Clinical Significance of the Genetic Landscape of Pancreatic Cancer and Implications for Identification of Potential Long-term Survivors

Pancreatic ductal adenocarcinoma (PDAC) is one of the most lethal solid malignancies and a major cause of cancer-related deaths in developed countries (1), with a >95% mortality rate. Most patients present with locally advanced or metastatic disease at initial diagnosis leaving relatively few as candidates for a potentially curative resection. Unfortunately, even in patients who undergo pancreatic resection, both local and systemic recurrences are common with a median postresection survival of less than 18 months (2).

The recent completion of the pancreatic cancer exome marked a notable milestone (3). The coding regions of 20,661 genes were sequenced in 24 PDACs indicating that these neoplasms contain an average of 63 genomic alterations, the majority of which are point mutations. Moreover, the genetic landscape of the PDAC genomes is notable for 4 frequently mutated genes, designated “mountains,” including KRAS, CDKN2A (p16), TP53, and SMAD4 (DPC4). Numerous candidate cancer genes altered at low frequency, designated “hills,” were also identified such as MLL3 and ARID1A (3, 4). These 4 mountain genes are well recognized as contributing to pancreatic carcinogenesis (5), and are thus classifiable as “driver” genes for this tumor type. Furthermore, based on comparative lesion sequencing these 4 genes are also classifiable as “founder” mutations in that they are present in the original parental clone that gave rise to the infiltrating carcinoma (6). Although additional genetic alterations accumulate during the ongoing clonal evolution of the carcinoma (“progressor” mutations), the constellation of founder mutations contained within the parental clone presumably constitutes the major characteristics for that carcinoma (6, 7).

The relationship between the genetic status of these 4 genes and clinicopathologic features, including survival have been previously studied. However, until now this work has focused on individual genes and has yielded conflicting results (8–14). Furthermore, although genetically engineered mouse models indicate that the concomitant expression of these mutated genes is crucial to progress to invasion and metastasis in PDACs (15–19), the extent to which the coexistence of 3 or more of these altered genes in the same PDAC influence the biologic behavior and survival outcome is unknown.

The objective of this study was to clarify the clinical significance of the genetic landscape of pancreatic cancer, specifically the genetic status of the KRAS, CDKN2A, TP53, and SMAD4 driver genes in a large series of nonfamilial advanced stage PDACs with known outcomes including patterns of failure and in a second set of xenografted PDACs. We now show that there are distinct patterns and prevalences to which these driver genes occur in the same carcinoma, and that these patterns are highly correlated with clinical features of patients.

Paraffin-embedded and snap-frozen tissue samples from 79 patients collected in association with the Gastrointestinal Cancer Rapid Medical Donation Program (GICRMDP) were used. This program was previously reported in detail (20). Among these 79 patients, 20 initially underwent surgical resection and the remaining 59 patients were initially diagnosed with Stage III/IV unresectable disease. On the basis of autopsy findings and clinical chart review, all patients died of causes directly related to their disease. The Johns Hopkins Institutional Review Board approved use of all patient samples for this study.

Snap frozen tissue samples were embedded in OCT compound (Sakura Finetek), sectioned by a cryostat and stained by hematoxylin and eosin. Tumor tissues were dissected macroscopically if the neoplastic cellularity was at least 50%, or microscopically using a PALM MicroLaser System (Carl Zeiss MicroImaging) for cases with low neoplastic cellularity. Genomic DNA from dissected tissues was extracted using phenol-chloroform, or QIAmp DNA Micro Kits if microdissected (Qiagen). Genomic DNA from microdissected tissues was quantified by calculating long interspersed nuclear elements (LINE) by real-time PCR as described previously (6) and whole genome amplification (WGA) was carried out using 10 ng total template gDNA and an illustra GenomiPhi V2 DNA Amplification Kit (GE Healthcare). PCR amplification was carried out using 20 ng of gDNA for KRAS exons 1 and 2, TP53 exons 5 to 9 and CDKN2A exons 1 and 2 using intronic primers flanking these exons (Supplementary Table 1). PCR products were sequenced by use of a M13F primer (5′-GTAAAACGACGGCCAGT-3′) or M13R primer (5′-CAGGAAACAGCTATGACC-3′) that was incorporated into the forward and reverse primer of each primer pair, respectively (Beckman Coulter Genomics). Sequencing data were analyzed with Sequencher 4.10 software (Gene Codes). Mutation analysis, confirmation and determination of somatic status were carried out using matched normal tissues from the same patient.

