Read the Full Video Transcript

Andrea Miyahira: Hi, everyone. I'm Andrea Miyahira at the Prostate Cancer Foundation. With me is Dr. Jeff Tosoian, an assistant professor at Vanderbilt University. He will discuss his recent paper, "Development and Validation of an 18-Gene Urine Test for Clinically Significant Prostate Cancer," published in JAMA Oncology. Dr. Tosoian, thank you for joining us today.

Jeffrey Tosoian: Thank you, Dr. Miyahira, and thank you to UroToday for having me. This work was, of course, only possible through wide collaboration within and across institutions and the kind and generous support of the PCF and the NCI Early Detection Research Network. Our study was focused on the diagnosis of prostate cancer, a clinical domain that has undergone quite an evolution over the past 5 to 10 years and continues to evolve. My only disclosure is that I was a co-founder and serve as a scientific advisor to Lynx Dx, which commercializes and provides cancer biomarker testing.

And so currently, serum PSA remains the first-line screening test to identify patients at increased risk of prostate cancer. Yet we know the limitations of PSA as a screening test. PSA is expressed by prostate epithelial cells, not cancer cells, meaning it lacks specificity for cancer and can be elevated due to a number of benign conditions. The clinical impact of this is apparent, looking at data from screening trials under the traditional diagnostic pathway where an elevated PSA alone triggered prostate biopsy. Under that approach, 76% of biopsies proved to be negative for cancer. And of the positive biopsies, half were positive for indolent, clinically insignificant disease.

In response to that, guidelines now suggest clinicians use additional tools in men with an elevated PSA to better define the probability of significant cancer. These tools include both MRI and blood or urine-based biomarkers. The biomarkers currently offered by NCCN guidelines are listed here on the slide. And by virtue of being cancer-specific, these biomarkers have consistently demonstrated improved accuracy relative to PSA. Clinically, the use of biomarkers has been shown to reduce the proportion of patients undergoing a biopsy and reduce the diagnosis of indolent cancers while maintaining high sensitivity for the clinically significant, higher-grade cancers we seek to detect and treat.

And while the biomarker components of available tests have improved upon PSA, the majority have not been shown to be truly specific for higher-grade, clinically significant cancers that are the focus of early-detection efforts. Meaning that on a gram-per-gram or potentially cell-for-cell level, these genes are not overexpressed by high-grade cancers relative to low-grade cancers. So we hypothesized that detection of higher-grade cancers would be improved through two mechanisms. First, and most uniquely, through the addition of novel biomarkers, specifically overexpressed by high-grade cancers. And second, through an ability to measure several fold more cancer-specific markers than the three to four biomarkers included in current assays.

And so we took a three-step approach to the work with the first being discovery, with the need to identify biomarkers that are uniquely associated with high-grade cancers. To do this, we quantified the expression of over 58,000 transcripts in tissue-based RNA-Seq data from 775 patients, and we used specific criteria to nominate genes that were either high-grade specific, meaning expression in high-grade cancer was significantly different from low-grade cancer, or more standard cancer-specific markers where expression in cancer, overall, differed from benign tissue.

Here are just several of the genes we found to be significantly overexpressed in higher-grade cancers. Those are shown in red, relative to their expression in low-grade cancers shown in green, with adjusted P values demonstrating statistical significance. And so ultimately, the discovery analysis identified 22 high-grade specific and 50 cancer-specific genes. But you'll recall that these findings were in tissue and our goal was to develop a non-invasive test that would preclude the need for tissue sampling through biopsy. So we set out to identify a urinary platform for detection and we selected the OpenArray multiplex qPCR platform as one that would allow for high throughput of gene expression. And these figures just show some early proof of concept data in which measurement of the high-grade marker, SChLAP1, was successfully demonstrated in clinical urinary specimens.

Credit for establishing an efficient and reliable workflow for the urinary assay goes to Dr. Lanbo Xiao at the University of Michigan, one of our co-investigators and lead authors. The markers that were unable to be measured by PCR, and those that were highly collinear, were removed and the resulting 44-gene candidate panel was supplemented with 10 additional cancer-specific markers identified in the literature. This yielded a 54-gene candidate panel for urinary testing and we proceeded to the development of the optimal testing model.

The development cohort included men presenting for prostate biopsy at the University of Michigan from 2008 through 2020. Prior to development, we considered the fact that prostate volume is known to provide meaningful information, but that clinically, it is only available in patients who have undergone a previous biopsy or MRI. So we developed two models in parallel, one including prostate volume for its use when available and one not including prostate volume. These were called MiPS2 and MiPS2+, or MPS 2+, which included prostate volume.

Model development was led by Dr. Yuping Zhang. The development cohort was split into four samples and this was repeated 10 times. Elastic-net regression models were fit to each of the 40 re-samplings. We were then able to tabulate the frequency with which each marker contributed to the optimal model across those 40 re-samplings. Here are the 17 genes that most consistently contributed to model accuracy. And these were included in the final models along with the reference gene KLK3.

In internal cross-validation, the novel models demonstrated an AUC of 80 to 82%. And as shown here, outperformed PSA, the PCPT Risk Calculator, and the original three-gene MPS test. The models were calibrated and locked for external validation. External validation was performed in a blinded manner, and the locked models were provided to investigators at the EDRN to carry out that assessment.

