Crizotinib Inhibits Metabolic Inactivation of Gemcitabine in c-Met–driven Pancreatic Carcinoma

Pancreatic ductal adenocarcinoma (PDAC) is the fourth leading cause of cancer-related deaths (1, 2). Despite extensive clinical efforts, the prognosis of this malignancy has not improved in the past decade, with less than 5% of patients alive 5 years after diagnosis (3). This dismal outcome may be explained by the lack of biomarkers for early screening/diagnosis and by the resistance to currently available chemotherapy regimens. Experimental therapeutic agents have thus far showed only limited effects. This may be explained by the selection of these drugs on the basis of their activity in preclinical models that do not recapitulate the complex molecular and histopathologic hallmarks of PDAC. Another explanation for some of the unfavorable clinical responses may be the lack of drugs that combine synergistically with gemcitabine, the standard of care treatment. Gemcitabine is metabolically inactivated by cytidine deaminase (CDA), limiting its therapeutic potential. Interestingly, Frese and colleagues reported that nab-paclitaxel potentiates the effect of gemcitabine by reducing the CDA-mediated inactivation of gemcitabine (4). In terms of preclinical models, the most commonly used PDAC models include (i) established cell line xenografts and (ii) genetically engineered mouse models (GEMM). Unfortunately, long-term culturing of PDAC cells results in distinct and irreversible loss of important genetic and biologic properties, including complex genomic aberrations affecting critical signaling pathways (5), maintenance of a distinct stem cell population (6, 7), and ability to metastasize (8). GEMMs of PDAC provide an essential tool to probe the function of specific genes in tumor initiation and progression (9). These models have the advantage to maintain an intact immune system and present tumors histologically similar to human PDACs, including a dense desmoplastic stroma reaction (10). However, several targets identified using GEMMs do not have human counterparts, while other important genes in human PDAC are not expressed in mice (11). Moreover, their dependence on limited critical genetic lesions, such as mutations in K-Ras, P53, and CDKN2A/P16, may not fully reflect the genetic diversity that characterizes human PDAC.

Early passages of primary PDAC cells may better mimic the genetic characteristics of the disease and might be better predictors of drug activity (12), including the standard treatment with gemcitabine (3). Recently, Hidalgo and colleagues developed primary models by implanting pieces of human PDACs subcutaneously into mice (13, 14). However, subcutaneously implanted tumors may not optimally recapitulate many of the essential features of tumor growth in patients, such as the ability to metastasize (15).

The aim of the present study was to develop orthotopic xenograft models using low passage primary PDAC cells injected in the pancreas of mice, preserving the genetic background and heterogeneity of human PDAC, while maintaining the macro- and microenvironmental interactions. The PDAC tumors were further characterized and monitored by bioluminescence, MRI, and high-frequency–ultrasound imaging. To demonstrate the feasibility of our PDAC models for drug discovery, we selected one of the models that was characterized by amplification and overexpression of c-Met—overactive in approximately 45% of patients with PDAC (16). Treatment with the combination of the c-Met/ALK inhibitor crizotinib and gemcitabine significantly reduced tumor growth and metastases. In addition, this PDAC model allowed for validation of a synergistic mechanism, i.e., crizotinib-mediated inhibition of CDA resulting in diminished gemcitabine inactivation.

Four primary PDAC cell cultures (PDAC-1/2/3/4) were isolated from 10 consecutive patients (40% efficiency) undergoing pancreaticoduodenectomy, according to a protocol approved by the Ethic Committee at Pisa University Hospital (Pisa, Italy). Sterile nonnecrotic tumors were rinsed and enzymatically dissociated. Medium was replaced every three days, until cell colonies were identified, which occurred after about 1 week. More information can be found in Supplementary Methods.

Within 10 passages, the resulting primary cultures were transduced with two lentiviral vectors encoding firefly-luciferase-mCherry (Fluc-mCherry, FM) and Gaussia-luciferase-CFP (Gluc-CFP, GC), as described previously (17, 18). Transduction efficiency was evaluated using Leica-MM-AF-NX fluorescence microscope (Leica) and by fluorescence-activated cell sorting (FACS) analysis (Becton Dickinson).

