Screening of Conditionally Reprogrammed Patient-Derived Carcinoma Cells Identifies ERCC3–MYC Interactions as a Target in Pancreatic Cancer

Pancreatic ductal adenocarcinoma (PDAC) affects 44,000 individuals yearly in the United States (1). This cancer is almost universally lethal, with chemotherapy having very limited efficacy and the vast majority of all diagnosed patients not surviving past the 5-year mark (1). The typical absence of drug-amenable oncogenic drivers in PDAC (2, 3) and the shortage of models representing clinical PDAC both contribute to the challenge of improving the status quo. Careful microdissection of PDAC tumors coupled with genomic analysis has shown that in addition to near universal mutations in KRAS and TP53, and common mutations affecting other regulators of cell cycle, such as CDKN2A (2, 3), deletions, amplifications, and mutations affecting transcriptional and chromatin remodeling, genes are important contributors to the mechanism of PDAC carcinogenesis (4). Among PDACs, the worst outcome is associated with MYC amplification, which has been detected in as many as 14% [provisional The Cancer Genome Atlas (TCGA), cbioportal.org (5)] to 30% (6) of these cancers.

Although targeted therapies have greatly improved response in some tumor types (7), evaluation of novel targeted agents in clinical trials for PDAC has typically been discouraging. Treatments such as the antiangiogenic agent bevacizumab (8), the Sonic Hedgehog (Shh) inhibitor vismodegib (9), or the EGFR receptor-targeting agent cetuximab (10) produced partial responses or limited stable disease in some individuals, but no statistically significant improvements in the general population of PDAC patients. With 30 to 41 new drugs granted FDA approval each year for cancer therapy (7), there has been a notable lack of progress in pancreatic cancer, with only four new agents becoming available to the patients in the past 20 years. These include gemcitabine (11); erlotinib (12) and nab-paclitaxel (ref. 13; both used in combination with gemcitabine); and the topoisomerase 1 inhibitor irinotecan, used in combination with fluorouracil and oxaliplatin (FOLFIRINOX; ref. 14), or with fluorouracil and nanoliposomal irinotecan (15). These last approaches have increased median survival to almost 1 year (14), in contrast to the prior 6.8 months with gemcitabine, but this outcome is still dismal.

A particular limitation of identifying new effective options for pancreatic cancer is the poor correlation of drug activity between long-term cultured cell line or xenograft models and that seen in clinical trials (16). Recent efforts to establish patient-derived xenograft (PDX) models, coupled with avoidance of in vitro passaging, have improved the correspondence observed between the antitumor activity of drugs in mouse models and in patients (17). Particularly, as the cost of PDX models limits their availability for high-throughput screening, the matching of PDX models with minimally cultured cells derived from genotypically matched primary tumors increases flexibility in drug screening, allowing better prioritization of drug potency before subsequent in vivo testing. Here, we used rapidly cultured, early passage PDAC cells from 6 patients to establish patterns of chemosensitivity to 866 clinical drugs. We subsequently validated the most promising drugs in vivo using PDAC PDX models, many derived from the same genotyped patients as used for the screening steps. As described below, this screening identified triptolide, an irreversible inhibitor of transcription that acts through binding to the ERCC3 helicase (18), as one of the most potent PDAC growth-inhibitory compounds in vivo in multiple PDX models, particularly in those with MYC amplification. Exploration of the mechanism of interactions between triptolide, MYC, and ERCC3 suggests a model in which some PDACs depend on integrated MYC-ERCC3–dependent transcription, providing a therapeutic vulnerability.

To identify potential new targets and mechanisms that could be effective for the treatment of pancreatic cancer, we conducted a systematic analysis of drug sensitivity. For this purpose, we developed a representative panel of PDAC cells rapidly expanded from primary tissue in vitro on a layer of irradiated feeder fibroblasts in the presence of Rho kinase inhibitor Y-27632 (19, 20). This methodology provided unselected cultures of PDAC cells within 36 to 61 days after initial tissue dispersal (Fig. 1A; Supplementary Fig. S1A), thus avoiding cell identity changes imposed by in vitro culture (21) while producing numbers of primary cells sufficient for drug sensitivity testing. We derived cultures of PDAC cells from primary surgical tumor samples taken directly from patients (3 cases), or from first passage PDAC xenografts implanted to the flanks of immune-deficient C.B17-scid mice (3 cases). These approaches yielded predominantly epithelial carcinoma cell populations, as judged by cellular morphology (Supplementary Fig. S1B), keratin expression, and the absence of reactivity with mouse-specific antibody or low to absent amplification of the mouse Alf gene by genomic qPCR (Supplementary Fig. S1C). The PDX models and the derived cellular cultures carried mutations identical to the patient's tumors (Supplementary Table S1).

