For patients presenting with uncertain primary tumors and who do not have a primary site identified even after exhaustive investigation empiric chemotherapy is rarely successful and often quite intolerable [7
]. Better survival is achieved when tissue-specific treatment strategies are utilized [11
]. Thus, availability of a test that identifies the tissue of origin would increase the chances of a patient receiving a more targeted and less toxic therapy. Depending on how such a test is integrated into the workup of uncertain primaries, it may also reduce the overall time and expense associated with the hunt for a primary tumor. In addition, as new tissue-targeted therapies are expected to be introduced, correct identification of the primary site will become even more important in guiding optimal patient management.
Recently, we reported that the Pathwork Tissue of Origin test showed robust performance in a large validation study with 547 tumors of known origin [29
]. We now have evaluated the performance of this test in a cohort of 21 CUP cases. In this cohort the Pathwork TOO Test was able to identify a probable single primary site in 16 (76%) of the cases. These results suggest that the TOO Test can significantly reduce diagnostic uncertainty in patients with CUP.
Three recent studies have reported on gene expression profiling as a strategy to determine TOO in patients with CUP [31
]. Two of these studies employed a microarray-based 495-gene-expression classifier (CupPrint, Agendia, Amsterdam, NED), which gave a confident and clinically valuable result in 14 (64%) of 22 CUPs from one study [32
] and a clinically feasible result in 18 (86%) of 21 CUPs from the other study [31
]. The third study evaluated a 10-gene RT-PCR-based expression assay (Veridex, La Jolla, CA USA) and identified a TOO in 23 (62%) of the 37 specimens classified as CUP after IHC analysis [33
]. Thus, expression tests with widely varying designs and features (e.g., number/types of genes; algorithm strategies; training set size; specimen handling protocols; number of tissue types on panel) have recently demonstrated an ability to issue a classification for a majority of CUPs.
This general uniformity in call rates for identifying a TOO in CUP patients among the various expression tests (i.e., 62%-86%) does not necessarily imply a uniformity of call accuracy or assay range. By definition, the gold standard for tumor calls in CUP cases is unknowable. Thus, accuracy in tissue identification can only be evaluated with the use of a large and diverse set of known tumor specimens. This type of evaluation is needed before any projections can be made about a test's clinical diagnostic value--much less its potential impact on therapy choices and outcome--in the setting of CUP. For example, in one of the studies cited above [32
], a parallel analysis of 84 tumors with known origin revealed total assay accuracy of 83%; however, the test misclassified 7 of 11 lung tumors and 3 of 3 pancreas tumors. Thus, if we extrapolate these results to CUP specimens, a large number of CUP tumors with lung or pancreas origin could yield a positive, albeit incorrect, tissue identification. Due to the small number of studied samples (in particular of the lung and pancreas groups) it is not possible to determine the performance of that test for individual tissue types. This same assay showed an accuracy of 87% in a previous study with 119 tumor samples; however, some tissue types were represented by only one or two specimens (e.g., breast, adrenal) [27
]. The 10-gene RT-PCR assay showed an accuracy of 76% in a set of 48 metastatic samples [26
]. Importantly, in a less rigidly defined set of 120 CUP specimens analyzed with the RT-PCR assay [33
], this test could not yield a result (due to insufficient mRNA quality or yield) or failed to assign a TOO (perhaps due to a limited 6-tissue panel) in 48% of the cases. When the Pathwork TOO Test was validated in a large multicenter study (n = 547), of poorly differentiated and undifferentiated primary cancers and metastatic tumors, it showed an overall sensitivity (positive percent agreement with reference diagnosis) of 87.8% for the 15 tissues of origin included in the panel [29
]. In this validation study, each of the 15 different tissue sites was represented by at least 25 specimens. These performance characteristics in known tissues must be considered when assessing the likelihood of expression test accuracy in CUP cases.
Our study has several limitations, chief among them the lack of a gold standard for comparison. This inability to confirm accuracy, which is unavoidable in CUP diagnostic studies, only heightens the importance of quality and rigor in the associated clinical and pathological investigations. In this study, patients and CUP specimens were characterized in clinics with extensive experience in oncology workups. Full histories, imaging results, lab records, and pathology reports (including an average of six IHC performed in each case) were available before the CUP diagnosis was issued. Thus, within the inherent constraints of CUP study design, this study provides a fair measure of test performance--certainly at least as fair as any of the seminal validation studies of IHC itself [17
]. Another limitation of this study was its small size. This is related to the difficulty in finding well-characterized fresh-frozen CUP specimens. Using frozen tissue as a basis for initial CUP performance testing makes sense since it yields more intact mRNA [37
]; however, we recognize that it is necessary to validate the TOO Test with formalin-fixed paraffin embedded (FFPE) specimens in order to allow the test to be applied widely in non-research clinical settings.
