Identification of Key mRNAs as Prediction Models for Early Metastasis of Pancreatic Cancer Based on LASSO

I propose a new method for variable selection and shrinkage in Cox's proportional hazards model. My proposal minimizes the log partial likelihood subject to the sum of the absolute values of the parameters being bounded by a constant. Because of the nature of this constraint, it shrinks coefficients and produces some coefficients that are exactly zero. As a result it reduces the estimation variance while providing an interpretable final model. The method is a variation of the 'lasso' proposal of Tibshirani, designed for the linear regression context. Simulations indicate that the lasso can be more accurate than stepwise selection in this setting.

Inflammation is a fundamental innate immune response to perturbed tissue homeostasis. Chronic inflammatory processes affect all stages of tumour development as well as therapy. In this Review, we outline the principal cellular and molecular pathways that coordinate the tumour-promoting and tumour-antagonizing effects of inflammation and we discuss the crosstalk between cancer development and inflammatory processes. In addition, we discuss the recently suggested role of commensal microorganisms in inflammation-induced cancer and we propose that understanding this microbial influence will be crucial for targeted therapy in modern cancer treatment.

Diagnosis of pancreatic ductal adenocarcinoma (PDAC) is associated with dismal prognosis despite current therapies; therefore new treatment strategies are urgently required. Numerous studies have suggested that epithelial to mesenchymal transition (EMT) contributes to early-stage dissemination of cancer cells and is pivotal for invasion and metastasis of PDAC 1-4 . EMT program is associated with phenotypic conversion of epithelial cells into mesenchymal-like cells in cell culture conditions, albeit such defined mesenchymal conversion (with spindle shaped morphology) of epithelial cells is rare with quasi-mesenchymal phenotypes occasionally observed in the tumor (partial EMT) 5,6 . Most studies exploring the functional role of EMT in tumors have depended on cell culture induced loss-of-function and gain-of-function experiments involving EMT inducing transcription factors such as Twist, Snail and Zeb1 2,3,7-10 . Therefore, the functional contribution of EMT program for invasion and metastasis remains unclear 4,6 and genetically engineered mouse models (GEMMs) to specifically address a causal connection are lacking. Here we functionally probed the role of EMT program in PDAC by generating PDAC GEMMs with deletion of Snail or Twist, two key transcription factors responsible for EMT. EMT suppression in the primary tumor did not alter the emergence of invasive PDAC, systemic dissemination and metastasis. Suppression of EMT led to an increase in cancer cell proliferation with enhanced expression of nucleoside transporters in tumors, contributing to enhanced sensitivity to gemcitabine treatment and increased overall survival of mice. Collectively, our study suggests that Snail or Twist induced EMT program is not rate-limiting for invasion and metastasis but highlights the importance of combining EMT inhibition with chemotherapy for the treatment of pancreatic cancer. We crossed Twist1L/L or Snai1L/L mice with Pdx1-Cre; LSL-KrasG12D; P53R172H/+ (KPC) to generate the Pdx1-Cre; LSL-KrasG12D; P53R172H/+; Twist1L/L (KPC; TwistcKO) and the Pdx1-Cre; LSL-KrasG12D; P53R172H/+; Snai1L/L (KPC; SnailcKO) mice, respectively. The resultant progeny were born in an expected Mendelian ratio, without overt phenotypic findings other than the anticipated emergence of spontaneous pancreatic cancer ( Extended Figure 1A ). Genetic deletion of Snai1 or Twist1 did not significantly delay pancreatic tumorigenesis, alter tumor histopathology features or local invasion ( Figure 1A-C and Extended Table 1 ). KPC; TwistcKO and KPC; SnailcKO mice displayed similar tumor burden compared to KPC control mice ( Extended Figure 1B ), and insignificant difference in overall survival ( Figure 1D ). Loss of Twist1 or Snai1 expression in the pancreas epithelium was confirmed by in situ hybridization coupled with CK8 epithelial immunolabeling ( Figure 1E and Extended Figure 1C ) as well as immunolabeling for Twist and Snail ( Extended Figure 1D ). Suppression of EMT program was significantly noted ( Figure 1F-G, Extended Figure 1E ). Lineage tracing ( Figure 1F ) and immunolabeling of the primary tumor ( Figure 1G ) showed a significant decrease in the frequency of epithelial cells with expression of the mesenchymal marker αSMA (EMT+ cells) and a decrease in expression of EMT inducing transcription factor, Zeb1 ( Figure 1H ). Global gene expression profiling of tumors revealed a decrease in expression of EMT associated genes (including Snai1 and Twist1) in KPC; SnailcKO and KPC; TwistcKO mice compared to