Introduction

Pancreatic ductal adenocarcinoma (PDAC) is an aggressive tumor with dismal prognosis. The overall 5-year survival rate is only 6% and after curative surgery less than 25%1, making PDAC one of the most lethal tumors among solid malignancies2. This poor outcome is related to multiple factors, including resistance to chemotherapy and the relatively late stage of diagnosis due to unspecific symptoms and aggressive tumor biology1.

Over the last decade major improvements have been made in understanding the mechanisms of molecular carcinogenesis in PDACs3,4,5,6,7. The first milestone was the discovery of the molecular fingerprint of PDAC that included the common mutations in KRAS, SMAD4, TP53 and CDKN2A/p168. Recent advances in gene sequencing9 by introduction of high-throughput molecular methods allowed to further address the genetic complexity of PDAC10. The first global analysis of 24 advanced PDACs using comprehensive exome sequencing revealed a high mutation rate with an average of 63 mutations per case connected to 12 core signaling pathways10. In 2011, Collisson et al. showed that PDACs and murine PDAC cell lines may be stratified by their transcriptional profiles into subtypes with different clinical outcome and drug response11. The stratification of these subtypes was associated with three patterns that were termed classical, quasi-mesenchymal and exocrine-like. The terminology of this subdivision, however, is hardly understandable regarding its morphological meaning. In other recent studies, it was reported that intact KRAS and the number of mutations in key genes may have prognostic value in PDAC patients12,13,14,15,16. However, it is unclear so far, whether there are genotype-phenotype correlations that are based on the identification of PDACs with special growth patterns or morphological variants with better survival than classical PDACs, beyond the established pathological parameters of staging and grading17.

This study that only involved PDAC patients in whom surgical treatment could be performed, focuses on detailed histological investigation and molecular examination of the mutational status of KRAS, CDKN2A/p16, TP53 and SMAD4 in correlation with survival and accurate morphological subtyping. The presented findings show that, according to the analyzed parameters, prognostic relevant subtypes of PDAC can be identified.

Results

The patients’ clinicopathologic features are summarized in Table 1. Female (45.8%) and male (54.2%) patients were equally represented with a median age at diagnosis of 68 years. Most patients presented with advanced stage of the disease (≥UICC 2b: 73.4%, 130/177). A minority of patients (9.4%) received neoadjuvant treatment (details see Supplementary Table 1). All patients were judged resectable and underwent major pancreatic surgery: pylorus-preserving partial pancreaticoduodenectomy (ppWhipple) 55.4% (98/177), partial pancreaticoduodenectomy (Whipple, classic) 9% (16/177), distal pancreatectomy 18.6% (33/177) and total pancreatectomy 17% (30/177).

Table 1 Clinicopathological features of 177 patients with resected pancreatic adenocarcinomas.
Full size table

Histological features

The results are summarized in Table 2 and examples for each pattern are shown in Figs 1 and 2. The majority of the cases (92.1%, 163/177) were classified as either conventional PDACs (91/177, 51%), predominantly graded G2 (44%) or G3 (48%), or combined PDACs (72/177, 41%), predominantly graded G2 (37%) or G3 (61%). Two of the conventional PDACs contained at the periphery concomitant small (<1.5 cm) gastric type IPMNs and four had retention cysts. The combined PDACs showed as dominant histologic feature a conspicuous cribriform (17/177), clear-cell (16/177), papillary (17/177), gyriform (8/177), micropapillary (2/177) or complex (12/177) component (Fig. 1A–F). While the cribriform and clear cell combined PDACs were all G2 tumors, the gyriform, papillary, micropapillary and complex combined PDACs accounted for most (90%) of the G3 tumors. One PDAC with a clear cell component was associated with a pancreato-biliary type IPMN, and another PDAC with a complex component with a gastric type IPMN. A minority of tumors (7.9%, 14/177) fulfilled the criteria of PDAC variants and special pancreatic carcinomas. There were two G3-adenosquamous carcinomas (Fig. 1G), two G1-colloid carcinomas (Fig. 1H), six G2-papillary carcinomas (Fig. 1I), one G2-medullary carcinoma (not shown), and three G1-tubular carcinomas (Fig. 2). One of the colloid carcinomas was associated with an intestinal type IPMN. Two thirds of the papillary carcinomas (4/6) were associated with IPMNs, either of gastric (1/6), intestinal (1/6) or pancreato-biliary type (2/6). The three tubular adenocarcinomas were composed of well-differentiated open tubules that infiltrated the parenchyma diffusely and were accompanied by abundant cellular desmoplastic stroma (Fig. 2). The tubules had oval, rounded or angular shapes and were lined by a single layer of mostly cuboidal cells, with little nuclear pleomorphism, inconspicuous nucleoli and scanty mitotic figures.

