Carcinogenesis of Pancreatic Ductal Adenocarcinoma

Abstract

Although the estimated time for development of pancreatic ductal adenocarcinoma (PDA) is more than 20 years, PDAs are usually detected at late, metastatic stages. PDAs develop from duct-like cells through a multistep carcinogenesis process, from low-grade dysplastic lesions to carcinoma in situ and eventually to metastatic disease. This process involves gradual acquisition of mutations in oncogenes and tumor suppressor genes, as well as changes in the pancreatic environment from a pro-inflammatory microenvironment that favors the development of early lesions, to a desmoplastic tumor microenvironment that is highly fibrotic and immune suppressive. This review discusses our current understanding of how PDA originates.

While mutations in ATM, BRCA1, BRCA2, CDKN2A, and others (see ref ¹) increase the risk of pancreatic cancer, mutation of the proto-oncogene KRAS is believed to be the most common initiating event². Tumorigenesis is accelerated by an increase in the mutant KRAS allele dossage³. Besides these genetic factors and age, other major factors are chronic pancreatitis, smoking, obesity, and type 2 diabetes (see refs ^{4, 5}).

Genetically engineered mouse models have confirmed that PDA development can be initiated by the acquisition of an oncogenic Kras mutations and upregulation of epidermal growth factor receptor (EGFR)/wild-type Kras signaling that leads to development of neoplasia^6–11. Equally important to the PDA development is chronic inflammation and inflammatory macrophages^12–16. Precancerous neoplasms progress to cancer by acquisition of additional mutations, often leading to loss of tumor suppressor function. At this stage, alternatively-activated macrophages (AAM) promote a spiral of signaling that elicits the desmoplastic reaction^{17, 18}, recruits Tregs, and creates a highly immunosuppressive TME¹⁹. The high abundance of cancer-associated fibroblasts (CAFs) then leads to the generation of a collagen- and hyaluronan (HA)-rich extracellular matrix (ECM) which promotes blood vessel collapse and increases hypoxia²⁰. Eventually, up to 80% of pancreatic tumor mass can consist of stromal matrix which surrounds islands of cancer cells, acting as a physical barrier for infiltrating lymphocytes and chemotherapy²¹. Increased matricellular-enriched fibrosis and tissue tension was associated with tumor progression and shorter patient survival²².

Although the estimated timeline of genetic changes that lead to transition from benign pancreatic neoplasia to malignant PDA is more than 20 years²³, patients usually receive a diagnosis of PDA after it already has metastasized. Increasing our understanding of how PDA develops and identifying markers of earliest cancerous lesions may allow the detection of this cancer at stages where it can be treated, or its metastasis can be prevented. Also, better understanding how neoplastic and cancer cells interact with the desmoplastic environment may lead to the development of new, more effective combination therapies.

The PDA Cell of Origin

Of the three major epithelial cell types in the pancreas (islet, acinar and duct), the cell(s) capable of giving rise to PDA is controversial. The ductal phenotype of PDA immediately suggests a duct “cell of origin”, but studies in mice have indicated that this might not be the case. The earliest models of pancreatic neoplasia that resembled the human disease were generated by directing Cre recombinase-dependent expression of an oncogenic form of Kras to pancreatic progenitor cells⁶; in these models, mutant Kras expression is maintained throughout organogenesis and in the adult parenchymal cells thereafter. The pancreata in these mice (designated “KC”) appear to develop normally, with metaplastic and pancreatic intraepithelial neoplasia (PanIN) appearing stochastically when mice are approximately 2 months old. Breeding the KC model with mice with other genetic modifications produces mice with neoplasia that resemble intraductal papillary-mucinous neoplasms (IPMNs)²⁴ or mucinous cystic neoplasms ²⁵. However, this ubiquitous and continuous expression of oncogenic KRAS throughout the pancreata has not helped to identify the PDA cell of origin, or how of the transformation of different cell types might affect the phenotypes of resulting neoplasms (Fig. 1). Use of inducible forms of Cre recombinase to initiate oncogenic Kras expression in specific cell compartments in the adult animal has allowed researchers to explore the consequences of oncogenesis arising from different cell types.

Figure 1. Cells of origin for PDA.

