Chemograms for Smarter Chemotherapy

Podcast Transcript

Dale Shepard, MD, PhD: Cancer Advances, a Cleveland Clinic podcast for medical professionals. Exploring the latest innovative research and clinical advances in the field of oncology. Thank you for joining us for another episode of Cancer Advances. I'm your host, Dr. Dale Shepherd, a Medical Oncologist and Co-Director of the Sarcoma Program at Cleveland Clinic.

Today, I'm happy to be joined by Dr. Jacob Scott, a Radiation Oncologist, Professor and Staff Physician Scientist at Cleveland Clinic. He's previously a guest on this podcast to discuss predicted survival using genomic associated radiation dose for HPV positive oropharyngeal squamous cell carcinoma. That episode is still available for you to listen to. He's here today to discuss how a chemogram advances genetic testing for cancer care. So, welcome back.

Jacob Scott, MD, DPhil: Thanks, Dale. Nice to be back.

Dale Shepard, MD, PhD: So, give us a little bit of an idea again about what kinds of things you do here at Cleveland Clinic.

Jacob Scott, MD, DPhil: Well, as you said, I actually am lucky enough to take care of sarcoma patients with you. So it's always fun to meet you on this different place, because we were talking shop just earlier. So I take care of sarcoma patients in a specialty clinic that we work in together using radiation therapy. And then I run a laboratory here that really thinks hard about how to realize the potential of personalized medicine.

And like you said in our last podcast, that was using data science methods to realize the potential for personalized medicine in radiation oncology. But today we're going to talk about that in medical oncology. And I think one of the things that... I think personalized medicine is a big, big buzzword. Lots of people are working on it. Mutations, targeted therapies, we all know. What we're working on here, and I think your listeners will like this. We all remember from med school, or the internists in the crowd will remember from their practice, an antibiogram. We all remember it, right? And it makes so much sense. You've got a patient, they're sick, they go septic, they spike a fever, you culture their blood, you send it out. Two days later, it tells you what antibiotic to give. Why don't we have that in oncology? And that's sort of the basis of this paper. The fundamental question is, can we develop a chemogram in the same way that we have an antibiogram?

Dale Shepard, MD, PhD: Excellent analogy. Now I have a crazy question. Sometimes on the podcast we get to get behind the scenes here. You are a radiation oncologist, so every time I have a radiation question, I reach out to you or Shauna. This is a chemogram. How as a radiation oncologist did you get involved in how we pick chemotherapies?

Jacob Scott, MD, DPhil: Well, we're all oncologists here, Dale. So in rad-onc, once you've figured out a method by which to optimize radiation therapy, where else do you go? So there's not radiation part two. I guess I could think about particulates and heavy ions, but instead I wanted to go to where most patients are treated. Almost every patient I see together with you, the vast majority are getting chemotherapy, but really we're still first line, second line, third line, kind of marching down the order with very little, and again, this is not in the targeted therapy realm, in patients without actionable mutations.

I'm not going to say that we're shooting blind, but we are sort of going on a population level basis rather than a personalized basis. And so what I realized over the course of the last, gosh, decade, was that the same mathematical approaches I was using to try to optimize radiation therapy using genomics I could use for chemotherapies, especially the sort of non-targeted cytotoxics that are still day rigor for the vast majority. I think something like 80% of patients are getting some older drugs that aren't targeted.

Dale Shepard, MD, PhD: Yeah. At some point in time in their treatment, they're going to see these.

Jacob Scott, MD, DPhil: Absolutely. And there's some exciting concepts in evolutionary oncology in particular, that over the course of a patient's journey, a chemotherapy that has failed them previously may actually come back and be useful again later on in their journey.

So really, how can we identify those maybe transient states of sensitivity to specific drugs? So it might be on day one, you're super sensitive to cisplatin. Great. If we knew that, we can give you cisplatin, but we also recognize that the reality in metastatic disease especially is that these tumors change and evolve over time. So after that cisplatin stops working for your patient, what next? And of course there's clinical trials that assess this, second line's this, third line's that, but it really might be that each patient has a different answer.

