Abstract
Pancreatic ductal adenocarcinoma (PDAC) is the most lethal common malignancy, with little improvement in patient outcomes over the past decades. Recently, subtypes of pancreatic cancer with different prognoses have been elaborated; however, the inability to model these subtypes has precluded mechanistic investigation of their origins. Here, we present a xenotransplantation model of PDAC in which neoplasms originate from patient-derived organoids injected directly into murine pancreatic ducts. Our model enables distinction of the two main PDAC subtypes: intraepithelial neoplasms from this model progress in an indolent or invasive manner representing the classical or basal-like subtypes of PDAC, respectively. Parameters that influence PDAC subtype specification in this intraductal model include cell plasticity and hyperactivation of the RAS pathway. Finally, through intratumoral dissection and the direct manipulation of RAS gene dosage, we identify a suite of RAS-regulated secreted and membrane-bound proteins that may represent potential candidates for therapeutic intervention in patients with PDAC.
Accurate modeling of the molecular subtypes of pancreatic cancer is crucial to facilitate the generation of effective therapies. We report the development of an intraductal organoid transplantation model of pancreatic cancer that models the progressive switching of subtypes, and identify stochastic and RAS-driven mechanisms that determine subtype specification.
See related commentary by Pickering and Morton, p. 1448.
This article is highlighted in the In This Issue feature, p. 1426
Introduction
Pancreatic ductal adenocarcinoma (PDAC) has one of the highest mortality rates among all types of cancer even when diagnosed at an early stage. For all stages combined, the 5-year survival rate is 9%. However, more than half of patients are diagnosed with metastatic disease, which is associated with a 5-year survival rate of only 3% (1). The development of early detection methods and effective therapies is needed to improve the outcome of patients with PDAC (2). Accordingly, the recent revolution in cancer genomics has led to the characterization of the genetic landscape of PDAC, and clinical trials in which treatment is matched to underlying mutations are currently under way (3). A new perspective relevant to therapeutic development has emerged from the recent identification of two major transcriptional subtypes of PDAC with characteristic histopathologic features and different prognoses (4–9). These include a “basal-like” (or squamous) subtype, which is poorly differentiated and carries a worse prognosis when compared with the better-differentiated “classical” (or progenitor) subtype (4, 7, 10). Basal-like and classical subtypes have been shown to share features of the squamous and progenitor subtypes, respectively, as defined by an independent PDAC classification system (5, 11, 12), which also can predict response to chemotherapy (6, 13–15). In addition, the basal-like subtype is associated with the activation of genes involved in the epithelial–mesenchymal transition (EMT), activation of the transcription factors MYC and TP63, and downregulation of endodermal identity markers, such as HNF4A and GATA6, whose expression are therefore hallmarks of the classical subtype (4, 6, 7). In addition, expression of the ΔN isoform of TP63 (ΔNp63) and GLI2 were recently shown to promote the basal-like identity in PDAC (16, 17). Further characterization of the molecular phenotype of these transcriptional subtypes may lead to the identification of tailored therapies with improved efficacy.
Beyond the genetic mechanisms of subtype specification (18, 19), epigenetic alterations have also been shown to dynamically regulate PDAC transcriptional subtypes. For example, paracrine communication from the tumor microenvironment can promote subtype specification (9, 20–22). In addition, mutations in several epigenetic regulator genes, including KDM6A, KMT2C, and KMT2D, have been associated with the basal-like subtype (4, 6, 23). However, the relative importance of genetic and epigenetic aberrations to promote the stable or dynamic regulation of subtypes is still unknown. Moreover, the development of subtype-based therapeutic strategies remains challenging due to the complex heterogeneous features of PDAC and the lack of preclinical models that recapitulate these subtypes.
To address these limitations, we sought to establish an in vivo xenograft model that could recapitulate the basal-like and classical subtypes of PDAC. As the pancreatic ducts are the sites where preinvasive pancreatic neoplasms progress in humans, we reasoned that seeding the murine pancreatic ductal system with human neoplastic cells would offer new cellular and molecular insights into the mechanisms promoting progression of this malignancy. We transplanted a panel of human patient-derived PDAC organoids, which genetically and transcriptionally reflect the diversity of PDAC (24), into the pancreatic ductal system of immunodeficient mice. The intraepithelial neoplasms that formed following xenotransplantation followed one of two patterns of progression: aggressive or indolent, which reflected the basal-like or classical subtypes of PDAC, respectively. Through analysis of the two progression phenotypes observed in our model, we have identified roles for both epigenetic and KRAS-driven signaling in promoting the progression to a basal-like subtype.
Results
Establishing an In Vivo Model That Recapitulates the Progression from Intraductal Epithelial Neoplasia to Invasive PDAC
We previously developed a three-dimensional organoid culture system to propagate patient-derived PDAC tissue from surgically resected primary tumors (T) and metastases (M), or fine-needle aspirates (F; ref. 25). To generate in vivo models of PDAC, we previously injected these organoids directly into the pancreatic interstitium of immunodeficient mice (Fig. 1A; refs. 25, 26). This “orthotopically grafted organoid” (OGO) model initially develops high-grade, extra-ductal neoplasms that resemble advanced pancreatic intraepithelial neoplasms (PanIN) and progress morphologically over several weeks to months to form frankly invasive tumors (25), markedly faster than the timescale of PDAC progression hypothesized to occur in humans (27, 28). Although this model is well suited for generating large PDAC tumors, these tumors arise within the interstitial space rather than within the pancreatic ductal system where human PDAC progression is believed to occur (29). We therefore sought to develop an intraductally grafted organoid (IGO) model that better recapitulates the progression from preinvasive cells within the duct system to tissue-invasive extraductal neoplasms with surrounding desmoplasia. To evaluate how IGO-derived cells develop tumors, we injected single-cell suspensions of organoids into the main pancreatic ducts of immunodeficient mice (Fig. 1A; Supplementary Fig. S1A) via a modified retrograde pancreatic intraductal injection method (30) and analyzed the location of cancer cells by immunofluorescence (IF) and three-dimensional imaging. At day 1 post transplantation, organoids were detected focally along the main pancreatic duct (Supplementary Fig. S1B). Within 1 to 2 weeks post transplantation, organoids were engrafted within the intercalated ducts (Fig. 1B and C, white arrow marks ducts labeled with CK19 staining) and the centro-acinar area (Fig. 1B and C, white dotted line marks the centro-acinar area; Supplementary Videos S1 and S2; Supplementary Fig. S1C) indicating that organoid cells were ultimately able to seed the terminal end of the duct, resulting in intraductal neoplasms focally connected to the host murine pancreatic duct (Fig. 1D; Supplementary Fig. S1D). Two distinct histologic features could be observed at 4 weeks following transplant: some regions derived from these transplants exhibited features of intraductal glandular prominence with columnar epithelium (Fig. 1E, white arrows), whereas other regions lacked glandular features and rather exhibited invasive features (Fig. 1E, white arrowheads; Supplementary Videos S3 and S4).
Thus, IGO-derived xenografts displayed a range of histologic features, with a subset of organoid cultures remaining restricted to the intraductal region.
To comprehensively characterize the difference in the phenotypes between the IGO and OGO models, we transplanted 21 distinct organoid cultures including three KRAS wild-type lines (24), using both methods, and analyzed differences in survival time of the transplant mice. All 21 organoids successfully engrafted in both models with variable efficiency (Supplementary Fig. S1E) except for two organoids (hF23, hF27) that exhibited a much lower engraftment rate (2/6 and 1/6) in the OGO model compared with that (7/8 and 7/7) in the IGO model. This result suggests that the ductal environment better supports survival of hF23 and hF27 compared with the extraductal environment.
A wide variation in survival times following PDAC organoid transplantation was observed (Fig. 1F). Interestingly, the survival times of the IGO transplants formed a bimodal distribution (Fig. 1G; Supplementary Table S1), whereas the survival of OGO transplants comprised a unimodal distribution (Fig. 1G). By approximating sampled survival times with two normal distributions, we defined survival thresholds for fast-progressing lines (“Fast-progressors”) and slow-progressing lines (“Slow-progressors”) in the IGO model (Fig. 1F and G; Supplementary Fig. S1F; and SM1). Mice bearing Fast-progressors (10/21 transplanted organoid lines) had mean survival times of fewer than 111 days, whereas the mean survival times of mice bearing Slow-progressors (9/21 transplanted organoid lines) were greater than 157 days (Fig. 1F and G; Supplementary Fig. S1F; and SM1). Although xenograft growth rates of each organoid in the OGO model showed similar trends as seen in corresponding organoids in the IGO model, survival times of Fast-progressors and Slow-progressors in the IGO model were shorter and longer, respectively, than survival times of the corresponding lines in the OGO model (Fig. 1H).
To determine whether the survival of mice bearing IGO transplants (IGO mice) reflected patient disease characteristics, clinical features of the patients from whom organoids were derived were compared with features of the IGO tumors. Eighty-six percent (6/7) of Fast-progressors were derived from stage 3/4 patients, whereas only 44% of Slow-progressors (4/9) were derived from stage 3/4 patients (Supplementary Table S2). The common metastatic sites for both IGOs and OGOs were lung and liver, followed by peritoneum (Supplementary Table S3), similar to patients with PDAC (31), and the metastatic rate of the Slow-progressor IGO models was less than the Fast-progressor IGOs, consistent with stage of the patients from whom the organoids were derived. We observed a slightly reduced metastatic frequency in the Fast-progressors in the IGO model compared with the same organoids engrafted in the OGO model (Supplementary Table S3), potentially reflecting the more rapid death of IGO mice from obstructive jaundice that temporally precluded the development of metastases (Supplementary Fig. S1G, left). At humane endpoint, 43% (16/37) of Slow-progressor IGOs (3 of 9 organoid lines) developed cystic tumors (Supplementary Fig. S1G, right, and S1H), although all IGOs except one (hT102) showed evidence of tumor progression with invasion into adjacent pancreatic tissues and/or distant metastasis (Supplementary Fig. S1H and S1I).
