Reconstituting development of pancreatic intraepithelial neoplasia from primary human pancreas duct cells

Development of systems that reconstitute hallmark features of human pancreatic intraepithelial neoplasia (PanINs), the precursor to pancreatic ductal adenocarcinoma, could generate new strategies for early diagnosis and intervention. However, human cell-based PanIN models with defined mutations are unavailable. Here, we report that genetic modification of primary human pancreatic cells leads to development of lesions resembling native human PanINs. Primary human pancreas duct cells harbouring oncogenic KRAS and induced mutations in CDKN2A, SMAD4 and TP53 expand in vitro as epithelial spheres. After pancreatic transplantation, mutant clones form lesions histologically similar to native PanINs, including prominent stromal responses. Gene expression profiling reveals molecular similarities of mutant clones with native PanINs, and identifies potential PanIN biomarker candidates including Neuromedin U, a circulating peptide hormone. Prospective reconstitution of human PanIN development from primary cells provides experimental opportunities to investigate pancreas cancer development, progression and early-stage detection.

P ancreatic ductal adenocarcinoma (PDA) typically presents at late stages with dismal overall survival. By contrast, fortuitous detection of early-stage disease localized to the pancreas can lead to curative treatment. Based on retrospective analysis of human tissue samples, the investigators postulate that a series of genomic mutations accumulate in pancreatic exocrine cells leading to dysplastic lesions called pancreatic intraepithelial neoplasia, PanIN1 and PanIN2, then PanIN3 (carcinoma in situ) before progressing to invasive PDA 1,2 . Among mutated genes, KRAS has been most closely associated with PDA and its precursors, with over 90% of PanINs and PDAs harbouring oncogenic KRAS mutations 3 . Loss-of-function mutations at high prevalence in 'tumour suppressors' encoded by CDKN2A (90-95%), SMAD4 (49-55%) and TP53 (50-84%) are coupled to protein loss and also tightly linked to PDA formation 4,5 . In human PDA, mutations in only one or two of these genes is infrequent; more commonly, three or four mutations are found in combination 4 . This suggests that multiple genomic alterations are required to initiate PDA development or progression. Collectively, mutations in KRAS, CDKN2A, SMAD4 and TP53 have been dubbed 'driver mutations' for human PDA formation 4,6 .
Findings from genetically engineered mouse models (GEMM) support this genetic PDA progression model. These findings include the observation that expression of oncogenic Kras alleles is sufficient to induce development of PanIN-like lesions in GEMM 7 and, depending on the developmental stage of Kras induction, to induce invasive PDA. The frequency and severity of invasive phenotypes can be increased in these genetic mouse models when oncogenic Kras expression is combined with other driver mutations or with experimental pancreatitis 8 . Despite impressive advances in genetically engineered mouse models of PDA development, there is no evidence that healthy human pancreatic cells can form PanIN or invasive PDA when similar driver mutations are introduced. Given the translational value of human PDA modelling, several groups attempted to generate human PanIN or PDA models using various cell sources such as an immortalized human ductal cell line 9 , human embryonic stem cells 10 or induced pluripotent stem cells 11 , or organoids derived from PDA patients 12 . However, none of these prior examples systematically introduced driver mutations in human pancreatic exocrine cells from healthy donors and reconstituted the features of human PanIN or PDA.
Here we report that recapitulating driver mutations in primary human pancreatic ductal cells reconstitutes development of lesions resembling PanINs. Lentiviral gene delivery combined with CRISPR-Cas9 genome-editing systems achieves permanent alterations in KRAS, CDKN2A, SMAD4 and TP53 in primary human duct cells. Cloned immortalized cells grow as epithelial monolayer spheres in three-dimensional culture. On orthotopic transplantation into adult mouse pancreas, these cells form structures with multiple cellular and molecular features of PanINs that do not progress after 6 months to invasive PDA. Thus, we generated a unique system to develop stable human PanIN-like lesions prospectively from healthy human pancreatic ductal cells. This will provide a robust experimental system for investigation of developmental, genetic and signalling mechanisms underlying formation of PanINs from healthy human duct cells.

