NEW YORK - Treating patients who present with metastatic cancer that has no clear origin (carcinoma of unknown primary) has long been a challenge in oncology, and one for which genomic and other molecular technologies have provided new hope: offering tools to predict where cancers may have originated and how best to treat them in each individual patient. But recent evidence has suggested that CUP patients are failing to benefit significantly even when such tests are implemented.
The Phase III GEFCAPI study presented last month at the annual meeting of the European Society for Medical Oncology, for example, concluded that patients treated on the basis of molecular tumor origin, or primary site testing, did no better than those whose oncologists treated them without this added information.
Another study, published earlier this year in the Journal of Clinical Oncology by investigators from a group of Japanese hospitals and academic centers, also concluded that gene expression profiling to determine CUP origin "did not result in a significant improvement."
Results like these have crystalized the challenge facing genomic test makers that are working on newer, and what they hope can be more powerful, CUP solutions. Companies that have recently described new methods and technologies in this area include genomics testing firm Caris Life Sciences.
At this summer's American Society of Clinical Oncology meeting, investigators from the company presented a poster on what the firm calls the Caris Molecular Disease Classifier. Researchers described their use of a machine-learning approach to create an algorithm that classifies unknown cancers using data from the company's 592-gene next-generation sequencing panel.
Authors wrote that they were able to make correct classifications of tumor samples with 85 to 95 percent accuracy. For cases of CUP, the classifier generated an "unequivocal result" in the vast majority.
"It's the same exact test that we've been running for years to identify therapy, it's just we're now able to give additional information on the likelihood of the origin of the cancer. And we validated it on more than 15,000 patients," David Spetzler, Caris' president and chief scientific officer, said this week.
In another poster presented at the recent ASCO meeting, precision medicine firm Tempus posted data on CUP tumor origin prediction: in its case, a transcriptome-based probabilistic classifier, which the company trained on over 10,000 tissue samples.
Investigators reported that the accuracy of the classifier was about 85 percent, ranging up to 88 percent for primary tumors and 77 percent for metastatic lesions. The tool was best at identifying colorectal cancers, followed by breast and lung cancers. Even in low-quality metastatic tumor samples where surrounding tissue can obscure a cancer's transcriptional profile, the tool still achieved 71 percent accuracy, the authors wrote.
Interestingly, among takeaways from the negative results presented at ESMO last month, investigators hypothesized that leaving aside prediction of a cancer's origin in the body altogether and focusing instead on identifying genomic biomarkers that make patients eligible for specific targeted therapies could be the better solution.
One limitation of the two gene expression tests used in the GEFCAPI trial — Pathwork's Tissue of Origin test followed by Biotheranostics' CancerTYPE ID after Pathwork folded in 2013 — is they don't read out the tumor profiling information that is increasingly being embraced to guide therapy across a range of advanced solid tumors.
But commenters at the meeting pointed forward to the Roche-sponsored CUPISCO trial, which is recruiting now and will randomize CUP patients to either standard treatment or to treatment guided by results of Foundation Medicine's Foundation One tumor profiling test.
Foundation Medicine began looking at how molecularly guided treatment might benefit CUP patients back in 2015. The company published a paper in Cancer Research, which analyzed samples from 442 patients with CUP, and researchers concluded that more than half of the participants had mutations that could inform some sort of personalized treatment decision. At ESMO, investigators updated that calculation, reporting that in a retrospective study of a few hundred cases, about a third would have been eligible for treatment in one of the precision medicine arms of the new CUPISCO trial.
Caris' Spetzler agreed that the evidence seems to suggest that understanding a CUP's origin alone doesn't do much to improve treatment. "It's really a challenging space because the nature of a CUP, almost by definition, means that it's a really poorly differentiated cancer, and those are the types [of tumors] that do the very worst," he said. "Part of it is that we don't have very many effective therapies out there. In that context, knowing what it is doesn't necessarily help because there's just a lack of good drugs."
However, he added, what Caris believes based on its own research, is that molecular target information alone — which is what is being explored in the Roche CUPISCO trial — is also not enough. Rather, both are important in Spetzler's view.
"We've seen over the years that just looking at where the tumor is coming from is really not sufficient to optimize treatment strategies, but the reverse is also true," he said. "Blindly just looking at mutations doesn't optimize treatment selection either."
BRAF mutations are a commonly cited example. Although BRAF mutations occur in a subset of melanoma and colorectal cancer patients with poor prognosis, the treatment approaches in these two tumor types are very different, suggesting that histology should not be ignored when deciding on molecularly targeted treatment approaches.
With a combined approach, Spetzler said, "not only are we able to test for all of the underlying mutations and alterations that can identify appropriate therapy, but insights into [the type of cancer] … is also important information that will go into treatment decisions and allows for a higher degree of optimization."
Caris is conducting a prospective outcome study using a combined approach, with both molecular drug targets and its newly developed Molecular Disease Classifier being read out from its core NGS test.
"We've enrolled almost 5,000 patients to date … but the final [results] will probably take a couple of years, just to allow the clinical outcome data to mature," he said. In the meantime, the company also has retrospective outcome data for about 1,200 CUP patients tested using the platform and is working on an analysis of that data. "It's not prospective, but we'll get some insight into the [outcomes] differences there much sooner," he said.
Caris is not alone in seeking to combine both origin information and molecular profiling. Belgium-based OncoDNA, for example, offers a CUP testing service that combines NGS and IHC assays to determine origin and also mine for targetable genomic alterations.
A small prospective study of that assay published last year found that the test found actionable mutations and while the results aren't mature, researchers reported that results did lead to treatment changes in a proportion of patients, with one entering a clinical trial, two receiving immunotherapy, and the rest moving to a different chemotherapy based on origin prediction.
That said, no patients had yet been directed to approved targeted therapies at the time of publication, highlighting "current barriers to accessing these treatments [in] patients with limited options and bad prognosis," authors wrote.
Spetzler said that Caris believes it is unique in being able to perform both analyses using a single platform. "It's not an additional test that we are performing. Rather, it's additional information that we can extract from our [existing] profiling tests."
One other area of test development that might affect CUP moving forward is liquid biopsy, or blood-based cancer profiling. In a commentary accompanying the CUP study in JCO this January, authors from the UK's NHS Foundation Trust wrote that because repeat biopsies are often especially difficult in CUP, and patients are at such an advanced stage of disease, turning to liquid biopsy methods might be beneficial.
Researchers have also begun to explore this. For example, investigators from UC-San Diego have conducted studies using Guardant Health's Guardant360 liquid biopsy assay that are similar in design to Foundation Medicine's study.
In 2017, that group wrote in Cancer Research, that among 442 patients with CUP tested using Guardant360, more than half had at least one targetable alteration — meaning a variant associated with either an FDA-approved drug or with an investigational clinical trial.