March 2009
Feature Story

Karen Titus

How important are origins? One need be neither Charles Darwin nor religious to find this question vexing. For pathologists and oncologists looking for the source of a tumor, the matter of origins can be just as fraught. Simply put, pathologists hate to issue a report of cancer of an unknown primary.

“Worst-case scenario, they give options: The metastasis may arise from the lung, breast, or colon,” says Catherine I. Dumur, PhD, assistant professor, Department of Pathology, Virginia Commonwealth University. Only when all options are exhausted will they report a cancer of unknown primary.

Creation stories, it seems, are a must. Recognizing this fundamental need for answers, several diagnostics companies are stepping into the breach, offering assays to help pinpoint carcinomas of unknown primary, or CUPs, although it may be more helpful to think of these assays succeeding in thy slightly different CUPs: tackling carcinomas of uncertain primary. By extension, they also contrive to answer questions about therapeutics. Because these assays are so new, it’s difficult to say with certainty how they’ll be used. But few doubt they’ll make their mark somehow.

Mat Moore, PhD, has watched the field evolve in the last seven or so years. Now vice president of research and development at Neo­Genomics Laboratories, Dr. Moore helped develop one of the earliest CUP assays, when he was a senior scientist at U.S. Labs; that company co-developed the assay with Arcturus Bioscience. (To trace this particular lineage, know that Arcturus begat bioTheranostics— which offers the Theros Cancer­Type ID assay—by way of bioMérieux, with a few name changes along the way.)

Dr. Moore says any discussion of these assays needs to start by acknowledging that CUPs are a clear diagnostic challenge, usually left untouched by traditional workups, and, without appropriate therapy, have a very poor survival rate. The next diagnostic steps are often launched by multivariate analysis, using a cumulative approach to look at hundreds or thousands of genes to determine which expression pattern best fits the tumor type to a database. “This seems to be the only practical way of getting some good information to this diagnostic dilemma,” Dr. Moore says. Two companies (Pathwork Diagnostics and bioTheranostics) use messenger RNA in their assays. A third company, Rosetta Genomics, uses micro­RNA. Each method, says Dr. Moore, has equal merit.

The tests

Rosetta Genomics has three ­micro­RNA tests for the clinical community, all with a miRview (pronounced MEER-view) prefix: miR­view mets, for identifying the tissue of origin of tumors; miRview squamous, for differentiating squamous from non-squamous non-small cell lung cancer; and miRview meso, for differentiating malignant pleural mesothelioma from adenocarcinomas in the pleural fluid.

Rosetta’s Dalia Cohen, PhD, chief scientific officer, and Ronen Tamir, chief commercialization officer, cite several reasons for using micro­RNAs, which are tiny (mature micro­RNAs are 22–25 nucleotides in length) molecules of noncoding, regulatory RNAs. Among other things, they’re thought to play key roles in differentiation during development and are involved in oncogenesis. “Micro­RNAs are called one of the major switches in the human body, because they regulate so many protein encoding genes,” Dr. Cohen says, explaining that they’re able to control gene expression after transcription occurs. They’re also remarkably tissue-specific, she adds. The company’s use of microRNAs to identify cancer tissue origin is discussed in a 2008 paper (Rosenfeld N, et al. Nature Biotechnology. 2008;26:462–469).

MicroRNAs also have the advantage of being well-preserved in paraffin, due to their stability, and extracting them from paraffin is relatively simple. The company developed protocols for extracting and processing microRNA.

The miRview mets test is performed in a CLIA-approved laboratory (at the company’s lab in Philadelphia), using formalin-fixed, paraffin-embedded samples. Currently, the test identifies 25 tumor origins by using only 48 micro­RNAs as markers.