Animal experiments were approved by the Committee on Animal Experiments of the VU University Amsterdam, the Netherlands. Orthotopic primary pancreatic cancer mouse models were generated via injection of 8 × 10⁵ primary cells into the pancreas of three 6- to 8-week-old female athymic nude mice per each PDAC-FM-GC cell culture type (Harlan). Tumor growth was monitored using bioluminescence imaging (BLI) of both Fluc and Gluc reporters, as described in detail in the Supplementary Methods. Mice were sacrificed upon discomfort or more than 10% weight loss and log-rank test was used to evaluate significant differences in survival. Pancreas and internal organs were removed and frozen or fixed in 4% paraformaldehyde.

Frozen sections from xenografts and originator tissues were stained with hematoxylin and eosin (H&E), and areas of tissue clearly affected by PDAC were dissected using the Leica-LMD6000 instrument, as described previously (19, 20). These laser-microdissected specimens were subjected to array Comparative Genomic Hybridization (array-CGH), in parallel with the primary culture and the originator tumor specimen, using Agilent 4 × 180 K platform (Agilent), as described, together with K-Ras mutation analyses, in the Supplementary Methods. Histopathologic analyses were performed on paraffin-embedded sections of pancreatic, liver, lung, and lymph node specimens. Immunohistochemistry (IHC) was executed according to standard procedures with the panel of antibodies, cytokeratin 7 (CK7), CK8/CK18, CK19, CEA, Ca19.9, EGF receptor (EGFR), vimentin, and chromogranin, routinely used in PDAC IHC (dilution 1:200). The percentages of hypovascular and hypoxic tumors were evaluated by IHC of CD31 and carbonic anhydrase IX (CAIX), respectively (21, 22). Visualization was obtained with the BenchMark Special Stain Automation system.

The copy number of the c-MET gene was determined by the TaqMan copy-number-assay (Hs04933308_cn). PCRs were performed on the ABIPRISM-7500 instrument, as described previously (23). Further details are reported in the Supplementary Methods.

RNA was extracted using the Ambion-RecoverAll kit, and cDNA obtained with the DyNAmo cDNA Synthesis Kit (Thermo Scientific). The samples were amplified using specific primers and probes for c-Met (TaqMan Hs01565584_m1) using the ABIPRISM-7500. Gene expression values were presented as |$2^{-{\rm \Delta \Delta}C_{\rm t}}$| value, calibrated to the value of HPNE immortalized ductal cells, and normalized to β-actin.

Protein expression of c-Met and phospho-c-Met was evaluated in all the human tumors as well as in the PDAC cells. IHC of the tissues was performed as described earlier, while cells were grown in Chamber-Slides System (Lab-Tek), incubated with specific monoclonal rabbit anti-human c-Met and anti-phospho-Y1003-c-Met antibodies (1:200 dilution; Santa Cruz Biotechnology), and then stained with avidin–biotin–peroxidase complex (Cell Marque HRP Detection System).

The cell growth inhibitory effects of three c-Met inhibitors, the non–ATP-competitive small molecule tivantinib (Bio-Connect), the ATP-competitive small molecule crizotinib (Bio-Connect), and the monoclonal antibody against the c-Met extracellular domain DN-30 (kindly provided by Dr. Paolo Comoglio, IRCC Candiolo, Italy), were determined by sulforhodamine-B (SRB) assay in PDAC cells, as described previously (23, 24).

To evaluate the modulation of c-Met and phospho-c-Met expression, Western blot analyses were performed in cells treated for 2 hours with 1 μmol/L crizotinib, after 10-minute exposure to 40 ng/mL of the c-Met ligand hepatocyte growth factor (HGF). Membranes were incubated with the anti-c-Met and anti-phospho-Y1003-c-Met antibodies described earlier (1:1,000 dilution), and mouse anti-β-actin (1:50,000; Sigma-Aldrich). Membranes were probed for 1 hour with goat anti-mouse InfraRedDye or goat anti-rabbit InfraRedDye (1:10,000; Westburg) secondary antibodies. Fluorescent proteins were detected by an Odyssey Infrared Imager (LI-COR Biosciences) and quantified with the Odyssey 3.0 software.

