Unknown primary cancer (UPC) represents a group of heterogeneous cancers and is defined by the presence of metastatic disease without an identifiable primary tumor site on presentation. It constitutes approximately 3–5% of all cancers, although the incidence can vary depending on the definition used and the extent of investigations performed.1–3 In a study conducted at The University of Texas M. D. Anderson Cancer Center (MDACC), the incidence was approximately 1.5% of all patients diagnosed with cancer.4 According to 2003 Surveillance, Epidemiology, and End Results data,5 30,000 (2%) new cases of cancers of “other and unspecified primary sites” were diagnosed in the U.S. To put this in perspective, approximately 30,000 new cases of pancreatic carcinoma were reported for that same year. There is a substantial and unmet need for basic research and therapeutic trials for UPC. With the increasing availability of new diagnostic modalities, physicians must determine how much diagnostic workup is sufficient before accepting a diagnosis of UPC. They must also consider whether a more precise diagnosis will have a substantial effect on the choice of treatment and overall outcome. In addition, the process can be anxiety provoking to the patient. In the current review, we describe the workup of patients with UPC, emphasizing the role of immunohistochemical (IHC) markers and the controversies associated with new imaging modalities, including positron emission tomography (PET) scan, the role of invasive studies, and the emerging role of gene expression profiling in the diagnosis of UPC.
Unknown primary cancer (UPC) is defined by the presence of metastatic disease for which a primary site is undetectable on presentation. Computed tomography scan of the body was performed routinely in search of the primary cancer and invasive procedures were pursued in selective cases. Magnetic resonance imaging of the breast enables identification of an occult breast primary tumor in ≤ 75% of women who present with adenocarcinoma in the axillary lymph nodes and can influence surgical management. Positron emission tomography scan also can be used in the diagnosis of UPCs, but its value is controversial. Cytokeratins 7 and 20 and thyroid transcription factor are some of the histochemical markers used in most patients who present with metastatic adenocarcinoma. Some of the newly discovered immunohistochemical markers further assist in narrowing the differential diagnosis. The role of molecular profiling to make the diagnosis, establish the prognosis, and assess the response to treatment in UPCs is evolving. The authors discuss the role of histochemical markers in the diagnosis of UPC and the most recent data regarding the use of imaging and invasive diagnostic modalities and gene expression profiles. Cancer 2004. © 2004 American Cancer Society.
DEFINITION OF UPC
The criteria for a diagnosis of UPC includes a biopsy-proven malignancy (for a cancer that could not have originated at the biopsy site), no primary tumor found after a thorough medical history or physical examination (including breast and pelvic examination in women and testicle and prostate examination in men), and normal laboratory test results, including the results of a complete blood count, blood chemistry, chest X-ray, computed tomography (CT) scan of the abdomen and pelvis, and mammography or prostate-specific antigen (PSA) test.4, 6 In most studies of UPC, investigators typically exclude lymphoma, metastatic melanoma, and metastatic sarcoma and concentrate on epithelial histologies, including adenocarcinoma, squamous cell carcinoma (SCC), poorly differentiated carcinoma, and neuroendocrine carcinoma.
PREVAILING HYPOTHESIS OF UPC
It is hypothesized that in patients who present with UPC, the primary tumor either remains microscopic and escapes clinical detection or disappears after seeding the metastasis. Naresh7 hypothesized that in UPC, the angiogenic incompetence of the primary tumor leads to marked apoptosis and cell turnover. This results in a biologically advanced tumor that acquires a metastatic phenotype. He further hypothesized that in the viscera, metastases may remain dormant until subclones with angiogenic phenotypes arise and lead to rapid growth of the metastatic tumor. Testing this hypothesis in a clinical setting presents a challenge. The traditional theory that metastases occur late in tumorigenesis is also debatable. There currently are emerging data to suggest that the propensity to metastasize might be hardwired early in tumorigenesis.8 In a breast carcinoma cell line, Kang et al.9 identified a specific set of genes that were responsible for bone metastasis when injected into mice. Therefore, it is quite possible that in UPC, the primary tumor has a “poor prognosis” signature from the start and is angiogenically incompetent to present clinically (either because of failed neoangiogenesis or because of locally secreted antiangiogenic factors), but it can metastasize to various organs.
