were correlated with SDMT score neither at baseline nor at 2-year follow-up assessment [ 19 ].
MicroRNAs
Among different microRNAs evaluated in a recent 1-year longitudinal study, miR-126.3p and miR-9.5p showed significant results [ 17 ]. MiR-126.3p correlated with EDSS and IPS performance, assessed with PST, at baseline, and miR-126.3p and miR-9.5p correlated with change in PST score at 1 year.
Parvalbumin (PVALB)
CSF levels of parvalbumin (PVALB) have longitudinally been correlated with cognitive impairment in a 4-year follow-up study [ 22 ]. Subjects were clustered into ‘cognitively normal’ (CN, 0 failed subtests), ‘mildly cognitive impaired’ (mCI, one or 2 failed subtests), and ‘severely cognitive impaired’ (sCI, 3 or more failed subtests), based on their performance at BRB-N. A global cognitive functioning index and domain-specific cognitive functioning indices were also calculated. Subjects with severe cognitive impairment at 4-year follow-up had significantly higher CSF PVALB levels at diagnosis, compared with those ones with normal cognitive functioning. Higher PVALB levels at baseline were significantly associated with worse global cognitive and memory functioning at the end of the follow-up time. Specifically, higher PVALB levels at diagnosis were associated with a worse performance in verbal learning and phonemic fluencies at the end of the follow-up period. Notably, a preliminary clinical threshold for PVALB concentrations at diagnosis was proposed (2.57 ng/mL), which maximizes the risk of showing cognitive impairment at follow-up, with a sensitivity of 91% (specificity 30%). Furthermore, subjects with higher levels of PVALB at diagnosis showed higher cognitive and global fatigue at follow-up.
Brain-derived neurotrophic factor (BDNF)
Higher BDNF levels in CSF at baseline were found in those subjects who experienced cognitive improvement (based on z -score change for each neuropsychological test), while subjects with disability improvement (EDSS) had higher serum BDNF levels at baseline [ 21 ]. Secondary analysis demonstrated significant correlations between serum BDNF and EDSS change, CSF BDNF and average z-score change and CSF BDNF and number of cognitive tests with improvement.
Beta-amyloid (Aβ42)
Aβ42 role in predicting cognitive impairment in MS was explored in only one of the selected studies, demonstrating that Aβ42, when combined with sNfL, cNfL and a clinical score (BREMSO), can predict an early cognitive decline for RRMS subjects at the moment of diagnosis [ 20 ]
Risk of bias
The Cochrane Risk of Bias Assessment Tool (ROBINS-E) [ 23 ] was used to assess the risk of bias of the articles included in this review. Figure 2 shows the summary of the risk of bias assessment. Most of the included studies ( n = 6) were reported as having “some concerns” in terms of risk of bias [ 15 , 17 – 20 , 22 ]. This was commonly due to concerns about inadequate adjustment for confounding variables, selection of participants, measurement of the outcome (e.g. only SDMT was administered, practice effect), and in the selection of reported results (e.g. significant results were emphasized). Two studies [ 16 , 21 ] were rated as having a “high risk” of bias due to concerns about missing data (subjects drop out during follow-up).
Discussion
This systematic review synthesizes current evidence on serum and CSF biomarkers as predictors of future cognitive decline in MS subjects.