Five days after surgery, mice were stratified on the basis of BLI intensities into four groups with comparable mean Fluc activity. Five mice per group were treated with 25 mg/kg crizotinib [dissolved in dimethyl sulfoxide (DMSO) and 0.5% methylcellulose, administered via oral gavage], 100 mg/kg gemcitabine (in sterile saline; intraperitoneally), or their combination, q3d × 4 (25, 26), whereas 5 control mice were not treated.

Additional imaging analyses to define tumor spatial characteristics and evaluate microenvironment structures, such as neovasculature, were carried out in 6 mice (2 PDAC-2-FM-GC, and 4 PDAC-3-FM-GC mice, including 2 mice treated with crizotinib), together with 2 healthy mice, by MRI (Achieva 3.0T X-series MRI, Philips) and high-frequency ultrasound including Power Doppler Mode (Vevo-2100; VisualSonics). All MRI scans were T1 and T2 weighted images. MR scanner bundled image analysis software (ImageJ 1.46) and the National Institute of Environmental Health Sciences (NIEHS) of NIH, MR-Visible-Mouse atlas, were used for analysis of the images.

HGF levels were quantified in medium from 1 × 10⁵ PDAC-1/2/3/4 cells using a specific ELISA kit (Quantikine Human HGF Immunoassay; R&D diagnostics). Similar measurements were performed in plasma samples from three PDAC-1/2/3/4-FM-GC mice, at day 14.

Accumulation of gemcitabine, gemcitabine nucleotides, and crizotinib was measured in cells, plasma, and pancreatic tumors of the PDAC-3-FM-GC model using liquid chromatography/tandem mass spectrometry (LC/MS-MS), as described in the Supplementary Methods. Plasma samples were drawn from 3 PDAC-3-FM-GC mice, 1 and 2 hours after treatment with gemcitabine, crizotinib, and their combination, at days 5, 8, and 14.

Because CDA is the main enzyme in gemcitabine catabolism, we evaluated its enzymatic activity in cells, as well as blood samples and tumor tissues from PDAC-3-FM-GC mice, as described in the Supplementary Methods. Moreover, because previous studies suggested that CDA can be degraded through reactive oxygen species (ROS; ref. 4), we evaluated modulation of ROS levels with a specific assay (Cell Biolabs).

All experiments were performed in triplicate and repeated at least twice. Data were expressed as mean values ± SEM and analyzed by a Student t test or ANOVA followed by the Tukey multiple comparison test. The level of significance was P less than 0.05.

In the present study, we first sought to develop novel orthotopic mouse imaging models of PDAC from human primary tumor cells, as outlined in Fig. 1A. Four primary cell cultures were successfully cotransduced using two lentiviral vectors, expressing Fluc-mCherry (FM) and Gluc-CFP (GC), with transduction efficacy of more than 90% (Fig. 1B, inset). The BLI signals increased proportionally to the cell number (Fig. 1B), indicating that both Fluc and Gluc signal intensities can be directly related to tumor cell growth. Next, because Gluc is secreted in the medium, we also monitored Gluc activity over time in the conditioned cell culture medium (Supplementary Fig. S1C), showing increased signal intensity over a span of time including at least one doubling time of these cells. Next, PDAC primary cells were injected orthotopically into the pancreas of 3 athymic mice. All four primary transduced PDAC cell cultures successfully engrafted, and the tumors developed in all the mice injected (100% take rate) and expanded over time, as shown by the increase in Fluc and Gluc signal intensities in Fig. 1C and D and Supplementary Fig. S1A and B. The PDAC-3-FM-GC mice had the shortest median survival (25 d) followed by 39 and 46 days in PDAC-2-FM-GC and PDAC-4-FM-GC, respectively, whereas mice engrafted with the PDAC-1-FM-GC cells had the longest survival (59 d), following the survival trend observed in the patients with PDAC (Table 1 and Supplementary Fig. S1C), although the number of models does not warrant a statistically supported survival correlation. To further define the spatial characteristics and vasculature of the PDAC tumors, we used MRI and high-frequency ultrasound with Doppler analysis, respectively. Twenty days after cell implantation, 2 mice bearing PDAC-2-FM-GC and PDAC-3-FM-GC tumors underwent MR scans, together with a healthy control mouse. MR images of both mouse models confirmed the localization of tumor cells in the pancreas, as well as retroperitoneal invasion (Fig. 1E), accompanied by enlargement of the mouse abdomen cavity in the sagittal and transversal MRI scans. Twenty-eight days after PDAC cells injections, we studied 6 mice bearing PDAC-2-FM-GC and PDAC-3-FM-GC with high-frequency ultrasound in 3D Power Doppler Mode enabling the measurement of tumor volumes and assessment of networks of vasculature, within living tissues (Fig. 1F and G and Supplementary Video S1). This imaging was able to visualize lesions less than 900 μm in diameter, revealing hypovascular pancreatic tumors.

