Main

Neoplasms with rhabdoid features occur in diverse organs. Their hallmark is ‘rhabdoid cells’ containing eosinophilic paranuclear cytoplasmic filamentous inclusions of intermediate filaments that displace the nucleus to the cell periphery.1, 2 Usually, these neoplasms, most of which occur in childhood, are also anaplastic, co-express cytokeratin and vimentin and are highly aggressive.3 In the pancreas, neoplasms with rhabdoid features have been reported under different names (pleomorphic adenocarcinoma, pleomorphic carcinoma, pleomorphic giant cell carcinoma, sarcomatoid carcinoma, anaplastic carcinoma),4, 5, 6, 7, 8 but have generally been included in the category of undifferentiated carcinomas, that are considered to be variants of ductal adenocarcinoma.9, 10, 11, 12 The term ‘rhabdoid’ was first applied to a pancreatic tumor by Nishihara in 1997.13 Since then, six other cases using this term have been reported.14, 15, 16, 17, 18, 19 The fact that only some of the undifferentiated carcinomas exhibit rhabdoid cells, and the notion that many reports describe not only neoplasms with anaplastic giant cell features, but also carcinomas with a sarcomatoid spindle cell appearance are strong arguments that the category of undifferentiated carcinomas of the pancreas includes a spectrum of morphologies that are probably of heterogeneous nature.

SMARCB1 is a member of the chromatin remodeling complex SWI/SNF located at chromosome 22q11.2, which probably functions as a tumor-suppressor gene.20 Its gene product SMARCB1 (INI-1) is ubiquitously expressed in all normal human tissue types and in all neoplasms with intact SMARCB1 locus.20 In several tumor types, rhabdoid cell morphology has been associated with complete loss of nuclear SMARCB1 as a result of deletions/mutations involving the SMARCB1 locus.21, 22 However, the concept of ‘rhabdoid neoplasms’ as uniform and specific entities has been challenged by the occurrence of similar cell types in otherwise differentiated neoplasms (‘composite rhabdoid tumors’), which only rarely showed chromosome 22q (SMARCB1) alterations.23, 24 The increasing availability of more sensitive molecular techniques, and improved immunohistochemical identification of SMARCB1 confirmed the existence of SMARCB1-deficient ‘true rhabdoid’ variants of different epithelial neoplasms in the pancreas,17 female genital tract,25 gastrointestinal tract,26 upper aerodigestive tract,27 and other rare sites. Currently, the main entities included under the umbrella of ‘SMARCB1-deficient neoplasms’ are pediatric atypical teratoid/rhabdoid tumors, malignant rhabdoid tumors of the kidney and of extrarenal soft tissue sites, proximal- and distal-type epithelioid sarcoma, renal medullary carcinoma, epithelioid malignant peripheral nerve sheath tumors, and subsets of extra-skeletal myxoid chondrosarcoma/myoepithelial neoplasms of soft tissue.28

Studying a series of undifferentiated carcinomas of the pancreas that were predominantly (>50%) composed of rhabdoid cells, we recognized that their morphology was heterogeneous and allowed the separation of two subtypes: the first with a pleomorphic and the second with a monomorphic anaplastic cell pattern. As it has been recently shown that pancreatic undifferentiated carcinomas frequently show KRAS amplifications,29 we aimed to investigate how the molecular changes in these neoplasms (ie, KRAS and SMARCB1 alterations) relate to their heterogeneous morphology. We herein report our experience with 14 pancreatic undifferentiated rhabdoid carcinomas, which we analyzed for SMARCB1 expression and KRAS mutations and/or amplifications. In addition, we performed a comprehensive review of the old and recent literature on comparable neoplasms in order to see how the tumors of our series fit into the clinicopathologic spectrum of this group of pancreatic carcinomas.

