MYC Instructs and Maintains Pancreatic Adenocarcinoma Phenotype

Pancreatic ductal adenocarcinoma (PDAC) is a lethal cancer with a 5-year survival of less than 9% (1). Its dismal prognosis is in great part due to its typically late presentation and its profound resistance to all forms of conventional chemotherapy and radiotherapy. Studies of resected tumors suggest that malignant ductal adenocarcinoma evolves from pancreatic intraepithelial neoplasms (PanIN), indolent precursors that are histologically classified into four stages (PanIN1A, 1B, 2, and 3) based on increasing degrees of nuclear and architectural atypia (2–4). Frank PDAC is characterized by its striking fibroinflammatory stroma, absent from normal pancreas, whose attendant dense desmoplasia typically constitutes some 90% of tumor bulk (5, 6). This fibroinflammatory stroma is a product of complex interplay between tumor cells and adjacent mesenchymal, endothelial, inflammatory, and immune cells and is thought to contribute to PDAC therapeutic recalcitrance by impeding vascular perfusion and oxygenation (7, 8) and suppressing antitumor immunity (9).

At the molecular level, PDAC is associated with a core of recurring, presumed causal, oncogenic mutations. Of these, activating mutations in KRAS are present in more than 90% of both PanINs and PDAC and are thought to be founder drivers. Other frequently recurring signature mutations inactivate CDKN2A (90%–95%), TP53 (50%–84%), and SMAD4/DPC4 (49%–55%), while recent, more granular, analyses reveal a plethora of additional sporadic PDAC mutations encompassing great functional diversity (10–12). As KRAS mutation is the most commonly recurring oncogenic mutation, many of the canonical features of PDAC are attributed to precocious KRAS signaling. KRAS is a multifunctional communication node with diverse outputs that encompass many downstream intracellular effectors (e.g., MAPK, PI3K, p38, JAK–STAT, Hippo, and NFκB) and a host of extracellular signals and their receptors [e.g., WNT, Notch, SHH, CXCL1, 2, and 5, CXCL8 (IL8), IL6, GM-CSF, and IL17RA]. In this way, KRAS mutations can influence diverse aspects of PDAC pathology, including tumor initiation, maintenance, drug sensitivity, metabolism, macropinocytosis, metastasis, desmoplasia, inflammation, and immunity (13, 14). Oncogenic Kras has been used as the initiating driver in multiple pancreatic cancer mouse models (15–23). However, all concur that, without cooperating oncogenic mutations, KRAS oncogenic potential is very poor, stalling at indolent PanIN with little desmoplasia, inflammation, or lymphocytic suppression (15, 24). In addition to its role in pancreatic tumor initiation (15), oncogenic KRAS also plays a role in maintenance of established PDAC (15, 17, 24, 25), where it appears to provide an obligate necessary platform upon which subsequent cooperating lesions exert their oncogenic impacts.

The prototypical cooperating oncogenic partner of KRAS is MYC (26), a pleiotropic transcription factor whose putative role is to mediate expression of the thousands of genes that, together, coordinate the diverse genetic programs needed for somatic cell proliferation. In normal cells, MYC serves as a functionally nonredundant and essential transcriptional node and is tightly controlled by mitogen availability (27). In contrast, aberrantly persistent MYC activation is implicated in the majority of, perhaps all, cancers, although it seldom involves mutation of the MYC gene itself and is more commonly an indirect consequence of its relentless induction by “upstream” oncoproteins—notable examples being WNT, Notch, and RAS itself (28–34). During pancreatic organogenesis and early postnatal growth, MYC is expressed in a population of multipotent pancreatic progenitor cells and, like its upstream activators β-catenin (34–36) and Notch (33), is required for exocrine pancreas development, maturation, and neonatal growth (32, 37–40). MYC is not appreciably expressed in normal adult, uninjured, nonproliferating pancreas, but a majority of PDACs exhibit demonstrable expression of MYC and its canonical target gene signature. A complication, however, is that endogenous MYC is obligatorily expressed in all normal proliferating somatic cells, so its mere expression in proliferating cancer cells is not itself evidence of oncogenic causation. However, overt amplification of the MYC gene is frequent in especially aggressive tumors (20, 32, 41–43) and multiple in vivo mouse studies indicate that deregulated MYC is causally oncogenic in exocrine pancreas (reviewed in ref. 32). For example, classic transgenic overexpression of MYC targeted to mature acinar cells via an elastase promoter induces widespread acinar-to-ductal metaplasia (ADM; ref. 7), whereas pdx1 promoter–directed overexpression of MYC in pancreatic progenitor cells through organ development elicits rapid emergence of ductal PanIN (44). Conversely, pdx1-Cre–driven excision of endogenous Myc concurrently with Kras^G12D activation blocks transition of PanINs to PDAC. Studies also indicate a persistent maintenance role for MYC in established PDAC. MYC knockdown inhibits proliferation of human PDAC cell lines in vitro (45), and direct (44) or indirect (46) inactivation of MYC triggers macroscopic regression of established PDAC in mouse models in vivo (47), albeit with persistence of dormant tumor cells (44). Taken together, these data indicate a critical role for MYC in both the progression from PanIN to PDAC and in subsequent PDAC maintenance (33). However, the nature of this role remains obscure, especially because MYC has no obvious functional overlap with any of the canonical CDKN2A, TP53, and SMAD4 mutants that typically associate with KRAS mutation in PDAC.

Here we use a rapidly and reversibly switchable genetic mouse model whereby Myc may be engaged or disengaged, at will, in indolent Kras^G12D-induced PanINs to address key questions concerning the mechanistic role of MYC in PDAC. Does Myc activation drive immediate transition of Kras^G12D-driven indolent PanIN to PDAC or, instead, merely raise the likelihood that happenstance progression eventually occurs? What is the phenotype of the pancreatic tumors that Myc activation elicits, and do they resemble spontaneous PDACs? Finally, we address why Myc activation is required for the maintenance of Kras^G12D-driven PDAC and by what mechanism such tumors regress upon Myc deactivation.

Conditional expression of oncogenic Kras^G12D in pancreas progenitor cells drives protracted and sporadic outgrowth of indolent PanIN lesions (18). Rarely, these progress to overt adenocarcinoma through sporadic accretion of additional lesions. Because deregulated and/or overexpressed MYC is a frequent feature of advanced PDAC, we first asked whether MYC deregulation is sufficient to drive progression of indolent Kras^G12D-driven PanINs to PDAC using the well-characterized pdx1-Cre;LSL-Kras^G12D/+ model (K^pdx1) in which Cre recombinase expression, driven by the pdx/IPF1 (Pancreas/duodenum homeobox protein 1) promoter in embryonic pancreatic and duodenal progenitor cells from around E8 (19), triggers expression of a single allele of Kras^G12D driven from the endogenous Kras2 promoter. K^pdx1 mice homozygous for Rosa26-LSL-MycER^T2 (48) were generated. In the resulting pdx1-Cre;LSL-Kras^G12D;Rosa26LSL-MycER^T2 (KM^pdx1) mice, activation of Cre recombinase in pancreatic progenitor cells induces expression of both Kras^G12D, driven at physiologic levels from its endogenous promoter, and the conditionally regulatable MycER^T2 variant of Myc (48–51). Of special note, MycER^T2 in KM^pdx1 pancreatic epithelia is driven at quasi-physiologic levels from the promiscuously active but weak Rosa26 promoter (48), levels comparable to that of endogenous Myc in proliferating neonatal mouse pancreas cells (Supplementary Fig. S1A and S1B). Hence, both Kras^G12D and Myc in KM^pdx1 mice are expressed at physiologic levels throughout.

