Fax: (617) 278–6921
Cytokeratin 20 expression identifies a subtype of pancreatic adenocarcinoma with decreased overall survival
Article first published online: 16 DEC 2005
Copyright © 2005 American Cancer Society
Volume 106, Issue 3, pages 693–702, 1 February 2006
How to Cite
Matros, E., Bailey, G., Clancy, T., Zinner, M., Ashley, S., Whang, E. and Redston, M. (2006), Cytokeratin 20 expression identifies a subtype of pancreatic adenocarcinoma with decreased overall survival. Cancer, 106: 693–702. doi: 10.1002/cncr.21609
- Issue published online: 20 JAN 2006
- Article first published online: 16 DEC 2005
- Manuscript Accepted: 19 JUL 2005
- Manuscript Revised: 10 JUL 2005
- Manuscript Received: 20 APR 2005
- pancreas cancer;
- cytokeratin 20;
- pancreatic intraepithelial neoplasia;
- tissue microarray;
Cytokeratins are markers of epithelial cell differentiation useful in determining histogenesis for malignancies with an unknown primary. Application of this principle to a single malignancy may identify cancer subtypes with altered developmental programs. Herein, we investigate the relevance of two widely used cytokeratins (CKs), 7 and 20, to subtype pancreas cancer and identify associations with clinical features.
A tissue microarray was constructed using tumor specimens from 103 patients who underwent resection for pancreatic adenocarcinoma with curative intent. A subset of resection specimens was evaluated for pancreatic intraepithelial neoplasia (PanIN) lesions. Tissues were immunostained by using specific anticytokeratin 7 and 20 monoclonal antibodies.
CK 7 and 20 expression was present in 96% and 63% cases of pancreatic adenocarcinoma, respectively. Ubiquitous CK 7 expression precluded further analysis. Tumoral CK 20 expression was not associated with any histopathologic parameter but correlated with worse prognosis when considered as either a dichotomous (P = 0.0098) or continuous (P = 0.007) variable. In a multivariate model, tumoral CK 20 expression remained a significant independent prognosticator. CK 20 expression was absent in all PanIN lesions from eight resection specimens in which the tumor component was negative for CK 20. In contrast, presence of tumoral CK 20 was highly concordant with its expression in corresponding PanINs.
CK 20 expression defines a subtype of pancreas cancer with important biologic properties. When present, CK 20 expression is an early event in pancreatic carcinogenesis identifiable in precursor lesions. Further studies to identify the underlying genetic changes associated with this altered developmental pathway are warranted. Cancer 2006. © 2005 American Cancer Society.
The mammalian cytoskeleton is composed of three different protein families: actin-containing microfilaments, tubulin-containing microtubules, and intermediate filaments. Intermediate filaments, comprising cytokeratins (CK), vimentin, desmin, neurofilaments, and glial fibrillary acidic filaments, are a diverse family of proteins linked to pathways of cellular differentiation.1, 2 CKs are the most complex members of the intermediate filament subfamily with expression being limited to epithelial cells.3–5 They can be divided into two subgroups: Type I (CKs 9–20), which are acidic and have lower molecular weight, and Type II (CKs 1–8), which have either a neutral or a basic charge.6 Association of members from each CK subgroup is necessary to form dimers, which, in turn, combine in an antiparallel fashion to form the basic building block of a keratin tetramer.7
As a diverse family of proteins, CKs have clinical importance because they serve as fingerprints for pathways of epithelial development and differentiation. Although they do not distinguish normal from malignant cells, the CK phenotype is generally retained during transformation from normal progenitor to malignant cells.4 Therefore, a current application of CKs is to determine the histogenesis of a tumor.8 Two of the more extensively studied CKs are 7 and 20 because of their complementary distribution in normal epithelium and malignant tissues: CK 7 is expressed in lung, ovary, endometrium, and breast, whereas CK 20 is found in gastrointestinal epithelium, urothelium, and Merkel cells.9–11
In addition to indicating the likely site of origin of a malignancy, the application of CKs to tumor classification has been recently extended to the identification of subtypes of a single malignancy. This is possible because all tumors from a single site may not uniformly recapitulate the CK profiles of their epithelial cell of origin. This is best exemplified in epithelial malignancies of the breast, where microarray-based transcriptional profiling studies have used hierarchical clustering to identify a distinct subgroup of tumors termed basal-like breast cancer.12, 13 Characterized by high histologic grade, expression of a variety of aggressive phenotypic markers, and a poor overall prognosis, these tumors are most readily identified by the aberrant expression of basal CKs 5 and 17.13, 14 These findings support the notion that variant CK profiles may be clinically important because they could signify important differences in biologically relevant developmental programs and differentiation pathways.
