Therapeutic Implications of Molecular Subtyping for Pancreatic Cancer


In this article, we review seminal articles that have evaluated the molecular architecture of pancreatic cancer. We compare the methods used and the molecular subtypes defined, and assess the predominant subgroups in order to better understand which therapies may improve patient outcomes.

The prognosis of metastatic pancreatic adenocarcinoma has recently begun to improve. In the last several years, first-line therapy with gemcitabine plus nab-paclitaxel or a regimen of fluorouracil, leucovorin, irinotecan, and oxaliplatin (FOLFIRINOX) has boosted the median overall survival (OS) duration to 8.5 months and 11 months, respectively, in patients with metastatic pancreatic cancer, compared with a historic OS of only 6 months prior to 2011. Moreover, sequencing these two regimens improved median OS to an unprecedented 18 months. Notably, as newer agents become available and undergo testing, there is some indication that certain subgroups of patients may benefit dramatically from therapies targeting specific pathways in pancreatic cancer. There have been several attempts to assess the molecular differences in the driving mechanisms of pancreatic cancers, and to link these to specific therapies that could be remarkably effective in selected patients. These molecular analyses-based primarily on assessment of DNA mutations but also incorporating RNA sequencing and, in some cases, protein expression analysis-are beginning to reveal specific subtypes of pancreatic adenocarcinoma. Identification of the appropriate therapy for these subtypes may lead to further improved OS in the relevant patient populations. In this article, we review seminal articles that have evaluated the molecular architecture of pancreatic cancer. We compare the methods used and the molecular subtypes defined, and assess the predominant subgroups in order to better understand which therapies may improve patient outcomes.


Pancreatic adenocarcinoma is a nearly universally fatal disease and is one of the few cancer types that continues to increase in incidence. In 2016 in the United States, pancreatic cancer was diagnosed in approximately 53,070 patients and caused an estimated 41,780 deaths.[1] Indeed, pancreatic adenocarcinoma, one of the most lethal solid tumors, is expected to become the second leading cause of cancer-related deaths in the United States in the next few years.[2] While the prognosis for patients with metastatic pancreatic adenocarcinoma continues to be very poor, improved outcomes have recently been demonstrated in patients treated with combination chemotherapy. With the availability of new therapies that have undergone clinical testing, it has become clear that agents targeting specific pathways can yield significant clinical benefits in certain subgroups of patients. Thus, there have been several attempts to assess the molecular differences in the driving mechanisms of pancreatic cancers, and to link these differences to specific therapies that could be remarkably effective for selected patients.

Conventional Pathologic Subtyping of Pancreatic Neoplasms

In general, the term “pancreatic cancer” encompasses a mix of pathologies in which the cells originate in the pancreas and are thought to be epithelial in origin; however, the question of the cellular origin of the disease continues to be debated.[3,4] Even with access to modern-day molecular profiling, histology that distinguishes the exact pathologic subtype of pancreatic lesions is still the most clinically informative piece of data (ie, for diagnosis and prognosis). Exocrine pancreatic cancers can have a gland-like appearance (when well differentiated) and are the most common types of pancreatic cancer; the majority are classified as pancreatic ductal adenocarcinoma (commonly referred to as PDAC or PDA). Other forms of exocrine cancers are rare but are typically cystic or can be acinar cell carcinomas. Endocrine cancers are a small subset of all pancreatic cancers (~5%). Pancreatic neuroendocrine tumors are classified as functional (hormone-producing) or nonfunctional (non–hormone-producing).

In Table 1, we have outlined the molecular and histologic features that are traditionally identified in the various pathologic subtypes.[5-9] While histologic features are still used to define and diagnose pancreatic cancer subtypes, more recently, modern molecular techniques such as next-generation DNA sequencing and RNA sequencing have been used to further classify pancreatic cancer subtypes. The hope is to discover better targets and improve the therapeutic options.

