Article Text


A genetic roadmap of pancreatic cancer: still evolving
  1. Faiyaz Notta1,2,
  2. Stephan A Hahn3,
  3. Francisco X Real4,5,6
  1. 1Princess Margaret Cancer Centre, University Health Network, Toronto, Canada
  2. 2Department of Medical Biophysics, University of Toronto, Toronto, Canada
  3. 3Department of Molecular Gastrointestinal Oncology, Ruhr-University Bochum, Bochum, Germany
  4. 4Epithelial Carcinogenesis Group, Spanish National Cancer Research Centre (CNIO), Madrid, Spain
  5. 5Departament de Ciències Experimentals i de la Salut, Universitat Pompeu Fabra, Barcelona, Spain
  6. 6CIBERONC, Madrid, Spain
  1. Correspondence to Dr Faiyaz Notta, Department of Medical Biophysics, University of Toronto, 27 King’s College Cir, Toronto, ON M5S, Canada; faiyaz.notta{at}, Professor Stephan A Hahn, Department of Molecular Gastrointestinal Oncology, Ruhr-University Bochum, Bochum 44801, Germany; stephan.hahn{at} and Dr Francisco X Real, Epithelial Carcinogenesis Group, Spanish National Cancer Research Centre (CNIO), Madrid 28029, Spain; freal{at}


A diagnosis of pancreatic ductal adenocarcinoma (PDA) is often fatal. PDA is widely recognised as one of the ‘incurable cancers’ because therapies against this tumour type are generally ineffective. The fatal nature of this tumour is due to its aggressive clinical course. Pancreatic cancer commonly presents at the metastatic stage; even in cases where tumours are localised to the pancreas at diagnosis, metastatic seeds have often been invariably been spawned off, frustrating surgical attempts to cure the cancer. The key principles of pancreatic cancer mutational development were outlined nearly two decades ago using the genetics of precursor lesions to position the various stages of tumour progression. Since then, there has been a cavalcade of new data. How these recent studies impact the classical perceptions of pancreatic cancer development is a work in progress. Given that significant improvements in patient outcomes are not in sight for this disease, it is likely that broadening the current perspectives and acquiring deeper biological insights into the morphogenetic route of tumour development will be needed to foster new strategies for more effective cancer control.

  • pancreatic cancer
  • cancer genetics
  • genetic instability
View Full Text

Statistics from


Any cell that divides generates mutations. Most mutations are benign and do not contribute to cell survival but, occasionally, a cell will undergo a mutation that will increase its likelihood to survive and expand clonally. Tumour surveillance mechanisms, either intrinsic (ie, tumour suppressors) or extrinsic to the cell (ie, immune system), have evolved to halt most aberrant clonal expansions from becoming cancerous. However, the implicit Darwinian nature of cancer generates extensive genetic diversity. This in turn gives rise to a wide array of phenotypes providing the necessary conditions to evade tumour suppressor barriers. The exact number of genetic alterations required for a normal cell to become an invasive cancer cell, and the time required to complete this process, still need to be resolved, but it is well accepted that a single genetic hit is insufficient to create a cancer and the malignant transformation process takes decades.

Historical landmark studies, such as those by Armitage and Doll, Knudson, Cairns and Nowell, initially put together the intellectual framework needed to understand cancer progression.1–3 This early narrative was the basis for the pioneering work by Fearon and Vogelstein on colon cancer in the 1980s.4 Fearon and Vogelstein made the critical connection that histological progression is related to genetic progression. By aligning specific driver mutations to cancer precursor stages (adenomas in the case of colon cancer), this group posited the first tractable model of tumour progression. Tracking mutations in physical precursors of invasive carcinoma provided a means to study the phases of tumour development and laid the groundwork for similar work in other tumour types. At the time, putative precancerous lesions in pancreatic ductal adenocarcinoma (PDA) had already been described,5–14 providing a morphological basis to apply the same concept. But the key genetic alterations that drive this disease needed to be identified to build the genetic roadmap of PDA progression. Almoguera et al15 identified KRAS as the first recurrently mutated gene in PDA, which strikingly was present in nearly all tumours. Cytogenetic analyses and chromosomal mapping studies subsequently demonstrated that allelic losses on chromosomes 9, 17 and 18 were also frequent and led to the identification of the major tumour suppressor genes (CDKN2A, TP53 and SMAD416–19). Together, these early studies have had a lasting effect: nearly two decades have passed and the view that PDA evolves through a restricted path involving these four genes has been cemented.

With critical genetic members in PDA identified, it was possible to develop a progression model. Akin to colon cancer, the mutation order of PDA was derived from the study of ductal precursor lesions, termed Pancreatic Intraepithelial Neoplasia or PanIN. The biology of PanINs is intimately linked to the PDA progression model. Other precancerous lesions not following the ductal progression lineage (ie, cystic tumours) have been proposed, but the focus of this review will be on PDA progression via PanIN.