Compared to breast and many other cancers, pancreatic adenocarcinoma is a relatively homogeneous disease with uniformly poor clinical outcomes and a low frequency of either molecular or histologic variants.15 An absence of disease subtypes is further reflected in uniform chemotherapy regimens.16 While keratins 7, 8, 18, and 19 serve as markers of normal pancreatic ductal epithelium and cancerous cells, CK 20 is absent from normal fetal and adult pancreas yet displays some expression in 30–60% of carcinomas.11, 17–19 These findings raise the possibility of altered histogenetic developmental programming in a subset of pancreas cancers that could be associated with important biologic or clinical significance.
To further investigate this possibility, we studied the CK 7 and 20 immunophenotype of resected pancreas adenocarcinoma in a large uniform cohort by using the tissue array format. Our purposes were to use CK expression in primary tissues to subtype pancreas cancer and to identify associations with clinical and pathologic features of the disease. Furthermore, to determine the timing of aberrant CK 20 expression during pancreatic neoplastic progression, we evaluated CK 20 expression in pancreatic cancer precursor lesions termed pancreatic intraepithelial neoplasia (PanINs).
MATERIALS AND METHODS
Patient and Pathology Data
After institutional review board approval, pathology reports were reviewed to identify patients who underwent curative surgical resection for pancreatic adenocarcinoma between 1991 and 2002 at the Brigham and Women's Hospital. Clinical information was gathered retrospectively from patient records, and time of death was determined using publicly available social security death indices. Pathologic findings were reviewed for each case with staging and grading based on the sixth edition of the American Joint Committee on Cancer (AJCC) guidelines. Ampullary carcinomas (resected cancers limited to ampullae or with associated ampullary dysplasia) as well as other histologic variants, such as colloid cancer, adenosquamous cancer, medullary cancer, were excluded from the study. None of the specimens had associated intraductal papillary mucinous neoplasms (IPMNs) or mucinous cystic neoplasms. Complete clinical data and adequate tissue for immunohistochemical analysis were available for 103 patients.
Tissue Microarray Construction
Original formalin-fixed, paraffin-embedded specimens were used to construct a pancreatic adenocarcinoma tissue microarray (TMA). Hematoxylin and eosin (H & E)-stained sections from each specimen were reviewed to identify appropriate normal and tumor regions. Two or three (0.6 mm in diameter) cylindrical tissue cores from both the tumor and normal adjacent pancreas were punched for each specimen using an automated tissue arrayer (Beecher Instruments, Sun Prarie, Wisconsin). Cores were subsequently transferred to a recipient block. Five μm sections were cut from the recipient block and stained with H & E to confirm the presence of tumor or normal tissue within each core.
For a subset of the 103 specimens, whole sections were reviewed to identify PanIN lesions in the nontumorous pancreas of resection specimens. PanINs were graded and differentiated from “cancerization” according to previously published criteria.20, 21 Whole sections were cut from representative regions containing PanIN Grades I–III and immunostained for CK 20 according to the protocol detailed below. This analysis was performed based on results from the TMA for 14 specimens, 8 cases with absent CK 20 expression and 6 cases with moderate to strong CK 20 expression in the primary tumor.