As previously described (18), by 15 weeks of age K^pdx1 control mice (i.e., Kras^G12D alone) each exhibited a small number of early-stage PanIN-1A and PanIN-1B lesions, with the vast majority of pancreatic ducts appearing histologically normal. These PanIN lesions were small, with little or no desmoplastic stroma (Fig. 1A). Activation of MycER^T2 alone for 3 weeks in the absence of Kras^G12D (M^pdx1 mice) had no significant histologic impact (Fig. 1A) save for a very modest increase in proliferative index (data not shown). In contrast, sustained activation of MycER^T2 for 3 weeks in KM^pdx1 mice (i.e., expression of Kras^G12D and Myc together) induced dramatic progression of invasive adenocarcinomas (Fig. 1A) that bore the stereotypical features of spontaneous mouse and human PDAC (Fig. 1B): highly proliferative nests of epithelial tumor cells of predominantly ductal phenotype embedded in extensive α-smooth muscle actin (α-SMA)–positive desmoplastic stroma. Tamoxifen alone had no discernible impact on pancreas architecture (Fig. 1B).

Unexpectedly, activation of MycER^T2 proved rapidly lethal in KM^pdx1 mice with only a few surviving to 3 weeks of sustained Myc activation. Lethality coincided with extensive MYC-induced overgrowth of KM^pdx1 duodenal epithelium, severely disrupted crypt–villus architecture, and emergence of multiple intestinal neoplasms. We attribute this to the known activity of the pdx1 promoter in relatively uncommitted E8 foregut progenitor cells that have yet to commit to pancreatic versus duodenal specification. To circumvent this problem, we confirmed our KM^pdx1 studies using the complementary p48-Cre;LSL-Kras^G12D; Rosa26LSL-MycER^T2 (KM^p48) mouse PDAC model, in which Cre expression is driven rather later and in more pancreas-specific progenitors by the p48/PTF1 promoter (18, 52). The histologic impacts of activating Myc in pancreata of KM^pdx1 and KM^p48 models were indistinguishable (Supplementary Fig. S2A). By 3 weeks of sustained Myc activation, both KM^pdx1 and KM^p48 pancreata exhibited an abundance of advanced PanIN3 lesions together with early invasive PDAC consisting of irregular malignant glands and solid clusters of neoplastic cells invading the stroma. Frank, highly invasive morphologically heterogenous pancreatic adenocarcinomas were also evident in some 30% of both KM^pdx1 and KM^p48 mice, rising to 90% of KM^p48 mice after 6 weeks of sustained Myc activation (Fig. 1; Supplementary Figs. S2A–S2C and S3A–S3D). No intestinal neoplasia was ever evident in KM^p48 mice, and any occasional lethality of KM^p48 mice after protracted Myc activation was attributed to the burden of grossly expanded, highly desmoplastic PDAC (Supplementary Fig. S2B and S2C). Accordingly, KM^p48 mice were used thereafter.

Although MycER^T2 is restricted solely to the PanIN epithelial cells in both KM^pdx1 and KM^p48 mice, Myc activation elicited widespread changes in stromal hematopoietic, endothelial, and mesenchymal cells. To address whether this MYC-induced stromal phenotype is an immediate, instructive consequence of MYC action or, instead, a delayed result of passive channeling by the host pancreatic tissue, we made use of the inherent rapidity and synchrony of MycER^T2 activation by tamoxifen in vivo (50, 51, 53, 54). Myc was activated in KM^p48 mice by systemic administration of tamoxifen for 0, 1, or 3 days, and pancreata were then harvested and analyzed. Within 24 hours of Myc activation, we observed a dramatic increase in proliferation (Ki-67) of the MycER^T2-expressing epithelial tumor compartment of all PanIN lesions accompanied by diverse stromal changes (Fig. 2): specifically, a marked influx of F4/80⁺ and CD206⁺ macrophages, Ly-6B.2⁺ neutrophils, and, to the periphery of each tumor, B220⁺ B lymphocytes, concurrent with loss of CD3⁺ T cells and induction of α-SMA in proximal stellate and fibroblastic cells, the latter indicative of myofibroblastic transdifferentiation. By 72 hours, progressive deposition of dense desmoplasia was evident together with loss of blood-vessel patency and upregulation of HIF1α, a marker of hypoxia, all presumed secondary consequences of prior stellate cell activation. Identical MYC-induced phenotypic changes, with identical kinetics, were observed in pancreata of the complementary KM^pdx1 switchable Myc model (Supplementary Fig. S4). Taken together, our data indicate that all the diverse stromal changes that MYC induces, all of them features of aggressive spontaneous PDAC seen in spontaneous mouse and human PDAC, are instructed by MYC from the outset.

As MycER^T2 is activated only in the pancreatic epithelial compartments of KM^pdx1 and KM^p48 mice, the immediate stromal changes that MYC elicits must be instructed by paracrine signals originating in those MycER^T2-expressing tumor epithelial cells. To identify such signals, we activated MycER^T2 in KM^p48 mice with a single systemic dose of tamoxifen, harvested pancreata 6 hours later, and then interrogated tissue extracts with growth factor/cytokine arrays (Fig. 3A and B) and isolated mRNA by RT-PCR (Fig. 3C). Various growth factors and regulators of inflammatory and immune cells were demonstrably induced. Some of these are already documented. For example, SHH was previously shown to be critical for PDAC desmoplastic deposition and maintenance (8). Similarly, macrophage influx into PanINs, clearly evident by 24 hours post Myc activation, was presaged by induction at 6 hours of the well-documented macrophage chemoattractant CCL9 (MIP1γ). MYC-induced influx of Ly-6B⁺ neutrophils was preceded at 6 hours by induction of CXCL5 (Fig. 3A), a well-characterized neutrophil attractant (55, 56) and one of several CXCR2 (IL8RB) receptor ligands. Myc activation also induced immediate upregulation of GAS6 (Fig. 3B), an AXL kinase ligand previously implicated in an unspecified role in PDAC-stromal signaling (57). GAS6 induction correlated with appearance of pAXL immunoreactivity in a subset of PanIN epithelial cells (Fig. 4A). Blockade of AXL kinase with a specific inhibitor, TP-0903 (ref. 58; Fig. 4B), profoundly inhibited MYC-driven neutrophil influx into tumors and B-cell influx into tumor peripheries and markedly suppressed induction of α-SMA in local stellate and fibroblastic cells (Fig. 4B). Comparable changes were observed following inhibition of GAS6 with low-dose warfarin (refs. 59, 60; Supplementary Fig. S5A). However, GAS6–AXL inhibition had only a modest inhibitory impact on macrophage influx and no discernible impact on MYC-induced depletion of intratumoral CD3⁺ T cells (Fig. 4B). Overall, blocking GAS6–AXL signaling profoundly retarded early MYC-driven tumor progression (reducing the prevalence of PanIN2 and 3 lesions in favor of PanIN1 and 2), reduced tumor burden, and inhibited proliferation of both tumor and stroma (Fig. 4C), in the main by suppressing proliferating neutrophils and fibroblastic cells (Supplementary Fig. S5B).