Our goal was to generate results we could rapidly translate to the clinic for the treatment of PDAC. Therefore, we systematically probed chemosensitivity of the in vitro PDAC models using both a focused library of 866 drugs from the NCI Clinical Collection (NCC; refs. 22, 23) and the FDA-approved drug collection (Supplementary Table S2). These drugs have been already tested in clinical trials for numerous oncologic and nononcologic indications, and have desirable drug-like properties including bioavailability, stability, and a well-characterized safety profile. We used the AUC as the most objective parameter of drug activity (24) which could be enumerated in all tested compounds. Drug cytotoxicity against PDAC cells was assessed at 6 therapeutically achievable concentrations ranging from 16 nmol/L to 10 μmol/L, using a CellTiterGlo viability assay measuring cellular ATP in rapidly expanded PDAC cell cultures from 6 patients (in 8 screens in total; Supplementary Table S2). To ascertain the reproducibility of the screen data, we independently derived parallel cell cultures from two separate sources: for patient 1, we cultured two first passage xenografts; for patient 15, we derived independent cultures from the primary tumor and the passage 1 xenograft. In both cases, we found highly correlated patterns of chemosensitivity between the biological replicates (Supplementary Fig. S1D).

Of the chemotherapy drugs that registered as the most potent agents in this assay (Fig. 1B), there was little variability in rank for growth inhibition between the 6 independent patient-derived cell cultures. Notably, the most potent drugs (Fig. 1B) included 5 anthracycline antibiotics (idarubicin, doxorubicin, epirubicine, mitoxantrone, and actinomycin D) which are known as transcriptional repressors based on their primary activity in binding the DNA minor groove (25). These also included a histone deacetylase (HDAC) inhibitor, romidepsin, which is involved in chromatin remodeling and transcription (26), and triptolide, a selective irreversible inhibitor of the transcriptional regulator ERCC3 [also known as xeroderma pigmentosum group B (XPB), a helicase in the TFIIH protein complex (18); Fig. 1B]. Together, DNA-targeting drugs comprised 8 of the 10 most potent hits.

Among the top 15% of most active compounds in PDAC screening (using an AUC threshold of 3), we subclassified drugs based on their described principal mechanisms of activity. Among those, drugs interfering with protein synthesis (homoharringtonine, HHT), folding (HSP90 inhibitor ganetespib), or degradation (proteasomal inhibitors) showed the greatest cytotoxicity in vitro (Fig. 1C). Among transcriptional repressors, triptolide and epirubicin had the greatest activity against PDAC cells in vitro. Microtubule poisons (MT) and drugs directly damaging DNA (DNA) were of substantially lower potency, potentially reflecting a limited single-agent activity of these drugs against pancreatic cancer.

To determine if the hits we observed were uniquely active in PDAC, we benchmarked the cytotoxicity profile we generated in the PDAC cell models to public data reporting systematic analyses of drug chemosensitivity for NCI-60 panel of cell lines (27) which are available in CellMiner (28, 29). The NCI-60 panel does not include PDAC cell lines. Of 189 drugs in common between the 866 compound library we screened directly, and those tested against the NCI-60 panel, 156 had cytotoxicity of greater than 30% (IC30; Supplementary Fig. S2A) relative to vehicle. Using viability ranking to normalize between the data sets, unsupervised clustering analysis determined that rapidly cultured PDAC cells differed distinctly in chemosensitivity pattern from the NCI-60 panel (Supplementary Fig. S2B). Compounds most active in both the NCI-60 analysis and PDAC screen included two proteasome inhibitors (bortezomib and carfilzomib), an anthracycline (actinomycin D), and romidepsin (Fig. 1B). Contrastingly, the ERCC3 inhibitor triptolide and other transcriptional inhibitors were among the top most potent cytotoxins in the PDAC screen but not in the NCI-60 panel.

To nominate candidates for further exploration in vivo, we further validated the IC₅₀ values for selected drugs representing these mechanistic classes in six patient-derived cell lines (Fig. 1D), selecting five candidates (romidepsin, triptolide, carfilzomib, bortezomib, and homoharringtonine) for further validation in PDX models.