DIAGNOSIS OF UPC
A complete medical history, physical examination, and detailed pathologic examination of the tissue biopsy sample are mandatory. It is also essential that the clinician communicate with the pathologist to ensure a directed and cost-effective workup. It is important for the clinician to be aware of the favorable and treatable subgroups of patients that are well recognized. These patients should not be missed because specific treatment options are available and, occasionally, they will have prolonged survival if treated appropriately. These subgroups include2 women with axillary adenopathy with adenocarcinoma or carcinoma (treated as AJCC Stage II breast carcinoma), women with peritoneal carcinomatosis (treated as ovarian carcinoma), patients with poorly differentiated or undifferentiated carcinoma (need thorough evaluation for extragonadal germ cell malignancy), patients with neuroendocrine carcinoma (which may have an indolent course), and patients with high cervical adenopathy with SCC (lymph node dissection and radiotherapy can provide long-term survival).
Medical History and Physical Examination
It is essential to conduct a complete medical history and physical examination in patients who present with UPC, paying attention to past biopsy results, history of surgery, and removed or regressed lesions. Family medical history is also important and may suggest the possibility of a familial predisposition to hereditary nonpolyposis colon carcinoma or hereditary breast carcinoma. The physical examination should include careful palpation of the thyroid, breasts, lymph nodes, and prostate and a digital rectal examination. The findings of a thorough physical examination can guide the physician to look for a specific primary tumor. For example, the presence of left supraclavicular adenopathy “Virchow's node” or periumbilical adenopathy or mass “Sister Mary Joseph's node” can prompt a thorough investigation of the gastrointestinal tract in patients who present with poorly differentiated carcinoma. In addition, the presence of a perianal mass may suggest anal carcinoma in a patient who presents with inguinal adenopathy.
Hematoxylin and eosin evaluation
Tissue specimens from a fine needle aspiration or core needle biopsy should be subjected to light microscopic examination after they are stained with hematoxylin and eosin. It is of great importance that the pathologist have adequate tissue specimens for the various studies. After light microscopy examination, approximately 60% of cases are reported as adenocarcinoma and 5% as SCC. In the remaining 35%, light microscopy allows less definitive conclusions—poorly differentiated adenocarcinoma, poorly differentiated carcinoma, or poorly differentiated neoplasm.
IHC markers help to define tumor lineage when peroxidase-labeled antibodies are used against specific tumor antigens. Communication between the pathologist and the clinician is extremely important for selecting the appropriate battery of antibodies and cannot be replaced by using a universal battery of markers. PSA, α-fetoprotein (AFP), and thyroglobulin are some of the older routine antibodies available. No IHC test is 100% specific, including PSA, which can be positive in patients with salivary gland carcinoma.10 It is important to use IHC markers for guidance in conjunction with the patient's presentation and imaging studies to select the best therapy.
There are 20 subtypes of cytokeratin (CK) intermediate filaments. These have different molecular weights and demonstrate differential expression in various cell types and tumors. Monoclonal antibodies (MoAb) to specific CK subtypes have been used to classify tumors according to their site of origin. The two most common CK stains used in UPC are CK20 and CK7.11–13 In addition to specific cytokeratins, some of the other IHC markers used to distinguish metastatic carcinomas include thyroid transcription factor (TTF-1), gross cystic disease fibrous protein (GCDFP), and uroplakin III (UROIII).
CK20 is a low molecular weight CK that is normally expressed in the gastrointestinal epithelium, urothelium, and Merkel cells.13, 14 CK7 is found in tumors of the lung, ovary, endometrium, and breast and not in lower gastrointestinal tract tumors.15 The CK phenotype CK20+/CK7− strongly favors colon primary tumors. Some studies reported that 75–95% of colon tumors demonstrate this pattern of staining.13, 15 Only 9–15% of lung carcinomas stain positively for CK20, which is helpful in distinguishing primary from metastatic adenocarcinoma in the lung. CK20-/CK7+ narrows the differential to lung, breast, biliary, pancreatic, ovarian, and endometrial carcinomas. Approximately 85% of lung carcinomas are positive for CK7. The use of TTF-1 and surfactant apoprotein further distinguishes lung primary tumors from other CK7-positive tumors.