Table 2 Histological types of pancreatic ductal adenocarcinoma: frequency and survival.
Full size table
Figure 1: Spectrum of histologic patterns in pancreatic ductal adenocarcinomas (PDACs).
figure 1

(A) Conventional PDAC. (B–F) Combined PDACs with a dominant histological component: (B) Clear-cell component. (C) Cribriform component. (D) Gyriform component. (E) Micropapillary component. (F) Complex component. (G) Adenosquamous carcinoma. (H) Colloid carcinoma. (I) Papillary carcinoma.

Full size image
Figure 2: Tubular adenocarcinoma of the pancreas.
figure 2

(A) Low power view showing groups of well-differentiated infiltrating tubules surrounded by small cuffs of desmoplastic stroma. (B) Infiltrating well differentiated neoplastic glands closely imitating normal ducts.

Full size image

No morphological differences were observed between patients who received neoadjuvant therapy and those without neoadjuvant treatment (data not shown).

Molecular features

KRAS, TP53, CDKN2A/p16 and SMAD4 mutations were found in various frequencies and combinations in all but one tumor (case # 62; confirmed by NGS). Most PDACs carried multiple mutations (two: 47/177, 25.6%; three: 85/177, 48%; four: 36/177, 20.3%), while tumors with one mutation (8/177, 4.5%) were rare.

KRAS was most frequently mutated (Table 3, Fig. 3), mainly in exon 2 (91%), and rarely in exon 3 (8.5%) and only once in both exons. No mutations were detected in exon 4. Almost all (98.8%, 162/164) KRAS mutations were detected by HRMA and Sanger sequencing (with expected detection limits of 10% and 15–20%, respectively (29)). Two low-level mutations (mutations with a low frequency in sequence reads) were identified by NGS (see also Fig. 3). All tumors with wildtype KRAS showed intact BRAF (codon 600).

Table 3 Molecular characteristics of pancreatic adenocarcinomas.
Full size table
Figure 3: KRAS mutation analysis (direct sequencing) of codon 12 of exon 2.
figure 3

(A) Hotspot of exon 2 shows KRAS wildtype, (B) KRAS mutation (p.G12D, arrow), or (C) low-level mutation (p.G12C, arrow, faint red signal), as confirmed by next generation sequencing (NGS) (case #190).

Full size image

TP53 wildtype and TP53 mutations were associated with distinct p53 immunolabelling patterns depicted in Fig. 4. TP53 Mutations were found in 139/177 tumors and detected in the known hotspots (exon 5–8), but not in exon 9. Mutation type 1 (defined as nuclear expression in ≥25% of tumor cells and missense mutation) was more common (52%, 92/177) than mutation type 2 (26.6%, 47/177; defined as absence of expression and presence of an intragenic deletion, a nonsense, a frameshift or splice site mutation) (Fig. 4).

Figure 4: TP53 analyses.
figure 4

(A,B) TP53 wildtype is associated with nuclear TP53 expression in up to 24% of neoplastic cells. (C,D) TP53 missense mutation (p.Y205C; exon 6) is associated with nuclear TP53 overexpression in ≥25% of neoplastic cells (mutation type 1). (E,F) TP53 intragenic deletion or nonsense, frameshift or splicesite mutations (mutation type 2; here represented by an insertion-frameshift mutation p.P153*fs28; exon 5) is associated with loss of nuclear TP53 expression in neoplastic cells.

Full size image

CDKN2A/p16 alterations, recorded in 75% of the tumors, were detected by loss of protein expression (130 tumors), which was confirmed by loss of heterozygosity (LOH) in 58 tumors. Three additional tumors showed intact expression despite presence of LOH. Loss of SMAD4 expression was recorded in one third of the cases (Table 3, Fig. 5).

