Transcription:
DNA Methylation Profiling in the Diagnosis of Brain Tumors

Dr. Andrew Wilner: Welcome to Better Edge, a Northwestern Medicine podcast for physicians. I'm your host, Dr. Andrew Wilner, Associate Professor of Neurology at the University of Tennessee Health Science Center and Division Director of Neurology at Regional One Health in Memphis, Tennessee.

Today, we're discussing the emergence of DNA methylation profiling for brain tumors. As of November, Northwestern Medicine is one of the first institutions worldwide to offer this advanced diagnostic technique on a routine clinical basis.

Joining me in this panel discussion are three physicians from Northwestern Medicine. Dr. Craig Horbinski is the Director of the Division of Neuropathology and a professor of Pathology and neurological surgery. Dr. Lawrence Jennings is the Director of Genomic Pathology and Associate Professor of Pathology. And Dr. Lucas Santana dos Santos is an Assistant Professor of Pathology and Director of Bioinformatics.

Doctors, welcome. This is really exciting for Northwestern Medicine and patients all over the Midwest. Dr. Jennings, I'd like to start with you. I remember learning in medical school that we classified brain tumors by histology. You know, mitosis, hypercellularity, blood vessels, necrosis. It seems we've come a long way since then. Would you describe DNA methylation?

Dr. Lawrence Jennings: DNA methylation refers to the addition of a methyl group to a small subset of cytosines throughout the genome. But these methylcytosines are not randomly dispersed, rather the methylcytosines inhibit transcription and function as part of the epigenetic regulation of the cells. It's the reason cells from different tissues express different genes, even though they had the exact same genetic sequence.

Dr. Andrew Wilner: Dr. Santos, can you bring me up to date on how we classify brain tumors in 2021? Where does DNA methylation fit in?

Dr. Lucas Santana dos Santos: Yeah, sure. For DNA methylation, we can distinguish between different tumor types with very high precision. Here at Northwestern Medicine, we built a classifier, artificial intelligent classifier, which can discern DNA methylation signatures of all these different tumors and, with very high confidence, it can predict and tell you what that actual tumor is. So with that, we get much more confidence and we can get much more precise diagnosis and distinguish between very similar tissue types and tumor types.

Dr. Craig Horbinski: Yeah. If I may step in, one of the things that really is interesting about this is like Dr. Jennings said, every cell has its own genome, of course, and each tissue type, each cell type has a unique methylation pattern. Certain genes are turned on, others are turned off. So for example, skin cells are going to have a different methylation pattern than lung cells than bowel cells, et cetera. And tumors also have their own unique methylation fingerprints really. They're fingerprints, so such that we can tell and discriminate effectively by methylation profiling between two different kinds of tumor, very different kinds of tumor, even if they look very similar under the microscope.

Dr. Andrew Wilner: So this is really an improvement on what we can see not with the naked eye, but with the eye, a trained eye looking through the microscope is going to look at this tumor and that tumor, but by adding DNA methylation, you can really discriminate with good confidence the tumor type. Is that correct?

Dr. Craig Horbinski: Yes. In fact, that's what the overall consensus is. In the research field, as well as on a prospective clinical basis, we might think a tumor is one thing histologically. But by methylation fingerprinting, it suggests something else entirely. And when you follow these patients over a long enough period of time to see how the tumor actually behaves, how the patient actually does, the methylation wins out virtually every single time.

Dr. Andrew Wilner: Okay. Dr. Horbinski, now that we can distinguish the tumor type, how does Northwestern Medicine use this information to direct therapy?

Dr. Craig Horbinski: Well, we're doing it now on a routine basis for all of our in-house cases, as well as any cases we receive on a consultation basis. And what we'll do is evaluate the tumor or the tissue under the microscope, just to get a sense of, "Okay, what are we dealing with here? Where's the best spot to analyze for genetic testing?" and we do next generation sequencing and methylation profiling in Dr. Jennings' laboratory in partnership with Dr. Santos. And then when we received the results, we integrate the methylation and the sequencing data in with what we saw pathologically under the microscope, as well as clinically, just based overall on the patient's age, sex, tumor location, et cetera, and issue an integrated diagnosis. And that integrated diagnosis takes into account all this disparate information, such that the clinicians, the treating physicians have a much clearer sense of exactly what this tumor is. And as a result, how to treat the patient's. So different diagnoses, of course, are going to suggest very different kinds of treatment.

One type of tumor might need radiotherapy, but maybe not chemotherapy. Another type might need both. Certain tumors might not need anything after surgical resection. Maybe it's just watchful waiting to see if the thing grows back. And that sort of methylation profiling is so powerful because it adds another layer of reliability of confidence in our diagnoses.

