In a quiet shift in how we understand a long-misunderstood illness, researchers at University College London and Queen Square Analytics have found that multiple sclerosis may not be one story but two distinct biological paths. Reported in the journal Brain and drawn from about 600 people living with the condition, their work suggests that the disease has been grouped by the symptoms it shows rather than by what it is doing within the body. Two patterns now stand apart: one in which damage surfaces early and moves quickly, and one in which the brain quietly thins before the telltale signals app
AI Uncovers Two Biological Subtypes of Multiple Sclerosis, Pointing Toward Tailored Treatment
MS is not one disease
Mark
So what's actually new here? Doctors already classify MS into types.
Mimi
Those types, like relapsing-remitting and secondary progressive, are based on symptoms. This work sorts patients by what's happening biologically, using MRI and a blood protein.
Mark
What separates the two subtypes?
Mimi
Order of events. In early sNfL, the protein rises early and the corpus callosum is damaged, with lesions forming faster. In late sNfL, brain shrinkage in areas like the limbic cortex comes first and the protein rises later.
Luke
Worth saying plainly: the treatment benefits are projections. The article says early sNfL patients could be offered stronger drugs sooner, but that's what might happen, not something tested here.
Mark
Is that why validation matters?
Mimi
Yes. The findings need confirming in further studies before clinics could rely on them.
Luke
Also, about 600 patients is the dataset size, and the article doesn't say how they were recruited or how diverse the group was. That's unknown from this reporting.
Mark
And the blood test being simple matters?
Mimi
Eshaghi stressed it's a widely available marker, which would make it easier to use alongside MRI.
Luke
Fair, though the article doesn't say whether sNfL testing alone could assign a subtype. The model used both together.
Le Pouls
- Multiple sclerosis has long been sorted by visible symptoms, leaving many patients matched to treatments that may not suit the tissue damage beneath.
- A team of UCL researchers used an AI model called SuStaIn to read brain scans and a blood marker of nerve damage across about 600 patients.
- Two distinct biological subtypes surfaced: an early, fast-moving form marked by early protein rise and corpus callosum damage, and a late form where brain shrinkage comes first.
- The early pattern could one day steer patients toward stronger treatments sooner, while the late pattern may call for therapies that protect brain cells.
- The findings, published in Brain, still await validation in further studies before they can change clinical practice.
In a quiet shift in how we understand a long-misunderstood illness, researchers at University College London and Queen Square Analytics have found that multiple sclerosis may not be one story but two distinct biological paths. Reported in the journal Brain and drawn from about 600 people living with the condition, their work suggests that the disease has been grouped by the symptoms it shows rather than by what it is doing within the body. Two patterns now stand apart: one in which damage surfaces early and moves quickly, and one in which the brain quietly thins before the telltale signals appear in the blood. The discovery offers the hope that, in time, a patient's treatment might follow the pattern of their own disease rather than the outward shape of their symptoms, though the finding must still be tested by others before it can reshape care.
Researchers at University College London and the company Queen Square Analytics have used artificial intelligence to identify two biological subtypes of multiple sclerosis that had not been recognised before. The work, published in the journal Brain, draws on data from around 600 people with the condition.
The need is long-standing. Multiple sclerosis affects millions worldwide, yet treatment decisions still rest largely on the symptoms a patient presents with rather than on what is happening in their tissue, so some people may receive therapies that serve them less well.
The team turned to a blood protein, serum neurofilament light chain, which nerve cells release when damaged. Its level can reveal how active the disease is. Using a machine learning model called SuStaIn, the researchers combined these measurements with brain MRI scans and traced how the changes unfolded over time.
Two patterns emerged. In the "early sNfL" group, the protein rose early, alongside damage to the corpus callosum, the bridge between the brain's hemispheres, and lesions appeared more quickly, suggesting a more active disease. In the "late sNfL" group, the order reversed: shrinkage in regions such as the limbic cortex and deep grey matter came first, and the protein rose later, pointing to a slower course.
Dr Arman Eshaghi, the study's lead author, said MS is not one disease and that current categories fail to capture the tissue changes clinicians need to understand. The practical hope is that early-pattern patients could be offered stronger treatments sooner and watched more closely, while late-pattern patients might benefit from therapies aimed at protecting brain cells.
The MS Society's Caitlin Astbury welcomed the work as exciting, noting that current definitions rest on clinical symptoms that often miss what is happening in the body. Around 20 treatments exist for relapsing MS, but many people with progressive forms still have few options. The results will need confirmation in further studies before they can guide practice.
Researchers at University College London and the company Queen Square Analytics have used artificial intelligence to find two biological subtypes of multiple sclerosis that had not been identified before. The work, published in the medical journal Brain, rests on data from around 600 people with the condition.
The problem it addresses is a familiar one. MS affects millions of people worldwide, yet treatment decisions still lean mainly on the symptoms a patient has, not on what is happening in the tissue. As a result, some patients may be given therapies that work less well for their particular form of the disease.
The team focused on a blood protein called serum neurofilament light chain, or sNfL. Nerve cells release it when they are damaged, so its level in the blood can signal how active the disease is. Using a machine learning model known as SuStaIn, the researchers combined sNfL measurements with brain MRI scans and looked for patterns in how the changes unfolded over time.
Two patterns emerged, which the team named "early sNfL" and "late sNfL." In the early group, high levels of the protein showed up early in the disease, together with damage to the corpus callosum, the structure that connects the two halves of the brain. These patients also developed brain lesions more quickly, which points to a more aggressive and active form of MS.
In the late group the sequence ran the other way. Brain shrinkage appeared in regions including the limbic cortex and deep grey matter before sNfL levels rose. That pattern looks slower-moving, with more visible nerve damage arriving later.
Dr Arman Eshaghi, the study's lead author and a researcher at UCL, said MS is not one disease and that current subtypes fail to describe the underlying tissue changes that clinicians need to understand in order to treat it. Combining an AI model with a widely available blood marker and MRI, he said, showed two clear biological patterns for the first time. He added that this could help clinicians see where a person sits on the disease pathway and who may need closer monitoring or earlier, targeted treatment.
The practical implication is that patients with early sNfL MS could in future be offered higher-efficacy treatments sooner and be watched more closely. Those with late sNfL MS might benefit from other approaches, such as therapies meant to protect brain cells and slow degeneration.
Caitlin Astbury, senior research communications manager at the MS Society, called the study an exciting development when speaking to The Guardian. She noted that it used machine learning to combine MRI scans and biological markers from people with relapsing-remitting and secondary progressive MS. Current definitions, she said, rest on clinical symptoms, and those categories often fail to reflect what is going on in the body, which can make the disease hard to treat effectively.
Around 20 treatment options exist for relapsing MS, and some therapies are starting to emerge for progressive forms, but many patients still have limited or no effective options. The findings will need validation in further studies before they can shape clinical practice.
Citations marquantes
By using an AI model combined with a highly available blood marker with MRI, we have been able to show two clear biological patterns of MS for the first time.— Dr Arman Eshaghi, lead author, UCL
Current definitions are based on clinical symptoms, and these categories often don't accurately reflect what is going on in the body.— Caitlin Astbury, MS Society (paraphrased, via The Guardian)
Envie de l'histoire complète ?
Lire l'original sur Euronews ↗
Technology