Pancreatic cancer epigenetics: adaptive metabolism reprograms starving primary tumors for widespread metastatic outgrowth

Abstract

Pancreatic cancer is a paradigm for adaptation to extreme stress. That is because genetic drivers are selected during tissue injury with epigenetic imprints encoding wound healing responses. Ironically, epigenetic memories of trauma that facilitate neoplasia can also recreate past stresses to restrain malignant progression through symbiotic tumor:stroma crosstalk. This is best exemplified by positive feedback between neoplastic chromatin outputs and fibroinflammatory stromal cues that encase malignant glands within a nutrient-deprived desmoplastic stroma. Because epigenetic imprints are chemically encoded by nutrient-derived metabolites bonded to chromatin, primary tumor metabolism adapts to preserve malignant epigenetic fidelity during starvation. Despite these adaptations, stromal stresses inevitably awaken primordial drives to seek more hospitable climates. The invasive migrations that ensue facilitate entry into the metastatic cascade. Metastatic routes present nutrient-replete reservoirs that accelerate malignant progression through adaptive metaboloepigenetics. This is best exemplified by positive feedback between biosynthetic enzymes and nutrient transporters that saturate malignant chromatin with pro-metastatic metabolite byproducts. Here we present a contemporary view of pancreatic cancer epigenetics: selection of neoplastic chromatin under fibroinflammatory pressures, preservation of malignant chromatin during starvation stresses, and saturation of metastatic chromatin by nutritional excesses that fuel lethal metastasis.

1. Introduction

Extracellular cues from the microenvironment and intracellular changes in metabolism are each capable of physiologically reprogramming eukaryotic epigenomes in cooperative ways that optimize most if not all cellular and systemic processes [1, 2]. Disturbances to these inputs can likewise pathologically reprogram epigenomes in patients with cancer and other diseases. Nutrient-derived metabolites, including acyl (RCO-) and methyl (-CH₃) groups, comprise chemical “ink” that is covalently bonded with the physical template of heredity (chromatin) to encode epigenetic information [3]. Chromatin is modified by “writer” and “eraser” enzymes that catalyze the formation and breakage of these covalent bonds. Epigenetic information is deciphered by various “reader” modules that recognize and dock to specific chromatin-bound metabolites. Importantly, different sets of readers can recruit distinct sets of downstream effector proteins that execute various chromatin-templated functions, including the same enzymes that write the epigenetic “signature” recognized by the reader modules [4].

Extracellular environmental cues influence epigenetic state by activating signal transduction pathways that dictate localization, activity, and composition of writers, erasers, and readers. Intracellular metabolism influences epigenetic state by manufacturing the epigenetic ink along with metabolite cofactors that many writers and erasers utilize to modify chromatin [5, 6]. In recent years. it has become clear that tumor cell autonomous genetic drivers, non-autonomous tumor:stroma crosstalk, and tumor metabolism can profoundly influence the epigenetic landscape of cancer [6–8]. As illustrated throughout this review, many of these influences are based on cooperative positive feedback interactions between the inputs that reprogram chromatin and the outputs from reprogrammed chromatin. To facilitate discussion of these complexities, we refer to cooperation between microenvironmental cues and chromatin as “symbiotic crosstalk” and cooperation between metabolism and chromatin as “metaboloepigenetics.” Although the terminology separates these processes for the sake of clarity, they are not necessarily mutually exclusive [9].

Pancreatic ductal adenocarcinoma (PDAC) is a premier example for how tumor cell-autonomous genetic drivers and non-autonomous stromal cues cooperate to rewire signal transduction pathways and metabolism in a manner that supports tumor cell growth and survival (“fitness”) during severe environmental stresses [10]. Such stresses include fibroinflammatory injury, extreme starvation, and hypoxia bordering on anoxia [11, 12]. Unfortunately, PDAC also remains a premier example of a recalcitrant cancer that remains one of the most lethal of all human malignancies due to lack of early detection and a high propensity to develop widely metastatic disease that typically presents suddenly and progresses rapidly [13].

Like other cancers, PDAC evolves through a multistep progression sequence (Fig. 1) beginning with benign neoplastic precursor lesions (pancreatic epithelial neoplasia: PanIN) that acquire genetic alterations in a recurrent set of cancer driver genes, most commonly KRAS, CDKN2A, TP53, and SMAD4 [14]. Neoplastic precursors that progress to malignancy form invasive primary tumors in the pancreas. A hallmark feature of primary pancreatic tumors is a dense fibroinflammatory (desmoplastic) stroma that encases malignant glands within a pressurized microenvironment. In most patients, PDAC progression terminates in widespread distant (hematogenous) metastasis, which progresses through a multistep metastatic cascade sequence. The genetics of PDAC progression has been deep sequenced in exquisite detail [14]. Beyond the genetics, each step of PDAC progression appears to be “co-driven” by other less intuitive gene:environment interactions that either activate fibroinflammatory signaling pathways or rewire metabolic adaptations to increase fitness during stress [11].

Fig. 1.

Stepwise progression of pancreatic cancer. The combination of chronic pancreatitis and activating KRAS mutations triggers neoplastic transformation of native ducts or during acinar to ductal metaplasia (ADM). Intraductal growth of neoplastic glands then progresses as benign microscopic precursor lesions known as pancreatic intraepithelial neoplasia (PanIN). PanIN progression proceeds over many years as additional genetic drivers accumulate and glands become increasingly disorganized and populated by cells with cytologic abnormalities (dysplasia). Malignant transformation occurs upon invasion into the surrounding pancreas. The resulting primary tumor is characterized by invasive glands encased within a dense fibroinflammatory stroma. Primary PDACs form glands lined by atrophic-appearing cells. Stromal stress stimulates invasive migrations that facilitate metastatic dissemination. Tumor cells that survive dissemination colonize the hepatic sinusoids, which are nutrient replete. Preparation of a pre-metastatic niche, systemic immunosuppression, and acquisition of nutrient-fueled metaboloepigenetic adaptations synergistically promote rapid outgrowth of hundreds of metastatic PDACs packed with biosynthetic-appearing cells