KPC control ( Extended Figure 1F ). Loss of Snail and Twist enhanced E-cadherin expression and suppressed Zeb2 and Sox4 expression in cancer cells ( Extended Figure 2A-C ). Snai2 (Slug) expression was restricted to early PanIN lesion in all the experimental groups with no observed expression in advanced tumors and was significantly reduced in KPC; SnailcKO and KPC;TwistcKO mice compared to KPC control mice ( Extended Figure 2D ). While desmoplasia, including extracellular matrix (ECM) and myofibroblasts content ( Figure 1I and Extended Figure 2E-F ), tumor vessel density ( Extended Figure 2G ), intratumoral hypoxia ( Extended Figure 2H ), CD3+ T-cell infiltration ( Extended Figure 2I ), and cancer cell apoptosis was unaffected with Twist/Snail deletion in KPC tumors ( Figure 2A ), the proliferation of cancer cells in mice with suppressed EMT program was significantly increased ( Figure 2B ), as shown previously in mouse models of breast cancers 11-13 . Immunostaining experiments further revealed that EMT+ cancer cells are largely Ki67− ( Extended Figure 3A ). Altogether, the data suggests that EMT program driven by Twist/Snail transcription factors is dispensable for initiation and progression of primary pancreatic cancer. Next, we investigated whether suppression of EMT program impacts invasion and metastasis. The number of YFP+ CTCs from lineage traced KPC and KPC; TwistcKO was found unchanged ( Figure 2C and Extended Figure 3B ), and expression of cancer cell specific KrasG12D mRNA in the blood from KPC, KPC; TwistcKO and KPC; SnailcKO was unaffected ( Figure 2D ), suggesting that suppression of EMT program in pancreatic tumors does not impact the rate of systemic dissemination of cancer cells. Extensive histopathological analyses, coupled with CK19 or YFP immunostaining of distant metastatic target organs, namely the liver, lung and spleen, indicated a similar frequency of metastasis in EMT suppressed tumors when compared to control tumors ( Figure 2E, Extended Figure 3C, Extended Table 1 , and Extended Table 2 ). The metastases were negative for Twist and Snail, and only a few KPC metastatic cells expressed αSMA or Zeb1 ( Extended Figure 3D-F ), while being positive for E-cadherin and Ki-67 ( Extended Figure 3G-H ). The proliferation rate of cancer cells in the metastases was similar in KPC, KPC; SnailcKO and KPC; TwistcKO mice ( Extended Figure 3H ). Collectively, the results indicated that the genetic deletion of Twist1 or Snai1 in PDAC GEMMs did not reduce metastatic disease. To evaluate whether cancer cells from the pancreas with and without EMT program differentially benefited from impaired proliferation to form secondary tumors, we isolated cancer cells from KPC, KPC; TwistcKO and KPC; SnailcKO mice to assay their organ colonization potential. Twist1 was significantly reduced and Snai1 expression was undetectable in cancer cells isolated from Twist and Snail deleted tumors, respectively ( Figure 2F ). Short-term potential to form tumor spheres (associated with putative cancer stem phenotype) appeared similar in TwistcKO and SnailcKO KPC cells when compared to control KPC cells ( Figure 2G ) 3,8,14-16 . Lung colonization frequency following the i.v. injection of KPC cancer cells (Twist or Snail deleted) were similar to the control KPC cancer cells ( Figure 2H ). These results suggest that a favored epithelial phenotype of cancer cells (via suppression of EMT program) did not impact the capacity to form tumor spheres or their ability for organ colonization 17 . Cancer cell EMT program is associated with gemcitabine drug resistance in PDAC patients and in the orthotopic mouse models of PDAC 1,2,8,9,18-23 . Moreover, enhanced frequency of EMT+ cancer cells in pancreatic tumors is associated with poor survival 24,25 . To determine whether EMT program suppression enhances PDAC sensitivity to gemcitabine chemotherapy, we tested the gemcitabine sensitivity of cancer cells with suppressed EMT program in KPC mice. Equilibrative nucleoside transporter ENT1 and concentrating nucleoside transporter CNT3 were significantly upregulated in cancer cells lacking Snail and Twist, while ENT2 expression was unchanged ( Figure 3A-C ). KPC, KPC; SnailcKO and KPC; TwistcKO mice were treated with gemcitabine and tumor burden was monitored by MRI ( Extended Table 3 ). Tumor progression was suppressed in KPC; SnailcKO and KPC; TwistcKO mice when compared to treated KPC control mice ( Figure 3D ). KPC; SnailcKO and KPC; TwistcKO mice treated with gemcitabine showed improved histopathology and increased survival ( Figure 3E-G ). Cancer cells isolated from the tumors of KPC; SnailcKO and KPC; TwistcKO mice showed epithelial morphology ( Extended Figure 4A ) and reduced expression of mesenchymal genes compared to KPC cancer cell lines ( Extended Figure 4B ), however, in tissue culture conditions (2D culture on plastic), equilibrative nucleoside transporters (ENT1/ENT2/ENT3) showed similar expression pattern and expression of concentrating nucleoside transporters (CNT1/CNT3) was not detected ( Extended Figure 4B ). Increased proliferation of KPC; SnailcKO and KPC; TwistcKO cancer cells compared to KPC control cells ( Extended Figure 4C ) likely accounted for the increased sensitivity to gemcitabine and erlotinib in this setting ( Extended Figure 4D ). Next, we crossed the Snai1L/L to the PDAC GEMM, Ptf1a (P48)-Cre; LSL-KrasG12D; Tgfbr2L/L (KTC) to generate Ptf1a (P48)-Cre; LSL-KrasG12D; Tgfbr2L/L; Snai1L/L (KTC; SnailcKO). The KTC model offers a reliable and penetrant disease progression rate with a consistent timeline of death due to PDAC. Similar to the KPC; SnailcKO mice, KTC; SnailcKO deletion exhibited suppression of EMT program but did not impact primary tumor histopathology, lifespan, local invasion, desmoplasia and frequency of apoptosis ( Figure 4F, Extended Figure 5A-E , and Extended Figure 6A ). KTC; SnailcKO mice presented with significantly reduced Zeb1 expression in cancer cells but enhanced proliferation and concentrating nucleoside transporter 3 (CNT3) expression ( Extended Figure 5E ). ENT2 and ENT1 expression were unchanged in KTC; SnailcKO mice compared to KTC mice ( Extended Figure 5E and Extended Figure 6A ). KTC; SnailcKO mice demonstrated enhanced response to gemcitabine therapy, with significant normal parenchymal area and reduced tumor tissue ( Figure 4A-C ). Gemcitabine therapy in KTC; SnailcKO reduced tumor burden ( Figure 4D ) and significantly improved overall survival ( Figure 4E ) of mice when compared to gemcitabine treated control KTC mice. Gemcitabine therapy specifically increased cancer cell apoptosis and removed enhanced proliferation observed in EMT program suppressed tumors ( Figure 4G and Extended Figure 5E ), without impacting the desmoplastic reaction ( Extended Figure 6B ). Overall, these results suggested an enhanced sensitivity of EMT− cancer cells to gemcitabine. Both the equilibrative nucleoside transporter 2 (ENT2) and the concentrating nucleoside transporter 3 (CNT3) were upregulated in EMT suppressed tumors ( Figure 4G ). These data support a possible mechanistic connection between EMT program and resistance to chemotherapy in PDAC. Collectively, our studies provide a comprehensive functional analysis of EMT program in PDAC progression and metastasis. Absence of either Twist1 or Snai1 did not alter cancer progression or the capacity for local invasion or metastasis to lung and liver in PDAC GEMMs. Metastasis occurs despite a significant loss of EMT program with either the deletion of Snail or Twist, and in both settings, Zeb1, Sox4, Slug and Zeb2 are also significantly suppressed. Nevertheless, it is likely that other EMT inducing factors may compensate for the loss of Snail or Twist to induce invasion and metastasis. While PDX-1 is expressed during the development of the pancreas (in early pancreatic buds: all three major lineages of the pancreas-ductal, acinar and beta-islets), its expression is largely repressed in the adult exocrine pancreas 26,27 . Therefore, deletion of Snail or Twist occurs at the embryonic stage and mice are born normal and exhibit normal pancreas histology prior to the onset of cancer. The GEMMs with Snail or Twist deletion develop PanIN lesions at the same frequency as the control mice. One could argue that suppression of EMT program starting from the inception of cancer could have launched compensatory mechanisms to overcome EMT program-dependent invasion and metastasis. However, such compensation is not observed with respect to chemo-resistance and previous studies have demonstrated that EMT program and cancer cell dissemination are observed even before PDAC lesions are detected in KPC mice 4 . Our study demonstrates that EMT program results in suppression of cancer cell proliferation, and suppression of drug transporter and concentrating proteins, therefore, inadvertently protecting EMT+ cells from anti-proliferative drugs such as gemcitabine. The correlation of decreased survival of pancreatic cancer patients with an increased EMT program is likely due to their impaired capacity to respond to gemcitabine, which is a standard of care for most patients 28,29 . Such diminished response to Gemcitabine will likely reflect on such patients also exhibiting higher metastatic disease. Collectively, our study offers the opportunity to evaluate the potential of targeting EMT program to enhance efficacy of Gemcitabine and targeted therapies 30 . Methods Mice Characterization of disease progression and genotyping for the Pdx1-Cre; LSL-KrasG12D; P53R172H/+ (herein referred to as KPC) and Ptf1a (P48)-Cre; LSL-KrasG12D; Tgfbr2L/L (herein referred to as KTC) mice were previously described 31-33 . These mice were bred to Snai1L/L (herein referred to as SnailcKO), Twist1L/L (herein referred to as TwistcKO), and R26-LSL-EYFP 33 . SnailcKO mice were