Results
Genetic modification of purified primary human duct cells. To investigate whether the genetic and cellular hallmarks of human PanIN development can be reconstituted in purified normal human pancreas cells, we used FACS to isolate pancreatic exocrine cells from human cadaveric donors without known pancreatic diseases ( Fig. 1a and Supplementary Table 1) 13 . FACS fractionation using CD133 antibody separated CD133 þ cells expressing ductal markers like KRT19 and CAR2, and CD133 À cells that include acinar and endocrine cells (Fig. 1b) 13 . We were unable to expand CD133 À cells in vitro (Fig. 1d, 'CD133 À '), precluding studies of PanIN or PDA development from human acinar cells. By contrast, duct cells survived and expanded as monolayer epithelial spheres cultured in Matrigel in a defined medium without serum or feeder cells up to 40 days, and then ceased growth (Fig. 1h) 13 . We found that purified CD133 þ cells could be infected with lentivirus harbouring the gene encoding the fluorescent protein H2B-mCherry, and genes conferring resistance to the drugs puromycin or neomcyin (Supplementary Fig. 1A and Methods section), indicating that genetic modification of primary duct cells using lentiviral methods is feasible.
Oncogenic Kras activation alone in pancreatic epithelial cells is sufficient to initiate PDA development in mice. However, Kras activation alone infrequently leads to development of invasive PDA, while combination with mutations in Cdkn2a, Smad4 or Trp53 can enhance the speed or frequency of progression to invasive disease in GEMM 8 . Moreover, it is not known whether such genetic changes are also sufficient to induce PDA development in human pancreatic exocrine cells. To study this question, we constructed lentiviruses expressing oncogenic KRAS G12V and lentiCRISPRv2 viruses (see Methods section) encoding the Cas9 nuclease and single guide RNA (sgRNAs) against the genomic loci for CDKN2A, SMAD4 and TP53 (KCST viruses, where K ¼ KRAS, C ¼ CDKN2A, S¼ SMAD4 and T ¼ TP53; Fig. 1c). Purified CD133 þ cells were infected with (1) control lentivirus producing H2B-mCherry and a neomycin resistance gene (Control-NeoR; CTRL), (2) virus expressing KRAS G12V and NeoR (KRAS-NeoR; KRAS), (3) a mixture of control viruses (Control-NeoR and lentiCRISPRv2-Control; CTRLmix) or (4) a combination of KCST viruses, and then grown in sphere cultures (Fig. 1d). Concurrently, CD133 À cells were also infected with KCST viruses. All the infection combinations ((1)-(4) above) of CD133 þ cells grew as spheres in 2 weeks and continued to grow up to 4 weeks after passaging; by contrast, CD133 À cells, including acinar cells 13 , failed to grow and expand (Fig. 1d,h). Genomic DNA PCR and quantitative real-time PCR (qRT-PCR) with the KCST virus-infected spheres confirmed the presence of all four transgenes (Fig. 1e) and mRNA expression of the oncogenic KRAS G12V (Fig. 1f). PCR amplification and DNA sequencing of the target genomic regions, followed by Tracking of Indels by DEcomposition (TIDE) analysis 14 , revealed high-efficiency genome editing with our lentiCRISPR reagents in CDKN2A, SMAD4 and TP53 loci, evidenced by a high prevalence of insertion or deletion (indel) mutations (49.5-86.2%; Fig. 1g and Supplementary Fig. 1D). Thus, our approach induced genetic and targeted genomic modifications of four PDA-associated 'driver' genes. We observed exponential growth of the KCST primary spheres over 5 months, while the spheres transduced with lentiviruses encoding Control-NeoR, KRAS-NeoR-Neo alone or a mixture of control viruses (CTRLmix) failed to expand beyond 30 days after infection (Fig. 1h). Haematoxylin-eosin staining of growing KCST spheres showed epithelial monolayers composed of cuboidal cells ( Fig. 1i and Supplementary Fig. 1C). The cytoplasm of these cells failed to stain with Alcian blue, which detects acid mucin production, a characteristic feature of PanINs ( Supplementary Fig. 1C). These data indicate that our lentiviral reagents efficiently induce genetic and genomic modifications of the PDA-associated genes (KCST) in purified normal human primary pancreatic ductal cells, and induce their immortalization.   