We recently showed in lung adenomas that MYC activation induces immediate T-cell expulsion via a mechanism requiring macrophage-specific expression of PD-L1 (also known as B7H1, encoded by CD274), a ligand for the PD-1 (CD279) receptor whose activation quells T lymphocyte immune functions. Myc activation rapidly induced PD-L1 expression on KM^p48 PanIN epithelial cells in vivo (Supplementary Fig. S6A), while systemic blockade of PD-L1 completely abrogated MYC-induced T-cell depletion in pancreatic tumors (Supplementary Fig. S6B and S6C), confirming a causal role for PD-L1 induction in driving exclusion of CD3⁺ T cells during the MYC-induced transition from PanIN to PDAC. However, PD-L1 blockade and consequent persistence of CD3⁺ T cells had no discernible inhibitory impact on the MYC-driven PDAC transition or subsequent tumor growth (Supplementary Fig. S6D). Myc activation also induced PD-L1 expression on KM^pdx1 PanIN epithelial cells in vivo (Supplementary Fig. S7A) and isolated KM^pdx1 PDAC cells in vitro (Supplementary Fig. S7B) while chromatin immunoprecipitation (ChIP) analysis located MycER^T2 protein on the CD274 gene (Supplementary Fig. S7C), all indicating that MYC-dependent PD-L1 induction in PanIN epithelium is cell-autonomous and that the CD274 gene is most likely a direct MYC target in pancreatic tumors.

We next addressed whether persistent Myc activity is required to maintain PDAC. MycER^T2 was activated for 3 weeks in 12-week-old KM^pdx1 or KM^p48 animals, inducing widespread multifocal aggressive PDAC in all mice. Subsequent MycER^T2 deactivation triggered ostensibly complete regression of all adenocarcinomas (Fig. 5A), culminating in pancreata largely indistinguishable from those in which Myc had never been activated. Even grossly expanded pancreatic tumors, resulting from 4 to 6 weeks of MycER^T2 sustained activation, underwent rapid regression following Myc deactivation (Supplementary Fig. S8A and S8B), although the gross architectures of the residual pancreata were often marred by extensive scarring and cyst-like dysmorphias (Supplementary Fig. S8C).

To confirm that MYC dependency is not solely an artefactual peculiarity of pancreatic adenocarcinomas driven directly by MYC, we also determined the consequence of inhibiting endogenous MYC in spontaneously evolved Kras^G12D-driven PDAC using the Omomyc dominant-negative MYC dimerization mutant to block endogenous MYC function (61). For this, we used pdx1-Cre;LSL-Kras^G12D;Trp53ER^TAM KI;TRE-OmoMyc;CMVrtTA (KPO) mice in which pdx1-Cre-driven activation of oncogenic Kras^G12D is combined with a background in which the endogenous p53 is replaced with the conditional Trp53ER^TAM allele (62). Because p53ER^TAM is functionally inactive in the absence of tamoxifen, KPO animals without tamoxifen are effectively p53^null, which greatly accelerates spontaneous transition of Kras^G12D-driven PanIN to PDAC (18, 19). KPO animals in addition harbor a doxycycline-inducible transgene encoding the dominant-negative MYC dimerization mutant Omomyc (61, 63). Accordingly, 10-week-old KPO mice harboring extensive, highly desmoplastic, spontaneous PDAC were treated with doxycycline for 4 weeks to induce Omomyc and inhibit endogenous MYC. As with deactivation of MycER^T2 in KM^pdx1 and KM^p48 mice, inhibition of endogenous MYC by Omomyc triggered quantitative regression of all spontaneous PDAC tumors, including their extensive attendant fibroinflammatory stroma (Fig. 5B).

To investigate the mechanics underlying the initiation of tumor regression post Myc deactivation, MycER^T2 was activated for 3 weeks in 12-week-old KM^p48 mice to generate early invasive PDAC and then acutely deactivated by tamoxifen withdrawal for 3 days (64). Myc deactivation triggered immediate tumor cell proliferative arrest (Fig. 6A) and initiated progressive reversal of stromal changes that Myc activity initially induced—specifically, rapid efflux of macrophages and neutrophils, rapid reinfiltration of CD3⁺ T cells concurrent with reduced expression of PD-L1, and deactivation of stromal stellate and fibroblastic cells (Fig. 6B and C). Of note, these stromal reversions all initiated after Myc deactivation with an immediacy comparable to their initial MYC-driven induction. Equivalent Myc deactivation–induced stromal reversal was also observed following Myc deactivation in KM^pdx1 animals (Supplementary Fig. S9).

One possible mechanism by which Myc deactivation might lead to overt loss of PDAC cells is by phenotypic reversal of neoplastic, predominantly ductal-like, dysplastic state back to their original acinar state and ductal organization. To investigate the potential for such ductal-to-acinar transdifferentiation, we used RNA-sequencing (RNA-seq) analysis to compare the transcriptional outputs of control pancreata isolated from 12-week-old KM^p48 mice in which Myc was never activated, versus KM^p48 pancreata in which Myc had been activated continuously for 2 weeks, versus pancreata in which Myc was activated for 2 weeks and then deactivated for 3 days (Supplementary Fig. S10). This clearly showed a marked induction of ductal markers (Egfr, Krt20, Tgfa), together with corresponding repression of acinar markers (Amy2b, Bhlha15, Ptf1a) in response to Myc activation and reversal back to acinar status upon subsequent Myc deactivation (Supplementary Fig. S10A). This was verified with KM^pdx1 cells in vitro using the amylase and cytokeratin as immunocytochemical markers of, respectively, acinar (65) and ductal (66) states (Supplementary Fig. S10B). Hence, Myc reversibly toggles PDAC epithelial cells between ductal (Myc ON) and acinar (Myc OFF) states.