Next, we sought to investigate whether differences in the murine microenvironment where the organoids were engrafted may affect the proliferation of cancer cells and shape the histologic features between IGO and OGO tumors. Accordingly, the number of actively proliferating cells was assessed by IF staining for Ki-67 on Fast- and Slow-progressor IGO and OGO tumor sections, and organoid-derived cancer cells were identified by expression of a human mitochondrial protein. We found that the Ki-67 content in intraductal lesions of four IGO-derived Slow-progressors (hF23, hF27, hT93, and hT102) was significantly lower than their corresponding OGO tumors, whereas proliferation in Fast-progressors was consistently comparable in both the IGO or OGO setting (Supplementary Fig. S1J and S1K). Notably, although many xenografted cells in the intraductal areas of Fast-progressors expressed Ki-67, in Slow-progressors, only human PDAC cells adjacent to murine duct cells were positive (Fig. 1I and J). OGO transplants uniformly displayed abundant interactions between cancer and stromal cells as well as prominent collagen deposition in comparison to tumors derived from IGO transplants (Fig. 1K). Invasive tumors following the progression of IGO transplants were oftentimes contiguous with the duct and accompanied by an abundant stromal reaction (Fig. 1K). Thus, we observed differences in both the neoplastic histology and the grade of stromal reaction present in IGO compared with OGO transplants. Our results suggest that the pancreatic microenvironment engraftment site may dictate the resulting histologic phenotypes.
Tumor Microenvironment Influences Transcriptional Subtypes of PDAC
We noted that a major difference between these two models was the microenvironment that the PDAC cells were engrafted in: whereas IGO transplants were exposed primarily to cues from the pancreatic ducts, OGO transplants were exposed to cues from the more varied tissue elements and cell types present in the interstitial space of pancreas, including blood vessels, nerves, connective tissue, and pancreatic stellate cells—a resident fibroblast type in the pancreas. To evaluate whether the microenvironment in which organoids engraft influences the transcriptome of cancer cells, we performed RNA sequencing (RNA-seq) using pairs of tumors derived from OGO and IGO transplants of 14 different organoid lines, taken at endpoint. We used Callari and colleagues' (32) method to discriminate between human cancer cells and mouse stroma transcriptomes. Gene expression changes in cancer cells were determined by analyzing those reads that mapped specifically to the human reference genome. Gene set enrichment analysis (GSEA) using the Cancer Hallmarks database indicated that genes involved in EMT and IFNα/γ response were enriched in the OGO relative to the IGO xenografts (Fig. 2A). We confirmed that the EMT marker gene Vimentin was highly expressed in OGO tumors compared with IGO tumors in hF2 and hT93 tumors (Fig. 2B and C; Supplementary Fig. S2A–S2C). The previously described “squamous PDAC” (17) and “basal-like” (7) gene signatures were enriched in the OGO xenografts (Fig. 2D). Conversely, the previously described “progenitor PDAC” (17) and “classical” (7) signatures were enriched in IGO-derived tumors compared with OGO-derived tumors (ref. 17; Fig. 2D). Moreover, the oxidative phosphorylation and fatty acid metabolism pathways, which have been associated with the PDAC progenitor/classical subtypes (4, 33), were upregulated in cancer cells from IGO tumors (Fig. 2A). These results suggest that differences in the microenvironment to which engrafted cells are exposed can influence the metabolic phenotype as well as the molecular subtype of PDAC. Next, to evaluate whether xenografted cancer cells influence the transcriptome of mouse stromal cells in the microenvironment, gene expression changes in stromal cells were analyzed by reads mapped to the mouse reference genome. GSEA revealed the enrichment of genes involved in EMT, WNT, and TGFβ response in mouse stromal cells in the OGO relative to the IGO xenografts (Fig. 2E). The extracellular matrix–rich (“ECM-rich”) gene signature (9) and “Moffitt-Activated” signatures (7) were also found to be enriched in the mouse stromal cells in OGO xenografts (Fig. 2F). In contrast, genes involved in oxidative phosphorylation and xenobiotic metabolism as well as a “Moffitt-Normal” signature (7) were enriched in mouse stroma cells in IGO-derived tumors compared with OGO-derived tumors (Fig. 2F).
We then sought to examine how the microenvironment influences tumor progression in the IGO model. At 4 to 8 weeks following transplant via the OGO model, the transplanted organoids had formed extraductal neoplasms with infiltration and activation of stromal cells (Fig. 2G; Supplementary Fig. S2D). In contrast, four weeks after IGO transplant, some organoids, such as hM1A, demonstrated invasion through the duct and generated lesions with abundant desmoplastic reaction, whereas other organoids, such as hF27, expanded intraductally and lacked an associated desmoplastic reaction (Fig. 2G; Supplementary Fig. S2D). Previous studies have shown that the microenvironment can influence the heterogeneity and aggressiveness of PDAC lesions through upregulation of signaling pathways including the TGFβ pathway, a major inducer of EMT (7, 8, 22, 34). As shown above, EMT and TGFβ pathways were upregulated in both the cancer cells and the stromal cells of OGO tumors. To assess the extent of TGFβ signaling during tumor development in IGO and OGO tumors, we analyzed the difference between IGO and OGO tumors at an early time point. To do so, we collected neoplastic cells from hM1A and hF24 xenograft tumors 4 weeks after transplant by either the IGO or OGO method, using human EPCAM-coated magnetic beads, and RNA was isolated for qPCR analysis. The hF24 organoid line is SMAD4-null (Supplementary Fig. S2E) and therefore lacks SMAD4-dependent TGFβ signaling. Accordingly, the TGFβ effectors SERPINE1 and SMAD7 were undetectable in both OGO- and IGO-derived transplants of hF24 (Supplementary Fig. S2F). Consistent with our hypothesis, SERPINE1 and SMAD7 were more highly expressed in the neoplastic cells arising from OGO transplants of hM1A compared with those from IGO transplants (Supplementary Fig. S2F), suggesting that TGFβ signaling is accentuated by the microenvironment present in the OGO model, compared with that of the IGO model. To confirm which cells express TGFβ ligands, we performed qPCR for the isolated tumor cells 4 weeks after transplantation (Supplementary Fig. S2G) and RNA-ISH for Tgfb1 and Tgfb2 in stromal cells using mouse-specific probes (Supplementary Fig. S2H). Both TGFB1 and TGFB2 were more highly expressed in cancer cells in OGO tumors compared with IGO tumors (Supplementary Fig. S2G). Moreover, increased expression of Tgfb1 and Tgfb2 was observed in stromal cells in OGO tumors compared with IGO tumors (Supplementary Fig. S2H). These results suggested that active autocrine and paracrine TGFβ signaling in both tumor and stromal cells may play a major role in tumor progression in OGO tumors.
To confirm that the transplantation method can influence the molecular subtype of the resulting tumor, we assessed OGO- and IGO-derived hM1A cells from tumors 4 weeks post transplantation for levels of molecular subtype markers (Supplementary Fig. S2I). S100A2, FAM83A, IGTA3, KRT5, and C16orf74, markers of squamous/basal-like PDAC (4, 7, 12, 17) were upregulated, whereas TFF2, AGR2, REG4, LGALS4, ANXA10, and DDC, markers of progenitor/classical PDAC (4, 7, 12, 17) were downregulated in hM1A OGO tumors compared with IGO tumors (Supplementary Fig. S2I), supporting our RNA-seq findings (Fig. 2D). In hF24, which is SMAD4-null, ITGA3 (squamous/basal-like marker) was upregulated and DDC (progenitor/classical marker) was downregulated in OGO tumors compared with IGO tumors (Supplementary Fig. S2I). Collectively, our results show that the microenvironment in which organoids are engrafted influences molecular characteristics of the cancer cells.
To evaluate which model better represents the diversity of transcriptional subtypes of human PDAC, we compared the proportion of subtypes in IGO and OGO tumors with those published in the COMPASS database (19), in which authors reported that previously reported “basal-like” and “classical” subtypes were split into “Basal-like-A” (BA) and “Basal-like-B” (BB), and “Classical-A” (CA) and “Classical-B” (CB), respectively. Genes from four signatures (19) were used for nonbiased hierarchical clustering (Fig. 2H), and we found that the IGO tumors predominantly comprised CA and CB subtypes, whereas the OGO tumors predominantly comprised BA and BB subtypes. The distribution of subtypes in IGO tumors best reflected the relative distribution of subtypes reported in the COMPASS database (Fig. 2I) To further confirm whether IGO models represent human PDAC, we performed IHC analyses for S100A2 (basal-like marker), TP63 (squamous marker), and GATA6 (progenitor marker) in human PDAC sections. The fine-needle aspiration sample from which the hF27 organoid was derived contained high GATA6 staining and low S100A2 and TP63 staining, whereas the original tumor from which the hM19B organoid was derived contained high S100A2 and low GATA6 staining (Supplementary Fig. S2J). IGO transplants of hM19B were classified as basal-like (Fig. 2H), consistent with the IHC results of subtype markers for the limited number of available samples. In addition, the IHC staining of subtype markers for IGO transplants of hM19B and hF27 was similar to the IHC results for their corresponding human samples (Supplementary Fig. S2K). Collectively, these results suggest that the IGO model can recapitulate the transcriptomic subtypes of human PDAC.
Indolent and Invasive Tumor Phenotypes Correlate with Molecular Subtypes
In the IGO model, organoids gave rise to distinct intraductal and invasive phenotypes that resemble the different histopathologic phenotypes observed in human PDAC progression (25). We sought to investigate whether observed differences in tumor kinetics were a function of intrinsic features of organoids observed in vitro. In vitro proliferation rates of 7 organoids were not significantly different and did not correlate with the in vivo tumor growth kinetics (Supplementary Fig. S3A and S3B). However, consistent with the findings of another group (35), hM1A and hF3 organoids (Fast-progressors) exhibited a dense, filled-in spherical morphology compared with the hollow cystic organoids formed by hF2 and hF27 (Slow-progressors; Supplementary Fig. S3C).