It all starts at PanINs

Hyperplastic lesions of pancreatic ductal epithelium were noted as early as 1936.5–14 Kozuka et al proposed the notion that PDA progresses through ‘sequential change from non-papillary hyperplasia through papillary and atypical hyperplasia to carcinoma’ in the 1970s.10 This was in line with classical pathological observations suggesting that epithelial malignancies progress through histological precursor stages to become invasive. Morphological heterogeneity among precursor lesions overlaid against lack of clear consensus and interobserver differences among pathologists hampered early work. To overcome this important shortcoming, a standardised nomenclature and classification system for PanIN nomenclature was established in 2001 and revised in 2015.20 Three distinct PanIN types were identified and thought to reflect different stages of neoplastic progression21: PanIN-1 being either flat (1A) or papillary (1B) hyperplastic lesions lacking dysplasia, PanIN-2 being papillary lesions with low to moderate dysplasia, and PanIN-3 showing high-grade dysplasia (ie, carcinoma in situ).22

KRAS mutations and PanIN

Much of the early genetic work on PanINs was performed before the consensus nomenclature was established. It rapidly became clear that KRAS mutations are highly prevalent regardless of the grade of precursor lesion. Together with the high frequency of KRAS mutations in PDA, this alteration appeared to be the cancer-initiating event. If a KRAS mutation was indeed the initiating event, its frequency would be expected to be similar at all precursor stages. However, initial studies showed that the prevalence of KRAS mutations increases with grade of the precursor lesions, being roughly 40% in PanIN-1 and gradually rising to greater than 80% in PanIN-3, a frequency similar to PDA.23 However, a recent report using a more sensitive KRAS mutation detection technique illustrated that all PanINs share a high frequency of KRAS mutations (>90%).24 The authors also found that the allele frequency of mutant KRAS increases with the PanIN grade, suggesting that PanIN-1—and to some extent PanIN-2—is often composed of a mixture of KRAS mutant and wild-type cells. It was unexpected that all cells sharing histopathological phenotype and pertaining to a single lesion are not clonally derived. High-grade PanIN-3 lesions showed higher KRAS mutant allele frequencies, indicating a single genetic origin. These data widely support that, in the majority of cases, KRAS is the incipient mutation in PDA development and that the proliferative changes underlying PanIN-1 formation are likely driven by the KRAS mutation. By extension, it would be expected that normal ducts should not harbour these mutations because they are histologically intact. In technically challenging experiments performed by Lüttges et al using a few cells from formalin-fixed specimens, KRAS mutations were detected in normal ducts located in the vicinity of the carcinoma or at the tumour-free resection margins.25 However, other studies have not identified KRAS mutations in normal ducts from patients with PDA.24 Although additional work is required to resolve this matter, it cannot be ruled out that additional changes to the KRAS mutation may be required to establish the PanIN-1 phenotype. More recently, a different precursor lesion designated as atypical flat lesion has been identified in human and murine pancreata, characterised by tubular structures with cuboidal cells and cytological atypia, and harbouring KRAS mutations.26

Limited clonal expansion of early PanINs

How KRAS mutations provide a clonal advantage against adjacent normal cells is a topic of intense debate. It may seem obvious that KRAS mutant cells induce PanINs by simply outproliferating normal ductal epithelium, which is quiescent. In mouse models (discussed in detail in the next section), KRAS mutant cells do not readily induce the PanIN-1 phenotype even when expressed in every epithelial cell of the pancreas. In fact, the majority of ductal cells carrying the mutation are normal for up to 2 months of life in these animals.27 This fits the model that KRAS mutated cells have no phenotype without additional changes, at least early in PDA development. ‘Additional’ could mean more mutations, epigenetic changes or non-genetic events, including changes in the microenvironment. To date, no recurrent mutation beyond KRAS has been linked to the PanIN-1 phenotype, suggesting that non-genetic events may be a critical driver of PanIN-1. In mouse models, changes in microenvironment as a consequence of activated KRAS—such as secretion of cytokines by epithelial cells that can recruit immunosuppressive myeloid cells (eg, granulocyte-macrophage colony-stimulating factor (GM-CSF))—have been described, supporting this notion. GM-CSF expression has been detected in human PanINs, implicating immune cells in propagating KRAS mutant cells.28 This notion is further supported by the fact that KRAS can induce IL17 receptor expression in PanINs, which can recruit IL17-producing T cells to establish what is referred to as ‘hematopoietic-to-epithelial signalling axis’.29 Accordingly, inhibition of this signalling axis effectively halted PanIN growth.29 Taken together these experiments highlight a potentially critical role of the microenvironment in early PanIN formation. Other mechanisms, more indirectly supporting the progression of ‘early’ KRAS mutated cells, have also been proposed, such as pancreatitis-induced inflammation or high-fat diet, which may contribute to modulation of KRAS activity.30–36

Beyond KRAS, PanINs also harbour shortened telomeres compared with normal pancreatic ductal epithelium.37–39 PanINs likely proliferate at an increased rate compared with normal ductal epithelium; because telomeres can shorten each time a cell divides, it is conceivable that shortened telomeres might induce a ‘telomere crisis’, thereby activating the p53 cell cycle checkpoint. In general, the extent of PanIN proliferation is thought to increase with PanIN grade, which seems to correlate with shorter telomeres in PanIN-2 and PanIN-3 compared with PanIN-1.37–39 A difficulty in the interpretation of these data lies in resolving the critical number of cell divisions that PanIN-1 undergo compared to normal ducts, and the activity of telomere maintenance mechanisms. An increasing rate of atypical mitosis, anaphase bridges and loss of CDKN2A, TP53 and SMAD4 tumour suppressors has been reported and supports growing genomic instability in higher grade PanINs.39–42 ATM–Chk2 checkpoint activation and p21 expression correlates with PanIN grades,43–45 which suggests a DNA damage response (DDR) has been activated, possibly driven by telomere attrition or accumulating mutations. Dividing cells also accrue mutations, and next-generation sequencing genomic analyses have shown that PanINs can harbour up to 50% of the somatic mutation burden observed in PDA.46