Based on his own extensive work with micro­RNAs, Mark Bloomston, MD, assistant professor of surgery, Division of Surgical Oncology, The Ohio State University, considers them to be a robust target for diagnostics. He’s looked into use of ­micro­RNA expression patterns to differentiate pancreatic adenocarcinoma from normal pancreas and chronic pancreatitis (Bloom­ston M, et al. JAMA. 2007;297:1901–1908). Peering into the future, he notes that recent papers have begun identifying microRNA biomarkers in serum. “That’s pretty exciting to me, as a pancreas and liver surgeon,” says Dr. Bloomston. “Being able to find something that’s predictive of who’s going to have a cancer is even more important, because early detection is key for those diseases.”

Dr. Dumur is another ­micro­RNA fan and uses it in her research. “Oh, I think it’s a great approach,” she says. “There’s been an explosion in knowledge about micro­RNAs in the last two, three years. MicroRNAs are versatile, and efficient for controlling gene expression after transcription.”

Research done by Marina N. Nikiforova, MD, assistant professor and associate director, molecular anatomic pathology, University of Pittsburgh, focuses on micro­RNA markers for diagnosis and prognosis of thyroid cancer and other cancer types (Nikiforova MN, et al. J Clin Endocrinol Metab. 2008;93:1600–1608). “Only a small number of miRNAs are required to differentiate cancer from noncancerous tissues,” Dr. Nikiforova says. She predicts microRNAs will be a powerful tool for cancer diagnosis in preoperatively obtained fine-needle aspiration samples and will help identify more aggressive types of thyroid cancer.

BioTheranostics uses a more ­traditional approach (if this field could be said to have one) with its RT-PCR CUP assay. It measures 92 genes (87 content, five normalization) in a 384-well plate format, says Mark Erlander, PhD, chief science officer, and was based on mining a 22,000-gene array and profiling close to 600 tumors—primary and metastatic—using both frozen and FFPE samples. The test is offered at the company’s CLIA-approved laboratory, for FFPE core biopsy samples. It classified 32 different tumor classes in the validation set of 119 FFPE tumor samples with an 87 percent success rate (Ma X-J, et al. Arch Pathol Lab Med. 2006;130:465–473).

Those who set out to create a CUP assay soon find themselves awash in a sea of numbers. One reasonable starting point for handling tumor classification is to create a set of genes that are specific for each tumor type, “kind of like IHC,” says Dr. Erlander.

“The problem with that,” says Xiao-Jun Ma, PhD, vice president of R&D, “is you’re dealing with hundreds of different tumor types. That becomes a very big gene list,” one that would need to be revised if the tumor panel were to expand.

Instead of looking for tumor-specific markers, the company relied on combinations of genes that work together to classify tumors, Dr. Ma says. This explains the assay’s rather compact gene set, which enabled the test’s developers to use RT-PCR to improve robustness. This should also open the gate to panel expansion; for now, the test can classify 39 different tumor types and 64 subtypes, the company reports. “We think it’s important to classify even more than that,” says Dr. Erlander.

Pathwork Diagnostics is also using mRNA for its assays. Its Tissue of Origin test has been FDA approved for use on fresh-frozen tumor specimens and is available through the company’s CLIA la­boratory. More recently, the company developed a similar assay for use on FFPE samples, for which it intends to seek FDA approval, says Deborah Neff, president and CEO. “Absolutely. We’ll be filing this test with the FDA for 510(k) clearance and working with them closely through that process this year.”

Moving to an FFPE test wasn’t easy, says David Henner, MD, PhD, chief medical officer. The formalin fixation process breaks RNA into small fragments (unlike with fresh-frozen samples, from which large molecular weight RNA can be readily extracted) and chemically modifies it. But the information is retained. “It’s as if you had a long sequence of numbers, but broken into smaller pieces.” The challenge was to move the information onto the microarray chips, then put the fragments back together to obtain good gene expression. Dr. Henner reports the result is an assay that performs almost identically to the company’s fresh-frozen assay, but obviously with much simpler handling requirements.