Survival Outcomes: Still Based on One-Size-Fits-All Treatment Plans

Despite the dismal prognosis of pancreatic cancer, real progress has been made in terms of improved radiologic response rates and survival duration.[10,11] Studies have also shown a widening of the overall survival (OS) curve, reflecting an increase in the number of patients for whom chemotherapy has provided an exceptional benefit. For example, combination therapy with fluorouracil, leucovorin, irinotecan, and oxaliplatin (FOLFIRINOX)[10] or with gemcitabine plus nab-paclitaxel[11] yielded much higher response rates than treatment with gemcitabine alone. Moreover, sequencing of these modern regimens improved median OS to 18 months.[12] Further, nearly 20% of patients treated with FOLFIRINOX were alive at 18 months, compared with only 6% treated with gemcitabine alone.[10] Notably, a small subgroup of patients in both trials survived beyond 3 years with metastatic disease.[10,11]

It is interesting to consider that, despite reported median progression-free survival times of only 6.5 months with FOLFIRINOX[10] and 5.5 months with gemcitabine plus nab-paclitaxel,[11] extended survival times were observed. These outcomes suggest that patients who survive with their disease for 18 to 24 months or longer either have exceptionally favorable prognostic factors or have experienced a significant therapeutic benefit that translated into an OS benefit of 1 year or longer after completion of treatment. In fact, given the statistically significant and clinically meaningful differences in the 18-month survival rate among patients treated with FOLFIRINOX or gemcitabine plus nab-paclitaxel rather than gemcitabine alone, the reported survival improvements are likely to be due specifically to characteristics that predict for (and have translated into) an enhanced benefit from chemotherapy. Consequently, there is an opportunity to further enhance patient outcomes by identifying the molecular characteristics of the patients who have shown a better response to each chemotherapy regimen.

It is also evident that the type of targeted therapy used plays a role in the treatment of selected patients with pancreatic cancer, with anecdotal and early-phase reports describing subgroups of patients who have experienced exceptional outcomes. Most notably, patients with genetic defects in the homologous recombination DNA repair pathway (HRD) have experienced significant clinical benefit from poly (ADP-ribose) polymerase (PARP) inhibitor–based and/or platinum-based therapies.[13-16] There is a growing body of evidence that patients treated with regimens that include PARP inhibitors have dramatically higher rates of response and longer survival than patients with typical pancreatic adenocarcinoma that is managed with standard chemotherapy. For example, O’Reilly et al recently presented outcomes for pancreatic cancer patients treated with the combination of gemcitabine, cisplatin, and veliparib.[16] The response rate among patients with confirmed BRCA1 or BRCA2 mutations was 66%, and mean survival time after enrollment was 8.4 months (with several patients alive more than 2 years from the time of diagnosis). The hope in pursuing molecular profiling of pancreatic cancers is to identify additional subgroups of patients with pancreatic cancer who achieve dramatic responses to treatment with targeted therapies-although these subgroups have yet to be defined. Table 2 highlights important studies of molecular subgrouping of pancreatic cancers. Table 3 compares signaling pathways and relevant genes identified in selected studies. Table 4 shows that the identified pancreatic subtypes were generally comparable across several studies, although the study authors sometimes used different labels to define a particular subtype. The Figure is a Venn diagram showing the core signaling pathways that were identified in several seminal pancreatic cancer molecular subtyping studies.

Seminal Studies of Molecular Profiles in Pancreatic Cancer

Jones et al, 2008

Jones et al conducted one of the earliest studies of molecular subtyping in pancreatic cancer.[17] The authors sequenced more than 20,000 genes, assessing samples for homozygous deletions and amplifications in the tumor DNA; initially they tested a discovery set of 24 samples, then evaluated a larger validation set of 90 samples. Using pathway analysis, they identified 12 core signaling pathways that were altered in most pancreatic cancer specimens. Molecular alterations were identified in genes and pathways that are classically associated with pancreatic cancers (such as KRAS, TP53, SMAD4, and CDKN2A).