To characterize our mouse models and to investigate whether they recapitulate the histopathology and immunophenotype of human PDACs, we performed both H&E and IHC analyses. In Fig. 2, we show the main histophatologic characteristics of PDAC-3-FM-GC tumors compared with surgical material resected from the human originator tumor, as representative example for all four models. Figure 2A–D and 2F–I show tissue sections stained with H&E from the human tumor tissue and orthotopic PDAC-3-FM-GC, respectively. All the PDAC-FM-GC mouse models showed key histopathologic features of human PDAC in terms of tumor infiltration (Fig. 2A vs. F), stroma and PDAC-associated desmoplastic reaction (Fig. 2B vs. G), ductal characteristics and adenocarcinoma differentiation (Fig. 2C vs. H), and inflammation (Fig. 2D vs. I). Representative H&E images of human and murine pancreatic healthy tissues (Fig. 2E–J) are depicted to highlight structural differences when compared with images of pancreatic tissue derived from the patients and mice with PDAC. Our PDAC models react positively to human specific antibodies directed against CK8/18, CK7, CK19, Ca19.9, EGFR, and CEA. Conversely, the tumors were negative for vimentin, and chromogranin (Supplementary Fig. S2A–H), and therefore at least partly recapitulate the characteristic human PDAC immunophenotype. In 70% of the xenografts, the blood vessels were organized in the stroma surrounding the tumor nests and did not invade into neoplastic masses. Similarly, most tumors showed high expression of the hypoxia marker CAIX (Supplementary Fig. S2I and J). Moreover, the tumors obtained from our models had increased mitotic activity, and were negative for additional IHC studies with mouse specific antibodies (Supplementary Fig. S3), demonstrating that all our PDAC-FM-GC mouse models are of human origin.

Of note, most of the orthotopic tumors metastasized to other organs, such as lymph nodes, liver, and lung as shown in Fig. 2K–M. The presence of these metastases was first detected via BLI and then confirmed via light microscopy of three nonsequential serial sections stained with H&E. The IHC analyses of these metastases showed staining similar to the primary tumors (Supplementary Fig. S2K–M). Macroscopic metastases were observed in all the livers of the PDAC-3-FM-GC and PDAC-2-FM-GC mice, whereas no liver metastases were detected in 33% of the mice of the PDAC-1-FM-GC and PDAC-4-FM-GC groups. Moreover, we visualized lung metastases in 2 of the PDAC-3-FM-GC and 1 of the PDAC-2-FM-GC animals. Thus, the in vivo metastatic potential of our models follows the typical routes of invasion to lymph nodes, liver, and lungs, as observed in patients.

After detecting a similar phenotype, we investigated whether the genetic signature of the human originator tumors was preserved in the cultured primary cells and in the mouse xenografts. DNA was extracted from cells and laser microdissection (LMD)–isolated tissues from human specimens and xenografts (Fig. 3A), and analyzed by array-CGH (Fig. 3B and Supplementary Fig. S4A and B). The statistical analyses (Fig. 3C) demonstrated the significant correlations between the different array-CGH datasets, indicating that both primary cultures and xenografts shared the same clonal origin of the human originator tumor. Moreover, the mutation analysis of K-Ras on the PDAC-3 originator tumor and xenograft detected the same mutation (GGT>GAT) in the codon 12 (Supplementary Fig. S4C). Future extensive mutational analysis of our models will further demonstrate whether the mutational spectrum among our tumor/cells/orthotopic PDAC models overlaps or diverges.