Table 2). Many lesions comprised tall columnar-shaped cells with basally located nuclei and mucinous cytoplasm, the typical features 1 of human PanIN1 (Fig. 2c,d,g,h). Alcian blue staining confirmed the presence of acid mucins (Fig. 2e,i). Activated oncogenic KRAS is associated with increased phospho-extracellular signal-regulated kinase (ERK), and immunohistochemistry detected increased phospho-ERK in these lesions at levels and in patterns comparable to those in native human PanINs (Fig. 2f,j and Supplementary  Fig. 3) 15 . By contrast, little to no phospho-AKT was detected in PanIN-like lesions ( Supplementary Fig. 3), consistent with prior reports that native human PanIN lesions show weak or non-detectable phospho-AKT immunoreactivity 16 (Fig. 3b,c). TIDE analysis revealed high-efficiency genome editing using our lentiCRISPR reagents (66.9-93.9%; Fig. 3d and Supplementary Fig. 2A and B). Similar to KCSTtransduced spheres, KECST-transduced spheres also grew exponentially over 6 months (Fig. 3e), indicating immortalization of the transduced KECST spheres. H&E staining revealed that monolayer cuboidal cells comprise growing KECST spheres, similar to features found in KCST spheres (Fig. 3f). After orthotopic transplantation of KECST spheres (n ¼ 9; Supplementary  phospho-ERK signals in these lesions, consistent with activation of KRAS (Fig. 4l,p and Supplementary Fig. 3). In addition, we found that these lesions maintained characteristic features of ducts, including production of cytokeratin 19 (CK19) detected by immunolabelling (Fig. 4e,f). We also confirmed that these lesions were derived from transduced human spheres by positive mCherry fluorescence and immunodetection of human nuclear antigen and human mitochondria ( Fig. 4g Table 3). In addition, we generated a clone with ERBB2 by infecting S1 KCST clone4 with ERBB2encoding lentivirus (Fig. 3a) to generate S1 KECST clone4 ( Supplementary Fig. 9C  After orthotopic transplantation, we observed development of PanIN-like lesions in cloned hiPanIN cell lines with KCST and KECST genotypes-S1 KCST clone3 , S1 KCST clone4 , S1 KECST clone4 and S2 KCST clone8 (Table 1, Fig. 5d,e and Supplementary Figs 9 and 10). For each of these clones, orthotopic transplantation produced lesions with features characteristic of 'late'-stage PanIN2 and PanIN3 lesions, including prominent nuclear abnormalities, mitotic figures, cribriforming, budding of small clusters of epithelial cells into the lumen or luminal necrosis (Fig. 5d,e and Supplementary Fig. 9A,B) 1 . By contrast, we observed development of normal duct-like structures in mice transplanted with S2 KCT clone3 cells (Table 1 and Supplementary Fig. 7C), suggesting an essential role for SMAD4 loss in hiPanIN development. Up to 4 months after transplantation, we did not observe evidence of invasive or metastatic PDA in any case. Thus, our studies reveal the potential of cloned hiPanIN cell lines to develop into stable lesions with characteristic features of early-stage PDA, including PanIN2 and PanIN3. In addition, we demonstrate that oncogenic KRAS G12V expression together with null mutations in CDKN2A, SMAD4 and TP53 in previously healthy human ductal cells is sufficient to produce lesions resembling human PanINs, but not invasive PDA within the framework of our experiments.    9 . We used the same lentiviral and lentiCRISPR reagents used for generating hiPanIN cells (Figs 1c and 3a) to generate HPDE cells harbouring genomic mutations in CDKN2A, SMAD4 and TP53 and expressing KRAS G12V and ERBB2 (HPDE KECST ), confirmed by genomic DNA PCR and qRT-PCR (Fig. 6a,b). TIDE analysis revealed relatively lower genome-editing efficiency than in primary pancreas duct cells (19.1-48.2%; Fig. 6c and Supplementary  Fig. 15A). After transplantation in NSG mice, dispersed HPDE control cells failed to grow (n ¼ 2). By contrast, all mice grafted with HPDE KECST cells developed multiple heterogeneous nodules in the pancreas within 8 weeks (Supplementary Table 5 and Fig. 6d). Histological analysis of the nodules revealed complex, moderate to poorly differentiated or poorly differentiated adenocarcinoma (Fig. 6d). Immunohistochemical analysis confirmed that tumours maintained expression of the ductal marker CK19 (Fig. 6d, CK19, green). Moreover, lung nodules were CK19 þ , consistent with a metastatic HPDE KECST origin (Fig. 6e). Similarly, we found the tumour development 8 weeks after orthotopic transplantation of HPDE cells harbouring genomic mutations in CDKN2A, SMAD4 and TP53 and expressing KRAS G12V (HPDE KCST ; Fig. 6f and Supplementary  Fig. 15B-E). When orthotopically transplanted into NSG mice, both HPDE KECST2 and HPDE KCST2 cells formed invasive adenocarcinoma in 8 weeks (Supplementary Table 5 and Supplementary Fig. 15F-H). Collectively, these data indicate that our lentiviral reagents successfully induce genetic and genomic alterations of the PDA-associated genes, and that such driver mutations are sufficient to generate HPDE cell-derived lesions resembling invasive or metastatic PDA after orthotopic transplantation.