Genetic rationale of PanIN progression towards PDA

The majority of PDA are considered to start as a PanIN-1. Even though PanIN-1 lesions harbour oncogenic KRAS mutations, an increased mutation burden, enhanced DDR, short telomeres and an activated senescence programme,31 35 these early precursors are generally considered to be benign because of their ubiquitous presence in the aged population.47 Inactivation of the tumour suppressor networks that halt clonal expansions in early PanINs is likely required for the histological progression. The CDKN2A locus encodes the CDK4/CDK6 inhibitor p16INK4A and the p53 inhibitor p14 ARF. p16INK4A is a major player in cell cycle control whose proficiency is thought to be critical for early PanINs and p14 ARF sequesters MDM2, which targets p53 for degradation. Inactivation of the CDKN2A locus was demonstrated at the level of aberrant methylation,48 mutation in combination with hemizygous deletions of chromosome 9 p,49 homozygous deletion50 or loss of protein expression51–53 in 30% of PanIN-1 and rises to over 80% in PanIN-3. The gradual increase in the frequency of CDKN2A alterations with histological grade suggests that inactivation of this gene is needed for PanIN-2 formation. This alteration likely releases a cell cycle block as shown by increasing Ki-67 expression,53 54 although upregulation of cyclin D was only reported to increase as early as in PanIN-2.53 Reports in the previous decade indicated a paucity of genetic or protein losses in SMAD4 and TP53 in PanIN-1s and PanIN-2s.42 53 However, at the PanIN-3 stage there was a sharp increase in the inactivation frequency in these genes.42 53 Because PanIN-3 is the equivalent of carcinoma in situ, it was implied that genetic alterations in SMAD4 and TP53 must follow CDKN2A in the mutational hierarchy of cancer progression.

Modelling PDA initiation and progression in mice

Genetically engineered mouse models (GEMMs) have proven to be a powerful tool to study PDA initiation and progression. Among the main issues addressed with these models are the initiating role of KRas mutations, the relevance of the spatiotemporal context and the cooperation of KRas mutations with tumour suppressor inactivation to recapitulate the proposed progression model.

Most studies have relied on activation of the two most common KRas mutations (G12V and G12D) in embryonic pancreatic progenitor cells using Pdx1-Cre or Ptf1a-Cre driver strains to activate recombination. Mutant KRas alone was found to drive the development of mouse PanINs (mPanIN) by adulthood, which morphologically mimic human PanIN. Low-grade mucinous PanINs (mPanIN-1) are the predominant precursor that emerge in this model, whereas high-grade PanINs (mPanIN-3) are rare. Using lineage tracing, low-grade mPanIN, high-grade mPanIN and PDA have all been shown to arise from KRas-mutant cells; however, definitive formal evidence of the mPanIN lineage progression (PanIN-1>PanIN-2>PanIN-3>PDA) is lacking. Overall, PDA penetrance in KRas-mutant only models is low and tumours arise at an advanced age (~1 year), supporting that additional changes beyond KRas are required for transformation, as in the human pancreas.

To model PDA, ductal cells were the obvious cell type in which to activate mutant KRas in the mouse pancreas. Conditional targeting of mutant KRas to ductal cells has proven difficult because of the paucity of mouse strains that selectively drive Cre expression in this cell lineage.55 Furthermore, efforts to induce PDA via activation of mutant KRas in ductal cells leveraging the regulatory sequences of Sox9 or Hnf1-β have had limited success in part because of the development of early tumours in other tissues (ie, lung).56 57 The availability of excellent tools to activate mutant KRas in pancreatic progenitors and in acinar cells, harnessing the selective expression of Elastase, Ptf1a, Cpa1 and Mist1, among others, has been exploited extensively.58 Several of these studies have made use of Ptf1a-null alleles in heterozygosity, which may have contributed to acceleration of tumourigenesis. Collectively, these studies show that mutant KRas activation in pancreatic progenitors27 59 or in acinar cells30 31 leads to acinoductal metaplasia (ADM), as well as to PanINs, both considered to be precursors to PDA.

The current thought is that ductal cells are more resistant to KRas signalling and PDA initiation,56 although mutant KRas activation in Trp53-mutant adult ductal cells led to efficient mPDA formation without PanINs.57 These studies have raised the important question of whether acinar cells are the cancer cell of origin of PDA in humans, and whether changes in their cell identity (metaplasia) play a role in PDA development. Of note, a study on KRAS mutations in ADMs with and without concurrent PanIN lesions in human tissue has shown that the former only harbour KRAS mutations when they are adjacent to PanINs.60

Another question addressed in mice is the timing of KRas mutations. In general, it is assumed that mutations initiating cancers that are diagnosed in older individuals occur during adulthood. By contrast, most studies using GEMMs have relied on the activation of mutant KRas during embryonic development.58 The relevance of the time at which mutations are introduced has been assessed carefully for acinar cells. Activating mutant KRas in committed acinar progenitors at mouse embryonic day 14.5, in the absence of other cooperating mutations, is sufficient to give rise to mPanINs as well as to PDA.30 By contrast, activation of mutant KRas in adult acinar cells is much less efficient, indicating that these cells are refractory to malignant transformation (see below). However, inducing damage/regeneration and inflammation after Kras activation in adult acinar cells can bypass the tumour suppressive mechanisms and leads to both mPanIN and PDA,30 and a role for cell differentiation/metaplastic processes has been proposed.61 The fact that the models in which only mutant KRas is used as a driver lead to malignant tumours at advanced age reflects the need to acquire additional genetic alterations. In summary, mutant KRas—but not the PDA tumour suppressors alone—can efficiently initiate PDA in mice with greatest efficiency when the mutations are introduced in progenitors and preacinar cells.

