Repression of the Type I Interferon Pathway Underlies MYC- and KRAS-Dependent Evasion of NK and B Cells in Pancreatic Ductal Adenocarcinoma

Cancers of the pancreas are projected to become the second most frequent cause of cancer-related mortality worldwide by 2030 (1). Ductal adenocarcinoma comprises up to 95% of pancreatic cancers, and neuroendocrine tumors (PNET) and other exocrine tumors account for remainder (https://www.pancreaticcancer.org.uk/; https://seer.cancer.gov/statistics/). The 5-year survival rate for pancreatic ductal adenocarcinoma (PDAC) is extremely low at 3% to 5%, and although that for PNET is considerably higher (30%–50%), neither cancer responds effectively to current treatment modalities (2). There is therefore a pressing need to further our understanding of the underlying biology of these cancers.

Activating mutations in KRAS occur in up to 93% of PDAC, while the majority of KRAS wild-type (WT) cases bear alternative genetic alterations predicted to result in ectopic RAS–ERK pathway activity, implying a critical requirement for this pathway in PDAC (3, 4). Experiments in genetically altered mice have however demonstrated that KRAS activation alone is insufficient for PDAC development: Cre-dependent activation of the endogenously expressed Lsl-Kras^G12D allele (5) alone gives rise predominantly to low-grade pancreatic intraepithelial neoplasms (PanIN), a subset of which may spontaneously progress to PDAC through the acquisition of additional mutations (6–8). Recent genomic characterization of cell lines generated from such spontaneously progressing tumors revealed increased KRAS^G12D allelic dosage as one of the most frequently recurring genetic alterations associated with progression (9), suggesting that the volume of signaling flux through the RAS pathway is rate-limiting for PDAC progression, as was recently shown for KRAS-driven lung cancer (10–13). Interestingly, cell lines that lacked increased KRAS^G12D allelic dosage contained amplifications of alternative oncogenes that encode proteins that may serve as potentially rate-limiting RAS effectors, including MYC, YAP1, and NF-κB (9, 14). MYC in particular is widely reported to be overexpressed in PDAC, with a propensity for low-level amplification suggesting that subtle changes in MYC expression may suffice to affect PDAC development (15, 16). Accordingly, expression of murine Myc from the Rosa26 locus was recently shown to accelerate progression to PDAC; however, the use of murine Myc cDNA in the transgene precluded a precise determination of the level of MYC deregulation required for tumor progression (17). Conversely, deletion/depletion of endogenous Myc profoundly delayed progression of PDAC driven by KRAS^G12D and loss of p53 (18, 19), and RNAi-mediated depletion of MYC in human PDAC cells suppressed proliferation in vitro and xenograft tumorigenesis in vivo (20), consistent with the hypothesis that MYC is a critical rate-limiting effector of KRAS in this cancer.

Here, we directly investigated the level of MYC deregulation required for pancreatic tumor development using conditional expression of human MYC from the Rosa26 locus at levels that approximate expression of endogenous Myc. We show that modestly deregulated expression of MYC alone suffices to drive PNET development and dramatically accelerates progression to metastatic PDAC when combined with KRAS activation. In PDAC, the combination of MYC and KRAS^G12D suppresses tumor infiltration of effector immune populations, as recently reported (21). Mechanistically, we reveal that MYC and KRAS^G12D cooperatively regulate gene expression, with particular convergence on repression of the type I interferon (IFN) pathway. We show that targeted suppression of the MYC–MIZ1 transcriptional repressor complex (22) restores IFN-related gene expression and consequent B cell– and natural killer (NK) cell–mediated immune surveillance. Our data reveal a mechanism of immune evasion driven by two of the most commonly dysregulated oncogenes in human cancer.

To define the level of MYC deregulation required for pancreatic tumor development, we used Rosa26^DM-lsl-MYC mice wherein the human MYC cDNA, preceded by a floxed translational stop cassette, was inserted into the murine Rosa26 locus by homologous recombination, as described previously (23). Use of the human cDNA facilitated the distinction between endogenously expressed murine Myc and Rosa26-driven MYC. RT-PCR demonstrated Cre-dependent expression of human MYC in Rosa26^DM-lsl-MYC/+- and Rosa26^{DM-lsl-MYC/lsl-MYC}-derived mouse embryonic fibroblasts (MEF) within 24 hours of infection with Adeno-Cre. Expression of one copy of human MYC mRNA was comparable with murine Myc, whereas two copies yielded modestly supraphysiologic expression (Fig. 1A). Consistent with previous reports that MYC regulates its own transcription (24), Rosa26^MYC/MYC MEFs exhibited reduced expression of endogenous Myc and were modestly sensitized to apoptosis upon serum withdrawal (Fig. 1B). At the protein level, activation of Rosa26^DM-lsl-MYC/+ drove modestly higher expression of MYC compared with WT MEFs. Notably, Cre-dependent activation of endogenously expressed Lsl-Kras^G12D also drove higher expression of MYC protein, comparable to that arising from Rosa26^MYC/+, and expression was higher again upon combined activation of both alleles (Fig. 1C).