kindly provided by S.J. Weiss, University of Michigan, Ann Arbor. TwistcKO mice were kindly provided by R. R. Behringer (UT MDACC, Houston, TX) via the Mutant Mouse Regional Resource Center (MMRRC) repository. The resulting progeny were referred to as KPC, KPC; SnailcKO, KPC; TwistcKO, KTC, and KTC; SnailcKO mice and were maintained on a mixed genetic background. Both males and females were used indiscriminately. Mice were given Gemcitabine (G-4177, LC Laboratories) via intraperitoneal injection (i.p.) every other day at 50 mg/kg of body weight. Hypoxyprobe was injected in a subset of mice i.p. at 60 mg/kg of body weight 30 minutes prior to euthanasia. For in vivo colonization assay, one million KPC, KPC; TwistcKO and KPC; SnailcKO tumor cells in 100 μL of PBS were injected intravenously via the retro-orbital venous sinus. Four to eleven mice were injected per cell line. All mice were euthanized at 15 days post-injection. All mice were housed under standard housing conditions at MD Anderson Cancer Center (MDACC) animal facilities, and all animal procedures were reviewed and approved by the MDACC Institutional Animal Care and Use Committee. Tumor growth met the standard of a diameter less than or equal to 1.5 cm. Investigators were not blinded for group allocation but were blinded for the assessment of the phenotypic outcome assessed by histological analyses. No randomization method or statistical sample size estimation was used. Histology and histopathology Histology, histopathological scoring, Masson's Trichrome staining (MTS), and Picrosirius Red were previously described 19,33 . Formalin-fixed tissues were embedded in paraffin and sectioned at 5 μm thickness. MTS was performed using Gomori's Trichome Stain Kit (38016SS2, Leica Biosystems). Picrosirius red staining for collagen was performed using 0.1% picrosirius red (Direct Red80; Sigma) and counterstained with Weigert's hematoxylin. Sections were also stained with hematoxylin and eosin (H&E). Histopathological measurements were assessed by scoring H&E stained tumors for relative percentages of each histopathological phenotype: normal (non-neoplastic), PanIN, well-differentiated PDAC, moderately-differentiated PDAC, poorly-differentiated PDAC, sarcomatoid carcinoma, or necrosis. When tumor histology was missing or of poor quality, the mice were excluded from all analyses and this was determined blinded from genotype information. A histological invasion score of the tumor cells into the surrounding stroma was scored on a scale of 0 to 2, with 0 indicating no invasion and 2 indicating high invasion, where invasion is defined as tumor cell dissemination throughout the stroma away from clearly defined epithelial “nests”. Microscopic metastases were observed in H&E stained tissue sections of the liver, lung and spleen. Positivity (one or more lesions in a tissue) was confirmed using CK19 and YFP immunohistochemistry. This data has been presented as a contingency table ( Figure 2E ) and represented as the number of positive tissues out of the number of tissues scored. The “Any” metastasis score is the number of mice positive for a secondary lesion found anywhere throughout the body out of the total number of mice scored. Immunohistochemistry and Immunofluorescence Tissues were fixed in 10% formalin overnight, dehydrated, and embedded in paraffin and 5 μm thick sections were then processed for analyses. Immunohistochemical analysis was performed as described 33 . Heat mediated antigen retrieval in 1 mM EDTA + 0.05% Tween20 (pH 8.0) for one hour (pressure cooker) was performed for Snail and Twist, 10 mM citrate buffer, pH 6.0 was performed for one hour (microwave) for Ki67 or 10 minutes for all other antibodies. Primary antibodies are as follows: αSMA (M0851, DAKO, 1:400 or ab5694, Abcam, 1:400), cleaved caspase-3 (9661, Cell Signaling, 1:200), CD3 (A0452, DAKO, 1:200), CD31 (Dia310M, DiaNova, 1:10), CK8 (TROMA-1, Developmental Studies Hybridoma Bank, 1:50), CK19 (ab52625, Abcam, 1:100), CNT3 (HPA023311, Sigma-Aldrich, 1:400), ENT1 (LS-B3385, LifeSpan Bio., 1:100), E-cadherin (3195S, Cell Signaling, 1:400), ENT2 (ab48595, Abcam, 1:200), Ki67 (RM-9106, Thermo Scientific, 1:400), SLUG (9585, Cell Signaling, 1:200), SNAIL (ab180714, Abcam, 1:100), SOX4 (ab86809, Abcam, 1:200), TWIST (ab50581, Abcam, 1:100), YFP (ab13970, Abcam, 1:1000), ZEB1 (NBP1-05987, Novus, 1:500), and ZEB2 (NBP1-82991, Novus, 1:100). Sections for pimonidazole adduct (HPI Inc., 1:50) or αSMA immunohistochemistry staining were blocked with M.O.M kit (Vector Laboratories, West Grove, PA) and developed by DAB according to the manufacturer's recommendations. Alternatively, for immunofluorescence, sections were dual-labeled using secondary antibodies conjugated to Alexa fluor-488 or -594 or tyramide signal amplification (TSA, PerkinElmer) conjugated to FITC. Lineage traced (YFP positive) EMT analysis was performed on 8 μm thick O.C.T. medium (TissueTek) embedded frozen sections. Sections were stained for αSMA (ab5694, Abcam, 1:400) followed by Alexa fluor-680 conjugated secondary antibody. Bright field imagery was obtained on a Leica DM1000 light microscope or the Perkin Elmer 3DHistotech Slide Scanner. Fluorescence imagery was obtained on a Zeiss Axio Imager.M2 or the Perkin Elmer Vectra Multispectral imaging platform. The images were quantified for percent positive area using NIH ImageJ analysis software (αSMA, Pimonidazole, SLUG, and CD31), percent positive cells using InForm analysis software (Ki-67 and CD3), or scored for intensity either positive or negative (CK19, YFP, ZEB1, ZEB2, SOX4, and Cleaved Caspase-3) or on a scale of 1-3 (E-cadherin) or 1-4 (ENT1, ENT2 and CNT3). In situ hybridization In situ hybridization (ISH) was performed on frozen tumor sections as previously described 34 . In brief, 10 μm-thick sections were hybridized with antisense probes to Twist1 and Snai1 overnight at 65°C. After hybridization, sections were washed and incubated with AP-conjugated sheep anti-DIG antibody (1:2000; Roche) for 90 min at room temperature. After three washes, sections were incubated in BM Purple (Roche) until positive staining was seen. Digoxigenin labeled in situ riboprobes were generated by in vitro transcription method (Promega and Roche) using a PCR template. The following primers were used to generate the template PCR product. Twist1; forward (5’-CGGCCAGGTACATCGACTTC-3’) and reverse (5’-TAATACGACTCACTATAGGGAGATTTAAAAGTGTGCCCCACGC-3’) Snai1: forward (5’-CAACCGTGCTTTTGCTGAC-3’) and reverse (5’-TAATACGACTCACTATAGGGAGACCTTTAAAATGTAAACATCTTTCTCC-3’) Gene Expression Profiling Total RNA was isolated from tumors of KPC control, KPC; TwistcKO and KPC; SnailcKO mice (n = 3 in each group) by TRIzol (15596026, Life Technologies) and submitted to the Microarray Core Facility at MD Anderson Cancer Center. Gene expression analysis was performed using Mouse Ref6 Gene Expression Bead Chip (Illumina). The Limma package from R Bioconductor 35 was used for quantile normalization of expression arrays and to analyze differentially expressed genes between cKO and control sample groups (p ≤ 0.05 and fold change ≥ 1.2). Gene expression microarray data was deposited in GEO (Accession number GSE66981). Genes up-regulated in cells acquiring an EMT program were expected to be down-regulated in the TwistcKO and SnailcKO tumors compared to control tumors. CTC assays Blood (200 μL) was collected from KPC;LSL-YFP and KPC; TwistcKO;LSL-YFP (ROSA-LSL-YFP lineage tracing of cancer cells) mice and incubated with 10 ml of ACK lysis buffer (A1049201, Gibco) at room temperature to lyse red blood cells. Cell pellets were resuspended in 2% FBS containing PBS and analyzed for the number of YFP+ cells by flow cytometry (BD LSRFortessa X-20 Cell Analyzer). The data was expressed as the percent YFP+ cells from gated cells, with 100,000 cells analyzed at the time of acquisition. Whole blood cell pellets were also assayed for the expression of KrasG12D transcripts, using quantitative real-time PCR analyses (described below). Primary pancreatic adenocarcinoma cell culture and analyses Derivation of primary PDAC cell lines were performed as previously described 36 . Fresh tumors were minced with sterile razor blades, digested with dispase II (17105041, Gibco, 4 mg/ml)/collagenase IV (17104019, Gibco, 4 mg/ml)/RPMI for 1 h at 37°C, filtered by a 70 μm cell strainer, resuspended in RPMI/20%FBS and then seeded on collagen I coated plates (087747, Fisher Scientific). Cells were maintained in RPMI medium with 20% FBS and 1% penicillin, streptomycin and amphotericin B (PSA) antibiotic mixture. Cancer cells were further purified by FACS based on YFP or E-Cadherin expression (anti-E-cadherin antibody, 50-3249-82, eBioscience, 1:100). The sorted cells, using BD FACSAria™ II sorter (South Campus Flow Cytometry Core Lab of MD Anderson Cancer Center) were subsequently expanded in vitro. All studies were performed on cells cultivated less than 30 passages. As these are primary cell lines no further authentication methods were applicable and no mycoplasma tests were performed. MTT and drug sensitivity assays MTT assay was performed to detect cell proliferation and viability by using Thiazolyl Blue Tetrazolium Bromide (MTT, M2128, Sigma) following the manufacturer's recommendations with an incubation of two hours at 37°C. For the drug treatment studies, a cell line derived from each of the KPC, KPC; SnailcKO and KPC; TwistcKO mice was treated with 20 μM Gemcitabine (G-4177, LC Laboratories) or 100 μM erlotinib (5083S, NEB) for 48 hours. The relative cell viability was detected using MTT assay with a cell line derived from each of the KPC, KPC; SnailcKO and KPC; TwistcKO mice. N value is defined as biological replicates of a single cell line. Control conditions included 1% DMSO vehicle for erlotinib. The relative absorbance was normalized and control (time 0 hour or vehicle treated) arbitrarily set to 1 or 100% for absorbance or drug survival, respectively. Quantitative real-time PCR analyses (qPCR) RNA was extracted from whole blood cell pellets following ACK lysis using the PicoPure Extraction kit as directed (KIT0214, Arcturus), or from cultured primary pancreatic adenocarcinoma cells using TRIzol (15596026, Life Technologies). cDNA was synthetized using TaqMan Reverse Transcription Reagents (N8080234, Applied Biosystems) or High Capacity cDNA Reverse Transcription Kit (4368814, Applied Biosystems). Primers for KrasG12D recombination are: KrasG12D forward (5’ ACTTGTGGTGGTTGGAGCAGC 3’), KrasG12D reverse (5’ TAGGGTCATACTCATCCACAA 3’). 