To be able to adequately interpret the prognostic significance of these putative markers, we have to correctly address and define the resection margins status, which is clearly one of the main prognostic factors of (locoregional) recurrence and outcome in pancreatic cancer [4-7].In the adjuvant setting, a large meta-analysis of five randomized trials reported that RCT was not associated with an increase in the risk of death [HR = 1.09; 95% CI: 0.89–1.32], median survivals being 15.8 months with, and 15.2 months without, RCT in the whole population [8]. In subgroup analysis, there was a significant difference in the effect of RCT dependent on R0 or R1 resection margins (p = 0.04); RCT being considered as more effective in patients with R1 resection. However, the rates of R1 resection reported in these five trials were very heterogeneous, varying from 17% to 82% [8]. Such heterogeneity is also observed in the literature with a large difference between series according to the centers and the procedures used for pathological examination, reflecting the absence of standardization [17-20]. In International Union Against Cancer (UICC) classification, R1 resection is referred to tumor cells observed by microscopy at one or more edges of the resection specimen [21], while the Royal College of Pathologists' guidelines defined R1 resection as the presence of tumor from 1 mm or less from the circumferential resection margin (CRM), or surface of the resection specimen, either by direct tumor invasion or lymph node involvement [22]. For anatomic reasons, and at the difference of adjacent organ and pancreatic transection margins, the CRM requires a specific pathological procedure to be correctly assessed. The Royal College of Pathologists' guidelines recommend evaluating the CRM in three different areas according to a plane perpendicular to the duodenal axis: the superior mesenteric vein groove separates the anterior CRM from the posterior CRM. In using this specific protocol, the rate of R1 resection is of about 80%, considerably higher in comparison with historical series [19,20]. However, the prognosis significance of R1 resection is probably not the same for all resection margins. In a recent retrospective study, involvement of the beds of the major mesenteric vessels and/or transection margins (n = 44, 44%) had a worse prognostic significance while there was no significant difference in survival between patients with only R1 anterior and/or posterior CRM (n = 61, 56%) compared with those with R0 resection (n = 39, 26%) [23].

Regarding the independent prognostic significance of R1 resection in randomized adjuvant trials [7,24], and its potential impact upon RCT indication, a rigorous and standardized examination of the resected pancreatic specimen is thus necessary. Moreover, as long as the prognostic significance of the different R1 resection margins is not clearly established, the real impact of others markers predicting patients' outcome cannot be adequately assessed.

Transforming growth factor-β (TGF-β) signaling pathways are involved in tumor pathogenesis and progression through different effects on cell differentiation, proliferation and invasion [25]. In PAC, a global genomic analysis has reported that perturbations of TGF-β signaling were found in 100% of the 24 analyzed tumors [26]. In the same way, it was shown that TGF-β ligands and the type II TGF-β receptor were overexpressed in cDNA microarray analysis as in quantitative proteomic analysis in PAC [27-29], and enhanced expression of TGF-β receptor II has been reported as a worse prognostic factor [30].

Functionally, the binding of TGF-β ligands induces the formation of heterotetrameric active receptor complexes that regulate the activation of downstream SMAD and non-SMAD pathways. In the SMAD pathway, phosphorylation of SMAD proteins by activated TGF-β receptor complexes lead to the formation of heteromeric complexes with the common partner SMAD4, also termed DPC4. Then, activated SMAD complexes translocate into the nucleus where they interact with other transcription factors to regulate expression of various genes [31]. SMAD4 was originally isolated from human chromosome 18q21.1 as a tumor suppressor gene, with about 55% of pancreatic cancers bearing deletions or mutations [32]. Subsequent to the demonstration that immunohistochemical labeling for SMAD4 is a sensitive and specific marker for SMAD4 alterations, the prognostic value of SMAD4 expression has been assessed in patients with resected or locally advanced PAC [33]. However, results of the two largest published studies were discordant, SMAD4 protein expression being reported as a good prognostic factor in one [34], but, contrarily, as a worse prognostic factor in the other [35]. More recently, in another study of 89 patients with resected PAC, SMAD4 gene inactivation (identified using high-density oligonucleotide array) was associated with shorter overall survival in multivariate analysis [36]. In fact, the population of patients included in these three studies was not comparable, particularly with regards to adjuvant treatment. SMAD4 inactivation was reported as a worse prognostic factor in a monocentric study in which patients received adjuvant RCT [n = 249; HR = 1.36, 95% CI: 1.03–1.81, p = 0.042] [35], and in a second study in which most of the patients received adjuvant RCT and 5FU chemotherapy [HR = 1.92, 95% CI: 1.20–3.05, p = 0.006] [36]. Whereas, SMAD4 inactivation was a good prognostic factor in a study in which only 11% of patients received 5FU adjuvant chemotherapy [34]. Consequently, interactions between SMAD4 status and the type of adjuvant treatment may be a confounding bias and may explain the discordant results reported.