We benchmarked these selected agents against a set of chemotherapy drugs currently approved by the FDA (www.NCCN.org) for the treatment of pancreatic cancer, albeit with relatively limited clinical activity. At 21 days, single-agent treatments with gemcitabine, 5-fluorouracil, or platinum agents caused approximately 50% growth delay of multiple pancreatic PDX models in vivo; nab-paclitaxel, mitomycin C, and irinotecan showed stronger (∼75%) tumor growth suppression (Fig. 2A and Supplementary Fig. S3A). Among the new candidate compounds, romidepsin (targeting HDAC-1 and -2) also caused approximately 75% growth delay of PDX tumors (Fig. 2A and Supplementary Fig. S3B) at MTDs (26). Homoharringtonine, a plant alkaloid and a ribosomal inhibitor used for the treatment of chronic myeloid leukemia (30), was inactive in vivo (Fig. 2A), as was carfilzomib (Supplementary Fig. S3C). The proteasome inhibitor bortezomib produced only a transient growth delay in vivo (Supplementary Fig. S3C), in accord with the negative outcome of a randomized clinical trial of bortezomib in combination with gemcitabine in metastatic pancreatic cancer (31).

In contrast, triptolide given orally at daily doses of 0.2 mg/kg for 21 days (32) showed the highest activity against 5 PDX pancreatic cancer models tested (∼90%; Fig. 2A) and in 2 cases, caused durable complete responses after cessation of dosing (Fig. 2B). This caused us to focus further analysis on this compound.

Testing of triptolide in conditionally reprogrammed PDAC cells revealed variable sensitivity between lines (Fig. 3A). More sensitive PDAC cells exhibit rapid induction of apoptosis after 24 hours in the presence of low nanomolar concentrations of the drug (Fig. 3B). Triptolide inhibits transcription by binding to ERCC3, a component of the TFIIH transcription complex (18, 33). We assessed global mRNA expression in 001 PDAC cells 6 hours after treatment with triptolide versus vehicle (Supplementary Fig. S4). We ranked the gene transcripts based on the fold change in their abundance in triptolide-treated PDAC 001 cells (Supplementary Fig. S4A and S4B). We further benchmarked the triptolide-dependent mRNA change in 001 cells against mRNA expression data for 9 cancer cell lines treated with triptolide for 6 hours, available through the NIH LINCS resource (34). Analyses of significantly downregulated gene transcripts in both the PDAC 001 and LINCS datasets showed MYC oncogene as a triptolide target among several Cancer Census genes affected (Supplementary Fig. S4C; Supplementary Table S3). In keeping with previously reported depletion of short-lived mRNA including MYC in A549 lung cancer cells treated with triptolide (35), we determined rapid downregulation of MYC mRNA (Fig. 3D) and protein (Fig. 3E) following triptolide treatment. MYC levels were undetectable after 1 hour of incubation with 50 nmol/L triptolide in the highly sensitive PDAC 001 and 017 cells. Triptolide dose titration experiments showed marked loss of MYC expression at concentrations as low as 6 nmol/L in 001 and 017 cells, but not in relatively refractory 015 and 007 cells (Fig. 3F; Supplementary Fig. S5A), at 24 hours after dosing. Triptolide effects on MYC depletion markedly greater than that of other established transcriptional repressors such as actinomycin D, romidepsin, and idarubicin (Supplementary Fig. S5B). MYC phosphorylation on T58 and S62 has been previously shown to increase MYC protein stability and contribute to KRAS-driven carcinogenesis (36). Emphasizing the relationship between triptolide response and short mRNA and protein turnover of the majority of cellular MYC, triptolide caused little change in the levels of T58- and S62-phosphorylated MYC protein. As expected for a transcriptional repressor (37, 38), we also observed that activity of triptolide induced the rapid decay of the RNA polymerase II subunit RPB1, and the dose- and time-dependent loss of phosphorylation of serines S2 and S5 in the RPB1 C-terminal heptapeptide repeat (ref. 39; Fig. 3F; Supplementary Fig. S5A). The levels of structural proteins such as alpha-tubulin and beta-actin were unaffected by triptolide treatment.