Paraffin-embedded samples of the primary carcinoma and matched metastases were immunolabeled for Cdkn2A, Tp53, and Smad4 as an adjunct to sequencing. At least 5 different distinct regions of the primary carcinoma were immunolabeled for each case to evaluate for potential heterogeneity. In the event of positive immunolabeling for Cdkn2A or Smad4 in the primary carcinoma, at least 5 different matched metastases, and local recurrences if available, were also labeled to assess for gene inactivation during disease progression. Immunohistochemical labeling was carried out using antibodies to Cdkn2A protein (ready-to-use, clone E6H4, MTM Laboratories), Tp53 protein (ready-to-use, Bp-53-11, Ventana) and Smad4 protein (clone B8, Santa Cruz Biotechnology) as reported (21). Nuclear labeling of Cdkn2A was scored as intact (positive, indicating the presence of an intact gene) or lost (negative, indicating a deletion, inactivating mutation, or promoter hypermethylation; refs. 22, 23). As previously described (21), Tp53 immunolabeling was considered abnormal when it showed robust nuclear accumulation of immunolabeled protein in ≥30% of the neoplastic cells compared with adjacent normal cells, or if the neoplastic cells showed a virtual absence of immunolabeling compared with immediately adjacent normal cells suggesting the presence of an intragenic deletion, nonsense or frameshift mutation (24–26). In all instances p53 labeling was evaluated within sections cut from at least 2 different paraffin blocks of the same carcinoma. Nuclear labeling of Smad4 was scored as intact (positive, indicating the presence of an intact gene) or lost (negative, indicating a deletion or inactivating mutation of the gene has occurred; ref. 27). Normal islets for Cdkn2A and normal acinar cells, islets, lymphocytes, and stromal cells for Tp53 and Smad4 were regarded as internal positive controls for each case. Negative controls for each of the antibodies were carried out using nonimmune serum instead of the primary antibody. Slides were scored by 2 of the authors (S.Y. and C.I.D.).

Dichotomous variables were compared using Fisher's exact test or the χ² test, and continuous variables were compared using the Student t test or the Mann–Whitney U test, where appropriate. Multiple groups were compared by the Kruskal–Wallis test or the chi-square test, where appropriate. Survival analyses were carried out by the Kaplan–Meier method or Cox regression and survival curves were compared with the logrank test. A P value of ≤0.05 was considered statistically significant. Statistical analyses were carried out using SPSS 20.0 software (SPSS).

The clinicopathologic features of all 79 patients with lethal pancreatic ductal adenocarcinoma whose tissues were collected in association with the GICRMDP are summarized in Table 1. Detailed findings at autopsy of 60 of these patients were previously described (28). Among all 79 patients, 56% were male and 81% of the primary carcinomas developed in the head or body of the pancreas. Most patients (75%) had advanced stage disease at diagnosis (Stage III or IV), and this corresponded to a median overall survival of 10 months for all 79 patients. Nonetheless, when stratified by stage at diagnosis the median overall survival was 24 months for Stage I/II, 11.5 months for Stage III, and 6.5 months for Stage IV patients. At autopsy, 17 (85%) of Stage I/II patients had a local recurrence although for 3 of these it was the only site of disease found. The liver was the most common site of metastatic disease among all patients and was found in 76% of patients. However, the extent of metastatic disease burden among all patients varied greatly (less than 10 to >100), a reflection of the inherent “metastatic efficiency” of each patient's pancreatic cancer (28).

DNA was extracted from snap frozen samples of normal tissue, primary infiltrating ductal adenocarcinomas and multiple matched metastases for all patients and sequenced for KRAS, CDKN2A, and TP53. Multiple samples taken from distinct regions of each primary carcinoma were analyzed (mean 5.9 samples per carcinoma), as well as multiple different metastases (mean 6.3 matched metastases per patient) corresponding to a total of 884 individual samples and greater than 2.5 million bases of sequencing data analyzed.