RAS pathway activity elevates MYC expression in pancreatic cancer, primarily via suppression of MYC protein turnover, and MYC was previously shown using RNAi to be required for proliferation of PDAC cells in culture (20, 25, 26). Use of the PICKLES database of CRISPR-mediated essentiality screening (27) confirmed the requirement for MYC to sustain the fitness of all PDAC lines examined, including several lines that do not require KRAS (Supplementary Fig. S1A). We asked whether MYC expressed from the Rosa26 locus is sufficient for tumor formation. We interbred Rosa26^DM-lsl-MYC mice (M) with pancreas-specific Pdx1-Cre mice (C), with and without the Lsl-Kras^G12D allele (K), and aged mice until humane clinical endpoints were reached. Pdx1-Cre–positive mice carrying one (MC) or two (M²C) copies of Rosa26^DM-lsl-MYC developed pancreatic tumors requiring euthanasia at a median age of 297 or 180 days, respectively. The combination of Lsl-Kras^G12D and one copy of Rosa26^DM-lsl-MYC (KMC) yielded a dramatically accelerated tumor phenotype requiring euthanasia at a median age of just 50 days, as compared with KC mice, which had a median survival >350 days (Fig. 2A–C). In contrast with KC mice, which developed mostly ductal epithelial PanIN lesions with infrequent progression to PDAC, as reported previously (6), MC and M²C pancreata showed no evidence of ductal epithelial phenotypes and instead presented with PNETs characterized by densely populated cells that stained strongly for the neuroendocrine marker synaptophysin and faintly positive for cytokeratin (Fig. 2D; Supplementary Fig. S1B). This PNET phenotype was largely masked upon inclusion of the Lsl-Kras^G12D allele: Rapidly arising tumors in KMC mice displayed a predominantly ductal adenocarcinoma phenotype (PDAC), complete with characteristic desmoplastic stroma, whereas <5% of tissue area exhibited PNET features (Supplementary Fig. S1B and S1C). PNET regions in KMC tumors expressed KRAS^G12D (Supplementary Fig. S1D), indicating that the persistence of this phenotype does not arise from failure to activate the Lsl-Kras^G12D allele in a subset of pancreatic tumor–initiating cells.

The Pdx1 promoter is active from embryonic day 8.5 (28), raising the possibility that blockade of pancreatic progenitor cell differentiation might explain the MYC-dependent neuroendocrine and accelerated PDAC phenotypes. To address this, we substituted the constitutively active Pdx1-Cre allele with a Pdx1-Cre-ER allele (C^ER) which expresses an inactive Cre–estrogen receptor ligand–binding domain fusion protein that can be activated by the synthetic ligand tamoxifen (29). KMC^ER mice were additionally interbred with mice carrying a Hprt-Lsl-IRFP Cre-reporter allele (30), to facilitate imaging of tumor populations (Fig. 2E). Cre-ER was transiently activated in mice ages 5 to 6 weeks by 3 days of tamoxifen injection, and mice were aged to clinical endpoint. As was found using the constitutively active Cre, all M²C^ER mice developed neuroendocrine tumors, reaching median endpoint at 275 days from the date of induction, although the majority of MC^ER mice failed to develop symptomatic disease within 400 days. KMC^ER mice all developed PDAC, reaching median endpoint at 128 days post-induction (Fig. 2F). Periodic sampling of KMC^ER pancreatic tissue following Cre-ER activation showed temporal progression of incipient tumors from PanIN-1 through PDAC (Fig. 2G). Furthermore, the IRFP allele revealed metastases in 6 of 10 mice, of which 4 had liver metastases and 4 had diaphragm metastases, including 2 mice with both (Fig. 2H). As observed with KMC tumors, KMC^ER tumors contained regions (∼15% of tumor area) of PNET (Supplementary Fig. S1C). All liver metastases histologically resembled the PNET phenotype and all stained strongly for synaptophysin. In contrast, diaphragm metastases all histologically resembled PDAC, but also contained discrete regions where cells stained strongly for either cytokeratin or synaptophysin, suggestive of phenotypic plasticity in the metastatic population (Fig. 2H). Accordingly, synaptophysin–cytokeratin double-positive cells could be identified in the ductal epithelium of primary tumors (Supplementary Fig. S1E), as previously reported using the embryonically active Pdx1-Cre allele (17). The accelerated tumor phenotypes observed upon MYC deregulation can thus arise in fully developed adult tissues.

Given the modest expression of Rosa26-driven MYC in MEFs, we asked whether autochthonously expressed Myc contributes meaningfully to the total pool of MYC protein in the KMC model. Note that whereas KRAS activation does not affect transcription from the Rosa26 locus, which we previously showed is refractory to growth factor signaling (31), post-translation regulation of MYC protein by RAS pathway activation should affect murine and human MYC protein equally (32). Accordingly, deletion of floxed murine Myc in MEFs reduced the total pool of MYC protein by approximately 50%, despite concurrent activation of Rosa26^DM-lsl-MYC and Lsl-Kras^G12D alleles (Fig. 3A). KMC mice interbred with Myc^fl/fl mice were aged until humane endpoints were reached. KMC mice heterozygous for floxed Myc showed significantly increased survival, and survival was increased further in homozygous Myc^fl/fl mice (Fig. 3B and C). Nevertheless, all mice developed PDAC. Examination of murine Myc by ISH revealed a mosaic pattern of continued expression of murine Myc in much of the ductal epithelium of all KMC-Myc^fl/fl tumors examined, indicating escape from Cre-mediated deletion and suggesting a selective pressure to retain expression of endogenous Myc despite the presence of Rosa26-driven MYC (Supplementary Fig. S2A and S2B). This MYC dose dependence was also evident in human PDAC cell lines, wherein depletion of MYC suppressed cell proliferation (Supplementary Fig. S2C), consistent with previous reports (20). These data strongly suggest that the level of MYC expression is rate-limiting for pancreatic tumor progression and, in KMC mice, both Rosa26-driven MYC and endogenously expressed Myc functionally contribute to the total pool of MYC protein.

