How Spinal Fluid Clues Can Predict Multiple Sclerosis

For most people who eventually develop multiple sclerosis (MS), the journey starts with a single, alarming episode of neurological symptoms—what neurologists call a clinically isolated syndrome (CIS). That might be sudden vision loss, numbness, or weakness that lasts days to weeks. The big clinical dilemma is what happens next: only about 63% of people with CIS will ever have a second attack and move on to a firm diagnosis of MS, while the rest may never relapse, even after 20 years. Being able to say who will convert—and who won’t—would completely change how confidently doctors could start, or withhold, early MS therapies. This is exactly the question tackled in this study, which used a “multi-omics” deep dive into cerebrospinal fluid (CSF) to search for more precise, individualized predictors of conversion from CIS to MS.

Why classic markers like oligoclonal bands aren’t enough
Today, diagnosis and risk stratification rely on MRI, clinical exam, and a few CSF tests, particularly oligoclonal IgG bands (OCGB). OCGB are very sensitive for MS, so their presence now counts toward a diagnosis in the 2017 McDonald criteria. But they’re poor at telling which CIS patients will actually relapse. In this cohort of 54 CIS patients followed for 2–10 years, 22 went on to have a second attack (converters) and 32 did not (non-converters). Every single converter was OCGB-positive—but so were 69% of the non-converters. That gives OCGB a perfect sensitivity (100%) but a very low specificity (31%) and an overall accuracy of only 59% for predicting a second clinical attack. In other words, OCGB can say “this looks like MS biology,” but they can’t reliably answer “will this person relapse?”

Multi-omics in spinal fluid: looking beyond a single marker
To go deeper, the authors collected CSF at the time of CIS (median three weeks from symptom onset), before any second attack or disease-modifying treatment. They then layered three types of data on top of standard clinical chemistry:

Cell counts and basic CSF measures (leukocytes, mononuclear vs polymorphonuclear cells, protein, albumin ratio)

Metabolomics using ¹H-NMR to quantify ~50 small molecules such as glucose, lactate, creatine and myo-inositol

Proteomics using the SomaScan platform, measuring >5000 protein analytes and ultimately focusing on 89 proteins linked to conversion

They used orthogonal partial least squares–discriminant analysis (OPLS-DA) plus rigorous cross-validation and permutation testing to separate converters from non-converters and to identify which features carried real predictive signal rather than random noise.

Simple CSF cell counts that outperform oligoclonal bands
One of the most striking “wins” came from a very simple measure that every lumbar puncture already generates: the CSF leukocyte count. Over half of the cohort had mildly elevated CSF cells (>4 cells/mm³), but this was much more common in people who later relapsed—72% of converters versus 41% of non-converters. When broken down by cell type, the signal almost entirely came from mononuclear cells (lymphocytes and monocytes), not polymorphonuclear cells. Mononuclear cell counts predicted conversion with an AUC of about 0.74 and an accuracy of 70%; total leukocytes performed similarly (AUC 0.73, accuracy 67%), both clearly better than OCGB alone. Importantly, these are cheap, routinely measured parameters—so simply paying closer attention to mononuclear pleocytosis could already sharpen individual-level risk estimates in CIS.

Metabolites and proteins as a molecular fingerprint of conversion
At the metabolite level, converters and non-converters had clearly different CSF “chemical fingerprints.” A small panel—particularly myo-inositol, glucose, plus lactate and creatine—drove most of the separation between groups. Changes in glucose and lactate are consistent with disturbed brain energy metabolism, something already implicated in MS through mitochondrial dysfunction and oxidative damage. Myo-inositol is enriched in myelin and glial cells; higher levels in converters likely reflect more intense demyelination and gliosis already underway at the CIS stage, even when MRI may look similar between patients. Individually, myo-inositol and glucose each predicted conversion better than OCGB (with accuracies around 72% and 63%, respectively), while lactate and creatine added value mainly in combination with the others.

Proteomics added another layer of resolution. The authors identified 89 proteins whose baseline CSF levels differed between converters and non-converters; 72 of them beat OCGB in AUC. Some high-performing examples include:

XRCC1, a DNA repair protein, lower in converters and the best single biomarker (AUC 0.84, 80% accuracy)

TPM3 and EFCAB14, whose low levels in converters gave 100% sensitivity for future relapse

MUSK, a receptor tyrosine kinase, whose low levels had 100% specificity—no person with high MUSK converted

Pathway analysis of the protein panel painted a biologically coherent picture: converters showed enrichment of cytokine–cytokine receptor signalling, TNF and interferon-γ pathways, leukocyte proliferation, and chemotaxis—essentially, an exaggerated CNS immune activation profile that looks very much like established MS. Non-converters, in contrast, were enriched for pathways related to extracellular matrix organisation, MAPK signalling, and even rheumatoid arthritis–like signatures, hinting that some CIS presentations may be driven by more peripheral or atypical inflammatory processes.

An integrated algorithm that beats any single test
The real power of “multi-omics” emerges when you don’t treat each biomarker in isolation. The authors systematically tested combinations of biomarkers and found that a five-variable model gave the best balance of performance and parsimony. That model combined:

MUSK (protein)

RPS6KA5 (a ribosomal S6 kinase)

DYNLT1 (a dynein light chain involved in neuronal structure)

CSF myo-inositol

CSF mononuclear cell count

Together, this panel predicted clinical conversion with an AUC of 0.94 and overall accuracy of 83%, clearly outperforming OCGB and any other single marker. Interestingly, OCGB itself did not make it into the optimal model; swapping mononuclear cell counts for OCGB actually worsened performance by about 11%. This underlines a key point: even beloved legacy markers may add little once more quantitative, pathway-level biomarkers are in the mix.

What this could mean for people with CIS
This study is still discovery-phase and based on a relatively small single-centre cohort, so its exact biomarker panel will need validation—and likely refinement—in larger, independent datasets before it reaches clinics. But conceptually, it shows that a short list of CSF features, spanning immune cell counts, metabolites, and proteins, can provide an individualized “risk portrait” of whether a person with CIS is on a trajectory toward definite MS. Clinically, such a tool could help neurologists justify early aggressive treatment for those at very high risk, while sparing lower-risk individuals from years of potent immunotherapy and its side-effects. Biologically, the work reinforces that early converters already have intense CNS immune activation, altered energy metabolism, and glial involvement at the time of their very first attack—changes that may be detectable long before disability accumulates.

Disclaimer: This blog post is based on the provided research article and is intended for informational purposes only. It is not intended to provide medical advice. Please consult with a healthcare professional for any health concerns.

References:
Probert, F., Yeo, T., Zhou, Y., Sealey, M., Arora, S., Palace, J., ... & Anthony, D. C. (2021). Integrative biochemical, proteomics and metabolomics cerebrospinal fluid biomarkers predict clinical conversion to multiple sclerosis. Brain Communications, 3(2), fcab084.