Materials and methods

Formalin-fixed and paraffin-embedded tissue blocks from 14 pancreatic tumors were retrieved from our routine surgical pathology files and from the consultation files of two of the authors (GK and AA). Several consultation cases (dating back to 1980) lacked detailed follow-up data. Included were nonendocrine neoplasms predominantly composed (>50%) of highly atypical tumor cells with eosinophilic filamentous paranuclear cytoplasmic (‘rhabdoid’) inclusions. Five of the fourteen tumors also showed a glandular component. Tumors containing spindle cells and squamoid cells as a dominant component or osteoclastic giant cells were excluded. Tissue sections were stained with hematoxylin and eosin and periodic acid–Schiff. Immunohistochemistry was performed on freshly cut 3-μm paraffin sections using a fully automated slide preparation system ‘Benchmark XT System’ (Ventana Medical Systems, Tucson, Arizona, USA). All reagents and buffers were retrieved from Ventana Medical Systems. The antibodies used in this study, epitope retrieval conditions, incubation time, secondary antibody information (catalog number, company, and dilution, incubation time) are given in Table 1. Antigen visualization was done using a Ultra View DAB-Kit (Ventana, catalog 05269806001). Positive and negative controls were used throughout. Loss of SMARCB1 expression was recorded, when the tumor nuclei showed ‘clean’ negative staining as opposed to unequivocal nuclear staining of the adjacent inflammatory, endothelial, stromal, and normal pancreatic cells. If only isolated tumor cells were SMARCB1 negative or showed weak positive staining, SMARCB1 expression was recorded as intact.

Table 1 Sources of antibodies and conditions of immunohistochemical staining
Full size table

Molecular Analysis and Assessment of Microsatellite Status

Mutational analysis of KRAS exon 2 was performed at the institutional molecular diagnostics laboratory using standardized protocols. In brief, mutation hotspots in exons 2 (codons 12, 13, 19) and 3 (codon 61) of the KRAS gene were analyzed using a single-nucleotide primer extension assay (SNaPshot) as described previously.30 KRAS amplification was assessed using a dual color probe (ZytoLight® SPEC KRAS/CEN12 Dual color Probe, Zytovision, Bremerhaven, Germany) according to the instructions of the manufacturer. A KRAS/CEP12 ratio of >2 was considered as amplification. Other alterations (monosomy, polysomy) were recorded too. The expression of the mismatch repair proteins was assessed in all cases using immunohistochemical staining for MLH1, PMS2, MSH2, and MSH6 (see Table 1 for detailed antibody sources and staining conditions). Unequivocal nuclear staining in ≥10% of tumor cells was considered retained (normal) expression. Normal mucosa, endothelial cells, and background inflammatory cells served as internal controls. In addition, microsatellite instability status was evaluated in two of the cases by PCR on tumor DNA extracted from formalin-fixed paraffin-embedded tissue samples using the methods described previously.31 One case was analyzed by electron microscopy using paraffin-embedded tissue.

Literature Review

For comparison of clinicopathological, demographic, and prognostic data with those of our tumor series, we performed a review of the MEDLINE literature dating back to 1968 using the keywords ‘undifferentiated carcinoma’, anaplastic carcinoma,’ pleomorphic carcinoma,’ ‘pleomorphic adenocarcinoma,’ ‘sarcomatoid carcinoma,’ ‘giant cell carcinoma,’ ‘rhabdoid carcinoma,’ and ‘carcinoma with rhabdoid features’. Included were only those cases whose clinicopathological and demographic data (age, gender, tumor site, size, metastases, treatment and outcome, presence or absence of well-differentiated component, as listed in Table 4) could be traced back to the individual patients and whose tumor histology fits the above-defined criteria. Cases with missing single criteria such as tumor size or site were included as well. Cases with rhabdoid phenotype seen only at metastatic sites were excluded.