There is evidence that PDAs develop from acinar and/or duct cells. In 3-dimensional organoid culture, duct cells that express oncogenic KRAS can develop into PanIN-like cells; when these are transplanted into mice they form low-grade dysplasias. After additional disruption of Brg1, duct cells with a KRAS mutation can progress to IPMN. Acinar cells are highly plastic and, in crosstalk with inflammatory macrophages, can transdifferentiate to a ductal phenotype (acinar to ductal metaplasia). With an oncogenic KRAS mutation these ADM cells stay locked in a ductal stage, show increased EGFR signaling, and progress to PanIN and PDA.

Ray et al²⁶ used an inducible Cre system (CK19^CreERT) to express oncogenic KRAS specifically in pancreatic ducts, resulting in the formation of mucinous, preneoplastic metaplasia with characteristics of low-grade PanIN. When ducts were isolated and placed in 3-dimensional culture, they gave rise to PanIN-like cells²⁷. However, acinar cells in the adult pancreas are highly plastic and can undergo acinar to ductal metaplasia (ADM) under stress conditions to form a duct-like cell type capable of giving rise to PanIN (reviewed in¹¹). In a study intent on resolving the likely PDA cell of origin, Kopp et al²⁸ expressed oncogenic Kras specifically in acinar and duct cells (using Ptf1a^CreERT and Sox9^CreER mice, respectively) and found that acinar cells were approximately 100-fold more susceptible to transformation than were duct cells, leading to the conclusion that acinar cells are the more likely cell to give rise to PDA in mice. However, it should be noted that in the acinar-specific Cre^ERT2 mice, the Cre^ERT2 transgene is knocked into a Ptf1a allele²⁹, inactivating the locus and making acinar cells inherently more susceptible to transformation³⁰. Although it is unlikely that disruption of a single allele of Ptf1a could account for the entirety of the increased transformation efficiency of acinar cells compared to duct cells, it is likely to have some effects on the outcome.

Consistent with the conclusion that oncogenic Kras is not sufficient to efficiently transform duct cells, studies have found that their transformation potential increases with additional genetic modifications, including those that compromise cell identity (disruption of Brg1 ³¹), increased phosphoinositide 3-kinase signaling within the KRAS pathway (disruption of Pten ³²), or with disruption of Tp53 ³³. Of note, compromising cell identity^{30, 34, 35} and increasing KRAS activity⁸ both promote transformation of acinar cells as well. Interestingly, in most models with duct cell-specific expression of oncogenic Kras, the resulting neoplasia either resembled IPMNs or otherwise bypassed low-grade PanINs^{31–33, 36}. This suggests that in some genetic contexts, the pattern of neoplastic progression differs based on what cell type in which oncogenic Kras is expressed.

While acinar cells seem to be the PDA cell of origin in mice, there is some evidence that not all acinar cells are equally capable of undergoing transformation and development into PDA. Westphalen et al³⁷ found DCLK1 to be a marker of a minor duct and acinar cells that act as facultative progenitor cells that act to repair pancreatitis-associated injury. DCLK-positive cells might be a subset of acinar cells that can become tumor-initiating cells, but further studies are needed.

Duct cells also may also be heterogeneous in their capacity to be transformed. Thayer et al identified a subset of duct cells that could be progenitors of IPMNs. Pancreatic duct glands were originally defined as small cystic ducts that branch from the pancreatobiliary gland and the main pancreatic duct³⁸. These structures are analogous to peribiliary glands that act as a reservoir for facultative progenitor cells, which expand in response to injury^39–41.

Neoplasia

Neoplastic precursors to PDA were defined by histopathological characteristics almost 2 decades ago (Fig. 2). Designating lesions as PanINs, grades 1–3, has been useful in determining the accuracy with which mice genetically engineered to develop pancreatic cancer model the human disease ⁴². In humans, PanIN1 and PanIN2 are not predictive for development of PDA ⁴³. In contrast, high-grade dysplasia (formerly PanIN3) is rarely found without associated PDA, so this equivalent of carcinoma in situ should be considered a bona fide harbinger of cancer. Detection of PanIN3-associated pathology is paramount to our efforts in early detection and imaging if we hope to make PDA preventable.

Figure 2. Progression of PDA and the associated microenvironment.