Activating mutations in KRAS were identified in 73 (92%) of 79 carcinomas analyzed (Supplementary Table 2). Mutations at codon 12 were most common (66 of 73 mutations, 90%), with G12D accounting for 38 (52%) of 73 carcinomas. For 6 carcinomas without a detectable KRAS mutation of codons 12, 13, or 61 we also analyzed for mutations of codon 146 (29), but no mutations were found.

High quality sequencing data were obtained for CDKN2A in 76 of 79 patient's carcinomas (Supplementary Table 2). Intragenic mutations were identified in 21 (28%) of 76 carcinomas analyzed, corresponding to 8 (38%) missense mutations, 7 (33%) nonsense mutations, and 6 (29%) frameshift mutations. All but 1 carcinoma with an intragenic mutation had loss of Cdkn2A protein expression. Because CDKN2A may undergo homozygous deletion or hypermethylation-induced silencing that would not be detected by sequencing (30), we also immunolabeled all 55 carcinomas in which no intragenic mutations were found. Of these, 48 (87%) had loss of Cdkn2A labeling. In total, loss of Cdkn2A secondary to any potential mechanism was detected in 72 of 79 (91%) carcinomas analyzed.

Inactivating mutations in TP53 were identified in 58 of 79 (73%) carcinomas, of which 28 (48%) were missense mutations, 11 (19%) were frameshift mutations, 9 (16%) were nonsense mutations, 6 (10%) were intragenic deletions, and 4 (7%) were splice-site mutations (Supplementary Table 2). Carcinomas found to be TP53 wild type by sequencing were also immunolabeled for Tp53 protein to assess for potential large homozygous deletions or mutations outside of the analyzed region. Of these, 3 had robust nuclear accumulation of Tp53 and 5 of 21 had complete absence of Tp53 protein. Overall, TP53 was altered in 66 of 79 (84%) carcinomas.

Finally, we also determined Smad4 immunolabeling patterns, which is a strong marker of SMAD4 genetic status (27, 31). Of 79 carcinomas analyzed, 39 (49%) showed loss of Smad4 immunolabeling consistent with inactivation of the SMAD4 gene (Supplementary Table 2).

In 73 patients analyzed (92%), there was complete concordance for genetic status and/or immunolabeling patterns of all genes in the primary carcinoma and the matched metastases. Of the remaining 6 patients, 1 showed intact Smad4 labeling in the primary carcinoma and peritoneal metastases, whereas the matched liver metastases in this patient showed loss of labeling, indicating genetic inactivation of SMAD4 occurred during subclonal evolution and metastatic progression (Fig. 1A and B). In an additional 5 patients intratumoral heterogeneity for Cdkn2A labeling was observed in the primary carcinoma in that regions of both strong positive and complete loss of labeling were seen (Fig. 1C–E). True heterogeneity versus a labeling artifact was confirmed by use of a second antibody to Cdkn2A raised against a different epitope of the protein that showed the identical pattern of labeling in these 5 carcinomas. One of these carcinomas contained a 6 bp in-frame deletion of the CDKN2A gene, and the matched liver metastases showed complete loss of Cdkn2A labeling. In the remaining 4 carcinomas no mutations were found, and the matched liver metastases also had loss of Cdkn2A labeling.

We next determined the specific genes altered in pancreatic cancers with 1, 2 and 3 total genetic alterations (Table 2), as well as the type of alterations for these genes in each category. Of interest, for 1 aggressive carcinoma (patient A68) only a KRAS mutation was found, despite analysis of 8 different microdissected samples of the primary carcinoma and 24 different matched metastases. Among carcinomas with 2 genetic alterations, all 14 had a KRAS or CDKN2A alteration and 9 of 14 (64%) harbored an alteration in both KRAS and CDKN2A. The remaining 5 carcinomas had either a KRAS or CDKN2A alteration in combination with a TP53 alteration. Among the 35 carcinomas with 3 genetic alterations, all 35 had a KRAS or CDKN2A alteration and for 28 of 35 (80%) carcinomas KRAS and CDKN2A were coexistent. Moreover, 25 of these 28 carcinomas (89%) contained TP53 as the third genetic alteration, and the remaining 3 carcinomas contained loss of SMAD4 as the third genetic alteration. The remaining 7 of 35 (20%) carcinomas had a KRAS or CDKN2A alteration in association with both TP53 and SMAD4 alterations.