Nonetheless, tumor regression induced by Myc deactivation cannot be simply ascribed to mere reversal of the events originally triggered by Myc activation. For example, Myc deactivation not only terminated further desmoplastic deposition but triggered its rapid degradation, quickly restoring blood vessel patency and local tissue oxygenation (Fig. 6C; Supplementary Fig. S9). Furthermore, Myc deactivation not only induced proliferative arrest and redifferentiation of pancreatic tumor cells but also triggered an abrupt wave of tumor cell death (Fig. 6A; Supplementary Fig. S9). To investigate whether such cell death was a consequence of the rapid PD-L1 downregulation–mediated influx of T cells that Myc deactivation induces, we systemically ablated CD4⁺ and CD8⁺ T cells. This efficiently abrogated all detectable circulating and splenic T cells (data not shown) and completely blocked repopulation of pancreatic tumors by CD3⁺ T cells, yet had no inhibitory impact on tumor cell death (Fig. 7A). In contrast, B-cell behavior was unexpected. Whereas CD20⁺B220⁺ B cells rapidly migrated to the PanIN peripheries upon Myc activation (Fig. 2), Myc deactivation triggered their immediate migration into each tumor mass (Fig. 6B). This influx of B cells was accompanied by concurrent influx of natural killer (NK) p46⁺ NK-like cells, and systemic depletion of either B cells or NKp46⁺ NK-like cells profoundly suppressed the abrupt wave of tumor cell death induced by Myc deactivation (Fig. 7B and C). Unexpectedly, B-cell ablation also blocked the Myc deactivation–induced influx of NKp46⁺ NK-like cells (Fig. 7D), suggesting that B cells play a critical role in initiating PDAC regression by facilitating recruitment of innate NK cells.

The low, quasi-physiologic levels of MYC used in our studies indicates that the mere deregulation of Myc, without any overt amplification or overexpression, is alone sufficient to drive progression of indolent Kras^G12D-driven PanIN lesions to aggressive PDAC. There is, apparently, no need for Myc to be overexpressed for any of its oncogenic activities to be manifest. An important implication of this is that MYC levels, specifically elevated MYC levels, cannot be used as surrogate indicators of a causal role for MYC in PDAC (or, indeed, any other cancer type). Rather, it is the aberrant persistence of MYC, not its elevation, that is its underlying oncogenic agency.

The tumor–stromal phenotype of the pancreatic adenocarcinoma lesions induced by activation of Myc in the PanIN epithelial compartment is marked by several notable features— its complexity, its immediacy, its tissue specificity, and, upon subsequent Myc deactivation, its rapid and complete reversibility. The complexity of the MYC-induced tumor phenotype is evident in the diversity of stromal cell types involved and in the complicated and highly organized choreography in which they participate. Myc activation within the PanIN epithelium triggers rapid influx of F4/80⁺CD206⁺ macrophages, Ly-6B⁺ neutrophils and/or granulocytic myeloid-derived suppressor cells (MDSC), and B220⁺ B cells; rapid efflux of CD3⁺ T cells; myofibrillar activation of neighboring fibroblastic cells, and consequent desmoplastic deposition, blood vessel compression, vascular occlusion, and hypoxia. Many of these stromal features are clearly evident within 24 hours of Myc activation in the epithelial compartment and are presaged by release of a distinct set of instructional paracrine signals, evident as early as 6 hours post Myc activation. Hence, although the adoption of the canonical phenotype of established PDAC tumors takes some longer time to establish, essentially all of the instructive signals and processes that culminate in the signature phenotype of established PDAC appear to be engaged almost immediately following Myc activation. These include potent chemoattraction of macrophages (67), Ly-6B⁺ neutrophils and/or granulocytic MDSCs, and B220⁺ B cells (68–71); expulsion of CD3⁺ T cells (67); activation and proliferation of fibroblastic cells; and extensive desmoplastic deposition with concomitant blood vessel compression, vascular involution, and hypoxia (5, 72, 73). The profound tissue specificity of this pancreatic MYC phenotype is evident from comparison with the adenocarcinomas induced in the lung by the exact same combination of Kras^G12D and switchable Myc (ref. 74; Supplementary Table S1). As in pancreas, the MYC-induced adenocarcinoma phenotype in the lung arises immediately following Myc activation, and involves rapid onset of tumor cell proliferation and invasion, influx of macrophages, and T-cell exclusion. However, although some stromagenic mechanisms appear broadly similar in both organs, for example, CCL9-dependent MYC-driven influx of macrophages, the MYC-driven adenocarcinoma transitions in pancreas and lung differ dramatically in several key respects, both with regard to the phenotypes induced and the MYC-induced signals that instruct them. Thus, the dramatic and rapid MYC-driven influx of granulocytic and B lymphoid cells into nascent pancreatic tumors is entirely absent in the lung. Indeed, Myc activation has precisely opposite impacts on B cells in lung versus pancreas, driving their expulsion in the former and influx to the tumor periphery in the latter. Such differences are reflected in the underlying paracrine signals responsible in each case. IL23 is responsible for MYC-driven exclusion of B (and T and NK) cells in the lung (74), yet it is not detectably induced in the pancreas: Conversely, the key GAS6–pAXL pathway required for neutrophil/granulocyte and B-cell chemoattraction in the pancreas appears absent from MYC-driven lung adenocarcinomas. Indeed, even when a common mechanism is ostensibly shared across both tissues, for example, the obligate role of PD-L1 in MYC-driven exclusion of CD3⁺ T cells, the mechanism of pathway deployment is quite different in the two tissues. In the pancreas, PD-L1 is expressed cell-autonomously in MYC-expressing PanIN epithelial cells, apparently through direct MYC-dependent transcription. In contrast, PD-L1 is never appreciably induced in MYC-expressing lung adenoma epithelial cells themselves and is, instead, ported into incipient tumors on incoming alveolar macrophages. Perhaps the most overtly different consequences of MYC action between the two tissues is the rapid activation and proliferation of local fibroblastic and stellate cells in the pancreas, triggering progressive deposition of desmoplastic stroma, leading to rapid blood vessel compression and hypoxia. In contrast, MYC activation in the lung elicits no appreciable activation of local fibroblastic cells or desmoplasia and instead engages a potent proangiogenic switch, fueled by macrophage-derived VEGF, which rapidly restores normoxia to the otherwise indolent, hypoxic adenomas (74).

A most remarkable feature of the complex, multicell lineage tumor–stroma phenotypes that MYC instructs in the pancreas and lung is that each closely matches the distinctive tumor–stromal phenotype of spontaneous adenocarcinomas arising from each of those same tissues. The inescapable inference is that adenocarcinoma phenotypes are principally specified by their host tissue of origin rather than by the peculiar ensemble of oncogenic mutations that drive them. Thus, RAS and MYC serve as common, tumor-agnostic downstream conduits of the multitude of different oncogenic mutations that lie upstream and, rather than specifying each tumor phenotype, Kras^G12D and Myc instead unlock a preconfigured tissue-resident program. The likely source of this tissue-resident program is suggested by the remarkable phenotypic overlap between PDAC (spontaneous and MYC-driven) and that of regenerating pancreas following injury. Both exhibit marked acinar–ductal metaplasia, epithelial proliferation, rapid infiltration of macrophages and neutrophils, exclusion of T cells, activation of stellate cells, and an extreme fibroinflammatory reaction (75–78), attributes so overtly similar as to frequently confound differential diagnosis between pancreatitis and early PDAC. To summarize: oncogenic MYC “hacks” the endogenous pancreatic regenerative program (79).

