FGFR2 Abrogation Intercepts Pancreatic Ductal Adenocarcinoma Development Available to Purchase

Pancreatic ductal adenocarcinoma (PDAC) remains one of the deadliest malignancies, with a 5-year survival rate of only 13% (1). Activating mutations of KRAS in the exocrine pancreas constitute the prevalent initiating event and drive the formation of precursor lesions called pancreatic intraepithelial neoplasia (PanIN; ref. 2). Analysis of pancreata from organ donors revealed that PanINs are common in the general healthy population and unlikely to progress to infiltrating carcinoma, given the relatively low incidence of PDAC (3). Additional signaling pathways, among which EGFR is the most well studied (4, 5), cooperate with mutant KRAS to drive PDAC initiation and progression. Inactivating mutations of tumor-suppressor genes (e.g., TP53, CDKN2A, and SMAD4) ultimately drive tumor formation (6, 7).

The extended time required for the progression of PanINs to PDAC provides a window of opportunity to intervene and block cancer development. Understanding the mechanisms that promote the early phases of pancreatic tumorigenesis is key to enable the development of strategies for better pancreatic cancer risk assessment, prevention and interception (8).

The FGFR gene family consists of four members, FGFR1 to FGFR4 (9). FGFRs are tyrosine kinase receptors (RTK) whose downstream pathways include the RAS/RAF-MAPK, PI3K/AKT, PLCγ, and STAT pathways. Context-dependent expression, ligand-binding capacities, and alternative splicing isoforms of the FGFRs determine differences in the pathways induced and consequent phenotypic changes. FGFRs are aberrantly activated through mutations, gene fusions, and copy-number amplifications in 5% to 10% of all human cancers but rarely in PDAC (10). Moreover, wild-type FGFRs can promote tumorigenesis and resistance to both chemotherapy and targeted therapies, including KRAS inhibitors (11–15). Efforts to inhibit FGFRs have led to the development of FGFR-targeted therapies, which include selective, nonselective, and covalent tyrosine kinase inhibitors, as well as monoclonal antibodies against the receptors and FGF ligand traps (10, 16).

While the oncogenic functions of FGFRs in invasive cancers have been extensively studied, their role in the early phases of tumorigenesis remains largely unexplored. In this study, we discovered that FGFR2 was upregulated in mutant KRAS-driven pancreatic metaplasia and precancerous lesions, supported PanIN formation, and FGFR2 inactivation restricted PDAC progression, making it a potential target for pancreatic cancer interception.

Kras^LSLG12D (RRID: IMSR_JAX:008179), Trp53^LSLR172H (RRID: IMSR_JAX:008652), Pdx1-Cre (RRID: IMSR_JAX:014647), and Rosa26^LSLYFP and Fgfr2^flox (RRID: IMSR_JAX:007569) strains in C57Bl/6 (RRID: IMSR_JAX:000664) background were interbred to obtain Pdx1-Cre (C); Pdx1-Cre; Fgfr2^flox (C Fgfr2^f); Pdx1-Cre; Rosa26^LSLYFP (CY); Kras^LSLG12D/+; Pdx1-Cre (KC); Kras^LSLG12D/+; Pdx1-Cre; Fgfr2^flox (KC Fgfr2^f); Kras^LSLG12D/+; Pdx1-Cre; Rosa26^LSLYFP (KCY); Kras^LSLG12D/+; Trp53^LSLR172H/+; Pdx1-Cre (KPC); Kras^LSLG12D/+; Trp53^LSLR172H/+; Pdx1-Cre; Fgfr2^flox (KPC Fgfr2^f); and Kras^LSLG12D/+; Trp53^LSLR172H/+; Pdx1-Cre; Rosa26^LSLYFP (KPCY) mice (17–20). C57Bl/6 mice were bred in house. All animal experiments were conducted in accordance with procedures approved by the Institutional Animal Care and Use Committee at Cold Spring Harbor Laboratory.

Primary KPC or KPCY tumor tissues were carefully dissected, avoiding adjacent normal pancreas or other tissue contamination. Tumors were minced and digested for 1 hour at 37°C in digestion buffer [DMEM, 5% FBS, penicillin, streptomycin, 2.5 mg/mL collagenase D (Sigma), 0.5 mg/mL Liberase DL (Sigma), and 0.2 mg/mL DNase I (Sigma)] while shaking. Single-cell suspensions were obtained by filtering through 100 µm nylon cell strainers and subsequent hypotonic lysis of red blood cells using ACK lysis buffer (Gibco). Preinvasive and invasive KPC cells were enriched using mouse Tumor Cell Isolation Kit (Miltenyi Biotech) by magnetic cell sorting according to the manufacturer’s instructions. To isolate preinvasive and invasive KPCY cells, cells were stained with 1 μg/mL 4′,6-diamidino-2-phenylindole (DAPI), and DAPI-negative yellow fluorescent protein (YFP)-positive cells were sorted using the FACSAria cell sorter (BD). Preinvasive and invasive cells were prepared for subsequent experiments or seeded in growth factor–reduced (GFR) Matrigel (BD).

Organoids were established as described in (21). T^m/+ and T^m/LOH organoid pairs were derived from passage 1 organoids (21). Cells were seeded in GFR Matrigel (BD). When indicated, GFR Matrigel (BD) was mixed with rat tail collagen I (Thermo Fisher Scientific). Once Matrigel was solidified, pancreatic organoid medium was added. Pancreatic organoid medium (“Complete medium”) contains AdDMEM/F12, 10 mmol/L HEPES (Invitrogen), Glutamax 1X (Invitrogen), penicillin/streptomycin 1× (Invitrogen), 500 nmol/L A83-01 (Tocris), 50 ng/mL murine EGF (mEGF), 100 ng/mL mNoggin (Peprotech), 100 ng/mL hFGF10 (Peprotech), 10 nmol/L hGastrin I (Sigma), 1.25 mmol/L N-acetylcysteine, 10 mmol/L Nicotinamide (Sigma), B27 supplement 1X (Invitrogen), and R-spondin conditioned medium (10% final). Pancreatic organoid minimal medium (“Minimal medium”) contains AdDMEM/F12, 10 mmol/L HEPES (Invitrogen), Glutamax 1X (Invitrogen), and penicillin/streptomycin 1× (Invitrogen). For the isolation of T^m/LOH organoids, cells were passaged 4 times in Complete medium supplemented with 10 µmol/L Nutlin-3a (Sigma).

For RT-qPCR assessment of TGFβ-induced transcriptional changes, T^m/+ organoids were passaged by dissociating and reseeding into Matrigel droplets. A portion of the cells were cultured with Complete medium, whereas a distinct portion of passage-matched cells were cultured in Minimal medium. Cells were cultured for 7 days in Minimal medium and then treated for 24 to 48 hours with 5 ng/mL recombinant mouse TGFβ1 (R&D Systems), unless otherwise specified, or vehicle control before being harvested for RNA extraction.

For RT-qPCR assessment of transcriptional changes upon Gata6 knockdown, T^m/+ organoids were cultured for 24 hours in Complete medium supplemented with 1 μg/mL doxycycline (Sigma) before being harvested for RNA extraction.

T^m/+ organoids were cultured for 1 day in pancreatic organoid reduced medium (no mEGF) and then treated for 48 hours with 1 µmol/L gefitinib (Selleckchem), 1 µmol/L erlotinib (MedChem), or DMSO control before being harvested for protein extraction.

Organoids were grown for 2 days in pancreatic organoid reduced medium (no mEGF) supplemented with 1 µmol/L erlotinib (MedChem) or vehicle. Organoids were harvested for protein extraction 1 hour after treatment with 100 nmol/L lirafugratinib (MedChem) or vehicle.