Results

Clinical and Demographic Features

The main clinicopathological features of the 14 patients are summarized in Table 2. Patients were seven males and seven females aged 44–96 years (mean age: 65; median: 65 years). Tumors were localized in the pancreas head (9), body (1), tail (3), and unspecified part of pancreas (1). Presenting symptoms included nonspecific abdominal symptoms, weight loss, and deteriorating general condition. One patient presented with leukocytosis (WBC: 37 000/mm3; granulocytes 97%) and died 2 days after a biopsy was obtained from the tumor. Nine patients underwent radical surgical procedures that varied based on the specific site of the tumor and achieved free surgical resection margins. One extensively necrotic tumor ruptured intraoperatively. Autopsy and biopsy specimens were obtained from three patients and two patients, respectively. Four of twelve patients in whom detailed data were available had positive regional lymph nodes, seven out of ten patients had intra-abdominal and/or liver metastases and one patient had a lung metastasis. Seven patients died of disease within 2 months (four before any therapy and three after surgical resection). One patient (case 2) is currently under palliative chemoradiotherapy 6 months after biopsy.

Table 2 Clinicopathological and molecular features of undifferentiated rhabdoid pancreatic carcinomas, own series (n=14)
Full size table

Pathological Findings

Grossly, the tumors were described as huge masses measuring 3–11 cm (mean size: 6; median: 6 cm), extensively infiltrating into the peripancreatic tissue and often invading adjacent structures, such as stomach, duodenum, and retroperitoneum. Their cut-surfaces were gray-whitish and friable with extensive areas of necrosis and hemorrhage. One tumor was submitted in pieces because of intraoperative rupture. Histological examination revealed two distinctive cytological patterns: pleomorphic giant cell and monomorphic anaplastic subtype. These two subtypes are described separately.

Pleomorphic Giant Cell Subtype

Ten of the fourteen tumors qualified as pleomorphic giant cell subtype. They showed highly pleomorphic neoplastic giant cells with abundant eosinophilic cytoplasm frequently containing rhabdoid inclusions (Figure 1a–d). The architectural patterns seen in the two subtypes were similar and varied from a ‘cytology slide’-like poorly cohesive mononuclear cell arrangement to a nested growth or diffuse sheets of compact tumor cells separated by thin fibrovascular septa lacking dense fibrous stroma and showing occasional acantholytic pseudoalveolar spaces. Cytokeratin immunostaining highlighted the striking variation in the size of the tumor cells (Figure 1e). All cases showed intact nuclear SMARCB1 expression (Figure 1f). In addition to strong and consistent cytoplasmic paranuclear expression of vimentin, all cases expressed pancytokeratin but in highly variable pattern (Figure 1e), usually accompanied by variable CK 7 and EMA staining (Table 3). All anaplastic cells of the tumors lost their membranous E-cadherin staining (Figure 2a), but retained it in the glandular components, especially in the lymph node metastasis (Figure 2b). Membranous β-catenin staining was lost in 10 and focally present in 2 tumors. A variable nuclear staining was seen in a few scattered tumor cells (<1%). The proliferation index (MiB1) exceeded 50% in all cases. Nuclear TP53 labeling was found in four out of seven tumors with assessable staining (Table 3). All other markers (desmin, protein S100, CD34) were not expressed in the tumor cells. The ‘rhabdoid’ inclusions were found to represent convolutes of intermediate filaments on ultrastructural examination of one case (not shown).

Figure 1
figure 1

Examples of the pleomorphic giant cell subtype of undifferentiated rhabdoid pancreatic carcinoma. (a) Highly pleomorphic tumor cells with variable nuclear sizes and frequent bi- and multinucleation. Note prominent cytoplasmic eosinophilia with rhabdoid inclusions. (b) Extreme example of cell size variation and eosinophilic cytoplasm. (c) Non-cohesive pseudoalveolar pattern. (d) Sarcomatoid spindle cells. (e) Cell size variation highlighted by pancytokeratin (note perineural carcinomatous glands lower left). (f) Intact nuclear SMARCB1 expression was seen in all cases of this subtype.

Full size image
Table 3 Immunohistochemical features of undifferentiated rhabdoid pancreatic carcinomas (n=14)
Full size table
Figure 2
figure 2

(a) Tumor with complete loss of E-cadherin (in contrast to stained normal glands on the left). (b) Membranous E-cadherin staining of the glandular tumor component in a lymph node metastasis.