Immunohistochemistry and Scoring
Whole section and TMA slides were deparaffinized and rehydrated through a series of xylenes and alcohols. Pretreatment of tissue sections with Proteinase K (DAKO, Carpinteria, California) was performed for 10 minutes at 37 °C and subsequently processed using the Envision System of DAKO. Following quenching of endogenous peroxidase activity and blocking of nonspecific binding, the slides were incubated with either monoclonal antihuman CK 7 or CK 20 antibody (DAKO) in 1:1000 and 1:50 dilutions, respectively, at room temperature for 40 minutes. Sections were then washed, and a secondary Envision antimouse antibody (DAKO) was applied for 30 minutes at 37 °C. Sections were then washed and reacted with diaminobenzidine and hydrogen peroxide. Slides were counterstained with hematoxylin, dehydrated, and mounted. Adjacent cores containing nonneoplastic tissues and pancreatic adenocarcinoma served as positive controls for expression of CK 7 and 20. Normal serum was used in place of primary antibody as a negative control.
Whole section and TMA slides were scored by two independent observers blinded to clinical and pathologic data (M.R. and E.M.). CK 7 and 20 expression was based on intensity of cytoplasmic staining as follows: 0: no staining; 1: weak intensity in any tumor cells with no moderate or strong intensity; 2: moderate intensity in any tumor cells with up to 10% of cells having strong intensity; 3: strong intensity in greater than 10% of cells. This scoring system was chosen because most cases were found to have relatively uniform staining of a given intensity across the majority of tumor cells. For statistical analyses, expression was dichotomized into a CK 20 low group (score, 0 or 1) and a CK 20 high group (score, 2 or 3).
The chi-square test was used to determine associations between CK 20 expression and individual clinicopathologic variables. Survival time was defined as the time from resection to death or censoring based on the date of last contact. Deaths within 30 days of surgery were excluded from analysis. Survival curves were plotted by using the Kaplan–Meier method and compared using the log-rank test. Multivariate analysis was performed using a Cox proportional hazards model. All statistical tests were two-tailed with a P ≤ 0.05 considered significant.
As shown in Table 1, the cohort included 103 patients with clinical and pathologic features representative of pancreatic adenocarcinoma. The average age at surgery was 64 years with almost equal representation of males and females (Table 1). Median tumor size was 2.50 cm with a range from 0.10 cm to 8.40 cm. Sixty percent of patients had lymph node metastasis at diagnosis, and most tumors were graded as moderate to poorly differentiated. Based on current AJCC criteria, the majority of patients had Stage II disease. The median overall postoperative survival time was 498 days with 1- and 5-year actuarial survival rates of 70% and 7%, respectively.
|Lymph node status|
|Tumor size, pathologic|
Pattern of CK 7 and 20 Immunohistochemical Expression in Normal Pancreas and Pancreatic Adenocarcinoma
CK 7 and 20 immunohistochemical expression were evaluated in ductal epithelium from nonneoplastic adjacent pancreas in resection specimens. CK 7 expression was present in all normal epithelium samples, whereas CK 20 was always absent (Fig. 1A). Pancreatic adenocarcinoma cores demonstrated strong CK 7 labeling in 96% of cases with absent expression in only 4 of 103 (Fig. 1B). In contrast, CK 20 tumoral staining had variable intensity, 37%: no staining; 43%: weak; 15%: moderate; and 5%: strong (Fig. 1C–F). When CK expression was dichotomized into groups of either present (any staining) or absent, the CK 7/20 immunoprofile was 2%: −/−; 2%: −/+; 35%: +/−; and 61% +/+. Due to the ubiquitous expression of CK 7 in evaluated normal pancreas and tumors, meaningful statistical analyses with CK 7 were unable to be performed; therefore, remaining analyses focused on CK 20.
