B (10:28)

Thank you very much, John, for having me. It's a pleasure.

A (10:31)

Your latest research suggests that the way we've been labeling Ms. For decades now as relapsing, remitting or secondary progressive doesn't tell the whole story. You've identified two new biological subtypes. Can you explain what these are and how they differ from the categories we're used to?

B (10:52)

You know, the way we categorize Ms. Disease courses is based on symptoms. It's been so since 1950s, so about maybe six, seven decades we have been relying on symptoms alone to. To classify patients. As people listening to a podcast may know better than we do. People living with ms, they are either labeled as relapsing, remitting Ms. Who later may develop a secondary progressive course. And also we have a separate group who may have progressive disease since the onset, patients who are cold and primary proflagellas. This is. It's been known that these sorts of categories have been outdated. We know this because they don't necessarily correspond with what we call biological changes or pathophysiological changes, that is basically changes in the tissues. Why is that important? Because as we move towards better diagnosis of Ms. And potential in the future preventing ms, it becomes important to move away from looking at Ms. Purely based on symptoms. So the idea would be to have a way of what I call biological classification of Ms. To stop the progression before it stops and potentially even prevent Ms. In the future. So that was the idea. And over the past five, six years, many different groups, including myself here in London, but also other groups in Germany, using blood samples, we've been looking at ways, if we can use artificial intelligence or AI to group patients in a way that doesn't necessarily have symptoms. So we asked computers if we could cluster, group or profile people with Ms. In similar groups that have similar trajectories on the data, or biomarkers, what we call biomarkers. So that is, for example, very specifically, as you mentioned, serum neurofilament, I'll use the short name, that is snfl. And what is probably also very available, that is mri, brain mri. So we know this, this is nothing new about dms. And previously we used MRI Alone to classify patients. So we, in that study in 2021, we found three different groups of patients. But now what we've done, we've gone first, we tried to simplify that classification. So in depth study we have patients who had early brain shrinkage, patients who had white matter plaques or white matter lesions, we call them lesion lead, and a third group of patients who had diffused damage in the brain, which we call them normal appearing white matter lead. Now recently, in the past month, the study published in Brain, we tried to simplify that, but the main question we answered is that adding snfl, the serum neurofilament improves that sort of classification, also simplifies that and is a step towards being able to have these source of biological classifications. And what we hope is to improve the classification of Ms. Based on the evidence provided in this study. Patients who had early SNFL elevations, as you mentioned, means patients who probably have a more active disease. Also these were patients who had early active MRI white methodes activity and patients who had late snfl. So patients had later elevation of serum neurofilament with a brain shrinkage later on. So using this source of classification, one would hope in the future to better find treatments or match treatments for specific persons living with ms, preventing disability progression before it happens, and also potentially. So that's a interesting topic, potential to prevent Ms. Before the symptoms emerge.

A (15:22)

Well, I want to get into some of that in a moment, but I'd like to ask you a little bit first about the AI tool you used, a machine learning tool, I believe it's called Sustain, to find these patterns for those of us who aren't tech experts, how did AI see something in these scans and blood tests that human doctors might have missed?

B (15:44)

Yeah, so we use the method you mentioned, Sustain. That is what we call unsupervised method. So in machine learning, AR parlance, we have usually two different ways of looking at data, one of which is supervised. Is like having a teacher teaching you what is what, what we call layoffs. For example, if you want to predict, let's say if you want to predict disability, we use supervised learning in the context of Ms. For example, this is unsupervised. That is we somehow, I mean in simple language, we throw data at the model, at the algorithm, and with some constraints, we look at how many clusters can best represent the data. Specifically in this case, the changes we observed was the patterns of change in MRI biomarkers. That includes brain volumes, but also white matter lesions, but also diffuse damage. Using a specific biomarker we call T1, T2 ratio. But without getting to those kinds of details, the idea here is that the AI is able to look at this huge amount of data from. In this case we had more than 600 patients, followed about 2, 3 years, depending on the clinical trial we used. These are changes that are first cannot be detected, especially on mri, cannot be detected by even expert radiologists that require quantification by AI tools. But also in case of SNFL or certain neurofilament on its own is a number. So I think that can be detected easily. But the important point is the pattern, so how it covariates, how it changes with other variables that in this case mri. So the idea is to have a very complex disease such as multiple sclerosis. We hope to be able, and that's what we're doing exactly, to use AI to better understand these changes. And I think the other promises to do so in a way that is unbiased. So it's not really, we don't. We had to select certain variables to input to the model, of course, but other than that, it's kind of, as I said, it's unsupervised. So it's very much, I can say unbiased, but it's less biased compared to manually or cherry picking, for example, certain aspects of the disease or virus.

A (18:23)

Your study identified two new subtypes called early SNFL and late snfl. The early SNFL subtype is characterized by high serum neurofilament light chain levels very early on and shows changes in a part of the brain called the corpus callosum. What does this mean for a patient's experience? Could this be considered a more aggressive form of ms?

B (18:51)

Potentially, yes. John. So this basically hints to a kind of disease that also was more active, that is, you know, is also more visible. When I say more visible means it shows itself early on as opposed to kind of, you know, Ms. Spectrum where it's more difficult to, to detect changes. As you know, we do have disability worsening or disability accumulation since the very beginning of ms, even in patients with relapsing rheumatic Ms. So that kind of disease, potentially those patients are likely to be in the late SNFL subtype, whereas patients who have relapses more active and also on mri, they're more likely to be in the early SNFL group. We knew about this. We knew some patients are more active than others. But how these two groups, the novelty and what is innovative in this study is really we bring this together. So you kind of what we call multimodal AI, you go above and beyond one aspect of the disease, that is either MRI or blood biomarkers, but you integrate them together to get a more holistic understanding of disease. And again, the idea is to improve whatever we can do. So improve prognosis, for example, improve in the future, potentially prescription of treatments, and as I said many times, potentially also moving forward, looking at how we can use these sorts of changes to prevent disability, worsening and potentially the diagnosis in the future.