Full size image

Monomorphic Anaplastic Subtype

Four tumors lacked significant cellular and nuclear pleomorphism and were rather monomorphic, with medium- to large-size vesicular nuclei, prominent nucleoli, and a moderate rim of eosinophilic cytoplasm frequently containing rhabdoid inclusions (Figure 3a). The cells were non-cohesive and loosely arranged without recognizable fibrous stroma. One tumor contained many infiltrating neutrophils (Figure 3b), with occasional cells showing emperipolesis. Another pattern seen in two cases were solid sheets of large epithelioid cells with a variable number of rhabdoid inclusions. This type of growth was frequently associated with prominent vessel-like clefts or hemorrhagic pseudocystic structures closely mimicking epithelioid angiosarcoma (Figure 3c) or proximal-type epithelioid sarcoma (Figure 3d). A mucoid/myxoid stromal pattern was seen focally in one case. Immunohistochemical staining showed similar expression of vimentin and cytokeratin as in the pleomorphic giant cell variant (Figure 3e; Table 3). Nuclear immunostaining for SMARCB1 was completely lost in all four cases (Figure 3f). Other immunohistochemical features were similar to the pleomorphic giant cell subtype. Nuclear TP53 labeling was found in one out of two tumors with evaluable staining (Table 3).

Figure 3
figure 3

Examples of the SMARCB1-deficient monomorphic subtype. (a) Small to medium sized monotonous rhabdoid cells in pseudoalveolar pattern. (b) Another case showed prominent neutrophilia and focal gland formation (upper left). (c) Epithelioid large cell pattern mimicking angiosarcoma. (d) Compact sheets of large cells with frequent rhabdoid inclusions mimicking proximal-type epithelioid sarcoma. (e) Characteristic paranuclear cytokeratin expression (KL1). Inset: loss of pancytokeratin in another case with focal expression in gland-like areas. (f) Complete loss of nuclear SMARCB1 expression (main image: same case as in a with retained expression in endothelial and stromal cells; inset: same case as in b, with prominent nuclear staining of neutrophils and stromal cells).

Full size image

Glandular Differentiation and Other Features

In the pleomorphic giant cell subtype, foci of neoplastic duct-like glandular structures were seen in four cases. One case showed only a focus of severely dysplastic pancreatic intraepithelial neoplasia (PanIN 3), but frankly glandular differentiation in a lymph node metastasis (Figure 2b). One tumor showed a focus of perineural glandular differentiation amid highly pleomorphic tumor giant cells (Figure 1e). In the monomorphic subtype, glandular features (seen in only one case) were very subtle and represented either single-scattered gland-like structures or pseudopapillary acantholytic gland-like spaces. The gland-like structures in the monomorphic variant were all SMARCB1 negative similar to the undifferentiated component. Extensive areas of tumor necrosis usually accompanied by severe hemorrhage, prominent lymphovascular invasion, infiltration of arterial vessels, and diffuse lymphoma-like infiltration of peripancreatic fatty tissue were found in all cases.

Molecular Findings

Molecular analysis revealed KRAS mutations in 6 out of 11 (54%) successfully examined tumors. All mutations were point mutations affecting exon 2 (Table 2). KRAS amplification was detected in 5 out of 13 (38%) cases successfully analyzed by FISH (Figure 4). One additional case showed polysomy (most cells displayed three signals for KRAS and CEP12). KRAS mutations and KRAS amplification coexisted in four out of six cases (66%). On the contrary, all four KRAS wild-type tumors successfully analyzed by FISH lacked KRAS amplification. KRAS amplification correlated with intact SMARCB1 expression in four out of five (80%) tumors. Taken together, KRAS alterations (mutations and/or copy number changes) highly correlated with intact SMARCB1 expression (7 out of 8 cases; 87%). On the other hand, a strong correlation was found between the absence of KRAS alterations and the loss of SMARCB1 expression (three of five KRAS wild-type cases were SMARCB1 deficient). All but one of the four SMARCB1-negative cases lacked any KRAS alterations.