Among the biomarkers analysed, NfL was the most extensively studied. NfL is a protein in the neuronal cytoskeleton and is particularly expressed at the axonal level. NfL is released into CSF and blood under normal conditions, but its concentrations proportionally increase following neuronal damage and axonal lesions [ 24 ]. Since neuronal and axonal damage are involved in the pathophysiology of MS, these processes seem to play a crucial role in the onset of cognitive dysfunction. Therefore, NfL may be a useful biomarker reflecting the degeneration of brain regions that are critical for cognitive functioning, especially in the presence of inflammatory activity. Previous works are in line with this hypothesis, demonstrating that higher NfL levels in subjects with MS are associated with brain and spinal cord volume loss, with a poor long-term outcome and cognitive impairment [ 25 – 27 ]. However, evidence resulting from the longitudinal studies showed controversial results. Some studies identified NfL as a feasible biomarker able to predict a future cognitive decline [ 15 , 16 , 18 , 20 ], while other studies did not, as well as in long follow-up periods [ 17 , 19 , 22 ]. Notably, subjects with higher sNfL levels seem to have a greater risk of experiencing an episodic verbal memory decline throughout the course of the disease and are more likely to experience a more severe disease course [ 18 ]. Furthermore, NfL levels in serum appear to be a better informative measure of cognitive decline than concentrations in CSF and this could be attributed to the different assay methodologies adopted in the studies. Beyond NfL, emerging biomarkers showed encouraging signals. PVALB is a calcium-binding protein expressed by fast-spiking GABAergic interneurons [ 28 ], which constitute a significant portion of cortical myelin and thus are particularly vulnerable to demyelination in MS [ 29 ]. Coherently with this statement, CSF PVALB concentrations reflect cortical atrophy in newly diagnosed subjects, also confirmed in postmortem observations [ 29 ]. Recent evidence indicates that CSF levels of PVALB can predict future cognitive deterioration in MS, showing longitudinal associations with poorer global cognition, impaired verbal memory, and greater fatigue. Importantly, a preliminary clinical threshold was also proposed by authors, suggesting a potential role in patient stratification [ 22 ]. Similarly, specific microRNAs (miR-126.3p, miR-9.5p) were significantly associated with processing speed and its decline over 1 year in MS subjects [ 17 ]. MicroRNAs are post-transcriptional regulators involved in inflammatory and neurodegenerative pathways. MiR-126-3p is mainly expressed in endothelial cells, where it regulates vascular homeostasis and angiogenesis. Through these mechanisms, it contributes to maintaining endothelial integrity and modulating neuroinflammation [ 30 , 31 ], whereas miR-9-5p plays an important role in the differentiation of neural stem cells, axonal growth, and synaptic plasticity. It is also involved in neuroimmune regulation, as it modulates microglial activation and cytokine production [ 32 , 33 ]. The exact role of these regulators in MS pathophysiology is still underexplored; however, the findings presented in this work support the potential role of specific microRNAs as prognostic biomarkers of cognitive impairment in MS. On the other hand, BDNF—a key molecule involved in remyelination, neuroplasticity and memory [ 34 ]—showed an opposite trend, with higher CSF levels linked to cognitive improvement [ 21 ]. Therefore, BDNF may play an important role in neurological and cognitive improvement acting as a neuroplasticity mediator in MS. Conversely, GFAP unexpectedly did not manage to achieve significant results [ 15 , 19 ]. GFAP is a structural protein found in the cytoskeleton of astrocytes, which is released in high concentrations into the CSF and bloodstream during the process of astrogliosis. Previous studies described its use in MS and reported correlations with disease severity, the extent of neuroinflammation and disease progression [ 35 , 36 ]. Furthermore, GFAP is currently being explored also in PD, where elevated plasma concentrations have shown a promising role not only in identifying subjects with cognitive impairment but also in predicting those who subsequently developed full-blown dementia [ 37 ]. However, inconclusive studies analyzed may arise from the scarce number of studies included and the different assay methodologies used. Furthermore, astrocytic reactivity (i.e. GFAP) may not be the correct pathological process underpinning cognitive decline in MS. In addition, even predominantly inflammatory biomarkers such as OPN, IL-6, CHI3L1, and CXCL13 provided insufficient longitudinal evidence on cognition [ 19 ]. While these molecules remain valuable for characterizing disease activity [ 38 – 41 ], their predictive role for cognitive outcomes may emerge only when combined with other biomarkers.
The main strength of our work lies in the exclusive inclusion of longitudinal studies, which are relevant for monitoring the relationship between biomarkers and cognitive performance over time. However, several limitations must be considered: the number of eligible studies was exiguous, and considerable methodological heterogeneity—including variability in cognitive assessment tools, biomarker assay methodologies (e.g., ELISA vs. SIMOA), and biological sources (CSF vs. blood)—represents a major limitation for direct comparison across studies. Moreover, most studies relied exclusively on SDMT or PST as measures of cognitive functioning. While these tools are well-established gold standards for a brief cognitive assessment in MS, their stand-alone use likely underestimates deficits in other domains, such as memory and executive functioning. Thus, this limitation highlights the importance of employing more comprehensive neuropsychological batteries in future research. Another limitation that should be emphasized is the lack of conversion of sNfL raw values into age-adjusted percentiles or z -scores based on a healthy control reference database. Converting raw sNfL concentrations into such standardized metrics provides a biologically normalized measure, allowing direct comparison across individuals and cohorts [ 42 ]. Unlike conventional statistical covariate adjustment, z -scores integrate demographic corrections at the biomarker level, improving interpretability and cross-study comparability. In our systematic review only two [ 18 , 20 ] out of eight studies applied this type of standardization.
