NEW YORK – A group of researchers at Memorial Sloan Kettering Cancer Center (MSK) and Epic Sciences has identified a chromosomal instability biomarker in circulating tumor cells (CTCs) that it believes can potentially help guide treatment for metastatic castration-resistant prostate cancer (mCRPC).
San Diego-based Epic Sciences may also pursue the development of a multi-marker panel integrating the new biomarker along with others for predicting prostate cancer treatment response, depending on the results of future validation studies.
The firm identifies CTCs by immunofluorescent staining for CD45, DAPI, and other biomarkers against a large background of normal cells. Researchers can then choose individual cells, pick them out of the slide, and use them in downstream analyses. The platform's workflow has a turnaround time of two to three days.
MSK oncologist Howard Scher said that his team originally applied Epic Sciences' CTC platform to develop an assay that detects the presence of androgen receptor splice variant 7 (AR-V7) in patient CTCs. The AR-V7 test helps clinicians decide whether to use taxane or androgen receptor signaling inhibitor (ARSi) therapy in patients with progressing mCRPC who were going to begin second- or third-line therapy.
In the new study, published last month in Cancer Research, the MSK and Epic researchers established another biomarker by building and testing an image-based algorithm on 10,240 CTCs in blood samples from 391 patients with progressing mCRPC prior to standard-of-care therapy. Patients were assigned to either training, analytical validation, or clinical validation cohorts.
"The goal is to see if you can take morphological features that will help you define the cell's genetics," Scher said. "In this case, we were [searching for] the feature of cells that have chromosome instability, which is usually associated with aggressive cancer and can have varied responses to different drugs."
Joseph Schonhoft, manager of translational research at Epic Sciences and first author on the paper, explained that chromosomal instability increases a tumor cell's ability to acquire chromosomal mutations, which tumor cells use to evolve and resist current cancer therapies.
Schonhoft and his colleagues initially studied the relationship between the number of large-scale transitions (LSTs) — genomic alterations based on chromosomal breakages in a CTC established by direct sequencing — and the cell's morphologic features.
The researchers used the training cohort of 26 patients to develop the phenotypic algorithm (pLST) to predict the actual number of LSTs in a CTC. Collecting 768 CTCs representing a range of sizes, shapes, and protein expression levels of cells and the amount of androgen receptor (AR) and cytokeratin in each cell, the team performed DNA whole-genome amplification and low-pass sequencing to calculate the number of LSTs within each cell. The group saw a success rate of 79 percent (608 cells), based on the number of contiguous regions of chromosomal breakage of at least 10 Mb.
Scher's team then used the imaging features to train the pLST algorithm to predict the number of LSTs identified by single-cell genomic sequencing. The group found a bimodal distribution of LSTs containing an LST-low group (eight or fewer LSTs) and an LST-high group (nine or more LSTs).
Identifying significant correlations between the CTC image features — such as AR intensity, CK intensity, nuclear entropy, and nuclear speckles — and the number of LSTs identified by single-cell whole-genome sequencing, the group used the relationship to build a computer vision algorithm to predict the number of LSTs in a single cell.
When applying a cutoff of nine or more phenotypic features to define an LST-high biomarker, the team saw that the algorithm had a single-cell clinical sensitivity of 68 percent and specificity of 84 percent.
"One of the really striking features is that when we looked down at the genomic level, all these CTCs had the [chromosomal instability] phenotype, but some of them were [also] driven by canonical drivers like PTEN loss as well," Schonhoft said.
Schonhoft and his colleagues then applied the pLST algorithm to an analytical validation cohort of 54 blood samples from 46 patients to separately assess the performance of the pLST's predictions compared to the number of LSTs determined by sequencing. The group found that the algorithm had a clinical sensitivity of 66 percent and a specificity of 74 percent.
While noting that the morphological features alone performed reasonably well, Scher said that his team also needed to establish a pLST biomarker-positive or pLST biomarker-negative cutoff based on the CTC number, as well.
Using a cutoff of three or more LST-high CTCs per mL of blood, the researchers found that the clinical sensitivity was 83 percent and specificity was 100 percent at the patient sample level.
Scher's team then applied the pLST algorithm in a clinical validation cohort to establish the link between the analytically validated pLST-based biomarker to a prostate cancer patient's overall survival. Collecting CTCs in 367 samples from 294 mCRPC patients, the team saw that about 22 percent of patients were positive for the chromosomal instability biomarker by applying the algorithm on their samples.