Our data show that Myc activation in KM^p48 or KM^pdx1 PanIN epithelial cells triggers release of a diverse set of paracrine signals that quickly and synchronously engage highly reproducible and coordinated changes across all stromal compartments. Macrophage influx, initiated within only a few hours of Myc activation, is long associated with especially poor PDAC prognosis. Pancreatic macrophages, which have diverse origins and functions (80, 81), are, in PDAC, implicated in promoting epithelial-to-mesenchymal transition and tumor cell migration, suppressing immunity, driving myofibroblastic transdifferentiation of stellate cells and cancer-associated fibroblasts (CAF) through release of multiple secondary signals including PDGF and TGFβ, and deploying various proteases that actively remodel stroma (81–85). Heavy infiltration with Ly-6B⁺ neutrophils and their G-MDSC kin is also associated with the most undifferentiated, aggressive, and metastatic forms of PDAC with the poorest prognosis (10, 55, 86, 87) and is thought to be a consequence of their secretion of multiple stromal remodeling proteases and diverse secondary signaling molecules including IL8, GM-CSF, CCL3, CXCL10, IL2, TNFα, and oncostatin-M (55), and through suppression of T-cell surveillance (88). CXCL5 is a well-characterized neutrophil attractant in PDAC (84, 86, 88) that is induced within 6 hours of MYC activation. However, prompted by the previously reported role of GAS6–AXL signaling in neutrophil dynamics (89, 90) and the rapid MYC-dependent induction of GAS6, a member of the Protein S family of vitamin K–dependent plasma glycoproteins, we uncovered an additional signaling pathway necessary for MYC-driven neutrophil influx. GAS6 is a ligand for the TAM receptor tyrosine kinase family (TYRO3, AXL, and MER) and its induction by MYC temporally correlated with appearance of pAXL in a proportion of PanIN epithelial cells. Blockade of either GAS6 with low-dose warfarin (59, 90) or AXL kinase with the selective AXL kinase inhibitor TP-0903 (58, 91, 92) blocked MYC-induced neutrophil influx and markedly suppressed proliferation in both tumor cell and stromal compartments, most especially stellate cells and CAFs, while exerting only a modest inhibitory impact on macrophage infiltration (Fig. 4). Unexpectedly, GAS6–AXL blockade also completely blocked the rapid influx of B220⁺ CD20⁺ B cells to the peritumoral stroma that MYC triggers. A pivotal role for B lymphocytes has lately emerged in PDAC progression and aggression, most notably innate B1b cells, hitherto implicated in physiologic processes such as microbial defense and removal of debris after tissue damage (93, 94). B cells are attracted to incipient PDAC via CXCL13 and, through their release of IL35, are thought to suppress T-cell immunity (95–97), promote PDAC cell proliferation and invasiveness, and inhibit apoptosis (98, 99). Diverse studies have demonstrated GAS6–AXL pathway activation in both tumor and stromal compartments of various solid tumors, including PDAC (57), and shown that GAS6–AXL pathway inhibition impedes growth of many cancers including PDAC (100) and promotes responses to chemotherapy (60, 91, 92, 100, 101). However, although GAS–AXL signaling is broadly implicated in regulating aspects of immunity (89, 92, 102–105) and tumor–stromal signaling (57), there is no previously described involvement of GAS6–AXL signaling in either neutrophil or B-cell dynamics in PDAC. Currently, it remains unclear what relative contribution of inhibition each of the individual target cell populations of GAS6–AXL blockade—neutrophils, B lymphocytes, or stellate cells—makes to MYC-driven PDAC progression, or whether each of these targets is affected independently or some are contingent upon another. Although our current analysis of signaling molecules is less comprehensive than recent proteome-based studies (57), our kinetic data provide unique causal information as to both the sources and sequence of signals and responses that assemble the PDAC tumor–stromal complex.

Concomitantly with the influx of macrophages, neutrophils, and B cells, Myc activation induced immediate expulsion of CD3⁺ T lymphocytes from PanINs, and this coincided temporally with induction of surface PD-L1 expression on PanIN epithelial cells in vivo. Moreover, ChIP analysis located MYC on the CD274 (PD-L1) promoter, implicating CD274 as a direct MYC target in the pancreas, as suggested in some other tissues (106). Although T-cell expulsion is not currently established as a mechanism of PD-L1 action, the very same physical location dependency of T cells on absence of PD-L1 has also been observed in lung adenocarcinoma progression. Moreover, the rapidity, immediacy, and reversibility of PD-L1 expression and T-cell exclusion suggest that there is a tight causal relationship between the two. Nonetheless, although systemic blockade of PD-L1 abrogated MYC-induced T-cell efflux, indicating that PD-L1 is causally required for MYC-induced T-cell efflux, the consequent persistence of T cells in the tumor masses had no discernible inhibitory impact on either PDAC maintenance or subsequent growth. These data are consistent with previous studies in both pancreas (107) and lung (74) that indicate that enforced persistence of T cells in tumors is, alone, insufficient to engage significant antitumor immune responses.

Activation (α-SMA) and mitogenesis of local stellate and fibroblastic cells was also evident rapidly following Myc activation in PanIN epithelial cells and was rapidly succeeded by progressive desmoplastic deposition, vascular occlusion and hypoxia. A more detailed future analysis will be needed to ascertain whether physical proximity of mesenchymal cells to PDAC epithelial cells (108) influences their MYC-dependent activation. In spontaneous PDAC, activated stellate cells and CAFs are known to be the principal agents of desmoplastic deposition (5, 57, 109–111). Although the role of PDAC desmoplasia in vascular compression and hypoxia is generally accepted (5, 6, 8, 72, 73, 109, 112, 113), its influence on disease progression and response to therapy remains contentious (8, 114–117).

The final notable feature of the MYC-driven PDAC phenotype is its rapid and complete reversibility upon subsequent Myc deactivation. Nor is such reversibility peculiar only to PDACs specifically driven by oncogenic MYC, because inhibition of endogenous MYC triggers identical regression in spontaneous Kras^G12D-induced (and p53-deficient) mouse PDAC. Such rapid and efficient reversal is notwithstanding the fact that Myc is deactivated only in the tumor epithelial compartment. It nonetheless has an immediate impact on the diverse cell types that comprise the dense PDAC desmoplastic stroma, unraveling the complex tumor maintenance signals shuttling between them. Indeed, we see no evidence whatsoever of any aspect of PDAC pathobiology that is self-sustaining and independent of requirement for continuous MYC activity within the driving epithelial tumor compartment. Just like MYC-driven progression, onset of regression is immediate upon Myc deactivation, with overt evidence of phenotypic reversal evident within only a few hours of Myc deactivation: Efflux of macrophages and neutrophils, as well as downregulation of PD-L1 and reentry of CD3⁺ T cells, is accompanied by rapid deactivation of local mesenchymal stellate and fibroblastic cells, degradation of desmoplasia and blood vessel dilation, and tissue reoxygenation. Sustained blockade of MYC function induces the quantitative and meticulous dismantling of even the most extensive tumor masses, restoring normal exocrine pancreas microarchitecture, albeit in a macroscopic organ badly disfigured by scars and cysts. In the main, PDAC disassembly reverses the protumorigenic processes that MYC initially induced, but with some important exceptions. Most notably, Myc deactivation triggers a sharp wave of tumor cell death, which is dependent upon comigration of peritumoral B cells and NK-like cells into the tumor mass. We observed a similar requirement for NK-cell activity in regression of lung adenocarcinomas following Myc deactivation in vivo (74), suggesting that innate immune lymphocytes play a general role in the regression process, as recently suggested (74, 118).