In conclusion, no single biomarker fully explains cognitive decline in MS. Nevertheless, individual biomarkers provide important insights into the complex pathophysiological processes underlying cognitive dysfunction. In our view, the most promising direction for future research is the development of integrated biomarker panels that could also add significant contributions to the clinical field. In this context, with the advent of more advanced and highly sensitive assay technologies such as SIMOA [ 43 ], blood-based biomarkers should be emphasized. Compared to CSF sampling, blood work is minimally invasive and easier to employ in routine clinical practice. Ultimately, the predictive value of these biomarkers will be critical for the early identification of subjects at risk of cognitive decline, enabling clinicians to anticipate, monitor, and potentially modify the cognitive trajectory of people living with MS.
Author contributions
Conceptualization, G.R.; methodology, G.R., G.M. and V.L.B.; validation, C.S., V.L.B., C.R., E.S. and G.D.; formal analysis, C.S.; investigation, G.R. and C.S.; resources, G.R, C.S. and S.S.; data curation, G.R.; writing—original draft preparation, G.R.; writing—review and editing, G.R and V.L.B.; visualization, V.L.B.; supervision, V.L.B and A.Q. All the authors read and approved the final manuscript.
Funding
This work was supported by Italian Ministry of Health on NextGenerationEU PNRR funds (Missione: M6/Componente: C2 Investimento: 2.1 Valorizzazione, CUP: I43C22000650007, Grant No. PNRR-POC-2022-12376360).
Declarations
Conflicts of interest
The authors declare no conflicts of interest.
References
1. Haki M, AL-Biati HA, Al-Tameemi ZS, Ali IS, Al-hussaniy HA (2024) Review of multiple sclerosis: epidemiology, etiology, pathophysiology, and treatment. Medicine (Baltimore) 103:e37297. 10.1097/MD.0000000000037297
2. McGinley MP, Goldschmidt CH, Rae-Grant AD (2021) Diagnosis and treatment of multiple sclerosis: a review. JAMA 325:765–779. 10.1001/jama.2020.26858
3. Walton C, King R, Rechtman L, Kaye W, Leray E, Marrie RA et al (2020) Rising prevalence of multiple sclerosis worldwide: insights from the atlas of MS, third edition. Mult Scler J 26:1816–1821. 10.1177/1352458520970841
4. Portaccio E, Bellinvia A, Fonderico M, Pastò L, Razzolini L, Totaro R et al (2022) Progression is independent of relapse activity in early multiple sclerosis: a real-life cohort study. Brain J Neurol 145:2796–2805. 10.1093/brain/awac111
5. Chiaravalloti ND, DeLuca J (2008) Cognitive impairment in multiple sclerosis. Lancet Neurol 7:1139–1151. 10.1016/S1474-4422(08)70259-X
6. Rao SM, Leo GJ, Ellington L, Nauertz T, Bernardin L, Unverzagt F (1991) Cognitive dysfunction in multiple sclerosis. Neurology 41:692–696. 10.1212/WNL.41.5.692
7. Doskas T, Vavougios GD, Karampetsou P, Kormas C, Synadinakis E, Stavrogianni K et al (2022) Neurocognitive impairment and social cognition in multiple sclerosis. Int J Neurosci 132:1229–1244. 10.1080/00207454.2021.1879066
8. Daneshvar S, Moradi F, Rahmani M, Golshaniniya P, Frounchi N, Seifimansour S et al (2025) Association of serum levels of inflammation and oxidative stress markers with cognitive outcomes in multiple sclerosis: a systematic review. J Clin Neurosci 132:110990. 10.1016/j.jocn.2024.110990
9. Rossi F, Giorgio A, Battaglini M, Stromillo ML, Portaccio E, Goretti B et al (2012) Relevance of brain lesion location to cognition in relapsing multiple sclerosis. PLoS ONE 7:e44826. 10.1371/journal.pone.0044826
10. Rademacher T-D, Meuth SG, Wiendl H, Johnen A, Landmeyer NC (2023) Molecular biomarkers and cognitive impairment in multiple sclerosis: state of the field, limitations, and future direction: a systematic review and meta-analysis. Neurosci Biobehav Rev 146:105035. 10.1016/j.neubiorev.2023.105035