Similar to reports that PanIN-2 and PanIN-3 harbour alterations in major tumour suppressor genes, there is unequivocal evidence that mutant KRas cooperates with p16, Trp53 or Smad4 for mPDA development in GEMMs. The most commonly used model in the field is the ‘KPC’ model, where mutant KRas is expressed in pancreatic progenitors together with the Trp53R172H mutant allele.59 KPC mice display increased mPanINs, develop tumours with a wide range of differentiation features and acquire metastases rapidly, in agreement with the role of p53 in cancer. PDA from KPC mice harbours aneuploid genomes and invariably loses the wild-type Trp53 allele, as in the human setting. The combination of mutant KRas activation and p16 inactivation also leads to undifferentiated, highly aggressive and metastatic PDA, although the latency period is longer than for the combination of KRas and Trp53 mutations.62 63 Distinguishing the role of the two tumour suppressor genes, acting through different mechanisms, in the Ink4A locus has required the use of gene-specific knockouts. These studies show that both p16 and Arf can contribute to PDA progression in mice.62 The activation of mutant Kras, with concomitant Smad4 inactivation, also leads to an accelerated development of PDA with a distinct IPMN-like morphology and more differentiated features.64 Several other predictions made from the genetic analyses of human lesions have been confirmed in these models, including the fact that inactivation of both alleles of the tumour suppressor genes is associated with a much more rapid appearance of PDA.65 Furthermore, the combination of Kras mutations with inactivation of multiple tumour suppressors (Trp53/Ink4a or Ink4a/Smad4) has also been shown to cooperate even when only one allele of each tumour suppressor is inactivated.62 Overall, these models have supported the crucial role that p16, p53 and Smad4 play in Kras-initiated cells to promote PDA development and have been critical to understanding PDA biology.

Critically evaluating the PanIN model

The PanIN progression model has been central to conceptualising how PDA becomes invasive. In reviews and textbooks, this model is pervasively used to describe the molecular pathogenesis of PDA. Considering the widespread impact of this progression model in the field and how it continues to guide the conceptions of PDA evolution, we would be remiss not to critically evaluate it. The essential question is whether PDAs emerge through a sequence of PanIN-1>PanIN-2>PanIN-3 prior to becoming invasive and metastatic. A relatively minor proportion of PDA arises through other types of mucinous lesions that are increasingly important from a clinical standpoint, but do not belong to the ductal progression model (ie, IPMNs), and are not considered here.66

For nearly a century, pathologists have reported that putative precursor lesions of cancer are present in many organs. Their frequency rises with age and far outweighs actual cancer incidence. This is also true for PanIN-1 lesions in the pancreas. In general, high-grade PanINs are rare and have mostly been observed in organs containing a PDA.9 11 14 67–69 This raises the question of whether PDA development indeed follows the suggested PanIN progression steps.70 71 A probable explanation for the lack of reports of advanced PanINs in non-PDA cases is that the whole pancreas is commonly not sampled in standard autopsy case studies. However, a recent report of extensive whole organ analysis via serial sections spaced at 5 mm identified advanced PanINs at a frequency of 4% in a hospital autopsy series47 using the current PanIN classification system. This is in agreement with data from a study from the 1980s reporting an incidence of 5% for moderate to severe dysplastic lesions.12 The accuracy of the estimates and the age-adjusted prevalence of such lesions need to be determined through additional autopsy series to control for potential selection biases. Another recent study on high-grade PanINs in pancreata from non-healthy individuals lacking PDA shows that the diameter of such lesions varies from the submillimetre to the few-millimetre range.69 Therefore, it seems plausible that small PanIN lesions (≤1 mm) may have gone unnoticed in previous histopathological studies, including those from Matsuda and Mukaka, due to the sampling strategy. These data clearly demonstrate that advanced PanINs do indeed exist.