And so we really are seeking in this paper to understand a genomic method by which to ascertain what your best sensitivity is. That's that chemogram. A ranking. Hey, you know what? For this patient, now that the drug has failed them in a first line drug, what to do next? Let's get another biopsy, maybe take some circulating tumor DNA, liquid biopsy, if you will, and can we predict that?

And so that's really the way we're going. So we're trying to do this in an agnostic way. Sorry, in a mutation agnostic way, looking at the transcriptome, to try to really figure out what existing drugs can we repurpose. And I think that's a real key here, because drug development is a decades long and billion-dollar enterprise, that one academic lab can certainly contribute to, but it's difficult to take from zero to hero, if you will.

However, if we can figure out clever ways using novel data science and artificial intelligence methods, for example, to repurpose drugs we already have in our formulary, we could help every patient.

Dale Shepard, MD, PhD: And I guess you mentioned time and money. And there have been people sort of trying to envision ways to pick the right drugs in other systems in the past. And countless people have been in the office like, "hey, we can do xenografts and we can set up animal models of your patient's tumor and you can figure out..." But it's like a month long process and my patient needs something yesterday. So tell us a little bit about how you approached it and things like, how long until you get results and things like that?

Jacob Scott, MD, DPhil: Yeah. You hit the nail on the head. So there's a whole bunch of different methods by which, like you said, you can take some tumor out. And almost exactly the same way that an antibiogram does it in real life, you see that tumor, maybe you put it on 20 different in vitro plates, maybe you put it in 20 mice, maybe you make organoids, treat them. But now you're talking about time, and you're talking about a lot of money.

And also the heterogeneity is going to start biting you because at some point you're now taking a tumor out, maybe it grows in the dish, maybe it doesn't, maybe it grows in the mouse, maybe it doesn't.

Dale Shepard, MD, PhD: Maybe it grows in the dish a different way that it's in the person.

Jacob Scott, MD, DPhil: Almost for sure it grows in the dish a different way than in a person, and same as the mouse. And so I think that those are incredibly important approaches, but like you said, they're going to take weeks and even in those weeks, the tumor in the patient is changing.

So three weeks later you give me an answer, it may no longer be the right answer. So that speed scale you're talking about, and not to mention cost scale is hugely important. So what we're suggesting here is that with new next generation sequencing techniques, we can get that rank list straight out of the transcriptome. Now, that's not instant and it's not free, but it's really fast compared to those avatar-based metrics and it's really getting cheaper and cheaper as we go.

So with $1,000 or $2,000 assay from a company, you could get the transcriptome back quickly and run an algorithm and have, within a couple of days, a rank list of drugs you already have access to, you know how to use, you know the safety and toxicity profiles of, and you can go from there.

Dale Shepard, MD, PhD: How did you go about correlating gene abnormalities to clinical or the sort of efficacy measures?

Jacob Scott, MD, DPhil: Yeah. Great question. So as in many of these, I'd say early developed tests, clinical decision support tools, what we did is harness the power of the big data that exists in the form of both the TCGA, which is the cancer genome atlas, which is a large scale patient derived tumor database of sequencing together with large scale pharmacogenomic assay platforms. In this case, we use the genomics of drug sensitivity database, GDSC, which is a massive collection of I think 1000s of cell lines tested against 100s of drugs with genomics and outcome.

And so we sort of are blending... You know, obviously there's drawbacks to in vitro experiments and cell line work, but there's also information there. So taking that big data source, sort of modulating it with what we understand about co-signaling within humans, putting them together and asking our algorithms to tease apart the signal from the noise there.

Dale Shepard, MD, PhD: And then you mentioned earlier about biopsies versus liquid biopsies. So are you trying to look at using a biopsy where you're looking at essentially more likely to get RNA or DNA, but at least something in the needle? Or the liquid biopsies where you get the big picture of the whole patient, but maybe less sensitivity?

Jacob Scott, MD, DPhil: So right now, we're thinking about needle biopsies of solid tumors. I think the future, I would much prefer to use circulating liquid biopsy surrogates, but I do think that you're yet one further step away. And I think the technology is not quite there. Because when you sample plasma, circulating tumor DNA, like you said, you're getting a big picture, but it's not deep. It's very hard to get the same information, the quality that we currently need, but that's where we want to go for sure.