11. Kumar A, Nemeroff CB, Cooper JJ, Widge A, Rodriguez C, Carpenter L et al (2021) Amyloid and tau in Alzheimer’s disease: biomarkers or molecular targets for therapy? Are we shooting the messenger? Am J Psychiatry 178:1014–1025. 10.1176/appi.ajp.2021.19080873
12. Yamashita KY, Bhoopatiraju S, Silverglate BD, Grossberg GT (2023) Biomarkers in Parkinson’s disease: a state of the art review. Biomark Neuropsychiatry 9:100074. 10.1016/j.bionps.2023.100074
13. Page MJ, McKenzie JE, Bossuyt PM, Boutron I, Hoffmann TC, Mulrow CD et al (2021) The PRISMA 2020 statement: an updated guideline for reporting systematic reviews. BMJ 372:n71. 10.1136/bmj.n71
14. Higgins JPT, Morgan RL, Rooney AA, Taylor KW, Thayer KA, Silva RA et al (2024) A tool to assess risk of bias in non-randomized follow-up studies of exposure effects (ROBINS-E). Environ Int 186:108602. 10.1016/j.envint.2024.108602
15. Barro C, Healy BC, Saxena S, Glanz BI, Paul A, Polgar-Turcsanyi M et al (2023) Serum NfL but not GFAP predicts cognitive decline in active progressive multiple sclerosis patients. Mult Scler J 29:206–211. 10.1177/13524585221137697
16. Bhan A, Jacobsen C, Dalen I, Alves G, Bergsland N, Myhr K-M et al (2023) Neurofilament and brain atrophy and their association with cognition in multiple sclerosis: a 10-year follow-up study. Acta Neurol Scand 2023:1–10. 10.1155/2023/7136599
17. Domínguez-Mozo M, Casanova I, Monreal E, Costa-Frossard L, Sainz-de-la-Maza S, Sainz-Amo R et al (2024) Association of microRNA expression and serum neurofilament light chain levels with clinical and radiological findings in multiple sclerosis. Int J Mol Sci 25:10012. 10.3390/ijms251810012
18. Friedova L, Motyl J, Srpova B, Oechtering J, Barro C, Vodehnalova K et al (2020) The weak association between neurofilament levels at multiple sclerosis onset and cognitive performance after 9 years. Mult Scler Relat Disord 46:102534. 10.1016/j.msard.2020.102534
19. Oset M, Domowicz M, Wildner P, Siger M, Karlińska I, Stasiołek M et al (2023) Predictive value of brain atrophy, serum biomarkers and information processing speed for early disease progression in multiple sclerosis. Front Neurol 14:1223220. 10.3389/fneur.2023.1223220
20. Tiu VE, Popescu BO, Enache II, Tiu C, Terecoasa E, Panea CA (2022) Serum and CSF biomarkers predict active early cognitive decline rather than established cognitive impairment at the moment of RRMS diagnosis. Diagnostics 12:2571. 10.3390/diagnostics12112571
21. Yalachkov Y, Anschütz V, Maiworm M, Jakob J, Schaller-Paule MA, Schäfer JH et al (2023) Serum and cerebrospinal fluid BDNF concentrations are associated with neurological and cognitive improvement in multiple sclerosis: a pilot study. Mult Scler Relat Disord 71:104567. 10.1016/j.msard.2023.104567
22. Ziccardi S, Tamanti A, Ruggieri C, Guandalini M, Marastoni D, Camera V et al (2024) CSF parvalbumin levels at multiple sclerosis diagnosis predict future worse cognition, physical disability, fatigue, and gray matter damage. Neurol Neuroimmunol Neuroinflamm 11:e200301. 10.1212/NXI.0000000000200301
23. Bero L, Chartres N, Diong J, Fabbri A, Ghersi D, Lam J et al (2018) The risk of bias in observational studies of exposures (ROBINS-E) tool: concerns arising from application to observational studies of exposures. Syst Rev 7:242. 10.1186/s13643-018-0915-2
24. Khalil M, Teunissen CE, Otto M, Piehl F, Sormani MP, Gattringer T et al (2018) Neurofilaments as biomarkers in neurological disorders. Nat Rev Neurol 14:577–589. 10.1038/s41582-018-0058-z
25. Barro C, Benkert P, Disanto G, Tsagkas C, Amann M, Naegelin Y et al (2018) Serum neurofilament as a predictor of disease worsening and brain and spinal cord atrophy in multiple sclerosis. Brain 141:2382–2391. 10.1093/brain/awy154
26. Williams T, Heslegrave A, Zetterberg H, Miszkiel KA, Barkhof F, Ciccarelli O et al (2022) The prognostic significance of early blood neurofilament light chain concentration and magnetic resonance imaging variables in relapse-onset multiple sclerosis. Brain Behav 12:e2700. 10.1002/brb3.2700