Media
Molecular comparison of hiPanINs to native PanIN and PDA.
Of the 92 upregulated, six genes (AGTR1, EBF4, MXRA5, PRSS1, PTGS2 and S100P) were previously reported to be induced in human PanINs by microarray analyses or immunohistochemistry [21][22][23][24][25] . Also, 23 upregulated genes including FN1, SEMA3A and GATA3 were shown previously, using mass spectrometry, RNA expression profiling and western blotting, to be induced in PDA (Supplementary Data 1). Among 48 downregulated genes, FXYD2 was previously reported to have reduced expression in two independent PanIN microarray analyses 21,22 , while five genes had been previously reported to have reduced expression in human PDA (Supplementary Data 1). Gene set enrichment analysis (GSEA) of our RNAseq data revealed statistically significant induction of genes involved in epithelialmesenchymal transition, G-to-M checkpoint and apoptosis (Supplementary Table 6), hallmark signatures related to cancer development. To further assess molecular similarities between our cultured hiPanIN clones and clinical PanIN or PDA specimen, we generated custom human PanIN/PDA gene sets with publically available microarray data and performed GSEA on our RNAseq data. As expected, we observed statistically significant enrichment of our RNAseq data in three published PanIN and PDA genesets (Fig. 7b). Collectively, these data suggest that our cultured hiPanIN clones show molecular similarities to clinical PanIN and PDA specimens. Neuromedin U (NMU) was originally identified as a neuropeptide that induces uterine smooth muscle contraction 26 ; recently, NMU has been suggested to be a circulating hormone that suppresses pancreatic islet insulin secretion 27 . Misexpression of pancreatic NMU protein has been previously reported in advanced human PDA, but not in precursor lesions like PanINs 28 . Our RNAseq analyses revealed NMU as one of the most highly elevated genes in all hiPanIN clones ( Fig. 7a and Supplementary Data 1), suggesting that NMU misexpression may initiate in PDA precursor stages. To address this possibility, we performed NMU immunohistochemistry on appropriate clinical tissue sections ( Fig. 7c; see Methods section). We did not detect NMU immunoreactivity in cases of normal pancreas (four out of four), chronic pancreatitis (three of three) or mucinous cystic neoplasms (MCN, five of five). However, we detected NMU protein production in intraductal papillary mucinous neoplasms (IPMNs; four of six cases), PanINs (six of six cases) and PDA (six of six cases; Fig. 7c). In each case of PanINs and IPMN, 50% of the pancreatic lesions labelled with NMU-specific antibody. In PanINs we observed this percentage of immunostaining in all PanIN grades 1-3. In contrast, more than 95% of individual PDA lesions were NMU-positive, suggesting that NMU immunoreactivity and expression increases as the precursor lesions progress to PDA. Thus, our  S1-CTRL S2 KECST C8 S1 KECST C4 S2 KCST C8 S1 KCST C4 S1 KECST C3 S1 KCST C3 NMU S100P PRSS1 PTGS2 AGTR1  identifying human PDA biomarkers to advance detection of early-stage disease. Recent studies have reported efforts to reconstitute PDA development in cultured human cells 11,12 , but the resources and conclusions from these studies are distinct from those reported here in several important ways. Kim et al. 11 reported reprogramming of cells derived from PDA resections to generate induced pluripotent stem cell lines, and teratomas formed by these after transplantation harboured early PanIN-like lesions that later progressed to invasive carcinoma. Since their starting material derived from resected advanced PDA, the genetic changes necessary for PanIN development in this system were not identified. Boj et al. 12 established human organoid cultures from unfractionated human pancreatic clusters and resected PDA samples from subjects with late-stage disease. However, systematic alterations of human PDA driver genes have not been performed in normal organoids; thus, the genetics of PanIN and PDA development were not directly addressed in that study. Huang et al. 10 generated human pancreatic exocrine spheroids from human ES cells and introduced oncogenic KRAS G12V and p53 R175H mutations to model PDA progression. Lesions produced in that study, however, did not recapitulate typical molecular and histological features of human PanIN or PDA either in culture or in transplantation settings. By contrast, we purified human pancreatic ductal cells from disease-free donors, and introduced driver mutations to address their necessity for PanIN or PDA formation in normal pancreatic duct cells. This strategy generated cloned hiPanIN cell lines with defined genetic changes that generate stable PanIN-like structures derived from duct cells.