A critical issue in almost all to date published studies of advanced PanIN is that they were isolated from regions adjacent to the tumour. The fact that advanced lesions are more readily identifiable in such areas is a cause for concern. This provides support for the long-standing controversy about whether a substantial proportion of reported PanIN-3s have been masquerading as ‘cancerization of duct’, that is, replacement of normal ductal epithelium by cancer cells. In the context of PDA, Murphy et al46 reported that 80%–90% of somatic variants are shared between PanIN-3 and PDA. In some cases, they found that PanIN-3s can have a higher somatic mutation burden than their corresponding PDA,46 which suggests that they may be PDA derivatives rather than being PDA precursors. The major implication that emerges is that the mutational hierarchy of PanIN progression has been derived from duct cancerisation events and, in consequence, a substantial proportion of the published data on PanIN-3, and possibly PanIN-2 as well, refer to cancer cells. This notion is further supported by a recent genetic analysis of PanINs by Hosoda et al, which largely failed to identify TP53 and SMAD4 mutations in PanIN-3 lesions extracted from non-PDA cases.69 Furthermore, mutational pattern analysis showed that the majority of early versus advanced PanIN were unrelated, suggesting no clonal relation within a linear progression model of these lesions. Lastly, some of the advanced PanIN lesion studies by Hosoda may not belong to the ductal progression lineage, since they harbour GNAS and RNF43 mutations that are typical for IPMNs.66 These findings stress that histomorphological grades of ductal lesions may not reflect their lineage identity. The lack of TP53 and SMAD4 inactivation in the majority of lesions fulfilling PanIN-3 criteria further suggests that these lesions may not be as advanced as previously anticipated. Clearly, an obvious explanation for the discrepancy between the data from Hosoda et al and previous TP53 and SMAD4 data from PanINs is again the aforementioned ‘duct cancerization’ argument. Furthermore, different PanIN-3 lesions may exist, with (more advanced) or without (less advanced) inactivation of TP53 and SMAD4, and histomorphological criteria may not be able to differentiate between them. It will be important to confirm the existence of such PanIN-3 variants (if they do exist) and define the clinicopathological conditions under which they arise.

Another explanation for the paucity of reported advanced PanIN-3 in the current literature could be that the time required for a less advanced lesion to progress to an advanced PanIN-3 is rather short70 71—in months rather than years or decades.72 73 This would dramatically reduce the prevalence of advanced lesions because they may evade detection due to being over-run by the invasive cancer because of their short half-life. In summary, to tackle inconsistencies in the classic PanIN progression model, high-resolution molecular and histopathological analysis combined with extensive sampling of entire pancreata are needed. This will be crucial in clarifying the role that the different PanIN precursors play in PDA development.

The genomics of pancreatic cancer

An important question that emerged from the PanIN studies was whether additional genetic players, beyond the ‘classic four’ (KRAS, TP53, CDKN2A and SMAD4), were required in PDA pathogenesis. To address this, global genomic analyses were needed and the birth of NGS technology facilitated this task. Jones et al were the first to interrogate the PDA genome globally.74 Analysis of the protein coding sequence of 24 cases revealed only 148 genes harbouring two or more mutations. No dominant recurrent mutation beyond the classic genes was observed, although the cohort size was too small to detect rare recurrent variants. These less frequent mutations did, however, collapse into defined signalling pathways. Notably, they found that nearly all tumours harboured mutations in Hedgehog/Wnt/Notch signalling components. Extending this analysis to over 100 PDA exomes, Biankin et al found that a fifth of PDA genomes harbour mutations and copy number aberrations in genes involved in axon guidance, a signalling pathway that provides attractive and repulsive cues in axon development.75 Whether these new PDA pathways can be leveraged for clinical targeting will be important to investigate because the number of recurrent actionable mutations in PDA is limited.

Using whole genome sequencing Waddell and colleagues found that PDA could be classified into four subtypes based on structural rearrangements: (1) stable, (2) locally rearranged, (3) scattered and (4) unstable. The unstable genotype was linked to defects in DNA double-strand break repair (DSBR), including germline and somatic alterations in BRCA1 and BRCA2.76 These tumours showed robust partial or exceptional responses to platinum-based therapy.76 Large-scale studies are underway to validate whether DSBR defects (~5%–10% of PDA) can be used as a biomarker for platinum-based chemotherapy such as FOLFIRINOX or PARP inhibitors in PDA ( identifiers NCT02184195, NCT01585805, NCT02498613). Altogether, this work identified several new genetic players that reside at low frequency (generally below 5%) in PDA, including MLL3, TGFBR2, ARID1A, EPC1, ARID2, SF3B1, ATM and RNF43. For a complete list, we refer readers to the following:  refs 74–77.

The resolution of these genomic analyses is confounded by the fact that PDAs are often highly desmoplastic. The exome study by Biankin et al revealed that mutation rates in PDA were lower compared with other solid tumours, raising the possibility that low tumour cellularity might have contributed to these results.72 Using needle microdissection, to enrich for epithelial tumour cells prior to exome sequencing, Witkiewicz et al found mutation rates of PDA were twofold to threefold higher compared with the previous exome study (67 mutations per case; n=109 cases).75 77 Their increased detection limit allowed identifying novel, infrequent, recurrent mutations (below 5%) in BCLAF1, IRF6, FLG, AXIN1, GLI3, PIK3CA and RBM10. Overall, these studies have revealed a striking mutational heterogeneity in PDA and have confirmed that, beyond the ‘classic four’ genetic players, recurrent mutations in PDA generally are rare.