Figure 5: Immunohistochemical analysis of CDKN2A/p16 and SMAD4.
figure 5

(A) Strong nuclear and cytoplasmatic staining of CDKN2A/p16 in neoplastic cells indicating the presence of an intact gene. (B) No labeling of CDKN2A/p16 in neoplastic cells indicating either a deletion, inactivating mutation, or promoter hypermethylation. (C) Nuclear SMAD4 immunolabeling of neoplastic cells indicating the presence of an intact protein. (D) Loss of SMAD4 expression in >90% of neoplastic cells indicating a deletion or inactivating mutation of the gene.

Full size image

Interestingly, altered CDKN2A/p16 was significantly more common in patients with positive lymph nodes (p = 0.02). No significant differences between lymph node negative and positive patients were observed for number of mutations, KRAS, TP53, SMAD4 and tumor morphology.

Four investigated metastases (3 from the liver and one from the peritoneum) had the same KRAS mutations as the primary. In case # 5 a low-level mutation in the primary tumor (p.G12D, 3%) was associated with a high-level mutation (p.G12D) in the liver metastasis. In case # 9 intact TP53 in the primary tumor was associated with a strong (>90%) nuclear labeling in the peritoneal metastasis.

No differences were observed in the mutational status of patients who received neoadjuvant therapy compared to the large group of patients without neoadjuvant therapy (data not shown).

Correlation of molecular with morphologic features

KRAS wildtype was significantly more commonly detected in variants than classical PDACs (conventional or combined) (p = 0.035) (Table 4). In detail, the group of KRAS wildtype tumors (7.3%, 13/177) included conventional PDACs (7/13), combined PDACs (3/13), and one colloid, one medullary and one tubular carcinoma. The KRAS wildtype tubular carcinoma and one KRAS-mutated tubular carcinoma also lacked alterations in the three other genes (Table 5), a result that was confirmed by additional extended gene analysis. One of the two KRAS-mutated tubular carcinomas harbored CDKN2A/p16 mutations (Table 5). All papillary carcinomas were associated with mutated KRAS and lacked mutated GNAS18, irrespective of the presence of an associated IPMN. CDKN2A/p16 alterations were more strongly associated (p = 0.016) with combined PDACs (60/71) than with conventional PDACs (29/91) (Table 4).

Table 4 Comparison between morphological phenotype, genotype, grading and survival.
Full size table
Table 5 Clinicopathologic and molecular features of pancreatic tubular adenocarcinoma.
Full size table

Clinical outcome correlated to morphologic and molecular features

At the time of survival analysis, 33 of 146 patients (22.6%) were alive. Median follow-up for all patients was 19.8 months (2.5–75.1 months) and median follow-up for patients alive was 49.4 months (25.9–75.1 months). Overall survival rates were 72% for one year, 29% for three years and 21% for five years.

Patients with colloid, medullary, tubular or papillary carcinoma survived significantly longer than patients with conventional and combined PDACs (p = 0.04) (Table 2). Longest survival was seen in two of the patients with a tubular carcinoma. They were still alive at the completion of the study, with a survival of >68.8 and >55.3 months, respectively. On the contrary, patients with adenosquamous carcinomas had an extremely poor outcome (4.1 and 10.0 months) (Table 2).

Detailed analysis of the large group of patients with conventional and combined PDACs revealed that patients with a conventional PDAC and those with a cribriform type combined PDAC showed the most favorable overall median survival (22.7 and 28.7 months), followed by combined PDACs with clear-cell and papillary components (17.6 and 13.9 months) (Table 2). Combined PDACs with gyriform and complex components were associated with poor survival (12.5 and 10 months).

The median patient survival correlated significantly with the number of mutations (p = 0.04, Fig. 6A). Patients with no or low number of mutations (no mutation >68.6; one mutation >45; two mutations >25.3 months) survived longer than patients with tumors that harbored three (17.6 months) or four mutations (14.5 months).

Figure 6: Kaplan-Meier survival curves in pancreatic adenocarcinomas correlated with molecular status.
figure 6

(A) Significant correlation of median patient survival with number of mutations. (B) Patients with KRAS wildtype have a significantly better overall survival than patients with mutated KRAS. (C) Patients with intact CDKNA2/p16 have a significantly better overall survival than patients with altered CDKNA2/p16.