Contemporary work has incorporated epigenetics into these conceptual frameworks. A common theme is activation of signaling pathways or metabolic adaptations whose primary function is to generate products that are beneficial within the microenvironmental conditions that neoplastic cells are exposed to. Remarkably, these processes often simultaneously generate byproducts that reprogram chromatin with outputs that reciprocally stimulate the microenvironment, thereby generating cooperative positive feedback loops. Here we summarize several recent studies in mouse and man that have begun clarifying how changing microenvironmental conditions globally reprogram the pancreatic cancer epigenome throughout disease progression, ranging from traumatic injuries that instill epigenetic predispositions to neoplasia (Fig. 2), symbiotic tumor:stroma crosstalk networks that reprogram neoplastic chromatin in precursor lesions (Fig. 3), a myriad of adaptive metabolic strategies that maintain malignant chromatin within the stromal stresses of primary tumors (Fig. 4), and, in a dramatic reversal of fortune, an abundance of nutritional excess that saturates metastatic chromatin with metabolites that fuel widespread distant metastasis (Fig. 5).

Fig. 2.

Epigenetic memories of injury imprint a sustained adaptive response into chromatin. A During injury-free homeostatic conditions, lineage defining transcription factors (NR5A2) activate transcription of genes that maintain pancreatic acinar (or ductal) differentiation. Injury-responsive chromatin is condensed and silent. B During transient pancreatitis, fibroinflammatory activation of MAPK signaling deploys injury-responsive transcription factors (AP-1), which transcriptionally activate tissue regenerating genes and index their regulatory elements with histone acetylation and increased accessibility for induction. Sequestration of master transcription factors (NR5A2) from their cognate promoters temporarily decreases expression of differentiation genes. C Indexed gene regulatory elements are persistently heritable even after pancreatitis has resolved and tissue regeneration is complete. Indexed chromatin is now permissive for rapid transcriptional induction during subsequent bouts of pancreatitis

Fig. 3.

Symbiotic tumor:stromal crosstalk selects the neoplastic epigenetic program. Glandular units that acquire inducible epigenetic injury programs (Fig. 2C and activating KRAS mutations during pancreatitis are predisposed for neoplastic transformation. Activated KRAS stimulates MAPK signaling and persistent transcription of injury inducible genes from reprogrammed chromatin, including Sonic hedgehog and inflammatory cytokines. Those factors are secreted by neoplastic epithelial cells retain fibroblasts and immune cells, which construct a rim of fibroinflammatory stroma around the neoplastic glands. Fibroinflammatory cells reciprocally secrete growth factors that increase fitness of neoplastic epithelial cells. The symbiotic positive feedback provides strong selective pressure that favors clonal expansion of KRAS mutated epithelial cells (re)-programmed with neoplastic chromatin, thereby giving rise to the birth of a neoplastic PanIN precursor lesion. These events support neoplastic transformation and presage future development of the desmoplastic stroma that will restrain malignant progression during invasive primary tumor growth

Fig. 4.

Adaptive metabolism maintains the malignant epigenetic program. Delivery of systemic nutrients and oxygen is impaired within a primary tumor due to interstitial stromal pressures that compress blood vessels. This necessitates a variety of metabolic adaptations to maintain growth, survival, and epigenetic fidelity in the face of depleted nutrient supplies. Primary PDACs attempt to compete with fibroblasts and immune cells to import limiting supplies of systemically delivered nutrients through cell surface receptors, although they mostly rely on alternative modes of indirectly extracting nutrients out other less desirable foodstuffs including extracellular matrix (e.g., macropinocytosis), damaged organelles (e.g., autophagy), and stromal cell waste (e.g., symbiotic metabolite exchanges). Growth factor signaling (phosphorylated ACLY) and hypoxic responses are also adopted to synergize with nutrient scavenging adaptations. This sustains production metabolites that fuel histone acetylation (acetyl CoA) and DNA/histone methylation (SAM) while simultaneously generating complementary “oncometabolites” (succinate, fumarate, L-2-hydroxyglutarate) that inhibit the reversal of DNA and histone methylation in a site-specific manner that activates oncogenes in euchromatin (green) and silences tumor suppressor genes in heterochromatin (red)

Fig. 5.

Nutrient-fueled metabolic adaptations reprogram chromatin for metastasis. Glucose is replete along the metastatic route (the hepatic sinusoids for example). High PGD catalytic rates deplete 6-phosphogluconate (6PG) substrates (not shown) which sequesters MondoA in the cytosol and prevents transcription of TXNIP. Loss of TXNIP retains glucose transporters at the cell surface, which raises glucose consumption rates. The excess glucose replenishes 6PG substrates to fuel additional rounds of PGD catalysis, and PGD products (NADPH and ribulose-5-phosphate) are obligate precursors for synthesis of biosynthetic metabolites (fatty acids, nucleotides). Excess glucose is also metabolized to acetyl CoA in parallel (TCA cycle and ACLY), which can further stimulate fatty acid synthesis in the cytosol or fuel histone hyperacetylation in the nucleus. Histone hyperacetylation is targeted to chromatin regions bound by pro-metastatic transcription factors (for example, FOXA1) to amplify transcriptional output of malignant genes. These collective events can synergize with other complementary mechanisms (especially immune suppression) to accelerate metastatic outgrowth

2. Pancreatic injury: trauma imprints epigenetic scars

Activating KRAS mutations are the earliest and most common genetic event that drives neoplastic transformation of pancreatic epithelial cells. In this context, KRAS activation stimulates signal transduction through the mitogen activated kinase (MAPK) pathways. MAPK signaling is important to resolve and repair fibroinflammatory tissue injury that occurs during acute or chronic pancreatitis. MAPK activation stimulates epithelial cell proliferation, survival, secretion of inflammatory cytokines, and acinar to ductal metaplasia (ADM) [15, 16]. ADM is a reversible cell fate switch, whereby zymogenic acinar cells partially transdifferentiate into metaplastic pancreatic ducts. It is thought that ADM helps resolve pancreatitis by suppressing zymogen production while simultaneously facilitating drainage of pooled secretions [17]. Pancreatitis is a strong risk factor for developing PDAC in humans. Emerging evidence indicates that pancreatic epithelial chromatin is stably imprinted with an epigenetic memory of injury during pancreatitis (Fig. 2) that cooperates with oncogenic KRAS mutations to facilitate neoplastic transformation (Fig. 3) after pancreatitis resolves, as detailed below and in the following section.