To gain insight into the functional roles of MYC in PDAC, we performed bulk-tumor RNA sequencing (RNA-seq) on six KC, six KMC, and four KMC-Myc^fl/+ end-stage tumors. Note that homozygous KMC-Myc^fl/fl mice were omitted from RNA-seq analysis given that the failure of Cre recombinase to efficiently delete both copies of endogenous Myc would likely confound interpretation. MetaCore GeneGo analysis revealed a pronounced reduction in immune cell–related gene expression in KMC relative to KC tumors (Supplementary Fig. S3). Specifically, we found reduced expression of genes encoding T-cell markers, including CD3, CD4, and CD8 (Supplementary Fig. S4A); B-cell immunoglobulin genes, including those encoding multiple constant heavy and light chains along with joining regions; and NK-cell genes, including natural killer triggering receptor (Nktr) and killer-type lectin receptor genes (Fig. 3D and E). IHC for markers of B (CD45R) and NK (NKp46) cells confirmed their reduced presence in KMC PDAC (Fig. 3F and G). Reduction of total MYC by deletion of endogenous Myc reversed the exclusion of B and NK cells but did not significantly influence T-cell infiltration (Fig. 3F and G; Supplementary Fig. S4B and S4C). RNA-seq analysis showed increased expression of multiple killer-type lectin receptor genes, indicative of NK-cell activation and memory (33), upon reduction of MYC in KMC tumors (Fig. 3H). Taken together with Figs. 1 and 2, these data show that subtle differences in MYC expression profoundly modulate long-term immune cell infiltration in PDAC, extending a recent report of acute immune modulation by MYC using a tamoxifen-dependent variant allele, Rosa26-Lsl-MycER^T2 (21).

MYC was previously shown to modulate the immune landscape in KRAS-driven lung adenocarcinoma via CCL9-dependent recruitment of protumor macrophages and IL23-dependent exclusion of T, B, and NK cells (34). Expression of Ccl9 was not significantly altered in our RNA-seq datasets and we found no difference in the number of macrophages present in end-stage KMC versus KMC-Myc^fl/fl tumors (Supplementary Fig. S4D and S4E). Functional IL23 comprises a heterodimer of p19 and p40 subunits, encoded by IL23a and IL12b, respectively (35). Although Il23a expression was higher in KMC versus KC tumors and significantly reduced in KMC-Myc^fl/+ tumors, the gene encoding its dimerization partner Il12b showed opposite regulation (Supplementary Fig. S4F), strongly suggesting that IL23 does not account for the immunosuppressive phenotype driven by MYC in PDAC. Expression of Gas6, recently reported to be induced upon acute MYC activation in PanIN lesions (21), was unchanged in end-stage tumors (Supplementary Fig. S4G). We therefore sought an alternative explanation for MYC-driven immunosuppression.

Type I IFN signaling via JAK–STAT was among the most strongly suppressed pathways (ranked 6th) in KMC tumors relative to KC tumors (Supplementary Fig. S3). Taking advantage of our ability to acutely induce expression of floxed MYC and KRAS^G12D in otherwise WT primary cells, we asked which pathways were regulated upon acute activation of MYC and/or KRAS in MEFs. Early-passage MEFs carrying Rosa26-Lsl-Myc^DM, Lsl-Kras^G12D, or both, were infected with Adeno-Cre overnight in full growth medium and their gene expression was compared with that of similarly infected WT littermate MEFs. Consistent with positive regulation of MYC by activated KRAS (Fig. 1), we found a pronounced overlap in the transcriptomic impact of activation of either oncogene alone (Supplementary Fig. S5A–S5C; Supplementary Tables S1–S3). Similar to our comparison between KMC and KC tumors, GeneGo analysis identified type I IFN signaling via JAK–STAT as the most significantly downregulated pathway upon combined activation of MYC and KRAS alleles (Supplementary Fig. S5D). Interrogation of the data at the single-gene level showed reduced expression of multiple IFN-induced genes, multiple IFN regulatory factors, and genes encoding STAT1 and STAT2, which combine with IRF9 to form the ISGF3 complex (36). Although several of these effects were evident with activation of either MYC or KRAS^G12D alone, regulation was most pronounced upon combined oncogene activation (Fig. 4A; Supplementary Fig. S5D).

We next compared the transcriptomic impact of acute depletion of MYC or KRAS from KMC-derived tumor cell lines. As was observed in MEFs, we found a pronounced overlap in significantly regulated genes altered in murine PDAC cells upon depletion of either Myc or Kras (Fig. 4B; Supplementary Fig. S5E). Notably, the type I IFN pathway was again among the topmost regulated pathways in both instances (Supplementary Fig. S5F), with increased expression of the ISGF3 complex subunits, multiple Irfs, IFN receptor genes, and IFN-induced genes detected upon depletion of either oncogene. These effects were typically stronger upon Kras depletion, and some Irf genes (e.g., Irf5) were regulated by KRAS but not by MYC in this analysis, perhaps reflecting the limits of transcript detection at the sequencing read-depth used (Fig. 4C). To extend these data to the human setting, we treated PDAC cell lines with the MEK1/2 inhibitor trametinib, which acutely reduced expression of MYC downstream of KRAS (Fig. 4D). In both AsPC-1 and DAN-G cells, trametinib increased expression of IRF5, IRF7, IFNB1, and a representative IFN-inducible gene, IFI44 (Fig. 4E). We conclude from these experiments that MYC and KRAS cooperate to suppress the type I IFN pathway.