PDA develops through different stages of low-grade dysplasia (PanIN1A/B, PanIN2). While PanIN1 lesions develop after acquisition of an oncogenic Kras mutation and upregulated EGFR signaling, further progression to PanIN2 requires additional gene mutations in tumor suppressors such as CDKN2A. NE-PanIN cells respond to neuropeptides and promote expansion of lesions. Increased fibrogenesis accompanies low grade dysplasia formation. This is mediated by AAM, which are attracted by PanIN cells and PSCs. AAM generate a tumor-promoting and immunosuppressive microenvironment by inducing fibrogenesis, inhibiting CD8+ T-cells, and by releasing factors that promote expansion of lesions. A strong desmoplastic reaction can be observed at PanIN3 lesions (carcinoma in situ) and PDA, with different types of cancer-associated fibroblasts (CAFs) that in part originate from activated PSC. CAFs stimulate TAMs and contribute to the presence of CD4+ T cells (such as Treg cells and Th17 cells) and B cells. At this stage, inactivating mutations in tumor suppressor genes (e.g. SMAD4, BRCA2, TP53) are observed.

There is a considerable cellular heterogeneity within pancreatic neoplasms. For example, although tuft cells are rarely found in normal pancreas, metaplastic epithelia and low-grade dysplasias contain a large number of this chemosensory cell type, which mediates inflammatory responses in other glandular tissues^44–47,^{48, 49}. Tuft cells express high levels of DCLK1⁵⁰ and often are conflated with the DCLK1-positive putative progenitor cells. However, unlike the DCLK-positive progenitor cells, tuft cells have a distinct actin cytoskeletal structure, terminating in apical microvilli-like tufts and express gustatory sensory pathway components⁵¹. In normal glandular tissues, tuft cells act to sense irritants⁵² and infection^44–47 and respond by emitting signals that promote an appropriate inflammatory response, either directly by signaling to inflammatory cells^44–46 or indirectly, by stimulating associated neurons^{47, 52}. Studies are underway to determine whether tuft cells in neoplasms serve a similar role in mediating the immune response to transformed cells.

Other cells found in PanIN lesions take on a neuroendocrine (NE) phenotype, though they have transdifferentiated from the transformed exocrine epithelia^{53, 54}. These NE PanIN cells (Fig. 2) respond to neuron signals to promote lesion growth⁵⁴, but also frequently delaminate and enter the surrounding stroma⁵³. Transdifferentiation to the NE phenotype is promoted by high activity of the oncogene c-Myc. Intriguingly, early dissemination of neoplastic cells in the absence of frank carcinoma is consistent with findings from Rhim et al⁵⁵ that neoplastic cells not only invade the surrounding stroma before progression to carcinoma, but can be found in the circulation and at secondary sites. The possibility that PDA metastases not only are seeded prior to the detection of cancer, but before progression to cancer, further emphasizes the importance of understanding the cellular plasticity evident at the earliest neoplastic stages of the disease.

Neoplastic and PDA Microenvironments

The microenvironment associated with PDA has complex functions, as it contributes to cancer development, progression to invasion and metastasis, cancer stem cell formation, and generates an ECM-composed barrier to perfusion and treatment⁵⁶. However, it also seems to restrain tumor cells, since tumors with reduced stroma are undifferentiated and more aggressive⁵⁷.

The desmoplastic stroma is organized by interactions among cell types including pancreatic stellate cells (PSCs), activated cancer-associated fibroblasts (CAFs), macrophages, other immune cells, precancerous or cancer cells, and the microvasculature (Fig. 2). Stromal cell composition varies among neoplastic and cancerous lesions⁵⁸, but also among PDA subtypes^{59, 60}.