It remains unclear why hiPanIN cells develop into PanIN-like lesions but not to PDA. In mice, oncogenic KRAS expression in exocrine cells is sufficient to promote both mouse PanIN (mPanIN) and PDA, when induced during embryonic pancreas development. However, when oncogenic KRAS expression is induced in adult mouse pancreas to model the adult onset of nearly all human PDA development, mPanIN or PDA failed to develop, even with additional deletion of p16 INK4A or Trp53 allele 30,31 . When chronic pancreatitis was induced, such mice showed development of mPanIN or PDA 30,31 , suggesting that both genetic and non-genetic acquired conditions may be needed to model invasive PDA development with hiPanIN cells. Our attempts to induce caerulein-mediated chronic pancreatitis in mice transplanted with hiPanIN clones have not produced invasive PDA, likely due to defective immune systems in NSG mice 32 . Consistent with this, survey of innate immune cells including macrophages (MAC-2 and F4/80), neutrophil (Ly6B.2) and mast cells (Toluidine Blue O) revealed the absence of these cells around the hiPanIN lesions, and weak or no immunoreactivity with phospho-signal transducer and activator of transcription 3 (STAT3) antibody was observed in the lesions (J.Lee and S.Kim, unpublished observations). Thus, it remains unclear whether chronic pancreatitis, inflammatory cells (which are mostly absent in NSG mice used for orthotopic transplants here) or other factors thought to promote PanIN-to-PDA progression could influence development of invasive-stage PDA-like lesions by hiPanIN cells. Studies on the pace of genetic and cellular changes in human pancreatic cancer suggest that the time from mutations that initiate PanIN formation to development of invasive disease and metastasis may span decades 18,33 . Therefore, compared to mice, the latency of PDA development by human cells may be intrinsically longer. The cell of origin may also be attributable to the resistance of hiPanIN to become PDA. Studies in mice have led to the hypothesis that both acinar cells and ductal cells have the potential to generate invasive PDA but via different precancerous developmental routes 34 . Therefore, it is possible that ductal cells in pancreas are more resistant to become PDA than acinar cells. Unfortunately, human pancreatic acinar cells did not grow in culture: thus, the PanIN/PDA-forming potential of human pancreatic acinar cells in our system remains unclear. hiPanIN progression to PDA may also require additional mutations or mutation combinations. In contrast to the spheres derived from primary human cells, the KCST combination of genetic changes promoted development of metastatic tumours by HPDE cells, a human pancreatic ductal line immortalized by human papilloma virus-16 E6/E7 that normally does not form tumours in xenotransplant models 9,20 . The parental HPDE line has multiple chromosomal deletions, in addition to expression of the E6/E7 oncogenes 35 ; thus, future studies should test the hypothesis that additional genetic modification of hiPanIN KCST and hiPanIN KECST duct cells could promote development of invasive or metastatic phenotypes. Alternatively, the tempo and order of mutations during native PanIN development may be important for PanIN-to-PDA progression of our hiPanIN models. As our hiPanIN model permits experimental modulation of the tempo and order of gene mutation, future studies should test this possibility. In summary, although the lack of progression to PDA limits our understanding of human pancreatic carcinogenesis, the stability of PanIN-like lesions in our system provides a unique platform and experimental advantages for identifying factors, mutations and conditions that promote invasive PDA development from late-stage PanINs.