Assays for transposase-accessible chromatin (ATAC) and chromatin immunoprecipitation (ChIP) followed by whole genome sequencing detected the appearance increased DNA accessibility and histone H3 lysine 27 acetylation (H3K27Ac) across numerous gene regulatory elements in murine pancreata during experimental pancreatitis [18]. Remarkably, these changes persist long-term with no apparent loss of quantitative signal intensity upon resolution of pancreatitis. Distal enhancers are often targets for long-term epigenetic memory. Indeed, many of the persistently reprogrammed chromatin sites encode enhancers with DNA binding sites for injury-inducible transcription factors (TFs) such as AP-1 family members FOS and JUN. Reprogrammed enhancers are accordingly situated near genomic regions encoding gene subsets involved in cell proliferation, cell survival, embryonic development, and inflammation [18–20]. Long-term persistence of reprogrammed enhancers after pancreatitis has resolved suggests that these events can function to facilitate efficient transcriptional induction from injury-responsive regions during future bouts of injury (Fig. 2). This constitutes a functional epigenetic memory of injury and passage through an ADM, a phenomenon referred to by the authors as a “sustained adaptive response.” Complementary studies indicate that some injury-responsive TFs can also physically interact with master regulator TFs such as NR5A2 that specify the acinar cell lineage (Fig. 2A, B). This property allows injury-induced TFs to sequester master TFs away from their cognate promoters to the injury-responsive loci, thereby temporarily repressing the acinar lineage program while simultaneously inducing the ADM repair program. Reversible induction of ADM is thereby reinforced [20] without sacrificing long term lineage fidelity (Fig. 2B–C). Finally, the repressive histone methyltransferase Enhancer of Zeste Homologue 2 (EZH2) is also acutely upregulated during pancreatitis and required for subsequent reversion of ADM and proper tissue regeneration [21]. This latter finding could reflect a requirement for maintaining EZH2-catalyzed histone H3 lysine 27 tri-methylation (H3K27Me3) within facultative heterochromatin encoding non-acinar lineage genes during semi-conservative replication of chromatin.

3. Neoplastic transformation: epigenetic scars scorch the soil

Activating KRAS mutations are necessary but not sufficient to increase the fitness of pancreatic epithelial cells relative to their wild-type counterparts. Indeed, although small numbers of KRAS mutated epithelial cells spontaneously emerge during transient bouts of pancreatitis, these cells are subsequently outcompeted and cleared by wild-type cells if the pancreatitis resolves in a timely fashion [22]. These findings imply that at least two additional events are required for neoplasia to develop. First, an external source of systemic pancreatitis must provide sufficient selection pressure to allow KRAS mutant cells to proliferate beyond a “neoplasia-permissive” numerical threshold. Otherwise, they would not be able to out-compete local populations of proliferating wild-type cells during the regeneration process. Second, localized or systemic pancreatitis must persist in close proximity to any cell populations that surpass a neoplasia-permissive threshold. Otherwise, in situ neoplasia (PanINs) will probably not progress further and may even regress altogether over time. Recent studies indicate that stromal cues released during pancreatitis cooperate with oncogenic KRAS mutations to install an epigenetic program that can help satisfy these neoplastic prerequisites. Evidence indicates that this occurs through a process of symbiotic tumor:stroma crosstalk: neoplastic epigenetic programs encoding fibroinflammatory paracrine signals are selected in cells with acquired KRAS mutations because both are beneficial during pancreatitis. Fibroinflammatory cells recruited by the paracrine signals reciprocate by releasing growth factors that benefit the epithelial cells and stably imprint the neoplastic epigenetic program into their chromatin. Once systemic pancreatitis resolves, continued production of paracrine signals from the expanded epithelial population “scorches” the surrounding tissue into a pancreatitis-like condition. The scorched stroma can then continue provision of fibroinflammatory-derived growth factors, thereby completing a crosstalk-driven positive feedback loop that facilitates neoplastic transformation and the birth of a PanIN (Fig. 3), as detailed below.

ATAC-seq experiments in transgenic mice detected the appearance of thousands of newly accessible chromatin locations in KRAS mutated epithelial cells during experimental pancreatitis [19]. The combination of KRAS mutations and pancreatitis greatly expands the repertoire of accessible regions beyond that observed in wild-type cells during pancreatitis [18, 20] or secondary to KRAS mutations alone [23]. Like the sustained adaptive response described above, targeted regions are largely injury-responsive enhancers such as those with binding sites for AP-1 TFs. However, a key difference is that the resulting transcriptional outputs are expanded to include additional genes encoding extracellular matrix proteins, secreted cytokines, cytokine and growth factor receptors, and positive regulators of cell motility [19]. A subset of these cytokines attracts fibroblasts and immune cells that secrete growth factors that reciprocally support cytokine-secreting, KRAS mutant epithelial cells [19, 24–26]. The reciprocal tumor:stroma signaling is therefore capable of maintaining pancreatitis-like conditions in the vicinity of KRAS mutant cells even upon resolution of systemic pancreatitis. Another defining feature of the neoplastic epigenetic program is that gene regulatory elements encoding binding sites for acinar master regulator TFs become permanently ATAC-inaccessible. This is coupled to redistribution of acetyl chromatin readers away from those sites to the newly accessible regulatory elements [19].

