Karin A. Oien and T.R. Jeffry Evans JCO Sep 20 2008: 4373–4375.

Despite recent advances in clinical cancer medicine, metastatic cancer of unknown primary site (CUP) remains a major challenge. Some 10% to 15% of cancer patients present with metastases, and the site of origin may not be initially apparent. In one third of these patients (approximately 4% of all cancers), even after extensive evaluation, no primary tumor is identified and they are designated as having CUP.

CUP comprises a range of cancer types and subtypes. Lymphoma, leukemia, melanoma, germ cell tumor, and sarcoma are usually excluded. Thus, CUP translates to carcinoma of unknown primary, of which 90% are adenocarcinomas or poorly differentiated carcinomas, 5% are squamous carcinomas and 5% are neuroendocrine carcinomas. Among the adenocarcinomas, the most common primary sites at autopsy are lung and pancreas (both comprising approximately 25%), followed by colon and stomach, then breast, ovary, prostate and the solid-organ carcinomas of kidney, thyroid, and liver.

Identification of the likely primary site may significantly influence optimal patient management. A number of therapeutic advances, including therapies targeted to specific pathways, may be applicable not only for breast and prostate cancers but also for GI cancers, especially colorectal. Diagnostic work-up includes clinical assessment, computed tomography scans of chest, abdomen, and pelvis, and targeted evaluation of any specific signs or symptoms. CUP is defined as the lack of any detectable primary site after full evaluation.

Biopsy of the metastasis establishes a diagnosis of malignancy and of the broad tumor type. However, it is not easy to specify the primary origin of an adenocarcinoma on routine hematoxylin and eosin staining. A number of differentiation-associated genes have emerged and entered clinical use during the last decade as markers for immunohistochemistry (IHC), detecting proteins such as CK7, CK20, PSA, ER, and TTF1, which, within larger panels, can classify carcinomas and predict the primary site of adenocarcinomas.

Large-scale mRNA expression profiling has been applied to a number of clinical questions in cancer research. CUP is no exception: multiple gene signatures for the prediction of primary site have been generated. This issue of Journal of Clinical Oncology contains two articles in which previously described mRNA-based gene signatures are evaluated in cohorts of patients with CUP to assess their potential contribution to patient management. It is inherently difficult to validate CUP assays because, by definition, the primary site remains unknown; the authors are to be congratulated for taking on this challenge. Varadhachary  evaluated the Veridex (Raritan, NJ) assay in 120 CUP patients and compared the expression profiles of the metastases with known clinicopathologic information and therapeutic response. Horlings  evaluated the Agendia (Amsterdam, the Netherlands) CupPrint assay in 84 known tumors and 38 CUPs and compared the results with known clinicopathologic information. All CUP patients had undergone full diagnostic evaluation. Both assays are mRNA-based and were applied to formalin-fixed, paraffin-embedded pathology samples.

The Veridex CUP assay comprises 10 genes tested by reverse transcriptase polymerase chain reaction (HUMPB, TTF1, DSG3, PSCA, F5, CDH11, MGB, PDEF, PSA, and WT1). It distinguishes six primary sites (lung, breast, colon, ovary, pancreas, and prostate) and thus focuses on true adenocarcinoma of unknown origin (ACUP). If the profile fits none of these, it is designated "other." Of the 120 patients studied, 78 were from a retrospective cohort who had undergone empirical CUP treatment, and 42 were from a prospective cohort, with therapy guided by the results of the diagnostic evaluation, including IHC but not the Veridex assay.

The assay yielded a result in 104 patients (87%). In the rest, the extracted mRNA was of insufficient quality or yield. The Veridex assay assigned a tissue of origin in 63 patients (61% of those with suitable RNA and 52% of the total). Fifty-five patients were designated lung, pancreas, or colon, which equates to 53% of the total population. This is consistent with previous CUP autopsy studies. The remaining eight patients were assigned to an origin in ovary, breast, or prostate. Thus, in 48% of patients, the assay either did not yield a result or was unable to assign an origin.