However, in their analysis of the findings, the group acknowledged that many samples harbored overlapping mutations of genes included in the 12 evaluated pathways; further, the direct therapeutic relevance of each of the subgroups is unclear, primarily because many of the pathways are not (currently) “targetable.” One of the major limitations of this study is that the discovery set comprised only 24 patient samples, none of which were primary tumor samples. In addition, the investigators used cell lines and xenografts generated from surgically resected patients and autopsy patients. Finally, given the importance of the HRD patient subgroup in pancreatic cancer, it was surprising that no tumors carried mutations in BRCA1, BRCA2, PALB2, ATM, ATR, RAD51, or FANCD2 genes.

Collisson et al, 2011

These investigators used DNA microarray–based gene expression profiling to analyze 27 resected tumor specimens from the University of California, San Francisco (for which survival data were available),[18] comparing the findings to those from a dataset from Badea et al.[19] Evaluating the clustering of gene expression patterns, they identified three specific subtypes, which they defined as classical, quasi-mesenchymal, and exocrine-like (Table 4). Consistent with this labeling, the classical subtype demonstrated high levels of expression of genes associated with epithelium and with cell adhesion, indicating a well-differentiated phenotype. The quasi-mesenchymal subtype showed high expression of mesenchymal-associated genes, suggesting a histology transitioning from well-differentiated to poorly differentiated cells. Finally, the exocrine-like subtype showed high expression of genes associated with digestion. These three subtypes were tested and validated across three additional published datasets.[20-22] Importantly, Collisson et al clearly demonstrated that the three subtypes provided robust prognostic information, since patients with resected classical tumors experienced better outcomes than those with resected quasi-mesenchymal tumors.

Collisson et al also explored the therapeutic implications of two of these three subtypes in vitro-having demonstrated that the same DNA microarray signatures could be used to subdivide 19 human and 15 mouse pancreatic cancer cell lines into classical and quasi-mesenchymal subtypes. In vitro, they demonstrated that the classical subtype was more dependent upon KRAS activity, and thus potentially more susceptible to treatment directed at KRAS. Similarly, knockdown of GATA6 expression reduced the growth of classical cell lines but not quasi-mesenchymal cell lines.[18] The authors’ attempts to evaluate the distinct pancreatic cancer subtypes as predictors of response to standard-of-care therapy were hampered by the limitations of the “standards” at that time: gemcitabine and erlotinib. Whereas minor differences in therapeutic response were observed for both agents (with gemcitabine being more active in the quasi-mesenchymal cells, and erlotinib having greater activity in the classical cells), the differences were small and the findings are not clearly applicable to the current modern regimens; the same subtype comparisons would need to be made in vitro but with the use of FOLFIRINOX or gemcitabine plus nab-paclitaxel.

Biankin et al, 2012

In 2012, these investigators published their first in a series of studies examining whole-exome sequencing and copy number analysis in tumor samples from patients with resected early-stage pancreatic cancer.[23] They examined 99 samples (that had > 20% cellularity and/or ≥ 10 validated somatic mutations) from 142 candidate patients, focusing on genes that were mutated in 2 or more samples. They demonstrated mutations identified in other analyses as commonly associated with pancreatic cancers, including KRAS, TP53, CDKN2A, SMAD4, MLL3, TGFBR2, ARID1A, and SF3B1. They also identified novel genes that were mutated, and were able to classify these according to their functional roles, including genes involved in chromatin modification (EPC1 and ARID2) and the DNA repair gene ATM, with the latter identified in 8% of cases. In addition, copy number analysis revealed abnormalities in similar genes, including the tumor suppressor genes CDKN2A and SMAD4. Interestingly, besides identifying mutations that are therapeutically relevant in DNA repair, in several genes involved in axonal guidance, the authors identified mutations reported to be associated with tumor cell invasion and metastasis (eg, SLIT2, ROBO2, SEMA3A, SEMA3E, and PLXNA1). However, they did not assess direct therapeutic targeting of the gene products in vitro.