The researchers also found that the biomarker in CTCs in a pretreatment sample was strongly linked to poor overall survival in patients treated with ARSi and taxane therapies.
"We also had to show that the results not only showed information about prognosis, but that it was reproducibly measured and that it provided unique data outside of clinical routine testing," Scher said.
However, he acknowledged that his team encountered multiple issues while identifying the biomarker in the study. The team had to initially develop software that would record the different phenotypic features so that it could reproducibly measure them in additional samples.
Scher said that the biggest challenge the researchers dealt with involved determining how to optimize the phenotypic biomarker best to establish the binary CTC and LST cutoff to guide treatment decisions.
"We saw that beyond three cells … and nine LSTs, we weren't getting much more performance," Scher said. "While you'd like them to be as different as possible, the cutoffs weren't perfect, but we looked at where they would be optimal."
Scher and his colleagues now plan to test the chromosomal instability biomarker by using the algorithm in a larger prostate cancer patient cohort to see if it can also serve as a predictive biomarker for establishing personalized treatment recommendations.
"We didn't get through the full prospective validation part for prediction in this study," Scher noted. "We'll be looking to see where [the biomarker] adds value in the sense of using, for example, different types of chemotherapy, platinum, or AR-V7 inhibitors."
Ajjai Alva, an associate professor of internal medicine in the hematology and oncology division at the University of Michigan, believes the study is a "step in the right direction" to develop biomarkers of patient response and drug resistance. Like Scher, he argued that the team will need to perform further validation work on the algorithm before the method is ready for clinical use.
"That will require more steps, but it is encouraging that the platform was able to be used to test for [chromosomal instability] markers," said Alva, who was not associated with the study. "I don't know if it's part of the eventual solution, but it's one more part of the profile for a comprehensive, 360-degree view to understand prostate cancer drug response and resistance."
Noting that chromosomal instability occurs in most cancer types, Scher believes that the pLST algorithm could potentially serve as a tool for prognostic purposes in other cancers. However, research groups will need to perform validation testing to establish the required number of CTCs and LSTs to identify the specific cancer in a blood sample.
"The main question is [if the biomarker] can be reproduced by another group to make sure that the data is not only possible at a sophisticated center like MSK, but elsewhere in the field as well," Alva said. "[You] may need independent groups or a larger consortium to test the method, then possibly prospective trials, where you have randomized designs with markers driving selection of therapy versus clinical standard selection of therapy."
Epic Sciences is also interested in partnering with Scher's team to determine whether the biomarker and other CTC-based biomarkers can be predictive of patient response to therapies targeting genomic instability. While the group is currently completing initial regulatory protocols for the prospective study, Schonhoft expects to begin collecting samples at the end of the year or by early 2021.
Depending on the results of the validation and prospective studies, Schonhoft said, Epic will potentially pursue the development of an integrated assay based off the chromosomal instability biomarker with its CTC detection platform for drug prediction. However, he noted that the chromosomal instability phenotype is one of several candidate biomarkers the team is considering for a larger multi-marker panel for prostate cancer treatment prediction.
Because late-stage mCRPC often has multiple drivers, Schonhoft believes that deconvoluting cellular heterogeneity is crucial for establishing a multi-biomarker panel for the disease.
"We will look at protein, genomic, and morphologic markers and use the concept of cellular heterogeneity to see which cellular phenotype disappears during therapy," Rick Wenstrup, CMO of Epic Sciences, explained. "We'll have to see which markers end up winning, whether it's pLST, a neuroendocrine inhibitor we've been monitoring, or an extremely promising immunofluorescence marker, especially for PARP inhibitors."
Alva said that for clinical use, researchers will need to establish a prostate cancer biomarker with high negative predictive value, especially in treatment selection.
"It's not surprising that we're being too simplistic in expecting one biomarker to yield a complete picture of the [prostate cancer] patient's tumor in every scenario," Alva said. "I think we are still in the discovery phase, and we might not be ready for any one marker to be used in prostate cancer to capture every aspect of the tumor."
Wenstrup said that the final clinical version of the assay will potentially be an "algorithmically-derived solution" for physicians treating prostate cancer patients, derived from the information collected from future studies. Depending on the clinical utility, he believes the firm would likely develop the assay for research-use-only purposes. However, he noted that Epic Sciences may also choose to develop a companion diagnostic with one of its pharma partners, pending the results of the validation studies.