27. Gaetani L, Salvadori N, Lisetti V, Eusebi P, Mancini A, Gentili L et al (2019) Cerebrospinal fluid neurofilament light chain tracks cognitive impairment in multiple sclerosis. J Neurol 266:2157–2163. 10.1007/s00415-019-09398-7
28. Hu H, Gan J, Jonas P (2014) Interneurons. Fast-spiking, parvalbumin+ GABAergic interneurons: from cellular design to microcircuit function. Science 345:1255263. 10.1126/science.1255263
29. Magliozzi R, Pitteri M, Ziccardi S, Pisani AI, Montibeller L, Marastoni D et al (2021) CSF parvalbumin levels reflect interneuron loss linked with cortical pathology in multiple sclerosis. Ann Clin Transl Neurol 8:534–547. 10.1002/acn3.51298
30. Cerutti C, Edwards LJ, de Vries HE, Sharrack B, Male DK, Romero IA (2017) MiR-126 and miR-126* regulate shear-resistant firm leukocyte adhesion to human brain endothelium. Sci Rep 7:45284. 10.1038/srep45284
31. Pan J, Qu M, Li Y, Wang L, Zhang L, Wang Y et al (2020) Microrna-126-3p/-5p overexpression attenuates blood-brain barrier disruption in a mouse model of middle cerebral artery occlusion. Stroke 51:619–627. 10.1161/STROKEAHA.119.027531
32. Yao H, Ma R, Yang L, Hu G, Chen X, Duan M et al (2014) MiR-9 promotes microglial activation by targeting MCPIP1. Nat Commun 5:4386. 10.1038/ncomms5386
33. Wu J, He J, Tian X, Luo Y, Zhong J, Zhang H et al (2020) Microrna-9-5p alleviates blood–brain barrier damage and neuroinflammation after traumatic brain injury. J Neurochem 153:e14963. 10.1111/jnc.14963
34. Miranda M, Morici JF, Zanoni MB, Bekinschtein P, Frontiers (2019) Brain-derived neurotrophic factor: a key molecule for memory in the healthy and the pathological brain. Front Cell Neurosci. 10.3389/fncel.2019.00363
35. Abdelhak A, Huss A, Kassubek J, Tumani H, Otto M (2018) Serum GFAP as a biomarker for disease severity in multiple sclerosis. Sci Rep 8:14798. 10.1038/s41598-018-33158-8
36. Högel H, Rissanen E, Barro C, Matilainen M, Nylund M, Kuhle J et al (2020) Serum glial fibrillary acidic protein correlates with multiple sclerosis disease severity. Mult Scler J 26:210–219. 10.1177/1352458518819380
37. Tang Y, Han L, Li S, Hu T, Xu Z, Fan Y et al (2023) Plasma GFAP in Parkinson’s disease with cognitive impairment and its potential to predict conversion to dementia. NPJ Parkinsons Dis 9:23. 10.1038/s41531-023-00447-7
38. Khademi M, Kockum I, Andersson ML, Iacobaeus E, Brundin L, Sellebjerg F et al (2011) Cerebrospinal fluid CXCL13 in multiple sclerosis: a suggestive prognostic marker for the disease course. Mult Scler J 17:335–343. 10.1177/1352458510389102
39. Bruno A, Dolcetti E, Azzolini F, Moscatelli A, Gambardella S, Ferese R et al (2022) Interleukin 6 SNP rs1818879 regulates radiological and inflammatory activity in multiple sclerosis. Genes 13:897. 10.3390/genes13050897
40. Talaat F, Abdelatty S, Ragaie C, Dahshan A (2023) Chitinase-3-like 1-protein in CSF: a novel biomarker for progression in patients with multiple sclerosis. Neurol Sci 44:3243–3252. 10.1007/s10072-023-06764-2
41. Marastoni D, Turano E, Tamanti A, Colato E, Pisani AI, Scartezzini A et al (2024) Association of levels of CSF osteopontin with cortical atrophy and disability in early multiple sclerosis. Neurol Neuroimmunol Neuroinflamm 11:e200265. 10.1212/NXI.0000000000200265
42. Benkert P, Meier S, Schaedelin S, Manouchehrinia A, Yaldizli Ö, Maceski A et al (2022) Serum neurofilament light chain for individual prognostication of disease activity in people with multiple sclerosis: a retrospective modelling and validation study. Lancet Neurol 21:246–257. 10.1016/S1474-4422(22)00009-6
43. Rissin DM, Kan CW, Campbell TG, Howes SC, Fournier DR, Song L et al (2010) Single-molecule enzyme-linked immunosorbent assay detects serum proteins at subfemtomolar concentrations. Nat Biotechnol 28:595–599. 10.1038/nbt.1641