These conflicting results are not easy to understand, and may also reflect the complex network of TGF-β signaling in cancer [25,37,38]. On the one hand, TGF-β, via SMAD pathway, is a potent inhibitor of epithelial cell growth and survival through modulation of cell cycle regulators and activation of apoptosis, but these effects are highly dependent on cellular context and tumor microenvironments [37]. On the other hand, and in using others pathways, TGF-β can promote proliferation, migration, and the epithelial-to-mesenchymal transition (EMT) which is implicated in metastatic phenotype [38].

Results of a study in murine model of PAC (with KRAS mutation) and in cell lines, have well described the different effects of TGF-β signaling according to the presence or not of SMAD4 [39]. Briefly, SMAD4 deletion enhanced PAC pathogenesis directed by KRAS mutation, but also altered the tumor phenotype: tumors with SMAD4 deficiency were well-to-moderately differentiated and expressed epithelial markers, whereas tumors with intact SMAD4 frequently exhibited markers of EMT. Moreover, cell line responses to TGF-β were different according to SMAD4 status. In cell lines from mice with SMAD4 deficiency, TGF-β had no effect on proliferation and increased modestly cell migration whereas, in cell lines from mice with intact SMAD4, TGF-β had the opposite effect according to cell differentiation: in well or moderately differentiated cells, it inhibited cell migration, cell growth, and promoted apoptosis, but it enhanced migration and increased proliferation in undifferentiated cells [39]. These results are consistent with the dual function of TGF-β pathways in cancer, and with results recently reported in breast cancer cell lines in which the type of motility (“single” versus “cohesive”) was different according to SMAD4 status [40].

In summary, these data support the fact that SMAD4 status may be involved in the risk of metastatic diffusion. This has been recently addressed in 76 patients who died from PAC in whom pathological features of tumors were analyzed after autopsy and correlated to the pattern of failure. In this study, the loss of SMAD4 was significantly associated with metastatic disease in comparison with local recurrence [41]. Considering the small number of patients included in this study, any definite conclusion is unlikely at this time. The exact prognostic versus predictive role of SMAD4/TGF beta, requires larger prospective studies.

S100A2 overexpression has recently been reported as a good prognostic marker in the largest published translational study in patients with PAC (n = 601) [42]. In this study, in which 17 candidate markers have been evaluated by immunohistochemistry or in situ hybridization, S100A2 overexpression was the only marker significantly associated with longer overall survival [in subgroup of resected PAC, HR = 1.87; 95% CI: 1.25–2.81; p = 0.0024]. However, despite the number of patients included, and the use of a training set then a validation set, results reported cannot really be conclusive because patients who received or did not receive adjuvant treatments (chemotherapy and/or radiotherapy) were mixed in the analysis [42]. Thus, S100A2 seems to be a promising marker, but its real prognostic and/or predictive value should be assessed in other studies with adequate methodology. Moreover, S100A2 function in PAC pathogenesis is unknown. S100 family seems implicated in complex network and various functions. S100 proteins have been reported to be implicated in the pathogenesis of many cancers, but were suggested as tumor suppressors in some and as tumor oncogenes in others [43]. Consequently, additional studies should be conducted to try to understand the function and role of the S100A2 protein in the pathogeneis of PAC.

Chemokines are a large cytokine subfamily of chemotactic cytokines, specifically acting on the superfamily of G-protein-coupled seven-span transmembrane receptors. Many chemokines and chemokines receptors have been described with involvement in cell differentiation, trafficking, and activation [44]. Some of them are implicated in the development or progression of cancer, in particular the chemokine CXCL12 (also termed stromal cell-derived factor-1, SDF1) which binds to two receptors, the CXC receptor 4 (CXCR4) and the CXC receptor 7 (CCR7) [45]. CXCL12 is a homeostatic chemokine, constitutively expressed in several organs, and, in particular, the liver, lung and in bone marrow. The CXCL12/CXCR4 axis seems involved in angiogenesis, metastasis, and survival; cancer cells expressing CXCR4 migrate and spread along the CXCL12 gradients [45-48].