Figure 4
figure 4

Example of KRAS amplification with >6 green signals of KRAS (fluorescence in situ hybridization dual-color probe).

Full size image

Literature Review

Our review of the English literature revealed 46 cases of pancreatic carcinoma that were described in details and fulfill the diagnostic criteria of rhabdoid differentiation as defined in this study. These neoplasms have been reported under different names: pleomorphic adenocarcinoma, pleomorphic carcinoma, pleomorphic giant cell carcinoma, round cell anaplastic carcinoma, sarcomatoid carcinoma, carcinoma with gemistocytes, rhabdoid carcinoma, and carcinoma with rhabdoid phenotype/features.5, 6, 7, 8, 13, 14, 15, 16, 17, 18, 19, 32, 33 Thus, including our cases, a total of 60 rhabdoid carcinomas of the pancreas have been described to date (Tables 2 and 4). Patients were 44 men and 16 women (M/F=2.8:1) aged 30–96 years (mean age: 65 years). In all, 71.6% of patients were ≥60 years. The tumor originated in the pancreas head (25), body (7), tail (10), diffuse/more than one part (11), and in nonspecified part of pancreas (7).

Table 4 Previously reported undifferentiated pancreatic carcinomas with rhabdoid features (n=46)
Full size table

Most patients presented with locally advanced or widely metastatic disease and received only supportive treatment or palliative surgery with or without chemoradiation. Seventeen patients received surgery as initial and definitive treatment, and three of them also received adjuvant radio/-chemotherapy. Of the remainder, 35 patients underwent biopsy and palliative treatment because of unresectable disease. The tumor was diagnosed at autopsy in five cases. Of four patients who received biopsy followed by radiochemotherapy, three died of disease within 3–13 months and one is alive, currently 6 months under palliative treatment. Lymph node metastases were identified in almost all cases with detailed information (>90% of all). Liver metastases were also seen in the majority of patients, usually shortly after surgical treatment. Follow-up ranging from 1 day to 19 months was available for 49 patients; all but 4 patients (92%) died of disease either postoperatively or within 1–19 months (median: 4 months; mean: 5.4 months). All but three patients died within 1 year (Table 2 and Table 4). The four patients who were reported alive had limited follow-up (≤6 months) or were recent cases.

In the tumor descriptions, rhabdoid cells dominated the tumor histology in most cases. Occasionally, signet ring-like cells with mucicarmine-positive vacuoles were noted. An adenocarcinoma component of varying degree of differentiation or high-grade intraepithelial neoplasia (PanIN3) were detected in 32 out of 41 (78%) of the cases, in which sufficient tissue was examined. In a few cases, mucinous adenocarcinoma, squamous differentiation, and spindle cell areas were reported. Unequivocal glandular differentiation in metastases in association with a primary tumor, which was devoid of such a component, was noted in four cases. The available immunohistochemical findings consistently demonstrated coexpression of vimentin and pancytokeratin, mostly confined to or highlighting the paranuclear rhabdoid cytoplasmic inclusions.

Molecular data were only available in a single case that showed missense mutation of SMARCB1 in the rhabdoid component.17 SMARCB1 expression and KRAS mutation were not tested in the 46 previously reported cases.

Discussion

According to the current WHO classification of pancreatic tumors, undifferentiated carcinoma is defined as ‘a malignant epithelial neoplasm in which a significant component of the neoplasm does not show a definitive line of differentiation.’10 Included in this definition are histologic variants, such as anaplastic giant cell carcinoma (composed of pleomorphic mononuclear cells admixed with bizarre-appearing eosinophilic giant cells), sarcomatoid spindle cell carcinoma, and carcinosarcoma with recognizable adenocarcinoma and high-grade spindle cell areas, and undifferentiated carcinoma with osteoclast-like giant cells. The ‘rhabdoid features’ that characterize the pancreatic undifferentiated carcinomas of this series and those described in several other reports5, 6, 7, 8, 13, 14, 15, 16, 17, 18, 19, 32, 33 have not been included into the descriptive terms of the WHO classification.10