TTF-1 is a 38-kilodalton (kD) homeodomain-containing nuclear protein that plays a role in transcriptional activation during embryogenesis in the thyroid, diencephalon, and respiratory epithelium.16–18 TTF-1 staining is typically positive in lung and thyroid carcinomas. Approximately 68% of adenocarcinomas and 25% of squamous cell lung carcinomas stain positively for TTF-1. This marker has been helpful in suggesting the primary tumor in patients who present with metastatic cervical lymphadenopathy and in patients with metastatic pleural effusions (lung vs. other adenocarcinomas).11
Roh and Hong11 studied the expression of TTF-1 and CK20 in 68 patients with metastatic disease to the cervical lymph nodes. Lung carcinoma was the primary tumor in 29 patients, colorectal carcinoma was the primary tumor in 3 patients, and primary tumors in other sites occurred in 23 patients. TTF-1 expression was detected in 69% of patients with metastatic lung carcinomas, but not in patients with metastatic gastrointestinal carcinomas. CK20 expression was detected in 69% of metastatic gastrointestinal carcinomas and no metastatic lung carcinomas. The authors concluded that TTF-1 has a specificity of 95% and a sensitivity of 69% for metastatic lung carcinoma, whereas CK20 has a specificity of 100% and a sensitivity of 69% for metastatic gastrointestinal carcinoma.11
Antibodies to the pulmonary epithelial cell-specific proteins surfactant proteins A and B (SP-A and SP-B) are sometimes helpful for diagnosing primary pulmonary nonsmall cell carcinomas. In a study by Bejarano et al.,19 SP-A and SP-B were detected in 54% and 63% of patients with lung adenocarcinomas, respectively. SCC specimens rarely stained positively when these markers were used. They were not as helpful in distinguishing between breast carcinomas metastatic to the lung and primary pulmonary carcinoma, however, because 46% of metastatic breast carcinoma specimens showed reactivity for both SP-A and SP-B. Surfactant apoproteins have also not been found to be helpful for differentiating lung adenocarcinoma from malignant mesothelioma.
Abutaily et al.20 studied 41 malignant mesothelioma and 35 lung adenocarcinoma specimens with commercial antibodies including calretinin, E-cadherin, N-cadherin, SP-A, TTF-1, thrombomodulin, and CK5/6. E-cadherin was expressed in all adenocarcinoma specimens and in 22% of mesothelioma specimens. TTF-1 staining was positive in 69% of the adenocarcinoma specimens and in none of the mesothelioma specimens. Polyclonal calretinin was positive in 80% of mesothelioma specimens but only in 6% of adenocarcinoma specimens. They concluded in their study that E-cadherin was 100% sensitive for pulmonary adenocarcinoma and TTF-1 was 100% specific for pulmonary adenocarcinoma.
Liver metastasis that presents as the only site of metastasis is not unusual in patients with UPC. The main differential diagnoses include hepatocellular carcinoma (HCC), metastatic cholangiocarcinoma, and metastatic adenocarcinoma. HCC usually can be diagnosed by its morphology and with the MoAb, hep par 1. Hep par 1, the expression of which is confined primarily to benign and malignant hepatocytes, has recently become commercially available and aids in the IHC diagnosis of HCC.21 Most cholangiocarcinomas stain negatively for hep par 1, although it is possible for metastatic adenocarcinomas with “hepatoid” features to be focally positive.22 Unfortunately, no IHC markers are effective at differentiating between cholangiocarcinoma and metastatic adenocarcinoma.
GCDFP-15, a 15-kD monomer protein, is a marker of apocrine differentiation and is specifically expressed in patients with breast carcinomas. It is also called prolactin-inducible protein, glycoprotein-17, and secretory actin-binding protein. It is also detected in the skin, salivary gland, bronchial gland, and the prostate and seminal vesicle.23 GCDFP-15 is detected in 62–72% of patients with breast carcinomas.24–26 Clark et al.27 studied the role of prolactin-inducible protein/GCDFP-15 messenger RNA (mRNA) as a marker for breast carcinoma metastasis. GCDFP-15 mRNA expression was detected in four of six breast carcinoma cell lines. In addition, GCDFP-15 mRNA expression was detected in 11 of 16 axillary lymph node metastases (69%). Higher GCDFP-15 mRNA levels were found to be correlated with estrogen receptor-positive, progesterone-postive, low-grade tumors and GCDFP-15 protein levels assessed by IHC tests.