These collective events and ectopic activation of ductal master regulator TFs effectively converts a reversible ADM into an irreversible “acinar to ductal neoplasia” with subsequent intraductal neoplastic outgrowth [19]. Thus, the epigenetic instructions for PanIN development and much of the future morphology of primary PDAC itself are encoded by a neoplasia-specific epigenetic program that requires the combined presence of oncogenic KRAS mutations and pancreatitis. In this scenario, if pancreatitis persists for some period time, it will specifically benefit KRAS mutated epithelial cells with injury-responsive epigenetic programs encoding fibroinflammatory paracrine signals. These will then eclipse a neoplasia-permissive threshold. At this point, their epigenetically encoded paracrine secretions can attain sufficient extracellular concentrations to continuously attract fibroblasts and immune cells that then remodel (scorch) the adjacent tissue into a fibrotic stroma that persists even after the systemic pancreatitis resolves. This model is supported by the experimental evidence presented above and by the invariable presence of concentric rims of scorched stroma that surround PanIN precursors in mouse and man [17]. A scorched stromal microenvironment is crucial because it increases fitness of KRAS mutant epithelial cells while simultaneously maintaining the neoplastic epigenetic program that generates symbiotic paracrine signals. In this way, symbiotic tumor:stroma crosstalk facilitates neoplastic transformation and stimulates subsequent PanIN precursor progression through a process of self-reinforcing positive feedback (Fig. 3).

KRAS addiction acts as a gateway for the gradual [27] or punctuated [28] acquisition of additional driver gene alterations that promote further precursor progression (dysplasia) towards malignant transformation. These are recurrent mutations or allelic losses targeting the core set of recurrently mutated tumor suppressor genes [14] along with additional lower frequency “backseat” genetic drivers that can influence evolutionary trajectories, differentiation state, and malignant propensities of the primary tumor [14, 17]. In addition to other canonical pro-tumorigenic outputs, emerging evidence indicates that late-appearing genetic drivers may also rewire metabolism to convert symbiotic crosstalk-driven neoplastic chromatin into a fully malignant epigenetic program.

Wild-type TP53 protein transcriptionally activates nutrient transporters [29] and metabolic enzymes [30] that can increase intracellular alpha ketoglutarate concentrations (α-KG). This influences α-KG-dependent dioxygenases, including ten-eleven translocation (TET) DNA demethylases and Jumonji-domain containing (JmjC) histone demethylases (Fig. 4). TET enzymes utilize oxygen and iron to couple oxidation of α-KG with hydroxylation of 5-methylcytosine (5mC) within genomic DNA. Oxidation of α-KG produces succinate which can inhibit the forward reaction [31]. Hydroxylation of 5mC produces transcriptionally active 5-hydroxymethylcytosine (5hmC) which can serve as a substrate for additional reactions that fully demethylate cytosine bases. In the setting of oncogenic KRAS, genetic inactivation of TP53 tips the balance from glucose- to glutamine-fueled anaplerotic α-KG production by tricarboxylic acid (TCA) cycle enzymes [30]. This decreases the intracellular α-KG to succinate ratio and inhibits TET enzymes, presumably because flux of glutamine-derived α-KG is directed “forward” into the TCA cycle to consume α-KG and produce succinate for export out of the mitochondria. Irrespective of the mechanism, the inverted ratio globally reverses TET-dependent 5hmC to silence expression of TP53 target genes and facilitate invasive primary tumor growth [30]. Collectively, these data imply that TP53 inactivation rewires TCA cycle flux to layer repressive DNA methylation and possibly activating histone methylation [32] upon the previously installed neoplastic program. The resulting malignant epigenetic program therefore combines elements of symbiotic tumor:stroma crosstalk with genetically-encoded metaboloepigenetics to facilitate transformation of PanIN precursor lesions into fully invasive primary pancreatic cancers.

4. Primary PDAC: epigenetic scars persist through famine and hypoxia

As increasing numbers of neoplastic cells leave the in situ precursor site and invade into the surrounding pancreatic tissue, they continue to proliferate and secrete stroma-activating paracrine signals encoded by the neoplastic epigenetic program. This causes the scorched stroma to expand and mature substantially during invasive primary tumor growth [33–35]. The resulting desmoplastic stroma is a hallmark feature of primary PDAC that effectively encases invasive epithelial glands within a dense network of stromal cells and extracellular matrix [17]. As a byproduct, rising interstitial pressures collapse the already sparse numbers of arterioles, capillaries, and venules that feed and drain the interior of the primary tumor mass [36, 37]. This results in sluggish delivery of systemically circulating oxygen, nutrients, and xenobiotics with impaired clearance of cellular waste products [37–39]. Primary PDACs therefore grow and evolve within a stressful metabolic ecosystem that is nutrient-deprived [37–39] and hypoxic [40, 41]. To cope with such stresses, invasive PDACs activate signaling pathways and unique metabolic adaptations that increase fitness under conditions of famine and hypoxia. The intricacies of these adaptations are reviewed elsewhere [11, 12, 42], and we will instead focus on how they globally impact malignant chromatin.

Most of the genomic locations that are reprogrammed during PanIN progression are spatially preserved in primary PDACs [19, 43]. Following malignant transformation, genetic backseat drivers or divergent geographic tumor:stroma ecosystems appear to shape the epigenetic landscape in more nuanced ways [44] that steer primary PDACs towards divergent subtypes [45] with characteristic lineage-defining morphologies [17], transcriptional landscapes [17, 46, 47], and stromal compositions [48, 49]. Like the crosstalk-driven neoplastic epigenetic program, PDAC lineage diversification is also largely specified by DNA-binding TFs that target chromatin writers and readers to specialized enhancer subsets [44, 50–55]. Of more general importance for primary PDACs is that most chromatin modifying enzymes require byproducts of metabolism to catalyze their reactions, while many of the resulting epigenetic products are themselves nutrient-derived metabolites. A major obstacle for starving primary PDACs therefore becomes maintaining global epigenetic state across large numbers of previously reprogrammed chromatin regions as feeder vessels collapse under the pressure of an expanding desmoplastic stroma. Emerging evidence suggests that symbiotic tumor:stroma crosstalk and unique metabolic adaptations cooperate to help maintain malignant epigenetic fidelity within the stressful conditions of a primary tumor (Fig. 4), as detailed below.