For dissociating organoids into single cell suspensions, organoids were incubated in TrypLE Express Enzyme (Thermo Fisher Scientific) for 15 minutes while shaking. To generate KPC-2D cell lines from tumor organoid cultures, organoids were dissociated into single cells as described above, resuspended with DMEM supplemented with 5% FBS, penicillin, and streptomycin, and plated on tissue culture plates. Brightfield images of organoids were taken using a Nikon eclipse TE2000-S microscope.

Genomic DNA from mouse tail and pancreas specimens was extracted using DNEasy Blood & Tissue Kit (Qiagen) following the protocol for tissue digestion. Organoids were harvested from two wells of a 24-well plate and centrifuged at 1,500 rpm for 5 minutes at 4°C. Genomic DNA from freshly isolated neoplastic cells or organoids was extracted using DNEasy Blood & Tissue Kit (Qiagen) following the protocol for cultured cells.

Each PCR reaction for p53 1loxP genotyping was performed in a 20 μL mixture containing 1× AmpliTaq Gold 360 Master Mix (Thermo Fisher Scientific), 0.5 µmol/L each primer, and 40 ng template DNA. The following primers were used: For AGC​CTG​CCT​AGC​TTC​CTC​AGG and Rev CTT​GGA​GAC​ATA​GCC​ACA​CTG. The PCR cycling conditions were 95°C for 5 minutes, followed by 40 cycles at 95°C for 30 seconds, 56°C for 30 seconds, and 72°C for 30 seconds, with a final extension step at 72°C for 5 minutes.

Each PCR reaction for genotyping of Fgfr2 alleles was performed in a 25 μL mixture containing 1× Taq Master Mix (Dye Plus; Vazyme), 0.4 µmol/L each primer, and 50 ng template DNA. The following primers were used: F1 ATAGGAGCAACAGGCGG, F2 TGC​AAG​AGG​CGA​CCA​GTC​AG, and F3 CAT​AGC​ACA​GGC​CAG​GTT​G. F1 and F2 produced a 142 bp and a 207 bp fragment from wild-type and Fgfr2^flox alleles, respectively. F1 and F3 produced a 471 bp fragment from the Fgfr2^Δ allele. The PCR cycling conditions were 95°C for 5 minutes, followed by 35 cycles at 95°C for 30 seconds, 60°C for 15 seconds, and 72°C for 45 seconds, with a final extension step at 72°C for 5 minutes.

Each PCR reaction for genotyping of Kras alleles was performed in a 20 μL mixture containing Advantage GC Polymerase (Takara), 1× GC 2 PCR buffer, 0.5 mol/L GC-Melt, 0.5 µmol/L dNTPs, 1.25 µmol/L each primer, and 50 ng template DNA. The following primers were used: For GGG​TAG​GTG​TTG​GGA​TAG​CTG and Rev TCC​GAA​TTC​AGT​GAC​TAC​AGA​TGT​ACA​GAG. Primers produced a 285 bp and a 315 bp fragment from WT and Kras^G12D alleles, respectively. The PCR cycling conditions were 98°C for 5 minutes, followed by 35 cycles at 98°C for 30 seconds, 58°C for 30 seconds, and 72°C for 30 seconds, with a final extension step at 72°C for 5 minutes.

PCR products were separated on a 2% agarose gel in 1× TAE buffer. Gel imaging was performed using a Syngene UV transilluminator or a ChemiDoc Imaging System (Bio-Rad).

Organoids were harvested from two wells of a 24-well plate and dissociated into single cells by incubating them in TrypLE Express Enzyme (Thermo Fisher Scientific) for 15 minutes while shaking. To analyze DNA content profile, 1 × 10⁵ freshly isolated neoplastic cells or organoid cell suspensions were resuspended in 1 mL of PBS and fixed by adding 2 mL of ice-cold absolute ethanol and kept at 4°C for at least 30 minutes. Cells were washed with 1 mL of 1% BSA in PBS and stained overnight with 50 μg/mL propidium iodide and 250 μg/mL RNaseA at 4°C. All FACS data were acquired using a LSRFortessa cell analyzer (BD) and analyzed with FlowJo software (TreeStar).

HEK293T (RRID: CVCL_0063) and KPC-2D cells were cultured in DMEM supplemented with 5% FBS, penicillin, and streptomycin. Cells were routinely tested for Mycoplasma contamination and were authenticated using short tandem repeat profiling.

Cas9-expressing T^m/+ organoids were established by infection with LentiV_Cas9_puro vector (RRID: Addgene_108100). For knockout (KO) experiments, single-guide RNAs were cloned into LRG2.1_Neo (RRID: Addgene_125593). Sequences of single-guide RNAs were TGCATCGAAAGGCAACCTCC for Fgfr2 KO1, GGAATACCTCCGAGCCCGG for Fgfr2 KO2, and GAAGATGGGCGGGAGTCTTC for Rosa26. For knockdown experiments, short hairpin RNAs (shRNA) were cloned into LT3GEPIR (RRID: Addgene_111177). Sequences of shRNAs were shGata6_1657 TGGAGTTTCATATAGAGCCCGC, shGata6_2857 TTTTCTTTTAAACAATTGGGAA, shGata6_3085 TAATGTAAACCAACCTGTGGGT, and shRluc CAGGAATTATAATGCTTATCTA. FLAG-tagged mouse Fgfr2 cDNA was cloned into LentiV_Blast (RRID: Addgene_111887) using Gibson assembly (New England Biolabs).

Lentivirus was produced in HEK293T cells using helper plasmids (VSVG and psPAX2) with X-tremeGENE 9 DNA Transfection Reagent (Roche). Organoids were dissociated into single cells by incubating them in TrypLE Express Enzyme (Thermo Fisher Scientific) for 15 minutes while shaking and spin-infected with the virus and 10 µg/mL polybrene (1,700 rpm for 45 minutes at room temperature). For organoid infection, the lentiviral supernatant was concentrated 10 times with Lenti-X concentrator (Takara). Media were changed at 24 hours after infection, and antibiotics (2 µg/mL puromycin, 1 mg/mL G418, or 10 µg/mL blasticidin) were added at 48 hours after infection.

Organoids were dissociated into single cells by incubating them in TrypLE Express Enzyme (Thermo Fisher Scientific) for 15 minutes while shaking. Cells were counted and diluted to 10 cells/μL in a mixture of pancreatic organoid medium (90% final concentration) and GFR Matrigel (BD; 10% final concentration). The 150 μL per well of this mixture (1,500 cells per well) was plated in 96-well white plates (Nunc), whose wells had been previously coated with poly(2-hydroxyethyl methacrylate; Sigma) to prevent cell adhesion to the bottom of the wells. Cell viability was measured every 24 hours, starting 1 day after plating, using the CellTiter-Glo assay (Promega) and a SpectraMax I3 microplate reader (Molecular Devices).

For assessment of growth changes upon drug treatment, cells were diluted to 30 to 50 cells/μL in a mixture of pancreatic organoid reduced medium (no mEGF; 90% final concentration) and GFR Matrigel (BD, 10% final concentration). 1 µmol/L lirafugratinib (MedChem) or 1 µmol/L erlotinib (MedChem) was added 24 hours after plating using the HP D300 Digital Dispenser. Compounds were dissolved in DMSO, and all treatment wells were normalized for DMSO content. Cell viability was assessed after 3 days of treatment using the CellTiter-Glo assay (Promega) and a SpectraMax I3 microplate reader (Molecular Devices).