In PAC, in vitro and in vivo data have reported that CXCR4 overexpression was frequent while normal pancreatic tissue does not express CXCR4 [47,49]. Moreover, CXCR4 was more frequently expressed in cell lines derived from distant lesions (metastases or ascites) than in those from primary tumors [47]. CXCL12 was also found in PAC tissue samples [49]. The effects of CXCL12 on CXCR4 positive cell lines have been well documented, and result in various effects: it stimulates cell migration in a chemotaxis assay, enhances cell transendothelial migration, and induces matrix metalloproteases activation [47,48]. In a murine model using nude mice and injection of cancer cells into tail vein, CXCR4 expression was associated with a dramatic increase in liver and lung metastases [48]. Most of these effects were abolished or significantly reduced with the use of an anti-CXCR4 monoclonal antibody, even in cell line which produce spontaneously CXCL12 [47,48]. Results of CXCL12 on proliferation and survival of cells expressing CXCR4 are discussed with discordant data published [47,48]. More recently, the CXCL12/CXCR4 axis was also described to be involved in pancreatic intraepithelial neoplasias with an expression frequency which increases during their progression [50]. Overall, these results suggest that an autocrine/paracrine loop involving this axis may play a major role in PAC pathogenesis and in metastatic diffusion.

Clinically, the prognostic value of CXCR4 expression has been evaluated in 71 patients with resected PAC. In multivariate analysis, CXCR4 expression was an independent and strong worse prognostic marker for overall survival [HR = 5.55; 95%CI: 1.92-12.31; p < 0.001] as well as lymph node metastases and undifferentiated histology. CXCR4 expression was also significantly correlated with the proliferative index (Ki67) and the type of relapse: liver metastases occurred in 59% of patients with high CXCR4 expression in their tumor, and in 28% of those with low CXCR4 (p = 0.016) [51].

Interestingly, recent data have reported that CXCR4 expression may be influenced by the TGF-β pathway and the phenotype of cancer cells. In a hepatocarcinoma cell line in which the presence of TGF-β induced EMT without suppressor effects, CXCR4 expression was stimulated by TGF-β [52]. In PAC stem cells (defined as CD133 positive), CXCR4 was markedly overexpressed in comparison with others cancer cells. Moreover, coculture of cancer stem cells with pancreatic stromal cells which express CXCL12 was associated with an increased migration and invasion ability of cancer stem cells, and these effects were significantly reduced by CXCR4 downregulation using RNA interference [53]. These data highlight the major role of stroma in PAC, and the necessity to try to understand the complex interactions between cancer cells and stromal cells.

In summary, CXCR4 may represent a specific marker of distant relapse risk, and appears as an attractive target for therapeutic options.

Permeation of gemcitabine through the plasma membrane by diffusion is low and requires specialized integral membrane nucleoside transporters proteins. The Human Equilibrative Nucleoside Transporters (hENT) and the Human Concentrative Nucleoside Transporters (hCNTs) play a key role in the intracellular uptake of gemcitabine and thus are key determinants of gemcitabine efficacy. hENTs mediate equilibrative bi-directional transport nucleosides across membranes [54,55], whereas hCNTs transport nucleosides against their concentration gradients driven by sodium and/or proton coupled electro-chemical agents [56]. Among these transporters, hENT1 and to a lesser degree, hCNT3 mediates the majority of gemcitabine transport in preclinical models [57-60].

Preclinical studies, including studies involving PAC cells lines, have suggested a positive correlation between hENT1 gene expression and chemosensitivity. Cultured human cells that are either pharmacologically or genetically deficient for hENT1 exhibit resistance to gemcitabine [59], suggesting that hENT1 abundance may be used as a predictive marker for response to gemcitabine. Several small retrospective studies have evaluated whether the expression of hENT1 into the tumor correlated with patients' outcome [61,62]. Although all these retrospective studies suggested a prognostic role for hENT1 in PAC treated with gemcitabine, none of them studied controls who did not receive gemcitabine. Consequently, the true predictive value of hENT1 in gemcitabine-treated PAC (i.e., the ability to identify those patients most likely to benefit from gemcitabine) could not be assessed in these studies. Recently, however, results of the translational study performed in a cohort of PAC patients from the large prospective randomized RTOG 9704 trial, support the concept of hENT1 as a predictive factor of gemcitabine benefit [63].