In many instances, MYC-dependent transcriptional repression involves binding to the MYC-interacting zinc finger protein MIZ1, encoded by ZBTB17 (18, 37). We therefore investigated the acute impact of Miz1 depletion in KMC PDAC cells in vitro. Transcriptomic analysis revealed that depletion of Miz1 upregulated approximately 60% of the same genes upregulated upon depletion of MYC from KMC cells (Fig. 5A; Supplementary Table S4). Using q-PCR, we determined that depletion of Miz1 or Myc specifically increased expression of Irf5, Irf7, Ifnb1, and Ifi44 (Figs. 5B; Supplementary Fig. S6A), as was found upon trametinib treatment of human PDAC cells. Chromatin immunoprecipitation (ChIP) analyses showed MYC binding to the promoters of STAT1, STAT2, IRF5, and IRF7 in DAN-G cells (Fig. 5C). ChIP–re-ChIP using a MIZ1 antibody showed efficient recovery of the same IRF and STAT promoter fragments initially precipitated a MYC but absent from control IgG precipitates, strongly suggesting that MYC and MIZ1 form repressive complexes on the STAT1, STAT2, IRF5, and IRF7 promoters (Fig. 5D; Supplementary Fig. S6B), as previously reported for the Stat1 promoter in murine KPC PDAC cells (18). To investigate the functional role of MIZ1 in vivo, we used Miz1^ΔPOZ mice (38) in which the sequence encoding the MYC-interacting POZ domain of MIZ1 is flanked by LoxP sites (Fig. 5E). Efficient deletion of the MIZ1 POZ domain in KMC-Miz1^fl/fl mice was verified by ISH and significantly extended the life span of tumor-bearing mice (Fig. 5F; Supplementary Fig. S6C). PDAC subtype analysis performed as per Bailey and colleagues (3) revealed that deletion of either Myc or Miz1 significantly increased the immunogenic subtype signature but did not significantly alter any of the other subtypes (Supplementary Fig. S6D). As was observed upon deletion of endogenous Myc, deletion of Miz1 restored tumor infiltration of NK and B cells (Fig. 5G). Antibody-mediated blockade of the type I IFN receptor IFNAR suppressed NK- and B-cell infiltration and negated the survival benefit of Miz1 deletion in tumor-bearing KMC mice (Fig. 5H–J), implicating IFN signaling in lymphocyte recruitment and in the survival benefit observed upon Miz1 deletion.

Type I IFNs were recently shown to induce expression of the B-cell chemokine CXCL13 (39). RNA-seq analysis showed extremely high expression of Cxcl13 in KMC-Miz1^fl/fl tumors, as compared with KMC tumors, which express very little Cxcl13 (Fig. 6A). ISH analysis of KMC-Miz1^fl/fl tumors showed Cxcl13 expression was restricted to a subset of F4/80-positive macrophages (Fig. 6B). Notably, only macrophages in close proximity to areas of ductal tumor epithelium stained positive for Ccxl13 mRNA, suggesting paracrine signaling from tumor epithelium to adjacent macrophages. A similar, albeit weaker, pattern of Cxcl13 expression in F4/80-positive macrophages adjacent to ductal epithelium was observed in KMC-Myc^fl/fl tumors but was absent from KMC tumors (Fig. 6A; Supplementary Fig. S6E). We used exogenous mouse IFNβ1 treatment of bone marrow–derived macrophages (BMDM) to confirm that IFN can stimulate Cxcl13 expression in macrophages (Fig. 6C). Next, we cultured macrophages with conditioned medium harvested from KMC tumor cells treated with nontargeting or MYC-depleting siRNA and found that only media from MYC-depleted tumor cells could stimulate Cxcl13 expression. Importantly, pretreatment of macrophages with an IFNAR1-blocking antibody completely abrogated Cxcl13 induction (Fig. 6D), confirming that type I IFN released from MYC-depleted tumor cells drives CXCL13 production in macrophages. Antibody-mediated depletion of CXCL13 from KMC-Miz1^fl/fl mice suppressed infiltration of both B and NK cells and negated the survival benefit of Miz1 deletion (Fig. 6E–G), whereas isotype control antibody had no effect (Supplementary Fig. S6F). Depletion of NK cells likewise negated the survival benefit, strongly suggesting that they actively restrain the tumor phenotype (Fig. 6H; Supplementary Fig. S6G). Accordingly, coculture of KMC tumor cells with in vitro–activated splenocytes enriched for NK cells showed that KMC tumor cells are efficiently killed by NK cells, equivalent to the established NK target cell line YAC1 (40), in contrast with primary fibroblasts that are largely resistant to NK-mediated killing (Fig. 6I). Taken together, these data delineate a pathway by which KRAS and MYC cooperate to evade NK-mediated antitumor activity in PDAC through suppression of the type I IFN pathway (Fig. 6J).

The evasion of antitumor immunity has emerged in recent years as a major hallmark of cancer that presents genuine therapeutic opportunities through the targeted suppression of immune-evasion mechanisms, often with long-term patient benefit. Most efforts to date have focused on mechanistic evasion of T lymphocyte–mediated tumor immunity, with targeted blockade of CTLA4 and PD-1/PD-L1 establishing a paradigm for successful therapeutic exploitation of immuno-oncology. Here, we present evidence of a mechanism of evasion of innate tumor immunity achieved through cooperation of the frequently dysregulated oncoproteins MYC and KRAS via suppression of the type I IFN pathway.