Organoids were dissociated into single cells by incubating them in TrypLE Express Enzyme (Thermo Fisher Scientific) for 15 minutes while shaking. A total of 500 cells/well were plated in 384-well white plates (Nunc) in 30 μL of pancreatic organoid reduced medium (no mEGF; 90% final concentration) and GFR Matrigel (BD, 10% final concentration) and supplemented with 1 µmol/L gefitinib (Selleckchem). Lirafugratinib was added 24 hours after plating using the HP D300 Digital Dispenser. Lirafugratinib was tested in six replicate wells/concentration, ranging from 1 × 10⁻¹⁰ mol/L to 5 × 10⁻⁵ mol/L. Lirafugratinib was dissolved in DMSO, and all treatment wells were normalized for DMSO content. After 3 days, cell viability was assessed using CellTiter-Glo as per the manufacturer’s instruction (Promega) on a SpectraMax I3 (Molecular Devices) plate reader. Changes in viability were assessed relative to DMSO-treated cells.

Three-dimensional (3D) culture of KC acinar-enriched explants in collagen was performed following the protocol described in 22. Briefly, pancreas from KC Fgfr2^+/+ and Fgfr2^f/f mice ages 6 to 10 weeks was harvested and rinsed twice in 5 mL cold Hank’s Balanced Salt Solution (HBSS; Gibco). Tissue was minced into 1 to 3 mm sized pieces and then centrifuged for 2 minutes at 300 g and 4°C. The buffer was aspirated and minced tissue was digested in 5 mL cold HBSS by addition of 100 μL 10 mg/mL collagenase P (Roche) for 15 to 20 minutes while shaking at 300 rpm at 37°C. During this time, mechanical dissociation by pipetting up and down was performed every 5 minutes. Collagenase P was inhibited by addition of 5 mL cold 5% FBS in HBSS. Cells were centrifuged for 2 minutes at 300 g and 4°C then washed three times with 5 mL cold 5% FBS in HBSS. Cells were passed through a 500 μm strainer (pluriSelect) and then through a 100 μm cell strainer (Corning) and then pelleted through 10 mL of 30% FBS in HBSS gradient. Cells were resuspended in media and incubated at 37°C for at least 4 hours prior to plating. Cells were cultured in 1× RPMI 1640 supplemented with 0.1 mg/mL soybean trypsin inhibitor (Thermo Fisher Scientific), 1 μg/mL dexamethasone (Sigma), 1% FBS, and penicillin/streptomycin. All media were sterilized through a 0.22-μm filter (VWR; Corning). Acinar-enriched clusters were processed for RNA and protein extraction or embedded in rat tail collagen I (Thermo Fisher Scientific) and plated in collagen-coated 24-well plates. Culture media was added on top of solidified matrix and changed on days 1 and 3 after plating. Cells/collagen discs were digested in HBSS adding 1:50 10 mg/mL collagenase P (Roche) for ∼30 minutes while shaking at 300 rpm at 37°C. After all of the collagen was digested, cells were centrifuged for 2 minutes at 300 g and 4°C then washed with HBSS. Cells were then processed for RNA and protein extraction. Cells/collagen disks were fixed with 10% neutral-buffered formalin for 15 minutes, followed by fixation with 70% ethanol overnight, and then processed for histology.

Brightfield pictures of the cultures were taken using an Echo Laboratories RVL-100-G microscope.

Cells were lysed with RIPA buffer (300 mmol/L NaCl, 5 mmol/L EDTA, 20 mmol/L HEPES, 10% glycerol, 1% Triton X-100) supplemented with protease inhibitors (cOmplete, Mini, EDTA-free Protease Inhibitor Cocktail, Roche) and phosphatase inhibitors (PhosSTOP, Roche). Cleared lysates were electrophoresed and immunoblotted with the following indicated primary antibodies: FGFR2 (Cell Signaling Technology, cat. #23328, RRID: AB_2798862), HSP90 (Millipore, cat. #07-2174, RRID: AB_10807022), p53 (Leica Biosystems, cat. #P53-CM5P, RRID: AB_2744683), p21 (Santa Cruz Biotechnology, cat. #sc-6246, RRID: AB_628073), cofilin (Cell Signaling Technology, cat. #5175, RRID: AB_10622000), amylase (Abcam, cat. #ab21156, RRID: AB_446061), phospho-p44/42 MAPK (ERK1/2; Thr202/Tyr204; Cell Signaling Technology, cat. #4370, RRID: AB_2315112), ERK1/2 (Cell Signaling Technology, cat. #9102, RRID: AB_330744), EPCAM (Cell Signaling Technology, cat. #93790, RRID: AB_2800214), and β-actin (Cell Signaling Technology, cat. #4970, RRID: AB_2223172). After incubation of the membranes with appropriate horseradish peroxidase–conjugated secondary antibodies (Jackson ImmunoResearch Laboratories), imaging was performed using an enhanced chemiluminescence detection kit (Cytiva) and autoradiography films (LabScientific).

Organoids were cultured as described above and quickly harvested on ice in Advanced DMEM/F12 medium supplemented with HEPES 1× (Invitrogen), Glutamax 1× (Invitrogen), and penicillin/streptomycin (Invitrogen). Organoids were dissociated to single cells with TrypLE Express Enzyme (Thermo Fisher Scientific). Cells were resuspended in 50 μL of GFR Matrigel (BD) diluted 1:1 with cold PBS. Mice were anesthetized using isoflurane and subcutaneous administration of ketoprofen (5 mg/kg). The surgery site was disinfected with iodine solution and 70% ethanol. An incision was made in the upper left quadrant of the abdomen. The 10⁵ cells were injected in the tail region of the pancreas of C57Bl/6 mice. The incision at the peritoneal cavity was sutured with coated Vicryl suture (Ethicon) and the skin was closed with wound clips (Reflex7, CellPoint Scientific). Mice were euthanized 39 days after surgery, at which point, the mice transplanted with T6^m/LOH organoids had reached the humane endpoint. Tumors were dissected and processed for histology.

All mice ages 6 to 10 weeks were food-deprived for 16 hours prior to the start of the study and given eight-hourly intraperitoneal injections of 80 μg per kg of cerulein (Sigma) or saline for 2 consecutive days. Mice were weighed prior to the start of the study.

KC mice ages 6 to 10 weeks were randomly assigned to the four treatment groups. All mice were food-deprived for 16 hours prior to the start of the study and given eight-hourly intraperitoneal injections of 80 μg per kg of cerulein (Sigma) for 2 consecutive days. Concomitantly, mice were treated via oral gavage for 10 consecutive days with 50 mg/kg erlotinib (MedChemExpress) every 24 hours, 30 mg/kg lirafugratinib (MedChem) every 12 hours, the combination or 0.5% methylcellulose + 0.1% Tween 80 as vehicle control. Mice were weighed prior to the start of the study and every 3 days.

All tissues were fixed with 10% neutral-buffered formalin overnight. Tissues were then processed with standard tissue processing protocol, embedded in paraffin, and 6 μm sections were cut and mounted on slides. Formalin fixed paraffin-embedded (FFPE) tissue sections were stained with hematoxylin and eosin or used for IHC/immunofluorescent (IF) labeling.

Analysis of diseased pancreas was performed using QuPath software (23) for percentage area on a digital image of the tissue section acquired using Aperio scanner (Leica Biosystems) or Olympus VS200 scanner. Hematoxylin and eosin sections were evaluated for the highest-grade lesion present in a blinded fashion.

The human PDAC and intraductal papillary mucinous neoplasm (IPMN) tissue microarrays used for IF and IHC labeling were kindly provided by the Sidney Kimmel Comprehensive Cancer Center at Johns Hopkins University. The human normal adjacent tissues were kindly provided by the Mount Sinai School of Medicine biorepository core facility. Human PanIN tissue microarrays were obtained from US Biomax (BIC14011b). All tissue donations and experiments were reviewed and approved by the Institutional Review Board of Cold Spring Harbor Laboratory and the clinical institutions involved. Written informed consent was obtained prior to acquisition of tissue from all patients, or consent was waived by local institutional review boards. The studies were conducted in accordance with ethical guidelines (Declaration of Helsinki). Samples were confirmed to be PanIN, IPMN, or PDAC upon pathologist assessment.