The predicted primary site was compatible in most patients with the clinicopathologic information. Direct comparison with IHC results for each patient would have helped to evaluate the clinical utility of the assay, but IHC was performed variably and not in all cases (unlike in Horlings’ study. It is stated that extensive IHC was performed in 37 patients suggesting a primary site in only four (11%) compared with 23 (62%) by the CUP assay. This would mean that the Veridex assay provided clinically useful additional information in 50% of patients. However, this IHC comprised only a minimum panel of CK7, CK20, and TTF1, with additional markers in a minority of cases, and may therefore be suboptimal for comparison.^1,4

One of the most interesting aspects of this report was the correlation of the expression profiles with the therapeutic response. Patients with a lung or pancreas profile responded poorly to treatment. In contrast, patients with a colon profile, usually on both Veridex assay and IHC, had a better outcome (sustained partial response) with colon cancer–specific therapy administered in the prospective cohort, than with the empiric platinum-/taxane-based CUP therapy administered in the retrospective cohort.

Horlings  from the Dutch Cancer Institute tested the Agendia CupPrint assay, which uses 495 predictive genes in a custom oligonucleotide microarray. CupPrint correctly classified 70 (83%) of the 84 known cancers tested. These were either primary tumors or metastases of known primary site, comprising 10 cancer types including breast, lung, colon, kidney, stomach, pancreas, prostate, ovary, thyroid, bladder, and dysgerminoma (not all of which are adenocarcinomas). CupPrint was excellent at classifying many primary sites, such as breast and colon. However, the misclassified metastases included seven (63%) of the 11 originating from lung and all three (100%) from pancreas, as well as two of three from stomach. This lack of sensitivity for lung and pancreas is an important limitation because these are the most common primary sites giving rise to ACUP.

Thirty-eight patients with ACUP were studied. Sixteen cases had already been resolved by IHC; in 15, the CupPrint prediction agreed. Twenty-two cases were described as previously unresolved, although almost all had tentative suggestions as to primary site from clinical investigation or pathology including IHC. In these, CupPrint predicted a single origin, which was mostly consistent with the clinicopathologic information, although some suggestions (such as rhabdomyosarcoma or germ cell tumor, in an elderly man and woman, respectively) were surprising. It is, nonetheless, a strength of the assay that all tumors yielded a prediction with CupPrint. It is a further strength of the study that all biopsies were reexamined by a single pathologist with application of an IHC panel comprising 10 to 12 genes.

To our knowledge, these are the first large-scale validation studies of mRNA-based gene signatures for the prediction of primary site in CUP patients. Two smaller studies have been reported.However, both studies in Journal of Clinical Oncology convincingly show that formalin-fixed, paraffin-embedded tissue is suitable for mRNA-based assays, even from small core needle biopsies and archival material (up to 10 years old), although successful retrieval of mRNA reduces with specimen age. In both studies, the predictions broadly correlate with existing clinicopathologic information, although CupPrint's lack of sensitivity for lung and pancreatic origins is unexpected. However, the underlying aim is to assess potential clinical utility, and it remains uncertain whether these CUP assays are confirming existing suspicions, with varying levels of confidence, or providing significant new diagnostic information. Both groups used pathology with IHC as the standard comparator and both suggest that IHC is superceded by their mRNA-based assays in up to half of cases. However, this can be argued only if IHC has been optimally selected, performed, and interpreted. Despite the authors’ best efforts, it is difficult to be certain in every case exactly what IHC was used, whether it was the best panel for the clinical context and morphology, and what was the resulting prediction.

This is the first time that retrospective correlation of molecular profile and therapeutic response has been undertaken in a significant CUP cohort. The outcome data comprise response rate rather than survival, and the patient numbers are still somewhat small. Nevertheless these results are exciting because they suggest that treating CUP patients according to their gene expression profile may improve outcome.

What we need now are prospective studies in which expression profiling results, whether at the mRNA or protein level or both, are used to direct tumor site–specific therapy, to determine whether this approach is superior to empirically selected CUP treatment regimens in terms of patient outcome. We will need to compare different and emerging (eg, miRNA or proteomic) profiling methodologies and different gene panels,^6-16 to determine which is the optimal technology and gene set, or whether integration of multiple profiles is superior. Ultimately, we will need to apply these techniques within the diagnostic and clinicopathologic framework of a multidisciplinary team for best management of these complex cases. And for full understanding, we must continue to investigate the biologic roles of these signature genes in both normal tissues and cancers.

As Horlings et al state, "Immunohistochemical analysis is a crucial component of the clinical work-up to identify the histogenetic origin of a malignancy." These two articles clearly demonstrate that RNA-based profiling is an additional and powerful analytic tool that may support the prediction of primary site, and thus enable tailored therapy and improve outcome. Expression profiling is raising the hopes and removing the nihilism for CUP patients.