Given the observations made in autopsied patients, we further explored the extent to which these driver gene alterations are coexistent in a second and more uniform set of xenografts derived from 84 pancreatic cancer patients with Stage I/II disease seen at our institution. The specific genetic features of KRAS, CDKN2A, TP53 and SMAD4 in these xenografts have previously been reported in association with whole exome sequencing of a large series of pancreatic cancers (3). These xenografts were also previously analyzed as part of a larger series of xenografted carcinomas evaluating the relationship of each of these genes to overall survival (8). However as the frequency and prevalence of coexistent mutations in xenografts from these patients were not addressed, we focused specifically on that aspect.

The genetic features of KRAS, CDKN2A, TP53, and SMAD4 in these xenografts were similar to that found for the autopsy cohort. All but 1 carcinoma (99%) had a mutation in KRAS with G12D the most common mutation identified in 40 of 84 (48%) carcinomas analyzed. Inactivating mutations or homozygous deletions of CDKN2A were found in 81 of 84 carcinomas (96%), and of TP53 in 71 of 84 (83%) of these same cases. Inactivation of SMAD4 by mutation or homozygous deletion was identified in 39 of 84 (46%) carcinomas and was most often seen in association with TP53 mutation (34 of 39, 87%). The frequency at which these driver gene alterations were coexistent in a single pancreatic cancer was also similar to the autopsy cohort, with the majority of carcinomas also having 3 (46%) or 4 (39%) coexistent alterations. Thus, our findings of the frequency and coexistence of driver genes in autopsied patients is likely correct and not an underestimate due to our sample type analyzed.

Given that SMAD4 loss was commonly seen in association with TP53 inactivation, we further explored this relationship. SMAD4 inactivation always occurred in association with 2 or 3 coexistent driver gene alterations, and the vast majority of SMAD4 inactive carcinomas had coexistent TP53 mutations (36 of 39, 92%). By contrast, TP53 alterations were equally likely to be found independent of SMAD4 inactivation with 35 of 66 (53%) in SMAD4 wild type carcinomas versus 31 of 66 (47%) in association with SMAD4 loss. SMAD4 status alone was significantly correlated with high metastatic burden (P = 0.008), as was TP53 status (P = 0.039). However, as these 2 gene alterations are commonly coexistent we compared the features among pancreatic cancers with TP53 alterations only, with SMAD4 alterations only, with alterations in both genes and in neither gene. To our surprise, TP53 alterations were similarly correlated with high metastatic burden disease when they occurred with or without coexistent SMAD4 alterations (P = 0.170), and differed from carcinomas without TP53 and SMAD4 alterations in which metastatic burden was more commonly oligometastatic (P = 0.008). To determine if the types of TP53 alterations differ among these groups to explain this observation, we assessed the frequency of TP53 missense versus null mutations (nonsense, deletion or frameshift) in the 58 carcinomas with complete sequencing data available. Of interest, null mutations were significantly more common in SMAD4 intact carcinomas (18/28, 64%) than in carcinomas with SMAD4 loss (7/22, 38%, P = 0.046). Collectively, this suggests that pancreatic cancers with high metastatic efficiency may be represented by at least 2 genetic subtypes, i.e., TP53 null mutant and TP53 missense mutant in association with SMAD4 loss.

Among all 79 carcinomas analyzed, 1 (1%) had a single detectable gene alteration, 14 (18%) had 2 gene alterations, 35 (44%) had 3 gene alterations and 29 (37%) had an alteration in all 4 genes analyzed (Table 2). Carcinomas with 1 or 2 alterations only were combined into a single group, as were carcinomas with 3 or 4 alterations, and the relationships of the number of genetic alterations to clinical features of each patient's carcinoma was analyzed (Table 3). There were no differences in mean age or gender distribution among patients in relation to number of gene alterations, nor were there differences in tumor size, location or differentiation at initial diagnosis. No relationship was found either with clinical stage at diagnosis, although 1 to 2 gene mutant carcinomas were twice more commonly observed in association with Stage I/II disease (30% of patients, vs. 15% of Stage III and 15% of Stage IV). By univariate analysis the number of altered genes was significantly correlated with both median disease free survival (P = 0.008) in patients with Stage I/II disease, and median overall survival (P = 0.041; Fig. 2) among all patients although this was not maintained when separated out by stage. However, a greater number of altered genes was also significantly correlated with high metastatic burden at autopsy with 10 of 15 (66%) patients with 1 to 2 altered genes having oligometastatic failure compared with 2 of 29 (14%) of patients with widespread metastatic failure (P = 0.002; Table 4). This relationship was also maintained when patients were stratified by tumor stage. Of interest, when controlling for clinical stage at diagnosis the number of altered genes remained significantly correlated to patient survival (P = 0.046; Table 5).