We identified the type I IFN pathway as a major target of oncogene-mediated suppression through the unbiased analysis of multiple RNA-seq datasets: comparing end-stage pancreatic tumors driven by KRAS^G12D alone with those driven by the combination of KRAS^G12D and MYC; comparing KMC-derived tumor cells depleted of Myc, Miz1, or Kras with mock-depleted parental cells; and comparing MEFs upon acute activation of Lsl-Kras^G12D, Rosa26^DM-lsl-MYC, or both, with WT littermate controls. In all such comparisons, the type I IFN pathway ranked among the pathways most significantly regulated. Although our mechanistic analysis here was performed exclusively in the context of pancreatic cancer, the conservation of this regulation in MEFs suggests that suppression of type I IFNs may be a general feature of cancers wherein MYC and KRAS are dysregulated. For instance, a strong negative correlation between tumors with an IRF8/9 gene signature and those with a MAX gene signature (MAX being the obligate heterodimerizing partner of MYC) was recently reported in mesothelioma (41). Moreover, the type I IFN gene cluster is syntenic with CDKN2A/B on chromosome 9 and is frequently co-deleted with the latter in a spectrum of human cancers: Across multiple cancers, patients with co-deletion show significantly reduced overall survival compared with those who retain the IFN gene cluster, providing further evidence of an antitumor role for this pathway and a selective pressure for cancers to reduce its activity (42).

Mechanistically, we present evidence of repressive MYC–MIZ1 complexes binding to the promoters of STAT1, STAT2, IRF5, and IRF7. These data are complemented by a previous report of MYC–MIZ1 binding to the Stat1 promoter in KPC (Kras^G12D;Trp53^R172H) murine pancreatic tumor cells (18) and MYC repression of STAT2 in breast cancer cells (43). Recent work has shown that basal expression of many IFN-stimulated genes is regulated by STAT2–IRF9 complexes independently of IFNAR signaling. Upon IFN-dependent activation of IFNAR, signaling through JAK family kinases leads to STAT1 binding to STAT2–IRF9 to form the ISGF3 complex (36), driving expression of, among other genes, IRF7, which in turn is strictly required for IFNα/β production (44). Note that other STAT family members may have very different effects, as recently evidenced by STAT6 driving expression of Myc during PDAC progression in response to T_h2-derived ILs (45). MYC–MIZ1 thus attenuates the type I IFN cascade at multiple points that would limit both basal and IFN-stimulated gene expression. The effects of KRAS on this pathway appear to be largely a consequence of MYC stabilization, at least in PDAC cells (25); however, signaling cross-talk between the RAS and JAK–STAT pathways likely contributes to some of the MYC-independent effects observed, or indeed to the stronger effects of KRAS modulation in these systems (46). Similarly, MIZ1 likely has additional roles in PDAC beyond its function as a MYC corepressor, evidenced by divergent regulation of some 40% of MYC-regulated genes upon acute depletion of MIZ1 in KMC cells. The fact that Miz1 could be efficiently deleted in ductal tumor epithelium, unlike endogenous Myc, argues that its role in PDAC is more limited and likely explains the stronger impact of the floxed Miz1 genotype upon Cxcl13 induction in vivo, as compared with the floxed Myc allele.

We demonstrate that derepression of the type I IFN pathway in PDAC tumor cells stimulates production in nearby macrophages of the canonical B-cell chemokine CXCL13, resulting in tumor infiltration by B and NK cells. Although the recruitment of NK cells may well be an indirect consequence of B-cell recruitment, as suggested by recent data (21), evidence is emerging of a direct role for CXCL13 in NK-cell recruitment, and a subset of NK cells express the CXCL13 receptor CXCR5 (47). Depletion of CXCL13 or blockade of the type I IFN receptor IFNAR each suppressed B- and NK-cell infiltration and reversed the survival benefit of Miz1 deletion in KMC tumors. Using coculture experiments, we showed that NK cells can efficiently kill KMC tumor cells, and that their depletion also negated the survival benefit observed upon Miz1 deletion. As is the case with cytotoxic T cells, NK cells express a variety of activating and repressive surface markers that are amenable to extrinsic modulation. As such, targeted activation of NK cells is emerging as an attractive therapeutic option, particularly in tumors with low mutation burden or lack of PD-L1 expression (48).

It would be implausible to suggest that the dramatic acceleration of PDAC development observed upon MYC and KRAS^G12D coexpression is entirely explained by these effects on the immune landscape. Indeed, these are highly pleiotropic oncoproteins with broad influence over many cell-intrinsic and cell-extrinsic biological features (21, 45). It is notable however that regulation of the type I IFN pathway occurs upon physiologic expression of KRAS^G12D or upon very modest overexpression of MYC, suggesting that this regulation may be present from the very outset of tumor initiation. This raises the exciting possibility of early intervention to reactivate IFN signaling. Although we show that pharmacologic inhibition of the RAS pathway can restore IFN-related gene expression in vitro, we would caution against such an approach in vivo given that immune cells use the very same RAS pathway to achieve their own rapid expansion (49). What is clear however is that a very modest increase in MYC suffices to dramatically accelerate KRAS-initiated tumors to metastatic PDAC, as we previously reported for lung adenocarcinoma (12), or indeed to alone drive PNET formation. The different etiologies of these phenotypes remain to be fully explored.