The copy-number variations observed in the PDAC cells and tissues were comparable with the genomic imbalances that we observed in a wide cohort of patients with PDAC (19), including copy-number gains in chromosomes 11q13.1 and 20q13.1 and losses in chromosomes 9p, 10q11.22, 18q12.2-23, and 19q13.31. Interestingly, the PDAC-3 specimens had high copy-number gain of the c-MET gene, located on the cytoband 7q31.2 (Fig. 4A), in contrast with the other PDAC models. These results were validated by copy-number analysis of the c-MET gene with a specific PCR assay (Fig. 4B). In accordance with these data, PCR and IHC analyses revealed that PDAC-3 tumors and cells were characterized by c-Met overexpression (Fig. 4C and D). Furthermore, PDAC-3-FM-GC mice had a significantly higher expression of HGF than the other PDAC-FM-GC models (Supplementary Fig. S5), which might be caused by both the HGF copy-number gain (cytoband 7q21.1 includes the HGF gene) and a potential HGF/c-Met feed-forward loop.

The concept of blocking c-Met is appealing because a number of studies have shown that the activated c-Met pathway is associated with poorer prognosis and chemoresistance. Therefore, we tested the activity of three novel c-Met inhibitors, tivantinib, crizotinib, and DN-30. All these compounds were more effective in the c-Met^high PDAC-3 cells compared with c-Met^normal cells, and crizotinib emerged as the most active inhibitor of cell growth, at all the tested drug concentrations, with an IC₅₀ value of 0.5 μmol/L in the PDAC-3 cells (Fig. 4E and F). Representative growth inhibition curves for PDAC-3 cells are shown in Fig. 5A. Median drug-effect analysis revealed a strong synergistic interaction of crizotinib with gemcitabine in these cells. Synergism was also observed in the other PDAC cells (Fig. 5B), in parallel with an increase in the accumulation of gemcitabine nucleotides (Fig. 5C). Because previous studies reported that gemcitabine is susceptible to CDA-mediated inactivation (4, 23), we evaluated CDA activity upon crizotinib treatment in the cells, demonstrating a significant reduction in CDA activity upon treatment with crizotinib in the PDAC-3 and PDAC-1 cells (Fig. 5D and Supplementary Fig. S6A). This explains, at least in part, the synergistic effect observed between crizotinib and gemcitabine in the PDAC-3 cells as well as the increase of gemcitabine levels, in both the PDAC-1 and PDAC-3 cells, suggesting that the crizotinib/gemcitabine combination might be synergistic regardless of c-MET status. The reduction of CDA could be caused by the degradation mediated by crizotinib-induced ROS (Fig. 5E and Supplementary Fig. S6B). Conversely, no modulation of CDA activity was determined by direct interaction with crizotinib (Supplementary Fig. S6C).

In the PDAC-3-FM-GC mouse model, starting from day 22, mean Fluc intensity was significantly (P = 0.002) higher in the nontreated mice than in mice receiving crizotinib alone (Fig. 6A). Moreover, the group receiving the combination had significantly lower Fluc intensity than mice receiving either gemcitabine or crizotinib alone (e.g., at day 30, 76% and 48%, respectively). Importantly, the results of measurement of Gluc intensities in blood were similar to Fluc read-out (i.e., at day-30, Gluc intensity in the combination was 4-fold decreased vs. control group; Fig. 6B). Tumor growth inhibition was reflected in a significant longer survival of mice treated with the combination than the single arm treatments and control mice (P < 0.01; Fig. 6C). Notably, the BLI signals quantified with our Fluc/Gluc double imaging reporter system were proportional to the tumor volumes as measured with high-frequency ultrasound in 4 PDAC-3-FM-GC mice. Moreover, BLI signals correlated with the size of necropsy specimens (Supplementary Fig. S7).