Dale Shepard, MD, PhD: And so, you had used the term targeted, and it's called a chemogram. Is this traditional cytotoxic chemo? Is there a thought that this could also incorporate sort of more the cell signaling targeted therapies, not the specific genomic things like a ALK inhibitor or an NTRK fusion, but more like an EGFR inhibitor or immunotherapies or things like that? Or is it just strictly chemo? And can we go to those other types of therapies as well with this?

Jacob Scott, MD, DPhil: That's a great question. So right now we have focused on the cytotoxics, mostly because there's a ton of data around them. I also think that many of the, let's say, pathway or signaling-based therapies, we can assay in other ways. So if you get the full transcriptome out, you're probably going to have information about those pathways and you could use it. You wouldn't need our chemogram-based approach.

And I would also say the same for immuno-oncology. I think that none of the drugs we looked at were immune oncology drugs. But that said, you think about Tim Chan's Lab here at Cleveland Clinic, there's really a good series of algorithms that exist already on tumor mutational burden. And a number of other things that are developed here as well that can really help us with those immuno-oncology drugs. But I think that our results at least have not extended yet to those.

But we're really actually trying to fill in that missing gap, because I think most of the work that's been done in personalized medicine to date has been finding mutations that have drugs that encode the same pathways. And then further the immuno-oncology drugs. And what's been left behind are the less sexy drugs, if you will, the sort of old workhorses.

Dale Shepard, MD, PhD: But they work. Patients don't like them, because they're like, "but that's 50-years-old." I'm like, "yeah, but it worked"

Jacob Scott, MD, DPhil: But it works, and it’s still first line therapy in the vast majority of situations, right? I think it's like 75% of patients, their first line therapy is going to be Gem, Cist, Carbo, Taxol, there's 10 of them that we looked at, and that's most of folks.

And I think that maybe to those of us who are at the tip of the spear, it's frustrating that 80% of patients, and probably more than 80% outside of academic centers, most folks who have metastatic disease are getting cytotoxics from 50 and 60 years ago. And so that's why we really approached it the right direction we could. Because if we can improve the outcomes in that set of patients, a 10% improvement in 80% of people is a big difference.

Dale Shepard, MD, PhD: Right. And oftentimes you, quite honestly, the first, second, third line of therapy is how they were developed, not necessarily how likely they are to work. And so it's a frequent conversation of this first line didn't work, but that doesn't mean your tumor's not going to respond better to what we traditionally use second.

Jacob Scott, MD, DPhil: Absolutely. And not even just the order of development, but the sort of intricacies of trial design over time. Many of these things are obviously... Evidence-based medicine is super important, it's our watchword, but it's double-edged sword, right? It cuts both ways, in that we have evidence that it's functional, but we're also bound by that evidence. And so I think one way to try to get around that is to use clinical decision support tools like this chemogram tool.

Dale Shepard, MD, PhD: And then I guess when we think about clinical decision tools, how much does... Well, everything that we have available in AI, how does this link to information when getting other sources, how does it help develop further this chemogram?

Jacob Scott, MD, DPhil: Yeah, great question. So I think we were... This chemogram project actually started with a gift from a family foundation looking at a rare cancer called Ewing sarcoma from the Carson Sarcoma Foundation. And they were really asking how in a rare disease can we... You can't run trials with 1000s of patients in these diseases, right? How can you understand personalized medicine?

And what we really sought to do there was to use experimental evolution to increase the size of our dataset. Because if you want to use some of these newer AI and machine-learning based tools, what they are is data hungry. So if you have a large enough dataset, you can pull signal from there that would otherwise be lost in the noise with older methods. But if you have a rare disease, the data sets are small.

And so what we've really been working together to do is combine novel evolutionary oncology laboratory techniques to expand our data sets together with these more modern AI and machine-learning tools to distill the information from within those. And I think that that's really going to be the way forward. The older data science tools are incredibly useful and have gotten us a long way, but in the newer data sets, the larger data sets, we really need to harness the powers of artificial intelligence.