To gain insight into the molecular mechanisms underlying the biological differences between Fast- and Slow-progressor IGO transplants, we performed global mRNA sequencing on 6 Fast- and 9 Slow-progressors. Differentially expressed gene (DEG) analysis identified 231 and 335 genes significantly upregulated in Fast- and Slow-progressors, respectively (“Fast-signature” and “Slow-signature,” respectively, Fig. 3A). To determine whether there was a correlation between our progression phenotype and previously identified transcriptional subtypes, we compared genes upregulated in Fast- and Slow-progressors with “squamous,” “basal-like,” “progenitor,” and “classical” gene sets. By GSEA, EMT and proliferation/cell cycle–related genes (i.e., MYC targets, E2F targets, and G2M checkpoint signatures) were enriched in Fast-progressors (Fig. 3B). In addition, KRAS and mTORC1 signaling genes, which have been associated with PDAC progression (36), were also enriched in the Fast-progressors (Fig. 3B), whereas bile-acid metabolism and fatty-acid metabolism signaling genes were enriched in the Slow-progressors (Fig. 3B). We also observed that the progenitor-type PDAC signature and classical signatures were significantly enriched in Slow-progressors, whereas the squamous PDAC signature and basal-like signature were significantly enriched in Fast-progressors (Fig. 3C). To confirm these results, we examined the expression levels of known progenitor/classical and squamous/basal-like marker genes by qPCR (Supplementary Fig. S3D). Among the progenitor/classical markers (4, 7, 17), LYZ, AGR2, and FOXA2 were significantly more highly expressed and GATA6 was modestly increased in Slow- compared with Fast-progressors, whereas CAV1 and S100A2 were more highly expressed among the squamous/basal-like markers (4, 7, 17) in Fast-progressors (Supplementary Fig. S3D). Moreover, IHC staining of the classical markers AGR2 and GATA6 and the basal-like marker S100A2 confirmed these results (Fig. 3D and E; Supplementary Fig. S3E–S3G).
To determine whether our Fast- and Slow-progressor gene signatures are relevant to human PDAC, we performed unbiased, hierarchical clustering of the PanCuRx Translational Research Initiative dataset of patients with locally advanced PDAC (37) based on expression of our Fast- and Slow-signature genes (Supplementary Fig. S3H). Interestingly, the Fast- and Slow-signature genes divided patients into two groups, which we classified as “Fast-signature” patients and “Slow-signature” patients (Supplementary Fig. S3H). Fast-signature patients exhibited significantly worse prognosis compared with Slow-signature patients (Fig. 3F). Moreover, tumors from 48 of 70 (65.7%) Fast-signature patients were classified as belonging to the basal-like subtype, and 86 of 96 (89.6%) Slow-signature patients were classified as the classical subtype (Fig. 3G). Taken together, these results show that our IGO-defined Fast- and Slow-signatures correlate with the basal-like and classical transcriptional subtypes of human PDAC, respectively, and with patient prognosis. Moreover, our results suggest that the growth kinetics of tumors derived from IGO transplants are predictive of PDAC transcriptional subtypes.
Transcriptional Shift Is Governed by Tumor Cell Plasticity In Vivo
Other groups have reported that transcriptional subtypes correlate with histologic manifestations and IHC staining for common squamous markers such as CK5, CK6, and TP63 (15, 18). We performed IHC for TP63 and S100A2, to distinguish basal-like and classical cell types in IGO tumors. We found that the immunostaining pattern of TP63 and S100A2 was heterogeneous (Supplementary Figs. S4A and S3F), suggesting that the squamous differentiation program may be dynamic. Previously, we reported that, based on the expression of genes in PDAC organoids cultured in vitro, organoids can be classified into classical and basal-like subtypes (24). Interestingly, the previously identified in vitro transcriptional subtypes were not always concordant with the in vivo transcriptional subtypes of IGO transplants (Supplementary Fig. S4B). The Fast-progressor organoids hF3 and hM19B were found to have a basal-like subtype both in vitro and in IGO transplants (Supplementary Fig. S4B). Accordingly, hF3 expressed the basal-like markers S100A2 and TP63 in vitro (Supplementary Fig. S4C). In contrast, the Fast-progressor organoid hM1A was classified as the classical subtype in vitro but demonstrated a basal-like phenotype in vivo (i.e., a “hybrid” phenotype; Supplementary Fig. S4B). hF24 was classified as classical subtype bothin vitro and in vivo and did not express the basal-like markers S100A2 and TP63 in vitro; however, hF24-IGO tumors contained S100A2- and TP63-positive cells in vivo (Supplementary Figs. S3F and S4A). The ability of the same organoid to manifest different subtypes depending on context suggests that transcriptional subtypes may be affected by microenvironmental cues.
Recent single-cell analyses have demonstrated the coexistence of basal-like and classical subtypes within a single tumor with the relative proportions of these cells determining the overarching subtype at the bulk transcriptome level (19, 38). In another recent study, evolutionary analysis of mixed phenotype tumors, comprising classical and basal-like features, from autopsy patients revealed that the basal-like subtype represented a subclone contained within an otherwise classical tumor (18). We questioned whether the hybrid phenotype of organoids in the IGO xenograft model might result from clonal selection or, alternatively, cellular plasticity driven by exposure to different environmental cues. To distinguish between these two possibilities, we labeled organoids with the Autobow system, which facilitates lineage tracing (Supplementary Fig. S4D; ref. 39). In this system, three fluorescent proteins with loxP sites and Cre are combined in a single cassette, and Cre activation results in the random and irreversible activation of various combinations of the three fluorescent proteins. We used the organoids hM1A and hF24 for these studies because they displayed a hybrid phenotype. Following introduction of the Autobow cassette to hM1A and hF24 and Cre activation in vitro, we confirmed that the system was working by detection of multicolor expression in cultured organoids (Supplementary Fig. S4E). Next, single-cell suspensions of these Autobow-expressing hM1A and hF24 organoids (hM1A-, hF24-Autobow) were transplanted by the IGO method. Four weeks post transplantation of hM1A-Autobow, each lesion expressed a single fluorescent protein, indicating that each individual lesion likely arose from a single engrafted cell (Supplementary Fig. S4F). At a later timepoint (8 weeks post transplantation) of hM1A-Autobow, multifocal and multichromatic tumors were observed (Supplementary Fig. S4G), indicating merging of individual clonal tumors. We observed both inter- and intraclonal differences in histologic grade, suggesting that individual clones retain the capacity to generate different histopathologic phenotypes. Notably, we observed intraclonal differences in the expression of basal-like and classical markers by IHC staining: the intraductal region of a monoclonal lesion expressed the classical marker GATA6, whereas an invasive region derived from the same clone expressed the basal-like marker S100A2 (Fig. 4A). These data suggest that the subtype identity of a transplanted single clone is not fixed and can be determined by the tumor microenvironment in vivo.
To exclude the possibility that different clones with the same color give rise to a single lesion by chance, hM1A and hF24 clonal organoids were derived from single cells in vitro and transplanted intraductally (Supplementary Fig. S4H). hM1A and hF24 clones formed tumors with varied intratumor histology consisting of invasive lesions that expressed the basal-like markers S100A2 and TP63 as well as intraductal lesions negative for these markers (Fig. 4B). These results demonstrate that shifts in PDAC subtype during the progression of tumors formed from hybrid organoids are likely due to cancer cell plasticity rather than altered clonality. To further determine whether these organoids are clonal, we performed laser capture microdissection (LCM) to isolate intraductal (S100A2-negative, lacking stromal interactions) and invasive (S100A2-positive, maintaining stromal interactions) lesions of IGO transplants of hM1A clones for DNA-seq analysis (Fig. 4C; Supplementary Fig. S5A). hM1A clones #1, #2, and #5 (CL1, CL2, and CL5) clustered together with LCM samples from both intraductal and invasive lesions of IGO tumors (Fig. 4C). We demonstrated that copy-number profile of the clonal hM1A CL2 organoid was similar to the copy-number profiles of intraductal and invasive lesions of its corresponding IGO transplants (Fig. 4D), suggesting that a single clone could indeed give rise to both intraductal (S100A2 negative) and invasive (S100A2 positive) lesions. CL1 and CL5 generated nearly identical copy-number profiles in vitro and in vivo, with minor focal differences that can be attributed to noise or slight differences in data analysis methods used for organoids and IGO tumors (Supplementary Fig. S5B). We also performed LCM to isolate intraductal and invasive lesions from IGO transplants of Autobow-hM1A clones for DNA-seq analysis (Fig. 4C; Supplementary Fig. S5C). Autobow-hM1A lesions #1, #2, and #3 clustered separately from clonal hM1A organoids, whereas intraductal and invasive lesions from same area clustered closely (Fig. 4C; Supplementary Fig. S5D), confirming that they likely originated from the same clone. We observed identical copy-number profiles between intraductal lesions and invasive lesions of IGO tumors of Autobow-hM1A (Supplementary Fig. S5D). Collectively, the presence of intraductal and invasive lesions derived from a single clone suggests that cell plasticity drives histologically distinct phenotypes.
Tumor–stromal interaction contributes to tumor heterogeneity (22) and may influence the dynamic shift of molecular subtypes which are associated with histologic phenotypes. To assess whether stromal interaction correlates with molecular subtype, we performed RNA-seq on LCM intraductal and invasive lesions of IGO transplants of hM1A clones (Fig. 4E). GSEA identified expression of EMT, TNFα signaling, and cell cycle–related gene sets to be enriched in invasive compared with intraductal lesions (Fig. 4F). EMT and proliferation signaling were previously found to be induced in cancer cells by interaction with stroma (22). Expression of the squamous PDAC signature and the basal-like signature were also enriched in invasive lesions, while the progenitor PDAC and classical signatures were enriched in intraductal lesions (Fig. 4G). Thus, IGO tumors derived from some PDAC organoids can simultaneously display squamous/basal-like and progenitor/classical phenotypes in a lesion-dependent manner, and the shift of subtypes is likely influenced by the tumor microenvironment.