Growth factors released from the stroma [56], growth factor signaling through the phosphatidyl inositol-3-kinase Ak strain transforming (PI3K-AKT) pathway [57], hypoxia [57, 58], and oncogenic KRAS [23] each converge on acetyl CoA metabolism to maintain global histone acetylation [23, 56, 57] of reprogrammed PDAC enhancers [19, 50] within the famine of the primary tumor. Acetyl CoA is an ancient intermediary metabolite that plays a central role in carbon metabolism and is the sole donor of acetyl groups for protein acetylation [59]. Nutrients (glucose, glutamine, fatty acids, branch chain amino acids) that are normally metabolized into acetyl CoA under replete conditions are depleted within the primary tumor [37–39]. A major adaptive mechanism acquired in PanINs that also benefits primary PDACs is secretion of fibroinflammatory ligands into the stroma. As detailed earlier, these paracrine signals stimulate an expanding pool of stromal cancer-associated fibroblasts (CAFs) to release epithelial growth factors (Figs. 3 and 4) [24, 60]. Growth factors released from the symbiotic CAFs that increase fitness of PDAC cells within a desmoplastic stroma also enhance efficient acetyl CoA synthesis during starvation. Docking of CAF-derived growth factors on epithelial cell surface receptors activates PI3K-AKT signaling [24], and AKT responds by phosphorylating the serine 455 residue of ATP citrate lyase (ACLY) [61]. ACLY catalyzes ATP-dependent conversion of cytosolic and nuclear pools of citrate into acetyl CoA and is accordingly required to maintain global histone acetylation under most circumstances [62–64]. Importantly, serine 455 phosphorylation has been reported to increase the V_max of the ACLY catalytic reaction and allows acetyl CoA production to continue under nutrient-limited conditions [61, 63]. In parallel with ACLY phosphorylation, crosstalk-driven PI3K-AKT signaling and various metabolic transport adaptations can maximize import of the trace amounts of potential extracellular acetyl-precursor nutrients that remain in the stroma including glucose [65, 66], glutamine [67], fatty acids [41], amino acids [68], protein [38, 69], and components of the dense extracellular matrix itself [70, 71] (Fig. 4).

Analogous to acetyl CoA and HATs, S-adenosylmethionine (SAM) is a nutrient-derived metabolite that serves as the sole methyl donor for DNA and histone methyltransferase (HMT) reactions. Sufficient intracellular SAM concentrations are therefore important to maintain DNA and histone methylation in a starving primary tumor. Evidence suggests that at least one subset of primary PDACs can rewire serine-glycineone carbon metabolism to repurpose serine and folate metabolism away from nucleotide synthesis to favor SAM synthesis and maintenance of global DNA methylation[72]. Other data indicate that PDACs acquire additional metabolic adaptations with potential to regenerate SAM pools or limit SAM consumption, including salvaging of key intermediary nucleotide cofactors such as NAD(P)[73], scavenging nucleotide bases through symbiotic exchanges with other cells[74], and importing extracellular cysteine precursors[75, 76] as briefly described later in the discussion. However, to our knowledge these complimentary processes have not yet been investigated in a metaboloepigenetic context.

Genetically driven adaptations that promote growth [65, 77, 78] and survival [40, 79] during hypoxia may indirectly allow primary PDACs to adopt other more generalized hypoxia-driven adaptations that help preserve the PDAC epigenome. Acetyl CoA synthetase 2 (ACSS2) is highly expressed in hypoxic cells and functions to catalytically ligate free acetate with CoA to yield acetyl CoA. ACSS2 is over-expressed in hypoxic solid tumors [58, 80, 81]. This includes transgenic mouse models of PDAC [57] where it concentrates in the nucleus to partially substitute for ACLY deficiency and maintain global histone acetylation [57, 63]. This activity may proceed through ACSS2-mediated import of extracellular acetate into cells for bulk production of acetyl CoA, “recapturing” acetate released from histone deacetylation reactions for re-use by HATs [58, 63, 82], or a combination of both (Fig. 4) [58]. Hypoxic conditions can simultaneously stimulate euchromatic and heterochromatic methylation of distinct histone target residues through a variety of mechanisms. These include lowering oxygen availability for enzymatic catalysis [83, 84], slowing aerobic respiration with accumulation of succinate and/or fumarate [85–88], down-regulating demethylase enzyme expression [89, 90], and causing spurious production of L-2-hydroxyglutarate [91, 92]. Each of these outputs can inhibit different TET and JmjC demethylase activities to varying degrees. Indeed, spurious production of L-2-hydroxyglutarate results in heterochromatic histone hypermethylation, silencing of differentiation gene subsets, enhanced self-renewal, and immune evasion in hypoxic primary PDACs [93]. This may exert similar outcomes as backseat genetic inactivation to components of the COMPASS histone methylation complex that have been documented in subsets of PDAC patients [55, 94, 95].

5. Metastatic PDAC: epigenetic scars saturate from insatiable gluttony

Distant metastasis is a multi-step cascade that requires vascular intravasation to escape the primary tumor, dissemination in the circulation to seed other organs, vascular extravasation to colonize those organs, and metastatic outgrowth to form new tumors. Several barriers exist at each step of the metastatic cascade that renders the overall process inefficient. This includes fluctuating nutrient compositions and metabolic demands that may diverge substantially from the primary tumor [96, 97]. Initial entry into the metastatic cascade may also largely depend on stochastic tumor:stroma interactions, tumor cell-of-origin, phylogenetic trajectories, and intra-tumor subclonal compositions [98]. Despite these complexities, a proclivity to assume migratory/invasive phenotypic states over others such as differentiation or proliferation appears to be a universal trait that enhances the probability that malignant cells will escape from the primary tumor [99]. Such proclivities are a matter of cell plasticity that is dictated by how pre-existing genetic drivers, tumor:stroma crosstalks, signaling pathways, metabolic dependencies, and epigenetic programs are collectively adapted to the external environment. For pancreatic cancer, the struggle to grow and survive within a dense desmoplastic stroma may represent a primal force that synergizes with chemotactic signals [100], symbiotic metabolite exchanges [68, 101], and premetastatic niches [102–104] to facilitate entry into the metastatic cascade.