A (20:31)

I think you've already started to answer my next question a little bit. It's about the late SNFL subtype, which is characterized by lower levels of serum neurofilament light chain early on. But the AI saw quiet shrinkage in the deep gray matter before those SNFL levels went up. And again, as a quick review for our listeners, you can think of that deep gray matter as sort of the mainframe that, that powers the brain. Dr. Ishaki, why is this quiet progression and deep gray matter so important to understand?

B (21:08)

Yeah, so this is probably the most mysterious and less understood part of Ms. And as I said, this happens early on, probably in all patients. So the silent part of disease, it has been coined differently by different people. Some people call it smoldering disease. We know about these days about period progression independent of relapse. This is the activity we used to call this disease progression. So there are lots of different ways of. I think if you look at the. Probably your podcast, but also in the literature, scientific literature, so it's very mysterious. We're starting to better understand it in our study. So we're specifically in this study, we looked at the clinical trial data. It's very short term to a maximum in some patients, three years of data. It's very short. To be able to capture some of these changes, specifically, the silent progression and the slow progression of disease usually requires 5 to 10 years of follow up. So it's very difficult to comment on that using this, you know, specifically this study. But we saw, as you mentioned, so early brain shrinkage, specific deep gray matter. So deep gray matter, as we say deep because it's really deep. If you open up the brain, it's really deep is in the center. And specifically this structure we call thalamus is the part of the brain that very much relates to walking ability and what doctors call motor impairment. You know, it could be thinking about walking, moving our hands and so on. These sorts of functions very much related to deep gray matter, actually. So again, this shows that there is a group of patients that go through this phase and as I said, it's very mysterious. So it's really one of those open questions in the ndms. And when I think about it, so we don't really know exactly what is driving that process and certainly we're not really able to stop it with the existing treatments.

A (23:20)

Your study found that people with early SNFL were 144% more likely to develop new lesions. Could knowing which subtype a patient falls under help to predict what their Ms. Might look like 5 or 10 years down the road?

B (23:38)

Yeah, so I think the key point exactly as you mentioned, if a. So again these are, you know, we talk about the future. If these were to deploy, for example in a clinical setting, if for example someone is labeled at early snfl, they're likely to be more active. So clearly in the conversations we may have in the future, if this comes into the conversation, it becomes part of the conversation. You will think of potentially prescribing higher efficacy, stronger treatments that can prevent disability before it happens. That's one part of the, the question. The other side is we know that you know, more activity requires, you know, is more aggressive disease, more what we call inflammation requires also a more efficacious treatment. So when you think of, you know, we have 20, 25 treatments depending on the country the audience are based in, but from those then one of the biggest question is, and there's always a risk associated with the treatment, is it worth, for example, starting with a higher efficacy treatment? And potentially we can or we should in the future be able to use AI or in general more information the better in this source of conversation to have this risk benefit ratio and make decisions based on that.

A (25:07)

That's exactly what I was wondering. As a matter of fact today with all of those approved treatments, it's not unusual for someone who's living with Ms. To be put on a few different disease modifying therapies before landing on one that seems to work best for them. So could integrating these new subtypes into clinical care help a neurologist find that best medication for a specific patient on day one?

B (25:37)

Absolutely, that's the hope. That's the hope to be able to do that. What I see in the future is not necessarily my study, but again the evidence is converging to be able to have algorithmic prediction to feed into the conversations in the future that people living with Ms. And the doctors nursing team and also people who care for people with Ms. Can have to improve that decision Making. That's exactly the idea. But also, as I said, we have treatments, but we are not really able to cure Ms. So in the future for new clinical trials, I think that's also important. If we are able to better profile the underlying biology, you can think about precision clinical trials. So think of clinical trials where we can use AI then to enrich trials with a specific treatment mechanisms that would target that specific biology. So I'm thinking about potential future remyelination trials, trials that look at the restoring, for example, lost myelin. If that happens in the future, you would want to have people who are most likely to respond to treatment. So it's really two aspects to it. One is the clinical care, which I really think it won't be, you know, I think it would be in the future, maybe we think in five, 10 years. It's a lot of work to get these sorts of ideas into the clinic. But also clinical trials probably is easier in terms of regulation to do. What you said in a clinical trial research setting earlier, the clinical care is probably, you know, we're thinking about probably a decade of work before we can get.

A (27:29)

It may be a decade out. But I think you've just given us a peek into what the future of Ms. Care is going to look like. And certainly in terms of being able to influence how clinical trials are designed and run, the better the trial design, the better the outcome in terms of developing new treatments faster. I want to congratulate you on this work. It increases our understanding of the biological mechanisms that are driving multiple sclerosis and it truly moves the entire field forward. Thanks so much for talking with me today.

B (28:06)

It's a pleasure, John. Thanks very much for having me.

A (28:08)

That's going to wrap up this episode of Real Talk Ms. Real Talk Ms. Is powered by the National Ms. Society and you can share this episode of the podcast by letting your friends or family members know that all they have to do is point their web browser@realtalkms.com 438. You'll find that link in today's show Notes, so you can easily copy and paste it right into an email or a text. I'm John Strum. Thanks for listening. Stay safe and make healthy choices.

B (28:49)

Sam.