Here we show that undifferentiated carcinomas of the pancreas exhibiting rhabdoid features vary greatly in morphology, but may be separated into two subtypes, one with a pleomorphic giant cell and another with a monomorphic anaplastic pattern. When these neoplasms were analyzed for KRAS alterations and SMARCB1 expression, a strong correlation was found between histology and molecular findings. KRAS mutations and/or amplification, identified by PCR/sequencing and FISH, respectively, and intact immunohistochemical expression of SMARCB1 (as a highly sensitive and specific marker for intact SMARCB1 locus) were linked to the pleomorphic giant cell subtype, whereas lack of KRAS alterations and loss of nuclear SMARCB1 was found in the majority of monomorphic anaplastic carcinomas.

Recently, KRAS copy number changes or polysomy of chromosome 12 were identified in 42% of undifferentiated carcinomas of the pancreas but not in ductal adenocarcinomas, and were also found to accompany the intratumoral transition from ductal adenocarcinoma to undifferentiated carcinoma.29 In addition, a trend toward more frequent KRAS amplification was observed among cases with mutant allele-specific imbalances.29 This study, however, did not focus on tumors with rhabdoid features or the expression of SMARCB1.

Several recent observations support the novel concept of SMARCB1-deficient neoplasms emerging as secondary ‘dedifferentiated’ clones in the background of a differentiated SMARCB1-positive ‘parent’ neoplasm (ie, an adenocarcinoma) in different organs including the pancreas.25, 26 Cho et al17 reported on a pancreatic mucinous adenocarcinoma with a ‘rhabdoid’ component. Molecular analysis revealed a missense SMARCB1 mutation in the rhabdoid component, but SMARCB1 immunostaining was not performed.

Our results confirm and extend the above-mentioned molecular findings in pancreatic undifferentiated carcinomas. First, they demonstrate that KRAS copy number changes on a background of intact SMARCB1 expression are typical for the pleomorphic anaplastic subtype of undifferentiated carcinomas with rhabdoid features. Second, they indicate that SMARCB1-deficient but KRAS-intact undifferentiated rhabdoid carcinomas are characterized by a monomorphic anaplastic histological cell pattern. These findings suggest that the molecular pathway leading to a morphological and biological shift of ductal adenocarcinomas to a highly aggressive pleomorphic anaplastic carcinoma with rhabdoid features is heterogeneous, depending, in case of the pleomorphic subtype, either on mutated and amplified KRAS on a background of intact SMARCB1 expression or, in case of the monomorphic subtype, on a loss of SMARCB1 in the presence of wild-type KRAS. The molecular mechanisms that follow these two distinct molecular changes and lead to a switch from a differentiated adenocarcinoma to an undifferentiated carcinoma with rhabdoid appearance and aggressive behavior are not known so far. However, it is likely that the concurrent loss of E-cadherin and β-catenin from the surface of the tumor cells, noted in this and other studies,34, 35 is a significant finding contributing to the highly infiltrating growth of the tumors. Concerning the rhabdoid features, which make these tumors so distinct, it can be concluded that their occurrence is not solely caused by loss of SMARCB1 but may also have other causes, as has also been noted in other neoplasms.26, 36 The exact role of SMARCB1 in the single KRAS mutated case in our series where its loss is likely to be a secondary event in tumorigenesis (as has been shown also in a subset of microsatellite instable colorectal cancer with secondary SMARCB1 loss26) remains therefore to be determined.

Identification of pancreatic undifferentiated carcinomas with rhabdoid cells in previously published reports is often difficult, because many of these tumors were lumped in the category of pleomorphic giant cell carcinoma or anaplastic carcinoma that also included cases without rhabdoid features. It seems that pancreatic undifferentiated carcinomas with rhabdoid features have first been described under the term pleomorphic carcinoma in 1954 by Sommers and Meissner, who found three such tumors among 142 autopsy cases of pancreatic adenocarcinoma (2%).4 In addition to these cases, we identified 46 other pancreatic tumors with rhabdoid features reported since 1968 (see Table 4). As our review covers a time period of more than 40 years, it is obvious that the relative frequency of these tumors is very low and probably in the range of 1%, particularly, if only surgically treated cases are considered.