Induction of a Basal-Like Phenotype in Invasive Areas of IGO Tumors Is Associated with Activation of the RAS Pathway
We next sought to understand the molecular mechanisms underlying the dynamic nature of the molecular subtypes observed in IGO-derived tumors. Allelic imbalance and high expression of mutant KRAS have been associated with aggressive undifferentiated histologic phenotypes in PDAC (19, 40). Therefore, we hypothesized that levels of KRAS signaling were higher in invasive compared with intraductal regions of IGO-derived lesions. To determine whether KRAS signaling levels varied in different lesions within IGO-derived tumors, we evaluated the levels of phosphorylated ERK (pERK), a downstream effector of KRAS signaling, in Fast- and Slow-progressor IGO tumors. Elevated levels of pERK were observed in Fast-progressor compared with Slow-progressor tumors (Supplementary Fig. S6A and S6B). Next, we sought to correlate the levels of pERK with molecular subtypes and histology in IGO-derived tumors of the Slow-progressor organoids hF2 and hF27. Whereas tumors derived from these organoids formed mostly well-differentiated intraductal lesions, smaller invasive lesions that express the basal-like marker S100A2 could also be detected (Fig. 5A; Supplementary Fig. S1I). Consistent with a role for KRAS signaling in promoting invasiveness, levels of pERK or phosphorylation of MAPK downstream target S6 (pS6) were higher in invasive areas compared with intraductal areas in hF2 tumors, and correlated with areas of high expression of S100A2 (Fig. 5A–C; Supplementary S6C and S6D). Similarly, pERK or pS6 levels were modestly upregulated in invasive lesions of hF27 tumors, which is consistent with the observed high expression of S100A2 (Fig. 5A–C; Supplementary S6C and S6D) and Ki-67 staining (Supplementary Fig. S1I and S1J). These results suggest that differential activation of the RAS pathway in intraductal and invasive lesions may contribute to the dynamic shifting of molecular subtypes in Slow-progressors.
To identify transcriptional features enriched specifically in invasive areas, we performed LCM and RNA-seq on intraductal and invasive regions of tumors derived from IGO transplants of the Slow-progressor organoids hF2 and hF27 (Fig. 5D; Supplementary Fig. S6E). GSEA using the Cancer Hallmarks database identified the E2F target genes, G2M checkpoint genes, and EMT genes as enriched among genes more highly expressed in invasive areas (Fig. 5E), consistent with the presence of a basal-like phenotype in the invasive regions of tumors generated from IGO transplants of Slow-progressor organoids. Accordingly, the classes of genes more highly expressed in the invasive compared with the intraductal regions of tumors generated from IGO transplant of Slow-progressor organoids (Fig. 5E) were quite similar to the classes of genes more highly expressed in Fast- compared with Slow-progressor tumors (Fig. 3B). For example, as was observed in the comparison of Fast- and Slow-progressors, genes from the squamous and basal-like PDAC and “Janes KRAS gene signatures” (41) were enriched in invasive areas relative to the intraductal areas of IGO transplants of Slow-progressor organoids (Fig. 5F; Supplementary Fig. S6F), whereas the progenitor and classical PDAC signatures were enriched in the intraductal regions (Fig. 5F; Supplementary Fig. S6F). These results support a role for increased RAS signaling in promoting a more invasive and basal-like phenotype in the IGO model.
KRAS Hyperactivation Promotes a Basal-Like Phenotype
To further determine whether differences in RAS signaling could underlie the different phenotypes of Fast- and Slow-progressors, we analyzed KRAS copy number (CN) in both groups using Fraction and Allele-Specific Copy Number Estimates from Tumor Sequencing (FACETS). KRAS CN was significantly higher in Fast-progressor compared with Slow-progressor organoids (Fig. 6A). In line with these results, expression of DUSP6, a target gene of KRAS signaling, correlated positively with KRAS CN (Fig. 6B). We next sought to determine whether increased KRAS signaling is sufficient to promote an invasive phenotype. To this end, we chose two Slow-progressor organoids harboring high CN of the classical subtype marker GATA6 and low CN of KRAS (KRASG12D; hF23 and hF27; Supplementary Fig. S7A). We transfected these organoids with a cDNA encoding an estrogen receptor fused to oncogenic KRASG12D that functions in a 4-hydroxytamoxifen (4-OHT)-inducible manner. Hyperactivation of oncogenic ER–KRASG12D in vitro by 4-OHT treatment resulted in increased levels of pERK and expression of DUSP6 (Supplementary Fig. S7B and S7C). In addition, hyperactivation of KRASG12D resulted in a change in hF23 and hF27 organoid morphology, characterized by the development of multiple cell layers resembling the morphology of Fast-progressor organoids cultured in vitro (Supplementary Fig. S3C; Fig. 6C and D). We observed no changes in proliferation rates upon hyperactivation of oncogenic ER–KRASG12D in vitro by 4-OHT treatment (Supplementary Fig. S7D), which reflected the lack of differences in the proliferation rates in Fast- and Slow-progressor organoids (Supplementary Fig. S3A and S3B).
Next, RNA-seq analysis was used to compare gene expression in organoids where ER–KRASG12D was hyperactivated by 4-OHT treatment (KRAS-ON) compared with untreated ER–KRASG12D organoids (Control). DEG analysis indicated that 446 and 291 genes were significantly upregulated and downregulated, respectively, in KRAS-ON compared with Control organoids [designated “KRAS-ON (in vitro)” and “Control (in vitro)” signatures]. Similar to the pathways found upregulated in invasive regions of IGO tumors, GSEA revealed the enrichment of EMT and TNFα signaling via NFκB pathways as well as the squamous and basal-like PDAC signature in KRAS-ON relative to Control organoids, whereas E2F signaling, IFNα response pathways, and the Progenitor PDAC signature were enriched in Control organoids (Supplementary Fig. S7E and S7F). Accordingly, S100A2 and GATA6 protein levels were increased in KRAS-ON and Control organoids, respectively (Supplementary Fig. S7G).
To determine whether the expression of genes associated with KRAS hyperactivation promotes basal-like features in our IGO model in vivo, we transplanted Empty vector (Empty) and ER–KRASG12D (KRAS)-expressing hF23 and hF27 intraductally and induced KRAS activation in vivo by tamoxifen treatment for 4 weeks (Supplementary Fig. S8A). Increased pERK- and S100A2-positive cells were observed in IGO tumors upon KRAS activation (Fig. 6E and F). There were no differences in tumor size and Ki-67 staining between Empty IGO tumors and KRAS IGO tumors (Supplementary Fig. S8B–S8D). Notably, however, we did observe an increased number of invading cells in KRAS-hF23 IGO tumors and a high-density level of collagen deposition in KRAS-hF27 IGO tumors compared with Empty IGO tumors (Fig. 6E and F; Supplementary Fig. S8C). This demonstrates that increases in oncogenic KRAS are sufficient to promote the invasive phenotype and desmoplastic reaction in vivo.
Next, RNA-seq analysis was used to compare gene expression in KRAS IGO tumors compared with Empty IGO tumors. DEG analysis indicated that 249 and 116 genes were significantly upregulated and downregulated in KRAS IGO tumors compared with Empty IGO tumors [designated “KRAS-ON (in vivo)” and “Control (in vivo)” signatures]. EMT and TNFα signaling via NFκB pathways were enriched in KRAS IGO tumors relative to Empty IGO tumors, whereas E2F and G2M signaling pathways were enriched in Empty IGO tumors (Supplementary Fig. S8E). In addition, the squamous and basal-like PDAC signatures were enriched in KRAS IGO tumors (Supplementary Fig. S8F). In contrast, the progenitor PDAC signature was enriched in IGO tumors (Supplementary Fig. S8F).
To identify common gene signatures induced by KRAS hyperactivation between in vitro and in vivo, we performed gene ontology analysis using overlapping genes between KRAS-ON (in vitro) and KRAS-ON (in vivo) (Supplementary Fig. S8G). Wound healing and cell migration pathways were upregulated both in KRAS-ON (in vitro) and KRAS-ON (in vivo) (Supplementary Fig. S8H). The cell junction organization pathway was upregulated only in KRAS-ON (in vitro), reflecting the morphologic change induced in vitro by KRAS hyperactivation (Fig. 6C; Supplementary Fig. S8H). IFN signaling and chemotaxis pathways were upregulated only in KRAS-ON (in vivo), suggesting cross-talk exists between cancer and stromal cells in vivo (Supplementary Fig. S8H). To determine whether KRAS-ON signatures are relevant to human PDAC, we calculated KRAS-ON (in vitro) and KRAS-ON (in vivo) scores of patients with PDAC in the PanCuRx dataset using single sample GSEA (ssGSEA). Patients with high KRAS-ON (in vitro) and KRAS-ON (in vivo) scores both exhibited poorer prognosis (Fig. 6G). Moreover, patients with tumors classified as basal-like subtype exhibited high KRAS-ON scores, suggesting that the KRAS-ON signatures correlate with a basal-like subtype (Fig. 6H).
To understand how KRAS hyperactivation contributes to the invasion and basal-like features in our model, we compared genes upregulated in invasive compared with intraductal areas of Slow-progressor tumors from our LCM expression data (Fig. 4F and G) with genes defining our KRAS-ON (in vivo) signature. Between the two gene sets, we identified 13 overlapping genes (Fig. 6I and J). Among them, AMIGO2 functions in axon guidance (42), and PMEPA1, ADAMTS9, MMP7, SERPINA1, TGM2, KLK6, and KLK7 function in EMT and ECM remodeling (43–49). For 4 of these 13 genes (AMIGO2, KLK6, PMEPA1, and TGM2), high expression was significantly correlated with poorer prognosis in patients from the PanCuRx dataset (Fig. 6K; Supplementary Fig. S8I). In addition, among those 4 genes, AMIGO2, KLK6, and TGM2 are known to be associated with the squamous phenotype (4, 17). Taken together, these results demonstrate that increased KRAS signaling induces EMT, axon guidance, and ECM remodeling pathways, and together these pathways may promote a basal-like subtype in PDAC. Importantly, these cellular and molecular features of advanced PDAC can be recapitulated and systematically evaluated in an intraductal xenotransplantation model (Fig. 6L).
Discussion
Personalized models of PDAC have the potential to provide insight into the heterogeneous biology of the disease and to facilitate the identification of novel biomarkers and more effective therapies. To date, studies of the early stages of human PDAC development and the progression of this disease have relied largely on findings from surgically resected specimens. However, these specimens cannot be perturbed for functional studies, and such specimens do not fully recapitulate the more advanced stages of PDAC, which generally are not resectable. Our previous studies have shown that our organoid culture platform can be exploited for genomic and functional studies to predict chemosensitivity and to inform personalized treatment (24). In addition, we previously demonstrated that a xenograft model in which organoids are injected into the pancreatic interstitial space (the OGO model) forms tumors that recapitulate the histologic heterogeneity of the tissue from which the organoids were derived (25). Here, we present a new xenograft model involving intraductal injection of dissociated organoids for the study of tumorigenesis and progression of human PDAC. This IGO xenograft model leads to initial engraftment of organoids as bona fide intraepithelial neoplasms in the host pancreatic duct system and to the development of intraductal and invasive lesions resembling human disease. Our IGO model offers distinct advantages over the prior OGO model because we can clearly identify demarcation of Fast- and Slow-progressors as determined by a bimodal distribution of survival. The Fast-progressors are hyperproliferative and invasive, and can activate cancer-associated fibroblasts when compared with Slow-progressors, and transcriptional and molecular analyses show that the Fast- and Slow-progressor phenotypes recapitulate the basal-like/squamous and classical/progenitor subtypes of human PDAC, respectively (4, 6, 7). Furthermore, the distribution of molecular subtypes in the IGO model reflects the relative distribution of subtypes of human PDAC tumors described previously (19). Therefore, the IGO model may be a new tool to explore PDAC subtypes with correlation to human specimens for the purpose of discovering fundamental insights and developing translational applications.