Waddell et al, 2015

These investigators analyzed whole-genome sequencing in 100 patients with resected pancreatic cancer.[24] This analysis led to classification into four distinct subtypes:

1) The stable subtype represented 20% of samples, and contained fewer (≤ 50 structural) variation events than the other subtypes. The prevalence of typical mutations identified in pancreatic cancer, such as KRAS, SMAD4, and TP53, was similar to that of other subtypes (with the rate of TP53 mutation only slightly lower, at 61% rather than 70%).

2) The locally rearranged subtype represented 30% of samples; it was termed “local” because a significant genetic abnormality was found on only one or two chromosomes. Notably, known oncogenes were amplified in one-third of locally rearranged cases. In many cases, these mutations were not therapeutically targetable (eg, in contrast to mutations in KRAS), but there were patients with amplifications in targetable genes, such as ERBB2, MET, CDK6, PIK3CA, and PIK3R3. Individually, these genes were amplified with low prevalence (1% to 2%) for the whole dataset population.

3) The scattered subtype represented 36% of all samples, and harbored a moderate number of structural variation events (< 200).

4) The unstable subtype, which represented 14% of the group, had a large number of structural variation events, suggesting defects in the maintenance of DNA integrity. Tumors of this subtype typically harbored mutations in genes known to be involved in DNA repair, such as BRCA1, BRCA2, PALB2, ATM,FANCM, XRCC4, and XRCC6-and clinical results from these samples suggested that patients in this subgroup were also the most responsive to platinum-based regimens, including some exceptional responders with radiologic complete response to therapy.

Moffitt et al, 2015

These investigators evaluated a wide range of pancreatic cancer samples-including 145 primary resected tumors, 61 metastatic samples, and 17 pancreatic cancer cell lines-as well as normal adjacent tissue from 46 pancreas specimens and 88 samples from distant tissues.[25] They also validated their findings using RNA expression analysis data from a variety of pancreatic cancer specimens, and from cancer-associated fibroblasts. Unique to this effort was a digital separation of gene expression between tumor, stroma, and “normal” patient tissue. The authors identified two tumor-specific subtypes: “classical” and “basal-like.” Patients with the basal-like subtype had a significantly worse prognosis than those with classical tumors. The analysis of the stroma was highly relevant, since it revealed two stromal subtypes: “normal” and “activated.” An activated stroma was associated with a significantly worse prognosis. Interestingly, specific combinations of tumor and stromal subtypes had a cumulative effect on prognosis; the prognosis was best for patients who had classical tumors with a normal stroma, while those with basal-like tumors with an activated stroma had the worst prognosis.

Witkiewicz et al, 2015

This investigation of 109 specimens of resected pancreatic cancer[26] was, in some ways, a modern update of the previously described study by Jones et al.[17] The authors microdissected the tumor specimens, then performed copy number analysis and whole-exome sequencing; all of the resulting data were able to be linked to patient outcomes. While the group did not attempt to identify specific subtypes of pancreatic cancer, they did identify several key signaling pathways in pancreatic cancer specimens, including the TGF-β; Notch; β-catenin (the Notch and Wnt categories were combined in the study by Jones et al[17]); hedgehog; retinoblastoma, or RB; SWI/SNF; and DNA repair pathways. KRAS alterations were prevalent, but KRAS signaling was not defined by the authors as a specific pathway, as in the case of Jones et al, for example. Several associations were also made between molecular alterations and prognosis. MYC mutations, for example, were associated with a poor prognosis. In addition, in many cases, the sequencing process revealed potentially targetable molecular abnormalities, such as alterations in DNA repair (15% of samples), and BRAF V600E alterations (3% of samples).

Bailey et al, 2016

These investigators used a combination of whole-genome sequencing and deep-exome sequencing to evaluate 456 specimens of resected pancreatic cancer.[27] They identified genetic mutations that were grouped into 10 mechanistic classifications. Importantly, the classifications (as in Jones et al[17]) were often overlapping-that is, these were not distinct subgroups. These included:

• 92% with KRAS mutations.

