Read the Full Video Transcript

Zachary Klaassen: Hi, my name is Zach Klaassen. I'm a urologic oncologist at the Georgia Cancer Center in Augusta, Georgia. I'm delighted to be joined today by Dr. Phuoc Tran, who is a radiation oncologist at the University of Maryland, and Dr. Tim Showalter, who is Chief Medical Officer for Artera. We're going to be discussing some ASTRO presentations featuring the ArteraAI prostate test. Gentlemen, thanks so much for joining us on UroToday.

Phuoc Tran: Thank you very much.

Zachary Klaassen: So Phuoc, I'm going to have you pull up the slides and run through the two presentations that were highlighted in the ArteraAI prostate test at ASTRO 2024.

Phuoc Tran: Of course. So we were very lucky to have two ORIOLE presentations accepted and that we were able to present at ASTRO 2024 in DC. The first was "Validation of a Digital Pathology-Based, Multimodal Artificial Intelligence Model in Oligometastatic Castration-Sensitive Prostate Cancer, Including Patients from STOMP and ORIOLE Phase II Randomized Clinical Trials." And this was presented by Dr. Philip Sutera. I have the pleasure of sharing with you folks.

So the background for this project was we wanted to look in a space that we are very interested in, that of oligometastatic castration-sensitive prostate cancer. And as many of you know, this is a space that has intense interest. Namely because we believe, at least as we have local therapists in this conversation right now with us, that local therapies may be able to play a role in "metastatic cancer." And depending on who you ask, the definition varies from three to five metastases. There's the CHAARTED, the STOMP definition of low-volume metastatic disease.

We don't have time to go into the finer details, but many of us believe that it's more about underlying biology. But nonetheless, just for operational purposes, we'll use three to five metastases. That is a space, at least in castration-sensitive prostate cancer, where we have emerging data that local therapies can actually change the natural history of the disease. However, what we also realize is that group of patients has highly heterogeneous disease biology in response to treatment. In particular, something known as metastasis-directed therapy.

To the right is just an example of that, a combination of meta-analysis of the STOMP and ORIOLE Phase II randomized trials demonstrating a benefit of metastasis-directed therapy for progression-free survival in these oligorecurrent castration-sensitive prostate cancer patients. We wanted to validate, explore, and validate the potential use of a digital pathology artificial intelligence multimodal biomarker developed by Artera, which many of you know was discovered, validated in localized prostate cancer and now recommended via NCCN for use in those patients. We called together a multi-international retrospective cohort of patients who had oligometastatic disease, again defined by five or fewer metastases. We worked with our colleagues at Artera to develop scores using their digital pathology MMAI algorithm. And then as our primary endpoint, we utilized overall survival stratified by median MMAI score.

In addition, we wanted to look at the utility of this MMAI biomarker as not only prognostic, but perhaps predictive in patients receiving metastasis-directed therapy because within this cohort of patients that we were able to pull together, we had both STOMP and ORIOLE Phase II randomized trials for this endpoint. Because overall survival is still a very rare endpoint, we decided to use something that is thought to be a very good surrogate for overall survival in this disease space, conventional metastasis-free survival. And what we wanted to test was whether this MMAI marker could be predictive for MDT use in patients with oligometastatic castration-sensitive prostate cancer, again using MMAI median scores as the stratification mark.

So what we found in the general cohort was that patients that had higher MMAI scores had higher PSA, higher Gleason, and higher proportion of patients who had the disease presentation known as synchronous metastasis, which is, as some folks know, we believe have a different biology and more aggressive disease natural history.

So the first result was that the MMAI score or biomarker was prognostic for overall survival in metastatic patients, metastatic castration-sensitive prostate cancer patients, in particular, these oligometastatic patients as shown here in this Kaplan-Meier curve. We then next proceeded to examine the possible predictive ability of this MMAI score. And what we found using just the subset of patients from the STOMP and ORIOLE trial, so small cohort of 51 patients, but nonetheless prospectively randomized between observation and MDT, we found that patients who had a high MMAI score seemed to derive a benefit as shown in the Kaplan-Meier curve to the left. However, those with a lower MMAI score derived no benefit from MDT as shown in the Kaplan-Meier curve on the right.