The cohort similarly included men referred for biomarker testing and prostate biopsy as part of a national EDRN trial. Uniquely, this cohort had data available for the Prostate Health Index and the original MPS test, which are two of the current guideline-endorsed biomarkers, allowing for direct comparison of the MPS2 models with the current standards of care. We also assessed the performance of two-and-three-gene models, including the components of two other commercially available urine tests. These were not the specific models, nor the platforms used by those tests and should not be interpreted to represent them, but simply was the optimal modeling of those component markers in this dataset.

For clinically significant cancer, the new models demonstrated a higher AUC than any of the existing tests with AUC values of 81 and 82% across the MPS2 models. And while AUC gives a general indication of accuracy, clinical consequences of testing are most critical. And so, as we and other authors have done previously, we assess test performance at the clinically useful threshold providing 95% sensitivity for clinically significant cancer. In the initial biopsy population, at the sensitivity of 95%, we found that the MPS2 models improve specificity by 5 to 15% relative to PHI and MPS. That translates clinically to an increased proportion of unnecessary biopsies that are avoided. In men undergoing repeat biopsy, the improvement offered by the MPS models was really quite striking, allowing for an avoidance of roughly one-half of unnecessary biopsies.

And this is biologically quite plausible because we know that the previous tests include PSA as one of the few component markers in those tests. And in the repeat-biopsy population, we know that PSA may not be the best marker of cancer. These patients have already undergone a biopsy due to PSA which was negative. And so it makes sense that with the use of 17 non-PSA markers in the new tests, the impact of the novel test would be particularly large in this population. And that is, in fact, what the validation showed. Importantly, the sensitivity and negative predictive value were also 99% for grade-group three and above cancers. This tells us that the rare false-negative test result of the MPS2 tests were almost uniformly more favorable grade-group two cancers that are least likely to cause harm.

Finally, we used decision curve analysis to demonstrate the clinical improvements offered by the MPS2 tests across the entire range of clinically relevant risk thresholds. The plot here shows the reduction in biopsies achieved through testing per 100 patients without missing a single diagnosis of clinically significant disease. And as shown in the figure, the MPS2 models offered the greatest net reduction in biopsies across all risk probabilities relative to the other tests.

And so, in conclusion, we added novel biomarkers for high-grade prostate cancer to a panel of cancer-associated markers to yield an 18-marker urinary panel which we've called MPS2. We found that at 95% sensitivity for clinically significant disease, MPS2 provided clinically significant improvements in unnecessary biopsies avoided in both the initial and repeat biopsy settings. And so, we believe these data suggest that MPS2 can provide meaningful improvement in the detection of clinically significant disease relative to currently available diagnostic testing options.

And so it's important to acknowledge the many people that contributed to this work, of course my collaborator and mentor on my PCF award, Dr. Arul Chinnaiyan, and many of our colleagues spanning several institutions, as well as our colleagues affiliated with the Early Detection Research Network. And so with that, I thank you very much for your time and am happy to take any questions.

Andrea Miyahira: Thank you so much, Dr. Tosoian. So, how do you see this being used in clinical practice? What would be the steps toward biopsy decision?

Jeffrey Tosoian: Absolutely. So, PSA has a first line, widely available and very inexpensive test to identify the majority of patients at risk. The question becomes, what then is the next step? And current guidelines allow for the use of either MRI or biomarkers. My opinion would be that I think MRI is an excellent tool, particularly at large institutions, academic institutions where specialists are able to review those scans and interpret them with a high degree of accuracy and reliability. However, in the population at large, I do think there is likely a role for biomarkers as that first line test with such high sensitivity and negative predictive value, to be able to really definitively rule out the need for additional testing in a proportion of patients.

The other part of that is that biomarkers are, of course, objectively measured and so are not dependent on interpretation by an expert radiologist and can be obtained in a matter of minutes in the standard clinical setting. I would then think that patients with a positive biomarker test would benefit from an MRI to then identify areas for additional sampling at the time of biopsy, the role for which MRI has been very well validated.

Andrea Miyahira: Thank you. So, do you see this as something that could be used in active surveillance settings?

Jeffrey Tosoian: Absolutely, and that's some of the work that is underway right now. We are looking at the entire 54-gene candidate panel in an active-surveillance population with the thought being that patients on active surveillance, we know, have a diagnosis of prostate cancer. And so it's quite possible that the cancer-specific markers are less informative than the newer high-grade cancer-specific markers. And that's something supported by the current literature where really both, blood, urine, and tissue-based biomarkers have not had great prognostic value, predictive value in active surveillance.

And so we are going to use the entire 54-gene candidate panel to explore each of those markers and potentially redefine a model for the active surveillance setting that can be used to ideally monitor patients in a more non-invasive manner so MRI and biopsies can be spread out even further during the course of monitoring.

Andrea Miyahira: Thank you. And what is the clinical development plan for this assay, and when might it be available for patients?

Jeffrey Tosoian: Yeah, so the test is actually available currently. And with the paper being published in April, the test and data will be submitted to the guideline panels, hopefully for inclusion in guidelines as part of the standard of care. And with that comes discussion with Medicare and payers regarding reimbursement. But the goal would be for it to be widely available to all patients and reimbursed by all insurances.

Andrea Miyahira: Okay, thank you. And thanks so much for coming on and sharing this with us today.

Jeffrey Tosoian: Thanks so much for having me. I really appreciate it.