Full size image

Correlation of patient outcome and molecular features revealed a significant survival benefit in patients with wildtype KRAS (7.3%, 11/146) compared with that of KRAS mutated patients (92.7%, 135/146) (p = 0.018, median survival >45 vs. 19.7 months, Fig. 6B). Likewise, patients with intact CDKN2A/p16 (22.8%, 33/145) lived significantly longer than those with altered CDKN2A/p16 (77.2%, 112/145) (p = 0.006; 36.9 vs. 18.8 months, Fig. 6C). Significant prognostic results were further obtained by the combination of CDKN2A/p16 and KRAS. All patients from the cohort followed for survival with an intact status of both genes (n = 4) were still alive at the end of the study. A median overall survival could therefore not be calculated. No prognostic significance was observed for mutated TP53 (with no prognostic difference between TP53 type 1 and type 2 mutations) and SMAD4.

Univariate survival analysis using log-rank test showed significant p-values for KRAS, CDKN2A/p16, number of mutations, pN and grading. In a backward selected multivariate Cox proportional Hazard model altered KRAS (HR 2.77, 95%-CI: 1.11–6.90; p = 0.019), pN1 (HR 2.66; 95%-CI: 1.67–4.26; p < 0.001) and grading G3 (HR 3.66, 95%-CI: 1.57–8.54; p = 0.001) were independent predictive variables for survival.

Discussion

Reliable prognostic markers for pancreatic ductal adenocarcinoma (PDAC) patients are so far rare. By analyzing the expression profile of primary PDACs, and human and mouse PDAC cell lines, Collisson recently identified molecular subtypes of PDAC that differed in clinical outcome and drug response11. Other studies revealed an association between expression as well as mutational status of key tumor suppressor and oncogenes and patient survival16,19. In all these studies, the molecular profile of PDACs is not or only vaguely correlated to the individual morphology of the tumors.

Here we present data of a correlative study on histopathology, molecular profile, and survival in PDACs and related carcinomas of 177 resected patients, with the aim to find prognostic relevant features. As expected, the overall outcome of our patient cohort was bad. Nearly two thirds of patients survived less than 24 months and the 5-year survival did not exceed 21%. However, within this cohort there were patients who survived for up to four years and longer, and whose tumors had a special histopathology and/or molecular status.

Histopathological heterogeneity in PDACs has long been recognized, but it has not been defined in detail, -with the exception of the definition of histological grade and histological variants. Here we classified the pancreatic carcinomas according to a defined growth pattern into three groups. The first group included the conventional type PDACs, which showed an equal mixture of various histological elements (for details see Material and Methods). The second group encompassed the combined PDACs, which were characterized by a dominant histological component (defined as involving more than 30% of the tumor area). The third group contained PDAC variants (i.e. adenosquamous carcinoma, colloid carcinoma, papillary carcinoma) and special adenocarcinomas such as medullary carcinoma and tubular carcinoma. By analyzing the survival of patients who were ascribed to the various morphological types, we found and confirmed that the colloid carcinoma, the medullary carcinoma and the tubular carcinoma showed a better outcome than conventional PDACs and particularly adenosquamous carcinoma20,21,22,23. Patient with a conventional PDAC or a PDAC with dominant cribriform component survived longer than patients with combined PDAC and other histological components, such as the clear cell, the papillary, the gyriform and, in particular, the complex component.

PDACs typically harbor KRAS mutations, followed by mutations of CDKN2A/p16, SMAD4 and TP5310. These genes are considered the driver genes of PDAC. In addition, there is a multitude of other, but much less frequent, gene alterations, as revealed by whole genome sequencing analysis6,7. In our patients, in whom the molecular results were obtained by Sanger methodology and completed in selected cases by NGS of a large PDAC gene panel including the 40 most commonly mutated genes, KRAS was found to be mutated in 92.7%, TP53 in 78.5%, CDKN2A/p16 in 72.9% and SMAD4 in 37.3%. Most of these alterations coexisted in individual tumors and two third of the PDAC patients (68%) harbored alterations in three or all four genes.

Correlation of the individual mutational status with the respective histopathology and survival data revealed a number of findings with prognostic significance. First, the number of mutations per tumor was of prognostic relevance. Patients with one, two or three mutations survived longer than those with alterations in all four genes (>45 vs. 25.3 vs. 17.6 vs. 14.5 months). Although these results confirm previous reports16, so far the correlation between gene status and phenotype has not been analysed in detail. Among the tumors with low number of mutations and prolonged survival, there are particularly colloid carcinomas, medullary carcinomas and tubular type carcinomas, while adenosquamous carcinomas, papillary carcinomas, and the combined PDAC with a complex pattern belong to the carcinomas with a high mutational frequency and poor survival (Fig. 7). Medullary and colloid carcinomas are known for their low prevalence of somatic mutations and good prognosis20,21,22,24. Conversely, adenosquamous carcinoma is well known for its many somatic mutations and aggressive behavior25,26. The papillary carcinoma variant seems in many cases to be the invasive component of an IPMN27, since IPMNs either of the pancreato-biliary (2/6), intestinal (1/6) or gastric type (1/6) were found in four of our six cases. However, it may also occur without an associated IPMN, since in two cases an associated IPMN was not found, and this was also true in another recently reported series of 10 papillary cystic PDACs28.