Association of CK 20 Expression with Clinical and Pathologic Variables
To identify associations of CK 20 expression with clinical and pathologic variables of the cohort, CK 20 expression was divided into two groups based on intensity of immunohistochemical staining, CK 20 low (absent or weak staining) and CK 20 high (moderate or strong staining). No significant correlations between tumoral CK 20 expression and any covariate including age, gender, tumor size, lymphovascular invasion (LVI), perineural invasion (PNI), microscopic margin, nodal metastasis, tumor location, tumor stage, tumor differentiation, or treatment with chemotherapy, were identified (Table 2).
|Variable||CK 20 Group||P value|
|Low N = 82||High N = 21|
|< 2.5 cm||42||11|
|> 2.5 cm||40||10|
When overall postoperative survival was stratified by CK 20 group, survival was significantly longer in patients whose tumors had low CK 20 expression relative to those with high levels of CK 20 (P = 0.0098). These results are represented graphically in Kaplan–Meier curves (Fig. 2A). If CK 20 expression is maintained as a continuous variable based on intensity of immunohistochemical staining, the Kaplan–Meier plot (Fig. 2B) demonstrates a progressively worse survival for increasing levels of CK 20 expression (P = 0.007).
To assess whether or not CK 20 expression was an independent predictor of overall postoperative survival, a Cox proportional hazards model was created in a forward fashion by including only those covariates that showed a statistically significant (threshold for inclusion, P ≤ 0.05) relation with postoperative survival. Univariate analysis demonstrated that location of tumor in the pancreatic tail, increasing tumor size, presence of lymph node metastasis, poor tumor differentiation, and CK 20 expression as either a continuous or dichotomous variable were predictors of poorer survival. Multivariate analysis (Table 3) demonstrated that after correction for confounding variables, CK 20 expression remained a significant prognosticator when either a continuous or a dichotomous covariate (P = 0.016 and 0.020, respectively).
|Hazard||95% CI||P value||Hazard||95% CI||P value|
|Tumor located in head of pancreas||0.40||0.17–0.95||0.037||0.38||0.15–0.96||0.040|
|Increasing tumor size||1.21||1.04–1.42||0.017||1.19||1.01–1.41||0.045|
|Presence of lymph node metastasis||1.86||1.13–3.05||0.015||2.08||1.15–3.75||0.015|
|Advanced tumor stage (T3/T4 vs. T1/T2)||1.81||0.85–3.86||0.122||—||—||—|
|Poor tumor differentiation||1.82||1.20–2.76||0.005||1.95||1.17–3.23||0.010|
|Presence of PNI||0.98||0.61–1.56||0.929||—||—||—|
|Presence of LVI||1.48||0.90–2.43||0.120||—||—||—|
|Presence of tumor at microscopic margin||1.45||0.90–2.33||0.122||—||—||—|
|CK 20 expression, continuous||2.67||1.12–2.13||0.008||2.41||1.09–2.25||0.016|
|CK 20 expression, dichotomous||2.14||1.19–3.89||0.012||2.15||1.13–4.12||0.020|
CK 20 Expression in PanIN Lesions
Presence of CK 20 in a subset of tumor tissues suggested that CK 20 might also be expressed in the precursor PanINs. To test this hypothesis, we selected subgroups of tumors with the highest labeling for CK 20 and with absent CK 20 to determine if CK 20 expression was present in PanINs. No CK 20 expression was identified in any of the PanINs present in eight resection specimens that had absent CK 20 expression in the cancer component (Fig. 3). In contrast, CK 20 positive PanINs were present in five of six resection specimens that had moderate to strong levels of CK 20 expression in the associated cancer component (Fig. 4). In general, PanINs stained at the same intensity as tumors from the corresponding specimen. CK 20 was expressed less frequently in Grade I PanINs, 3 of 5 (60%), than in high-grade PanINs (Grades II and III) 7 of 8 (88%). Whole sections of CK-20–negative specimens also allowed a more complete evaluation of a subset of the specimens. Importantly, no focal CK 20 positivity was seen in any of these cases.
Cancers of the pancreas are fourth on the list of cancers frequently occurring in men and women in the United States.15 Over 90% of patients diagnosed with the disease present at later stages, thus making curative resection impossible. Of those patients that undergo curative resection, less than 10% are alive at 5 years.22 This uniformly poor clinical outcome is accompanied by relatively homogeneous pathologic features, characterized by moderately to poorly differentiated adenocarcinomas with a prominent associated desmoplastic stromal reaction. Efforts toward identifying the molecular underpinnings of pancreas cancer have identified a number of genetic alterations, such as mutations in K-ras, SMAD4, and p16, that are acquired during the progression through PanINs to invasivecarcinoma.23, 24 It has also been suggested that some of these changes may aid diagnosis and prognostication of pancreatic cancer. Herein, we report for the first time that one of the cytokeratins, CK 20, is useful to identify subtypes of pancreas cancer with distinct clinical features.