We next determined whether the endogenous expression of MYC or ERCC in PDAC models predicted their susceptibility to triptolide. Direct comparison of MYC protein levels in early passage PDAC cultures showed significant variability between lines, with the highest levels observed in two in vitro sensitive 001 and 017 cells, whereas the levels of ERCC3 were nearly similar across all models tested (Fig. 4A). Further, overexpression of exogenous MYC using rapid lentiviral transduction and puromycin selection conferred 2- to 5-fold increase in sensitivity to triptolide. The drug-sensitizing effect of MYC overexpression was especially prominent in PDAC cells 050 which are triptolide refractory suggesting a direct relationship of MYC status with triptolide response (Fig. 4B). Contrastingly, overexpression of the triptolide target ERCC3 conferred resistance to 2 most sensitive (001 and 017) out of 3 PDAC cells, supporting a direct on-target effect of triptolide (Fig. 4B and Supplementary Fig. S6A). In contrast, expression of neither MYC nor ERCC3 affected PDAC cells sensitivity to EZH2 histone methyltransferase inhibitor UNC1999 (Supplementary Fig. S6B), indicating specific interaction with triptolide.

The effect of triptolide in vivo on the growth of pancreatic xenografts was highly variable (Fig. 4C) from rapid complete responses to transient growth delay. We further probed the relationship of MYC genomic copy number to the PDAC sensitivity to triptolide based on the established MYC addiction in MYC-amplified cancers (40). Of all PDAC models, only tumor tissue of patient 001 was MYC-amplified and carried 12 copies of MYC gene by comparative genomic hybridization and genomic qPCR (Fig. 4D). Thus, we additionally tested two chemotherapy-refractory PDX models, gastric (CTG0353) and ovarian (CTG0258), with high levels of MYC amplification (with 42 and 70 copies of MYC gene, respectively; Fig. 4D). Oral administration of triptolide for 21 days in all three MYC-amplified PDX models demonstrated complete regression of all tumors (Fig. 4C). Contrastingly, pancreatic or other PDX models with a normal copy number of MYC (015, 017, and CTG0302) showed regression but no complete responses, and tumors resumed growth after the triptolide treatment stopped. In contrast to the high sensitivity to triptolide in the short-term culture (Fig. 3A), PDAC 017 xenografts carrying diploid MYC gene showed only transient growth delay (Fig. 4D).

To further investigate the dependence of MYC on ERCC3 function and to rule out a potential off-target activity of triptolide, we silenced ERCC3 with siRNA and measured MYC expression. Interestingly, depletion of ERCC3 significantly reduced MYC protein expression (by 30%; Fig. 4E), and MYC-silenced 001 cells markedly downregulated ERCC3 protein (at 50%; Fig. 4E). Extending this latter result, we used a tetracycline repressor-regulated model of MYC expression (41) to confirm that MYC suppression led to loss of ERCC3 protein expression (Supplementary Fig. S6C), suggesting a reciprocal dependency of MYC and ERCC3 to regulate oncogenic transcription.

Despite the dramatic initial meltdown of MYC-amplified tumors following a 21-day course of triptolide treatment, in most animals, the tumors recurred after a prolonged remission (Fig. 4C and F). Re-treatment of the recurrent tumors with the same dose of triptolide caused regression of all but a few tumors, which grew in the presence of the drug. Remarkably, the resistant tumor showed markedly upregulated expression of ERCC3 and MYC (Fig. 4G), which is consistent with our proposed model of a critical dependency between ERCC3 and MYC. Triptolide-refractory 001 cells obtained from the refractory tumors grew in the presence 50 nmol/L triptolide, whereas the naïve 001 cells experienced complete growth inhibition at approximately 5 nmol/L (Fig. 1D). This acquired triptolide resistance was selective, as sensitivity to the BRD4 inhibitor JQ1 and UNC1999 was comparable in triptolide-sensitive and -refractory cells (Supplementary Fig. S6D and S6E).

Previous reports had indicated that inactivating mutations in ERCC3 or GTF2H4 confer triptolide resistance (33, 42). We ruled out such acquired mutations by sequencing the RNA from resistant cells, finding no mutations in any components of the TFIIH or RNA polymerase II complex (Supplementary Fig. S7). However, these triptolide-refractory 001 cells had markedly increased levels of ERCC3 and were resistant to triptolide-mediated MYC depletion (Fig. 4C). Furthermore, silencing ERCC3, but not MYC, reversed the triptolide resistance (Fig. 4D), indicating that overexpression of drug target is responsible for the acquired resistance. These findings are also consistent with the results of ERCC3 overexpression (Fig. 4B).