FFPE sections were deparaffinized and rehydrated. For antigen retrieval, slides were incubated with 10 mmol/L citrate buffer (pH 6.0) in a pressure cooker for 6 minutes. To perform IHC, endogenous peroxidase activity was quenched in 3% H₂O₂ for 20 minutes. Tissues were blocked in 2.5% Normal Horse Serum blocking solution (Vector Laboratories) for IHC or 5% BSA (Sigma) in Tris-buffered saline with Tween 20 (TBST) for IF and subjected to staining with the following antibodies overnight at 4C: p53 (Leica Biosystems, cat. #P53-CM5P, RRID: AB_2744683 1:100), GFP/YFP (Abcam, cat. #ab6673, RRID: AB_305643 1:100), amylase (Sigma-Aldrich, cat. #A8273, RRID: AB_258380 1:250), Dolichos biflorus agglutinin fluorescein (Vector Laboratories, cat. #FL-1031, RRID: AB_2336394 1:200), FGFR2 (Abcam, cat. #ab58201, RRID: AB_2103519 1:1000), cytokeratin 19 (CK19) Alexa Fluor 647 (Abcam, cat. #ab205446, RRID: AB_3674091 1:500), CK19 (Sigma-Aldrich, cat. #MABT913, RRID: AB_2892523 1:500), GATA6 (R&D Systems, cat. #AF1700, RRID: AB_2108901 1:500), S100A2 (Abcam, cat. #ab109494, RRID: AB_10859000 1:250), FGF7 (OriGene, cat. #ta321423, RRID: AB_3674092 1:100), FGF10 (Millipore, cat. #ABN44, RRID: AB_11204345 1:100), phospho-p44/42 MAPK (ERK1/2; Thr202/Tyr204; Cell Signaling Technology, cat. #4370, RRID: AB_2315112 1:250), Ki67 (Thermo Fisher Scientific, cat. #14-5698-82, RRID: AB_10854564 1:500), Ki67 Alexa Fluor 488 (Abcam, cat. #ab281847, RRID: AB_3674093 1:500), p19^Arf (Abcam, cat. #ab26696, RRID: AB_776947 1:1000), S100A4 (Abcam, cat. #ab27957, RRID: AB_2183775 1:500), E-cadherin (BD Biosciences, cat. #610181, RRID: AB_397580 1:500), clusterin (CLU; Thermo Fisher Scientific, cat. #PA5-46931, RRID: AB_2608081 1:100).

ImmPRESS alkaline phosphatase and ImmPRESS horseradish peroxidase IgG polymers (Vector Laboratories) were used as secondary antibodies for IHC. ImmPACT DAB peroxidase substrate and Vector Blue alkaline phosphatase substrate (Vector Laboratories) were used as substrates. Hematoxylin (Vector Laboratories) was used as counterstain. Cover slides were mounted with Cytoseal 60. Brightfield images were obtained using a Zeiss Axio Imager.A2 microscope.

Secondary Alexa Flour antibodies for IF were obtained from Thermo Fisher Scientific. Nuclei were stained with 1 μg/mL DAPI for 1 hour. Slides were mounted with ProLong Gold Antifade Mountant (Thermo Fisher Scientific). Z-stack images were obtained using a Leica SP8 confocal laser microscope at ×40 magnification. Large mosaic stitched images were obtained using a Zeiss Observer inverted fluorescence microscope. Tissue sections were imaged at 20× using an Olympus VS200 scanner.

RNA in situ hybridization combined with IF was performed according to the manufacturer’s instructions (RNAscope Multiplex Fluorescent Reagent Kit, v2 323100; ACD) using probes specific for Fgfr1 (443491-C2; ACD), Vim (457961-C4; ACD), and Cdkn2a/ARF (503811-C1; ACD). Once completed, the samples were washed in PBS and then blocked for 1 hour with 20% donkey serum at room temperature. Primary antibodies were incubated overnight at 4°C. Secondary Alexa Flour antibodies (1:500 in blocking buffer) were incubated for 1 hour at room temperature, and the samples were washed three times in PBS. Slides were counterstained with DAPI and mounted with ProLong Gold Antifade Mountant (Thermo Fisher Scientific). Z-stack images were obtained using a Zeiss LSM780 confocal laser scanning microscope at ×20 magnification.

Quantification of staining was performed with QuPath software (23).

RNA was extracted using TRIzol Plus RNA Purification Kit (Thermo Fisher Scientific) per manufacturer’s instructions. RNA samples were treated on column with PureLink DNase (Thermo Fisher Scientific).

Total RNA was isolated as described above. cDNA was produced using TaqMan reverse transcription reagents (Thermo Fisher Scientific). Ten ng of cDNA was used for RT-qPCR reactions with TaqMan Universal Master Mix II, no UNG (Thermo Fisher Scientific) and the following TaqMan probes: Fgfr2 Mm01269930_m1, Gata6 Mm00802636_m1, Fgfr1 Mm00438930_m1, Snai1 Mm00441533-g1, Fn1 Mm01256744_m1, and Hprt Mm00446968_m1 (Thermo Fisher Scientific).

Z-score values were derived from 2^−ΔC_t values.

The quality of purified RNA samples was determined using a Bioanalyzer 2100 (Agilent) with RNA 6000 Nano Kit. RNAs with RNA integrity number values greater than 8.5 were used to generate sequencing libraries. Libraries were generated from 1 μg of total RNA using TruSeq Stranded Total RNA Library Prep Human/Mouse/Rat Kit (48 Samples; Illumina) or KAPA mRNA HyperPrep Kit (Roche) per manufacturer’s instructions. Libraries were quality checked using a Bioanalyzer 2100 (Agilent) with High-Sensitivity DNA Kit and quantified using PicoGreen (Thermo Fisher Scientific). Equimolar amounts of libraries were pooled and subjected to single- or paired-end, 75 bp sequencing at the Cold Spring Harbor DNA Sequencing Next-Generation Shared Resource using an Illumina NextSeq 500 or 550 instrument.

RNA sequencing (RNA-seq) read quality was first quantified using FastQC (http://www.bioinformatics.babraham.ac.uk/projects/fastqc/). Reads were then trimmed using Trim Galore. Reads were mapped to transcript annotation GENCODE M16 (organoid RNA-seq) and GENCODE M35 (KC acinar-enriched explant RNA-seq; ref. 24) using Spliced Transcripts Alignment to a Reference (25). RSEM (26) was used to extract counts per gene and transcripts per million (TPM). RNA-seq tracks were generated from aligned reads using deepTools2 (27) and visualized using UCSC Genome Browser (28).

Differential gene expression analysis was performed using Bioconductor package DESeq2 (29). A prefiltering step was performed to remove genes that had reads in less than two samples. T^m/+ and T^m/LOH organoids were considered paired datasets. A prefiltering step was performed to remove genes that had reads in less than two samples. At this step, all genes not classified as protein coding according to BioMart were discarded (30). Only genes with an adjusted P value less than 0.05 were retained as significantly differentially expressed genes. A principal component analysis was performed using the 500 most variable protein-coding genes after variance stabilizing transformation using DESeq2 software. The graphical representation of the two most important components was created using the CRAN package ggplot2 (31).

Default parameters were used to perform GSEAPreranked analysis using the gene set enrichment analysis (GSEA) software (32, 33). Hallmark and curated gene sets were downloaded from the MSigDB database, which include pathways from the Kyoto Encyclopedia of Genes and Genomes (34, 35).