Pathwork’s tumor panel has 15 different tissues of origin that represent 58 different tumor morphologies. This covers about 90 percent of metastatic tumors, Dr. Henner says. Although the company intends to expand its panel eventually, “We’re quite pleased with what we have now.” The company issues a similarity score for each of the 15 tumors on the panel, ranked from most likely match to least likely, with the total score adding up to 100. Dr. Henner reports that for the fresh-frozen assay, a match can be made 95 percent of the time between the unknown sample and one of the tissues in the bank. Just as important, he says, the test can rule out at least 12 tissues.

When standardizing the microarray data for the assay, the company used more than 5,000 tumor and tissue samples. For the classification algorithm, Pathwork used 2,000 data files. And the validation, Dr. Henner reports, involved 545 samples. The latter step relied mostly on difficult cases. “We tried, as much as possible, to include metastatic samples or poorly differentiated samples in our validation set,” he says. “We wanted to give the test the maximum stress.”

The users

Origins are never simple for long. Even the best creation stories can’t make up their minds, and quickly splinter into differing accounts, raising all sorts of questions. Why should medicine be any different? As they make their way into practice, the CUP assays are creating trails of anecdotes that may offer insight into their potential use.

One year into his use of the bioTheranostics assay, what does Bradford A. Tan, MD, have to say? “I’m not 100 percent sold on this assay,” says Dr. Tan, medical director, Cancer Treatment Centers of America at Midwestern Regional Medical Center, Zion, Ill.

He’s used it on 16 cases. In nine cases, he says, the assay’s results agreed with the pathologist’s impression and clinical findings.

The other seven cases “were all over the place and don’t make sense,” Dr. Tan says. In one case, there was disagreement over the report of a squamous cell carcinoma when in fact it was morphologically a metastatic adenocarcinoma. “Those things don’t help us,” Dr. Tan says.

How did Dr. Tan and his colleagues come to choose this particular test? “They were basically the only game in town at that time,” he says. With other choices now available, he’s looking at other assays.

For now, Dr. Tan plans to use the assay as a last resort. “The basic H&E still helps you a lot,” he says. Nonetheless, there’s no doubt in his mind that these assays represent the future of testing. “We will be going to the molecular level. It’s just a matter of time,” he says.

Pathologist Robert McGee, MD, PhD, has run a half dozen samples on the bioTheranostics assay, with five of the six confirming Dr. McGee’s suspicions. The sixth “came back with something that ­didn’t make clinical sense to us. We suspected a neuroendocrine tumor, and the result we got back was for a soft tissue tumor,” says Dr. McGee, of Randolph Hospital and Randolph Cancer Center, Asheboro, NC.

Dr. McGee chalks up such cases to the work-in-progress nature of the assays. “It may have been one of those cases where they don’t have enough data in their library for the particular tumor we were dealing with. So they gave us the best match they could at that point, without a lot of certainty.” If that was indeed the case, Dr. McGee says, such matters will improve as the company expands its library.

Even a handful of success stories are enough to cheer Patrick Acevedo, MD, attending physician, Florida Cancer Institute, New Hope Practice, Spring Hill, Fla. He first turned to the bioTheranostics test for a patient with a fair burden of disease from what easily could have been any kind of GI cancer. The test called it a renal cell carcinoma. “This was quite helpful from a medical oncologist’s standpoint, because of the number of target therapies that are now available for that.” The patient was treated, which kept his disease stable for a time before progression.

“This can really open up the armamentarium we can use,” says Dr. Acevedo. With CUPs, on the other hand, oncologists have fairly fixed single or dual chemotherapies, most of them antiquated, he says. He and his colleagues have used the test in 15 to 20 patients, he estimates. More than half the time, “We’re able to pick out regimens that would be more specific than had we not done the test.”

CUP tests give oncologists an important, though at times overlooked, benefit: If pathologists hate to write “carcinoma of unknown primary” on their reports, clinicians hate to convey that same news to patients. “It’s so difficult,” says Dr. Acevedo.