As with MYC-driven PDAC genesis, the rapidity, reproducibility, efficiency, complexity, and tight choreography of PDAC regression all suggest it is a manifestation of some evolved physiologic process. Clues to the origin of this process lie in multiple studies of the postregenerative resolution phase of pancreas injury by which normal pancreas mass, architecture, and function are restored. Such injury resolution involves ductal-to-acinar transdifferentiation and the involution of fibrosis, and requires both the rapid exclusion of F4/80⁺ macrophages and neutrophils and the influx of lymphocytes (75, 119, 120). It seems plausible that, just as activating Myc hacks into the endogenous pancreas regenerative program, inhibiting MYC (or upstream oncogenic signals) hacks into a physiologic injury-resolution program that evolved to terminate postinjury regeneration, prune excess cells, and reinstate normal tissue architecture. Indeed, it is possible that enforced engagement of this same injury-resolution program is the principal mechanism by which targeted inhibition of oncogenic signaling exerts its curative impact in cancers.

LSL-Kras^G12D, p48^cre/+, Pdx1-Cre, Trp53ER^TAM, TRE-Omomyc, and Rosa26-lsl-MER^T2 mice have all been described previously (18, 48, 61, 62, 121, 122). CMVrtTA [Tg(rtTAhCMV)4Bjd/J] mice were purchased from Jackson Laboratories. Mice were maintained on a 12-hour light/dark cycle with continuous access to food and water and in compliance with protocols approved by the UK Home Office guidelines under a project license (to G.I. Evan) at the University of Cambridge (Cambridge, United Kingdom). Deregulated Myc activity was engaged in pancreatic epithelia of KM^pdx1 and KM^p48 mice by daily intraperitoneal administration of tamoxifen (Sigma-Aldrich, TS648) dissolved in sunflower oil at a dose of 1 mg/20 g body mass; sunflower oil carrier was administered to control mice. In vivo, tamoxifen is metabolized by the liver into its active ligand 4-Hydroxytamoxifen (4-OHT). For long-term treatment, mice were placed on tamoxifen diet (Harlan Laboratories UK, TAM400 diet) as described previously (64); control mice were maintained on regular diet. Omomyc expression was systemically induced in KPO (pdx1-Cre;LSL-Kras^G12D; Trp53ER^TAM KI;TRE-Omomyc;CMVrtTA) mice by addition of doxycycline (2 mg/mL plus 5% sucrose) to their drinking water versus 5% sucrose for control mice.

Mice were euthanized by cervical dislocation and cardiac perfused with PBS followed by 10% neutral-buffered formalin (Sigma-Aldrich, 501320). Pancreata were removed, fixed overnight in 10% neutral-buffered formalin, stored in 70% ethanol and processed for paraffin embedding. Tissue sections (4 μm) were stained with hematoxylin and eosin (H&E) using standard reagents and protocols or with Gomori Trichrome Stain (Polysciences, 24205), according to the manufacturer's instructions. For frozen sections, pancreata were embedded in OCT (VWR Chemicals, 361603E), frozen on dry ice, and stored at −80°C.

One hundred microliters of fluorescein-conjugated Lycopersicon esculentum lectin (FL-1171, Vector Laboratories) was administered intravenously three minutes before euthanasia. Pancreata were harvested, and deparaffinized tissue sections were counterstained with Hoechst dye and imaged by fluorescence microscopy.

For IHC or immunofluorescence (IF) analysis, paraffin-embedded sections were deparaffinized, rehydrated, and either boiled in 10 mmol/L citrate buffer (pH 6.0) or treated with 20 μg/mL Proteinase K to retrieve antigens, depending on the primary antibody used. For IF studies, OCT-embedded sections were air-dried and fixed for 30 minutes in 1% paraformaldehyde. Primary antibodies used were as follows: rabbit monoclonal anti–Ki-67 (clone SP6) (Lab Vision, Thermo Fisher Scientific, 12603707), rat anti–Ki-67 (Clone SolA15, Thermo Fisher Scientific, 15237437), rat monoclonal anti-CD45 (clone 30-F11, BD Pharmingen, 553076), rat monoclonal anti-neutrophils (Clone 7/4, Cedarlane, CL8993AP), rat monoclonal F4/80 (clone Cl:A3-1, Bio-Rad, MCA497R); goat polyclonal anti-MMR/CD206 (R&D Systems, AF2535), goat polyclonal anti-NKp46/NCR1 (R&D Systems, AF2225), rabbit monoclonal anti-CD3 (clone SP7, Thermo Fisher Scientific, RM-9107-RQ), rabbit polyclonal anti-CD274/PD-L1 (Abcam, ab58810); rat monoclonal anti-CD45R/B220 (clone RA3-6B2, Thermo Fisher Scientific, MA1-70098), rabbit polyclonal anti-αSMA (Abcam, ab5694), rabbit polyclonal CD31 (Abcam, 28364), rabbit polyclonal anti-HIF1α (Abcam, ab82832), rabbit polyclonal anti-pAXL (Y779, AF2228, R&D Systems), mouse monoclonal anti Pan-Keratin (clone C11, NEB, 4545S). Primary antibodies were incubated with sections overnight at 4°C except for anti-CD3, which was applied for 20 minutes at room temperature. For IHC analysis, primary antibodies were detected using Vectastain Elite ABC HRP Kits (Vector Laboratories: peroxidase rabbit IgG PK-6101, peroxidase rat IgG PK-6104, peroxidase goat IgG PK-6105) and DAB substrate (Vector Laboratories, SK-4100); slides were then counterstained with hematoxylin solution (Sigma-Aldrich, GHS232). For IF analysis, primary antibodies were visualized using species-appropriate cross-adsorbed secondary antibody Alexa Fluor 488 or 555 conjugates (Thermo Fisher Scientific); slides were counterstained with Hoechst (Sigma-Aldrich, B2883) and mounted in Prolong Gold Antifade Mountant (Thermo Fisher Scientific; P36934). TUNEL staining for apoptosis was performed using the Apoptag Fluorescein in situ Apoptosis Detection Kit (Millipore; S7110) according to the manufacturer's instructions. Images were collected with a Zeiss Axio Imager M2 microscope equipped with Axiovision Rel 4.8 software.