Figure 7: Risk stratification after surgical resection of PDAC.
figure 7

Proposed scheme to identify PDACs with low/intermediate and high biological aggressiveness based on morphological classification and molecular testing of key genes.

Full size image

Second, patients with either wild type KRAS or CDKN2A/p16 had a better outcome than those with mutations in these genes, while loss of SMAD4 and intact or altered TP53 was of no prognostic relevance. The statistical significance of wildtype KRAS proved to be so strong that multivariate analysis identified the mutational status of KRAS as an independent prognostic marker. This result confirmed the recently reported survival benefit for PDAC patients with intact KRAS12,13,14,15. The largest of these studies by Sinn, included 153 PDAC patients and reported a significantly decreased median survival of 12.7 months for patients with mutated KRAS versus 20.7 months for patients with wild type KRAS15. In our study, the corresponding data were 19.7 months for mutated KRAS versus >45 months with wild type KRAS. The longer survival in our patients may be due to a better selection on the basis of a more extensive KRAS analysis including exon 2 to 4, in contrast to Sinn’s study that only focused on KRAS exon 2. This methodological difference probably also explains, why in Sinn’s study the KRAS mutation rate of 68% is much lower than in our study with a rate of 93%.

Among the tumors with a wildtype KRAS status were a medullary carcinoma, whose particular genetic and biologic features have been described20,21,22,23, and a tubular adenocarcinoma. This latter type of carcinoma, which corresponds by grade to a G1 tumor (WHO 2010), may be considered a conventional PDAC with a strictly well differentiated tubular differentiation. It shows great similarities regarding its morphology, low frequency, rare mutations and long survival to the tubular type carcinoma of the breast29,30,31,32. One of the three tubular adenocarcinomas that we identified, not only had an intact KRAS gene, but also harbored no alterations in the 40 most commonly altered PDAC genes, including CDKN2A/p16, TP53 and SMAD4. Moreover, the patient with this tumor is still alive (see Table 4, case # 62), with a follow-up for >68.6 months. The other two patients with tubular carcinomas had KRAS and/or CDKN2A/p16 alterations, but no loss of SMAD4 or TP53 mutations (details see Table 4), and survived for >55.3 and 19.3 months.

The relevance of mutated KRAS as a prognosticator in PDAC, a feature shared with bile duct cancer33, is biologically most likely related to its driver function. Evidence that KRAS, in interaction with TP53, CDKN2A/p16 and SMAD4, is a driver gene comes from genetically engineered KRAS mouse models of pancreatic cancer10,34,35,36,37,38,39,40,41. Moreover, recently developed mouse models in which mutated KRAS can be switched on and off, have impressively demonstrated that continuous oncogenic KRAS signaling is essential for both progression and maintenance of PDAC42,43 and its metastases44.

Patients with wildtype CDKN2A/p16, like patients with intact KRAS, lived significantly longer than those with altered genes (p = 0.006; 36.9 months vs. 18.8 months). Though multivariate analysis failed to identify this gene constellation as an independent prognostic factor, CDKN2A/p16 seems to play a role in patient outcome. All our patients without alterations in CDKN2A/p16 and KRAS were still alive at the completion of the study. On the other hand, loss of CDKN2A/p16, as was recently reported, seems to be associated with lymphatic invasion and widespread metastasis19 and was significantly associated with lymphatic spread in our study. Interestingly, subgroup analysis revealed a significant higher CDKN2A/p16 mutation rate in combined PDACs with a dominant histological component compared to conventional PDACs. This suggests that a driver gene might be linked to the presence of dominant histological components in PDACs.