Our data suggested that ERCC3 expression, like MYC expression, might be related to survival in PDAC patients. Using the survival information from the TCGA dataset (cbioportal.org; refs. 2, 43), we determined that PDAC with high ERCC3 expression had significantly worse overall survival (OS; median of 13.8 vs. 16.8 months; log-rank test P = 0.029; Fig. 5A). ERCC3 expression was independent of the MYC copy-number status of the tumors (Pearson 0.067). We then examined whether the predictive value of ERCC3 expression was influenced based on tumor genotype, again using data from cbioportal.org and querying a set of common genetic lesions in PDAC. Genetic loss of CDKN2A, encoding p16, is common in PDAC and associated with deregulation of cell-cycle control (2, 3). Although homozygous deletions of CDKN2A did not independently predict OS (Fig. 5B), this genotype has a striking interaction with ERCC3 expression. Elevation of ERCC3 strongly predicted shorter OS in CDKN2A-null tumors (Fig. 5C, median of 4.8 vs. 19.9 months; log-rank test P = 0.005), but lost predictive power in CDKN2A wild-type tumors (Fig. 5D).

Taken in sum, our data demonstrate in multiple PDX and cell culture models for PDAC that inhibition of ERCC3 with triptolide is a useful therapy, particularly in the context of amplified or overexpressed MYC. They also demonstrate specific interaction between ERCC3 and MYC, reflecting interdependent protein expression required for cell viability and resistance to triptolide. Together, these results suggest the idea that inhibition of the transcription factor ERCC3 is particularly important in fast-proliferating cells and tumors lacking growth restriction checkpoints: an idea further supported by our identification of a relationship between high ERCC3 expression and poor overall survival for PDAC, particularly in the context of a common CDKN2A gene disruption. Our data thus imply triptolide, in addition to MYC-amplified carcinomas, may also be particularly effective in the subset of PDAC with CDKN2A deletions.

This study, for the first time, used a recently described rapid culture method with limited passaging (19, 20) to develop tools for the evaluation of PDAC chemosensitivity in physiologic models. Subsequent screening with these tools evaluated many distinct pharmacologic drug classes in the currently available pharmacopeia. This emphasized a particularly high sensitivity to agents targeting the transcriptional apparatus and protein turnover, some of which were comparable to or outperformed conventional chemotherapy drugs currently in use for PDAC. Not all of these agents were effective in subsequent in vivo testing: For instance, homoharringtonine (also known as omacetaxine mepesuccinate) showed lower activity in vivo than in vitro. This may reflect the lower proliferation index of tumors growing in desmoplastic microenvironments, relative to the more exponential growth conditions pertaining in cultured cells, which reduces absolute dependence on maximally efficient mRNA and protein expression. Alternatively, it may reflect poor penetration of the drug in tumor tissue, a limiting factor for some otherwise promising compounds. However, the ERCC3-targeting drug triptolide was highly effective in vivo, with activity strongly linked to disruption of MYC expression. This activity may reflect the irreversible modification of ERCC3 by triptolide, resulting in an inescapable brake on the core transcriptional machinery.

The transcriptional machinery has been assuming greater importance as a target of therapy in PDAC. Genomic analyses identifying common mutations in the SWI/SNF chromatin remodeling complex genes including the DNA binding subunits (ARID1A, ARID1B, PBRM1) and enzymatic subunits (SMARCA2, SMARCA4; ref. 44) emphasized the role of chromatin remodeling in PDAC etiology. The role of epigenetic mechanism was also demonstrated for the histone H3K36 demethylase KDM2B, which is overexpressed in PDAC (45). Preclinical studies targeting chromatin remodeling with a bromodomain 4 inhibitor, JQ1 (46), or with a combination of JQ1 and a HDAC inhibitor (47) have demonstrated efficacy in PDX and in a genetic mouse model of PDAC. Importantly, control of tumor growth with these therapies involved limiting activity of MYC, which is a critical promoter of pancreatic carcinogenesis (48), and for which high expression is known to be a poor prognostic factor for PDAC (4). Our data suggest further evaluation of triptolide in combination with other means to deregulate MYC as a promising approach.