And in the best-case scenario, the individual impact can be stunning. BioTheranostics reports a case from a user in which the test correctly identified a germ cell seminoma, which had previously gone unidentified despite 20 immunostains (one of which was negative for seminoma). The patient, who had been in hospice, had complete resolution of disease on a PET-CT.

The evaluators

Since the assays are so new, among those with the most experience using them (outside of the companies) are pathologists who’ve been involved in validation and other studies.

Iris Schrijver, MD, falls into this category. Dr. Schrijver is director, molecular pathology laboratory, Stanford University Medical Center, Lucile Packard Children’s Hospital, and associate professor of pathology and (by courtesy) pediatrics, Stanford University School of Medicine Pathology Department. Stanford was one of four sites that participated in a Pathwork effort to study the assay performance and reproducibility in 60 samples of archived tissue specimens from poorly and undifferentiated tumors, both metastatic and primary (Dumur CI, et al. J Mol Diagn. 2008;10:67–77).

Dr. Schrijver found it intriguing to get an early look at what’s required to run the assay in a diagnostic setting. “It’s not trivial,” she says. Hands-on time is considerable, and total time to result is about 4 days, she reports. Even with frozen samples, Dr. Schrijver says, messenger RNA degrades fairly quickly; if a sample is not frozen right away, the assay could be compromised. “We’ve had a few samples fail,” she says.

The test also left her feeling a bit adrift in terms of specimen control. Performing the wet work in her lab, then sending the specimen to Pathwork for data analysis and a report, was an unusual departure. “It’s always a bit unsettling for a lab director to have direct responsibility but not full control over the testing that comes out of your laboratory,” she says, though she adds that she’s fairly confident in the version of the assay used on fresh-frozen specimens because of its FDA approval.

Dr. Schrijver found the reports helpful, with some caveats. Higher similarity scores give relatively high confidence that the site of origin has been correctly identified; low similarity scores help rule out certain tissues of origin. But the picture becomes more complicated, she says, when you remember the panels for these assays contain limited numbers of tumors—a panel that does not contain uterine cancer, for instance, may respond to uterine tumors with a similarity score for ovarian cancer. “You need to be aware that you might get a result despite the fact that the actual tissue of origin is not part of the panel,” Dr. Schrijver cautions. For the foreseeable future, these tests should be seen as one piece of the puzzle, she suggests.

Clearly, the nuts and bolts of validation should absorb any pathologist con­sidering a CUP assay. Dr. Dumur, of VCU, was involved in the same study as Dr. Schrijver and adds her thoughts on the subject.

Dr. Dumur notes that reproducibility studies on other microarray-based studies have used the same RNA-extracted materials across participating institutions. This could create good reproducibility, but left undefined by such studies is the question, What is the reproducibility from the starting material?

The Pathwork study addressed this by having all four institutions share the same tissue material—those 60 frozen samples—giving a more realistic experience. “This is ultimately the material we’re going to be using in the clinical lab,” Dr. Dumur says.

Dr. Dumur then proceeds to take a backstage tour through the Pathwork data, with a larger goal in mind: conveying to pathologists the tough questions they need to ask about the assays. “I’m an analytical person,” she says simply.

Though the labs used different technologies to isolate the RNA, results were reproduced. She credits the strong algorithm used to normalize the data, developed by Pathwork, which outperforms the MAS5 algorithm that will be familiar to users of the Affymetrix GeneChip. The Pathwork algorithm is based on 121 normalization genes, Dr. Dumur explains—not necessarily housekeeping genes, but genes that are quite invariable across different tissue types.

“Even though we were sharing pieces of tissue from the same specimen, you can have different cellular compositions—slight differences, or dramatic differences. So having the reproducible results was very surprising to me,” says Dr. Dumur.