PDA subtypes defined by mutational signatures

How mutations accrue in lineage development and tumour progression is fundamental to understanding cancer evolution. Many mutations can be attributed to specific aetiologies that underpin distinct environmental exposures or biological processes.78 79 Over the life history of any tissue lineage or a cancer, mutations can accumulate from a mixture of endogenous and exogenous sources. Endogenous sources of mutation include spontaneous chemical changes in DNA (ie, cytosine deamination), DNA polymerase errors, reactive oxygen species, DNA/RNA editing enzymes, retroviral transposition and mitotic errors. Cigarette smoking, chemotherapeutic drugs, ultraviolet and ionising radiation are examples of exogenous sources of mutation. These biological processes can leave traces on DNA and, because they often act in a sequence-specific context (referred to as a ‘mutational signature’), they can be traced by sequencing through the use of pattern recognition algorithms.80 One such algorithm is non-negative matrix factorisation, which is commonly used for signature deconvolution.81 Thus far, approximately 30 mutational signatures, featuring distinct aetiologies, across cancer have been found.81–83 Signature analysis has allowed identifying distinct mutational mechanisms linking lifestyle exposures (ie, tobacco) with specific tumour types (ie, lung vs pancreatic cancer).84 By applying this methodology to PDA, Connor et al85 identified four dominant signatures in PDA related to biological processes, including ageing, DSBR, mismatch repair (MMR) and C>A transversions, which have also been linked to defects in DSBR. Concurrent RNA sequencing analysis from these cases also revealed DSBR and MMR PDA subtypes were associated with higher expression of antitumour immunity molecules.84 85 Whether these subtypes may benefit from immune checkpoint inhibitors remains an important avenue of investigation.

‘Catastrophic’ PDA evolution via mitotic errors: polyploidy and chromothripsis

Our perspectives on how cancers develop have been considerably revised over the past decade, with new evidence coming into view, in part, due to advances in DNA sequencing. Several novel mutational processes have been identified contending that a fraction of tumours do not arise via a ‘classical’ stepwise progression. In some tumours, progression is marked by a catastrophic genomic phenomenon such as chromothripsis.86 In chromothripsis, parts of a chromosome or whole chromosomes get trapped during cell division in micronuclei,87 where they are broken into tens to hundreds of pieces simultaneously, and then are haphazardly put back together disfiguring the genome (box 1 and figure 1). Micronuclei-independent processes related to anaphase bridging have also been identified as an additional mechanism of chromothripsis.88 On rare occasions when a cell survives such an insult, the outcome can be oncogenic.86 Other catastrophic events that can elicit extensive genomic damage in an abbreviated time frame are telomere-based breakage–fusion–bridge cycles,89 and chromoplexy,90 a set of mostly balanced chains of rearrangements often involving several chromosomes (box 1). Polyploidisation, where the whole genome is duplicated or triplicated, is another process that facilitates rapid adaptation and evolution under conditions of cellular stress (box 1).91 Collectively, these mutational processes, linked to errors in mitosis, can accelerate tumour evolution and promote aneuploidy, and are associated with aggressive tumour behaviour.

Box 1

Glossary of genetic events referred to in the text


Polyploidisation is a mutational process by which the whole genome is duplicated or triplicated. This phenomenon affects a third of human cancers106 and is an important precursor to aneuploidy. There are several mechanisms of polyploidisation, including cytokinesis failure, endoreplication and cell fusion.91 A recent cell-to-cell invasion process referred to as entosis107 could also be a potential mechanism of polyploidisation.

Chromothripsis and chromoplexy

Chromothripsis is a mutational process that involves the break and repair of tens to hundreds of chromosomal segments. From the Greek words ‘chromos’ for ‘chromosome’ and ‘thripsis’ for ‘shattering into pieces’, this phenomenon was initially found in patients with chronic lymphocytic leukaemia where DNA rearrangements on certain chromosomes could not be reconciled to have accumulated gradually through iterative clonal expansions.86 Although chromothripsis is infrequent among all cancers (~5%), certain tumour types such as adenocarcinomas of the oesophagus108 and pancreas94 are particularly enriched for this phenomenon. There are two distinct mechanisms of chromothripsis. Zhang et al have demonstrated that chromothripsis is due to aberrant DNA replication in nuclear structures called micronuclei, where chromosomes that lag during cell division are trapped and aberrantly repaired.87 A second mechanism involves cells undergoing telomere crisis, and the formation of chromatin bridges between chromosomes that snap as a cell undergoes cytokinesis.88

Another mutational process able to generate large clusters of DNA rearrangements is chromoplexy.90 Initially observed in prostate cancer, chromoplexy encompasses a group of balanced chains of rearrangements often involving several chromosomes.90 The mechanistic basis of chromoplexy has not been elucidated.

Figure 1

Chromothripsis and mechanisms leading to chromothripsis. Chromothripsis results from the shattering of a chromosome and the resulting attempt of the cell to repair this event. This typically leads to rearrangements, with fragments of chromosomes bound together in the correct or inverse orientation, and to gains or losses leading to +1 or −1 copy number changes (A). Two mechanisms that have been shown to lead to chromothripsis are shown in the figure. In panel (B), telomere erosion leads to chromosome fusion events that, during nuclear division, result in chromatin bridges, and eventually in chromothripsis (88). In panel (C), micronuclei resulting from chromosome breaks are not segregated into the nuclei of daughter cells and undergo pulverisation in the cytoplasm. Some of the fragments generated can remain in the cytoplasm, whereas others can become integrated in the nuclear genome in subsequent cell divisions (87).