By identifying regulatory cross-signaling between ERCC3 and MYC, our results will help guide the future clinical development of solubility-enhanced triptolide analogs (ClinicalTrials.gov identifier NCT01927965; ref. 32). Of particular interest, this work is the first to suggest that increased ERCC3 expression may be a hallmark of aggressive PDAC in the context of the cell-cycle deregulation associated with loss of CDKN2A. More work is required for better understanding whether ERCC3 overexpression contributes to a poor outcome based on its role in transcription, or through provision of increased resistance to the conventional DNA damaging therapies used to treat PDAC. Whatever the mechanism, this observation will also guide the clinical development of triptolide analogs.

We obtained surgical pancreatic adenocarcinoma specimens from consented patients in accordance with the Fox Chase Cancer Center Institutional Review Board protocol 12-822. Tumor tissue was placed in ice-cold sterile transport medium and immediately implanted in Matrigel (Corning Life Sciences) to both flanks of C-B17.scid mice as described (16) and/or cultured with irradiated J2 feeder fibroblasts as detailed below. In our experience, take rate from metastatic sites was higher at 6/10 (60%), compared with the primary PDAC where implantation success was at 11/56 (19.6%).

The antibodies used in Western blot were pThr58-Myc (Santa Cruz Biotechnology; sc-135647), pS62-Myc (Abcam), ERCC3 (A301-337A; Bethyl Laboratories); antibodies to MYC (#5605), Rpb1-CTD (#2629), pSer2-CTD of Rpb1 (#8798), pSer2-CTD of Rpb1 (#8807), β-actin (#4967), and α-tubulin (#2125) were purchased from Cell Signaling Technology. Triptolide was purchased from Cayman Chemical and dissolved in DMSO as 1 mmol/L stock. Chemotherapy drugs were purchased from the Fox Chase Cancer Center pharmacy.

Fragments of freshly obtained tumor tissues were dissociated using collagenase/hyaluronidase and Dispase (StemCell Technologies) at 37°C for 3 hours with occasional shaking as per the manufacturer's protocol. PDAC cells were then placed in flasks layered with irradiated 10⁶ J2 NIH-3T3 fibroblast feeder cells in full medium. The cell culture medium included complete DMEM with freshly added supplements (GIBCO-Thermo Fisher Scientific): 10% FBS, penicillin, streptomycin and glutamine, F12 nutrient mix (all at 1X), 25 μg/mL hydrocortisone, 125 ng/mL EGF, 5 mg/mL insulin, 250 μg/mL fungizone, 10 mg/mL gentamycin, 10 μg/mL nystatin, 11.7 μmol/L cholera toxin, and 5 μmol/L Rho kinase inhibitor Y27632 (Sigma-Aldrich). The medium was replaced every 2 days. The cells were cultured for a period of 1 to 2 months and split twice a week using standard trypsinization procedure (19). Typically, 10⁶ irradiated feeder J2 cells were plated at each passage with 1 to 2 × 10⁶ PDAC cells in a T75 flask. Prior to compound screening, feeder cells were detached following a brief incubation with trypsin, which caused their rapid detachment while PDAC cells remained adherent to the flask.

The library included a panel of 730 compounds of the NCC (22, 23) and 101 FDA-Approved Oncology drug Set IV from NCI/DTP Open Chemical Repository (http://dtp.cancer.gov) encompassing the agents that have been in phase I–III clinical trials and are part of the NIH Roadmap Molecular Libraries Screening Centers Network (49). Controls and several screening compounds were added to the plates manually (full list in Supplementary Table S1). PDAC cells were plated in 96-well flat-bottom plates (Sarstedt) at 2,500 cells per well 48 hours prior to adding the drugs. The compound library was maintained as 10 mmol/L stock in DMSO which has been used for further serial dilution using robotic high-throughput plate handling automation (CyBio) to create 2 replicates of each library plate at 10 μmol/L, 2 μmol/L, 0.4 μmol/L, 0.08 μmol/L, and 0.016 μmol/L final drug concentrations. Sensitivity was assessed on day 6 using standard viability assay (CellTiter-Glo, CellTiter-Blue, Promega; ref. 50).

Cryopreserved in DMSO PDX tumor fragments were quickly thawed, washed in RPMI, and resuspended in 1:1 mix of RPMI and Matrigel (Corning Life Sciences) on ice until the moment of implantation. Tumor fragments were subcutaneously injected using 1 mL syringe and 18G 1½ needle in both flanks of 5- to 8-week-old C-B17.scid mouse. Animals with established tumors (around 150 mm³) were randomly divided to receive indicated drugs or vehicle with indicated dose schedules and routes of administration. Animals were weighed twice a week. Tumor size was measured with a digital caliper twice a week, and volumes were calculated using the modified ellipsoid formula as described (50) until the maximum size of 1,500 mm³ or if animals exhibited distress, >20% weight loss or if tumors ulcerate, in which case mice were humanely euthanized.