Given the absence of a gold standard to evaluate CUP assay performance, a surrogate marker of accuracy utilized by prior studies has been the correlation of the TOO prediction with existing clinicopathologic information [31
]. In our study, 10 cases show consistency with the clinicopathologic-based differential diagnosis (Table ) and in six cases the test's suggested diagnosis was inconsistent. Review of cases where TOO Test positive results were not consistent with IHC and/or tentative clinical suggestions sheds further light on the trustworthiness of gene expression profiles. None of these six TOO Test results were deemed implausible or in absolute contradiction to any known clinicopathologic findings. Even the TOO Test call of breast cancer in a male (Case 18), while surprising, cannot be considered patently incorrect given current estimates of approximately 2000 such new cases in the U.S. every year with a rising incidence [39
] and also considering reports of male breast cancer presenting as CUP [40
]. In addition, the accompanying TOO Test negative results add weight to the plausibility of surprising positive calls by ruling out several of the originally suspected primaries. In fact, the availability of "rule outs" is a unique feature of TOO Test design that may eventually prove valuable in case management; other tests provide the result of the most molecularly similar tumor but do not report information that allows ruling out specific tissue types. Thus, although six of the TOO Test results were unexpected, given the published performance characteristics of the test and recalling the performance deficits of IHC as cited previously, they were still clinically plausible TOO sites. Based on the overall clinical contributions of TOO Test results, we estimated that the expression test would have helped to inform patient management decisions in the majority of cases (Table ): definitively in the 16 positive cases by identifying a single primary site, and to varying degrees in the five indeterminate cases by eliminating potential primary sites.
Another measure of the clinical relevance of TOO identification is the ability to evaluate response to therapy in those tumors that were treated with tissue-specific approaches. In this regard, Varadhachary and co-workers reported that patients with a molecular signature of colon carcinoma showed better response to colorectal-specific therapies than to empiric CUP therapy [33
]. Unfortunately, the size of our study and the unavailability of full outcome and treatment information for all 21 patients make it impossible for us to quantify how the expression results might have translated into changed therapy or improved outcomes. However, review of the treatment data available for about half of the cases indicated that most patients got the type of supportive care with or without empiric chemotherapy that is typical in CUP [4
]. Of the 11 Mayo patients, for example, four received at least 4 cycles of chemotherapy and one received radiation therapy (4 radiation sequences, 800 cGy). Many of these patients likely would have received a more tissue-specific therapy if a trusted expression test result had been available at presentation. In Case 11, for example, one of the options for more ovarian-focused chemotherapy might have been given instead of paclitaxel/carboplatin if the TOO Test result of ovarian cancer had been available. In Case 6, doxorubicin and paclitaxel might have been given rather than only supportive care if the TOO Test result supporting a breast origin had been known. Similarly, in Cases 4 and 8, patients might have received bevacizumab instead of a broad-based chemotherapy (paclitaxel/carboplatin with or without gemcitabine) if the TOO Test result indicating colorectal cancer had been available at the time of diagnosis.
Recently, based on retrospective and prospective analyses [12
], some researchers have already advocated using expression profiles to guide targeted therapy in patients with colorectal cancer profiles. Such changes in CUP therapy directed by molecular profiling might reasonably be expected to equate to improvement in outcomes, but this important hypothesis needs to be tested in prospective studies. Only carefully designed studies will reveal whether tumors originally deemed CUPs respond to specific treatment in the same manner as more well-differentiated tumors of the same tissue type. It may be, for example, that hard-to-identify CUP-like tumors actually possess distinct genetic/phenotypic aberrations (while maintaining tissue-specific expression traits) that limit their susceptibility to therapies tested only in non-CUP tumors [41
]. In this context, because of the TOO Test's high accuracy in poorly differentiated tumor types and also because of the high number of genes assayed, this test may be particularly well suited not only to classifying tissue origin in CUP but, in the future, to providing information about the tumor's susceptibility to specific therapies.