Beside the nucleoside transporters, several enzymes involved in gemcitabine metabolism, such as deoxycytidine kinase (dCK), cytidine deaminase, RRM1, and RRM2, may have also a key role in altering intracellular disposition of the drug and determining response to gemcitabine. These proteins have had limited evaluation as biomarkers in PAC. Single, small, retrospective, clinical studies have suggested the poor prognostic value of high levels of RRM1 or RRM2 [64,65] gene expression and low deoxycytidine kinase [66,67] protein expression in patients with pancreatic cancer treated with gemcitabine.

In our efforts to guide treatment decision for patients with PAC, we might better define gemcitabine sensitivity and resistance by an integrated analysis of the expression of dCK, hENT1 and hCNT3 in larger cohorts of patients, ideally within the context of a randomized controlled trial. In this manner, it may be possible to more precisely define those subgroups of patients who would derive particular benefit from gemcitabine, and those patients who warrant investigation with experimental therapies or should be treated with non gemcitabine-based combinations.

Therapeutic options in metastatic PAC have recently increased with the results of the PRODIGE 4/Accord 11 trial that demonstrated the superiority of FOLFIRINOX regimen in comparison with gemcitabine alone [16]. Only patients with a performance status of 0–1 were included in this study, and treatment decision making could be clinically based on performance status; gemcitabine alone remaining the standard in patients with grade 2 performance status. However, apart from concerns of toxicities, FOLFIRINOX is probably efficient, whatever the performance status, and specific predictive biomarkers of this combination should therefore be evaluated. Moreover, erlotinib and gemcitabine combination was associated with a modest efficacy on overall survival in the whole study population while a subgroup of patients probably benefit from erlotinib addition [15]. The same reasoning can be applied to derivative platinum combinations [12,13]. In this context of increasing options, identification of predictive markers may be of great help to personalize the use of these drugs and to decide on the use or not of a gemcitabine-based regimen.

Erlotinib is a human epidermal growth factor receptor (EGFR) tyrosine kinase inhibitor. In the Phase III randomised trial in patients with advanced PAC, patients who received the erlotinib and gemcitabine combination had a significantly longer overall survival than those treated with gemcitabine alone (6.24 months vs. 5.91 months, p = 0.038) [15]. In subgroups analysis, the hazard ratio for survival was significant only in the subgroup with metastatic disease [HR = 0.79; 95% CI: 0.65–0.97], and not in the subgroup with locally advanced disease [HR = 0.94; 95% CI: 0.63–1.39]. In this study, EGFR level evaluated on archival tissue by immunohistochemistry had no prognostic or predictive value in both arms. However, the occurrence of skin rash was a predictive factor of overall survival in multivariate analysis [HR = 0.74; 95% CI: 0.56–0.98] [15]; this was confirmed in another large Phase III study [93].

Normally, activation of EGFR leads to a cascade of phosphorylation in many pathways: the RAS/RAF/MAPK, PI3K/AKT, and STAT pathways which are implicated in cellular proliferation, adhesion, angiogenesis, migration, and survival [94]. Notably, mutations of genes of these pathways are frequently found in most cancers. In PAC, KRAS mutations are observed in 75–90% of cases [95–97], EGFR mutations in 1%-2%, and EGFR amplification in 40%-50% [97,98]. In colon cancer, the negative predictive value of KRAS mutation status for the anti-EGFR antibodies efficacy has been reported first in retrospective studies then definitively confirmed by retrospective analysis of tumors of patients included in large prospective trials [99,100]. In wild-type KRAS colon cancer, EGFR amplification has also been reported as a predictive marker [101]. In non-small cell lung cancers, EGFR mutations, wild-type KRAS, and EGFR amplification have been described as predictive marker of erlotinib efficacy [102,103].

In PAC, the predictive value of KRAS mutations and EGFR amplification for erlotinib efficacy has been assessed from tumor samples of 26% of patients included in the prospective randomized Phase III trial [97]. The hazard ratio of death between the two arms was 0.66 [95% CI: 0.28–1.57] for patients with wild-type KRAS, and 1.07 [95% CI: 0.28–1.57] for patients with mutated KRAS (interaction; p = 0.38). There was also no significant difference between the two arms according to EGFR amplification. One of the reasons for these disappointing results is the possibility of interactions between gemcitabine and erlotinib effect in the combination arm and the confounding role of KRAS status as prognostic or predictive.