1/ΔCt values are presented to show KrasG12D expression in indicated experimental groups, statistical analyses were assayed on ΔCt. Primer sequences for EMT related genes are listed in Supplemental Table 1, GAPDH was used as an internal control. The data is presented as the relative fold change and statistical analyses were assayed on ΔCt. Tumor sphere assay Tumor sphere assays were performed as previously described 33 . Two million cultured primary tumor cells were plated in a low-adherence 100mm dish (FB0875713, Fisherbrand) with 1% fetal bovine serum, Dulbecco's modified Eagle's medium, and penicillin/streptomycin/amphotericin. Cells were incubated for seven days and formed spheres were counted at 100x magnification. Three, two and three cell lines were analyzed for KPC control, KPC; TwistcKO and KPC; SnailcKO group, respectively, five field of views per cell line were quantified. MRI Analyses MRI imaging was performed using a 7T small animal MR system as previously described 37 . To measure tumor volume, suspected regions were drawn blinded on each slice based on normalized intensities. The volume was calculated by the addition of delineated regions of interest in mm2 × 1 mm slice distance. None of the mice had a tumor burden that exceeded 1.5 cm in diameter, in accordance with institutional regulations. All mice with measurable tumors were enrolled in the study (see Extended Table 3 ). Mice were imaged twice, once at the beginning of the enrollment (Day 0), and a second time 20 days (Day 19) afterwards. Surviving animals were euthanized at end point (Day 21) for histological characterization. Statistical analyses Statistical analyses were performed on the mean values of biological replicates in each group using unpaired two-tailed or one-tailed t-tests (qPCR only), one-way ANOVA with Tukey's multiple comparisons test using GraphPad Prism, as stipulated in the figure legends. χ2 analyses, using SPSS statistical software, were performed comparing control to cKO groups for metastatic or colonization frequency across multiple histological parameters in all mice and mice ≥ 120 days of age. Fisher's Exact P value was used to determine significance. Results are outlined in Extended Table 2 . Kaplan-Meier plots were drawn for survival analysis and the log rank Mantel-Cox test was used to evaluate statistical differences, using GraphPad Prism. Data met the assumptions of each statistical test, where variance was not equal (determined by an F-test) Welch's correction for unequal variances was applied. Error bars represent s.e.m. when multiple visual fields were averaged to produce a single value for each animal which was then averaged again to represent the mean bar for the group in each graph. P < 0.05 was considered statistically significant. Extended Data Extended Figure 1 A Representative H&E images of small intestine (SmInt), kidney, and heart (scale, 100μm). B Pancreatic mass of (n = 29, 13, and n = 28 mice; s.d.; one-way ANOVA). C Merge of Twist1 or Snai1 in situ hybridization (black) followed by CK8 (red) immunolabeling in tumors from KPC and KPC; TwistcKO or KPC; SnailcKO mice, respectively. White arrows highlight positive cells in the stroma while yellow arrows highlight negative epithelium (scale, 50 μm). D Twist or Snail immunostaining in KPC and KPC; TwistcKO or KPC; SnailcKO tumors, respectively. Black arrows highlight positive cells in the stroma while red arrows highlight negative epithelium (scale, 20 μm). E Channel separations of the representative images of αSMA immunolabeling in YFP lineage traced tumors found in Figure 1F (scale, 50 μm). F EMT gene expression signature analysis in KPC, KPC; TwistcKO and KPC; SnailcKO cohorts (n = 3 mice). Red arrows indicate reduced Twist1 and Snai1 expression in KPC; TwistcKO and KPC; SnailcKO cohorts, respectively. Extended Figure 2 A E-Cadherin immunolabeling and quantification of primary KPC (n = 5 mice), KPC; TwistcKO (n = 5 mice) and KPC; SnailcKO (n = 4 mice) (scale, 100 μm). B Zeb2 immunolabeling and quantification of primary KPC (n = 6 mice), KPC; TwistcKO (n = 5 mice) and KPC; SnailcKO (n = 7 mice) (scale, 50 μm; inset scale, 20 μm). C Sox4 immunolabeling and quantification of primary