Precision Drug Combinations for Advanced Prostate Cancer Treatment

April 20, 2022 Brad Power

Ally Perlina, Chief Science Officer at CureMatch, presented combinations of approved drugs that are the best fit for advanced prostate cancer patient Brian McCloskey based on his cancer’s unique molecular profile. CureMatch is a San Diego-based company which takes biomarkers identified by most any diagnostic sequencing test and matches them with approved drugs in combinations.

Their distinctive value is using multiple drugs in combination to achieve a better fit with the unique characteristics of a patient’s cancer, and therefore better patient outcomes. The CureMatch system starts with input from a patient’s molecular profile, derived from DNA and/or RNA sequencing of a patient’s tumor tissue or liquid biopsy sample.

The test will identify biomarkers or variants of interest (variants from normal cells, potential drivers of the cancer), usually 4 to 6. Not all biomarkers or variants go into the algorithm. Some may be rejected because they are not pathogenic (don’t drive the disease), and some because there are no therapies targeting that biomarker.

The physician has latitude to include or exclude biomarkers or treatment options due to any reason, such as medical history. For example, Brian had pembrolizumab, so he might decide to remove it as a treatment option. Then CureMatch analyzes the patient’s selected biomarkers against the roughly 300 drugs that the FDA has approved in combinations of 3, 2, or 1 drugs. They do not include clinical trials in their recommendations.

For example, Brian has a variant (B7-H3/CD276) which has several clinical trials targeting this pathway, but it would not be included in a CureMatch recommendation because there are no approved drugs for it yet. Off-label uses of drugs (drugs that have been approved but for a different indication) are included in the treatment combinations.

5 million possible combinations of the roughly 300 approved drugs are then scored on the extent to which they address the patient’s selected biomarkers. If a drug combination addresses 4 of 6 biomarkers, the score is 67%, 3 of 6 would be 50%, and 2 of 6 would be 33%. The drug combinations are ranked on their scores, with explanations and links to support the choices.

For example, Brian had 6 actionable markers, 16 on compendia drugs, 54 matching drugs, and 24,857 relevant combinations. CureMatch’s top option, with a score of 32%, was a 3-drug combination of apalutamide (FDA approved, AR target), olaparib (FDA approved, FANCA target via PARP-1, PARP-2), and trametinib (off-label, BRAF target via MAP2K1, MAP2K2, and MAP2K2 target).

As Emma Shtivelman, an experienced PhD molecular biologist and Chief Scientist at Cancer Commons summed up, “... someone like me looks more for clinical trials rather than off-label treatment options. And when I look for clinical trials, I keep in mind the previous treatments... CureMatch has this assumption that a physician will be able to prescribe off-label combinations of two or three drugs.

This happens rarely, even with the best specification. I know two patients who had recommendations from CureMatch and their physicians worked with the recommendations, and it worked great. But it's an exception... ” Advanced Prostate Cancer Patient Panel Report Brian McCloskey shared highlights from a meeting of advanced prostate cancer patients who have registered with the Prostate Cancer Lab community.

CureMatch Recommendations for Treatment Combinations for Brian McCloskey Brad Power: To set up Ally's CureMatch analysis, we're going to have Brian briefly set the stage with his medical history. Brian McCloskey: I was diagnosed in 2016. I had a prostatectomy. Unfortunately, it came back, and then I went on this wild ride. I went on first line hormone therapy, then Lupron, then radiation, and then we saw an uptick in my PSA.

Then I went to a second line hormone therapy with apalutamide, and saw an immediate response to that, which was great. I took a bit of a holiday in December of 2019, you can see at step 5 there. Unfortunately, I had a biochemical recurrence at that juncture, but it became visible as soon as I went off of all hormone therapy. In August of 2020, point 6 on this graph, we discovered I had six metastatic lesions in my peritoneum.

We did surgery, and then I started a combination of chemo and pembro. You can see my PSA began to drop during all of that. I took six rounds of chemo and stayed on pembro for about 12 months. I began to see a bit of an uptick in September of last year. We did some more imaging and found I had 3 metastatic lesions in the peritoneum. Prior to that there were no visible mets after the surgery and after chemo. I am now on abiraterone. 45.

Saed Sayad: What was the Gleason score? Brian McCloskey: It was a 4 + 3, with grade 5. Ally Perlina: Thanks for having me. I hope what I have will be helpful. I spoke to Brian last year. We met virtually. I know there is a history of his getting a different CureMatch report in 2017. It was different data then, and there was different knowledge out there in the world.

You can think of CureMatch as knowledge representation and reasoning, as opposed to a machine learning kind of AI. We are not trying to make a prediction. It's more like scoring molecular fitness of treatments (therapies, drugs) to the markers of molecular type in each individual patient profile. As our founder and chief medical advisor Dr. Razelle Kurzrock said, it's like a snowflake, everybody's cancer is unique.

The idea of her CureMatch brainchild that was born in 2015 was to be able to emulate molecular tumor board reasoning, with all the knowledge at hand, so that for any unique new case, even if some mutations have never been seen before “Drug Combinations Promise Precision” (Ally Perlina, CureM in the arrangement that they align in a give patient, no matter how unique a case is, if there is a way to therapeutically address or drug the target, then the match will be considered by our system.