The origin of PDAC is still debated, but several studies suggest that a subset of preneoplastic cells migrate from the intraductal lesions throughout the duct system, resulting in multiple lesions characterized by spatial and genetic divergences (50, 51). PanIN lesions are well-known precursor lesions to PDAC, and high-grade PanIN commonly harbors signature mutations in KRAS and CDKN2A, and less commonly in TP53 or SMAD4, that are prevalent in invasive PDAC (28). Invasive PDAC has been reported to spread into and within the duct system, which is known as cancerization of the pancreatic ducts (29). Such lesions recapitulate the histopathologic features of high-grade PanIN lesions but harbor genetic alterations more characteristic of late-stage, invasive PDAC. In an alternate model of PDAC progression, Notta and colleagues demonstrate that a subset of PDAC tumors may not progress through a linear series of histologic stages associated with the gradual accumulation of driver mutations (52). Instead, within this subset, catastrophic mitotic events occur with the development of invasive cancer. Using our IGO model, we analyzed how cancer cells with preexisting driver mutations behave within the nontransformed murine ductal system in the normal pancreata, where cancer cells arise originally, and found that injected cancer cells engraft at the peripheral ducts before expanding intraductally or invading into tissues. The propensity for cancer cells to invade into tissues is driven by cancer cell–intrinsic features, as IGO tumors derived from some organoids quickly invaded tissues, whereas tumors derived from other organoids were more likely to remain in the duct. We found that a subset of proliferating cancer cells invades through the basement membrane, replacing acinar cells and inducing an abundant stromal reaction. Alternatively, neoplastic cells that expand intraductally migrate throughout the duct, histologically resembling cancerization of the ducts (50). These results support two possible behaviors of intraductal expansion of tumor cells.
We previously reported the presence of molecular subtypes (basal-like and classical) in organoids cultured in vitro. In this study, we uncovered discordance between the molecular subtypes of some organoids when transplanted as IGO tumors compared with those observed in vitro. We found increased expression of squamous and basal-like markers in the invasive areas of Fast-progressor tumors despite low or undetectable expression in most of the corresponding in vitro organoid cultures (4, 17, 23). On the basis of the concordance of molecular subtypes observed in vitro and in tumors derived from IGO, we were able to further classify organoids as having a fixed or hybrid subtype. The hybrid subtype exhibited heterogeneity of molecular subtypes within a single tumor consistent with recent single-cell analyses and evolutionary analyses (18, 19, 38). Our findings suggest that molecular subtypes are regulated by both genetic alterations and dynamic epigenetic modulations consistent with the existence of different trajectories of PDAC progression observed in humans (51, 52). In further support of the dynamic nature of PDAC subtypes, single clones of Fast-progressor organoids, labeled by the Autobow system, could form both intraductal and invasive lesions in the IGO model. This finding supports the notion that cell plasticity and not clonality determines the heterogeneity in histologic and molecular phenotypes observed in the hybrid subtype. By comparing transcriptional signatures and molecular characteristics of the intraductal and invasive lesions, we have identified a set of genes and pathways that determine invasiveness. We hypothesize that these new genes may represent therapeutic targets that when antagonized will either prevent conversion of classical cells to basal-like or cause a reversion of basal-like to classical cells. Future research will better delineate the roles of these invasiveness genes and will test whether targeting such pathways is an effective therapeutic strategy.
The changes in molecular subtypes that we have observed following intraductal transplantation may be attributed to the influence of the microenvironment on the PDAC cells. In support of this hypothesis, although TGFβ signaling is suppressed in traditional organoid culture medium, in tumors derived from IGOs, TGFβ signaling is more highly activated in cancer cells that have invaded through the duct and are embedded in a dense stromal environment compared with those localized within an intraductal environment. EMT and proliferation signaling have been found to be invasive features of cancer cells induced by the stromal environment (22), and these pathways were enriched in invasive lesions compared with intraductal lesions. Further analysis of the cross-talk between stromal cells and cancer cells in the IGO model will help to determine the extent of stromal influence on the development of PDAC heterogeneity (26, 53).
Increased allelic frequency and gene dosage of KRAS, a gene mutated in most PDAC tumors, have been shown to alter cellular metabolism, drug sensitivity, and histopathologic/clinical outcome (9, 19). We identified a RAS–EMTpathway as upregulated in our bulk analysis of Fast- compared with Slow-progressors derived from IGO. These pathways were also upregulated in the microdissected invasive areas compared with intraductal areas of Slow-progressor tumors. Consistent with these data, we find that Fast-progressors carry higher KRAS copy number compared with Slow-progressors, and 4 out of 5 of the Fast-progressors harbor the stronger G12V allele, whereas 6 out of 8 of Slow-progressors harbor the weaker G12D and G12R alleles (Supplementary Table S2; ref. 54). Moreover, increased mutant KRAS dosage is sufficient to induce basal-like/squamous features in Slow-progressor organoids. Similar to previous reports correlating gene dosage with clinical outcome, we find that expression of a “KRAS-ON” gene signature, which we identified using our KRAS overexpression model, correlates expression of a basal-like/squamous gene signatures and poor prognosis in patients with PDAC (40). However, increased KRAS signaling is not required for the basal-like/squamous subtypes, as 2 out of 3 of our IGO tumors derived from wild-type KRAS organoids are Fast-progressors (Supplementary Table S2), and prior work has shown that TP63 and GLI2 can dictate a basal-like state in PDAC cells (16, 17). Furthermore, KRAS-addicted cells have been previously characterized as more classical and epithelial in monolayer cell cultures (6, 55), and KRAS ablation was recently reported to induce a basal-like phenotype in surviving cells in vivo (16). Further experiments are warranted in our IGO system to examine how KRAS addiction controls the molecular subtype and invasive properties of PDAC cells. We identified a set of 13 genes that are upregulated in the invasive areas in Slow-progressors derived from IGO and upon increasing KRAS dosage in Slow-progressor organoids. This gene set comprises secreted and membrane-localized proteins, which are involved in regulating cell shape, axon guidance, and ECM remodeling—pathways associated with poor patient survival in genomic studies (56). In addition, we found that a subset of 4 genes (AMIGO2, KLK6, PMEPA1, and TGM2) correlate with poor prognosis in patients with locally advanced PDAC. Among them, KLK6, a secreted protein with trypsin-like serine peptidase function, has been implicated in the invasion of ovarian cancer, melanoma, and KRAS-mutant colon cancer cells, and is associated with poor prognosis in multiple human cancers (48, 57–60). KLK6 has demonstrated enzymatic activity against fibrinogen and collagens I and IV and may represent a targetable vulnerability for treatment of PDAC (61). Future studies will be aimed at investigating the functional role of KLK6 and other KRAS-ON/invasive signature genes in subtype switching and PDAC progression.
The IGO model provides, for the first time, an opportunity to follow the trajectory of human pancreatic cancer cells as they proliferate and migrate throughout the pancreatic ductal system to develop extensive intraductal lesions and invasive carcinoma. The histologic and transcriptional differences found in Slow- and Fast-progressors may be attributed to differences in the progression of tumors secondary to either the evolution of the tumor microenvironment, the acquisition of early catastrophic genetic events, or the gradual accumulation of driver mutations. As such, the IGO model can also serve as a platform for identifying biomarkers for early detection and potential candidates for therapeutic intervention.
Methods
Animal Experiments
All animal experiments were performed according to procedures approved by the Institutional Animal Care and Use Committee at Cold Spring Harbor Laboratory. NOD.Cg-Prkdcscid Il2rgtm1Wjl/SzJ (NSG) mice were bred in-house or purchased from Jackson Laboratories. Mice were enrolled in experimental studies between 3 and 10 weeks of age. For transplant experiments, a single-cell suspension of human patient-derived organoids was generated as described previously (25). For intraductal grafting of organoid (IGO) transplants, 50,000 to 100,000 cells, suspended in 25 μL of PBS, were infused into the pancreatic duct using the modified retrograde infusion technique described previously (30). Mice were anesthetized with 2.5% to 3% (weight/volume) isoflurane by inhalation. The duodenum was rotated and stretched to expose the biliary–pancreatic duct and its junction with the duodenum, which appeared pale. A microclamp (S&T Vascular Clamps 00396-01, F.S.T) was placed on the common bile duct above the branch point of the pancreatic duct, to prevent infusion into the liver. A small hole opposite to the sphincter of Oddi, 1 to 2 mm away from the ampulla of Vater, was made with a 27-gauge needle, and a 27-gauge blunt-ended Hamilton syringe (80508, Hamilton) was passed through the small hole that has been created by the needle in the duodenal wall and then into the sphincter of Oddi. The tip of the catheter was positioned at the origin of the branch connecting the pancreatic duct to the common bile duct. If the anatomy was unclear, the tip of the catheter was placed just distal to the sphincter of Oddi. Another microclamp was placed at the sphincter of Oddi to prevent backflow or leakage during the infusion. Cells were injected by manually depressing the plunger into the syringe for 2 to 3 minutes. The incision point in the duodenum was closed by suturing (6-0 silk, CP Medical Sutures) or using VetClose surgical glue (Henry Schein CA1999). The pancreas and other organs were carefully placed back into the abdomen before suturing the peritoneal cavity and skin incision (separately). Macroscopic fluorescent images were taken using the Leica LMD7000 system. For orthotopically grafted organoid (OGO) transplants, 50,000 cells, suspended in 50 μL of Matrigel (Matrigel, BD Biosciences) diluted 1:1 with cold PBS, were injected into the tail region of the pancreas using insulin syringes (30 gauge) as described previously (25). Mice were sacrificed at humane endpoints. For hF23 and hF27, mice were sacrificed when their IGO counterparts reached humane endpoints because the formation of palpable masses was uncommon following OGO transplant. Engraftment was confirmed histologically. One mg/kg of body weight of tamoxifen was dosed by oral gavage for consecutive 5 days for 4 weeks.