All experiments involving mice were approved by the local animal welfare committee [Animal Welfare Ethical Review Body (AWERB)] and conducted under U.K. Home Office licenses PE47BC0BF, 70/7950, and 70/8375. Mice were maintained on a constant 12-hour light/dark cycle, fed and watered ad libitum, and all were mixed background (FVBN and C57BL/6). The following genetically modified mice were described previously: Lsl-Kras^G12D (5); Rosa26^DM-lsl-MYC (23); Pdx1-Cre (6); Pdx1-CreER (29); Myc^fl (50); Hprt-Lsl-IRFP (30); and Miz1^ΔPOZ (38). All genotyping was performed by Transnetyx Inc. To induce allele recombination, CreER^T2 was activated in the pancreas of 5- to 6-week-old mice by intraperitoneal injection of 2 mg/kg tamoxifen (dissolved in peanut oil) for 3 consecutive days. For overall survival analysis, cohorts of mice were monitored and euthanized when clinical endpoint was reached. Endpoint monitoring was performed by facility staff without knowledge of genotype. For histologic analysis, mouse tissues were fixed with 10% neutral buffered formalin overnight. At euthanasia, a small portion of pancreatic tumor was snap-frozen for RNA analysis. To determine the influence of CXCL13, a cohort of randomly selected KMC-Miz1^ΔPOZ mice were treated with CXCL13-blocking antibody (0.5 μg, i.v., AF470, R&D Systems) or isotype control (goat IgG, AB-108-C, R&D Systems) from 6 weeks of age, twice weekly for 3 weeks, and sacrificed at clinical endpoint. To determine the influence of IFN signaling, a cohort of randomly selected KMC-Miz1^ΔPOZ mice were treated with IFNAR1-blocking antibody (clone MAR1.5A3, BE0241, BioXCell) intraperitoneally, 200 μg/20 g body mass on first dose and 100 μg/20 g for the following six doses, from 6 weeks of age, once every 3 days for 3 weeks, and sacrificed at clinical endpoint. For NK-cell depletion, 6-week-old mice were treated with anti-NK1.1 (clone PK136, BP0036, BioXCell), intraperitoneally, 100 μg/20 g, two doses in the first week and once per week for the following 3 weeks. To verify NK-cell depletion following the NK1.1 antibody treatment, blood was sampled by tail bleed from mice prior to and after 2 weeks of treatment. Cells were stained with CD45 (10311, BioLegend), CD3 (100327, BioLegend), NK1.1 (108707, BioLegend), NKp46 (137612, BioLegend), and Zombie NIR10311, (423106, BioLegend) after red blood cell lysis (eBioscience, 00-4333-57) and analyzed by flow cytometry (BD Fortessa). FACS profiles were generated and quantified using FlowJo (Tree Star). Rosa26^DM-lsl-MYC mice are available from JAX Mice at https://www.jax.org/strain/033805.

All IHC and ISH staining was performed on 4-μm formalin-fixed, paraffin-embedded sections, which had previously been heated to 60°C for 2 hours. Peroxidase blocking was performed for 10 minutes in 1% H₂O₂ diluted in H₂O, followed by heat-mediated or enzyme-mediated antigen retrieval. Nonspecific antibody binding was blocked with up to 3% BSA or up to 5% normal goat serum for 1 hour at room temperature. The following antibodies were used at the indicated dilution and indicated antigen retrieval method: synaptophysin (ab8049, 1:50, pH 6), CD45R (ab64100, 1:200, ER2 Leica), NKp46 (aF2225, 1:200, pH 6), F4/80 (ab6640, 1:100, Enzyme 1 Leica), MYC (ab32072, 1:100, pH 6, Abcam), KRAS^G12D (CST14429, 1:50, pH 9, Cell Signaling Technology), and pan-cytokeratin (MS-343, 1:100, pH 6, Thermo Fisher Scientific). NKp46, pan-cytokeratin, and synaptophysin were stained on a Dako AutostainerLink48, CD45R and F4/80 were stained on the Leica Bond Rx Autostainer, and KRAS^G12D and MYC were stained manually. Mouse EnVision (Agilent), and goat ImmPRESS Kit and rabbit IgG (Vector Laboratories) were used as secondary antibodies. The horseradish peroxidase (HRP) signal was detected using liquid DAB (Agilent and Invitrogen). Sections were counterstained with hematoxylin and cover-slipped using DPX mount (CellPath). ISH detection of Cxcl13 (ACD 406318), Miz1 (ACD 520288), and PP1β (ACD 313918; Advanced Cell Diagnostic) mRNA was performed using RNAscope 2.5 LS Detection Kit (ACD). Detection of murine Myc (ACD 712368) was performed using a BaseScope 2.5 LS Detection Kit (ACD). Both techniques were performed on a Leica Bond Rx Autostainer, strictly adhering to ACD protocols. Tumor area was calculated using HALO Software (Indica Labs) as the percent area of pancreas occupied by PDAC and PNET, measured on hematoxylin and eosin–stained sections. To quantify the immune infiltration, positive cells were counted manually using QuPath cell (https://qupath.github.io/) counter function and normalized to tumor area (PDAC + PNET) calculated as described above.