I'm going to tell you how it fits in the therapy-matching space, and then go to the reports. We don't attempt to call variants or markers as pathogenic or not. We are not trying to assign the initial clinical interpretation and variant curation of significance the way that labs do. We go next in line after an NGS report, and also consider molecular data on a proteomic or gene expression level from other types of reports.

Unlike some companies that do a beautiful job of showing what are the known therapies associated with actionable markers, we also cover novel combinations. We work with the over 300 or so available cancer drugs to consider the millions of ways to mix and match them into customized two-drug or three-drug combinations, which is what molecular tumor boards often recommend.

They take drugs that are available and combine them in ways that have possibly not been tested in clinical trials or approved. But the drugs themselves are available. The novel combinations cover about 99% of the top molecular matches for each patient. Brad Power: Can you please define some terms? NGS is for Next Generation Sequencing. They are typically looking at an oncopanel of about how many genes?

Ally Perlina: We are not limited by the panel or number of genes that the lab looks at. Larger panels from CLIA-certified labs may be better, but we will take any sample type, liquid biopsy or solid tissue, any company or lab that does NGS analysis, whether they interpret the results or therapies or not. We will take the molecular profile markers that are discovered, from DNA level or RNA level tests, and and will CureMatch that.

Brad Power: To get a sense of scale, one of these companies, like Foundation Medicine, looks at 200 or 300 genes, and then identifies variants in a half dozen or dozen markers that might be useful? Ally Perlina: Usually what we see are 5 or 6 markers on average. When patients turn to CureMatch is when the decision is not trivial to choose from among millions of possible combinations to come up with the best combination of two to three drugs.

Often this is too late. We hope that NGS will be ordered sooner and more frequently, and CureMatch will become the standard of care. We get quite complex cases. We see anywhere between 2 to 15 or even 20 markers coming out of the tests, particularly if you have RNA expression data, which adds markers. Rick Stanton: On the slide where you have "known therapies less than 1% of matches", let me say what I think that might mean.

If I have a mutation that has a known therapy that is directly targetable, that is kind of rare. And for the other 99%, you need the CureMatch system to target it. Ally Perlina: It's not just mutation to drug matches that are one-to-one.

What this is saying is that out of all the reports we have done, we have seen one therapy address the overall molecular profile on the top (with the best score) 1% of the time, and we have seen a combination address all of the pathogenic markers 99% of the time. You need to take the drugs out of the 300 and mix and match them in customized, individualized ways that are molecularly precise and justifiable.

Jeff Waldron: Does CureMatch incorporate drugs that are investigational, under clinical trials or available to patients through compassionate use or expanded access? Ally Perlina: This came up in my correspondence with Brian. We just work with available drugs. Investigational drugs are not included in our system. If it is approved for some indication, even a non-cancer indication, then it is included.

Our clients, doctors of patients, have not been turning to us for clinical trial matching for investigational drugs. When they turn to us, they know that they are going to be trying something different and the case is very complex and often advanced, and they don't know if the patient will qualify for a clinical trial.

We take care of what we do uniquely, which is taking care of the complexity of choosing from millions of possible combinations, then filtering and scoring for targeting the overall profile. We will expand from there. We are focused, and we are a small startup. Glenn Sabin: You are also looking at non-cancer agents that are also FDA-approved? Ally Perlina: There are relatively fewer of those, but they are included.

John Laird: When you say that you identified 5 or 6 pathogenic markers, what does that mean? Not every mutation is pathogenic. To be chasing mutations that are irrelevant to the progression of the disease would be a concern. Ally Perlina: Sometimes we don't have enough markers to produce a report.

I participated in several of these hackathons, and at one stage, there weren't enough markers to run a report, and then later more markers were identified, and then there was. And there are pathogenic markers that are not actionable. In Brian's cas case there is a fusion that is not considered actionable. But there are other markers that are actionable.

This example is not Brian's report, but in this case, there were 5 actionable markers, but there were 6 or 7 pathogenic or clinically significant markers reported from the NGS results. Unless the clinical team overrides this and says that this or that is pathogenic, we will take whatever the lab report says, and we will CureMatch that. We don't reinterpret CLIA-certified results. Here are the report results for Brian.

The report gives you a score, the "PreciMatch Score". This score represents how well any given option is covering all of the pathogenic markers with the drugs that are included. We show top 3-drug and 2-drug combinations and monotherapies. Physicians like to see that, and sometimes they score quite well. When we have something that covers therapeutic markers of the profile of the patient 100%, then the score is 100.

If there is no molecular matching for the treatment, then the score will be zero. There has to be molecular justification or indication for a drug. With this molecular score, the higher the score, the better we have seen progression-free survival and overall survival outcomes, and the worse the score, the worse the outcomes. This CureMatching and scoring has merit.

We have been able to collect evidence that shows that our algorithm and knowledge base is predictive of outcomes. That is where we come in with intelligent therapeutic decision support. 5 million doesn't change, but the number of potentially relevant two- or three-drug combinations that the engine considers does change. Here there were 6 actionable markers. But on the second page, there are more than 6 markers.

For example, if somebody had 20 markers, but you could only hit 8 with a combination of 3 drugs, then you're not going to get a score of 100%, because you're at best hitting under 50% of the markers. How many birds can you shoot with three stones? These drug combinations shown do not address the marker that is the fusion because we're not counting it as an actionable marker. We do not consider it targetable.

All other markers from transcriptomic and genomic data were taken into account.