And using a statistical test to examine the ability for the MMAI score to be predictive, known as an interaction statistic, you can see there that the P interaction is significant at 0.02, that the MMAI score may be able to predict patients who benefit the most from MDT, again in this case versus observation. So the summary from that first abstract is that the Artera MMAI score is not only prognostic for overall survival in these metastatic, specifically oligometastatic castration-sensitive prostate cancer patients, but from a subset of patients from the STOMP and ORIOLE trial, it seemed as though that the MMAI score actually may predict which patients benefit from metastasis-directed therapy.

I'll quickly move on to the second abstract, which was entitled "Digital Pathology Multimodal Artificial Intelligence Algorithm is Associated with Prometastatic Genomic Pathways in Oligometastatic Prostate Cancer." This was presented by Dr. Yong Song.

Just some background as a consort diagram showing again the cohort of patients that we utilized for the study. This was the same 221 patients from the prior abstract, so a multi-institutional, multi-international retrospective cohort of patients. In the original 221 patients of those, we had 168 patients, again, 130 who were metachronous, 35 who were synchronous who not only did we have clinical information, but those patients underwent germline and somatic profiling of their primary tumors and in some cases a small number of their metastatic lesions. Again, those patients' H&E slides are digitized and sent to Artera to develop their MMAI Artera score.

Farther down in the consort diagram, you can see the type of profiling, and it was a hodgepodge of standard-of-care genomic profiling from the major vendors. So just some technical details. DNA mutational analysis was utilized to commonly used databases to determine whether patients had mutations in genes of interest. Again, we used cutoffs of the MMAI score by median or quartile. Fisher's exact was used for statistical significance as the test. And then for transcriptional RNA work, we used two algorithms that are commonly used in the industry as listed there.

So first set of results using DNA mutational analysis, what we found was that the mutations in patient samples correlated with MMAI score. In this case, we looked at SPOP mutations. SPOP is a tumor suppressor that when mutated is known to confer or correlate strongly with good prognosis. And so keeping with the concept that the MMAI score from zero to one with one being the most aggressive, we found an inverse correlation.

So patients with SPOP mutations tended to have lower MMAI scores, which is consistent with what we know about the biology of SPOP mutation-containing prostate cancer patients. Another pathway that many folks have shown to be correlating not only with poor prognosis but the metastatic state in both castration-sensitive but those with castration-resistant prostate cancer is the DNA damage response pathway that is, as shown there, a number of genes are correlated with that pathway, ATM, BRCA2 being the two most prominent. And as you can see, patients with higher MMAI scores, at least in this way shown in the quartile, have higher numbers of mutations, at least from a frequency perspective. If we focus on the two most commonly mutated genes in this case, ATM and BRCA2, what we find is that patients have a statistically significant higher number of mutations in those two genes.

And that also correlates by Fisher's exact test with higher quartile MMAI score. And keeping with metastatic pathways, the WNT pathway, which is a developmental pathway also co-opted by cancer for metastatic reasons, mutations in that set of genes correlate also with higher MMAI scores shown here.

So to summarize all the past two abstracts, this newly minted digital pathology, multimodal artificial intelligence biomarker from Artera was again, prognostic in overall survival for oligometastatic castration-sensitive prostate cancer patients, perhaps predictive for metastasis-directed therapy benefit in these same patients. And then a host of DNA mutations correlating with metastatic pathways, in this case, the DNA damage response, WNT pathway, was also correlated with higher MMAI score as well as—I didn't get a chance to show it—but our transcriptional data suggested that a pathway associated with metastasis, in this case, the EMT or epithelial-mesenchymal transition pathway at the RNA transcriptional level also correlated with high MMAI scores.

So with that, I'd like to thank you for allowing us to present this data on behalf of the team involved, not only in our group but also from Artera. Thank you.

Zachary Klaassen: That's fantastic. So Phuoc, congratulations to your investigational team and the Artera team. It seems like every major conference, whether it's AUA, ASCO, ASTRO, we're seeing more and more data in different disease spaces from the ArteraAI process. That's very exciting and really just continuing to push the boundaries in how we're stratifying patients. And so just by way of discussion, but I'd like to just focus on that STOMP and ORIOLE analysis. And it's really interesting if you think about this from a high level, we're really getting close to precision oncology for these patients. We scan them with PSMA PET, we're able to identify those lesions. We know that if we treat all those lesions, they're going to do better if we treat none of them or some of them. And now we have a biomarker in the MMAI model that says these patients are going to potentially do really well. How do we translate this? And how close are we to getting precision oncology in this disease space?