Pancreatic tissues were ground into powder in liquid nitrogen and proteins extracted using standard protocols. Total protein lysates were electrophoresed on an SDS-PAGE gel and blotted onto Immobilon-P membrane (Millipore). Membranes were blocked with 5% nonfat milk and primary antibodies incubated overnight at 4°C. Secondary antibodies were applied for 1 hour followed by chemiluminescence visualization. The following primary antibodies were used: MYC (ab32072; Abcam, 1:2,000 dilution) and β-actin (sc-69879, Santa-Cruz Biotechnology, 1:5,000 dilution). HRP-conjugated secondary antibodies: goat anti-rabbit (sc-2301; Santa-Cruz Biotechnology 1:7,500 dilution) and goat anti-mouse (A4416, Sigma-Aldrich, 1:7,500 dilution).

Pancreata were collected at appropriate time points and snap-frozen in liquid nitrogen. Whole pancreas protein extracts were isolated and incubated with mouse inflammation/cytokine antibody arrays (40 targets, Abcam, ab133999; 97 targets, Abcam ab169820), according to the manufacturer's instructions. Signal intensity was determined using ImageJ software. Differential expression of cytokine RNA was assessed by qRT-PCR. For this, total pancreas RNA was isolated using a Qiagen RNeasy Plus Isolation Kit followed by cDNA synthesis (High Capacity cDNA RT Kit, Applied Biosystems, 4374966). qRT-PCR was performed using TaqMan Universal Master Mix II (Thermo Fisher Scientific, 4440038), according to the manufacturer's protocol. Primers used were: Cxcl5 (Thermo Fisher Scientific, Mm00441260_m1), Gas6 (Thermo Fisher Scientific, Mm00441242_m1), Ccl9 (Thermo Fisher Scientific, Mm00436451_g1), Ccl2 (Thermo Fisher Scientific, Mm00490378_m1), Tbp (Thermo Fisher Scientific, Mm00446973_m1), Shh (Thermo Fisher Scientific, Mm00436528_m1). Samples were analyzed in triplicate on an Eppendorf Mastercycler Realpex 2 with accompanying software. Tbp was used as an internal amplification control.

Total RNA was isolated from KM^p48 mouse pancreas using a PureLink RNA Mini Kit (Thermo Fisher Scientific) and PureLink DNAse set (Thermo Fisher Scientific). The quality, quantity, and integrity of RNA were assessed by NanoDrop1000 spectrophotometer and 2100 Bioanalyzer (Agilent Technologies). RNA libraries were generated using TruSeq Stranded mRNA Library Prep Kit following the manufacturer's instructions (Illumina). RNA Libraries were run on the Ilumina NextSeq 500 using the 75 cycle high-output kit (single-end sequencing). The quality of the sequencing data was analyzed by the bioinformatics team at the Cambridge Genomic Services (CGS) at the University of Cambridge (https://www.cgs.path.cam.ac.uk) using FastQC v0.11.4. In summary, reads were trimmed using Trim-Galore v0.4.1 and those less than 20 bases long were discarded. Reads were mapped using STAR v2.5.2a. The Ensembl Mus. Musculus GRCm38 (release 95) reference genome was used with annotated transcripts from the Ensembl Mus. musculus GRC38.95.gtf file. The number of reads mapping to genomic features was calculated using HTSeq v0.6.0. Differential Gene Expression Analysis using the counted reads employed the R package edgeR v3.16.5 and the paired design model as suggested in the edgeR user's guide. RNA-seq data have been deposited in the ArrayExpress database at EMBL-EBI (www.ebi.ac.uk/arrayexpress) under accession number E-MTAB-8467.

Pancreatic tumors were isolated as described previously (8), but with modifications. Briefly, 12-week-old KM^pdx1 mice were placed on tamoxifen diet for 3 weeks. Pancreatic adenocarcinomas were then harvested, minced, and digested at 37°C in Hank's Balanced Salt Solution containing 2 mg/mL type V collagenase (Sigma-Aldrich, C9263) for 30 minutes under constant agitation. Digested tumors were centrifuged and the pellet further digested with 1× Trypsin-EDTA (0.05% trypsin, 0.02% EDTA; Sigma-Aldrich, 59418C) for 5 minutes. Proteases were inactivated by the addition of DMEM (Thermo Fisher Scientific, 41966-029) containing 10% FBS (Thermo Fisher Scientific, 10270106). Established cell lines were propagated in DMEM containing 10% FBS and maintained with 100 nmol/L of 4-OHT (Sigma-Aldrich H7904); they were routinely evaluated for Mycoplasma contamination using the VenorGeM Classic Kit (Mycoplasma Detection Kit for Conventional PCR) according to the manufacturer's protocol (Minerva Biolabs, 11-1050). Cells stored at early passage were thawed and cultured for up to 7 weeks in total. For immunofluorescence analysis, cells were fixed in 4% paraformaldehyde and permeabilized with 0.4% Triton X-100. Primary antibodies used were: mouse monoclonal anti-Amylase G-10 (Clone G-10, Santa Cruz Biotechnology, sc-46657), mouse monoclonal anti–pan-cytokeratin (clone AE1/AE3, Abcam, ab27988). Primary antibodies were incubated with fixed cells overnight at 4°C, and were visualized using species-appropriate cross-adsorbed secondary antibody Alexa Fluor 488 or 555 conjugates (Thermo Fisher Scientific) as described above. For gene expression analysis, total RNA was isolated, reverse transcribed, and amplified by RT-PCR as described above. Primers used were for Cd274/pdl1 (Thermo Fisher Scientific, Mm00452054_m1), Tbp (Thermo Fisher Scientific, Mm00446973_m1). Tbp was used as normalization control.

ChIP was performed as described in ref. 123, with modifications. Briefly, 10⁷ cells cultured with 100 nmol/L 4-OHT or ethanol control for 6 or 24 hours were incubated with 1% formaldehyde (Sigma-Aldrich, F8775) at room temperature for 10 minutes. Fixation was quenched by addition of glycine to a final concentration of 125 mmol/L for 5 minutes at room temperature. Cells were then washed with ice-cold PBS, pelleted, and resuspended in lysis buffer (10 mmol/L Tris-HCl pH 8.0, 85 mmol/L KCl, 0.5% NP40, 1 mmol/L DTT) with protease inhibitors (Sigma-Aldrich, 11836153001). Cross-linked lysates were sonicated to shear DNA to an average fragment size of 100 to 500 bp. The following antibodies were then used for immunoprecipitation: rabbit polyclonal anti–c-MYC (Cell Signaling Technology, 9402), rabbit polyclonal anti-ERα (clone HC20; Santa Cruz Biotechnology, sc-543), normal rabbit IgG as a background control (Cell Signaling Technology, 2729). Chromatin regions enriched by ChIP were then identified by qPCR in triplicate using SYBR Green Reaction Mix (Thermo Fisher Scientific, 4385616) according to the manufacturer's protocol. Pdl1 primers used for amplification were 5′-GTTTCACAGACAGCGGAGGT-3′ (forward) and 5′-CTTTAAAGTGCCCTGCAAGC-3′ (reverse) with NCBI reference sequence NM_021893.3 and described in ref. 124. Calculations of average cycle threshold (C_t) and SD for triplicate reactions were performed and each DNA fraction normalized to its input to account for chromatin sample preparation differences.

