The diagnosis of the metastatic carcinoma of unknown origin can be very difficult. The determination of the primary site of the metastasis is a challenge to both oncologists and pathologists, having potentially important clinical and therapeutic consequences [1–3]. In the setting of carcinomas of unknown primary, clinicopathological correlation and a panel of standard immunostains help define the primary site, and direct appropriate treatment [4, 5].
Cytokeratins are group of approximately 20 proteins that consist of a type of intermediate filament and are differentially expressed in epithelia of various sites. The cytokeratins most often used are CK7 and CK20 [7–10]. CK7 is found in the glandular epithelium and epithelial tumors of lung, ovary, endometrium and breast, but is not found in GI epithelium. Conversely, CK20 is expressed principally in the normal glands and epithelial tumors of the GI tract, urothelium, and Merkel cells. The cytokeratin 7/20 profile of a particular tumor has proved to be a useful aid in differential diagnosis of carcinomas, since primary and metastatic tumors tend to retain the cytokeratin profiles of the epithelium from which they arise . In his review article, Tot summarized the results of 29 studies containing more than 3500 reported cases of adenocarcinomas stained with CK20 and CK7. This review stated that metastatic colorectal, gastric and pancreatic adenocarcinomas have similar CK7 and CK20 staining ratios as their respective primary tumors. Only gastric adenocarcinomas showed statistically significant differences in CK20 expression when the primary and metastatic locations were compared .
Normal epithelium of the small bowel, appendix and colorectum, and adenocarcinomas from these sites, are almost consistently CK7-/CK20+, helping to distinguish these adenocarcinomas from adenocarcinomas of many other primary sites [9–15]. The CK7-/CK20+ pattern was identified in 65% to 95% of the colorectal adenocarcinomas in different series [11, 12, 20–23, 31]. On the other hand approximately one third of gastric and less than 10% of pancreatic adenocarcinomas also show this pattern [11, 12, 23]. The CK7-/CK20+ immunophenotype was expressed by 75 of 118 (64%) colorectal and 3 of 59 (5%) gastric tumors and was not observed in any pancreatic adenocarcinomas in the present study. Therefore, it has been presumed that CK7 is not typically expressed by colonic epithelial tumors. Interestingly, several reports have described CK7 expression in colorectal adenocarcinoma, with expression ranging from 5% to 74% [11, 12, 22, 23, 31]. The reasons for this discrepancy are unclear. However, this may be the result of differences in the studied population or the interpretive criteria that was used. In our study, CK7 expression was detected in 22% (26/118) of colorectal adenocarcinomas.
In comparison with the CK7-/CK20+ immunoprofile, the CK7+/CK20+ immunoprofile is commonly present in urothelial carcinomas, gastric carcinomas and tumors of the pancreatobiliary tract [11, 12, 15]. Gastric adenocarcinomas are the most heterogeneous subgroup of carcinomas with respect to their CK7/CK20 immunophenotype. While most gastric adenocarcinomas are CK20+, they may or may not be CK7+ [11, 12, 23]. The results of CK7/CK20 immunohistochemistry for cholangiocarcinomas, gall bladder carcinoma and pancreatic ductal adenocarcinoma are conflicting. While all studies have found CK7 immunopositivity in these tumours, many studies have found the majority are CK20- [40, 41], while others have found the majority to be CK20+ [11, 12]. In the present study the largest proportion of gastric carcinomas was of the CK7+/CK20+ phenotype (48%), and a substantial proportion was of the CK7+/CK20- phenotype (32%). CK7+/CK20- immunoprofile was the most common pattern, accounting for 75% of pancreatic adenocarcinomas. The CK7+/CK20+ immunophenotype was expressed in 20% of colon, 48% of gastric and 22% of pancreatic adenocarcinomas, which was not helpful in the differential diagnosis. However, CK20 reactivity was diffuse (more than 50% of cells were positive) in the majority of colorectal carcinoma cases and mainly focal ( < 50% ofcells were positive) in gastric and pancreatic adenocarcinomas as in previous studies [22, 23, 31, 41].