KPC (n = 7 mice), KPC; TwistcKO (n = 6 mice) and KPC; SnailcKO (n = 8 mice) (scale, 50 μm; inset scale, 20 μm). D Slug immunolabeling and quantification of primary KPC (n = 4 mice), KPC; TwistcKO (n = 4 mice) and KPC; SnailcKO (n = 4 mice) tumors (scale, 50 μm; inset scale, 20 μm). E Sirius Red staining and quantification of primary KPC (n = 21 mice), KPC;TwistcKO (n = 8 mice) and KPC;SnailcKO (n = 11 mice) (scale, 200 μm; s.d.) F αSMA immunolabeling and quantification of primary KPC (n = 5 mice), KPC;TwistcKO (n = 5 mice) and KPC;SnailcKO (n = 5 mice) (scale, 100 μm). G CD31 immunolabeling and quantification of primary KPC (n = 4 mice), KPC;TwistcKO (n = 4 mice) and KPC;SnailcKO (n = 3 mice) (scale, 200 μm, inset scale, 100 μm). H Pimonidazole staining and quantification of primary KPC (n = 4 mice), KPC; TwistcKO (n = 4 mice) and KPC; SnailcKO (n = 4 mice) (scale, 100 μm). I CD3 immunolabeling and quantification of primary KPC (n = 5 mice), KPC;TwistcKO (n = 5 mice) and KPC;SnailcKO (n = 5 mice) (scale, 100 μm; inset scale, 25 μm). Unless otherwise indicated error bars represent s.e.m, and significance determined by One-way ANOVA. *P < 0.05, ** P <0.01, *** P <0.001. ns, not significant. Extended Figure 3 A Immunolabeling of primary tumors (n = 3 mice) for αSMA (red), CK8 (green), Ki-67 (white) and DAPI (blue); yellow arrows point to EMT+ cells (scale, 20 μm). B Representative dot plots of circulating YFP+ cells. C Images of serial sections of KPC; LSL-YFP lung and liver metastasis stained for H&E or immunolabeled for CK19 or YFP. Yellow dashed box represents magnified areas in panel below (scale, 200 μm; magnification scale, 100 μm). D KPC metastatic tumors stained for Twist and Snail (n = 3 mice; scale, 50 μm; inset scale, 20 μm). E Zeb1 immunolabeling and quantification of metastatic KPC (n = 4 mice), KPC; TwistcKO (n = 3 mice) and KPC; SnailcKO (n = 4 mice) (scale, 50 μm; inset scale, 20 μm). F αSMA immunolabeling and quantification of metastatic KPC (n = 3 mice), KPC; TwistcKO (n = 3 mice) and KPC; SnailcKO (n = 3 mice) (scale, 50 μm; inset scale, 20 μm). G E-Cadherin staining on serial sections of αSMA immunolabeling and quantification of metastatic KPC (n = 4 mice), KPC; TwistcKO (n = 3 mice) and KPC; SnailcKO (n = 4 mice) (scale, 50 μm; inset scale, 20 μm).H Ki-67 immunolabeling and quantification of metastatic KPC (n = 7 mice), KPC; TwistcKO (n = 3 mice) and KPC; SnailcKO (n = 3 mice) (scale, 50 μm). Unless otherwise indicated error bars represent s.e.m, percentages indicated represent percent decrease from control, and significance determined by One-way ANOVA. * P <0.05, ** P <0.01, *** P <0.001. ns, not significant. Extended Figure 4 A Brightfield micrograph of cultured primary KPC, KPC; TwistcKO and KPC; SnailcKO cells (scale, 50 μm). B EMT and gemcitabine transport related gene expression shown by qPCR analysis in KPC (n = 3-4 cell lines), KPC; TwistcKO (n = 5 cell lines) and KPC; SnailcKO (n = 5-6 cell lines) (s.d., one-tailed t-test, * P < 0.05, numbers list non-significant P values. nd: not detected, ns: not significant). C MTT assay showing cell proliferation in KPC, KPC; TwistcKO and KPC; SnailcKO cells (n = 8, 8, and 8 biological replicates of a cell line for each genotype). D Relative cell viability (MTT assay) in cultured KPC, KPC; TwistcKO and KPC; SnailcKO cells treated with gemcitabine or erlotinib (n = 8, 8, and 8 biological replicates of a cell line for each genotype). Unless otherwise indicated error bars represent s.e.m, significance was determined by one-way ANOVA. ** P <0.01, *** P <0.001, **** P <0.0001. Extended Figure 5 A Representative H&E images (scale, 100 μm). B Relative percentage of each histological tissue phenotype of KTC (n = 8 mice) and KTC; SnailcKO (n = 6 mice) primary tumors (s.d.). C Primary tumor invasiveness in KTC (n = 8 mice) and KTC; SnailcKO (n = 6 mice) (s.d.). D Pancreatic mass in KTC (n = 5 mice) and KTC; SnailcKO (n = 6 mice) (s.d.). E Immunolabeling and quantification of primary KTC (n = 5 mice), KTC; SnailcKO (n = 4 mice) for αSMA (red), CK8 (green) and DAPI (blue); white arrows indicate double positive cells (scale, 20 μm), Zeb1 (scale, 50 μm; inset scale. 20μm), cleaved caspase-3 (scale, 50 μm; n = 4 and 4 mice), Ki-67 (scale, 100 μm), ENT2 (scale, 100 μm) and CNT3 (scale, 100 μm). Unless otherwise indicated error bars represent s.e.m, and significance determined by two-tailed t-test. * P <0.05, *** P <0.001. ns, not significant. Extended Figure 6 A-B Staining and quantification of (A) KTC (n = 5 or 6 mice), KTC; SnailcKO (n = 4 or 5 mice) (B) KTC + GEM (n = 4 or 5 mice), KTC; SnailcKO + GEM (n = 5 mice) for Masson's Trichrome Stain (MTS) (scale, 200 μm), Sirius Red staining (scale, 200 μm), and ENT1 (scale, 100 μm). Error bars represent s.d. (MTS and Sirius Red) or s.e.m. (ENT1), and significance determined by two-tailed t-test. ns, not significant. Extended Table 1 Pathological spectrum