As the primary tumor expands, the pressurized interior becomes progressively more famished and hypoxic. Although primary PDACs are presumably well-adapted, nutrient starvation, hypoxia, or infectious pathogens can each revert any sufficiently plastic eukaryotic cell into a primordial state of invasive migratory behavior known as the integrated stress response (ISR) [99, 105]. A general feature of the ISR is suppression of protein synthesis and proliferation (quiescence) so that limited supplies of ATP can be dedicated to generating the cytoskeletal forces required to migrate away from stressful environments. Indeed, hypoxia [106, 107], nutrient depletion [108, 109], and the ISR [108, 110] have all been shown to facilitate PDAC migration, invasion, and metastatic dissemination. Invasive migrations and tumor cell disseminations can be deployed using multiple modalities, including collective migration [111], epithelial-to-mesenchymal transition (EMT) [112–114], and partial/hybrid EMTs [115]. Irrespective of the mode, stress-induced migration of oxygen-or nutrient-seeking PDAC cells to well vascularized tumor periphery and beyond may represent an important driver of locally advanced disease that simultaneously facilitates entry into the metastatic cascade (Figs. 1 and 5).

Cells that pioneer better oxygenated or nutrient replete tissues would be expected to strengthen nutrient-fueled euchromatic aspects of the PDAC epigenome through metaboloepigenetic reinforcement. Conversely, stress-fueled heterochromatic aspects could weaken as malignant cells escape the stromal selection pressures imposed by symbiotic tumor:stroma crosstalk. Indeed, some modes of invasive migration (EMT) trigger global opening of chromatin [116] with increased euchromatic histone H3K4/K36 methylation [116, 117]. Those events can be coupled to partial loss of heterochromatic H3K9 methylation [116], although ISR-related transcription factors can partially counteract this by directly recruiting the H3K9 methylation machinery to preserve heterochromatin fidelity [118]. These data along with unique antioxidant adaptations that facilitate survival during metastatic dissemination [119, 120] introduce the possibility of adaptive metaboloepigenetic programs that are operative at each step of the metastatic cascade. Adaptive metaboloepigenetic programs during the early steps of metastasis await discovery and may require applying single cell analytic technologies to heterogenous samples with low tumor cellularity [111, 113]. The terminal and lethal step of the cascade is metastatic outgrowth. This step has proven amenable to bulk epigenomic analyses due to the high tumor cellularity [17, 121, 122] and monoclonality [123–126] of established metastatic tumors, as detailed below.

Integrated ChIP-, bisulfite-, and RNA-seq experiments examining distant metastatic PDACs from patient samples [127] and transgenic mice [128] both detected quantitative increases in histone acetylation across thousands of gene regulatory elements with massive up-regulation of overtly malignant, pro-metastatic mRNA transcripts from these regions [121, 127–129]. These events correspond to up-regulation of the TF FOXA1 [121, 127, 128], with experimental evidence that FOXA1 directly targets saturating enrichments of hyperacetylation to histones within oncogenic ATAC-seq accessible chromatin [128]. These include many sites previously indexed for accessibility during installation of the neoplastic epigenetic program [19, 56]. In patient samples, quantitative reductions in DNA and H3K9 methylation, spreading of histone acetylation, and evidence of prior genome instability were also detected within large heterochromatin domains, including those adjacent to amplified KRAS alleles [127] that predispose to metastatic dissemination. Thus, independent studies in mouse and man indicate that a selectable, pro-metastatic epigenetic program consisting of globally hyperacetylated and hypomethylated chromatin is acquired as a terminal event during malignant progression (Fig. 5). Although the bulk sequencing approaches employed cannot distinguish whether the quantitative increases are uniform or heterogenous across metastatic cell populations, the sheer magnitude implies near saturating enrichments of hyperacetylated histones. This is supported by our own unpublished observations that HDAC inhibitors only slightly increase global H3K27Ac ~ 5–10% beyond the high baseline amounts in distant metastatic PDACs. More importantly, these events specify a bona fide metastasis-intrinsic epigenetic program since they are not observed in primary tumors and are selectively required for widespread metastatic outgrowth [121, 127, 128, 130, 131].

Many of the metaboloepigenetic adaptations in primary PDACs are genetically encoded and heavily influenced by desmoplastic conditions including starvation and hypoxia. However, although PDAC genetic drivers are shared between metastatic PDACs and the primary tumor that seeds them [123, 124, 132], stromal desmoplasia is not well-developed in distant metastases [17, 121, 122, 133]. Thus, changing environmental conditions along the metastatic route may necessitate acquisition of metastasis-intrinsic metabolic (or crosstalk) adaptations that are not encoded by newly acquired genetic drivers [134]. Unique metabolic adaptions are in fact implied by PDAC morphologies and clinical behaviors. Primary PDACs are often populated by sparse numbers of atrophic-appearing tumor cells with modest amounts of cytoplasm, relatively small nuclei, and hyper-chromatic (closed) chromatin. In contrast, widely metastatic PDACs are often packed with large numbers of biosynthetic-appearing tumor cells with plump cytoplasm, enlarged nuclei, and hypochromatic (open) chromatin [17]. Unlike primary PDACs which grow silently as solitary lesions in the pancreas, distant metastases present suddenly and progress rapidly to diffusely fill the liver and lungs with hundreds to thousands of metastatic tumors [126, 135]. In transgenic mice, ablation of the desmoplastic stroma stimulates vascular neogenesis with similar PDAC morphologies and aggressive clinical behaviors as seen in human metastasis [136]. These collective observations suggest that PDACs may acquire nutrient-fueled biosynthetic adaptations that can cooperate with pre-existing genetic drivers and blunted anti-tumor immunity to accelerate disease progression during distant metastasis. Supporting this notion is emerging evidence for pro-metastatic adaptive metaboloepigenetic programs that are supported by nutrient-fueled positive feedback rather than new genetic mutations (Fig. 5) [121, 130, 131, 137, 138], as detailed below.