SNP calling was based on the GATK RNA-seq short variant discovery pipeline (36). More specifically, the reads were grouped per sample and duplicated reads were marked using Picard MarkDuplicates (http://broadinstitute.github.io/picard). Then, the GATK toolkit (37) was used—in the following specific order—to span splicing events and to reassign map quality (SplitNCigarReads program), to realign indels (RealignerTargetCreator and IndelRealigner programs), to recalibrate the base quality scores (BaseRecalibrator program), and to call and filter variants (HaplotypeCaller and VariantFiltration programs) across all samples. Read count and quality were extracted, for variants of interest, using the mpileup option from SAMtools (38). Using a home-made script, reads with mapping quality filter value (MAPQ value) < 5 and sequencing quality score (phred value) < 20 were removed before counting the number of reads (coverage) for each allele present at the position of interest.

The single-cell RNA-seq (scRNA-seq) plot was generated with ggplot2 (31). A cell was considered as expressing the gene when the transcription count was higher than zero. The cell clusters corresponded to the one defined in the associated study.

GraphPad Prism 10 was used for graphical representation of data. Heatmaps were generated using GraphPad Prism 10 or RStudio and Bioconductor package ComplexHeatmap (39). Statistical analysis was performed using the tools within Prism 10 indicated in the figure legends. Asterisks denote P value as follows: *, P ≤ 0.05; **, P ≤ 0.01; ***, P ≤ 0.001; ****, P ≤ 0.0001; ns, not significant. Figures were prepared using Illustrator (Adobe).

RNA-seq data generated in this study have been deposited in the Gene Expression Omnibus (GEO) database under the accession number GSE278148 (KC acinar-enriched explant RNA-seq) and GSE278203 (organoid RNA-seq).

Publicly available data reanalyzed for this study are available in the GEO database under accession codes GSE195914 (preinvasive and invasive KPC cell scRNA-seq; ref. 40), GSE93326 (human PDAC laser capture microdissection RNA-seq; ref. 41), GSE71729 (human PDAC microarray data; ref. 42), and GSE250519 (CUT&RUN; ref. 40) and in the European Genome-Phenome Archive (EGA) under accession code EGAS00001002543 (human PDAC LCM RNA-seq; ref. 43). Expression and mutational data for PDAC studies were obtained from the cBioPortal (paad_tcga; ref. 44).

All other raw data generated in this study are available upon request from the corresponding author.

To investigate the molecular mechanisms that promote PDAC progression, we utilized the Kras^LSLG12D/+; Trp53^LSLR172H/+; Pdx1-Cre (KPC) mouse model in which Kras^G12D and Trp53^R172H were expressed in the pancreas by virtue of Cre recombinase under the control of the Pdx1 promoter (17, 19). In this genetically engineered mouse model, the transition from premalignant lesions to invasive cancer is facilitated by loss of heterozygosity (LOH) of the wild-type Trp53 allele (45). Indeed, we confirmed by PCR-based genotyping that loss of the wild-type Trp53 allele occurred in vivo in neoplastic-enriched cell populations isolated from KPC tumors (Fig. 1A; refs. 17, 46). However, in all samples, we observed a faint band corresponding to the wild-type copy of Trp53, suggesting the presence of precancerous cells within the tumor.

To investigate whether the tumor mass contained precancerous cells, we analyzed p53 expression in tumor sections of KPCY mice (Kras^LSLG12D/+; Trp53^LSLR172H/+; Pdx1-Cre; Rosa26^LSLYFP), in which all Kras^G12D-expressing cells presented the YFP lineage label (18). Under basal conditions, p53 levels are undetectable by IHC due to its rapid turnover (17, 47, 48). Mutations in Trp53 result in highly stabilized mutant p53 proteins only in the absence of the other wild-type Trp53 allele (17, 45, 49). Therefore, only Trp53 LOH cells present positive staining for p53 by IHC. IHC labeling for YFP and p53 in KPCY tumors revealed two populations of YFP-positive cells: one with increased nuclear expression of p53, which represented Trp53 LOH invasive cells, and the other in which p53 was not detected, representing the precancerous cells (Fig. 1B). Indeed, cells that did not present p53 accumulation displayed histological features of acinar-to-ductal metaplasia (ADM) and murine PanINs (mPanIN), whereas cells with detectable p53 expression exhibited features of invasive carcinoma.

We previously described the establishment of pancreatic ductal organoid cultures from multiple primary tumors from KPC mice and showed that tumor “T” organoids did not exhibit Trp53 LOH (21). Given that Trp53 LOH was prevalent in freshly isolated KPC neoplastic cells (Fig. 1A), we sought to determine when Trp53 LOH cells were lost in organoid culture. We performed PCR-based genotyping of neoplastic-enriched cell populations after isolation and upon passaging as organoids (Fig. 1C). Precancerous cells harboring both mutant and wild-type Trp53 alleles outcompeted the Trp53 LOH cells in organoid cultures in a few passages. This growth advantage was independent of medium or matrix composition and associated with the ploidy status (Supplementary Fig. S1A–S1C). KPC tumors were enriched in aneuploid and tetraploid cells, which were outcompeted in organoid cultures by diploid and quasi-diploid cells.

To isolate the Trp53 LOH cells, we treated the organoids with Nutlin, a small molecule that mediates p53 stabilization by inhibiting the p53–MDM2 interaction (50). Nutlin treatment induced the death of the cells that retained wild-type p53 and allowed for the isolation of the Trp53 LOH cells (Fig. 1D; Supplementary Fig. S1C–S1E). Using this strategy, we established matched organoid cultures of KPC preinvasive cells that retained the wild-type Trp53 allele (T^m/+) and KPC invasive cells that had lost the wild-type Trp53 allele (T^m/LOH).

Once stable cultures were established, T^m/+ organoids were diploid, whereas some T^m/LOH organoids showed a slight increase in DNA content by propidium iodide staining (Supplementary Fig. S2A; ref. 21). While T^m/+ and T^m/LOH organoids proliferated at similar rate in vitro, T^m/LOH organoids formed tumors with significantly higher penetrance following orthotopic implantation in syngeneic mice (Fig. 1E; Supplementary Fig. S2B). Mice transplanted with T^m/+ organoids developed either no tumor or very small tumors when pancreata were collected at the humane endpoint of mice transplanted with the matched T^m/LOH organoids.

To investigate the transcriptional differences between KPC pre-invasive and invasive cells, we performed RNA-seq. When subjected to principal component analysis, T^m/+ and T^m/LOH organoids separated based on their LOH status and paired organoids did not cluster together (Fig. 1F; Supplementary Fig. S2C). Comparison of T^m/+ and T^m/LOH organoids identified 1,514 differentially expressed genes: 996 upregulated and 518 downregulated genes in T^m/LOH relative to T^m/+ (Supplementary Table S1). GSEA confirmed repression of the p53 signaling pathway in T^m/LOH organoids (Fig. 1G). Among the upregulated genes, we identified several genes that were previously associated with pancreatic cancer progression, such as Prdm1/BLIMP1, Foxa1, Gata5, Twist1, Glul, and Soat1 (Supplementary Fig. S2D; refs. 46, 51–53). Among the genes induced in T^m/+ compared with T^m/LOH organoids, we noted Fgfr2 isoform IIIb (from now on, Fgfr2), which we had previously shown to be expressed in KPC precancerous cells (Fig. 1H; Supplementary Fig. S2E; ref. 40). Of note, Fgfr2 was the only Fgfr to be highly expressed in T^m/+ organoids.

To investigate FGFR2 expression during pancreatic cancer progression, we performed IF labeling. In healthy pancreata, FGFR2 was not expressed in normal ducts or acinar cells, but was weakly expressed in blood vessels (Fig. 2A). Notably, the percentage of acinar cells that expressed FGFR2 significantly increased as soon as KRAS^G12D initiated the transformation of these cells to metaplastic ductal-like lesions in young Kras^LSLG12D/+; Pdx1-Cre (KC) mice. In these mice, the majority of CK19-positive precancerous cells expressed FGFR2 at the plasma membrane (Fig. 2B). In KPC tumors, FGFR2 expression was observed in precancerous cells in which p53 was not or barely detectable, whereas it was absent or reduced in invasive cells with nuclear accumulation of p53 (Fig. 2C; ref. 40).