Macrophage Populations in Tumor Development and Progression

Seminal work from the Barbacid laboratory has shown that besides an oncogenic Kras mutation, pancreatitis (pancreatic inflammation), as an inhibitor of oncogene-induced senescence, is a crucial factor for PDA development^{12, 13}. Adult acinar cells are not very susceptible to transformation by oncogenic Kras alone and ADM initiation requires additional inflammatory signaling events. For example, partial deletion of Nr5a2 (Nr5a2^+/− mice), a gene which encodes an orphan nuclear receptor that among other factors maintains acinar cell identity, mediates a transcriptional switch controlling expression of pro-inflammatory genes⁶¹, linking the loss of cell identity with an inflammatory response. Oncogenic Kras also drives further pro-inflammatory signaling in precancerous neoplasia through activation of STAT3^62–65, NF-κB^{9, 66–68} and GSK3/NFAT signaling^{69, 70}. Moreover, acinar cells that express activated KRAS (with the G12D mutation) induce local inflammation by upregulating chemoattractants for inflammatory macrophages⁶⁷. Inflammatory macrophages contribute to acinar cell dedifferentiation and ADM and preneoplastic lesion formation by secretion of inflammatory mediators such TNF and CCL5 ^{16, 67}. In addition, inflammatory macrophages also upregulate tissue inhibitor of metalloproteinases or matrix metalloproteinases (MMPs), which contribute to re-organization of the acinar microenvironment by promoting ADM^{16, 71}. However, Kras-driven local inflammation is an inefficient driver of oncogenic progression to PDA and requires additional inflammatory insults and genetic alterations^{12, 13, 72}.

Pre-cancerous PanIN1 lesions actively shape the microenvironment by secreting factors such as IL-13 to initiate a macrophage phenotype switch towards alternatively-activated M2 macrophage populations¹⁸, which (in mice) are best characterized by YM1, Arg1, FIZZ1 and IL-1ra expression¹⁷. IL-13 depletion in KC mice shows that this macrophage population (often labeled as YM1⁺ macrophages) has a role in suppressing inflammation and promoting lesion growth through secretion of CCL2 and IL-1ra, but also is a crucial driver of fibrogenesis^{17, 18}.

In pancreatic tumors, tumor-associated macrophages (TAMs) comprise inflammatory and alternatively-activated macrophage populations. TAMs regulate immunosuppression through secretion of immunosuppressive cytokines and chemokines, and by promoting Treg recruitment and providing co-stimulatory signals to block T-cell proliferation^73–75. In addition, TAMs regulate fibrinogenesis, vascularization and angiogenesis^{76, 77}, and promote epithelial-to-mesenchymal transition (EMT), invasiveness and metastasis ^78–80. Moreover, these immunosuppressive populations are expanded in the microenvironment, after epithelial necroptosis in response to chemotherapy ⁸¹, or after radiation-induced damage of cancer cells⁸².

Due to the contributions of macrophages to tumor initiation and progression, strategies to target these cells might reduce fibrosis and increase the anti-tumor T-cell response, which might be effective in combination with chemotherapy ⁸³ (Fig. 3).

Figure 3. Strategies for Therapeutic Targeting of PDA.

Chemotherapy alone is not very effective against PDAs, so strategies have been developed to target cells that support tumor growth and progression. These include targeting CD4+ T cells and strategies to shift macrophage populations to an inflammatory phenotype, to generate an inflammatory microenvironment. Strategies to decrease fibrosis include depletion of AAMs or TAMs, or direct targeting of fibroblasts.

CAFs

The ECM in the stroma of PDAs is mainly produced by CAFs and comprises many components, including collagen I, collagen IV, fibronectin, laminin, hyaluronan, glycosaminoglycan, growth factors, cytokines and chemokines, proteinases and other factors^{20, 84–86}. CAFs can be of different origin and result from pancreatic stellate cells (PSCs), tissue resident fibroblasts, bone marrow-derived mesenchymal cells, adipose tissue and epithelial cells that underwent EMT (discussed in more detail in⁸⁷).

Major contributors to the desmoplastic reaction in PDA are PSCs⁸⁸. PSCs in the normal pancreas present as periacinar cells, where they are quiescent^{89, 90}. PSCs are activated during acute or chronic inflammation^91–93, changing their morphology into myofibroblast-like cells which express αSMA and are proliferative⁹⁴. The resulting activated fibroblasts mediate enhanced production of ECM molecules, orchestrate MMP and TIMP-mediated matrix reorganization, and crosstalk with tumor cells via growth factors and cytokines such as EGF, basic fibroblast growth factor (bFGF), and Wnt family members^95–97. The high abundance of CAFs and generation of a collagen- and HA-rich ECM promotes constriction of vasculature, creating a hypoxic microenvironment²⁰. This requires tumor cells to adapt their metabolism away from oxidative phosphorylation⁹⁸, often relying on metabolites produced by other cells within the microenvironment, including CAFs, for biosynthesis and energy^99–101.