UROIII, high molecular weight cytokeratin, thrombomodulin, and CK20 are the markers typically used for diagnoses favoring a urothelial origin. Uroplakins (Ia, Ib, II, and III) are specific differentiation products of terminally differentiated superficial urothelial cells. Kaufman et al.28 reported that the MoAb to UROIII (clone, AU 1) was positive in 60% of primary urothelial carcinomas and 53% of metastases. The sensitivity for this marker was approximately 57%. In another study by Parker et al.,29 the results of IHC tests were reported for 112 archived paraffin-embedded urothelial neoplasms. Fifty-seven percent of the 112 tumor specimens stained positively for UROIII, approximately 69% stained positively for thrombomodulin, and 80% stained positively for high molecular weight CK. CK20 was positive in 48% of the tumor specimens. The expression of the four markers appeared to vary with tumor grade and stage. All small cell carcinoma specimens were negative for all markers. UROIII was not expressed in tissue cores of nonurothelial tissue specimens. Thrombomodulin was expressed in 10 of 37 nonsmall cell lung carcinoma (NSCLC) specimens (27%) and in 2 of 36 lymphoma specimens (5%). High molecular weight CK was expressed in 44% of NSCLC specimens. These data show that UROIII is a highly specific marker for tumors of urothelial origin even though it is expressed in approximately 50–60% of these tumors. The authors concluded that coexpression of these markers strongly suggests urothelial origin. Their study illustrates the role of an antibody panel that includes these four markers (UROIII, thrombomodulin, high molecular weight CK, and CK20) in the diagnosis of urothelial tumors.29 This panel is especially helpful for UPC.
Figure 1 delineates a simple algorithm for the IHC approach for UPC. Table 1 lists additional tests necessary for further defining the tumor lineage. A more complicated and comprehensive algorithm most likely could improve the accuracy of prediction but would make the process very confusing. Most pathologists prefer to start with the markers delineated above that provide maximum information. With the use of IHC markers, electron microscopy (which can be time-consuming and expensive) is rarely needed.
|Urothelial carcinoma||UROIII, THR, HMWCK|
|Breast carcinoma||GCDFP-15, ER, PR|
|Lung (mainly adenocarcinoma)||TTF-1, surfactant A and B|
|Medullary thyroid carcinoma||TTF-1, Calcitonin|
|Merkel cell carcinoma||CD117|
|Hepatocellular carcinoma||Hep par-1|
|Prostate carcinoma||PSA, PAP|
Serum Tumor Markers and Cytogenetics
It is accepted that men with adenocarcinoma and bone metastasis should be evaluated with a PSA level. β-human chorionic gonadotropin (β-HCG) and AFP levels usually are measured in men with a diagnosis of undifferentiated carcinoma or a poorly differentiated carcinoma (especially with a midline tumor). AFP is also useful in the diagnosis of HCC. Contrary to previous beliefs, elevated β-HCG and AFP levels do not necessarily predict responsiveness to chemotherapy or better survival.30 Other tumor markers, including carcinoembryonic antigen (CEA), CA 125, CA 19-9, and CA 15-3, are not helpful in establishing the site of the primary tumor. Koch and McPherson31 reviewed CEA levels in 32 patients with UPC in whom the primary site was later established and concluded that a CEA value > 10 ng/mL suggested the site to be the breast or ovary and in the presence of liver metastases, to be the large bowel or pancreas. In the MDACC series, 41 of 147 consecutive patients had a CEA value > 10 ng/mL and it did not help to establish the primary site (unpublished data). In conclusion, most serum tumor markers are nonspecific and, at most, play a prognostic role rather than a diagnostic role in UPC.
Motzer et al.32 described the molecular and cytogenetic results for 40 patients with poorly differentiated carcinoma and UPC. In 17 patients (42%), a diagnosis was provided by genetic analysis, including 12 patients (30%) with cytogenetic changes characteristically observed in germ cell tumors (isochromosome 12p-i [12p], increased 12p copy number or deletion of the long arm of chromosome 12). Patients with chromosomal abnormalities of germ cell tumors also responded better to cisplatin-based chemotherapy compared with patients without that diagnosis (75% vs. 18%). In another study by Pantou et al.,33 20 UPC samples were studied using various genetic techniques. In five samples (four of lymphoma and one of Ewing sarcoma), cytogenetics aided in the diagnosis of the primary. The other samples revealed multiple complex cytogenetic patterns. The overexpression of various genes, including bcl-2 (40%),34 HER-2 (11%),35 and p53 (26–53%),34, 36 have been studied in UPC samples without an impact on response to therapy or survival.