Human pancreatic cell lines were obtained from ATCC and were maintained in DMEM (MIA PaCa-2) or RPMI (AsPC-1, DAN-G) supplemented with 10% FBS and penicillin–streptomycin. All cell lines were validated using an approved in-house validation service (CRUK-BICR) and tested periodically for Mycoplasma. Cell lines were thawed from primary stocks maintained under liquid nitrogen and cultured for a maximum of 8 weeks (<20 passages), during which time all experiments were performed. Cells were treated with 10 nmol/L (DAN-G) or 50 nmol/L (AsPC-1) trametinib (MedChem Express) for 16 hours and used for protein and RNA analysis. Primary MEFs were generated from E13.5 embryos by interbreeding mice carrying Rosa26^DM-lsl-MYC, Lsl-Kras^G12D, and Myc^fl/fl to obtain desired genotypes. MEFs were cultured in DMEM with 10% FBS and penicillin–streptomycin in 3% oxygen. To activate floxed alleles, MEFs were infected with 300 pfu/cell of Adeno-Cre (University of Iowa, Vector Core Facility, Iowa City, IA) and harvested at 24 hours. Mouse pancreatic tumor lines were generated from end-stage KMC mice. Tumors were disintegrated mechanically and cultured in DMEM supplemented with 20% FBS and penicillin–streptomycin. Once established, KMC cells were maintained in DMEM supplemented with 10% FBS and penicillin–streptomycin. For cell death measurements, cells were trypsinized, quenched with 1% BSA followed by replacement of original supernatant, and centrifuged at 300 × g for 5 minutes; 200 μL of Annexin binding buffer (10 mmol/L HEPES pH 7.40, 140 mmol/L NaCl, 2.5 mmol/L CaCl₂) and 2 μL of Annexin V-APC (BioLegend 640920) were added to the pellet and incubated for 15 minutes. Propidium iodide (10 μg/mL) was added immediately prior to FACS analysis. For immunoblotting, whole-cell lysates were prepared in RIPA buffer (150 mmol/L NaCl, 50 mmol/L Tris pH 7.5, 1% NP-40, 0.5% sodium deoxycholic acid, 1% SDS, plus complete protease and phosphatase inhibitor cocktail) followed by sonication (40% Amp for 5 seconds). MYC (ab32072), vinculin (ab129002), Antibodies to histone H2B (ab1790, Abcam), pERK1/2 (CST 4370), ERK1/2 (CST 4695, Cell Signaling Technology), and Actin (SC 47778, Santa Cruz Biotechnology) were used as primary antibodies. Secondary HRP-conjugated antibodies (α-mouse IgG NA931V and α-rabbit IgG NA934V, both GE Healthcare; and α-goat IgG, Vector Laboratories PI-9500) were detected by chemiluminescence (Bio-Rad Western blotting substrate 1705060). To deplete MYC in human pancreatic cell lines, the following siRNAs were used: siMYC 5 (Qiagen SI00300902) and siMYC 9 (SI03101847). To deplete MYC in mouse pancreatic tumor lines, a combination of mouse and human siRNA was used as follows: Pool 1 (SI01321012 and SI00300902), Pool 2 (SI01321012 and SI03101847), Pool 3 (SI01320991 and SI00300902; all from Qiagen). To deplete Miz1 in mouse lines, SI01320991 (Qiagen) was used. To deplete Kras in mouse lines, SI02742439 (Qiagen) was used.

Bone marrow cells were isolated from WT mice by snipping the end of the femur and centrifuging at 5,000 rpm for 1 minute. Monocytes were differentiated into macrophages by culturing in non–TC-treated plates (Corning, 430597) for 6 days in RPMI medium containing M-CSF (10 ng/mL). Medium containing M-CSF was replaced after 3 days. Where indicated, BMDMs were treated with recombinant mouse IFNβ1 (8234-MB, R&D Systems) for 4 hours (10 ng/mL). KMC tumor cells were used to generate conditioned medium. Fresh medium (DMEM + 20%FBS) were added 24 hours after transfection with siMYC or nontargeting control siRNA. Twenty-four hours later, conditioned medium from MYC or nontargeting control–depleted cells was collected, centrifuged (1,200 rpm, 5 minutes), and supplemented with M-CSF and, where indicated, anti-IFNAR1–blocking antibody (20 ng/mL, BE0241, BioXCell) or mIgG (BE0083) prior to BMDM treatment. BMDMs were additionally pretreated with anti-IFNAR1 or mIgG overnight. BMDMs were treated with conditioned medium for 24 hours. To detect Cxcl13 expression in BMDMs, cDNA was synthesized with Oligo-dT primer (M510, Promega), and preamplified (40 nmol/L of each primer, SYBR Green buffer, 1 mg/mL BSA, 2.5% glycerol) for GusB and Cxcl13 for 20 cycles. Diluted cDNA (1:20 in 10 mmol/L Tris and 1 mmol/L EDTA) was quantified by real-time PCR using SYBR Green method (VWR QUNT95072).

NK cells were activated ex vivo as described previously (51): Freshly isolated splenocytes of naïve mice at 3 × 10⁶ cells/mL were treated for 72 hours with 50 ng/mL of IL2 (BioLegend, catalog number 575404). Target cells, stained with a cell trace dye (Thermo Fisher Scientific, catalog number C34567 or V12883) following the manufacturer's instructions, were cocultured with different ratios (0, 5, and 10) of IL2-stimulated splenocytes, enriched for activated NK cells. After 4-hour coincubation, cells were harvested, stained with Zombie NIR (BioLegend, catalog number 423106), and analyzed by flow cytometry (BD Fortessa). FlowJo (Tree Star) software was used for analysis.