To verify that crizotinib and gemcitabine do reach the tumor in vivo, we used a specific validated LC/MS-MS method (27). Crizotinib and gemcitabine were detected in all samples (Supplementary Fig. S8A and B). In the tumor samples from mice treated before their sacrifice, we detected crizotinib and gemcitabine in the range of 2.5 (gemcitabine in the gemcitabine-alone group) to 21.5 (crizotinib in the combination group) ng/mg, respectively. Considering the conversion of 1 g tissue to 1 mL liquid, the highest tissue concentration of crizotinib was about 100-fold higher than the IC₅₀ of crizotinib in cell culture (i.e., 47.7 μmol/L). In the combination group, the concentration of crizotinib in pancreatic cancer specimens was considerably higher than in the mice treated with crizotinib alone, whereas gemcitabine concentration was about 2-fold higher in the combination group than in the mice treated with gemcitabine alone (Fig. 6D). Moreover, in the blood samples of the combination group, 2 hours after drug administration, we observed a significant increase of the concentration of crizotinib compared with the samples taken after 1 hour (P < 0.05). The concentration of crizotinib was also 1.4-fold higher in mice that received the combination than single-drug treatment (Supplementary Fig. S8C), suggesting a time-dependent accumulation of the drug as well as a favorable pharmacokinetic interaction with gemcitabine. Similarly, we observed a significant increase of gemcitabine in the combination group with respect to the mice treated with gemcitabine alone. In keeping with the results obtained in vitro, we observed that the enzymatic activity of CDA was significantly reduced in all the samples from mice treated with crizotinib and crizotinib–gemcitabine combination (Fig. 6E and F). These results could explain why the accumulation of gemcitabine was significantly increased both in the blood and in the tissue samples from mice treated with the combination with crizotinib compared with gemcitabine alone.

Here, we describe the development and use of orthotopic PDAC mouse imaging models for the identification of new PDAC drug combinations. First, we determined the histopathologic characteristics of the primary PDAC tumors in the mouse pancreas as well as of the systemic metastases. In particular, the PDAC tumors developed a stromal reaction, and both histopathologic and high-frequency ultrasound analyses revealed a hypovascular tumor tissue. Because several studies in patients with PDAC and GEMM PDAC suggest that poor drug delivery and response is attributable to highly desmoplastic and hypovascular tumors (10, 28), it is important that our models recapitulate these environmental factors.

Most of the anticancer drugs in clinical trials are selected on the basis of their activity in preclinical models (29), which may not recapitulate the molecular and histopathologic characteristics of human PDAC. This includes established cell lines cultivated as monolayers or their xenografts, which undergo substantial changes that deviate from their originator tumors because of clonal selection over many passages in vitro. This genetic drift leads to a homogeneous population of cells that may not model the intratumoral heterogeneity of human PDAC, which is critical for experimental testing of anticancer drugs, because heterogeneity provides the foundation for the selection of resistant subpopulations (30). Recently, primary PDAC xenograft models were developed, in which resected human PDAC tissues were implanted subcutaneously in immunodeficient mice (14). These models preserved tumor genetics and heterogeneity, representing an optimal model to study the drug interaction with complex human gene alterations. The engraftment efficiency was about 61% (14), a higher efficiency rate than that we report here for the establishment of primary PDAC cell cultures. However, several studies demonstrated that subcutaneously xenograft models often do not recapitulate the essential features of tumor growth, native tumor microenvironment, and locally invasive and metastatic disease (31), supporting the development of orthotopically implanted tumors as a critical preclinical tool for testing new agents.

Another crucial step for the development of the PDAC mouse models consisted of the application of BLI, a low-cost longitudinal in vivo imaging method (32). Recently, BLI was used in the evaluation of delivered cell response in pancreatic islet stem-cell transplantation, as well as in an orthotopic model of MiaPaCa-2 pancreatic carcinoma cells (33, 34). Most BLI studies used Fluc, which catalyzes the oxidative decarboxylation of luciferin in the presence of ATP, O₂, and Mg²⁺, producing yellow–green light. In the present study, this method was complemented with the Gluc assay, which provides another sensitive and quantitative assessment of transduced tumor cells through the analysis of naturally secreted Gluc that can be measured in the blood of mice and thereby facilitate less-invasive monitoring (17). Indeed, the combination of Fluc and Gluc assays provided a reliable indicator for localizing and quantifying orthotopic pancreatic tumors and metastases, and was compatible with MR and ultrasound imaging methods.