Isolation of Cancer Cells from Xenograft Tumors
The murine pancreas was isolated, minced, and digested for 20 minutes at 37°C in digestion buffer [DMEM, 10% FBS, 1% penicillin, 1% streptomycin, 2.5 mg/mL Collagenase D (Sigma-Aldrich), 0.5 mg/mL Liberase DL (Sigma-Aldrich), 0.2 mg/mL DNase I (Sigma-Aldrich)] with triturating every 5 minutes. After washing with flow buffer (PBS, 2% FBS), cell suspensions were collected and run through 100-μm cell strainers. Red red blood cells were lysed using ACK lysis buffer (Gibco). Cancer cells were enriched with human EPCAM-coated beads (130-061-101, Miltenyi Biotec) according to the manufacturer's instructions. Samples were collected in TRIzol reagent for RNA extraction.
Plasmid Construction and Transfection
For generation of pBOBI-PGK-puro-EF1-LS (luciferase-2A-mStrawberry), the pBOBI-EF1-LS plasmid was a gift from Scott K. Lyons (Cold Spring Harbor Laboratory; ref. 62), and the puromycin resistance cassette was cloned into pBOBI-EF1-LS. p-lenti-PGK-ER-KRASG12V was a gift from Daniel Haber (Massachusetts General Hospital Cancer Center and Harvard Medical School; Addgene plasmid #35635; http://n2t.net/addgene:35635; RRID:Addgene_35635). To generate the p-lenti-PGK-ER-KRASG12D construct, ER–KRASG12D was PCR amplified and empty linker was annealed to this construct. Then, the annealed construct was subcloned into p-lenti-PGK-puro and confirmed using Sanger sequencing. pCAG-Autobow was a gift from Joshua Sanes (Harvard University; Addgene plasmid #45182; http://n2t.net/addgene:45182; RRID:Addgene_45182) For generation of the Autobow vector (pPB-CAG-Autobow-neo), Autobow cDNA from the pCAG-Autobow vector was subcloned into the pPB-CAG-EBNXN (Sanger Institute) vector, and a neomycin resistance cassette was added. Organoids were transfected with pCMV-hyPBase (Sanger Institute) and pPB-CAG-Autobow-neo using Xtremegene9 (Roche), and antibiotic selection with Geneticin (1 mg/mL) was initiated 3 to 5 days post infection.
Organoid Culture, Lentivirus Production, Lentiviral Infection, and 4-OHT Treatment
Establishment and DNA sequencing of human organoids used in this study were reported previously (24). Human organoids were cultured as described previously (24). Briefly, cells were resuspended in 100% growth factor–Reduced Matrigel (Corning) and plated as domes overlaid with Human Complete Feeding Medium (hCPLT): advanced DMEM/F12, 10 mmol/L HEPES, 1× Glutamax, 500 nmol/L A83–01, 50 ng/mL hEGF, 100 ng/mL mNoggin, 100 ng/mL hFGF10, 0.01 mmol/L hGastrin I, 1.25 mmol/L N-acetylcysteine, 10 mmol/L Nicotinamide, 1X B27 supplement, 10% (vol/vol) R-spondin1–conditioned medium, 50% (vol/vol) Afamin/Wnt3A conditioned media (35, 63). No other authentication of organoids was performed except for DNA sequencing. HEK293T cells were cultured in DMEM with 10% FBS and 1% penicillin/streptomycin. All organoid and cell lines were routinely tested for Mycoplasma with MycoAlert Mycoplasma Detection Kit (LT07-318, Lonza) at Cold Spring Harbor Laboratory.
Lentivirus was produced in HEK293T cells by transfecting lentiviral constructs with helper plasmids [pMD2.G (Addgene #12259) and psPAX2 (Addgene # 12260)] using Xtremegene9 (Roche) per manufacterer's instructions. For HEK293T cells in 10-cm dish, 1.25 μg of plasmid DNA, 1.25 μg of pMD2.G, 2.5 μg of psPAX2, and 15 μL of Xtremegene9 were mixed. Lentivirus-containing supernatant was collected at 24, 48, and 72 hours post transfection, filtered, and pooled together. Lenti-X concentrator (Clontech) was used to concentrate the lentivirus per manufacturer's instructions. The viral pellet was resuspended in hCPLT containing 8 μg/mL polybrene. For organoid infections, single-cell suspensions of organoids were generated (24) and were spinoculated with virus at600 relative centrifugal force (RCF) for 30 minutes at room temperature (100,000–250,000 cells). Cells were then incubated at 37°C for 4 to 16 hours, resuspended in Matrigel, and plated under standard organoid culture conditions. Selection was initiated 3 to 5 days post infection. Organoids were treated with Geneticin (1 mg/mL, Life technologies) or puromycin (2–3 μg/mL, Sigma-Aldrich) for selection. To induce KRAS hyperactivation, ER–KRASG12D organoids were treated with 4-OHT (0.1 μmol/L, Sigma-Aldrich) for 48 hours, and untreated organoids were used as a control.
Proliferation Assays
For proliferation assays, organoids were dissociated as described previously (24), and 400 single cells were plated per well in black clear-bottom 384-well plates in 30 μL of 10% Matrigel in hCPLT media. Proliferation was measured every 24 hours for 5 to 6 days using CellTiter-Glo (G7573, Promega) according to the manufacturer's instructions.
Histopathology, IHC, and Immunofluorescence
Formalin-fixed, paraffin-embedded (FFPE) blocks were sectioned and deparaffinized. Antigen retrieval was performed in Citrate buffer, pH 6.0 for 6 minutes in a pressure cooker unless otherwise specified, then rinsed in tap water for 10 minutes. For IHC staining, endogenous peroxidase was quenched in 3% hydrogen peroxide in H2O for 15 minutes at room temperature. The slides were then rinsed in 1× TBS + 0.1% Tween-20 (TBST). Slides were blocked in 5% Horse serum for 1 hour at room temperature and incubated in primary antibody at 4°C overnight. Slides were washed three times in TBST, then incubated in either horseradish peroxidase–conjugated or alkaline phosphatase–conjugated secondary antibodies for 30 minutes (for IHC), or fluorescent secondary antibodies for 1 hour at room temperature (for immunofluorescence staining). For IHC, slides were washed twice in TBST and once in H2O. ImmPACT DAB (Vector) or ImmPRESS-AP (Vector) was added for 1 to 5 minutes followed by an H2O wash, and slides were incubated in hematoxylin counterstain for 20 seconds. Slides were then dehydrated, and a coverslip was added using Cytoseal 60 adhesive. Brightfield images were obtained with an Axio Imager.A2 (ZEISS; Improvision). For IF, slides were washed three times in PBS. Slides were counterstained with DAPI (D8417, Sigma-Aldrich), mounted in Prolong Gold Antifade Reagent (P10144, Invitrogen), and imaged on a Leica TCS SP8 laser scanning confocal (Boulder Grove Il), controlled by the LAS AF 3.3.10134 software. This confocal was mounted on a DMI 6000 CS inverted microscope, equipped with 4 laser lines and 2 photomultiplier tubes.
Primary antibodies used for IHC were: pERK (4370, Cell Signaling Technology), TP63 (39692, Cell Signaling Technology), S100A2 (ab109484, Abcam), GATA6 (AF1700-SP, R&D Systems), AGR2 (13062, Cell Signaling Technology), and αSMA (M0851, Dako). Primary antibodies used for IF were: human mitochondria (MAB1273, Millipore), CK19 (TROMA III, DSHB), Ki-67 (RM-9106, Thermo Fisher Scientific), VIM (5741, Cell Signaling Technology), GFP (ab13970, Abcam), YFP (EVN-AB606-C200, Axxora), RFP (EMU107, Kerefast), and F-actin (A12379, Invitrogen). Secondary antibodies used for IF were anti-mouse Alexa Fluor 488 (A21202, Thermo Fisher Scientific), anti-rabbit Alexa Fluor 647 (A21244), anti-rat Alexa Fluor 568 (A11077), anti-chicken Alexa Fluor 488 (A11039), anti-rabbit Alexa Fluor 546 (A11035), and anti-guinea pig Alexa Fluor 647(A21450).
For sequential IHC, DAB (SK-4105, Vector Laboratories) and Red Alkaline Phosphatase substrate (SK-5105, Vector Laboratories) were used for different primary antibodies. Hematoxylin was used as nuclear counterstain. Hematoxylin and eosin (H&E), Masson trichrome, and Alcian Blue staining were performed according to standard protocols.
Whole-Mount Immunofluorescence of Organoid Cultures
Organoids were plated in small (25 μL) Matrigel domes in an 8-well chamber slide (Ibidi #80826) and allowed to grow to 80% confluence. Organoids were fixed with 4% paraformaldehyde (PFA) at room temperature for 20 minutes. After washing with PBS/Glycine buffer (0.75% Glycine in PBS, pH 7.4) three times for 10 minutes each and washing with IF buffer (0.1% BSA, 0.05% NaN3, 0.2% Triton-X, 0.05% Tween-20 in PBS, pH 7.4) three times for 10 minutes each, samples were incubated in blocking buffer (10% goat serum in IF buffer) at room temperature for 1 hour. Samples were incubated with primary antibodies diluted in blocking buffer at 4°C overnight, followed by washing with IF buffer three times at room temperature for 10 minutes each. Samples were incubated with secondary antibodies diluted in blocking buffer at room temperature for 1 hour. Slides were DAPI counterstained and mounted in Prolong Gold antifade reagent (P10144, Invitrogen). Organoids were imaged on a Leica TCS SP8 laser scanning confocal as described above.