• 78% with cell cycle checkpoint mutations (eg, TP53, CDKN2A, and TP53BP2).

• 47% with aberrations in transforming growth factor (TGF)-β signaling (eg, SMAD4, SMAD3, TGFBR1, TGFBR2, ACVR1B, and ACVR2A).

• 24% with mutations leading to histone modification (eg, KDM6A, SETD2, and activating signal cointegrator 2 complex [ASCOM] members MLL2 and MLL3).

• 14% with mutations in the switch/sucrose nonfermentable, or SWI/SNF, complex (eg, ARID1A, PBRM1, and SMARCA4).

• 17% with germline or somatic mutations in the BRCA pathway (eg, BRCA1, BRCA2, ATM, and PALB2).

• 5% with Wnt signaling pathway defects (RNF43 mutation).

• 16% with defects in RNA processing genes (eg, SF3B1, U2AF1, and RBM10).

• Tumors with defects in Notch signaling.

• Tumors with defects in Slit-Robo signaling.

The investigators employed RNA expression analysis to identify transcriptional networks related to these gene mutations. For the initial unsupervised analysis, they used a subgroup of 96 samples with a high epithelial tumor content. The RNA expression analysis enabled the clustering of the cancer specimens into four distinct subtypes that were also found to be present in a subsequent analysis of 232 specimens, irrespective of tumor cellularity. The four subtypes were: squamous; pancreatic progenitor; immunogenic; and aberrantly differentiated endocrine exocrine, or ADEX.

The squamous subtype of tumors harbored a greater rate of mutation in TP53 and KDM6A and demonstrated mutations in gene networks involved in inflammation, hypoxia response, metabolic reprogramming, TGF-β signaling, MYC pathway activation, autophagy, and upregulated expression of TP63ΔN and its target genes. Most significantly, the squamous subtype of tumors carried a poor prognosis. The pancreatic progenitor subtype of tumors aberrantly expressed genes involved in pancreatic development (eg, PDX1, MNX1, HNF4G, HNF4A, HNF1B, HNF1A, FOXA2, FOXA3, and HES1). There was also an enrichment in inactivating mutations of TGFBR2. The authors did not report on any correlations between expression of specific genes and prognosis.

The immunogenic subtype of tumors was characterized by expression of genes involved in pathways mediating acquired immune suppression. Tumor cells had significant immune infiltration and upregulation of cytotoxic T-lymphocyte–associated antigen 4 and programmed death 1 immune suppression pathways. These findings suggest that patients with this pancreatic cancer subtype may benefit from treatment with immunotherapeutics designed to overcome these two immune suppression pathways. The ADEX subtype of tumors was characterized by expression of genes involved in later stages of both pancreatic exocrine and endocrine cell development.

Discussion: Therapeutic Implications of the Subtypes

Pancreatic cancer remains an almost universally fatal disease. However, with the progress made in recent years, there is unquestionable heterogeneity in patient outcomes-with some resected patients being cured of their disease, and some with metastatic disease experiencing prolonged survival. The deep genetic analyses detailed in this review aim to genetically categorize pancreatic tumors into distinct biologic subgroups, with the goal of discovering robust prognostic and predictive biomarker signatures to enhance patient outcomes. Comparison of these key studies has revealed that we have the capacity to categorize pancreatic cancer patients based on different techniques and from independent cohorts, but unfortunately much work remains to be done in translating this genetic information into clinical practice.