Subclassification of human malignancies has been the foundation of progress in understanding cancer behavior and response to conventional and novel treatments. In some cases, these subclassifications link a single gene alteration to targeted therapeutics, such as the use of trastuzumab in breast cancers with HER2 amplifications and gefitinib in nonsmall cell lung cancer with EGFR mutations.25–27 In other cases, the subclassification broadly separates cancers with widely divergent molecular profiles, such as the dichotomy between mismatch repair proficient and deficient colorectal cancers, the latter revealing widespread tumor DNA microsatellite instability, characteristic clinical pathologic features, and better overall patient survival.28
Transcriptional profiling studies have been undertaken in pancreatic cancer, primarily with the goal of identifying novel tumor-specific markers, and clinically relevant subclassifications have not yet been identified.29, 30 Investigations in other tumor systems have been more successful, including identification of so-called basal-like subtype of breast cancer. Characterized by high pathologic grade, brisk mitotic activity, estrogen and progesterone receptor negativity, and poor patient outcome, this subset of cancers is defined by expression of CK 17 and the basal CKs 5 and 6.12, 13 The frequent association with germline BRCA1 mutations further supports the contention that these tumors progress along a distinct molecular pathway and gives some insight into the molecular basis of this pathway.31 There are other examples of keratin profiles that play a role in identifying cancer subtypes, such as the use of high molecular weight cytokeratins to distinguish between follicular variant papillary carcinomas and follicular carcinomas of the thyroid.32
Our study lends support to the hypothesis that cytokeratin immunophenotypes may identify subsets of cancer with distinct clinical outcomes despite similar pathologic characteristics. We find that relative to patients with tumors of high CK 20 expression, patients who have tumors with low CK 20 expression have a significantly longer overall postoperative survival, and that this relationship persists even after adjusting for other confounding variables. These findings are robust as the relation is maintained whether CK 20 expression is dichotomized or maintained as a continuous categorical variable. In fact, none of the patients whose tumors expressed moderate or strong levels of CK 20 survived beyond 26 months. This convincing stratification of clinical outcome is somewhat surprising because CK 20 expression has historically been used primarily as a marker of site of origin for unknown primaries. In colorectal cancer, however, CK 20 negativity predicts tumors with mismatch repair deficiency, indicating a possible role in demarcating major clinically relevant pathogenetic subgroups.33 Overall, these data strongly suggest that tumoral CK 20 expression not only indicates a pancreatic tumor subtype with aberrant epithelial differentiation and divergent natural history, but its expression may also be indicative of a distinct pathogenetic pathway of neoplastic progression.
CK 20 is associated with the mammalian gastrointestinal tract because of its presence in mature enterocytes and goblet cells. Within the colon, CK 20 expression is limited to more terminally differentiated cells as opposed to less differentiated epithelial cells at crypt bases.1 Yet despite this fact, CK 20 expression remains highly conserved in colorectal cancer and is largely unaffected by underlying genetic alterations associated with malignant transformation. One important exception is the loss of CK 20 expression in some mismatch repair deficient colorectal cancers. In sharp contrast, CK 20 is not expressed in either normal fetal or adult pancreas tissue but is present in a significant percentage of pancreas cancers.17 Therefore, CK 20 expression must be aberrantly acquired during progression of a subset of pancreatic cancers. Although there are no apparent differences in histologic morphology of CK 20 positive pancreatic cancers, this shift in cytokeratin profile may indicate an underlying alteration in epithelial differentiation programming. Investigations of normal tissue samples have demonstrated widespread variations in gene expression profiles across different types of mature epithelia. It is thus conceivable that aberrant CK 20 expression could be a marker of broad differences in molecular pathogenesis.