Dr. Andrew Wilner: Dr. Horbinski, tell me a little bit more from a practical point of view. So it sounds like I need a tissue sample, So I need a biopsy. Is that right?

Dr. Craig Horbinski: Yes. Especially in the brain, if it's a difficult to reach spot or there's other reasons to just be conservative, they'll take a small tissue biopsy. But if they're pretty confident that the patient would benefit from a larger surgical resection, they will just go right in and do that sometimes too.

Dr. Andrew Wilner: All right. So we've got a piece of the tumor, a technician runs it down to your laboratory. The doctor's upstairs waiting, how long is it gonna take to get a result?

Dr. Craig Horbinski: Well, the interoperative consultation, we're just taking a quick look at it under the microscope to see if the tissue, the biopsy is potentially diagnostic, that should only take a matter of minutes, maybe 10, 15 minutes total from start to finish.

Dr. Andrew Wilner: And to get the methylation result, how long is that process?

Dr. Craig Horbinski: Well, that's got to take days. Doctors Jennings and Santos can speak to that.

Dr. Lawrence Jennings: So from the time we receive a sample, usually in the form of a formalin-fixed paraffin-embedded tissue on a microscope slide, it's a minimum of five days technical work. Our turnaround time therefore is measured in days between seven and twelve days. During that time, we're usually also doing sequencing on the same tissue sample to port back to Dr. Horbinski and his team, both the NGS results as well as the methylation array results.

Dr. Andrew Wilner: Now, this test is not routinely available in most places. Is that because it's difficult to do or expensive or how come everybody isn't doing it? Dr. Santos, would you like to tell us?

Dr. Lucas Santana dos Santos: Yeah, sure. So we use a microarray chip to run those samples. Technically, that's not complicated, but it's complicated in the AI. We use our AI algorithms, artificial intelligence, to create a classifier, and that is computation intensive. And that requires a lot of expertise to train an algorithm and make sure the algorithm is validated and actually it's outputting high confidence results.

Dr. Craig Horbinski: If may step in, so the idea here is this test is powerful because it consists of a library of thousands of known tumors, known brain tumors, with each of which of course has its own methylation fingerprint, so a methylation profile. So when you take a new case and you run it through and you get the methylation results, doctors jennings and Santos, what they do is they compare it with the existing library to see if there is a match. And if there are some matches, what is this new tumor? What is this unknown tumor matching up against in the known library? So it is an example of machine-based learning of artificial intelligence, so that the larger the library, the more variants of a given tumor that are in that library, the greater the likelihood that this new unknown tumor is going to find a match. And that's where we're fortunate to have doctors Jennings and Santos because not every institution has that level of expertise.

Dr. Andrew Wilner: So is this a library that you can share with other institutions or is it something you develop all by yourself?

Dr. Lucas Santana dos Santos: So that library was data that was published. So we use the data that was published by the Hanover group and that publically available. On top of that, we use our own samples to boost training of that algorithm. And there is maintenance of that algorithm that needs to happen, so every so often we retrain their algorithm to make it better and more robust and able to account for the variability of new samples coming in and new classes. So there's refinement that goes in that process. So it is a very much of a continuous process for improvement.

Dr. Andrew Wilner: So in fact your library is updated with each case and will become more sensitive and specific as time goes on. Is that right?

Dr. Lucas Santana dos Santos: Yes. As time goes on, what happens is we do retraining of the algorithm, usually we do in batches. It doesn't happen every day, but it happens. When we have enough cases, then we retrain the algorithm, retest it, revalidate it. And then we release a new version of that algorithm. And that new version will be more robust than the previous version because you have more samples, more variants, different things that we never saw before, and that's how it works. The plan is to retrain the algorithm every quarter or so.

Dr. Craig Horbinski: And that's the beauty of this kind of concept as opposed to most other things, is the more you do it, the better the tool gets. It's a self-learning, self-correcting process. As a result then, right now, of course, we're only offering it in the context of central nervous system tumors but this approach can be used for all kinds of tumors, you know, difficult to diagnose sarcomas, other kinds of cancers, metasticies with an unknown primary site, you know, where if somebody has metastatic disease, but we're not, you know, we're not sure where it's coming from. This methylation profiling can be extraordinarily powerful in those contexts.

Dr. Andrew Wilner: I'd like to put this technique in perspective for referring physicians. For example, how many brain tumor cases do you look at each year approximately? And then how often for that number of cases would you use this DNA methylation technique?

Dr. Craig Horbinski: Well, in our neuropath division here at Northwestern Medicine, we have a pretty robust service. We're looking right now at about a thousand brain tumors, almost a thousand brain tumors of all kinds, both primary and metastatic tumors, and that nervous system tumors as well.