She also urges pathologists to look at the number of each tumor type used in any assay’s validation studies. Doing so allows one to feel “analytically confident,” as she puts it. A test that reports a 100 percent match for a particular tumor type may or may not be impressive—getting a correct match on only one sample in a tumor class is not a confidence booster. More powerful would be larger sets, of all types—25 samples or greater per class, says Dr. Dumur. She cites breast cancer as an example and begins ticking off a list: “You have ductal, lobular, PR, ER, luminal, basal—so many different varieties. If you don’t cover them all during the validation, then you cannot name with certainty that a metastatic specimen arises from a breast primary tumor.”

What’s becoming evident is that what first appears to be a molecular matter is actually a bioinformatics one. “Yes!” says Dr. Schrijver. “But I don’t think it’s feasible, with this type of assay, for pathologists to individually look at the raw data, unfortunately.” This will plague future assays as well, she says. Pathologists may find themselves sailing among the data like so many Darwins, who on his voyage aboard the HMS Beagle amassed samples of, well, everything, and seemed to leave no observation unrecorded in his journals, from fossils to weather to the hospitality of South American generals. “At this point in laboratory medicine it is relatively easy to generate a ton of data,” says Dr. Schrijver, “and it’s relatively difficult to interpret these data into something that’s clinically useful.”

Kevin Halling, MD, PhD, has also been involved in validating the Pathwork assay, work that has been presented in abstracts and appears to have left him with more questions than answers.

“Here’s the dilemma,” he starts, then laughs. “Um, the dilemma is there are multiple dilemmas,” says Dr. Halling, codirector of the clinical molecular genetics and molecular cytology laboratories, Mayo Clinic.

The first involves discrepancies between test results and clinical impression. “What’s the truth?” As Drs. Tan and McGee have already learned, this is no academic question. In Dr. Halling’s experience, such splits were not resolved in favor of the test or the clinical impressions with any predictability. A fellow pathologist at Mayo, Fabiola Medeiros, MD, would sometimes find that further exploring the clinical history would tilt the answer in favor of the test; in other cases, her own pathologic impressions simply became stronger. “It’s difficult to get at the gold standard. How do you know what’s right?” Dr. Halling asks.

It’s possible, he continues, that discrepancies could be attributed to shortcomings in the groups of tumors represented by the test. The various types of ovarian cancers, for example, may have different gene expression profiles. Then again, he says, “We’re comparing this to morphology, and regardless of your pathologic impression, it may be that the assay is more accurately predicting how that tumor would respond to a chemotherapy. It may be that discrepancies are sort of irrelevant.”

Mahesh M. Mansukhani, MD, was involved in validating Rosetta assays at Columbia University Medical Center, New York, where he is associate professor, clinical pathology, and director, CUMC molecular pathology laboratory. Rosetta reports that his work on the squamous test was set to be published online in the Journal of Clinical Oncology on March 9.

One challenge of validating new molecular markers, says Dr. Mansukhani, is having to start with cases where pathologists agree on tumor type. There is no other gold standard, he says. “How do you then extrapolate from that data to cases where there’s a real question as to what the tissue of origin is?”

With molecular assays, he says, “There’s always a question of nontumor tissue contributing to whatever data you get. Whereas with in situ assays, you can differentiate between tumor tissue and nontumor tissue, just by morphology.

“For instance, we had a case where there was lots and lots of mesothelioma in colon, and there was a small amount of colon tissue,” Dr. Mansukhani says. The ­micro­RNAs of normal colon tissue overwhelmed the mesothelioma, producing the signal of an adenocarcinoma. “But when we microdissected out tumor from normal, and didn’t include the normal, then we just got signal from the mesothelioma, and it showed up as a mes­othelioma based on the assay. So when using primary tumors in validation studies, and stating that we’ve got 100 percent sensitivity for these tumors, the question is, Is the sensitivity driven by the normal tissue, or is it driven by the tumor?” This is a problem especially in metastatic tumors. “Often you can actually see the signal of the organ in which the tumor is, as opposed to the organ from which the tumor came.” On CUP assay reports, this might show up as the second choice.