FGFs are produced by fibroblasts and bind to heparin and heparan sulfates, which protect them from proteases and enhance the formation of an active signaling complex with the FGFRs (54). We found that FGFR2’s ligands FGF7 and FGF10 often colocalized with FGFR2, and the percentage of double positive cells increased with disease progression, potentially due to the rise in fibroblast numbers (Supplementary Fig. S3A and S3B; ref. 15).

Next, we examined whether the expression pattern of FGFR2 observed in mice was conserved in human. Consistent with the finding in mice, human precursor lesions IPMN and PanIN expressed FGFR2 (Fig. 2D). However, differently from mice, FGFR2 was expressed at high levels in most neoplastic cells in classical PDACs, in fewer cells and at lower levels in intermediate coexpressor (IC) PDACs, and was absent in basal-like PDACs, as defined by immunolabeling for the classical marker GATA6 and the basal-like marker S100A2 (Fig. 2E; ref. 55). In line with these results, analysis of expression data of human PDACs showed significantly higher expression of FGFR2 in classical compared with basal-like PDACs (Supplementary Fig. S3C).

Because GATA6 and FGFR2 were highly expressed in classical PDACs, we wondered whether GATA6 regulated FGFR2 expression. In T^m/+ organoids, Gata6 knockdown using three different doxycycline-inducible shRNAs resulted in Fgfr2 downregulation (Supplementary Fig. S3D). In addition, CUT&RUN profiling in the classical PDAC cell line HPAF-II revealed the binding of GATA6 to the promoter region of FGFR2 (Supplementary Fig. S3E; ref. 40). Together, these results demonstrated that FGFR2 is progressively upregulated in human pancreatic precancerous lesions and classical PDAC and repressed in basal-like PDAC (Fig. 2F) and that its expression is regulated by GATA6.

Our immunolabeling analysis in KPC tumors indicated that FGFR2 expression was weak or absent in invasive cells with nuclear accumulation of p53. To explore whether the loss of p53 mediates FGFR2 repression, we analyzed our previously published T^m/+ organoids in which we inactivated the wild-type Trp53 allele using CRISPR (51) and found that FGFR2 expression was retained following LOH of Trp53, indicating that FGFR2 downregulation in invasive cells was initiated by subsequent events (Supplementary Fig. S4A).

Increased activation of the TGFβ pathway has been reported in invasive KPC cells and in basal-like PDACs, in which it is known to contribute to epithelial-to-mesenchymal transition (EMT; refs. 40, 43, 56). Because FGFRs are also key regulators of EMT during embryonic development and in neoplastic cells during cancer progression (57), we sought to evaluate whether TGFβ signaling drives FGFR2 repression, as previously observed in mammary epithelial cells (58). Therefore, we treated three T^m/+ organoid lines with TGFβ and analyzed the transcriptional changes by RT-qPCR (Fig. 3A). Organoids cultured in “Complete” medium expressed Fgfr2 at higher levels compared with organoids grown in “Minimal” medium without any additives, including the TGFβ receptor inhibitor A83-01 and the FGFR2 ligand FGF10. The addition of TGFβ to the “Minimal” medium further reduced the expression of Fgfr2 and induced the upregulation of the mesenchymal genes Fgfr1, Snai1, and Fn1. The switch from Fgfr2 to Fgfr1 expression was dose-dependent and time-dependent (Fig. 3B; Supplementary Fig. S4B).

To assess Fgfr2 and Fgfr1 expression in relation to cell states, we analyzed our scRNA-seq data of KPC preinvasive and invasive cells (Fig. 3C; ref. 40). Fgfr2 expression was detected in few of the epithelial cells in cluster 1, whereas Fgfr1 was expressed by the mesenchymal cells in cluster 3. Using RNA in situ hybridization (RNA ISH) in combination with IF for p53 and Cdkn2a/ARF as markers of invasive cells, Vimentin as a marker of mesenchymal cells, and CK19 as a marker of ductal cells, we confirmed that FGFR2 was expressed in precancerous ductal lesions in murine tumors, while Fgfr1 was expressed in invasive mesenchymal cells and some stromal cells (Fig. 3D and E; Supplementary Fig. S4C and S4D). Although PDAC cells occupy a continuum of epithelial-to-mesenchymal expression states (40, 56), Fgfr2 expression was restricted to precancerous cells with strong epithelial features, whereas Fgfr1 was uniquely expressed by mesenchymal cells. The majority of neoplastic cells that exhibited a partial EMT phenotype expressed neither gene.

Finally, we found that FGFR2 expression was significantly increased in SMAD4-altered human PDACs compared with those expressing wild-type SMAD4, further supporting a role for TGFβ signaling in downregulating FGFR2 (Fig. 3F).

FGFR2 upregulation in mutant KRAS-driven pancreatic metaplasia and precancerous lesions could be part of a tumor-promoting program or an inflammatory response. Because FGFR2 was implicated in the repair of the skin, intestine, liver, and lung (59), we tested whether FGFR2 was required for pancreas recovery following injury. We inactivated Fgfr2 in the pancreas by crossing Pdx1-Cre (C) to Fgfr2^f/f mice in which the alternatively spliced FGF-binding Ig domains IIIb, IIIc, and transmembrane domain of Fgfr2 are flanked by loxP sites, resulting in an inactive protein after recombination (Supplementary Fig. S5A and S5B; ref. 20). The pancreata of C Fgfr2^f/f mice did not present any overt defects upon histological evaluation (Fig. 4A). Using an established model of pancreatitis with the cholecystokinin analog cerulein, we found that 1 day after treatment, the pancreata of both C Fgfr2^+/+ and Fgfr2^f/f mice presented transient pancreatic inflammation, with edema, acinar cell damage, ADM, and infiltration of immune cells. By day 7, the tissue integrity in both strains was almost completely restored, reaching full recovery by day 30. Immunolabeling was consistent with the histological analysis, revealing a transient decrease of the acinar marker amylase and concomitant increase of the ductal marker CK19 at day 1 and reversion to pre-cerulein treatment levels by day 30 in both C Fgfr2^+/+ and Fgfr2^f/f mice (Fig. 4B). These results suggested that Fgfr2 is dispensable for pancreas recovery following injury.

Cerulein-mediated injury to the pancreas accelerates mutant KRAS-driven metaplastic lesions and mPanIN formation (60), as confirmed by histological and immunolabeling analyses of the pancreata from KC in comparison with C mice (Supplementary Fig. S6A and S6B). FGFR2 expression was not detected by IHC upon transient pancreatitis and following recovery (Fig. 5A). However, in KC mice, FGFR2 was upregulated in acinar cells, metaplastic lesions, and mPanINs upon KRAS^G12D-driven malignant transformation, indicating its potential involvement in promoting early progression. Intriguingly, analysis of normal adjacent tissue from patients with PDAC revealed areas of morphologically normal acinar cells, ADM, and PanINs that stained positive for FGFR2, which was reminiscent of what we observed in KC mice (Supplementary Fig. S6C).