The pancreatic cancer progression model illustrates the approximate timing of accumulation of genetic alterations during PanIN progression (32). KRAS mutations are an early event and are followed by inactivating mutations in CDKN2A, whereas TP53 and SMAD4 alterations occur relatively later during PanIN-3. Although our data are in agreement with this model, they also suggests that this mode of genetic progression likely occurs for only a subset of patients in that only 37% to 39% of carcinomas contain alterations in all genes. Thus, a more complete understanding of the extent to which alterations of these genes are coexistent in pancreatic cancer should not only provide insight into the dynamics by which they occur during pancreatic carcinogenesis, but also the biologic features of the infiltrating carcinomas that developed from those precursors.

The major clinical implication of this work is that knowledge of the gene status of the 4 major driver genes in pancreatic cancer, and specifically the extent to which they are coexistent in an individual patients cancer, provides distinct information regarding patterns of disease progression, metastatic failure and survival outcome. It is important to emphasize that other genes also play an important role in the biology of pancreatic cancer, for example inactivating BRCA2, PALB2, or FANC gene mutations that may confer susceptibility to cisplatin or PARP inhibitors (33, 34). However, because mutations in those genes are relatively uncommon our rationale was to identify genetic factors that influence outcomes for a greater number of patients. For example, among Stage I/II patients' carcinomas with 2 driver gene alterations were associated with relatively longer median disease-free survival, and carcinomas with 2 driver gene alterations were significantly more likely to develop oligometastatic failure. Ultimately, although the demographics of these patients are entirely in keeping with the epidemiology and clinical features of larger cohorts of patients in well-controlled studies, additional validations in a controlled setting will be necessary.

The most common initiating genetic events in pancreatic cancer are oncogenic mutations in KRAS and inactivating mutations, deletions or methylation of CDKN2A (30), and the sole identification of these 2 driver genes accounted for many of these cases. However, in other carcinomas the 2 driver gene alterations corresponded to alternative combinations, for example KRAS and TP53, but importantly never included SMAD4. Overall, these carcinomas with “two” driver genes had significantly longer disease free and median overall survival, suggesting the subset of patients whose carcinomas have these genetic features may be enriched for long-term survivors. Of note, it is highly likely that additional genes may be mutated in the TP53 (apoptotic) and TGFβ pathways in these carcinomas that were not evaluated by our approach. For example, Jones and colleagues proposed that the significance of genetic alterations in pancreatic cancer were largely for their indication of the core signaling pathways they occurred in, and that although more than 1 gene may be targeted in a pathway only 1 gene of the pathway is targeted per carcinoma (3). Moreover, it is conceivable that these alternative genetic alterations may not have the same effects on survival or progression as for TP53 and SMAD4 that are the most frequent genetic targets in their respective pathways. Consistent with this notion, Blackford and colleagues found that among all members of the TGFβ signaling pathway that may be genetically inactivated in pancreatic cancer, only SMAD4 loss is associated with worse overall survival (8). By contrast, in 1 patient in our study only a KRAS mutation was found despite careful methodology, and this patient had widespread metastatic disease at autopsy following a mere 5 month overall survival, suggesting relatively rare genetic events occurred during carcinogenesis leading to a particularly aggressive phenotype (35).