Three-Dimensional IF Staining
A modified FLASH method (64) was used. The entire pancreas was divided into 3 to 4 pieces less than 4 to 5 mm in a diameter and fixed in 4% PFA overnight at 4°C. The PFA-fixed tissues were then washed two times with PBS-T [0.4% Triton X-100 (Sigma-Aldrich) in PBS] at room temperature for 1 hour each, and incubated in 200 mmol/L boric acid (Sigma-Aldrich) with 4% SDS (Sigma-Aldrich) pH 7.0 overnight at 54°C. For immunostaining, tissues were blocked in FLASH blocking buffer [1% BSA (Sigma-Aldrich), 5% DMSO (Sigma-Aldrich), 10% FBS (Gibco), 0.02% sodium azide (Sigma-Aldrich), 0.2% Triton X-100 in PBS] for 1 hour, before incubation in primary antibodies for 48 hours at room temperature. Secondary antibodies were incubated for 48 hours at room temperature. Cleared tissues were then prepared using a tissue-clearing kit (ab243298, Abcam) according to the manufacturer's instructions and imaged using a Z-stack scan on a Leica TCS SP8 laser scanning confocal (Boulder Grove Il), controlled by the LAS AF 3.3.10134 software.
Histologic Quantification
Slides were scanned using an Aperio CS2 scanner (Leica). Tumor and nontumor areas on H&E-stained slides were manually annotated and were measured using ImageJ software. The number of Ki-67–positive and total nuclei of tumor cells in intraductal area and the tumor cells adjacent to murine duct cells in Fig. 1I and J were counted manually. The number of Ki-67–positive and total nuclei in organoids in Supplementary Figs. S3C and S7D and in invasive and intraductal areas of IGO and OGO tumors in Supplementary Fig. S1I and S1J were counted automatically using ImageJ software in at least three representative high-powered fields. To quantify AGR2, S100A2, and pERK IHC, the percentage of strong positive pixels was counted relative to the entire section using ImageJ software.
Protein Isolation and Immunoblot Analysis
Organoids were harvested in Cell Recovery Solution (Corning), incubated for 15 minutes on ice, pelleted, and washed twice in ice-cold PBS. Resulting cell pellets were lysed in 1% Triton X-100, 50 mmol/L Tris-HCl pH 7.5, 150 mmol/L NaCl, 5 mmol/L EDTA with 1× protease (11836170001, Roche), and phosphatase inhibitor cocktails (4906845001, Roche) on ice for 30 minutes. The lysate was clarified at 16,000 RCF for 10 minutes at 4°C, and aliquoted into new tubes for storage at −80°C.
Protein lysates were separated on 4% to 12% Bis-Tris NuPAGE gels (Life Technologies) run with MOPS Running Buffer (Life Technologies), transferred to polyvinylidene difluoride membrane (Millipore), and blocked in 5% BSA (Sigma-Aldrich) in TBST for 1 hour. The membranes were then incubated in primary antibody overnight with rocking at 4°C followed by incubation in appropriate secondary antibodies conjugated to horseradish peroxidase for 1 hour and detection by enhanced chemiluminescence (GE Healthcare).
Primary antibodies used were: TP63 (39692, Cell Signaling Technology), S100A2 (ab109484, Abcam), GATA6 (AF1700-SP, R&D Systems), HSP90α (07-2174, EMD Millipore), SMAD4 (sc-7966, Santa Cruz Biotechnology), pERK1/2 (4370, Cell Signaling Technology), and ERK1/2 (4695, Cell Signaling Technology).
Genomic DNA Isolation from Organoid
Organoids were harvested in Cell Recovery Solution (Corning), incubated for 15 minutes on ice, pelleted, and washed twice in ice-cold PBS. Genomic DNA from organoids was collected using DNeasy Blood & Tissue Kit (Qiagen) and was sonicated using a Covaris Focus Acoustics System. Libraries were prepared using the NEBNext Ultra II DNA Library Prep kit (NEB). Libraries were evaluated using BioAnalyzer (Agilent 2100) and sequenced using the Illumina NextSeq platform (Cold Spring Harbor Laboratory Genome Center).
Genomic DNA Isolation from FFPE Sections of IGO Tumors
FFPE blocks were transferred to PEN membrane glass slides. Adjacent sections were stained with H&E, S100A2, and GATA6 for pathology review. Laser capture microdissection was performed on a LMD7000 system (Leica) to isolate intraductal lesions with GATA6 staining and invasive lesions with S100A2 staining. DNA was extracted using Arcturus PicoPure DNA Extraction Kit (Thermo Fisher Scientific) and was sonicated using a Covaris focus acoustics system. Libraries were prepared using the NEBNext Ultra II DNA Library Prep Kit (NEB). Libraries were evaluated using BioAnalyzer (Agilent 2100) and sequenced using Illumina NextSeq platform (Cold Spring Harbor Laboratory Genome Center).
KRAS Copy-Number Analysis Based on Organoid Exome Sequencing
In addition, we made use of copy-number analysis performed using whole-exome sequencing data from organoids as described previously (24). Whole-exome sequencing data for organoids were processed by the New York Genome Center somatic preprocessing pipeline. Copy-number variants were detected using FACETS v0.5.2 (65) for exome, resulting in segmented profiles where the copy number is approximated by a piecewise-constant function of the genomic position.
Copy-Number Analysis of Bulk DNA from Organoids and FFPE Sections
The copy-number analysis was performed as described previously (66). Briefly, the genome was divided into either 5,000 or 20,000 bins. After the normalization, the normalized bincount vectors were segmented using Bioconductor DNAcopy package version 1.50.5 as described previously (66). DNAcopy parameters used were alpha = 0.02, nperm = 1000, undo.SD = 0.5, and min.width = 3. Copy number heat maps were made using the heat map function in R using the segmented bin values. The distance function used was “manhattan,” and the hierarchical clustering agglomeration method used was “ward.D2.”
RNA Extraction and Quantitative RT-PCR
RNA was extracted from samples in TRIzol Reagent (15596-018, Thermo Fisher Scientific) using the PureLink RNA Mini Kit (12183018A, Thermo Fisher Scientific) per the manufacturer's instructions. cDNA was synthesized using 1 μg of total RNA and TaqMan Reverse Transcription Reagents (N808-0234, Applied Biosystems). qRT-PCR analysis was performed with gene-specific Taqman primers (Supplementary Table S4; Thermo Fisher Scientific) and Taqman Universal Master Mix II, no UNG (Thermo Fisher Scientific) on a QuantStudio 6-Flex Real time-PCR instrument (Applied Biosystems). Relative gene expression quantification was performed using the ΔΔCt method with the QuantStudio Real-Time PCR software v1.1 (Applied Biosystems). Expression levels were normalized to HPRT or ACTB.
Tumor RNA-seq Experiments
For RNA-seq comparison of OGO- to IGO-derived tumors, OGO and IGO transplants of hM1A, hF2, hF3, hT3, hF23, hF24, hT30, hF32, hF43, hT85, hT93, hT102, hM17D, and hM19B were performed. Transplant-derived tumors were harvested at endpoint. For RNA-seq comparison of Fast versus Slow IGO tumors, 6 IGO transplants of Fast-progressors (hM1A, hF3, hT3, hF24, hF43, and hM19B) and 9 IGO transplants of Slow-progressors (hF2, hF23, hF27, hT30, hF32, hT85, hT93, hT102, and hM17D) were performed, and transplant-derived tumors were harvested at endpoint. RNA from bulk tumors was extracted using the PureLink RNA Mini Kit with TRIzol Reagent as described above.
RNA-seq libraries were constructed using a TruSeq Stranded Total RNA Kit with RiboZero Human/Mouse/Rat (RS-122-2202, Illumina) with 0.2 to 1 μg RNA per sample, per manufacturer's instructions. Libraries were evaluated using BioAnalyzer (Agilent 2100) and then sequenced using the Illumina NextSeq platform (Cold Spring Harbor Laboratory Genome Center).
RNA-seq of Laser Capture Microdissected IGO Tumors
FFPE blocks were transferred to PEN membrane glass slides. Adjacent sections were stained with H&E and S100A2 for pathology review. Laser capture microdissection was performed on a LMD7000 system (Leica) to isolate regions containing cancer cells within the duct (intraductal, “INTRA”) and outside the duct (invasive, “INVASIVE”) from IGO transplants of hM1A clonal organoids, cancer cells within the duct (intraductal) and outside the duct (invasive) from IGO transplants of hF2 and hF27, and cancer cells from IGO transplants of Empty vector (Empty) and ER–KRASG12D (ER–KRAS)-expressing hF23 and hF27. RNA was extracted using NucleoSpin totalRNA FFPE XS (Clontech), and libraries were prepared using the SMARTer Stranded Total RNA-Seq Kit v2 - Pico Input Mammalian (Takara). Libraries were evaluated using BioAnalyzer (Agilent 2100) and sequenced using Illumina NextSeq platform (Cold Spring Harbor Laboratory Genome Center).
Analysis of Bulk Tumor, Laser Capture Microdissected Tumor, and Organoid RNA-seq Data
The quality of the RNA-seq reads was quantified using FastQC version 0.11.5 (Babraham Bioinformatics). Next, to classify reads as coming from the mouse stroma or the human cancer cells, we employed the strategy presented by Callari and colleagues (32). A custom reference genome composed of the GRCm38.p6 mouse genome and GRCh38.p12 human genome was generated, and a combined human and mouse transcriptome annotation composed of GENCODE 28 and M17 (67), respectively, were generated. Reads were aligned against the composite genome using STAR version 2.5.2b (68) by RSEM software version 1.3.0 (69) with the following parameters: rsem-calculate-expression –star. Counts per gene were extracted, using the combined human and mouse transcriptome annotation. For the organoid RNA-seq, the alignment was done on the GRCh38.p12 human genome, and counts per gene were extracted from human transcriptome using STAR and RSEM as described above. Differential gene expression analysis was performed using Bioconductor package DESeq2 version 1.24.0 (70), with a prefiltering step to retain genes with reads in more than one sample. Human and mouse transcriptomes were split to be analyzed separately. In presence of paired data, the DESeq2 design was adapted consistently. Only genes with an adjusted P value < 0.05 and an absolute log2 fold change ≥ 1 were retained as significantly differentially expressed.
The Bioconductor package ComplexHeatmap version 2.0.0 (71) was used to generate heat maps of the DeSeq2-normalized counts shown in Fig. 3A, standardized by gene, of the significantly differentially expressed genes. Hierarchical clustering of the genes was based on the complete linkage method using the Spearman correlation coefficient. For the clustering of the samples, Euclidian distance was used.