IMAGING STUDIES IN UPC
Chest X-Ray and Computed Tomographic Scan
Chest X-ray and CT scans are included in the routine tests for diagnosis of UPC. In a patient with a negative chest X-ray, a question often arises regarding the need for a chest CT scan. A small retrospective study by Latief et al.37 assessed the advantages of a chest CT scan compared with a chest X-ray in 32 of 925 patients who presented with brain metastasis without a known primary tumor. They found that in 12 patients (38%), the chest X-ray was reported as negative or nonspecific, but a CT scan of the chest revealed a primary neoplasm. As expected, lesion size was the major factor in detecting a primary tumor on chest X-ray. Considering that lung carcinoma is one of the most common identifiable primary tumors in patients who present with UPC (followed closely by pancreatic carcinoma), a CT scan of the chest is usually performed, especially if the pathologic tests are suggestive of a lung primary tumor. A CT scan of the abdomen also is performed routinely to look for an occult primary tumor38 and in women a pelvic CT scan often is needed.
Mammograms should be performed for all women who present with metastatic adenocarcinoma regardless of the results of the pathologic evaluation.
Positron Emission Tomographic Scan
The PET scan has proved to be a valuable diagnostic technique for various indications, from workup of a solitary lung nodule and mediastinal staging of lung carcinoma to the detection of tumor recurrence in various malignancies. In general, the PET scan reveals a primary tumor in 8–53% of patients with UPC and has a false-positive rate of approximately 20%. The combination of functional and anatomic imaging (PET and CT scans) is currently being evaluated and will very likely reduce the false-positive rate. To our knowledge there are a substantial number of studies of the utility of the PET scan in patients with occult primary head and neck carcinomas. Most trials have included a small sample of patients and a primary tumor has been identified in approximately 21–30% of the patients.39–45 Most physicians believe that an 18F-fluorodeoxyglucose-PET scan is useful in this patient group46 and may help to guide the biopsy, determine the extent of disease, and determine the appropriate treatment. Outside of these indications, PET scans cannot currently be recommended for the standard workup of all patients with UPC. Trials that have evaluated the effectiveness of PET scans in patients with UPC and negative conventional diagnostic tests are presented in Table 2. In addition, to our knowledge the cost-effectiveness of using 18F-fluorodeoxyglucose-PET scans in patients with UPC for the diagnosis of primary tumors has not been studied to date.
|Trial||No. of patients||Primary tumors suggested on PET/total cases (%)||Primary tumors detected on further investigation/total cases (%)||Comments|
|Braams39||13||4/13 (31)||4/13 (31)||Metastatic cervical lymph nodes|
|Kole et al.40||29||7/29 (24)||7/29 (24)||Survival not altered|
|Lassen et al.41||20||13/20 (65)||9/20 (45)||—|
|Safa et al.42||14||3/14 (21)||3/14 (21)||Metastatic cervical lymph nodes|
|Bohuslavizki et al.43||28||16/28 (57)||9/28 (32)||Metastatic cervical lymph nodes|
|Jungehulsing et al.44||27||7/27 (26)||7/27 (26)||Metastatic cervical lymph nodes|
|Johansen et al.45||42||20/42 (48)||10/42 (24)||—|
Magnetic Resonance Imaging Scans
The general role of magnetic resonance imaging (MRI) scans in the diagnosis of UPC is unclear unless there is a contraindication to a CT scan. Some authors advocate an MRI scan of the neck for diagnosis of metastatic disease, but to our knowledge the choice of an MRI scan compared with a CT scan has not been studied carefully and its use most likely would not change the treatment plan for most patients.
An MRI scan is an accepted test for patients with isolated axillary lymph node metastases and suspected occult primary breast carcinoma.47 Morris et al.48 described 12 women who presented with isolated axillary lymphadenopathy pathologically confirmed to contain metastatic adenocarcinoma. In nine patients (75%), enhancement was observed on magnetic resonance images. Surgery revealed that these patients had primary tumors at the same site. Mastectomy in two patients with negative MRI findings did not reveal any breast tumor.