The prognostic relevance of SMAD4 has been the subject of controversial discussions over the last years. While several studies associated the loss of SMAD4 with poor prognosis or early metastatic disease14,19,45,46, others were unable to confirm these results16,47,48. Likewise, no differences in the survival data in patients with intact and loss of SMAD4 were observed in our study (p = 0.15, median overall survival 22.2 vs. 17.6 months).

In summary, our findings specify the prognostic relationship between the histopathology and molecular profile, based on the morphologic stratification of PDACs in subtypes and variants, and the mutational status of the four driver genes, KRAS, CDKN2A/p16, SMAD4 and TP53. Because our data suggest that PDAC subgroups can be identified (low to intermediate aggressive PDACs versus highly aggressive PDACs, see Fig. 7), tailored therapy options may be discussed. Patients with altered CDKN2A/p16 might benefit from more aggressive preoperative therapies, whereas patients with wildtype KRAS might best be treated by upfront surgery followed by adjuvant therapies. In view of these potential therapies it should be considered to include the status of the four driver genes into the pathological reporting of PDACs in the future12,13,14,15,49. Given the potential clinical implications of our results, validation in independent PDAC cohorts is of utmost importance.

Material and Methods

The study was approved by the ethics committee of the TU München, Germany (documents no. 1926/2007 and 126/2016S). Written informed consent was obtained from all patients. All methods were performed in accordance with the relevant guidelines and regulations.

Patients

From 07/2007 to 07/2011 200 patients underwent an elective pancreatic resection at the Department of Surgery, Klinikum rechts der Isar, TU München, Germany, with a final histopathologic diagnosis of PDAC. Associated and concomitant intraductal papillary mucinous neoplasms (IPMN)50 were reported. IPMNs with only minimally invasive component were excluded. TNM and grading followed the WHO recommendations17. Patients with a family history of PDAC were excluded from the study.

Clinical data and follow up

Clinical, demographic and macroscopic information was obtained from a patient database and by reviewing the medical charts and pathology reports. Follow up on the patients’ conditions was obtained by clinical record, directly contacting the patients and/or their physicians. Patients with neoadjuvant therapy (n = 14, details see Supplementary Table 1), distant metastasis/UICC stage IV disease (n = 11, thereof six liver metastases, four peritoneal metastases and one pulmonary metastasis, details see Supplementary Table 2) and/or arterial resection (n = 4), perioperative death/death due to complications (n = 1) and recurrence surgery (n = 1) were excluded from survival analysis. Three patients were lost to follow-up.

Histologic analysis

All PDACs were histologically classified into conventional PDACs, combined PDACs in which, in addition to the classical tubular growth pattern, a special histologic component was present in more than 30% of the tumor area, and variants with a special pattern in at least 50% of the tumor area. PDACs with a conventional morphology were largely composed of well- to moderately developed tubular and duct-like structures and showed only few other structures, such as glands with clear cell morphology, cribriform architecture, papillary epithelial lining and individual pleomorphic cells17. Combined PDACs with dominant histological features showed either a clear-cell, cribriform, gyriform, papillary, micropapillary or complex component against a background of tubular architecture (see Fig. 1). The complex component was characterized by small irregular glands mixed with solid or cribriform cell sheets and individual pleomorphic cells. PDAC variants included colloid, adenosquamous, and papillary carcinoma (see Fig. 1A–F). Among the pancreatic carcinomas that have not yet been regarded as PDAC variants are medullary and tubular carcinoma. The tubular adenocarcinoma is separated from conventional PDAC by its entirely well differentiated tubular architecture that is characterized by small tubular glands diffusely infiltrating the pancreatic parenchyma and difficult to distinguish from equally sized normal ducts (see Fig. 2).

Immunohistochemical analysis

All stainings were run on an automated immunostainer with an iVIEW DAB detection kit (Ventana Medical Systems, Roche, Mannheim, Germany) according to the company’s protocols for open procedures with slight modifications, for details see Supplementary materials and methods.

Molecular analysis

Molecular analysis was performed on extracted DNA from manually microdissected FFPE tumor tissue, for details see Supplementary materials and methods and Supplementary Table 1.

Statistical analyses

The statistical analyses performed are described in Supplementary materials and methods.

Additional Information

How to cite this article: Schlitter, A. M. et al. Molecular, morphological and survival analysis of 177 resected pancreatic ductal adenocarcinomas (PDACs): Identification of prognostic subtypes. Sci. Rep. 7, 41064; doi: 10.1038/srep41064 (2017).

Publisher's note: Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.