We also found CK 20 expression in precursor PanIN lesions adjacent to CK 20 positive carcinomas, suggesting that this aberrant profile is most commonly acquired before development of invasion. CK 20 expression was present across a spectrum of PanIN lesions, including some PanIN I lesions, supporting progression of low- to high-grade PanINs through invasive carcinoma. Therefore, the switch to a CK-20–positive epithelial differentiation pathway seems to occur very early in neoplastic progression and is maintained through to development of invasion. In contrast, there was no CK-20–positive PanIN in cases with CK-20–negative carcinomas. Recent investigations of intraductal papillary mucinous neoplasms have revealed a similar divergence in differentiation and progression of noninvasive neoplasms. In this setting, so-called intestinal-type IPMNs are much more likely to be CK 20 positive and also more likely to be associated with mucinous adenocarcinomas when malignant progression is present.34 The only other setting in which epithelial differentiation and cytokeratin profiles have been extensively examined is in the process of acinar-ductal metaplasia occurring in chronic pancreatitis.18 In our series, although CK 7 expression was markedly increased in nonneoplastic lobular tissue, we saw no evidence of aberrant CK 20 expression in nonneoplastic pancreatic ducts (data not shown). These findings suggest that transition to CK 20 positivity occurs at a PanIN stage and support the notion that CK-20–positive pancreatic duct epithelial neoplasms could be progressing along a distinct carcinogenic pathway. In the future, as pancreatic cancer research evolves toward identification and management of high-risk noninvasive epithelial lesions, further studies will be able to investigate the role of CK 20 expression in predicting risks of neoplastic progression.
It remains unclear if deregulated CK 20 expression represents a specific functionally important biologic alteration in pancreas tumors or is simply a downstream target of other genetic alterations or activated transcriptional pathways. There is strong evidence for a pathogenic role of other CKs in nonmalignant diseases of skin and liver, and increasing basic research and clinical data to suggest a role in tumorigenesis.35, 36 In vitro studies demonstrate that cell lines transfected with either CK 8 or 18 have higher migratory and invasive ability relative to controls.37, 38 In vivo experiments with CK 8 overexpressing transgenic mice identified alterations of the exocrine pancreas including cellular dysplasia, dedifferentiation of acinar tissue, and ductal metaplasia.39 Deregulated protein expression of CK 18 in primary breast tumors is associated with advanced pathology and worse clinical outcome.40 Additional studies will be required in pancreatic neoplasia model systems to determine whether CK 20 expression has a direct biologic role in malignant transformation
Despite recent advances in delineating the underlying molecular genetic alterations in pancreatic cancer, there has been little progress in abating the aggressive natural history and poor overall survival. One continuing limitation is the general approach of considering pancreatic cancer as a single disease. In contrast, advances with other human malignancies have led to subclassifications that are highly relevant for treatment algorithms and etiologic insights. Using an outcome-linked tissue microarray of resected pancreatic cancer, we have found that CK 20 expression predicts reduced survival. The acquisition and maintenance of this altered epithelial differentiation from early noninvasive PanINs raises the possibility that CK 20 positivity may demarcate a distinct pathway of carcinogenesis. These findings have implications for the use of CK 20 as a novel prognostic marker in the clinical management of pancreatic cancer and could lead to new insights into etiology and pathogenesis of this deadly disease.
The authors acknowledge Mei Zheng for her help with immunostaining.
- 7Chemical cross-linking indicates a staggered and antiparallel protofilament of desmin intermediate filaments and characterizes one higher-level complex between protofilaments. Eur J Biochem. 1992; 206: 841–852., , .
- 21The pancreatic duct and its arteriovenous relationship: an underutilized aid in the diagnosis and distinction of pancreatic adenocarcinoma from pancreatic intraepithelial neoplasia. A study of 126 pancreatectomy specimens. Am J Surg Pathol. 2004; 28: 613–620., .
- 22Surgical management of early-stage pancreatic cancer. Cancer Control. 2004; 11: 23–31., , , .