Prognostic Value of Molecular Subtypes in Pancreatic Cancer
Pancreatic ductal adenocarcinoma is one of the most aggressive human cancers with 338,000 new cases and 331,000 deaths in 2012 worldwide.1 Its incidence is rising.2 The 5-year survival is dismal, inferior to 10%. Complete surgical removal of the tumor followed by adjuvant chemotherapy is the only curative treatment, but less than 20% of patients are candidate to surgery, and even after such treatment, most of them will have a relapse and die. Despite the testing of multiple agents for many years, only few chemotherapy drugs (gemcitabine with or without nab-paclitaxel, FOLFIRINOX regimen combining 5-fluorouracil, leucovorin, oxaliplatin, and irinotecan) and 1 targeted therapy (erlotinib) have shown some degree of efficacy in the unselected whole population and are approved in clinics. However, the survival benefit is modest, making crucial not only the development of novel drugs, but also the identification of patients most likely to benefit from one or another treatment, as exemplified in non–small cell lung cancer with epidermal growth factor receptor inhibitors.3
Clinical heterogeneity of cancers has been better understood during the last few years, thanks to high-throughput molecular analyses that allowed the discovery of previously unrecognized tumor subtypes.4 For example, gene expression profiling revealed a new molecular classification of breast cancer including at least 5 biologically and clinically relevant subtypes,5 linked to major molecular alterations of disease and to mammary cell types, as well as to epidemiological specificities, different rates of therapeutic response, and different outcomes. Today, breast cancer is no longer regarded as a single disease, but as a collection of separate diseases.6
Regarding pancreatic cancer, during the last decades, research efforts have identified some key molecular alterations, such as KRAS, TP53, SMAD4, CDKN2A, and ARID1A mutations and GATA6 amplification,7,8 but without clinical application to date. Recently, a transcriptional classification of pancreatic ductal carcinoma including 4 tumor subtypes associated with different molecular pathways was reported.9 The subtypes were named squamous, pancreatic progenitor, immunogenic, and aberrantly differentiated endocrine exocrine (ADEX) based on the differential expression of transcription factors and downstream targets important for lineage specification and differentiation during pancreas development and regeneration. Three of these subtypes directly overlapped with another molecular prognostic classification previously reported,10 suggesting their robustness. Interestingly, the Bailey’s subtypes were associated with overall survival (OS): the squamous subtype was an independent poor-prognosis factor, but the series was relatively small with 96 informative samples.
To extend such prognostic value in a larger cohort, we gathered clinicopathologic and gene expression data of clinical pancreatic carcinoma samples from 11 public data sets, collected from the National Center for Biotechnology Information/Genbank GEO, ArrayExpress, and TCGA databases (Supplementary Table 1, http://links.lww.com/MPA/A568). Samples had been profiled using DNA microarrays and RNASeq. Before analysis, expression data were normalized within and between all data sets as previously described.11 The pooled data set contained 609 primary cancer samples clinically annotated. We applied Bailey’s classification to our pooled data set: 231 samples were identified as squamous (38%), 142 as pancreatic progenitor (23%), 90 as immunogenic (15%), and 146 as ADEX (24%). We then searched for correlations between this classification and clinicopathologic data, including survival. No association was found with patient’s age and sex, American Joint Committee on Cancer stage, and pathological type, tumor size, and lymph node status, whereas a correlation was found with pathological grade (Supplementary Table 2, http://links.lww.com/MPA/A569).
Regarding prognosis, OS was available for 364 patients: it was defined as time from diagnosis to death, irrespective of cause. With a median follow-up of 9 months (range, 1–84 months), 188 patients died, and the 2-year OS was 40% (95% confidence interval [CI], 34%–47%) (Fig. 1A).