Olson et al.49 studied 40 women with metastatic disease to the axillary lymph nodes and no primary tumor was found on mammograms. In 28 women (70%), a primary tumor was found on an MRI scan using a dedicated breast coil. Twelve women had negative MRI scans of the breast. Five of these 12 women underwent modified radical mastectomy and 4 of them had no evidence of tumor in the mastectomy specimen. That study showed that MRI scans of the breast enabled identification of the site of the primary tumor in ≤ 75% of patients suspected of having occult primary breast carcinoma and that MRI findings can influence surgical treatment. Negative breast MRI scans predicted a low yield at mastectomy. The common practice is to irradiate the breast instead.
ROLE OF INVASIVE PROCEDURES IN STAGING UPC
Bronchoscopy, upper endoscopy, and colonoscopy cannot be recommended for routine workups in the diagnosis of UPC in asymptomatic patients. Their use should be tailored to the particular disease presentation of each patient. Physicians should use their judgment before subjecting a patient to these procedures, especially if the results would not change the patient's overall treatment plan. For example, a triple endoscopy (including a rigid laryngoscopy, bronchoscopy, esophagoscopy) is usually performed in the routine workup of patients who present with metastatic cervical adenopathy. Colonoscopy is recommended for patients who present with resectable liver metastasis (adenocarcinoma).
STAGING ISSUES IN SPECIFIC PRESENTATIONS
Role of Bilateral Tonsillectomy in Staging Squamous Cell Carcinoma in Cervical Lymph Nodes
Studies show that in approximately 3–10% of patients who present with metastatic SCC in the cervical lymph nodes, a primary tumor cannot be detected after a routine workup. With systematic staging studies, including directed biopsies, this number could decrease to < 3%. In the absence of widely metastatic disease, finding a primary tumor in the head and neck area is of significance for the following reasons: 1) postoperative radiation ports can be reduced, which can decrease early and delayed complications, including xerostomia; 2) surveillance for recurrent disease improves; and 3) such a finding helps with prognostic stratification. CT or MRI scans of the neck usually help to determine the extent of metastasis and levels at which lymph nodes are involved but rarely help with detection of the primary tumor. Panendoscopy with directed biopsies of the suspicious areas and inconspicuous areas (nasopharynx, base of the tongue, pyriform sinus, and tonsils) is the standard of care in many institutions. Small tumors of the palatine tonsil can originate in the deep crypts. The rich lymphatic drainage allows early spread to the lymph nodes. A superficial biopsy of the tonsil may not detect a small primary tumor. An ipsilateral tonsillectomy has been recommended for all patients presenting with cervical UPC. In 1 retrospective study50 of 87 patients who underwent a tonsillectomy as part of the workup for cervical lymph node metastasis presenting as UPC, 26% had a tonsil primary. In this group, 67 patients presented with a single cervical lymph node and 31% of these had a tonsil primary tumor. The involved lymph node was subdigastric in 38%, submandibular in 28%, and midjugulocarotid in 23%. The authors concluded that tonsillectomy detects the primary tumor and helps to avoid irradiation of the normal larynx in 26% of patients who have a cervical lymph node with an unknown primary tumor. Therefore, tonsillectomy should be performed in patients presenting with a single lymph node involving the subdigastric, midjugulocarotid, or submandibular areas or in patients with bilateral subdigastric adenopathy. Randall et al.51 suggested that in patients who have previously undergone tonsillectomy, all residual tissue, if any, should be removed, or a biopsy should be performed on the tonsilar fossa. In 2001, Koch et al.52 presented a case series of 41 patients with contralateral tonsil carcinoma. In four patients, disease was detected only after staging the tissue sample from a bilateral tonsillectomy. Bilateral tonsillectomy does not increase the risk of postoperative bleeding or other complications.