Consistent with clinical experience using F ¹⁸-labeled glucose PET scans to image tumors [139], distant metastases consume excessive amounts of glucose relative to primary PDACs and peritoneal implants [121]. A portion of the excess glucose is metabolized through remnants of the ascorbic acid biosynthetic pathway to manufacture 6-phosphogluconate (6PG) and related precursors outside the rate-limited oxidative pentose phosphate pathway (PPP) [130, 140]. 6PG is a substrate for the biosynthetic enzyme phosphogluconate dehydrogenase (PGD) [130] which is itself acetylated into a fully active conformation by components of glucose-driven acetyl CoA metabolism [141]. PGD converts 6PG substrates and NADP cofactors into ribulose-5-phosphate (R5P) and NADPH [127], respectively. Both products are pro-tumorigenic [127, 142], since R5P is an obligate precursor for de novo nucleotide biosynthesis while NADPH is required for de novo synthesis of fatty acids, cholesterol, and antioxidants. Thus, in the metastatic context, high glucose consumption can provide the post-translational modifications and metabolite substrates that support abnormally high PGD catalytic rates with accelerated downstream production of pro-tumorigenic anabolic metabolites.

High PGD catalysis consumes 6PG as it is produced. While this stimulates forward flux of glucose into 6PG biosynthesis pathways [130], the resulting (over)consumption depletes steady state concentrations [127]. This bears consequences for nutrient sensing transcription factors such as MondoA [143–146]. These nutritional sensors monitor intracellular concentrations of glucose-derived phosphorylated sugars as surrogates of glucose consumption. Rising glucose consumption rates normally increase steady state concentrations of phosphorylated sugar derivatives, which triggers MondoA nuclear translocation. PGD-driven overconsumption of 6PG instead creates a state of glucose pseudostarvation that retains MondoA in the cytosol (Fig. 5) [121, 137]. Failure of nuclear MondoA translocation prevents transcriptional activation of its target gene thioredoxin-interacting protein (TXNIP) [121], and this in turn prevents TXNIP-mediated endocytosis of glucose transporters off the cell surface (negative feedback) [147]. Thus, high PGD catalysis suppresses TXNIP-driven endocytosis, glucose transporters are retained at the cell surface, glucose consumption rates rise, the excess glucose temporarily replenishes 6PG substrates, additional rounds of PGD catalysis are stimulated, and the process initiates again (positive feedback, Fig. 5) [121, 137]. A selectable byproduct of this feedback is provision of excess glucose to produce the bulk acetyl CoA required to fuel ACLY-dependent hyperacetylation of metastatic chromatin (Fig. 5) [121, 131]. Thus, available data indicate that high PGD catalysis, excessive glucose import, production of bulk acetyl groups, and transcription factor targeting of HAT activities synergize to reprogram the PDAC epigenome into a globally hyperacetylated chromatin state that is permissive for up-regulation of the metastatic transcriptome (the pro-metastatic epigenetic program; Fig. 5). Interestingly, global loss of histone H3K9 methylation from large heterochromatin domains is also installed into the pro-metastatic epigenetic program [127], although the mechanisms and functional consequences await clarification.

6. Perspectives and future directions

PDAC research has blossomed into a paradigm for how tumor:stroma crosstalk, adaptive metabolism, and epigenetic reprogramming cooperate to drive neoplastic progression during changing microenvironmental selection pressures. We conclude this review with a concise series of perspectives, beginning with a specific hypothesis for each step of PDAC progression. Although currently speculative, each of these hypotheses are formulated upon existing data and will likely be either supported or refuted in timely follow-up studies. Those discussions are followed by several potentially fundamental yet largely unexplored questions that could shape this emerging field of study for years to come.

During the sustained adaptive response, epithelial progenitors and their differentiated progeny are stably programmed with a MAPK-inducible epigenetic memory of injury. Entire populations of acinar cells are therefore poised to enter ADM, replicate, and survive during otherwise lethal or senescent fibroinflammatory conditions. Although this synchronizes cells to repair tissue damage [18], it may also expand the pool of epithelial cells with actively replicating DNA during the repair process. A testable hypothesis is that the epigenetic memory of injury spatially embeds an intrinsically mutagenic “field effect” into pancreatic tissue that increases the risk of stochastic replication error(s) that serendipitously activate KRAS. Moreover, this would occur under the precise environmental and metabolic pressures that favor selection and clonal expansion of such cell(s). Experiments addressing such epigenetic field effects could help clarify precisely how chronic injury increases the risk of developing neoplasia.

In PanIN precursors, reduced mitochondrial aerobic respiration (“Warburg effect”) could cooperate with TP53-driven accumulation of succinate (or fumarate) to induce a state of pseudohypoxia [85, 148]. Since fibroinflammatory wound healing is itself a hypoxic condition, a testable hypothesis is that TP53-driven pseudohypoxia synergizes with KRAS-driven symbiotic crosstalk to amplify pro-tumorigenic stromal remodeling, acetyl CoA sparing, or immune evasion [149–151].

Primary PDACs may need to evolve multiple redundant or synergistic metaboloepigenetic adaptations to maintain the total or local concentrations of nutrient-derived epigenetic donor and inhibitor metabolites within the K_D range(s) of specific chromatin modifying enzymes [63, 152]. Although acetylated regulatory regions number in the thousands, each one only spans at most a few nucleosomes. Thus, a testable hypothesis is that combining a diverse repertoire of acetyl CoA-sparing metabolic adaptations ensures that any acetylation-based mode of targeted gene regulation can remain operative even within the famine of a primary tumor.