The relative inefficacy of currently available therapeutic strategies in PDAC has also been attributed to the high rate of genetic alterations affecting multiple pathways, whose multilevel cross-stimulation can overcome drug activity (5). Therefore, after finding a similar phenotype, we performed a whole genome investigation to evaluate whether the genetic signatures are preserved in the primary tumor cells cultured in vitro and in mice xenografts, as compared with the human originator tumors. All our preclinical models showed genetic abnormalities similar to the respective originator PDAC tumors, and genotypic heterogeneity was reflected by an average of more than 50 different aberrations. Hence, these models can be used as reliable tools for understanding the role of complex PDAC genetic characteristics to test targeted drugs and to ultimately improve patient care.

Among these genetic aberrations, we validated by PCR analysis the copy-number gain of the gene encoding for the tyrosine kinase c-Met, resulting in c-Met mRNA and protein overexpression in the PDAC-3 model. This transmembrane receptor is encoded by a gene located at chromosome 7, which has been identified as a proto-oncogene (35, 36). c-Met plays an important role in the control of tissue homeostasis under normal physiologic conditions (37). Conversely, its abnormal stimulation, through overexpression (with or without gene amplification or mutations) or aberrant autocrine or paracrine HGF ligand production, mediates the activation of a wide range of different cellular signaling pathways involved in proliferation, scattering, and invasion (38). Accordingly, c-Met has been shown to be upregulated in many solid tumors, including PDAC (39, 40), in which its stimulation has been associated with poor prognosis and chemoresistance (38, 41). Moreover, it has been shown that c-Met plays a key role in the cancer/stroma interaction (10), and is a marker for pancreatic cancer stem cells, representing an optimal target to prevent the development of metastases (42). Against this background, studies on c-Met inhibitors can provide optimal novel anticancer drugs in c-Met–overexpressing tumors. Therefore, as a proof-of-principle for the feasibility and applicability of our orthotopic PDAC-FM-GC models to test novel treatment strategies, we administered gemcitabine alone or in combination with crizotinib in mice with PDAC-3-FM-GC tumors. Crizotinib has recently been approved for the treatment of locally advanced or metastatic anaplastic lymphoma kinase (ALK)–positive non–small cell lung cancer (NSCLC), as detected by a U.S. Food and Drug Administration–approved biomarker test. However, crizotinib suppresses autophosphorylation of both c-Met and ALK receptor tyrosine kinases and competes with ATP binding in both kinases through its interaction at the hinge region of c-Met (43). Indeed, Ou and colleagues showed that patients with NSCLC with c-MET amplification, but without ALK rearrangement, experienced a rapid and durable response to crizotinib, demonstrating its therapeutic role also as a bona fide c-Met inhibitor (44). Previous studies in gastrointestinal tumors, including PDAC, did not find cancers positive for expression of EML4–ALK (45), but Sennino and colleagues showed that simultaneous inhibition of c-Met and VEGF signaling by crizotinib reduced tumor growth, invasion, and metastasis in pancreatic neuroendocrine tumors (46). In our recent in vitro study, we showed the therapeutic potential of crizotinib in several PDAC cells, unraveling its ability to specifically target cancer stem cell–like subpopulations, interfere with cell proliferation, induce apoptosis, reduce migration, and synergistically interact with gemcitabine (23). Considering all these pieces of previous evidence, in the present study, the tumor volumes of PDAC-3-FM-GC mice that received crizotinib, or crizotinib and gemcitabine combination were significantly reduced compared with the untreated mice. Importantly, survival analysis showed that mice treated with crizotinib alone survived longer than nontreated mice, and mice treated with the combination of crizotinib and gemcitabine survived significantly longer than all the other groups.

Finally, to verify whether crizotinib and gemcitabine do reach the tumor and have positive pharmacokinetic and pharmacodynamic interactions, we evaluated their concentrations using a LC/MS-MS method that allows measuring both drugs in as little as 5 μL of blood and 5 μg of tissue samples. These analyses showed significantly higher concentrations of gemcitabine in mice treated with the gemcitabine–crizotinib combination than those treated with gemcitabine alone. Correspondingly, crizotinib reduced CDA activity both in tissue specimens and blood samples. A recent study showed increased intratumoral gemcitabine levels attributable to a marked decrease in CDA protein expression caused by its degradation through ROS induced by nab-paclitaxel (4). Because intact c-Met signaling is a critical factor in the protection against excessive generation of endogenous ROS (47) and c-Met inhibitors, such as PHA665752, increased ROS (48), we demonstrated that a similar mechanism of posttranscriptional degradation induced by ROS was responsible for CDA reduction after exposure to crizotinib, resulting in the increased stabilization of gemcitabine. These findings may have critical implications for the rational development of innovative regimens including the administration of crizotinib and gemcitabine in selected subgroup of patients.