Cross-comparison of the studies discussed in this article does demonstrate that reproducible biologic subgroups are emerging in pancreatic cancer. While the nomenclature differs, as shown in Table 4, there are strong biological similarities between the classical (Collisson and Moffitt), stable (Waddell), and pancreatic progenitor (Bailey) subtypes. Likewise, the quasi-mesenchymal (Collisson), unstable (Waddell), squamous (Bailey), and basal-like (Moffitt) subtypes are biologically similar. Finally, the exocrine-like (Collisson) and ADEX (Bailey) subtypes have overlapping characteristics. However, these subgroups are by no means perfectly overlapping. For example, Bailey et al[27] point out that only 50% of their squamous subtype samples overlap with the basal-like subgroup from Moffitt.[25] Furthermore, Waddell[24] and Bailey delineate other subtypes that are not well defined by Collisson[18] and Moffitt, including the locally rearranged and scattered subtypes described by Waddell, and the immunogenic subtype detailed by Bailey. Finally, as Moffitt et al point out, the distinctions may be due, in part, to the tissues assessed. Because Moffitt et al separated tumor tissue from stroma, and compared these cells against normal tissue, it appears that some aspects of the subtype designation may be confounded by the gene expression of the predominant tissue.[25] In particular, the genes expressed in Collisson’s quasi-mesenchymal subtype overlap in part with genes used by Moffitt et al to distinguish subgroups of the stroma.

Nevertheless, a critical question is whether these subgroups offer more prognostic/predictive clinical information about pancreatic cancer than conventional pathology information. Indeed, comparisons of the defined subtypes suggest some prognostic relevance. In particular, the prognosis of the quasi-mesenchymal/squamous/basal-like subtype is worse than that of the classical/pancreatic progenitor subtype. In addition, there are potential predictive patterns suggesting that the quasi-mesenchymal/squamous/basal-like subtype seems to be more responsive to chemotherapy than the classical/pancreatic progenitor subtype. Validation of these observations could have a tremendous impact: because we know that, despite “standard” chemotherapy, the vast majority of patients with resected pancreatic cancer will experience recurrence and ultimately die from their disease, there may be an opportunity to focus on the group of patients who can actually benefit from adjuvant chemotherapy. Perhaps, for example, patients with resected classical/pancreatic progenitor tumors do not actually gain any benefit from current standard adjuvant chemotherapy, and novel adjuvant therapies need to be identified. In contrast, perhaps standard adjuvant chemotherapy should be reserved only for patients with resected quasi-mesenchymal/squamous/basal-like subtypes. Adjuvant chemotherapy is clearly standard of care,[28,29] so these hypotheses would, of course, need to be tested in a clinical trial.

Furthermore, there is the question of whether patients with certain subtypes of pancreatic cancer are more appropriate candidates for novel therapies. For example, Bailey et al[27] inferred that the immunogenic subgroup will be more responsive to immunotherapies. Their reasoning is biologically sound, but given the recent history of a lack of benefit from immunotherapies in pancreatic cancer, this hypothesis, too, must be thoroughly tested in clinical trials, or at least by retrospective analyses. Biankin et al[23] did define a novel subgroup of patients with potentially targetable molecular abnormalities, focusing on the aberrant expression of the axonal guidance genes. Clinical trials utilizing appropriate novel agents in patients with aberrant Slit-Robo signaling would be worth pursuing, but a review of the database shows that no such trial has yet been initiated.

In fact, there is not a clear connection between defining pancreatic cancer subtypes genetically and identifying therapeutically targetable pathways. Nevertheless, sweeping genetic analyses of the types performed in the key studies discussed in this article do reveal several key therapeutically relevant groups of tumors. The most relevant subgroup by far is the 12% to 17% of patients with mutations in the HRD pathway. While this group is often characterized as having BRCA pathway abnormalities, the tumors can also harbor multiple mutations that result in similar deficiencies in achieving DNA repair through homologous recombination. These include tumors with mutations in BRCA1 and BRCA2, ATM (which Biankin et al reported in up to 8% of patients[23]), PALB2, ATR, RAD51, and others.