Since, occasional colorectal carcinomas may show significant CK7 expression and conversely, expression of CK20 may be seen in a variety of non-colorectal adenocarcinomas, there is interest in the development of new and more specific markers of intestinal differentiation. Human CDX2 protein is a member of the homeobox genes that encodes an intestine-specific transcription factor. This protein regulates intestinal development and is expressed in the nuclei of epithelial cells throughout the intestinal tract in embryonic and postnatal life [25–27]. Expression of CDX2 mRNA has been shown to be highly restricted to intestinal epithelium . The sensitivity and specificity of antibodies to CDX2 protein as a marker of colonic adenocarcinoma has been recently evaluated in various studies with reported sensitivity and specificity of greater than 90% [28–33]. Werling et al  examined CDX2 expression across 476 samples of human tumors and concluded that it is an excellent marker of adenocarcinomas arising in the GI tract, particularly the duodenum and colon. These authors reported that high levels (> 75% positive cells) of CDX2 expression were found almost exclusively in adenocarcinomas of the colorectum, and intermediate levels (26%-75% positive cells) of immunostaining were found in many adenocarcinomas arising elsewhere in the GI tract. They also demonstrated that primary and metastatic colorectal carcinomas showed remarkably similar scoring patterns. All primary and 25 of 26 metastatic colonic adenocarcinomas showed high levels of CDX2 expression (2+ or 3+) in this study. In another study, Kaimaktchiev et al  observed a greater than 80% concordance for CDX2 expression in the analysis of matched primary and lymph node metastases. In addition, all 17 colorectal metastases examined by whole sections were CDX2 positive in this study. Using tissue microarrays, Moskaluk et al  analyzed CDX2 staining in 745 samples of human cancer and arrived at similar conclusions. Barbareschi et al  compared CDX2 expression in primary and metastatic tumors found in the lung and concluded that this marker is highly selective for tumors originating from the colon and rectum, but also stains metastases from the stomach and ovary. In our study, CDX2 was expressed in 114 of 118 (97%) colorectal, 36 of 59 (61%) gastric, and 5 of 32(16%) pancreatic adenocarcinomas. The majority of cases (92/114, 81%) demonstrated strong and diffuse immunostaining in more than 50% of cells in colorectal tumors. Among the CDX2 positive gastric carcinomas (36/59), reactivity was focal in 22 cases (22/36, 61%). Among the 32 cases of pancreatic adenocarcinoma, only 5 cases were focally positive for CDX2.
Among colorectal adenocarcinomas, the relationship between tumor grade and CDX2 staining has been controversial. Hinoi et al  demonstrated that a rare subset of poorly differentiated colonic carcinomas termed large cell minimally differentiated carcinoma or medullary carcinoma are characterized by microsatellite instability and loss of CDX2 expression. Kaimaktchiev et al  recently studied tissue microarray samples of 1109 colorectal adenocarcinomas and found a lack of CDX2 reactivity in 14 (28%) of 50 poorly differentiated tumors. They concluded that CDX2 expression decreases with tumor differentiation. Other series, however, failed to find a strong correlation between CDX2 expression and the level of differentiation in colorectal adenocarcinomas. In the study of Werling et al , 74 of 75 colonic carcinomas showed high levels of CDX2 expression (2+ or 3+). Although several high-grade tumors showed scores of 2+ (26%-75% positive cells) compared with scores of 3+ (> 75% positive cells) that were observed in all well-differentiated carcinomas, the authors concluded that the expression of CDX2 did not appear to correlate with the level of tumor differentiation. Saad et al  also showed that CDX2 expression was not influenced by tumor grade. In this study, there was no significant association between CDX2 expression and tumor differentiation in colorectal carcinomas (98% of low grade tumors and 91% of high grade tumors were positive for CDX2). Our semiquantitative scoring system did not, however, take into account the intensity of immunostaining, but focused exclusively on the fraction of cells positively immunostained. It is likely that other methods of assessing absolute levels of CDX2 expression might show differences related to tumor differentiation.