We have previously shown that SMAD4 status of the primary carcinoma correlates with patterns of failure in pancreatic cancer (28), and now extend this observation by illustrating that SMAD4 loss is most often seen in the setting of coexistent mutations in TP53. In this regard, SMAD4 loss is a marker of genetically complex pancreatic cancers (i.e., those with all 4 driver gene mutations). These data also clarify prior observations that not all patients with widespread metastatic disease at autopsy have SMAD4 loss, and provide evidence that mutations that specifically abolish TP53 gene expression may also promote widespread metastatic failure independently of SMAD4 loss in some patients. Thus, determinations of both SMAD4 and TP53 status may have value in identifying patients at risk for widespread metastatic failure. Furthermore, as additional genes are functionally validated as drivers in this tumor type (3, 4), it is conceivable that they will provide added information regarding prognosis and risk of metastatic failure for pancreatic cancer patients.

KRAS mutations in normal cells leads to replicative senescence (36), and it has been suggested that CDKN2A inactivation provides a selective advantage to KRAS mutant cells by allowing cell division to proceed unhampered through the G1 checkpoint (37). That the vast majority of pancreatic cancers in this study have coexistent KRAS and CDKN2A mutations (89%) provides support to this concept. Beyond KRAS and CDKN2A, the frequencies by which alterations in TP53 or SMAD4 occur are relatively lower. SMAD4 loss most often occurred in a background of TP53 mutations yet TP53 mutations occurred at similar frequency in the presence or absence of SMAD4 loss, suggesting SMAD4 inactivation follows TP53 during the genetic progression of PanINs. In this context SMAD4 loss may provide a selective advantage to cells with coexistent KRAS, CDKN2A, and TP53 mutations. In support of this hypothesis, we noted that TP53 null mutations were less commonly found in association with SMAD4 inactivation suggesting that TP53 null mutations select against SMAD4 loss. Alternatively, TP53 null mutations may have similar “potency” in progressing to an infiltrating carcinoma as coexistent TP53 missense mutations and SMAD4 loss. Consistent with this concept the metastatic burden of patients whose carcinomas corresponded to these 2 genetic categories (KRAS/CDKN2A/TP53-null vs. KRAS/CDKN2A/TP53-missense/SMAD4) were similar to each other and significantly different from carcinomas that did not have TP53 or SMAD4 mutations.

The significance of exomic sequencing can only be realized by translational studies that include well-annotated patient data. We now show the clinical significance of such data for patients with pancreatic cancer. In time, these data may also have value for personalized approaches to management of pancreatic cancer patients.

No potential conflicts of interest were disclosed.

Conception and design: S. Yachida, D. Laheru, J.M. Herman, V.E. Velculescu, C. Wolfgang, C.A. Iacobuzio-Donahue

Development of methodology: S. Yachida, C. White, Y. Zhong, R.A. Morgan, C.A. Iacobuzio-Donahue

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): S. Yachida, C. White, Y. Naito, J.A. Brosnan, A.M. Macgregor-Das, R.A. Morgan, T. Saunders, S. Jones, C.A. Iacobuzio-Donahue

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): S. Yachida, Y. Zhong, J.A. Brosnan, A.M. Macgregor-Das, R.A. Morgan, R.H. Hruban, A.P. Klein, S. Jones, C. Wolfgang, C.A. Iacobuzio-Donahue

Writing, review, and/or revision of the manuscript: S. Yachida, J.A. Brosnan, D. Laheru, J.M. Herman, R.H. Hruban, A.P. Klein, S. Jones, V.E. Velculescu, C. Wolfgang, C.A. Iacobuzio-Donahue

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): S. Yachida, R.A. Morgan, J.M. Herman, C.A. Iacobuzio-Donahue

Study supervision: C.A. Iacobuzio-Donahue

Supported by National Institutes of Health grants CA140599, CA101955, CA62924, and CA121113, The Uehara Memorial Foundation, The Alfredo Scatena Memorial, The George Rubis Endowment for Pancreatic Cancer Research, The Michael Rolfe Pancreatic Cancer Foundation, Sigma Beta Sorority, The Joseph C. Monastra Foundation, The Gloria Swan Pancreatic Cancer Foundation, The Skip Viragh Pancreatic Cancer Center, The Patty Boshell Pancreas Cancer Foundation, and a Stand Up To Cancer Dream Team Translational Cancer Research Grant, a Program of the Entertainment Industry Foundation (SU2C-AACR-CT0109; V.E. Velculesco and C.A. Iacobuzio-Donahue).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.