Patients with these HRD defects (at least those with known BRCA mutations) have responded robustly to PARP inhibitor–based therapy,[13-16] and routine, systematic genetic evaluation of patient tumors for these abnormalities is arguably as important as screening lung cancer tumors for abnormalities in EGFR, ALK, or ROS1. Secondarily, patients with tumors that harbor HRD defects, particularly BRCA mutations, should also be screened for germline mutations of the identified genes. Such findings could have important clinical implications for cancer screening in patients and their family members. Furthermore, Waddell et al identified a subtype of patients (locally rearranged) with distinct, targetable molecular abnormalities, including amplifications in ERBB2, MET, CDK6, PIK3CA, and PIK3R3.[24] Similar targetable abnormalities were also identified by Witkiewicz et al.[26] While an individual amplification/mutation may represent 1% or less of the pancreatic cancer population, collectively as many as 5% to 10% of tumors may harbor a mutation, the targeting of which may lead to improved patient outcomes.


The body of work performed by the groups reviewed herein is a tour de force, and represents our collective ability to dive deep into the molecular makeup of a patient’s tumor. However, the goals of such efforts need to remain focused on trying to determine how best to improve outcomes for patients with a disease as deadly as pancreatic cancer. In the ongoing identification and evaluation of distinct molecular subtypes, clinicians continue to identify key prognostic factors, and to determine the predictive value of response to standard chemotherapies. Other relevant questions remain that may be obstacles in moving this subtyping to the clinic: Can subtypes change within a tumor (ie, how can we address intratumoral heterogeneity)? Can exposure to therapy change the subtype of a tumor? Can we design novel targeted drugs that can benefit all subtypes? Additionally, a byproduct of these deep genetic analyses will inevitably be to identify distinct, targetable molecular abnormalities; in order to benefit an individual patient, however, these abnormalities will need to be identified by real-time methods.

Financial Disclosure: The authors have no significant financial interest in or other relationship with the manufacturer of any product or provider of any service mentioned in this article.


1. American Cancer Society. Cancer facts and figures 2016. Accessed February 8, 2016.

2. Rahib L, Smith BD, Aizenberg R, et al. Projecting cancer incidence and deaths to 2030: the unexpected burden of thyroid, liver, and pancreas cancers in the United States. Cancer Res. 2014;74:2913-21.

3. Kopp JL, Sander M. New insights into the cell lineage of pancreatic ductal adenocarcinoma: evidence for tumor stem cells in premalignant lesions? Gastroenterology. 2014;146:24-6.

4. Kopp JL, von Figura G, Mayes E, et al. Identification of Sox9-dependent acinar-to-ductal reprogramming as the principal mechanism for initiation of pancreatic ductal adenocarcinoma. Cancer Cell. 2012;22:737-50.

5. Borazanci E, Millis SZ, Korn R, et al. Adenosquamous carcinoma of the pancreas: molecular characterization of 23 patients along with a literature review. World J Gastrointest Oncol. 2015;7:132-40.

6. Wu J, Matthaei H, Maitra A, et al. Recurrent GNAS mutations define an unexpected pathway for pancreatic cyst development. Sci Transl Med. 2011;3:92ra66.

7. Hackeng WM, Hruban RH, Offerhaus GJ, Brosens LA. Surgical and molecular pathology of pancreatic neoplasms. Diagn Pathol. 2016;11:47.

8. Distler M, Aust D, Weitz J, et al. Precursor lesions for sporadic pancreatic cancer: PanIN, IPMN, and MCN. Biomed Res Int. 2014;2014:474905.

9. Klimstra DS, Pitman MB, Hruban RH. An algorithmic approach to the diagnosis of pancreatic neoplasms. Arch Pathol Lab Med. 2009;133:454-64.

10. Conroy T, Desseigne F, Ychou M, et al. FOLFIRINOX versus gemcitabine for metastatic pancreatic cancer. N Engl J Med. 2011;364:1817-25.

11. Von Hoff DD, Ervin T, Arena FP, et al. Increased survival in pancreatic cancer with nab-paclitaxel plus gemcitabine. N Engl J Med. 2013;369:1691-703.