CDX2 represents the latest in a series of transcription factors that have found important applications in diagnostic surgical pathology as highly specific and sensitive markers of specific cell and tumor types. Nuclear transcription factors have several distinct advantages over cytoplasmic ''differentiation'' markers: (1) transcription factors generally yield an ''all or none'' signal, with most of the positive cases containing positive signal in > 90% of the tumor cell population; (2) given the nuclear localization of the signal, it is much less likely to be confused with biotin or other sources of false positive cytoplasmic signals; and (3) lack of association between the levels of expression of nuclear transcription factors and the state of differentiation of the tumor . For example, in the study described here, 114 of 118 cases of colonic adenocarcinoma were CDX2-positive, independent of tumor grade.
Expression of CDX2 in tumors other than colorectal carcinoma has been previously reported [28, 29, 32–35, 40, 41]. CDX2 expression has been documented in gastric adenocarcinoma by several different groups [28, 32, 43–46]. Werling et al  reported scores of 2+ (26%-75% positive cells) and 3+ (> 75% positive cells) positivity in 17 (70%) of 24 cases. These authors also reported that no association between any histological subtypes within pancreatic or gastric tumors and CDX2 expression could be discerned. In the study of Kaimaktchiev et al , CDX2 staining was observed in gastric adenocarcinomas (16 of 71), more commonly in the intestinal-type than in the diffuse-type (28.9 vs 11.5%). Our results are entirely consistent with these studies in that CDX2 staining was observed in 61% of gastric adenocarcinomas and significantly favored in the intestinal-type tumors over the diffuse variants (77% versus 45%). Park et al  reported that, CDX2 expression was decreased in early gastric cancers, when compared with dysplasia, and was even more reduced in advanced cancers. Similarly, Kim et al  reported lesser CDX2 expression in early gastric cancers compared to advanced tumors. Liu et al  also showed that CDX2 expression is progressively decreased in gastric intestinal metaplasia, dysplasia, and cancer. We didn't find any association between CDX2 expression and stage of gastric adenocarcinomas. As for CDX2 expression in pancreatic ductal adenocarcinomas, there appears to be somewhat less agreement in the literature. Werling et al  reported scores of 2+ (26%-75% positive cells) and 3+ (> 75% positive cells) positivity in 7 (32%) of 22 cases, Moskaluk et al  found 1+ ( < 25% positive cells) expression in 8 (33%) of 24 cases, and in the series of Chu et al , CDX2 reacted with 10 (22%) of 46 cases. In contrast, Kaimaktchiev et al  found that only 3 of 70 cases were positive for this marker. In the present study, among the 32 cases of pancreatic adenocarcinomas, only 5 cases were focally positive for CDX2.
Based on the studies mentioned above CDX2 expression alone does not reliably distinguish between colorectal adenocarcinomas and adenocarcinomas arising elsewhere in the GI tract, particularly pancreatobiliary and gastric adenocarcinomas, although the sensitivityof CDX2 for colorectal cancer is significantly higher than for these latter tumors. Qualitatively, focal and weak CDX2 expression in a given tumor favors extra-intestinal origin whereas uniform intense expression favors intestinal origin. In comparison with the CK7-/CK20+ immunoprofile Tot  found that CK7-/CK20+ expression pattern was more specific for colonic adenocarcinoma metastases than CDX2 alone (98.7% vs 90%), but less sensitive (79.5% vs. 84%). We also evaluated the sensitivity, specificity, positive predictive value, and negative predictive value of CDX2 expression and CK7-/CK20+ immunophenotype to differentiate colorectal adenocarcinoma from pancreatic and gastric adenocarcinomas. Determining the CK7/CK20 phenotype proved to be more specific in differentiating colorectal adenocarcinoma from pancreatic and gastric adenocarcinomas (specificity 96.7%) than the expression of CDX2 was. The CK7-/CK20+ phenotype had a superior positive predictive value (96.2%) in these circumstances. CDX2 expression at both cut-off levels (> 5% and > 50%) had a higher sensitivity (96.6% and 78%) and higher negative predictive value (92.6% and 74.8%) than the CK phenotype. The specificity of CDX2 expression did not reach the level of specificity of CK7/CK20 phenotype at a > 50% level, either (84.6%).