Findings from our study also suggest that important intercellular or signalling interactions in PDA development are recapitulated by hiPanIN cells after orthotopic transplantation. A characteristic feature of PanIN and PDA development is the prominent fibrotic response found encasing hyperplastic or dysplastic pancreatic epithelial cells, and we observed this in all our orthotopic transplants. We also observed striking changes of hiPanIN cell nuclear and cytoplasmic morphology induced by transplantation, supporting the possibility that relevant reciprocal signalling interactions occur between hiPanIN cells and surrounding stroma. Although the contribution of stroma in PDA development, growth and survival appears well established 36 , it remains unclear whether cues from stroma or other non-epithelial components are also required for normal-tonative human PanIN development. Our hiPanIN cells are generated in defined medium without serum or feeder cells; thus, modulation of these cells with purified signalling factors or non-epithelial human cells should be feasible, and could accelerate studies of human stromal-epithelial signalling in PanIN and PDA progression. In our transplantation studies with the parental S1-3 hiPanIN cells, which represent a mixed population of mutant and normal cells, we did not observe production of lesions with features of 'more advanced' PanIN3, which is thought to precede development of invasive PDA. One interpretation of this finding is that cells with less than four mutations form low-grade PanINs and are found more frequently than those with all four in clonal populations. Consistent with this possibility, we found that transplantation of specific cloned cell lines (S1 clones) produced lesions with features of PanIN2 and PanIN3. Together, these findings support the hypothesis that genetic heterogeneity of epithelial cells within individual PDA lesions could affect histopathological development 19 . In summary, targeting a small number of genes in primary pancreatic ductal cells from human donors without known pancreas disease was sufficient to induce clones capable of generating PanIN-like structures. This unique resource should be useful for identifying human PDA biomarkers to advance detection of early-stage disease, and for investigating fundamental genetic or signalling mechanisms that promote PanIN development into invasive PDA.

Lentiviral infection and post-infection culture.
To infect HPDE cells, media were replaced with fresh media supplemented with 8 mg ml À 1 polybrene (Sigma-Aldrich, St Louis, MO) and lentiviruses were added directly into the media. Media were replaced after 3 days and puromycin selection began at a concentration of 0.5 mg ml À 1 for 5 days.
Sorted human primary ductal cells were washed twice with Advanced DMEM/F-12 media (Invitrogen) and resuspended 2-5 Â 10 5 cells in 300 ml of sphere culture media-Advanced DMEM/F-12 media supplemented with recombinant human epidermal growth factor (50 ng ml À 1 ; Sigma), rhR-spondin I (500 ng ml À 1 ; R&D Systems, Minneapolis, MN), recombinant human fibroblast growth factor 10 (50 ng ml À 1 ; R&D Systems), recombinant mouse Noggin (100 ng ml À 1 ; R&D Systems), 10 mM nicotinamide in PBS and Pen/Strep. The resuspended cells were placed in a well of 24-well Ultra Low Cluster Plate (Costar 3473, Corning, New York) and mixed with lentivirus for suspension infection overnight at 37°C. Next day, infected cells were washed twice with PBS and resuspended with 120 ml sphere culture media. Two hundred microlitres of growth factor-reduced Matrigel (BD Biosciences) was then added and the mixture was placed around the bottom rim of each well of 12-well plate. After solidification at 37°C for 60 min, each well was overlaid with 2 ml of sphere culture media. Media were replaced twice a week and the spheres were collected after 2 to 3 weeks for passaging. Static images of spheres were collected using Zeiss Axiovert 200 inverted microscope (Carl Zeiss, Germany). For collecting spheres, 1 ml of 3U ml À 1 dispase (Life Technologies) solution containing 0.1 mg ml À 1 DNaseI in PBS was added in each well and the Matrigel was mechanically disrupted by pipetting and incubated at 37°C for 1 h. The released spheres were collected, washed twice with PBS and used for subsequent applications. For passaging spheres, the collected spheres were trypsinized at 37°C for 5 min followed by quenching with fetal bovine serum. The dispersed cells were then used for cell counting with a haemocytometer or were plated for sphere culture as described above.
To isolate sphere clones, individual spheres were picked with 20 ml micropipette and transferred in microfuge tubes. The isolated spheres that represent each clone were dispersed into single cells with 100 ml of 0.5% Trypsin-EDTA and plated for sphere growth for 2 to 3 weeks. To ensure the purity of the clones, spheres were picked again from the growing spheres and subject to dispersion for second sphere culture (Fig. 6a).