To explore the role of FGFR2 in pancreatic tumorigenesis, we crossed KC mice with Fgfr2^f/f mice (Supplementary Fig. S6D). Next, we modeled mutant KRAS-induced malignant transformation of acinar cells using 3D cultures of KC acinar-enriched epithelial explants embedded in collagen (22). Under these culture conditions, many acinar cells were lost because of cell death, but some acinar clusters and associated ductal cells formed ductal structures (Fig. 5B). We found that acinar-enriched clusters from KC Fgfr2^+/+ mice formed ductal structures more efficiently than the ones from KC Fgfr2^f/f mice. Protein analysis at days 0, 2, and 5 of culture showed a progressive decreased expression of the acinar marker amylase and conversely an increased expression of the ductal marker EPCAM; however, EPCAM upregulation was greater in KC Fgfr2^+/+ compared with KC Fgfr2^f/f cells (Fig. 5C; Supplementary Fig. S6E). Furthermore, analysis of MAPK signaling by detection of the phosphorylation of ERK1/2 revealed activation of the pathway from day 2 of culture, which was stronger in KC Fgfr2^+/+ compared with KC Fgfr2^f/f cells (Fig. 5C; Supplementary Fig. S6E). mRNA expression analysis by RNA-seq showed progressive repression of acinar genes and upregulation of ductal genes, including Fgfr2, over time in culture (Supplementary Fig. S6F; Supplementary Table S2). GSEA revealed upregulation of KRAS signaling at day 2 and proliferation-associated programs, such as G₂M checkpoint, E2F and MYC targets, at day 5 in KC Fgfr2^+/+ compared with KC Fgfr2^f/f cells (Fig. 5D and E; Supplementary Fig. S6G; Supplementary Table S2). Altogether, these results suggested that FGFR2 promotes proliferation and enhances MAPK signaling during mutant KRAS-driven acinar-to-ductal–like cell transformation ex vivo. To corroborate these findings, we performed IF for FGFR2, Ki67, phospho-ERK1/2, and CLU, a marker of acinar transformation (61), in pancreata from KC mice at days 7 and 30 after cerulein treatment. We found that CLU- and FGFR2-positive cells presented with a stronger signal for phospho-ERK1/2, and a higher fraction of these cells were Ki67-positive (Fig. 5F and G), further reinforcing that FGFR2 supports mutant KRAS-induced malignant transformation of acinar cells by promoting proliferation and MAPK signaling activation. This suggested that FGFR2 is a key factor in pushing KRAS signaling over the threshold necessary for disease progression.

To assess the effect of FGFR2 inactivation on tumorigenesis in vivo, we analyzed metaplastic lesion and mPanIN formation during spontaneous and cerulein-accelerated tumorigenesis in KC Fgfr2^f/f compared with Fgfr2^+/+ mice. In both models, we found that KC Fgfr2^+/+ mice presented a significantly larger fraction of diseased pancreatic tissue compared with age-matched KC Fgfr2^f/f mice (Fig. 6A–D), indicating that mutant KRAS-driven pancreatic tumorigenesis was hindered upon inactivation of FGFR2. Immunolabeling was consistent with the histological analysis, revealing smaller areas of mPanINs and larger areas of acinar tissue in KC Fgfr2^f/f compared with Fgfr2^+/+ mice (Supplementary Fig. S7A and S7B).

In addition to having lower fraction of diseased pancreas, KC Fgfr2^f/f mice harbored lower grade mPanINs. In the spontaneous tumorigenesis model, the KC Fgfr2^+/+ mice exhibited mPanIN-1B and mPanIN-2 lesions, whereas KC Fgfr2^f/f mice harbored mPanIN-1A, mPanIN-1B, or no lesions (Fig. 6E). Similarly, in the cerulein-accelerated tumorigenesis model, KC Fgfr2^+/+ mice exhibited mPanIN-1A, mPanIN-1B, and mPanIN-2 lesions, whereas KC Fgfr2^f/f mice harbored lower grade mPanIN-1A and mPanIN-1B (Fig. 6F). Immunolabeling analysis revealed similar levels of phosphorylated ERK1/2 and percentage of proliferating precancerous cells in mPanINs from KC Fgfr2^+/+ and Fgfr2^f/f mice (Supplementary Fig. S7C–S7F).

Given the differences that we observed in premalignant lesion formation in the KC pancreatic cancer model with Fgfr2 inactivation, we evaluated tumor formation in the KPC model. At 4 months of age, KPC Fgfr2^+/+ mice presented a significantly larger fraction of diseased pancreas tissue compared with KPC Fgfr2^f/f mice (Fig. 6G and H). Notably, three of four of the KPC Fgfr2^+/+ mice examined had developed a tumor mass, while KPC Fgfr2^f/f mice harbored mPanIN-1A or mPanIN-1B lesions (Fig. 6I). Next, we followed a cohort of mice to humane endpoint and found that Fgfr2 inactivation in the KPC model led to a significant extension of median survival from 143 to 195 days in KPC Fgfr2^f/f, although KPC Fgfr2^f/f mice eventually all succumbed to PDAC (Fig. 6J; Supplementary Table S3). Genotyping analysis of organoids derived from KPC Fgfr2^f/f tumors showed recombination of the Kras, Trp53, and Fgfr2 loci, thus excluding the possibility of incomplete recombination of Fgfr2 as a potential escape mechanism (Supplementary Fig. S8A). Histological evaluation of the tumor tissue revealed a higher frequency (21 vs. 38%) of cellular and mesenchymal PDACs in KPC Fgfr2^f/f mice (Supplementary Fig. S8B; Supplementary Table S3). IF analysis for E-cadherin, as a marker of epithelial differentiation, and S100A4, as a marker of EMT, revealed a higher percentage of double positive cells in tumors from KPC Fgfr2^f/f mice (Supplementary Fig. S8C). These results suggested that the acquisition of a partial EMT phenotype may represent a bypass mechanism to promote disease progression in the absence of FGFR2. However, this did not result in a significant increase in metastasis formation (Supplementary Fig. S8D and S8E).

Altogether, these data demonstrated that abrogation of FGFR2 delays KRAS^G12D-induced tumorigenesis, with alternative adaptative pathways likely compensating for the inactivation of FGFR2 later in tumor progression.

Previously, other groups showed that EGFR signaling is required for mutant KRAS-driven pancreatic cancer development (4, 5). However, given that pharmacologic inhibition of EGFR restricted but did not abrogate premalignant lesion formation (4, 62), we wondered whether EGFR and FGFR2 cooperated in promoting pancreatic tumorigenesis. We first examined FGFR2 expression and function upon EGFR inhibition. Treatment of T^m/+ organoids with the EGFR inhibitors gefitinib and erlotinib resulted in FGFR2 protein upregulation, suggesting that FGFR2 may compensate for EGFR inhibition (Fig. 7A). Indeed, the growth of Fgfr2 knockout T^m/+ organoids was strongly inhibited by erlotinib, while it was not consistently reduced compared with control organoids in normal culture conditions (Fig. 7B; Supplementary Fig. S9A). Overexpression of FGFR2 in T^m/+Fgfr2^f/f organoids rescued the growth inhibition caused by erlotinib treatment (Fig. 7C).

Dose escalation with the FGFR2 inhibitor lirafugratinib reduced the growth of gefitinib-treated control but not Fgfr2 knockout T^m/+ organoids in a dose-dependent manner, confirming the specificity of this small molecule for FGFR2 (Supplementary Fig. S9B; ref. 63). To investigate MAPK signaling downstream of FGFR2, we induced FGFR2 upregulation by inhibiting EGFR via erlotinib treatment for 2 days in T^m/+ organoids and acutely inhibited FGFR2 for 1 hour with lirafugratinib. While lirafugratinib treatment alone did not affect ERK1/2 phosphorylation, sequential treatment with erlotinib followed by lirafugratinib resulted in a significant reduction in ERK1/2 phosphorylation (Fig. 7D). In line with the genetic results, FGFR2 inhibition by lirafugratinib did not affect T^m/+ organoids growth; however, combined inhibition of FGFR2 and EGFR by lirafugratinib and erlotinib, respectively, further reduced cell growth compared with EGFR inhibition alone (Fig. 7E; Supplementary Fig. S9C). Together, these results indicated that combined inhibition of FGFR2 and EGFR reduces MAPK signaling and proliferation in precancerous organoids.