Dan-G cells were cross-linked with formaldehyde (1% final concentration) for 10 minutes at 37°C. Cells were scraped in PBS containing protease inhibitors and centrifuged at 300 × g for 5 minutes (4°C). Cells were resuspended in lysis buffer 1 (5 mmol/L PIPES pH 8, 85 mmol/L KCl, 0.5% NP40) and incubated on ice for 20 minutes. After lysis, cells were centrifuged at 300 × g for 5 minutes (4°C) and the pellets were resuspended in lysis buffer II (RIPA). After 10-minute incubation, lysates were sonicated to fragment DNA. Chromatin was precleared and immunoprecipitated with 5 μg of MYC (N262, SC764) or MIZ1 (B10, SC136985) antibody overnight at 4°C followed by incubation with 60 μL of Protein G beads (17-0618-01, GE Healthcare). Rabbit IgG (CS2729) or mouse IgG (BE0083) was used as antibody control. DNA was then eluted using elution buffer (1% SDS, 0.1 mol/L NaHCO₃), de–cross-linked and purified using Qiagen PCR Purification Kit for q-PCR analysis. For Re-ChIP, α-MYC or RIgG control immunoprecipitated chromatin was eluted with Re-ChIP elution buffer (1× TE, 2% SDS, 15 mmol/L DTT) at 37°C for 30 minutes, immunoprecipitated with 5 μg of MIZ1 (B10, Sc136985) antibody overnight at 4°C. The immunoprecipitated DNA was eluted as described above. The following primer sets were used: IRF7 promoter: gacaccagcctgaccaacatag, acaatcttggcccaccacaac; IRF5 promoter: aagagcaagagttaccaagcga, taaagaacctcaccccagaacc; IRF5 intronic region: ctctggctttctcctgcagacc, cccattgaagccctgggtact; STAT1 promoter: gctggtcgtcactctcacaa, tcgcctactcttaaggggct; STAT2 promoter: tccaggctcctcaagctagt, gcactttctacgaggggagg; and VAMP4 promoter: cagtggttgttcctcccta, ccgagccctattcacctaaa.

Ribosome-depleted total RNA was used for analysis of MEF gene expression. Poly-A enriched RNA was analyzed for bulk tumor and KMC cell line gene expression. Datasets are available from ArrayExpress under accession numbers E-MTAB 6824 (MEFs); E-MTAB 8792 (KMC cell lines), and E-MTAB 8797 (PDAC tumors). Full details of next-generation sequencing protocols and downstream analysis are provided in the Supplementary Materials and Methods.

Raw data obtained from qRT-PCR, FACS, and growth curves were copied into Excel (Microsoft) or GraphPad prism spreadsheets. All mean and SEM values of biological replicates were calculated using the calculator function. Graphical representation of such data was produced in Excel or GraphPad Prism. Statistical significance was determined by the Student t test. For multiple comparisons, ANOVA was used with a post hoc Turkey test or post hoc Fischer LSD test. For non–normally distributed data (e.g., survival benefit and immune cell infiltration), Mantel–Cox (two-way comparison), or Kruskal–Wallis (multiple comparison) tests were performed. For RNA-seq data, adjusted P values calculated in R are shown. *, P < 0.05; **, P < 0.01; ***, P < 0.005.

O.J. Sansom reports receiving commercial research grants from AstraZeneca, Novartis, and Cancer Research Technology. D.J. Murphy reports receiving commercial research grants from Puma Biotechnology and Merck Pharmaceutical Group. No potential conflicts of interest were disclosed by the other authors.

Conception and design: N. Muthalagu, S.B. Coffelt, O.J. Sansom, D.J. Murphy

Development of methodology: N. Muthalagu, R. Wiesheu, S. Neidler, S.B. Coffelt, O.J. Sansom

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): N. Muthalagu, T. Monteverde, X. Raffo-Iraolagoitia, R. Wiesheu, D. Whyte, S. Laing, B. Kruspig, S.A. Karim, K. Gyuraszova, L.M. Carlin, J.P. Morton

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): N. Muthalagu, X. Raffo-Iraolagoitia, R. Wiesheu, D. Whyte, A. Hedley, R. Upstill-Goddard, R. Shaw, C. Rink, A.V. Biankin, S.B. Coffelt, J.P. Morton, D.J. Murphy

Writing, review, and/or revision of the manuscript: N. Muthalagu, T. Monteverde, A. Hedley, R. Shaw, C. Nixon, A.V. Biankin, S.B. Coffelt, O.J. Sansom, J.P. Morton, D.J. Murphy

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): C. Nixon, W. Clark, S.B. Coffelt

Study supervision: N. Muthalagu, S.B. Coffelt, O.J. Sansom, J.P. Morton, D.J. Murphy

Other (supervised the work on this paper of one of the co-authors): L.M. Carlin

The authors thank all members of the Murphy Lab past and present who contributed to the development of this work. We would like to thank the core services at the CRUK Beatson Institute with particular thanks to the Biological Services Unit and histology core facility. Thanks to Kirsteen Campbell, Florian Bock, and the Stephen Tait laboratories for helpful discussions and to Catherine Winchester for critical reading of the manuscript. Artwork was provided by Sara Zanivan using images from https://smart.servier.com/. Miz1^ΔPOZ mice were provided by Martin Eilers. Funding for this work was provided by Worldwide Cancer Research grant AICR 15-0279; Pancreatic Cancer UK Future Leaders Academy 2017; European Commission Marie-Curie actions PCIG13-GA-2013-618448, SERPLUC, and H2020-MCSA-IF-2015-705190-NuSiCC; and Cancer Research UK grants A21139 (to O.J. Sansom) and A23983 (to L.M. Carlin). N. Muthalagu received a L'Oreal/UNESCO Women in Science 2018 award. T. Monteverde was funded through a British Lung Foundation studentship (BLF-APHD13-5). Additional support was provided by Wellcome Trust grant 105614/Z/14/Z; CRUK Glasgow Cancer Centre grant A25142; and CRUK Beatson Institute core facilities grant A17196.