Genomic DNA RNA and cDNA preparation for PCR and qRT-PCR. Genomic DNA was prepared using the DNeasy Blood and Tissue Kit (Qiagen Sciences, MD). A total of 30-50 ng of genomic DNA was used per the PCR reaction. PCR amplicons were used for agarose gel electrophoresis for assessing the presence of transgenes, or purified with DNA Clean and Concentrator Kit (Zymo Research, Irvine, CA) for sequencing and subsequent TIDE analyses to assess indel efficiency 14 . Total RNA was prepared with QIAGEN RNeasy micro kit (Qiagen Sciences) or PicoPure RNA Isolation Kit (Thermo Fisher Scientific), and used for cDNA synthesis using QIAGEN Omniscript RT Kit (Qiagen), according to the manufacturer's protocol. Relative mRNA level was measured by qRT-PCR of each cDNA in duplicate with gene-specific probe sets (Applied Biosystems, Foster City, CA) with TaqMan Universal PCR Master Mix (Applied Biosystems) and the ABI Prism 7500 detection system (Applied Biosystems). Normalizations across samples were performed using b-actin primers. PCR was performed with Accuprime Pfx (Thermofisher) in the presence or absence of 4 M betaine (Sigma-Aldrich) in 35 cycles, each cycle composed of a denaturing step at 95°C for 1 min, an annealing step at 60°C for 1 min and an extension step at 68°C for 1 min, followed by final extension at 68°C for 5 min. Primer sequences are shown in Supplementary Table 7.
Off-target effect analysis. Off-target site prediction was performed using Cas9 online designer (http://cas9.wicp.net) 38 . The potential off-target sites with the off-target scores more than 0.1 were selected, which resulted in two potential offtarget sites for sgRNA CDKN2A#1, one for SMAD4#1 and no sites were detected for TP53#2. PCR primers encompassing the potential off-target sites were designed and the PCR amplicons were purified and sequenced for TIDE analyses as described above. PCR was performed with Accuprime Pfx (Thermofisher) in the presence or absence of 4 M betaine (Sigma-Aldrich) in 35 cycles, each cycle composed of a denaturing step at 95°C for 1 min, an annealing step at 60°C for 1 min and an extension step at 68°C for 1 min, followed by final extension at 68°C for 5 min. Primer sequences are shown in Supplementary Table 7.
Orthotopic transplantation. Orthotopic transplantation was performed in the pancreas of NSG mice 39 . A total 0.75-1 Â 10 6 of dispersed single cells of human spheres or HPDE were resuspended in 100 ml of HBSS with 1% Matrigel and injected into the splenic lobe of pancreas. We estimated a minimum of three mice per each cell type for a proper evaluation of their growth as tumour or PanINs. Therefore, three to five male mice with the age of 10-14 weeks were randomized, given unique ID numbers for blinding and used per cell type for transplantation.
RNA isolation and RNA-Seq assays. RNA quality and quantity was measured using Bioanalyser instrument (Agilent Technologies). RNA-Seq libraries were built using KAPA Stranded mRNA-Seq Kit (KK8400, KAPA Biosystems, Wilmington, MA). Barcoded libraries were multiplexed and sequenced as single-end 75 bp reads on Illumina NextSeq 500 sequencer. RNA-Seq data sets were aligned to the Human Genome Assembly hg19 and analysed using DNAstar ArrarStar and Qseq version 12.3.1 build 53 (DNASTAR Inc., Madison, WI) with default parameters. Only the genes with more than 1 average RPKM (reads per kilobase million) value of all clones were used for fold change and GSEA analyses. For GSEA, GSEA v2.2.2 Software (Broad Institute) was used with Hallmark gene sets (H) in the Molecular Signatures Database (MSigDB) v5.1. Custom human PanIN and PDA genesets were generated by downloading publically available expression data sets that compare human PanIN or PDA with normal pancreas or normal ducts from Pancreatic Expression Database 41 . Due to minimum geneset size limit (16 or more genes per geneset), total 34 genesets with geneset size ranging from 16 to 334 were generated from 20 PanIN and PDA expression studies. The custom human PanIN and PDA genesets are available on request. In all cases, only the signatures showing normalized enrichment scoreZ1.5, P valueo0.01 and false discovery rate q-value o0.05 were considered significant.
Study approval. Institutional review board approval for research use of human tissue was obtained from the Stanford University School of Medicine and Johns Hopkins School of Medicine. Written informed consent was received for all human tissues used in this study. All animal experiments and methods were approved by the Institutional Animal Care and Use Committee of Stanford University.
Data availability. All sequencing data that support the findings of this study have been deposited in the National Center for Biotechnology Information Gene Expression Omnibus (GEO) and are accessible through the GEO Series accession number GSE88997. All other relevant data are available from the corresponding author on request.