Dale Shepard, MD, PhD: You used a couple of large data sets to kind of put together the correlations and things. What's being done clinically to sort of check on how well this works, how predictive it is?

Jacob Scott, MD, DPhil: Great question. So I think what we're working on now is prospective registry trials. So in the absence of, or in the run-up to interventional trials, what we're able to do with a decision support tool like this is simply ask the question in a treating physician's hands.

And so once you get outside of the standard of care, obviously if you're first and second line therapy, you're following evidence-based medicine. However, as well as I know, many of the patients that we see have already gone through the standard lines and are in uncharted territory. And so the treating physician has a number of choices, usually based on gestalt, maybe the patient's toxicity profiles or pharmacogenomics when it comes to how they'll respond, or if we're lucky to have a targeted one mutation.

But for the most part, I won't say it's guesswork, but it's educated guesswork. And so what we can do is we can say, "hey, let's take this tool, assay the patient's tumor, rank the chemotherapies we have." And then ask how? Whatever the physician chose, we won't tell them how to choose it. Whatever the physician chose, we can then compare and say, "hey, they chose what we think was the worst. Did the patient do well? Did patient do poorly and vice versa?" And so I think that's a bridge between where we are now, which is an informatics tool, and an interventional trial is that middle step, and that's what we're doing now.

Dale Shepard, MD, PhD: Are there any particular cancers that you think this has been more predictive than others or is it kind of across the board?

Jacob Scott, MD, DPhil: You know, we started...There's a paper upon which this one stands, that it was a cisplatin sensitivity score by itself, and that was designed to look at any squamous cancer. And so I would say in this case, we have stronger evidence in the... And of course, squamous are across the body. But I think that's where we're going to start, it's a little bit, it's not quite pan-cancer, but it's a large swath of them.

Dale Shepard, MD, PhD: And then I guess the other question would be, if we're thinking about giving a systemic therapy, it could be neoadjuvant, could be adjuvant, could be metastatic. Are you thinking this primarily has a role right now in a metastatic setting?

Jacob Scott, MD, DPhil: Yeah. I think for sure at this point. Again, the double-edged sword of evidence-based medicine, I think really for tools like this, novel tools like this, we're not going to be in the first and second line setting together with patients who are likely metastatic disease. If we're in the realm of curability in first line and localized disease. Outside of a robust clinical trial platform, I think that this would be not what you'd want to use right now. You want to stick with what we know.

Dale Shepard, MD, PhD: Makes sense. What's next with this?

Jacob Scott, MD, DPhil: Yeah, great question. So I think, like I mentioned, we're interested in going into observational registry trials. Specifically, we're going to start in Ewing sarcoma, and then maybe move into some other disease sites. Maybe we could talk about sarcoma in our group, and then really thinking about what other data sources for external validation exist.

So looking at existing clinical trials that have happened, can we dig down into those and see what our chemogram would predict, and if that really plays out in the existing trials that have happened. Further, in the setting of the Biobank that we have here at Cleveland Clinic, we now have this massive repository of patient samples who've been treated, we have their treatment histories, and so linking together with the sequencing of those tumors, together with their treatment history and outcome, it's a goldmine there. Because we really have the ability to look at that real world data without even having to intervene on patients.

There's an infinite number of clinical trials that have been done, if you want to think of it that way, in that Biobank. It's just a matter of distilling the data out. So that's where we're going next.

Dale Shepard, MD, PhD: Excellent. Jake, always a pleasure, always good to talk to creative thinkers and appreciate you coming and give us some insight.

Jacob Scott, MD, DPhil: Thanks for being here, Dale. Appreciate you.

Dale Shepard, MD, PhD: To make a direct online referral to our Cancer Institute, complete our online cancer patient referral form by visiting clevelandclinic.org/cancerpatientreferrals. You will receive confirmation once the appointment is scheduled.

This concludes this episode of Cancer Advances. For more podcast episodes, visit our website, clevelandclinic.org/canceradvancespodcast. Subscribe on Apple Podcasts, Spotify, or wherever you listen to podcasts.

Thank you for listening. Please join us again soon.