The gender distribution and mean age of the patients with pancreatic undifferentiated carcinoma with rhabdoid features as well as the tumor location in the pancreas are similar to the clinicopathological features of pancreatic ductal adenocarcinoma.10, 11, 12 The prognosis, however, whether the tumor contained a glandular component or not, is worse.5, 6, 7, 8 According to our review, almost all patients died of their disease or its complication within a mean of 4 months with >20% dying immediately or within a few weeks after biopsy or surgery. Tschang et al summarized the ‘usual course of events’ in these patients with the words ‘hospitalization, laparotomy and biopsy, and death.’7 This seems to be true until today, irrespective of the treatment modality used. The rapid progress of the tumors may be explained by the complete loss of the adhesion molecules E-cadherin and β-catenin from the surface of the cells that probably results in a highly aggressive, non-cohesive (lymphoma-like) tumor cell growth, with easy invasion of adjacent organs, vessels, and the retroperitoneum. Adverse systemic effects caused by cytokines produced by the tumor cells seems to be an adverse prognostic factor in some patients. Nakajima et al33 reported one patient with rhabdoid carcinoma who presented with a peripheral leucocyte count of 91 500/mm3 (neutrophils: 87.5%) and elevated serum granulocyte colony-stimulating factor. The patient died 11 days after diagnosis and expression of granulocyte colony-stimulating factor could be demonstrated in tumor cells. Similar to that case, one of our patients presented also with elevated leucocyte count of 37 000/mm3 (neutrophils: 97%) and died 2 days after biopsy. Histologically, prominent granulocytic infiltrates were seen within some tumors.

From a differential diagnostic view point, undifferentiated pancreatic rhabdoid carcinoma needs to be distinguished from proximal-type epithelioid sarcoma of the viscera37, 38 and from microsatellite instable pancreatic medullary carcinoma with DNA replication errors.39 SMARCB1-deficient undifferentiated rhabdoid carcinoma is indistinguishable from rhabdoid cell-dominated epithelioid sarcoma on morphological ground alone. However, identification of a glandular component in 50% of cases and of a driver KRAS mutation in one of our four cases that are SMARCB1 deficient highlights the epithelial derivation of these neoplasms and argues against a visceral variant of epithelioid sarcoma. Pancreatic medullary carcinoma with DNA replication errors is a rare variant of poorly differentiated non-gland-forming carcinoma that is characterized by wild-type KRAS.39 In sharp contrast to our series, this subtype shows expanding borders, syncytial growth of tumor cells, lacks rhabdoid cell features, consistently shows loss of mismatch repair proteins and microsatellite instability, and is characterized by prolonged survival.39

In summary, our study highlights the phenotypic and molecular heterogeneity of the pleomorphic/anaplastic variant of pancreatic undifferentiated carcinoma, which has undergone rhabdoid transformation. The clear-cut correlation between tumor histology and molecular findings (KRAS alterations and SMARCB1 expression status) highlights the existence of at least two independent molecular pathways, each associated with its own morphological subtype. The rhabdoid phenotype seems to unify these two subtypes. Although the pathogenesis remains in most parts unclear, KRAS mutation/amplification on the one side and loss of SMARCB1 on the other side seem to have an essential role for the phenotypic shift from differentiated ductal adenocarcinoma to undifferentiated carcinoma with rhabdoid changes. The results of this study let us therefore assume that the pancreatic undifferentiated carcinomas not only differ in their morphology but also in their molecular profiles. If this assumption is correct, a careful morphological separation of the variants using a clear terminology is needed. The nomenclature that has been so far used for these neoplasms is often confusing and may impede appropriate future treatments.