A related matter: It’s known, for example, that within renal cell carcinoma, the different tumor subtypes exhibit different expression patterns. Perhaps, Dr. Mansukhani suggests, the CUP assays will have different specificity and sensitivity for each of the different histology types—which again leads to the matter of how many different subtypes were included in the validation process.

He also wonders how carefully oncologists read the literature, especially since they’re often the ones who first learn of the test and then push for it. In his experience, data often rest on idealized cases, a fact he says is not always fully appreciated by those who only glance at validation studies. “They don’t always have a full understanding of how reported sensitivity and specificity apply to individual cases,” he says.

What did Federico A. Monzon, MD, find when he worked on the Pathwork validation? He says the test has a “pretty robust” performance when samples fit the established parameters in terms of tumor content and tissue viability. “We also learned the performance is actually fairly equivalent between metastatic tumors and poorly differentiated primary samples; metastatic samples performed a tiny bit lower than primary tumors, but very close,” says Dr. Monzon, medical director of molecular diagnostics, The Methodist Hospital, Houston. This validation study has been accepted for publication in JCO, he says.

He too expresses concern over the number of specimen types used in validation studies for different assays—a 100 percent match for one specimen is not enough to hang one’s hat on. Pathologists will need to do a little poking around for validation information. “I don’t think it’s as readily available as people think,” Dr. Monzon says. “You can certainly go and dig out the publications and figure it out yourself—it’s probably not going to be prominently displayed in the promotional materials.”

“We’re now—and this applies to many molecular tests in the cancer area—starting to get inundated with different tests from different laboratories, each one claiming to be the best,” says Dr. Monzon. “We need to be very critical about this. It’s the responsibility of the pathologist to determine which test will best serve our patients.”

The possibilities

Dennis Sgroi, MD, director of breast pathology and associate pathologist, Massachusetts General Hospital, is a spirited supporter of these assays. Like Dr. Moore, he helped build the bioTheranostics-by-way-of-Arcturus test. “This is more accurate. If you do IHC, why wouldn’t you do this?” he asks.

Well? He doesn’t use a CUP assay at MGH. Ask him why, and he ponders the question a bit before floating several possible reasons.

Pathologists in academic settings “are so used to interpreting immunohistochemical stains and trying to figure out tumors based off IHC—that’s their knee-jerk reaction. They’re most comfortable with that, so that’s what they order,” Dr. Sgroi says. And unlike IHC, reimbursement policies for CUP assays are unresolved for now.

In community settings, he suspects, CUP assays are more welcomed. “They’re frustrated by sending out a case, getting the $1,200 immunohistochemical workup, and not being much further than they were when they started,” says Dr. Sgroi, who is also associate professor, Department of Pathology, Harvard Medical School.

He shoulders a bit of the blame himself. Any new test or procedure needs a champion, like a legislator guiding a bill through Congress, and Dr. Sgroi says he hasn’t been that person. “It’s been a side interest of mine, so I haven’t pushed it. I’ve been pursuing other, treatment-predictive biomarkers instead.”

Other reasons for the lack of interest, given by other pathologists, are a bit stickier, and range from institutional politics to fear of losing business to what one person terms “billing selfishness.”

None of these reasons are unique to CUP assays. They’re part of the usual greeting given to almost any new medical technology, a vague mix of reality and anxiety. It must drive diagnostics companies nuts.

But most observers say the CUP assays will catch on if they help match patients to appropriate therapy; if they minimize potential toxicity associated with administering the wrong agents; and if they cut costs.

Intones Dr. Sgroi: “If the right clinical situation comes up, they will order it.”