There are several subtypes of CAFs in the PDA microenvironment, with different functions ¹⁰². Myofibroblastic CAFs (myCAFs) express high levels of αSMA and are in close proximity to tumor cells, whereas inflammatory CAFs (iCAFs) express low levels of αSMA and high levels of chemokines and cytokines and are located farther away from cancer cells^{87, 103}. iCAFs contribute to an immune-suppressive environment by excluding CD8⁺ T cells^103–105. A putative immune-modulatory population of antigen presenting CAFs (apCAFs) expresses MHC class II and CD74¹⁰². In organoid cultures and in KPC mice, CAF heterogeneity can be induced by factors produced by tumor cells, such as TGFB and IL1. TGFB promotes their differentiation to myCAFs, whereas IL1 promotes generation of iCAFs¹⁰⁶. CAF populations are therefore dynamic and convert among phenotypes as lesion- or tumor-associated macrophages^{102, 106}.

In addition to their role in tumor development and progression, CAFs promote resistance to chemo- or radiation therapy through release of exosomes¹⁰⁷ and through Hedgehog signaling^{108, 109}. However, CAF and fibrosis depletion for clinical purposes (Fig. 3) is controversial, because it may lead to more aggressive tumors¹¹⁰. This is because iCAFs, which are low in αSMA and IL6, promote tumor growth, whereas myCAFs, which express high levels of αSMA, can function in restraining tumor growth^{27, 103, 111}.

B Cells and T Cells in the TME

In KC mice, early-stage pancreatic lesions have a low abundance of T cells ^{18, 67, 80}. In patients, infiltration of PDAs by CD4+ T cells and decreased infiltration of CD8+ T cells have been associated with shorter survival times ^{19, 112}. CD4+ T-helper 17 (Th17) cells are pro-inflammatory cells that are recruited by CAFs. They produce IL17 and IL23, which contribute to invasion and metastasis as well as microvessel density¹¹³. Expansion of Th17 cells is promoted by mast cells¹¹⁴, which are recruited by PDA cells and also contribute to neovascularization and metastasis by releasing VEGF and FGF2¹¹⁵. CD4+, FOXP3+ and CD25+ Treg cells or T- suppressor cells are found in high numbers in the TME and produce TGFB and IL10. Presence of Treg cells correlates with poor survival of patients. Treg cells, along with Th2 cells, which exhibit tumor-promoting functions by secreting IL-13¹¹⁶, can generate an immunosuppressive environment.

CD8+ T cells in the PDA microenvironment are associated with increased survival¹¹⁷, although PDA cells embedded in fibrosis can escape the cytolytic effects of these T cells. CD8+ T cell responses are repressed by increased presence of immunosuppressive macrophages⁸², which can produce CCL2, the ligand for CCR2¹⁸. In mice, inhibition of CSF1R or CCR2, reduced numbers of Treg cells and increased numbers of CD8+ T cells, which increased the effects of chemotherapeutic agents and radiation therapy ^118–121 (see Fig. 3).

Studies of mice and human tumor PDAs have found evidence for infiltration by B cells at later stages of tumor development and in metastases^{122, 123}. B cell recruitment to the stroma can be mediated through CXCL13, which is expressed by CAFs. Stromal B cells produce IL35 to promote tumor cell proliferation ¹²³. Moreover, B cells in the TME contribute to immune suppression¹²⁴; and depletion of B cells reduces PDA progression and metastasis in mice¹²².

Molecular Subtypes and Personalized Treatment

Tumors were historically grouped based on their morphologic characteristics and associated with patterns of progression and patient outcome, based largely on the relative levels of differentiation of the cancer cells. Genome and transcriptome analyses have recently complemented pathology analyses, assigning tumors to molecular subtypes, with the aim to better predict outcomes and select treatment. In PDA, there are four major driver mutations, KRAS, TP53, CDKN2A and SMAD4, all currently undruggable¹²⁵. Interestingly, metastases largely reflect a subclone of the primary tumor¹²⁶, suggesting that none of the major driver mutations are determinant of the switch to metastatic behavior. Instead, this transition appears to be strongly promoted by metabolic rewiring and epigenetic changes¹²⁷, suggesting novel vulnerabilities for metastatic PDA.