of disease and metastasis in KPC, KPC; TwistcKO and KPC; SnailcKO cohorts. Pathological Spectrum within cohorts ID AGE PDA Differentiation Histology 1 Histology 2 Liver Lung Spleen Any Moribund KPC (104) 1 158 Y W S G Y Y N Y Y 2 165 Y W G N N N N Y 3 148 Y P S G N N - N Y 4 135 Y M S G Y N Y Y Y 5 95 Y M G N Y N Y N 6 42 Y M G N N N N Y 7 55 Y P G S Y N N Y Y 8 91 Y M G N N N N N 9 87 Y W G N N N N N 10 63 Y P G Y Y Y Y N 11 108 Y P S G Y N N Y FD 12 110 Y W G N N N N N 13 104 Y W G Y N N Y Y 14 54 Y W S G N N N N Y 15 108 Y P S G N Y N Y Y 16 42 Y P S G N N N N Y 17 68 Y W G N N N N N 18 107 Y P G N N N N N 19 87 Y P G N N N N N 20 48 Y P G S N N N N Y 21 109 Y P G S Y Y N Y FD 22 81 Y P G Y Y N Y Y 23 151 Y W G N Y N Y Y 24 47 Y M G S N Y N Y Y 25 143 Y P G S N Y N Y Y 26 122 Y W G Y N N Y N 27 115 Y P G Y Y N Y N 28 76 Y W G N Y N Y N 29 122 Y M S G Y N N Y Y 30 97 Y P G N N N N N 31 107 Y W G N N N N N Totals (Median) 31/31 11/31 11/31 2/30 17/31 % 100.0% 35.5% 35.5% 6.7% 54.8% TwistcKO (111) 1 148 Y W G S Y N N Y N 2 151 Y P S G Y Y Y Y N 3 140 Y P G Y Y N Y Y 4 53 Y P G S N N N N Y 5 43 Y P G N N N N Y 6 117 Y P G S N N N N N 7 90 Y P S G Y N N Y Y 8 52 Y P G S N N N N Y 9 104 Y P G N N N N N 10 218 Y P G S N N Y Y Y 11 153 Y P G N Y N Y Y 12 45 Y P G S N N N N Y 13 77 Y P G S Y N N Y Y 14 126 Y P G S Y Y N Y Y Totals (Median) 14/14 6/14 4/14 2/14 8/14 % 100.0% 42.9% 28.6% 14.3% 57.1% SnailcKO (103) 1 144 Y W G N Y N Y N 2 51 Y P G S N N N N Y 3 105 Y P G S N Y N Y Y 4 111 Y P G N N N N N 5 106 Y P G S Y N Y Y Y 6 129 Y P G N N N N N 7 102 Y P G S N Y - Y N 8 98 Y P G S Y N Y Y N 9 47 Y P G S N N N N Y 10 54 Y W G Y Y N Y FD 11 59 Y M G Y N N Y N 12 103 Y P G Y N N Y N 13 60 Y P S G Y N Y Y Y 14 77 Y P G Y N N Y Y 15 57 Y M S G Y N N Y FD 16 130 Y P G Y Y N Y FD 17 76 Y P G S N N N N FD 18 111 Y P G N Y N Y Y 19 100 Y P G S Y N Y Y FD 20 104 Y P G S Y N N Y Y 21 124 Y M G N N N N FD 22 88 Y P G S N N N N Y 23 192 Y W G Y Y N Y Y 24 122 Y P G N N N N Y 25 60 Y W G S N N N N Y 26 112 Y W G N Y N Y N 27 48 Y P G S N N N N Y 28 48 Y P G S N N N N Y 29 124 Y P G S Y Y Y Y N 30 215 Y W G N N N N N Totals (Median) 30/30 13/30 9/30 5/29 18/30 % 100.0% 43.3% 30.0% 17.2% 60.0% Key: (Y) yes. (N) no, (W) well, (M) moderate, (P) poor, (G) glandular, (S) sarcomatoid, (FD) found dead, (-) no tissue Extended Table 2 Results of χ2 analysis reporting Fisher's Exact P value. χ2 Analysis Group Perameter Fisher's Exact P value Differentiation All Ages Control vs. TwistcKO Early Tumor progression 0.458 Control vs. SnailcKO 0.106 Control vs. TwistcKO Late Tumor progression 0.458 Control vs. SnailcKO 0.106 Control vs. TwistcKO Sarcomatoid 0.108 Control vs. SnailcKO 0.446 Differentiation ≥120 days Control vs. TwistcKO Early Tumor progression 0.580 Control vs. SnailcKO 0.569 Control vs. TwistcKO Late Tumor progression 0.580 Control vs. SnailcKO 0.569 Control vs. TwistcKO Sarcomatoid 1.000 Control vs. SnailcKO 0.119 Metastasis All Ages Control vs. TwistcKO Liver Metastasis 0.744 Control vs. SnailcKO 0.358 Control vs. TwistcKO Lung Metastasis 0.743 Control vs. SnailcKO 0.786 Control vs. TwistcKO Spleen Invasion 0.581 Control vs. SnailcKO 0.254 Control vs. TwistcKO Any Metastasis 1.000 Control vs. SnailcKO 0.797 Metastasis ≥120 days Control vs. TwistcKO Liver Metastasis 0.627 Control vs. SnailcKO 1.000 Control vs. TwistcKO Lung Metastasis 0.592 Control vs. SnailcKO 1.000 Control vs. TwistcKO Spleen Invasion 0.559 Control vs. SnailcKO 1.000 Control vs. TwistcKO Any Metastasis 0.473 Control vs. SnailcKO 0.608 Extended Table 3 Pathological spectrum of disease and metastasis in KPC, KPC; TwistcKO and KPC; SnailcKO cohorts treated with Gemcitabine KPC Gemcitabine cohorts ID Start Age (Days) Start Volume (mm3) End Volume (mm3) Survival (Days) KPC + GEM (89) (13) 1 148 1610.351 D 7 2 72 29.736 D 13 3 72 439.795 902.759 21 4 80 44.14 D 14 5 100 536.304 592.31 21 6 89 166.968 D 2 7 94 52.734 D 7 6 122 90.211 D 14 9 164 217.919 D 8 10 143 212.817 D 18 11 84 323.829 897.217 21 12 58 76.734 D 4 13 58 116.186 D 8 Mean (Median) 301.4 797.4 Stdev 406.9 145.1 TwistcKO + GEM (79) (21) 1 117 243.0 644.2 21 2 75 47.2 180.0 21 3 75 45.4 460.9 21 4 78 54.6 47.5 21 5 46 53.7 66.5 21 6 96 63.1 D 13 7 90 23.9 D 13 8 79 101.0 D 14 9 52 28.5 D 14 10 52 49.4 98.706 21 11 104 43.4 127.0 21 12 104 53.5 12.1 21 13 68 56.7 D 15 14 122 650.1 164.1 21 15 104 181.8 78.6 21 Mean (Median) 113.0 187.9 Stdev 154.8 193.0 SnailcKO + GEM (96) (21) 1 188 255.2 D 12 2 181 854.7 D 4 3 127 32.0 59.6 21 4 127 58.7 107.4 21 5 142 109.8 D 14 6 54 33.6 57.2 21 7 89 17.0 D 13 8 78 54.9 39.6 21 9 78 3.1 D 15 10 104 209.7 134.3 21 11 96 220.0 280.2 21 12 96 24.1 46.2 21 13 119 711.0 D 18 14 126 655.6 805.4 21 15 119 168.6 D 18 16 82 453.8 517.4 21 17 82 56.7 74.1 21 18 90 40.0 D 16 19 67 80.5 D 10 20 66 49.5 226.2 21 Mean (Median) 204.4 213.4 Stdev 250.7 231.7 Key: (D) died Supplementary Material Supplemental Table 1