GSEA was performed using the GSEA program version 4.0.3 (Broad Institute), with ranked log2 fold change values as input. Hallmark gene sets and C6 oncogenic signatures were downloaded from the Molecular Signatures Database (MSigDB; http://www.broadinstitute.org/gsea/index.jsp). Squamous-PDAC and Progenitor-PDAC Identity signatures (17), basal-like, classical, Moffitt-Normal, and Moffitt-Activated signatures (7), Immune-rich and ECM-rich signatures (9), and the JANES-KRAS signature (41) were described previously.
For the heat map showing transcriptional subtypes of IGO and OGO tumors in Fig. 2H, consensus clustering was performed on tumor expression signatures using the top 100 ranked genes from the four malignancy-related NMF components (19) using ConsensusClusterPlus v.1.40.0 (72). Using Pearson correlation distance, clustering was performed with Pearson correlation distance (partitioning around medoids) with 10,000 resamplings to generate the final consensus.
For the heat maps showing expression of genes in tumors taken from PanCuRx patients (i.e., Supplementary Fig. S3H), heat maps of gene expression (log10(FPKM+1)) were generated with the clustermap function in the Python data visualization library Seaborn version 0.9.0 (https://github.com/mwaskom/seaborn/releases/tag/v0.9.0). Each row shows the gene expression scaled to have zero mean and unit variance. Scaled values were clipped to be within the range [−5, 5] for visualization. Hierarchical clustering was performed using an average-linkage method using Pearson correlation coefficients between each patient or gene signature.
Imputation of Survival Time for Censored Mice
For mice in the survival study that were censored (because the mouse was sacrificed prior to humane endpoint or was still alive at the last observed time), the survival time was imputed by using the median of the last observed survival times for all mice (transplanted with any organoid) that survived as long as or longer than the censored mouse. The survival plot in Fig. 1F was generated with package ggplot2 version 3.1.0 (73).
Classification of Fast- and Slow-Progressors
Supplementary Figure SM1 displays the workflow of the classification method. For the purpose of determining the Fast- and Slow-progressor classification, when multiple organoid lines from a single patient were used in the study (e.g., hM1A, hM1E, and hM1F; or hM19A, hM19B, hM19C, and hM19D) only a single organoid line per patient (hM1A and hM19B) was retained. For each organoid line, three survival times (corresponding to the number of times available in the group with the smallest number of replicates) were randomly selected. If censoring was present, the imputed survival time was used. For each organoid line, the mean and the SD of the three sampled survival times were used to estimate a normal distribution of survival times, and 10 data points (survival times) were randomly selected from the distribution. All selected survival times from all organoid lines were pooled together and used to generate a global distribution of the survival times. The package mixtools version 1.1.0 (74) was then used to detect the two normal curves that best approximated the global distribution of the survival times. The mean (mean survival of 75 days) was associated with the Fast-progressors, whereas other curve (mean survival of 263 days) was associated with the Slow-progressors. To determine the upper threshold in survival times for the Fast-progressors, we selected the position in the right tail of the first curve where the cumulative normal probability becomes 95% (111 days). To determine the lower threshold in survival times for the Slow-progressors, we selected the position in the left tail of the second curve where the cumulative normal probability becomes 5% (157 days). The area between those two positions is considered unclassified.
Survival Analysis of PanCuRx Data
KRAS-ON (in vitro) score and KRAS-ON (in vivo) score were obtained using ssGSEA through GenePattern web tool (https://cloud.genepattern.org/gp/pages/login.jsf). GraphPad Prism software was used for Log-rank (Mantel–Cox) test.
Data Availability
The RNA- and DNA-sequencing data discussed in this publication have been deposited in NCBI's Genotypes and Phenotypes database and are accessible through dbGaP study accession number: phs002045.v1.p1.
Statistical Analysis
Statistical analysis was performed using GraphPad Prism software using unpaired Student t test and Log-rank (Mantel–Cox) test for survival data unless otherwise indicated in figure legend. P values <0.05 were determined to be significant. Asterisks denote P values as follows: *, P < 0.05; **, P < 0.01; ***, P < 0.001.
Disclosure of Potential Conflicts of Interest
D. Plenker reports grants from German Research Foundation (DFG) (PL894/1-1) outside the submitted work. R.H. Hruban reports a patent for GNAS licensed to Thrive Earlier Detection (R.H. Hruban has the right to receive royalty payments from Thrive Earlier Detection for the GNAS invention in a relationship supervised by Johns Hopkins University). J. Gillis reports grants from NIH (R01LM012736) and grants from NIH (R01MH113005) during the conduct of the study. C.R. Vakoc reports grants from Boehringer Ingelheim, personal fees from KSQ Therapeutics, and personal fees from Third Rock Ventures outside the submitted work. D.A. Tuveson reports grants from NIH-NCI, Lustgarten Foundation, Simons Foundation, V Foundation, Pershing Square Sohn Foundation, Thompson Family Foundation, PANCAN, DFG, BCRF, and NY State during the conduct of the study, and ONO Therapeutics outside the submitted work; personal fees from Leap Therapeutics, Surface Oncology, Cygnal Therapeutics, and Mestag Therapeutics; is an SAB member for Leap Therapeutics, Surface Oncology, Cygnal Therapeutics, and Mestag Therapeutics, and has Stock in Leap Therapeutics, Surface Oncology, and Cygnal Therapeutics. No potential conflicts of interest were disclosed by the other authors.
Authors' Contributions
K. Miyabayashi: Conceptualization, resources, data curation, funding acquisition, validation, investigation, visualization, methodology, writing-original draft. L.A. Baker: Data curation, funding acquisition, validation, investigation, visualization, project administration, writing-review and editing. A. Deschenes: Data curation, software, formal analysis, validation, investigation, visualization, methodology, writing-original draft, writing-review and editing. B. Traub: Funding acquisition, validation, investigation, writing-review and editing. G. Caligiuri: Validation, investigation. D. Plenker: Resources, funding acquisition, validation, investigation, writing-review and editing. B. Alagesan: Data curation, funding acquisition, validation, investigation, visualization, writing-original draft, writing-review and editing. P. Belleau: Data curation, software, formal analysis, validation, visualization, methodology. S. Li: Data curation, software, formal analysis, validation, investigation, visualization, methodology, writing-review and editing. J. Kendall: Data curation, software, formal analysis, validation, investigation, visualization, methodology, writing-review and editing. G.H. Jang: Data curation, software, formal analysis, validation, investigation, visualization, methodology. R.K. Kawaguchi: Data curation, software, formal analysis, validation, investigation, visualization, methodology. T.D.D. Somerville: Resources, data curation, funding acquisition, validation, investigation, writing-review and editing. H. Tiriac: Resources, data curation, validation. C.-I. Hwang: Funding acquisition, validation, writing-review and editing. R.A. Burkhart:Resources, funding acquisition, validation, writing-review and editing. N.J. Roberts: Resources, validation. L.D. Wood: Resources, funding acquisition, validation, writing-review and editing. R.H. Hruban: Resources, supervision, funding acquisition, validation, writing-review and editing. J. Gillis: Data curation, software, formal analysis, supervision. A. Krasnitz: Data curation, software, formal analysis, supervision, validation, methodology. C.R. Vakoc: Resources, supervision, funding acquisition, validation, writing-review and editing. M. Wigler: Data curation, software, formal analysis, supervision, funding acquisition, validation. F. Notta: Resources, data curation, software, formal analysis, supervision, funding acquisition, validation, investigation, methodology, writing-review and editing. S. Gallinger: Resources, supervision, funding acquisition. Y. Park: Data curation, funding acquisition, validation, investigation, visualization, writing-original draft, project administration, writing-review and editing. D.A. Tuveson: Conceptualization, supervision, funding acquisition, validation, project administration, writing-review and editing.
Acknowledgments
We thank Dr. Chen-Hua Chuang and Dr. Monte M. Winslow for the instruction in retrograde pancreatic intraductal injection. We acknowledge the Cold Spring Harbor Laboratory Animal, Animal Imaging, Histology, Microscopy, and Next Generation Sequencing Shared Resources, which are supported by the NIH Cancer Center Support Grant P30CA045508. We acknowledge Ela Elyada for the construction of pBOBI-PGK-puro-EF1-LS plasmid. We also acknowledge Vincenzo Corbo for generation of mStrawberry-expressing organoids. We thank members of the Tuveson Laboratory for their assistance and advice. D.A. Tuveson is a distinguished scholar of the Lustgarten Foundation and Director of the Lustgarten Foundation-designated Laboratory of Pancreatic Cancer Research. D.A. Tuveson is also supported by the Cold Spring Harbor Laboratory Association, the V Foundation, the Thompson Foundation, and the NIH (P30CA45508, P20CA192996, U10CA180944, U01CA224013, U01CA210240, R01CA188134, and R01CA190092). D.A. Tuveson and C.R. Vakoc are supported by the Simons Foundation (552716) and the NIH R01CA229699. Y. Park is supported by NCI R50CA211506. B. Alagesan is supported by NCI F30CA200240. C.R. Vakoc was supported by Pershing Square Sohn Cancer Research Alliance, the Cold Spring Harbor Laboratory and Northwell Health Affiliation, the NCI 5P01CA013106-Project 4 and 1RO1CA229699, and the 2016 Pancreatic Cancer Action Network-American Association for Cancer Research Career Development Award, supported from the Daniel and Janet Mordecai Foundation, Grant number 16-20-25-VAKO. K. Miyabayashi is supported by the Daiichi Sankyo Foundation of Life Science and the Uehara Memorial Foundation, Japan. B. Traub is supported by the German Research Foundation (DFG; TR 1663/1). D. Plenker is supported by the German Research Foundation (DFG; PL 894/1-1). T.D.D. Somerville was supported by a grant from the State of New York, contract no. C150158. L.A. Baker was supported through the NIH (F32CA192904). C.-I. Hwang was supported by NIH F32CA180717. L.D. Wood is supported by the Sol Goldman Pancreatic Cancer Research Center. This work was supported by grants to M. Wigler from the Simons Foundation, Life Sciences Founders Directed Giving-Research (award numbers 519054), and The Breast Cancer Research Foundation (award number 18-174). The pancreatic cancer patient study was conducted with the support of the Ontario Institute for Cancer Research (PanCuRx Translational Research Initiative) through funding provided by the Government of Ontario. This work was also supported by the Cold Spring Harbor Laboratory and Northwell Health Affiliation (Project Lazarus) as well as the Northwell Health Tissue Donation Program.