12. Portal A, Pernot S, Tougeron D, et al. Nab-paclitaxel plus gemcitabine for metastatic pancreatic adenocarcinoma after FOLFIRINOX failure: an AGEO prospective multicentre cohort. Br J Cancer. 2015;113:989-95.

13. Lowery MA, Kelsen DP, Stadler ZK, et al. An emerging entity: pancreatic adenocarcinoma associated with a known BRCA mutation: clinical descriptors, treatment implications, and future directions. Oncologist. 2011;16:1397-402.

14. Kaufman B, Shapira-Frommer R, Schmutzler RK, et al. Olaparib monotherapy in patients with advanced cancer and a germline BRCA1/2 mutation. J Clin Oncol. 2015;33:244-50.

15. Pishvaian MJ, Wang H, Zhuang T, et al. A phase I/II study of ABT-888 in combination with 5-fluorouracil (5-FU) and oxaliplatin (Ox) in patients with metastatic pancreatic cancer (MPC). J Clin Oncol. 2013;31(suppl 4):abstr 147.

16. O’Reilly EM, Lowery MA, Segal MF, et al. Phase IB trial of cisplatin (C), gemcitabine (G), and veliparib (V) in patients with known or potential BRCA or PALB2-mutated pancreas adenocarcinoma (PC). J Clin Oncol. 2014;32(suppl 5s):abstr 4023.

17. Jones S, Zhang X, Parsons DW, et al. Core signaling pathways in human pancreatic cancers revealed by global genomic analyses. Science. 2008;321:1801-6.

18. Collisson EA, Sadanandam A, Olson P, et al. Subtypes of pancreatic ductal adenocarcinoma and their differing responses to therapy. Nat Med. 2011;17:500-3.

19. Badea L, Herlea V, Dima SO, et al. Combined gene expression analysis of whole-tissue and microdissected pancreatic ductal adenocarcinoma identifies genes specifically overexpressed in tumor epithelia. Hepatogastroenterology. 2008;55:2016-27.

20. Balagurunathan Y, Morse DL, Hostetter G, et al. Gene expression profiling-based identification of cell-surface targets for developing multimeric ligands in pancreatic cancer. Mol Cancer Ther. 2008;7:3071-80.

21. Pei H, Li L, Fridley BL, et al. FKBP51 affects cancer cell response to chemotherapy by negatively regulating Akt. Cancer Cell. 2009;16:259-66.

22. Grutzmann R, Pilarsky C, Ammerpohl O, et al. Gene expression profiling of microdissected pancreatic ductal carcinomas using high-density DNA microarrays. Neoplasia. 2004;6:611-22.

23. Biankin AV, Waddell N, Kassahn KS, et al. Pancreatic cancer genomes reveal aberrations in axon guidance pathway genes. Nature. 2012;491:399-405.

24. Waddell N, Pajic M, Patch AM, et al. Whole genomes redefine the mutational landscape of pancreatic cancer. Nature. 2015;518:495-501.

25. Moffitt RA, Marayati R, Flate EL, et al. Virtual microdissection identifies distinct tumor- and stroma-specific subtypes of pancreatic ductal adenocarcinoma. Nat Genet. 2015;47:1168-78.

26. Witkiewicz AK, McMillan EA, Balaji U, et al. Whole-exome sequencing of pancreatic cancer defines genetic diversity and therapeutic targets. Nat Commun. 2015;6:6744.

27. Bailey P, Chang DK, Nones K, et al. Genomic analyses identify molecular subtypes of pancreatic cancer. Nature. 2016;531:47-52.

28. Neoptolemos JP, Palmer DH, Ghaneh P, et al. Comparison of adjuvant gemcitabine and capecitabine with gemcitabine monotherapy in patients with resected pancreatic cancer (ESPAC-4): a multicentre, open-label, randomised, phase 3 trial. Lancet. 2017 Jan 24. [Epub ahead of print]

29. National Comprehensive Cancer Network Guidelines. Pancreatic adenocarcinoma (version 2.2016). Accessed February 14, 2017.