Resolving the disarray of DNA copy number in aneuploid genomes from whole genome sequencing data—a problem that is heightened when tumour cellularity is low, as is the case in PDA—has been a major challenge. To gain insights into these mutational processes, a novel bioinformatics tool that could accurately decipher DNA copy number in aneuploid genomes was recently applied to over 100 tumour cell-enriched PDA genomes. This analysis revealed that mutational phenomena associated with mitotic errors are pervasive in PDA—75% of tumours were either polyploid or harboured at least one chromothripsis event, and approximately 40% harboured both. Interestingly, chromothripsis events were found to occur both before and after polyploidisation, suggesting that polyploidisation likely exacerbates ongoing instability in these tumours.92 Chromosome 18 was most commonly affected by chromothripsis, invariably resulting in losses of chr18q, where SMAD4 resides. Combined, approximately 20% of chromothripsis events included allelic losses in chr9q (CDKN2A), chr17p (TP53) and chr18q (SMAD4), indicating that common PDA drivers are affected by these events. High rates of catastrophic events have been noted in other aggressive tumours, such as oesophageal cancer and medulloblastomas, related with germline TP53 mutations,93 and are in line with the clinical behaviour of PDA.

The high frequency of DNA rearrangements has provided an opportunity to reconstruct events in PDA progression. In a subset of cases (~14%), rearrangements led to allelic losses in CDKN2A, TP53 or SMAD4 simultaneously either via chromothripsis or single unbalanced translocations. These findings raise two critical questions: (1) When do such events occur relative to histological progression (PanINs or other histological lesions)? (2) Do cases with catastrophic events outside of the most critical PDA drivers follow the classical PanIN model? Although more information will be required to resolve such questions, such genomic events likely occur early in PDA tumourigenesis—for example, at the time PDA becomes invasive or soon after—because they are present in the clone that emerges from the most recent common ancestor of the tumour. To resolve this matter, the genomic ancestors of PDA will need to be more accurately deciphered. Polyploidisation and chromothripsis provide a rapid burst of genetic change and are potential candidate cancer-transforming events in PDA. Genomic analysis of the available mouse models could also be exploited to further understand the role of complex genomic rearrangements such as chromothripsis in the evolution of PDA.

Dynamics of PDA progression

Aligning with the catastrophic hypothesis,70 94 there are recent data indicating that once PDA is detected, clinical progression from locally invasive (T1) to fully metastatic (T4) is rapid.95 Comparative genomic analysis of primary tumour and matched metastases has led to the proposal that PDA takes nearly two decades to develop following the KRAS initiating event. The estimates suggest that a preneoplastic phase lasting approximately one decade is followed by a period of invasive growth of 7 years.73 96 97 Metastatic potential develops during the invasive phase, and metastatic seeds require an additional 3 years to manifest completing the 20-year cycle. The broad implications are that metastases occur ‘late’ in PDA progression, leaving a large window of opportunity for early detection (~7 years), where the primary tumour would be invasive but non-metastatic, and where surgical resection would possibly be curative.97

Whether metastases occur ‘early’ or ‘late’ in PDA progression remains a topic of intense debate. At the crux of the problem is the difficulty to exactly define when a precursor becomes invasive and when an invasive cancer becomes metastatic. Rhim et al98 have demonstrated, using the KPC mouse model, that PDA cells can circulate and colonise the liver prior to the emergence of frank malignancy. This supports that metastasis is an early event in PDA. In essence, the model of PDA progression based on catastrophic genomic events also supports this notion. In catastrophic tumour progression, there are two critical events: the cancer-initiating event and the cancer-transforming event. Because of the long period that elapsed after the initiating event (eg, KRAS mutation), metastases will always appear to be genomically ‘late’ as most mutations have accrued after cancer initiation but before cancer transformation events.97 What is more clinically relevant is the time for metastases to appear once invasion has occurred. If a catastrophic event is indeed transforming, it might confer a cell with both invasive and metastatic potential. In this scenario, there would be a very short latency between the birth of the locally invasive phase clone and its ability to spawn metastatic seeds. These experimental data would suggest that PDA lacks an extended locally invasive. This is consistent with the observation that a large fraction of patients with stage I/II PDA undergoing surgical resection succumb to metastatic disease within 2 years,99 100 and that approximately one-third of incidental PDA cases with small primary tumours already harbour metastases.101 Moreover, recent analyses of disseminated lesions indicate little genetic heterogeneity in metastases, suggesting that there is short latency period between the ability of an invasive clone to seed metastases.102 Importantly, the overall time scale proposed by Yachida et al does not disagree with the catastrophic model of PDA progression. Whether PDA evolves as ‘one giant step’ or a series of ‘small steps’ may or may not impact the absolute time it takes for this disease to emerge in patients. The functional outcome of chromothripsis or other catastrophic phenomena is likely to vary widely because of the diversity in the set of genes affected in each case. Notably, the genes involved in such events, and not the catastrophic event itself, are the critical factor that will determine how such events impact tumour evolution and progression. This scenario can have implications for how long it takes for the metastatic phenotype to emerge in PDA progression. The missing piece of the puzzle is how catastrophic genomic events influence the invasive phenotype. In the future, resolving this matter will help us to better understand metastatic progression.

Future perspectives

As George Box put it, ‘All models are wrong but some are useful’.103 There is no doubt that the PanIN>PDA  progression model put forward close to two decades ago has been highly instructive, providing a stimulating framework through which the understanding of the genetic basis of PDA has made its way.