Consequently, predictive markers of erlotinib efficacy should be assessed in patients with PAC treated by erlotinib alone, or in combination with predictive markers of gemcitabine efficacy, both in the setting of adjuvant and advanced disease clinical trials.

To our knowledge, no study has evaluated predictive markers of oxaliplatin and cispatin efficacy and/or toxicity in patients with PAC. However, results reported in others cancers could lead to future research.

Oxaliplatin and cisplatin act through formation of DNA adducts which interfere with DNA replication. DNA repair enzymes, in particular, proteins of the nucleotide excision repair (NER) pathway, are thought to repair DNA damage caused by platinum agents, and play a major role in preventing cell death. The excision repair cross-complementation group 1 (ERCC1) protein functions in a complex with others NER proteins, and is required for the excision of the damaged DNA [104]. In vitro, the expression level of ERCC1 mRNA has been shown to be correlated with derivative platinum cytotoxicity [105,106]. Moreover, it has been suggested that the NER was implicated in the cytotoxic synergism observed between gemcitabine and cisplatin [107]. Clinically, many retrospective studies have reported a significant correlation between the expression level of ERCC1 mRNA and the efficacy of derivative platinum based chemotherapy in colorectal, esophageal, and gastric cancers [108-110]. Furthermore, in non-small-cell lung cancer, the expression level of ERCC1 assessed by immunohistochemistry has been reported as a significant predictive marker for overall survival in a retrospective analysis of an adjuvant Phase III trial in which patients were randomized between an arm of cisplatin-based adjuvant chemotherapy and an observation arm [111].

Several polymorphisms of NER enzymes have also been described. In patients with metastatic colorectal cancer, ERCC1 Asn118Asn (354C > T) polymorphism has been reported as a predictive marker of objective response in one study [112], and of overall survival in another [113]. However, results of two large retrospective studies in patients with colorectal cancer included in prospective trials did not confirm a predictive value of ERCC1 codon 118 polymorphism [85,114].

Glutahione S-transferases (GSTs) are a family of Phase II detoxification enzymes implicated in cellular resistance to drugs. Several enzymes are found in this class, with various activities and tissue specificities. Out of them, polymorphisms of GSTP1, GSTT1 and GSTM1 are associated with variations in their protein activity [115]. The predictive value of these polymorphisms has recently been assessed in three retrospective analyses in patients with metastatic gastric (one study) or colorectal cancer (two studies) [85,114,116]. In two of these studies, GSTP1 polymorphism was significantly associated with oxaliplatin-related neurotoxicity [114,116]. Whereas, the value of GSTT1 and GSTM1 polymorphisms was less clear [85,114,116]. Because of discordant results, additional studies in a larger cohort of patients are necessary to definitively conclude about the predictive value of these polymorphisms. In patients with metastatic PAC, their value to predict the toxicities from the FOLFIRINOX regimen should be evaluated.

Irinotecan is a topoisomerase I inhibitor of which the active metabolite SN38 is eliminated predominantly by glucuronidation. This reaction is principally mediated by UDP-glucuronosyltransferase 1 family polypeptide A1, which is encoded by the UGT1A1 gene. Activity and expression level of UGT1A1 depends on polymorphisms of TA repeats number in the TATA element of UGT1A1 gene [117]. The wild-type allele (UGT1A1*1) has six TA repeats, and the variant allele seven (UGT1A1*28). Patients who are homozygous for UGT1A1*28 have less capacity to eliminate SN38, and many studies have reported a significant increase risk of grade 3-4 neutropenia and/or diarrhea in those patients treated by irinotecan for metastatic colorectal cancer. However, a meta-analysis demonstrated that the risk of grade 3–4 neutropenia in patients with a UGT1A1*28/*28 genotype was correlated with the dose of irinotecan [118].

Irinotecan is active in pancreatic cancer cells [119] and was shown to have an activity in first- and second-line in patients with advanced PAC [120-122] Moreover, recent results of the PRODIGE 4/Accord 11 trial support the use of irinotecan in PAC [16]. To our knowledge, no study has evaluated the risk of neutropenia according to UGT1A1 polymorphisms in patients receiving FOLFIRINOX. In this regimen, the irinotecan dose is 180 mg/m². Whatever the type of regimen, the predictive value of UGT1A1 polymorphisms on outcome has not been yet established in PAC. Considering the high number of neutropenic events observed with FOLFIRINOX in the above study, this marker can be helpful.