I'd say roughly around 40%, 45% of those cases are consult cases. They're coming from outside, mostly from molecular diagnostic workup. And we're anticipating doing methylation profiling on the majority of them, certainly all the gliomas or anything suspected of being a glioma and all the other kinds of embryonal tumors, like medulloblastoma that sort of thing. And certainly at least some of the higher risk meningiomas or meningiomas that we suspect are going to behave more aggressively, we're going to do it on those. And then on some of the remainder, it'll be an ad hoc basis. If we see a tumor that we're not entirely sure what it is, our thresholds for ordering methylation profiling is going to be low because it can be so helpful.

Dr. Andrew Wilner: I'd like to follow up on that. How can Northwestern Medicine partner with referring providers using methylation profiling?

Dr. Craig Horbinski: Well, it's very easy actually, because we already have an existing pipeline whereby people can send us consult cases. Like I said, 40% to 45% of our volume currently are consultations from outside hospitals, and this is from all over the country. And in some cases, even overseas. So we already have a nice pipeline and an infrastructure whereby we receive external material. And the nice thing is it's such a robust assay. All we need are some unstained slides. And then we received the case, we look at the histology, the basic histology, get a sense of what this thing might be. And we send the remaining material to doctors jennings and Santos and they do their magic and that's it. So it's a fairly straight forward operation.

Dr. Andrew Wilner: In my reading, there was a suggestion that pediatric tumors are more difficult to diagnose than adult tumors. Is that correct?

Dr. Craig Horbinski: Well, many of them are. In fact, it's interesting you should mention that because we just came out with the fifth edition of the World Health Organization classification for central nervous system tumors. And probably the one section that underwent the most substantial revisions compared to the 2016 edition was the pediatric tumors, because so many of them are now being defined by specific genetic alterations, like histone-mutant tumors, but there's others as well. And the embryonal tumors in particular have undergone substantial revisions or reclassifications based on otherwise esoteric fusions and mutations that nobody in their right mind would have ever thought existed, but we're finding that a lot of these pediatric tumors have those driver mutations, these really weird driver mutations that do have clinical significance. And so as a result though, under the microscope, it's virtually impossible to distinguish these different kinds of tumors with different driver mutations because they look so similar. Methylation profiling is proving to be extraordinarily robust at discriminating between those different kinds of tumors with those different driver mutations.

Dr. Andrew Wilner: Well, that's great. You know, I've got to add that this discussion is inspiring to me because what I remember from medical school a long time ago was that there was a long way to go in sorting this stuff out. And it looks like a definite progress has been made. Dr. Santos, what's the long-term vision for methylation profiling and the work you're doing at Northwestern Medicine?

Dr. Lucas Santana dos Santos: There are a couple of things that we want to explore. Following up on what Dr. Horbinski has said previously, one of the things that we want to do, expand that to different tumor types, right? To have like different assays for multiple tumor types. The other interesting thing is also related to machine learning and algorithms, right? We are always interested in getting different algorithms, see how they perform and add new features. For example, adding new data types, like adding images, adding different types of genomic data that's not methylation, to see if we can come up with even better and more advanced and more robust algorithms.

And the other is research. Like, there's a lot interesting in research. DNA methylation can really shed some light in how dynamics, cellular dynamics occur, stem cell expression, how things interact, and gives a much bigger picture in terms of biology of the cancer, tumorigenesis, drug resistance and things like that. So we're trying to explore these three different things using methylation at the beginning of that.

Dr. Andrew Wilner: Dr. Horbinski, as we wrap up, what else would you like the audience to know about methylation profiling or referring patients to Northwestern Medicine?

Dr. Craig Horbinski: We are open for business. We can take consults from all over. Like I say, we have a pipeline, we do this on a regular daily basis. And all you have to do is check out our website, Northwestern Pathology Consultations. It'll come right up with all the instructions and everything. Very straight forward. And we're glad to work with you.

One thing in particular that I think sets us apart is that not only you're going to get the methylation profiling, of course, but you get it integrated into the overall picture. So it's integrated with histology, it's integrated with the overall clinical profile. And if there's any difficulties, any tricky cases, that sort of thing, anything at all that's unusual, we're very aggressive-- we're very prospective, let's put it that way, about reaching out contacting, discussing with the outside pathologist with the outside treating physicians, just to make sure everybody's on the same page.

Dr. Andrew Wilner: Well, thank you, doctors, for this very interesting discussion.

And that wraps up this episode of Better Edge, a Northwestern Medicine podcast for physicians. To refer your patient or for more information, head on over to our website at breakthroughsforphysicians.nm.org/neuro to get connected with one of our providers. Please remember to subscribe, rate and review this podcast and all the other Northwestern Medicine podcasts.