To test the cooperation between FGFR2 and EGFR in promoting pancreatic tumorigenesis, we performed an in vivo study. Two-month-old KC mice were treated with cerulein to expedite tumorigenesis and concomitantly received lirafugratinib and erlotinib alone or in combination (Supplementary Fig. S9D). The study was limited to 10 days because longer term treatment led to excessive weight loss in the combination arm (Supplementary Fig. S9E). Even in this short course of treatment, we observed a significant reduction in the fraction of diseased pancreatic tissue in the mice that received both lirafugratinib and erlotinib compared with vehicle-treated mice, which was confirmed by immunolabeling for CLU, CK19, and amylase (Fig. 7F–H; Supplementary Fig. S9F). Notably, analysis of phosphorylated ERK1/2 by IF revealed a significant decrease upon erlotinib treatment alone or the combination with lirafugratinib (Fig. 7G; Supplementary Fig. S9F). However, only the combination therapy was able to significantly reduce the percentage of Ki67-positive cells (Fig. 7H). These results indicated that EGFR and FGFR2 cooperate to facilitate mutant KRAS-driven tumorigenesis by promoting proliferation and MAPK pathway activation.

PanINs, the precursor lesions of PDAC, are common in the general healthy population and rarely develop into carcinoma through less-defined mechanisms (3). An increase in gene dosage of mutant KRAS was implicated in driving early tumorigenesis, suggesting that neoplastic progression may be initiated when KRAS signaling exceeds a critical threshold (64). In lung cancer, activation of p53 was triggered only by enhanced oncogenic KRAS signaling, driving the selective pressure against p53 tumor-suppressive function (65). Interestingly, 3D modeling of human PanINs revealed the expansion of a clonal TP53 mutation as low-grade progressed to high-grade PanINs (2). Moreover, analysis of mouse models indicated that pancreas-specific oncogenic KRAS expression in organogenesis gave rise to a normal, functional pancreas and only sporadically promoted ADM and mPanIN formation (19, 66). Furthermore, these mPanINs progressed to PDAC at low frequency. Inflammatory conditions enhanced pancreatic epithelial cell plasticity and selected for subpopulations with increased mutant KRAS activity, which initiated malignant invasion (67). Altogether, these findings indicated that disease progression is determined by mutant KRAS signaling intensity.

In this study, we discovered a novel role for the RTK FGFR2 in enhancing mutant KRAS signaling in early pancreatic tumorigenesis. We found that FGFR2 was induced as soon as KRAS^G12D initiated malignant transformation of acinar cells but was neither expressed in the healthy pancreas nor after injury. Inactivation of Fgfr2 reduced KRAS^G12D signaling activity measured by phosphorylation of ERK1/2 in acinar-enriched epithelial explants and the spontaneous as well as cerulein-accelerated formation of premalignant lesions in Kras^G12D mice. Notably, the surrounding fibroblasts provided the ligands necessary for FGFR2 activation, highlighting the importance of a precancer fibroinflammatory microenvironment to enhance mutant KRAS activity and initiate transformation.

Cancer interception was defined by the Nobel Laureate and molecular biologist Elizabeth Blackburn in 2011 as “the active way of combating cancer and carcinogenesis at earlier and earlier stages” (68). As such, cancer interception includes the use of approaches to prevent malignant development in high-risk patients that carry oncogenic driving mutations and blocking premalignant lesions from evolving to invasive disease (8). Our mouse models carry the oncogenic Kras^G12D mutation in the pancreas, which leads to the development of PDAC with full penetrance (19). Our experiments provide evidence for the prevention of precancerous lesion formation by FGFR2 abrogation and a consequent extension in survival. However, our data do not indicate whether FGFR2 inactivation in established premalignant lesions would block or delay the progression to PDAC. This could be addressed in future studies by performing long-term treatment with the FGFR2 inhibitor or, alternatively, by genetic deletion in dual-recombinase mice. Furthermore, as FGFR2 was found to be highly expressed in classical PDACs, it will be interesting to explore FGFR2 inhibition as a personalized medicine approach for patients with this PDAC subtype. This is of great relevance in light of the finding that mesenchymal and basal-like PDAC cells display increased sensitivity to KRAS inhibition compared with classical PDAC cells, which constitute a reservoir for disease relapse (69, 70).

RTKs present substantial overlap in signaling cascades, allowing for the augmentation of downstream signaling. The RTK EGFR is required for oncogenic KRAS-driven malignant transformation of acinar cells and is currently being explored as a potential cancer interception target (4, 5, 71). Importantly, in this study we found that FGFR2 was upregulated and promoted the growth of precancerous organoids upon EGFR inhibition. Similar results were previously described in non–small cell lung cancer (72). Reciprocally, EGFR was shown to mediate resistance to FGFR2 inhibition in FGFR2 fusion–positive cholangiocarcinoma (73).

Using a mouse model of early pancreatic tumorigenesis, we showed that only combined inhibition of EGFR and FGFR2 was able to reduce mutant KRAS-driven tumorigenesis and proliferation of pancreatic cells. However, a major hurdle of the combination therapy and of RTK inhibitors in general is their toxicity. Drug development approaches and preclinical studies will be required to identify therapeutic strategies that are not overly toxic. Current efforts with EGFR inhibitors have raised the possibility of reducing side effects by employing intermittent dosing, weekly dosing, lower dosing, or a combination of agents (71).

With the increasing number of FGFR2 inhibitors entering the clinic, this study lays the foundation to explore the potential use of these compounds in combination with EGFR inhibitors for PDAC interception. However, further work is required to determine the target population with high risk of developing pancreatic cancer that should receive these therapies and potential adaptative escape mechanisms associated with the redundancy of RTK signaling.

D.A. Tuveson reports grants from the Lustgarten Foundation, Thompson Foundation, Pershing Square Foundation, Simons Foundation, and NIH during the conduct of the study, as well as other support from Leap Therapeutics, Xilis, Mestag Therapeutics, Dunad Therapeutics, Cygnal, grants and personal fees from ONO, and grants from Mestag Fibrogen outside the submitted work. No disclosures were reported by the other authors.

C. Tonelli: Conceptualization, data curation, formal analysis, funding acquisition, investigation, visualization, methodology, writing–original draft, writing–review and editing. A. Deschênes: Data curation, formal analysis. V.A. Gaeth: Investigation. A. Jensen: Investigation. N. Vithlani: Investigation. M.A. Yao: Investigation. Z. Zhao: Resources. Y. Park: Supervision, funding acquisition, writing–review and editing. D.A. Tuveson: Conceptualization, supervision, funding acquisition, writing–review and editing.

We thank Drs. Klingbeil, Caligiuri, and Shakiba for critical reading of this manuscript. This work was performed with assistance from the Cold Spring Harbor Laboratory shared resources, which are supported by the NIH (Cancer Center Support Grant 5P30CA045508: Bioinformatics, DNA Sequencing, Flow Cytometry, Microscopy, Animal, and Animal and Tissue Imaging Shared Resources). This work was performed with assistance from the U.S. NIH Grant S10OD028632-01. The authors are supported by the NIH (Cancer Center Support Grant 5P30CA045508) and the Lustgarten Foundation, where D.A. Tuveson is a distinguished scholar and Director of the Lustgarten Foundation–designated Laboratory of Pancreatic Cancer Research. D.A. Tuveson is also supported by the Thompson Foundation, the Pershing Square Foundation, the Cold Spring Harbor Laboratory and Northwell Health Affiliation, the Northwell Health Tissue Donation Program, the Cold Spring Harbor Laboratory Association, and the NIH (5P30CA045508, U01CA210240, R01CA229699, U01CA224013, 1R01CA188134, and 1R01CA190092). This work was also supported by a gift from the Simons Foundation (552716 to D.A. Tuveson). C. Tonelli was a fellow of the American–Italian Cancer Foundation. Y. Park is supported by the NCI (R50CA211506).