Identifying Tumors of Uncertain Origin

With advances in microarray-based genomic diagnostics, pathologists could be on the brink of a new era of clarity and efficiency in diagnosing challenging cases

By Shawn H. Becker, MD

Few clinical issues are more frustrating for the pathologist than metastatic tumors where the primary site remains uncertain following a diagnostic workup. A recent study highlighted the case of a woman with an ovarian mass, which a biopsy sample showed to be a clear cell carcinoma of the ovary. Investigators suspected that the tumor was a possible kidney metastasis, but her thorough workup, including immunohistochemistry (IHC), was inconclusive. Such ambiguity leaves physicians with the unenviable quandary of how to diagnose and, thus, best treat the patient.

The Problem

Metastatic tumors of uncertain origin—where a presumptive diagnosis can be made, but the physician is unsure—are a common problem. Cancer of Unknown Primary (CUP), for example, constitutes a heterogeneous group of metastatic tumors for which no primary site can be

detected following a thorough medical history, physician examination and diagnostic work-up. CUP represents the seventh to eighth most frequent type of cancer worldwide and the fourth most common cause of cancer death among males and females.¹ According to Surveillance, Epidemiology and End Results data, in the United States, CUP accounts for 2.3 percent of all cancers in both sexes,1 or around 30,000 new cases each year.² A recent analysis performed by Tong et al³ concluded that the number may actually be much higher than previously thought, showing that, in the United States, there are 53,000 new CUP Medicare patients each year.

The importance of identifying the tumor of origin in CUP cases has been documented in the literature. In a prospective clinical study, Abbruzzese et al concluded that "the survival duration of patients in whom the primary tumor was diagnosed was superior to that of patients in whom the primary tumor remained unknown."⁴ Further, the National Comprehensive Cancer Network's guidelines emphasize the significance of identifying the tissue of origin so that cancer-specific treatment recommendations can be followed.

Current pathological techniques, including IHC and imaging studies such as x-rays and computed tomography (CT), identify the primary tumor site only 25 percent of the time in living patients.⁵

A more robust diagnostic solution is needed for identifying uncertain primary tumors. Gene expression profiling–using microarrays, in particular–has been discussed as a possible solution. Advances in microarray-based genomic diagnostics could mean that pathologists are on the brink of a new era of clarity and efficiency in diagnosing these challenging cases.

Limitations of Existing Technologies

CUP cases are especially frustrating because the pathologist deploys every known diagnostic tool in his armament, but is still unable to provide a diagnosis. No one knows the biological reason these tumors are so difficult to detect. Hypotheses include that the primary tumor remains microscopic and escapes clinical detection or that it disappears after creating the metastasis. Another theory is that replicating cancer cells tend to metastasize to distant sites early and actually fail to grow at the site of origin. Histology alone does not allow the origin of CUP specimens to be clearly identified due to morphological similarity to other tumor types, or because they are so poorly differentiated that they are difficult to relate to any specific tumor types.⁶ As a result, while many of the pathologist's diagnostic tools have increasing power, they either do not provide enough information or their utility is limited to specific clinical situations.

IHC studies can sometimes identify the primary tumor, even in poorly differentiated metastases.¹ However, many available IHC markers do not address the whole range of potential tumor types. Further, even the most commonly used staining phenotypes (CK7 and CK20) produce enough false positives and false negatives to make a definitive diagnosis difficult based on this technique alone.¹

In a 2004 review article, Mintzer et al asserted, "Although extremely helpful, tissue antigens such as CK7 and CK20 themselves rarely specifically identify a tumor site. But they can help suggest a primary site through the process of elimination."⁷

Cytogenetic methods, such as fluorescent in situ hybridization (FISH), which assess chromosomal abnormalities to pinpoint the primary tumor site, can provide insights in a number of specific situations, but not in a comprehensive manner. This technique is limited; only a few diagnostic chromosomal abnormalities have been identified to date.¹

Imaging is also a standard tool used to assess CUP patients. CT scans (particularly of the abdomen and pelvis), mammography, magnetic resonance imaging (MRI) and fluorodeoxyglucose (FDG) positron emission tomography (PET) can all be helpful. But, too often, these various imaging techniques are only applicable to specific situations based on the anatomic location of the tumor and can be ineffective at determining the origin of the tissue.

There are many reasons for the failure of imaging techniques to determine the site of origin of a tumor—very small tumor size, confounding structural abnormalities and limitations of each imaging modality.

Many of the tests described above are often run in parallel, yet still fail to identify the primary tumor site in CUP cases. This can lead to increased diagnostic costs, while the patient still may not receive optimum treatment.

Genomic Tools

Genomic tools are particularly well-suited to this challenge. In recent years, starting with seminal articles such as that of Alon et al8 in 1999, scientists have explored using gene expression signatures to identify a tissue. Alon and his colleagues showed that, using sophisticated algorithms, tissue types (e.g., cancerous and noncancerous) can be separated on the basis of subtle distributed patterns of genes, which individually vary slightly between the tissues.⁸

Since then, the literature has mushroomed, as scientists have repeatedly demonstrated similar phenomena across many tissue types. In a landmark study, Yeang et al demonstrated the feasibility of performing clinically useful tumor classification from samples of multiple tumor types. The authors state, "In principle, tumor gene expression profiles might serve as molecular fingerprints that would allow for the accurate classification of tumors."⁵

In nearly all cases, multiple genes are required to generate a statistically valid signature, which in turn has driven growth of the tools for analyzing multiple genes in parallel–namely, the microarray.

Microarrays

Microarrays consist of small DNA fragments called probes that are arranged or "arrayed" on a small glass platform. The probes comprise a short section of a specific gene's entire DNA chain. RNA is extracted from the tissue of interest, labeled with a detectable marker and spread over the microarray. The RNA from the tissue of interest will bind to complementary gene-specific probes on the array, known as hybridization. The relative fluorescence intensity (which can be measured with tools such as lasers) of each gene-specific probe reflects the level of expression of the particular gene. The greater the degree of hybridization, the more intense is the signal, implying a higher relative level of expression.⁹

With full sequencing of the human genome completed in 2003, single arrays can now represent the entire human genome. This has prompted two fundamental changes that have enabled creation of diagnostic tests based on this technology. First, it has accelerated discovery of genomic profiles; all of the possible genes are present and it becomes a matter of sophisticated informatics to determine which genes are ultimately useful for identification. Second, once molecular profiles have been discovered, it is now possible to answer multiple diagnostic questions with one array. A prototypical application determines to which of several possible tissue types a specimen belongs.

Another approach to gene expression profiling is polymerase chain reaction (PCR), which involves amplifying (or copying) small segments of DNA for molecular analysis. Although microarrays are neither as sensitive nor as quantifiable as methods like PCR, the advantage of microarrays is that they can enable a far greater number of genes to be evaluated at one time, providing more robust diagnostic information to the clinician. Additionally, on a practical level, reagent costs remain fixed with microarrays.

Clinical Application of Microarrays

Multiple studies have demonstrated that, using microarrays, gene expression profiling is a feasible and accurate way to classify cancer of unknown origin in a research setting.^8,10 However, several challenges have prevented this research from translating to the clinical setting, including: