Pancreatic cancer biology and genetics from an evolutionary perspective

KRAS

Activating mutations of the KRAS oncogene on chromosome 12p are the most common genetic abnormality, present in approximately 95% of pancreatic tumours analysed^44,47. KRAS encodes a member of the RAS family of GTP-binding proteins that mediate many cellular functions, including proliferation, cell survival and cytoskeletal remodelling⁴⁸. Most mutations in KRAS are believed to cause a constitutively active protein, although recently some KRAS mutants, specifically KRAS^G12C, have been demonstrated to have nucleotide cycling activity that is draggable^49,50. In approximately 4% of pancreatic cancers KRAS amplification occurs together with the oncogenic mutation⁴². BRAF, the signalling mediator immediately downstream of KRAS, is mutated or amplified in a mutually exclusive manner from KRAS in 3–4% of cases^42,51. Intriguingly, although KRAS mutations are found in 99% of PanIN-1s³⁷, no more than 95% of pancreatic cancers have a KRAS or BRAF mutation, supporting the notion that a KRAS mutation is not strictly required for the development of pancreatic cancer.

CDKN2A

The tumour suppressor gene CDKN2A encodes p16^INK4A and p19^ARF through a common locus on chromosome 9p⁵². The genomic structure of CDKN2A is highly complex in that it produces two mRNAs, each with a unique first exon but sharing exons 2 and 3. However, exon 2 in the mRNA that encodes p19^ARF is derived from a different reading frame from that of the mRNA encoding p16^INK4A; thus, the two proteins are not isoforms⁵³. The high frequency at which this locus is inactivated in pancreatic cancers (>90%)⁵⁴ raises the question of which tumour suppressor is being selected for inactivation⁵⁵. Evidence in mice and humans points to p16^INK4A as the primary target because mutations in exon 1 — which is used in the transcript encoding p16^INK4A — that would leave p19^ARF functional have been reported in both pancreatic cancers and melanomas⁵⁵. However, large homozygous deletions often inactivate both proteins so that loss of either may contribute to pancreatic carcinogenesis by different mechanisms. For example, p16^INK4A inhibits cell cycle progression through the G1/S checkpoint mediated by CDKs such as CDK4 and CDK6 (REF. 56). The loss of this important restraint leads to unchecked CDK4 and CDK6 expression and cell cycle progression through the G1/S checkpoint. Telomere shortening in concert with the loss of the G1/S checkpoint creates an environment that facilitates chromosome instability and the accumulation of structural rearrangements, including fold-back inversions, which are a form of mutation relatively specific to pancreatic cancer^10,57. By contrast, p19^ARF induces cell-cycle arrest independently of CDKs by binding to the E3 ubiquitin ligase MDM2 to inhibit p53 degradation; loss of p19^ARF abrogates p53-induced apoptosis and cell cycle arrest⁵⁵. In a small number of pancreatic cancers that retain CDKN2A, somatic mutations in F-box and WD repeat domain containing 7 (FBXW7), which encodes a ubiquitin ligase that targets cyclin E for degradation, or the gene encoding the ring E3 ubiquitin ligase anaphase promoting complex sub-unit 2 (APC2, also known as ANAPC2) that regulates chromosome segregation and mitotic fidelity have been reported^43,51.

TP53

p53 is a latent transcription factor that is activated by stimuli such as DNA damage or stress. Upon activation, p53 performs many functions, including regulation of the G1/S checkpoint, maintenance of G2/M arrest to enable DNA repair, and apoptosis⁵⁸. TP53 is somatically mutated in up to 85% of pancreatic cancers⁴⁷. As many as 66% of TP53 mutations in pancreatic cancer are missense mutations that affect the DNA binding domain^43,47. Although not completely inactivating, such mutations provide oncogenic gains of function compared with the normal protein⁵⁸, often in association with nuclear accumulation of p53 in the neoplastic cell³⁹. Previous studies based on immunolabelling for the p53 protein product in fixed tissues have drastically underestimated the frequency with which TP53 is inactivated by not accounting for somatic alterations that result in a loss of protein expression^39,59. However, it is clear that nonsense mutations, frameshifts and homozygous deletions not detected by p53 immunolabelling are notable mechanisms of TP53 inactivation in pancreatic cancer^43,44. In one study of late-stage pancreatic cancers, up to half of all mutations in TP53 were predicted to cause a loss of protein expression leading to null alleles⁴⁷. Among cancers that have no detectable TP53 mutation, numerous genes, such as excision repair cross-complementation group 4 (ERCC4), ERCC6, E1A binding protein p300 (EP300) or RAN binding protein 2 (RANBP2), are mutated at low frequencies and may provide alternative inactivation of one or more p53 functions. Of particular importance is ATM, a gene also implicated in familial pancreatic cancer that may also be sporadically mutated⁶⁰; ATM phosphorylates p53 directly and has pivotal roles in responding to cell stress and maintaining genome integrity⁶¹.

SMAD4

SMAD4 is inactivated in approximately 55% of pancreatic cancers, either by homozygous deletion (30%) or by an intragenic mutation in association with loss of the second copy (25%)⁶². The SMAD4 protein is a crucial co-transcription factor and mediator of the transforming growth factor-β (TGFβ) canonical signalling pathway for cellular growth, differentiation and maintenance of tissue homeostasis⁶³. The TGFβ pathway is notable for its dualistic nature in cancer; during the early stages of the clonal expansion phase (PanIN-1 and PanIN-2) it restrains neoplastic cell growth, whereas in later stages of clonal expansion (PanIN-3 and invasive cancers) TGFβ signalling promotes growth, in part owing to the loss of SMAD4 and the canonical arm of the TGFβ pathway^64,65. Up to 10% of pancreatic cancers without SMAD4 alterations harbour an inactivating mutation in TGFβ receptor 1 (TGFBR1), TGFBR2, activin A receptor type 1B (ACVR1B) or SMAD3, providing alternative mechanisms to inactivate TGFβ signalling^43,44.

An important facet of SMAD4 inactivation is the context in which it occurs. One study that evaluated the patterns of coexistence of driver gene mutations found that most pancreatic cancers with SMAD4 inactivation had coexistent TP53 gain-of-function alterations, whereas pancreatic cancers that retained SMAD4 were more likely to have TP53 loss-of-function alterations⁴⁷. This relationship probably reflects the interdependence of p53 and TGFβ for transcriptional gene activation (discussed in greater detail in the next section). Thus, TP53 mutant pancreatic cancers can be segregated into two types, those with TP53 loss of function and wild-type SMAD4, and those with TP53 gain of function and SMAD4 loss of function. Validation of these genetic subtypes in independent cohorts and their relationship to therapeutic responses remain to be discerned, including the extent to which these genetic subtypes overlap with other biological subtypes that have been described^66–68 (BOX 3). Such efforts would be particularly worthwhile in the context of clinical trials of patients with pancreatic cancer for which pretreatment tissues are available⁶⁹.

Synergistic effects between mutations

Although the genes described thus far reveal those pathways that are disrupted, it is unlikely that each gene exerts its effects independently. Mutant KRAS has been shown to impede TGFβ signalling by inhibiting receptor SMADs^70,71 and inhibiting p53 by blocking its amino-terminal phosphorylation⁷². In turn, TGFβ signalling interacts at many levels with the RAS-RAF-ERK pathway⁷³. p53 is required for TGFβ target gene transactivation by binding to distinct cis-enhancer elements in the same gene promoters through association with receptor SMADs⁷⁴. Furthermore, mutant p53 and SMADs form a complex that inhibits p63, enabling aggressive features of cancer cells⁷⁵. Thus, disruption of crucial driver genes creates a complex tumorigenic network that is expected to greatly alter the systems biology of the cell. An improved understanding of this altered system could be exploited to identify unique vulnerabilities and target specific mutant proteins or pathways⁷⁶.

The desmoplastic stroma of pancreatic cancer

The epithelial wound healing response is particularly robust in the pancreas, as evidenced by histology findings in pancreata from patients with chronic pancreatitis⁸⁵. This response is evolutionarily conserved to support the multicellular state and is coordinated in large part by TGFβ (REF. 86). Features associated with a wound healing response include fibroblast activation, immune suppression, remodelling of the ECM and trophic signals to promote re-epithelialization⁸⁷. That the stroma en masse has an influence on the neoplastic epithelium is undisputed, yet the extent to which each cell type supports rather than restrains neoplastic growth is an area of immense interest.

Co-option of the stromal response by cancer indicates that the stroma provides paracrine signals that select tumour cells with certain properties. Such paracrine signals originate from various sources but have mostly been described for α-smooth muscle actin (αSMA)⁺ myofibroblasts. Myofibroblasts are derived from normally quiescent pancreatic stellate cells (PSCs) in the pancreas. Upon activation, PSCs lose their cytoplasmic lipid, transdifferentiate into αSMA⁺ myofibroblasts with proliferative capacity, secrete various growth factors, such as TGFβ, fibroblast growth factor (FGF) and platelet-derived growth factor (PDGF), and substantially increase production of ECM components⁸⁸. Moreover, unlike the non-neoplastic setting, in which proliferative signals are eventually quelled with the culmination of repair following injury, PSCs and other stromal mesenchymal cells are continuously activated by the neoplastic epithelium itself, which secretes PDGF, TGFβ and sonic hedgehog (SHH)^89–91. Evidence supporting the tumour-promoting role of PSCs and their myofibroblastic derivatives comes from mouse studies in which pharmacological inhibition of PSC activation by the vitamin D analogue calcipotriol led to stromal collapse, smaller tumours and improved chemotherapeutic delivery⁹². Analogous results are seen following stromal ablation by short-term inhibition of Hedgehog signalling⁹³ or enzymatic ablation of hyaluronic acid (HA), a major constituent of the ECM that is secreted by myofibroblasts^94,95. By contrast, two recent studies using a mouse model of pancreatic cancer indicated that stromal ablation by conditional deletion of Shh (chronic inhibition)⁹⁶ in or of αSMA⁺ myofibroblasts themselves⁹⁷ led to more aggressive tumours. This suggests that distinct components of the myofibroblastic secretome have tumour-restraining properties, although one cannot entirely rule out that the secretomes of other stromal cell populations (for example, macrophages) have tumour suppressive features as well. These opposing forces occur over geographic space and time and in part may underlie the formation of intratumoural heterogeneity by favouring selective sweeps of one clonal population at the expense of another.

The effects of the microenvironment on pancreatic cancer cells go beyond stromal cells. The abundant ECM produced by αSMA⁺ myofibroblasts is rich in HA, fibrillar collagens and secreted protein acidic and rich in cysteine (SPARC), which act as a physical barrier to the neoplasm^94,95. HA is a large negatively charged glycosaminoglycan that binds to large amounts of water, leading to high hydrostatic pressure and interstitial fluid pressure (IFP)⁸⁰. The swelling caused by high concentrations of HA stresses collagen fibrils tethered to cancer or endothelial cell surface receptors, which contract in response, leading to pathological IFP, widespread vascular collapse and hypoperfusion. Although this is problematic from the point of view of therapeutic delivery^93,94, such a phenomenon itself may cause geographic isolation of neoplastic cells that are already in a nutrient-restricted environment, thus enforcing allopatric evolution and intratumoural heterogeneity. IFP also leads to hypoxia, a pervasive feature of the pancreatic cancer microenvironment that serves as yet another powerful selective force⁹³. Pancreatic cancer cells can adapt to these environmental pressures through metabolic reprogramming and shunting of resources; these adaptations occur in association with KRAS mutations well before the onset of invasion and are continuously refined with subclonal evolution⁸³’⁹⁸. This is analogous to ecological studies that have shown that the most successful invasive species are those that are predisposed to the most efficient use of available (limited) resources⁷⁸.

The immune system in pancreatic cancer

The immune system represents yet another highly complex programme that has evolved to support the multicellular state⁹⁹. In the context of cancer evolution immune cells represent native predators. Abundant evidence indicates that the pancreatic cancer microenvironment is immunosuppressed at multiple levels, some of which occur in association with the clonal expansion phase of the neoplasm and themselves may enforce genetic bottlenecks in a temporal and spatial manner¹⁰⁰. In general, the pancreatic cancer microenvironment is notable for T cell suppression by several mechanisms, including an accumulation of CD4⁺ forkhead box P3 (FOXP3)⁺ regulatory T cells (T_reg cells), M2 tumour-associated macrophages (TAMs), myeloid-derived suppressor cells (MDSCs) and fibroblast activation protein (FAP) ⁺ fibroblasts, a type of stromal cell distinct from αSMA⁺ myofibroblasts in the pancreatic cancer microenvironment^84,101,102. The endogenous cytotoxic T cells do not seem to be dysfunctional, as mechanisms to bypass their suppression unmask latent immune responses and promote intratumoural accumulation of T cells^101,103.

Metastasis of pancreatic cancer

Although metastasis is managed clinically as a distinct stage, from an evolutionary standpoint it is a reflection of clonal competition and fitness levels in the primary site (FIG. 1c). Four lines of evidence support this notion. First, dissemination itself has not been shown to be a rate-limiting step for the formation of metastases, as 99% of cells that enter the circulation do not survive beyond 24 hours¹⁰⁴; moreover, dissemination occurs from the earliest stages of pancreatic carcinogenesis⁷⁹. Second, no metastasis-specific genes have been found in pancreatic cancer; instead, a substantial proportion of metastatic efficiency is determined by the genetic alterations that arise during the clonal expansion phase itself (KRAS, CDKN2A, TP53 and SMAD4), before the moment of invasion^47,65. Thus, the genetic features of the parental clone play an important part (albeit they are not the only factor) in determining the extent to which the clone will successfully adapt and survive in foreign microenvironments should metastasis occur. Third, in two independent studies encompassing 13 unique patients, metastatic subclones were shown to arise from large populations of cells in the primary tumour^10,105. These subclones can be identified by their unique set of passenger mutations or structural rearrangements, which are genetic markers of the life history of that lineage⁹. Finally, mathematical models predict at least 5–10 years for the emergence of a metastatic subclone following development of the parental clone, again implying the importance of time for adaptation within the microenvironment⁹. The complement of features of the pancreatic cancer cell or its microenvironment that dictate metastasis to the liver, lungs or peritoneum has yet to be determined. However, recent data using lineage tracing in a mouse model of pancreatic cancer indicate that multi-clonal seeding is required to initiate metastases in an organ-specific manner¹⁰⁵.

Is KRAS the only initiating driver gene in sporadic pancreatic cancer?

KRAS is undoubtedly important for pancreatic cancer biology, and extensive efforts are under way to target this oncoprotein¹⁰⁶. What is debatable is whether KRAS mutations are the only initiating event in pancreatic cancer.

Oncogenic KRAS mutations can be found in the pancreata of patients with no evidence of PanINs or pancreatic cancer, suggesting that they are necessary but not sufficient to initiate pancreatic carcinogenesis¹⁰⁷. Moreover, people with germline mutations in KRAS have not been reported to have a higher risk of developing pancreatic cancer or other non-neoplastic pancreatic sequelae; instead they develop diseases related to developmental delay, bone marrow failure and syndromic cardio-facio-cutaneous disorders, all probably a result of oncogene-induced senescence¹⁰⁸. Thus, it seems paradoxical that KRAS mutations that may cause senescence can initiate pancreatic cancer. One explanation for this paradox is that the spectrum of KRAS germline mutations differs with respect to the codons affected and thus they do not lead to KRAS hyperactivity to the same extent as oncogenic mutations, for example, G12S, K117R and A146T mutations, in patients with Costello syndrome compared with G12D mutations in those with pancreatic cancers¹⁰⁸.

Compelling experimental evidence that supports the notion that mutant KRAS in preductal epithelial cells can initiate pancreatic cancer is its ability to inhibit immune-induced senescence and promote localized immunosuppression^27,109,110. Oncogenic KRAS has also been shown to induce expression of functional interleukin-17 (IL-17) receptors on transformed epithelial cells while stimulating infiltration of IL-17-producing T helper 17 (T_H17) cells and γδ T cells into the adjacent microenvironment. As a result, the transformed epithelial cells undergo paracrine stimulation by the secreted IL-17, which supports clonal expansion of the KRAS mutant population¹¹¹. Thus, although the random occurrence of an oncogenic KRAS mutation may cause senescence in most instances, occasionally the KRAS mutant cell may survive long enough to incite a local immunotolerant environment that supports its clonal expansion into a large enough population for additional genetic events to occur¹¹² (FIG. 1a). This scenario is consistent with mathematical models that predict that at least a decade is required from initiation to formation of the clonal population that will eventually breach the basement membrane and become an infiltrating carcinoma⁹.

An alternative possibility that should be considered is that mutations in KRAS are not always the initiating event but may be a driver gene alteration that is selected for in the clonal expansion phase following a tumour cell of origin first acquiring a different driver gene mutation. An example of a strong candidate for an alternative initiating driver gene is CDKN2A, as CDKN2A mutations are linked to an inherited risk of pancreatic cancer²¹. Inherited mutations in DNA damage repair genes such as BRCA2 do not negate the possibility that genes such as CDKN2A could be an alternative initiating driver, as they act by increasing the number of potentially deleterious genetic events per cell division and hence the chance that inactivating mutations in these driver genes occur. Consistent with this interpretation, there is no difference in the genetics of familial pancreatic cancers compared with those in which the disease occurred sporadically²¹. Finally, although not experimentally studied, constitutional epimutation may be a mechanism of pancreatic carcinogenesis¹¹³. This would be supported by reports that many patients with a strong familial pattern of inheritance do not have an identifiable germline genetic event²¹. It is crucial to understand these possibilities in light of limited success in screening for pancreatic cancer or in developing chemopreventive strategies thus far. For example, anti-inflammatory or immunomodulatory agents may have a greater preventive effect for KRAS-initiated pancreatic cancers than for those that arise as a result of loss of a tumour suppressive mechanism¹¹⁴.

What is the importance of mutations in epigenome regulatory genes?

Perhaps the biggest revelation from high-throughput sequencing of many cancer types, including pancreatic cancer, has been the identification of recurrent somatic mutations in genes encoding epigenome regulators, specifically members of the SWI/SNF complex and the histone-lysine N-methyltransferase 2 (KMT2) family. Individually, members of each gene family are somatically mutated in a small percentage (<5%) of pancreatic cancers analysed^42–45. Collectively, the frequency of somatic alterations for any member of these gene families is higher, suggesting that a common epigenomic phenotype is selected for by various genotypes. For example, up to 30% of pancreatic cancers were shown to have an alteration in one of five different members of the SWI/SNF complex in a mutually exclusive manner¹¹⁵.

The SWI/SNF family of genes encodes proteins that make up one of two complexes, the BRG1- or HRBM-associated factor (BAF) complex and the polybromo-associated BAF (PBAF) complex. Each complex relies on ATP hydrolysis to directly disrupt histone-DNA contacts. In general the key role for SWI/SNF complexes is to control the balance between differentiation and sternness and to antagonize the action of the Polycomb repressive complex¹¹⁶. Although mutations in several SWI/SNF family members have been described in pancreatic cancer, the most frequently mutated genes are AT-rich interactive domain-containing 1A (ARID1A) and SWI/SNF-related matrix-associated actin-dependent regulator of chromatin subfamily A member 2 (SMARCA2), both components of the BAF complex^42,44. Similar to genes related to SWI/SNF signalling, the KMT2 genes encode proteins that are components of multi-subunit complexes. KMT2 proteins methylate histone H3 on lysine 4 (H3K4) to promote genome accessibility and transcription¹¹⁷. In pancreatic cancer, KMT2C (also known as MLL3) is the most commonly mutated member of this family, although mutations in KMT2D (also known as MLL2) and KMT2A (also known as MLL) are also seen^42,44. Like SWI/SNF, KMT2 genes have pervasive roles in regulating sternness and differentiation.

To date, the temporal occurrence of these gene alterations has not been explored, thus it remains to be seen whether they represent mutations acquired during the clonal expansion phase or subclonal events that are selected for their fitness advantage in the primary tumour microenvironment. This distinction is crucial, as events acquired during early carcinogenesis are expected to be more targetable than those that are subclonal in nature¹¹. In addition, unlike CDKN2A, TP53 or SMAD4, in which both alleles are targeted, members of the SWI/SNF and KMT2 gene families require loss of function of only a single allele for their effects in cancer^116,117. A determination of the requirement of the wild-type allele for cancer cell survival would be fruitful, as it may provide a therapeutic vulnerability using synthetic-lethal approaches^118,119. The temporal occurrence of these alterations also has importance from the evolutionary perspective. Mutations that arise during the intraductal clonal expansion phase provide a clue to the survival advantage required for the neoplasm to develop, and suggest that at the moment of invasion the parental clone was already maximally equipped for survival through rapid epigenetic adaptation. By contrast, mutations that arise in a subclonal manner after invasion occurs may be a reflection of the spatial heterogeneity of microenvironmental selection factors. Such an instance could be exploited to better understand the heterogeneity of the micro environment in general and in relation to stromal ablation therapies (FIG. 2).

What are the clinically relevant aspects of heterogeneity?

Heterogeneity is a loosely used term in cancer biology. At one extreme it may be used to describe inter-patient heterogeneity with respect to biological subtypes of the disease that differ in their aetiology²¹, biology^66–68 or response to therapy⁴² (BOX 3). The success of any personalized intervention depends on the specific genotype and microenvironment, and the immune and metabolic phenotypes of each patient. There is no doubt that a better understanding of such phenotypes will provide rapid improvements in clinical management, as it has already in BRCA-mutant ovarian cancers¹²⁰. At the opposite extreme is intra-tumoural heterogeneity, most often described in relation to genetics⁹, although epigenetic or phenotypic variants of pancreatic cancer can be described with this term¹²¹.

Broadly in the field of cancer research there is a lack of distinction between genetic subclonal heterogeneity within a primary tumour in general, in metastasis-initiating cells of the primary tumour specifically, or within a metastasis, each of which may have distinct clinical and therapeutic implications at a particular stage of disease¹²² (FIG. 3). A deeper understanding of these different types of heterogeneity will help to define clinically relevant subclones, and the contexts in which subclonal heterogeneity is most meaningful biologically and therapeutically (FIG. 2). Moreover, there is little distinction between heterogeneity of unequivocal driver gene alterations and heterogeneity of somatic alterations with predicted consequences in passenger genes. However, the latter provide unexplored territory with regard to the importance of spatially distinct passenger mutations within a single neoplasm in relation to immunoediting¹²³, the mini-driver model of polygenic cancer evolution¹²⁴ or recurrent regions of haploinsufficiency¹²⁵. A counterintuitive view also asks to what extent is heterogeneity reduced during tumour evolution? Although mutations and cell divisions supply heterogeneity over time, there probably exist several bottleneck events that reduce overall diversity during cancer evolution, for example, fixation in evolutionary stage 1 (initiation) and colonization in evolutionary stage 3 (introduction to foreign microenvironments).

What are the evolutionary effects of treatment?

Currently, the only potentially curative therapy for pancreatic cancer is surgical removal of the neoplasm³, causing an evolutionary effect akin to near total decimation of the cancer cell population. However, most patients who undergo surgery will develop recurrent disease, providing evidence that small populations of cancer cells are left behind either locoregionally or systemically and as predicted by computational models¹²⁶. The evolutionary dynamics by which these residual cells survive, divide and develop into clinically evident populations of cancer cells while under the selective pressures of systemic chemotherapy or radiation is unknown.

The same can be stated for locally advanced, unresectable or metastatic pancreatic cancer. It is reasonable to assume that radiation, cytotoxic chemotherapies and targeted agents that constitute the standards of care for this disease all influence cancer cell evolution¹³. However, at each stage of disease the extent to which different treatment modalities contribute to genetic botdenecks, the selection of resistant clones or the de novo formation of resistant clones remains unknown despite our general knowledge of resistance mechanisms in cancer¹²⁷. Only with dedicated studies that rely on post-treatment tissues at the time of progression can these questions begin to be addressed.