EVOLVING ROLE OF DNA MICROARRAY IN DIAGNOSIS OF UPC
If the primary tumor has not been detected, developing therapeutic strategies and determining the prognosis of the heterogeneous subsets of patients with UPC are quite challenging. The diagnostic yield with IHC tests and imaging studies is approximately 20–30%.1, 53 Molecular cancer diagnostics is a very promising way to improve this yield. Comprehensive gene expression databases that have become available for common malignancies may also be useful for the diagnosis of UPC. Su et al.54 described the use of large-scale RNA profiling and supervised machine learning algorithms to construct a first-generation molecular classification for the 11 malignancies that account for 70% of all cancer-related deaths. The predictor gene subsets included genes whose expression is specific for the tissue of origin as well as genes whose expression is elevated in patients with cancer. They used a set of 100 primary carcinomas (training set) from 10 common tumor types. A predictive algorithm was developed using 110 of the 9198 genes that were expressed minimally in these tumors. The algorithm then was tested against an additional 75 blinded samples including 12 metastatic samples. It accurately predicted the tumor of origin in > 90% of cases. Eleven of the 12 metastasis test cases were classified correctly. Ramaswamy et al.55 evaluated 218 tumor tissue specimens (14 common tumor types) and 90 normal tissue samples for oligonucleotide microarray gene expression analysis. They used the relative levels of expression of 16,063 genes and expressed sequence tags to develop a predictive support vector machine algorithm. The algorithm then was tested on an independent group of 54 tumor specimens. The overall prediction accuracy was 78%. Of the 54 independent tumor specimens tested, 8 were metastatic tumors. Of these eight tumor specimens, six were accurately identified with regard to origin, suggesting that tumors retain the markers of their tissue of origin throughout the process of metastatic evolution. However, to our knowledge there are not enough data available to fully support this statement. More studies are needed on metastatic samples, preferably from different sites, to compare the transcript profiles of primary tumors and their metastases. In addition, most of the tumor types that could not be classified accurately were moderately or poorly differentiated (high-grade) carcinomas.
Dennis et al.56 used public expression data from serial analysis of gene expression libraries and the literature regarding tumor markers and their differential expression to identify 61 candidate tumor markers (genes) whose expression patterns were predicted to be characteristic of the site of origin. They tested 11 of these genes using reverse transcriptase–polymerase chain reaction in primary adenocarcinoma samples (breast, ovary, stomach, pancreas, and lung), 7 (64%) of which were site restricted. They concluded that these tissue-specific and tissue-restricted tumor markers could be used to predict the likely site of primary tumors in patients with UPC.
Bhattacharjee et al.57 reported the results of gene expression analysis of 186 lung tumor specimens (including 127 primary lung adenocarcinomas and 12 carcinomas metastatic to the lung on the basis of clinical history). Hierarchical and probabilistic clustering of expression data defined biologically distinct subclasses of lung adenocarcinoma. The authors identified 1 distinct hierarchical cluster of 12 samples that were most likely metastases from the colon. Of the 10 samples in this group for which clinical and histopathologic data were available, only 7 samples had been diagnosed previously as colon metastases. Another adenocarcinoma, which expressed several breast-associated markers, was a known breast metastasis. One sample, which was not identified as a metastasis, showed markers associated with the liver. The authors concluded that clustering identified suspected metastases, some of which were not previously identified from medical history and histopathologic tests, suggesting a role for gene expression analysis in the diagnosis of lung tumors.
The application of gene expression studies to the diagnosis of UPC requires the availability of a training set of gene profiles of known primary tumors that represents the tumor types that are believed to be present in the study population. Several questions arise when the utility of gene expression analysis is considered for UPC. Because, by definition, the primary tumor site is not identifiable, the validation of site prediction in this setting can be a problem. To our knowledge, there are insufficient data concerning gene expression samples from metastatic sites of known primary tumors, especially poorly differentiated tumors. Can one assume that UPCs are no different from other metastatic lesions and that they retain the molecular signature of their site of origin? Does a poorly differentiated tumor that presents as a UPC have a similar genetic profile as its site of origin?
It is possible that incorporating the site prediction provided by microarrays into a larger statistical algorithm of IHC markers and patterns of metastasis may improve the overall accuracy of such a model. A well designed prospective study on gene expression profiles on a large set of metastatic samples is needed to determine the effectiveness of this approach, preferably from different sites to compare the transcript profiles of primary tumors and their metastases.
UPC is the name for a group of heterogeneous tumors in patients who present with metastatic disease without an identifiable primary tumor site. An aggressive diagnostic workup usually is not cost-effective. Clinicians need to know when a sophisticated pathologic or radiologic evaluation will be helpful and proceed accordingly. Our understanding of the role of molecular targets and DNA microarrays in UPC is evolving and will help us not only to better understand the pathogenesis of metastases but also to exploit novel therapeutic agents for this disease entity.