DNA and histone methylation at gene regulatory elements may operate by similar principles as acetylation [153]. However, unlike acetylation much of the malignant methylation program in primary PDAC is targeted to large heterochromatin domains spanning hundreds to thousands of kilobases [43, 72, 127]. Multiple redundant or synergistic adaptations that simultaneously target methylation writers and erasers may therefore be required to maintain large heterochromatin domains during stress. A testable hypothesis is that SAM-sparing adaptations are co-selected with those that inhibit demethylases to preserve malignant patterns of global methylation in starving primary tumors. As introduced in earlier sections, intuitive possibilities include combining rewired one carbon metabolism [72] with nucleotide salvaging [73] or scavenging [74] to (re)direct folate flux into SAM cycles or combining these maneuvers with import of extracellular cysteine precursors [75, 76] that decrease SAM consumption in the transsulfuration pathway. Another testable possibility is that glucose, amino acid, protein, and phospholipid scavenging adaptations [68–70, 154] could also help support SAM synthesis and DNA methylation [153, 155] by extracting methionine and methionine precursors from the stroma or by slowing SAM consumption during de novo phospholipid biosynthesis [156].

The limited available evidence suggests that metastatic PDACs can tap nutrient-replete reservoirs in metastatic niches to support rapid metastatic outgrowth. At least one way this can be achieved is through adaptive metaboloepigenetic programs that rely on positive feedback between hyper-active biosynthetic enzymes [130] and excessive nutrient import mechanisms [121]. A testable hypothesis to extend this concept further is that adaptive metaboloepigenetic programs are themselves co-selected with other complementary symbiotic crosstalk or metabolic adaptations [109, 138, 157, 158] that amplify nutrient-fueled biosynthesis [56, 65, 127], global epigenetic reprogramming [43, 50, 51, 56, 63, 121, 127, 128], malignant transcriptional outputs [121, 127–129], and systemic immunosuppression [104] beyond what can be achieved in a malnourished primary tumor. This would have the effect of synergistically accelerating widespread metastatic outgrowth as seen in patients.

Beyond the focused hypotheses introduced above, there are many other open ended questions that remain largely unexplored. Symbiotic crosstalk between stromal cues, tumor cells, and genetic drivers are crucial inputs that establish and maintain the PDAC epigenome. However, the precise mechanisms for how these inputs specifically target global changes to thousands of genomic locations awaits clarification. Additional signaling pathways, metabolic adaptations, chromatin modifications, and transcriptional regulators beyond those currently described undoubtedly participate in this process, leaving ample room for future progress. Intracellular trafficking and nuclear compartmentalization of nutrient-responsive transcription factors, metabolic enzymes, enzymatic cofactors, and metabolites is an emerging yet underappreciated concept [159] that could prove universally important for gene regulation and cancer. Metaboloepigenetic regulation of DNA repair [160, 161] could heavily impact PDAC genetic diversity, patterns of genome instability, and treatment responses. Yet this possibility remains largely uninvestigated to our knowledge. Stromal inflammation, immune suppression, and immune evasion are crucial for malignant progression and resistance to immune checkpoint inhibitors [162]. To our surprise, it is unknown if inflammatory cues or cytokine signaling pathways such as JAK-STAT [163] symbiotically influence the PDAC epigenome. Nor is it clear if the PDAC epigenome might encode transcriptional outputs that symbiotically signal back to immune cells.

Although a dense stroma is not well-developed in most widely metastatic PDACs [17, 121, 122], a delicate stromal matrix is nevertheless present [17, 164] that is capable of maturing from necrotic interiors [17] as distant metastases grow [165] and outstrip nutrient supplies. In addition, oligometastatic PDACs and peritoneal implants retain a similar dense stroma as primary PDACs [17, 121]. The degree to which the pro-metastatic epigenetic program remains dependent on inputs from symbiotic stromal cues verses transitioning to inputs from nutrient-replete reservoirs is therefore an open question that could vary according to metastatic site(s) and outgrowth kinetics. Another important question is whether symbiotic tumor:stroma metabolite exchanges might regulate PDAC [68, 74, 101, 166] and stromal cell[166, 167] epigenomes in a reciprocally cooperative manner within primary or metastatic tumors. Distinguishing metaboloepigenetic adaptations that are universally required throughout PDAC progression from others that uniquely support specific steps [168] is also sparsely investigated but could prove fundamental. This issue is especially pertinent to the metastatic cascade, including steps of tumor cell dissemination [100, 106, 112, 114, 117, 119], organotropic colonization [102, 103, 169], and metastatic outgrowth [104, 121, 123, 127, 128, 130, 131]. Are nutrient-fueled positive feedback loops that suppress negative feedback opposition (e.g., PGD-TXNIP [121, 137]) a common mode of epigenetic regulation in metastatic (or primary) PDACs? If not an isolated nuance, how then do such non-genetic adaptive metaboloepigenetic programs emerge in the first place? Are they operative in other aggressive cancers and can they co-opt other nutrients besides glucose, other metabolic enzymes besides PGD-TXNIP, and other epigenetic inputs besides hyperacetylation? Finally, characterizing symbiotic tumor:stroma crosstalk and adaptive metaboloepigenetic mechanisms conferring treatment resistance is arguably the most important translational path forward, especially since inhibitors are available against many metabolic and chromatin modifying enzymes.

As illustrated above, the spatial and temporal epigenetic landscape of pancreatic cancer progression holds strong potential for fundamental discoveries. However, the sheer number of open questions that remain for just this one facet of the disease should temper expectations regarding how close we truly are to understanding the full biological spectrum. The biology of pancreatic cancer has in fact proven far more difficult to understand than first anticipated from the early discoveries of a small and seemingly manageable set of recurrent genetic drivers. This is due in no small part to the unexpected number of additional complexities, several of which have been cited here. Common themes emerging from the literature indicate that navigating these complexities may require a convergence science approach that uncovers interdependencies between biological processes that have traditionally been investigated in isolation. To that end, discovery of mutually supportive positive feedback interactions between genetic drivers, tumor microenvironments, signal transduction pathways, metabolic adaptations, epigenetic programs, and systemic immune status may represent a high yield path toward a more holistic and actionable understanding of pancreatic cancer and other human diseases.