No one knows for sure what the right situation will look like. Should the tumor panels take on the most difficult cases, or the most common ones? Will more tumors and subtypes be added to the panels? When? Will targeted therapies be available? What information will prove to be most valuable—tissue or more refined targets, such as receptors or pathways? Or both? Will the tests receive FDA approval? What’s better, fresh-frozen tissue or FFPE samples? Will testing be handled primarily at companies’ own labs, or will later versions of the assays move into academic settings, reference labs, and even larger community hospitals? These are a lot of moving targets.

As the discussion continues, predictions abound. It’s a little like listening to a flock of fortune-tellers, all of whom, granted, have advanced degrees.

The goal, as Dr. Sgroi sees it, is to start with the most common tumor types, then refine the panels over time. The assays now tackle ovarian cancer as one category. That may be OK, at least for now, says Dr. Sgroi. Treatments for the major subtypes are fairly clearly established, and pathologists can recognize major subtypes relatively easily. The rare subtypes aren’t easy to pick up, but there may not be specific treatments targeting them right now, either.

Dr. Halling suggests thinking of CUP assays as “fancy immunohistochemistry. It’s like having the ability to look at 1,500 immunostains.” He sees value in reflexing to a CUP test if traditional morphology and a battery of IHC stains fail to yield an answer.

The University of Pittsburgh’s Dr. Nikiforova plumps for an in-house test, arguing that molecular information should not be separated from pathology diagnosis, IHC, and other pathology techniques. CUP assays will require pathologist involvement, no ifs, ands, or buts.

The blunt fact remains: Cancer can behave like almost any disease, Dr. Schrijver says. For truly difficult brainteasers—perhaps one percent of the cases encountered by pathologists at Stanford, she says—the added clinical value of a CUP assay may be limited without a backup method to confirm the finding. Of course, that lack might be the reason for turning to a CUP assay to begin with. “These are just very, very difficult cases,” she sighs.

For now, Stanford is running the Pathwork assay only on a research basis. “We want to look at what the added clinical value of the assay might be,” Dr. Schrijver says. While noting that every institution is different, she says the demand for such an assay at Stanford has been small initially (perhaps, she suggests, because few oncologists even know the assay exists). That could change now with the recent availability of Pathwork’s FFPE assay, since most of the tissue specimens are processed this way. “That opens up a whole new pool of samples to order this test on,” she says.

NeoGenomics’ Dr. Moore is among those who expect the CUP assays to slowly shift from the more difficult cases to the more routine ones. He draws an analogy to flow cytometry, which initially was applied only to cases presenting difficult diagnostic dilemmas. Now, he notes, flow is basically the starting point for most hematologic malignancies.

Next, he says, it would be smart for the companies to start incorporating response-to-therapy components into their assays. “Clearly the next logical step,” Dr. Moore says. The tests are already measuring the RNA expression, which, in many cases, correlates to protein expression. “So it’s not a big stretch.”

Dr. Monzon likes the idea of using the assays to rule out tumors. “Let’s say you have a tumor in the bladder, and the appearance is fairly unusual for your run-of-the-mill tumor. You might want to exclude that it’s coming from anywhere else. That’s a situation I ran into not infrequently” at a previous institution, a major cancer center. “It’s uncommon for tumors to have these unusual features, but cancer is so common in our population now that you end up with a significant number of cases.”

Dr. Mansukhani asks if the primary purpose of a CUP test will be to direct treatment. If so, then the ultimate proof of their value will be outcome following treatment, he says, rather than correlation to specific histology. Or is the goal to speed up the initial workup and direct further investigations to identify the site of origin, with savings to the system? Such questions have not been answered. “They’re not even being addressed,” Dr. Mansukhani says.

Of course, these assays are still in their toddler years. As Dr. Moore says, this is an “evolutionary process. I don’t think anyone has drawn a line in the sand and said, ‘We have the perfect assay.’”

He’s right. No company is saying that, and all are working at a brisk pace to improve on their early offerings.

“But obviously we have to walk before we can run,” says Dr. Erlander.