Phuoc Tran: Yeah, it's obviously a question that's burning, I think, in many people's minds, and they want the answer to that. I would say it's a little bit premature to use these data just as standard of care, obviously it was a very small cohort. But absolutely the concept of precision medicine and the overall concept of risk stratification in metastatic patients, and obviously all of the metastatic patients, is one that many of us, particularly our group, are interested in. And I think we are very close, but obviously this same type of work needs to be validated in larger cohorts, ideally integral biomarker-designed prospective trials. I'm not sure if we absolutely need that to put the last nail in the coffin, but I think we're very close. But I would caution that this is still relatively early work, and we need to see it validated at least a few more times before we can implement it in the clinic.

Zachary Klaassen: And even probably, I guess, arguably earlier in the pipeline of using in the clinic is this very elegant work on the metastatic genomic pathways, which is really cool work. Looking at alterations we know that are associated with high risk or even lower risk protective with SPOP, but taking again this biomarker and then further stratifying these patients, how do you see this translating to the clinic at some point? What's the next steps for this work?

Phuoc Tran: I think there's a few different ways it can help us in the clinic. I think the first is, and this kind of gets into the weeds of how the MMAI, the Artera MMAI was developed, is that it used, at least in some aspects of its development, what are known as unsupervised approaches. And when you have an unsupervised algorithm, many times the features, what the AI is seeing, can be what is known as non-human interpretable. And just clinicians, I think all of us are just—we're innately intuitive in how we do things, and if we can't understand how something works, I think it's much less likely that we're going to utilize it.

And I think the most important part of this work that we just talked about was that we were able to pin some of these MMAI features from the Artera score to concrete biological pathways that make sense to us as clinicians. So I think that's the first point that I think would help is that it might address some of that lingering hesitancy to use a test where it's like, well, how is it even picking up? What is it even picking up?

I think from a biologist's perspective, the next point, and from a therapist's perspective, the next point I think is really interesting is that it is up potentially novel features that may lead us to new biology, one. And then two, from a therapist's standpoint, it may lead us to new targets. So open up even new avenues to improve the outcomes of our patients.

Zachary Klaassen: Yeah, that's a great answer. Tim, I'd love for you to give us a couple of take-home points maybe from these studies, but also in terms of the ArteraAI prostate test in general. It's been a great few months of seeing new data.

Tim Showalter: Yeah, I will add that I was particularly excited by both of these abstracts because I think they fill really important needs for the evidence foundation for how to incorporate the MMAI test into practice. One of which I think, as we pointed out, is that in the first abstract, Phuoc and his team were able to demonstrate that the Artera test is validated in an oligometastatic cohort. Previously, we and our collaborators have shown that the test is validated for prognostic performance in localized as well as metastatic prostate cancer and even castrate-resistant non-metastatic prostate cancer. And this fills a really important area currently to demonstrate that even in the oligometastatic setting, that prognostic performance carries through. And I think it really completes that picture for confirming that this is a clinically relevant tool. And the second piece, as Phuoc alluded to, is the real clear need for AI explainability.

I mean, we get asked all the time, how can we be sure that the test works as well as it appears to, and I think that both of these abstracts speak to that. On the one hand, we've shown yet another data set that the prognostic performance is compelling, there's a clinical use for it, and then the other, there's a tight connection with biological underpinnings of what really drives cancer behavior and likelihood of benefiting from therapy. I think on both of those fronts, really excited to see the mounting evidence applied in this situation. And I think that this is important work that should be highlighted.

Zachary Klaassen: Absolutely. It's been a great discussion. Congratulations to both groups from the ArteraAI group and as well as Phuoc and your colleagues as well. We'll look forward to more exciting Artera data coming down the pipeline, I'm sure. So thank you both for your time and your expertise on UroToday.

Phuoc Tran: Thank you.

Tim Showalter: Thank you.