Although genomic information might be used to select treatment with in a small population of PDA patients, it has not been robust for molecular subtyping PDAs as a whole. However, transcriptomes have been used to assign PDAs into meaningful categories. Collisson et al¹²⁸ performed microarray analyses of gene expression profiles of laser-capture microdissected pancreatic cancers, to exclude confounding signals from the fibroinflammatory stroma. Moffitt et al subtyped PDA samples using array hybridization and RNA-seq analyses, and separated tumor cell signatures from parenchymal and stromal cell signatures⁶⁰. Bailey et al performed RNA-seq analysis of 96 PDAs with high epithelial cellularity⁵⁹. Thus far, the only consensus subtypes that have been identified by these approaches are the well-differentiated pancreatic progenitor/classical subtype, which retains pancreatic and epithelial identity, and a quasi-mesenchymal, basal-like, squamous subtype, which is undifferentiated, shows downregulation of pancreatic and epithelial identity factors, and is associated with poorest prognosis. In mice treatment with blocking antibodies against CSF1R¹²⁹ or CXCR2¹³⁰ can shift the basal-like subtype to a more differentiated phenotype, suggesting the intriguing possibility of inducible plasticity within subtypes.

Tiriac et al performed transcriptome analysis of PDA patient-derived organoids and identified gene-expression signatures associated with responses to specific therapies¹³¹. Taking advantage of the in vitro nature of the system, the authors then subtyped the organoids based on their responses to chemotherapeutic agents, which they called pharmacotyping. The authors were then able to identify transcriptomes that associated with specific pharmacotypes. The authors found that these transcriptomes correlated with patient responses to therapy in a retrospective data set. Interestingly, the pharmacotypes did not correlate to their basal and classical signatures. Whether these pharmacotypes can be used to predict patient response to therapy in a prospective study is yet to be determined.

Future Directions

Pancreatic cancer is the deadliest of the common cancers, and survival times have increased only marginally in the past several decades. Our attempts to effectively treat patients with pancreatic cancer have been thwarted by our inability to detect PDA at an early stage, as well as the lack of efficient combination therapy.

Although mutation of KRAS is one of the earliest events in pancreatic tumorigenesis, oncogenic KRAS both activates cell intrinsic carcinogenic pathways while also promoting interactions among epithelial cells, immune cells and fibroblasts that collectively create an immune suppressive fibroinflammatory stroma. Once mutant KRAS expression is blocked rapid stromal remodeling occurs¹³². Mutant KRAS also induces inflammation, which conversely promotes the plasticity of neoplastic cells at all stages of tumor progression^{18, 129, 130}. Understanding this coevolution of tumor development and the stromal response may be the key to early detection and treatment. For example, early detection has focused largely on detecting unique byproducts of the tumor cells. However, markers of the fibrotic and inflammatory responses might also be used for early detection. Moreover, desmoplasia acts as a physical barrier to drug perfusion¹⁰⁸ and the interstitial fluid pressure created by hyaluronan deposition collapses the vasculature necessary for drug delivery^{20, 133}. This all suggests that successful treatment strategies will need to target these stromal compartments.

Taken together, work from recent years emphasizes the importance of dissecting the complex, vicious circle of progression of pancreatic tumors as a whole, not just the cancer cell component. Increasing our understanding of the complex interactions between cells of the stroma and low-grade and high-grade dysplasia cells might lead to new approaches to both detection and therapy.

Abbreviations

AAM

alternatively-activated macrophage

ADM

acinar to ductal metaplasia

CAFs

cancer-associated fibroblasts

CCL2

chemokine (C-C motif) ligand 2

CCR2

C-C chemokine receptor type 2

CSF1R

colony stimulating factor 1 receptor

CXCR2

CX-C chemokine receptor type 2

DCLK1

doublecortin like kinase 1

ECM

extracellular matrix

EGFR

epidermal growth factor receptor

EMT

epithelial to mesenchymal transition

IPMN

intraductal papillary-mucinous neoplasms

MMP

matrix metalloproteinase

NE

neuroendocrine

PanIN

pancreatic intraepithelial neoplasia

PDA

pancreatic ductal adenocarcinoma

PSCs

pancreatic stellate cells

SMA

smooth muscle actin

TAM

tumor-associated macrophages

TGFB

transforming growth factor beta

TIMP

tissue inhibitor of metalloproteinases