With the establishment and extensive characterization of our double bioluminescent patient-derived orthotopic mouse PDAC models, we aimed at improving the preclinical PDAC model systems. An important limit is that a certain number of cell isolations failed. In addition, it is currently unclear the degree to which the absence of a functional immune system in our immunocompromised mice will influence treatment response of these models (11). Future parallel investigations in mice with reconstituted human immune system may answer such questions. However, we envision our models as an important research platform, which can be used to test selected drugs whose efficacy might then be coupled to specific genetic profiles. Availability of similar models should empower researchers to conduct “preclinical-clinical trials,” testing candidate therapeutics more efficiently and less expensively than traditional trials in patients. Moreover, with the decreasing costs of whole genome sequencing and the subsequent identification of novel “driver mutations,” it should be possible to use representative models to realize the ultimate goal of personalized medicine for each patient. This strategy seems particularly suitable for heterogeneous cancers such as PDAC, for which a combination of drugs that hit a variety of key targets is more likely to be effective (5).

In conclusion, extensively genetically and histopathologically characterized models provide a robust platform to explore novel therapeutic strategies and for further exploration of the biologic and mechanistic insights that can ultimately be applied to the future clinical practice. Here, we used our models to identify crizotinib and gemcitabine as a promising drug combination, acting synergistically via simultaneous targeting of key intratumoral genetic features and increase of drug delivery by CDA modulation, and warranting further investigation for the treatment of PDAC.

No potential conflicts of interest were disclosed.

Conception and design: A. Avan, V. Caretti, N. Funel, G.J. Peters, T. Würdinger, E. Giovannetti

Development of methodology: A. Avan, V. Caretti, N. Funel, E. Galvani, M. Maftouh, R.J. Honeywell, O. Van Tellingen, D. Campani, T. Würdinger, E. Giovannetti

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): A. Avan, V. Caretti, V. Caretti, N. Funel, E. Galvani, M. Maftouh, R.J. Honeywell, T. Lagerweij, O. Van Tellingen, D. Fuchs, H.M. Verheul, U. Boggi, T. Würdinger, E. Giovannetti

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): A. Avan, V. Caretti, N. Funel, M. Maftouh, R.J. Honeywell, D. Fuchs, T. Würdinger, E. Giovannetti

Writing, review, and/or revision of the manuscript: A. Avan, V. Caretti, N. Funel, E. Galvani, M. Maftouh, G.-J. Schuurhuis, G.J. Peters, T. Würdinger, E. Giovannetti

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): N. Funel, T. Lagerweij, D. Fuchs, G.J. Peters, E. Giovannetti

Study supervision: N. Funel, D. Campani, H.M. Verheul, G.-J. Schuurhuis, U. Boggi, G.J. Peters, T. Würdinger, E. Giovannetti

The authors thank Arjan van der Velde (Integrative Bioinformatics Centre, VU University, Amsterdam), Dr. Carla Molthoff (Nuclear Medicine and PET Research VU University Medical Center, Amsterdam) and Dr. Jithin Jose (VisualSonics, Amsterdam, the Netherlands) for the technical support with the array-CGH analysis, animal studies, and high-frequency ultrasound system, respectively.

This work was partially supported by grants from Netherlands Organization for Scientific Research, Veni grant (E. Giovannetti), CCA Foundation 2012 (E. Giovannetti, A. Avan, and G.J. Peters), AIRC/Marie Curie International Fellowship (E. Giovannetti), FIRC International Fellowship (E. Galvani), Italian Minister of Research - PRIN-2009 (U. Boggi, N. Funel, and E. Giovannetti), and Istituto Toscano Tumori (U. Boggi, N. Funel, and E. Giovannetti).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.