At diagnosis, only 10% to 15% of patients with a PAC can have a curative resection, and these resected specimens constitute the best samples for translational research, both quantitatively and qualitatively, allowing analyses at the DNA, RNA, or protein level. Consequently, with the exception of patients with locally advanced tumor who had an exploratory laparotomy with tumor biopsies, most recent studies which reported prognostic and/or predictive markers have only included patients with resected PAC. However, considering that the large majority of patients have an advanced disease at diagnosis, new techniques to provide sufficient quantity/quality tumors samples should be developed in those patients. Such progress could allow inclusion of more patients in translational research studies, to accelerate them, and to assess if results reported in patients with resected tumors are valuable in patients with advanced tumors. In this way, development of robust and reproducible methods to obtain tumor samples from endoscopic ultrasound-guided fine-needle aspirations or biopsies, or from transparietal biopsies should be a priority [123]. Furthermore, besides cytopathologic samples, evaluation of systemic blood markers such as miRNA or SNPs, seem promising [124,125].

As we discussed earlier, most recent studies which reported prognostic markers have included patients with resected PAC. However, because of the relatively few number of resectable patients, most of these studies have included a mixed population of patients who received various adjuvant treatments and who were resected or not. Indeed, a prognostic marker discriminates a population of patients by the overall survival in the absence of treatment, or despite it, and consequently, the best setting to evaluate a prognostic marker remains a population of patients who did not receive adjuvant treatment. However, a predictive marker separates a group of tumors or patients by the response or resistance to a specific treatment. The best method to assess the predictive value of a marker for a specific treatment is to evaluate the benefit of this treatment in the different subgroups of patients stratified for this marker. For the assessment of the prognostic value of a marker, analyzing a mixed patients' population who received or did not receive adjuvant treatment can therefore result in discordant results and interpretation. As an example, microsatellite instability (MSI) phenotype, which is a good prognostic marker but a negative predictive marker of adjuvant 5FU benefit in stage II-III colon cancer [126,127], has initially and, surprisingly, been reported as a positive predictive marker of adjuvant 5FU benefit [128]. In this first study, authors compared patients with MSI phenotype tumors to those with microsatellite stability (MSS) tumors in subgroups defined by adjuvant 5FU administration. In the subgroup of patients who did not receive adjuvant 5FU, no survival difference was observed between MSI and MSS tumors (5-year survival: 37% vs. 32%; p = 0.82) whereas in the subgroup of patients who received adjuvant 5FU, patients with MSI tumors had a significantly better five year survival than those with MSS (90% vs. 35%; p = 0.0007) [128]. In this article, there is no clear explanation for the absence of MSI prognostic value in patients who did not receive adjuvant 5FU but this result was confounding. Whereas, the better survival of patients with MSI tumors in subgroup treated with adjuvant 5FU was certainly due to the good prognostic value of MSI and not to a predictive one. Subsequently, most studies that further evaluated the benefit of adjuvant 5FU in subgroups of patients, defined by MSI/MSS phenotype, showed that MSI phenotype was a negative predictive marker of adjuvant 5FU [126,127].

Prospective randomized clinical trials remain the gold standard to validate prognostic and/or predictive markers and avoid confounding interpretations on their relative signification depending on the setting of treatment [129].

During the last decade, Phase III randomized trials which evaluated cytotoxic combinations or targeted therapies in advanced PAC have accumulated negative results. Consequently, it seems today essential to modify our practice and to perform well-designed Phase II trials to identify new drugs or combinations that will harbor a high chance of success in subsequent Phase III studies [130]. Ideally, such studies should take into account our knowledge of PAC carcinogenesis, involved signaling pathways and potential predictive markers. For this, neoadjuvant Phase II trials in resectable patients are promising and may rapidly and safely discover signals of efficacy of an experimental regimen. During the window interval between diagnosis and surgery, the administration of a short neoadjuvant chemotherapy course would allow to evaluate treatment activity and constitute a test for subsequent adjuvant chemotherapy. Moreover, analysis of consecutive modifications in tumor and stroma could contribute to the identification of new targets or biomarkers, and to progress in our understanding of this disease. In the same way, prospective evaluation of predictive and/or prognostic markers should be an integral part of clinical trials in PAC with potentially a collection of tumors or blood samples from all enrolled patients [130].