In this review, we have considered several recently published data that should help to move forward in this important and challenging area. The new findings come from a wide variety of realms and include detailed clinical observations and autopsy studies, next-generation sequencing genomic analyses, and computational tools. In box 2 we share our views on some of the current urgent needs in the area covered by this review, elicited by the more recent studies, and in figure 2 we propose a revised version of the PDA progression model that considers the new findings discussed here.

Box 2

What is needed?

  • Assess the prevalence of all reported types of putative preneoplastic lesions (ie, PanINs, AFLs) in individuals without PDA and in subjects with PDA

  • Establish international consortia to compare characteristics and prevalence of these lesions in different geographical and ethnic settings

  • Develop tools to acquire sociodemographic, lifestyle, genetic and other types of epidemiological information in a standardised manner

  • Establish standard operating procedures to obtain, process and isolate biological material from these samples and—whenever possible—obtain other biological material from the same subjects

  • Characterise putative preneoplastic lesions at the genomic level (whole genome sequencing, single cell sequencing, single cell RNA-Seq)

  • Identify optimal markers characteristic of the lesions more likely to represent true PDA precursors

  • Define better the similarities and differences between ‘the ductal progression pathway’ and ‘the IPMN pathway’

  • Develop strategies for the non-invasive identification of PDA precursors, including biological fluid based-assays and imaging

  • Identify molecular targets that could be harnessed for PDA chemoprevention

  • Identify the genetic/epigenetic mechanisms involved in the resistance to therapy to better understand selective pressures driving tumour evolution

  • Establish the mechanisms involved in the resistance of PDA to immune therapies to improve their antitumour efficacy

  • Optimise the use of genetic mouse models to identify mechanisms involved in PDA development/progression that could be exploited in the clinical setting

  • AFL, atypical flat lesion; PanIN, Pancreatic Intraepithelial Neoplasia; PDA, pancreatic ductal adenocarcinoma.

Figure 2

A morphogenetic model of PDA development. In this model, development of preneoplastic lesions, PanIN and possibly other types of precursors, and potential routes of progression to PDA, are considered in the context of ageing (blue) and cancer development (yellow). Telomere attrition and KRAS mutations occur in a normal epithelial cell and are cancer-initiating events. From there, LOH at chrs 9 p, 17 p and 18q are major contributors for progression to PanINs or PDA. Mutations in CDKN2A, TP53 and SMAD4 are key events that distinguish PanIN and invasive carcinoma. Catastrophic genomic events, shown in red, can occur at various time points along the progression model. CIN, chromosomal instability; LOH, loss of heterozygosity; PanIN, Pancreatic Intraepithelial Neoplasia; PDA, pancreatic ductal adenocarcinoma.

The implications are broad and timely since PDA’s societal weight increases with the life expectancy of the global population, and it is predicted that this tumour may become the second cause of cancer death in the Western world in the next decade.104 105


The authors thank the members of their laboratories for valuable discussions.


  1. 1.
  2. 2.
  3. 3.
  4. 4.
  5. 5.
  6. 6.
  7. 7.
  8. 8.
  9. 9.
  10. 10.
  11. 11.
  12. 12.
  13. 13.
  14. 14.
  15. 15.
  16. 16.
  17. 17.
  18. 18.
  19. 19.
  20. 20.
  21. 21.
  22. 22.
  23. 23.
  24. 24.
  25. 25.
  26. 26.
  27. 27.
  28. 28.
  29. 29.
  30. 30.
  31. 31.
  32. 32.
  33. 33.
  34. 34.
  35. 35.
  36. 36.
  37. 37.
  38. 38.
  39. 39.
  40. 40.
  41. 41.
  42. 42.
  43. 43.
  44. 44.
  45. 45.
  46. 46.
  47. 47.
  48. 48.
  49. 49.
  50. 50.
  51. 51.
  52. 52.
  53. 53.
  54. 54.
  55. 55.
  56. 56.
  57. 57.
  58. 58.
  59. 59.
  60. 60.
  61. 61.
  62. 62.
  63. 63.
  64. 64.
  65. 65.
  66. 66.
  67. 67.
  68. 68.
  69. 69.
  70. 70.
  71. 71.
  72. 72.
  73. 73.
  74. 74.
  75. 75.
  76. 76.
  77. 77.
  78. 78.
  79. 79.
  80. 80.
  81. 81.
  82. 82.
  83. 83.
  84. 84.
  85. 85.
  86. 86.
  87. 87.
  88. 88.
  89. 89.
  90. 90.
  91. 91.
  92. 92.
  93. 93.
  94. 94.
  95. 95.
  96. 96.
  97. 97.
  98. 98.
  99. 99.
  100. 100.
  101. 101.
  102. 102.
  103. 103.
  104. 104.
  105. 105.
  106. 106.
  107. 107.
  108. 108.
View Abstract


  • Contributors FN, SAH and FXR wrote the paper together, reviewed all its contents and share responsibility for authors' views. All three authors share correspondence.

  • Funding FN is supported by grants from the Princess Margaret Cancer Foundation, Pancreatic Cancer Translational Research Initiative at the Ontario Institute for Cancer Research (OICR) through support from the Ontario Ministry of Research and Innovation, and an Investigator Award from the OICR. Work in the laboratory of FXR is supported by grants from Ministerio de Economía y Competitividad (SAF2015-70553R) and Asociación Española Contra el Cáncer.

  • Competing interests None declared.

  • Provenance and peer review Commissioned; externally peer reviewed.

Request Permissions

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.