For PD-L1 blocking studies, KM^p48 mice were injected intraperitoneally with 160 μg of LEAF-purified anti-mouse CD274/PD-L1 (clone 10F.9G2, BioLegend, 124309) or rat IgG2b isotype control (clone RTK4530, BioLegend, 400644) every 2 days for 2 weeks, starting 1 day prior to Myc activation. Mice were euthanized after 2 weeks and pancreata harvested for histology. For CD4/CD8 blocking studies, KM^p48 mice were administered tamoxifen diet for 3 weeks. Commencing four days prior to returning the mice to normal (tamoxifen-free) diet they were injected twice intraperitoneally with 200 μg of rat anti-mouse CD4 (Clone GK1.5, BioXCell, BE003-1) four days apart, and four times with 200 μg rat anti-mouse CD8 (Clone 2.43, BioXCell, BE0061) every other day. Control mice were injected with an equivalent amount of rat IgG2b isotype control (Clone LTF-2, BioXCell, BE0090). Mice were euthanized 3 days post Myc inactivation and pancreata harvested for histologic analysis. CD20-expressing B cells were blocked with 250 μg of Ultra-LEAF purified anti-mouse CD20 antibody (clone SA271G2, BioLegend, 152104) or Ultra-LEAF purified Rat IgG2b isotype control (clone RTK4530, BioLegend, 400644). Two days before cessation of tamoxifen administration, mice were injected intravenously with 250 μg of antibody or isotype control. They were then euthanized 3 days post Myc inactivation and pancreata isolated for histologic analysis. To block NKp46⁺ cells, KM^p48 mice were given tamoxifen diet for 3 weeks. Starting four days prior to returning the mice to normal (tamoxifen-free) diet, mice were intravenously injected every other day (total of 4 injections) with 120 μL of Ultra-LEAF Purified anti-Asialo-GM1 Antibody (Clone Poly21460, BioLegend, 14602) or 120 μg of rabbit IgG isotope control (Biotechne, AB-105-C). Mice were euthanized one day after the last antibody injection coincident with third day post tamoxifen diet cessation.

For warfarin blockade of GAS6, KM^p48 mice were randomized to receive either normal drinking water or water supplemented with 1 mg/L warfarin (Sigma-Aldrich, A2250) starting 1 day prior to Myc activation. Warfarin-containing water was replenished every 2 days. Mice were euthanized 3 days following Myc activation and pancreata harvested for histology. For AXL receptor tyrosine kinase inhibition, the specific inhibitor TP-0903.tartrate (Tolero Pharmaceuticals, Inc.) was dissolved in 5% (w/v) Vitamin E TPGS + 1% (v/v) Tween 80 in water and 55 mg/kg administered by oral gavage to KM^p48 mice every 2 days, starting 2 hours before initial Myc activation. Mice were euthanized 5 days later and pancreata harvested for histologic analysis.

KM^p48 mice were anesthetized by continuous infusion of 3% isofluorane using a veterinary anesthesia system (VetEquip, California). The ventral, lateral, and dorsal abdominal surfaces were shaved and mice then injected intraperitoneally with 2 mL of sterile saline. Tumors were imaged using a Vevo 2100 imaging system (Visual Sonics). Serial images were collected at 0.2-mm intervals throughout the entire tumor. Vevo2100 v1.6.0 software was used to assess tumor size at each time point from 3-D scans, identifying in the Z plane and calculating maximal width and length.

For quantification of tumor burden, H&E sections were scanned with an Aperio AT2 microscope (Leica Biosystems) at 20× magnification (resolution 0.5 μm per pixel) and analyzed with Aperio Software. Statistical significance was assessed by Student t test, with the mean values and the SEM calculated for each group using Prism GraphPad software. Kaplan–Meier survival curves were calculated using the survival time for each mouse from each group, with the log-rank test used to calculate significant differences between the groups using Prism GraphPad software. Student t test was employed in statistical analyses of ChIP and RT-PCR. IHC and IF staining were quantified using Fiji open source software; statistical significance was determined by Student t test using Prism GraphPad software; data were represented as box-and-whisker plots, which show upper extreme, upper quartile, median, lower quartile, lower extreme. P values (ns, nonsignificant; *, P < 0.05; **, P < 0.01; ***, P < 0.001; ****, P < 0.0001) for each group (the number of animals in each cohort varied between 4 and 7 for each category).

L. Soucek is CEO at Peptomyc S.L. and has ownership interest (including patents) in the same. G.I. Evan is an advisory board member at AstraZeneca. No potential conflicts of interest were disclosed by the other authors.

Conception and design: N.M. Sodir, L. Soucek, G.I. Evan

Development of methodology: N.M. Sodir, R.M. Kortlever, G.I. Evan

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): N.M. Sodir, R.M. Kortlever, T. Campos, L. Pellegrinet, S. Kupczak, P. Anastasiou, L. Brown Swigart

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): N.M. Sodir, R.M. Kortlever, V.J.A. Barthet, T. Campos, L. Pellegrinet, M.J. Arends, G.I. Evan

Writing, review, and/or revision of the manuscript: N.M. Sodir, R.M. Kortlever, T. Campos, L. Soucek, M.J. Arends, T.D. Littlewood, G.I. Evan

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): N.M. Sodir, P. Anastasiou, L. Brown Swigart, G.I. Evan

Study supervision: L. Brown Swigart, T.D. Littlewood, G.I. Evan

Other (genotyping of animals, IF staining, assisting with other experiments): P. Anastasiou

Other (interpretation of the histopathologic data): M.J. Arends

We are indebted to the members of the Evan laboratory, Drs. Daniel Von Hoff (TGEN), Ronald Evans and Michael Downes (Salk Institute), Aarthi Gopinathan and Christine Feig (Cambridge Cancer Institute), Daniel Masso (Vall d'Hebron Institute), and Fernanda Kyle Cezar (King's College London) for invaluable discussion and advice. We also thank Alessandra Perfetto and Daniel Masso for assistance with animals, Stephanie Whike and Michaela Griffin for assistance with histology, and the Pre-clinical Genome Editing Core at the Cambridge Cancer Research Institute (CI) for assistance with ultrasound scans. We are grateful to Tolero Pharmaceuticals, Inc. for their generous gift of TP-0903. This study was supported by program grants Cancer Research UK C4750/A12077 and C4750/A19013A, the European Research Council (294851), and a Stand Up To Cancer-Cancer Research UK-Lustgarten Foundation Pancreatic Cancer Dream Team Research Grant (grant number: SU2C-AACR-DT20-16; all to G.I. Evan). Stand Up To Cancer (SU2C) is a division of the Entertainment Industry Foundation and the research grant is administered by the American Association for Cancer Research, the Scientific Partner of SU2C.