The DX tumor tissues were lysed with T-PER reagent (Thermo Fisher Scientific) containing double concentration of Halt protease and phosphatase inhibitors (Thermo Fisher Scientific). For Western blot, cells were seeded on 6-well plates at density of 30,000 cells per well in RPMI-1640 supplemented with 1% FBS, 2 mmol/L l-glutamine, 25 mg/L Insulin, and 100 mg/L penicillin/streptomycin. Next day, the cells were treated with indicated drugs for 24 hours and lysed with RIPA lysis buffer (Santa Cruz Biotechnology) containing Halt protease and phosphatase inhibitors (Thermo Fisher Scientific). Protein concentrations were measured using BSA assay. Western blot membranes were scanned using Odyssey infrared imaging reagents including blocking buffer and secondary antibodies (LI-COR).

We used the Human Multicopy Reference assay (QIAGEN) to determine MYC gene copy number by qPCR in PDAC cells using an ABI Prism 7700 Detection System (Applied Biosystems) with the following primers: MYC, forward-TTCTAACAGAAATGTCCTGAGCAATC, reverse- TCAAGACTCAGCCAAGGTTGTG; ALB (albumin), forward-CATTTATTGGTGTGTCCCCTTG, reverse-ACACCAGTGAAAACAATTTAAGCC. The results were analyzed by the comparative Ct method to establish relative expression curves. MYC copy number was quantified relative to ALB based on the relative stability of ALB chromosomal locus 4q11-q13 in PDAC (3). PDAC cells were transfected with validated siRNA at 50 nmol/L concentrations targeting human MYC (CTCGGTGCAGCCGTATTTCTA) and ERCC3 (CCGGTTCACCTCCGATGCCAA) genes and the firefly luciferase GL2 (AACGTACGCGGAATACTTCGA) as a non–targeting-negative control in triplicates with siRNA mixed with HiPerFect Transfection Reagent (QIAGEN) according to the manufacturer's reverse-transfection protocol. For ERCC3 overexpression in HEK-293 cells, the full-length ERCC3 cDNA was amplified using forward-ATGGGCAAAAGAGACCGAGCG and reverse-CTCTTCAAGCGCTTTAGGAAATGA primers and Gateway-cloned to pLEX mammalian expression vector (OHS4735, Thermo Scientific). Cells were transfected with plasmid DNA using Lipofectamine 2000 as per the manufacturer's (Thermo Scientific) protocol. For lentiviral expression, full-length cDNA of ERCC3 and T58A-MYC was PCR amplified and cloned to pLEX vector using Gateway cloning as per the manufacturer's protocol (Thermo Scientific). Infected patient-derived PDAC cells (at multiplicity of infection of 0.5) were briefly selected with puromycin and used at early passages 5–8.

We used AUC to assess the drug's cytotoxicity as described (24). The CellTiter-Glo viability data were normalized using the mean of DMSO control wells on each plate. An open source R package (www.r-project.org) was used to model fitting of dose response curves for each compound using a four-parameter log-logistic formula. The extracted fitted values were used to calculate AUC and to estimate IC₅₀ for each compound.

Comparisons between the chemosensitivity of rapid cultures of pancreatic carcinoma cells and the NCI-60 cell lines panel were made using the publicly available data from the CellMiner database (http://discover.nci.nih.gov/cellminer). Using the unique NSC identifying numbers for each compound, 189 CellMiner drugs overlapped with our screening library. Of those, we included in the analysis 156 drugs showing at least 50% cytotoxicity, and for which the data were available in at least 70% of cell lines of the NCI-60 panel. MD-MBA-453 cell line on the NCI-60 panel was excluded from the analysis because it had drug screening data on less than 50% of drugs. Compound activity values were expressed as −log₁₀ and ranked. For data visualization, the hierarchical clustering we used average linkage and distance methods to build heatmaps in R software. We used Student t test to analyze the differences between the biological repeats in experiments for statistical significance.

No potential conflicts of interest were disclosed.

Conception and design: K. Paz, E.A. Golemis, I. Astsaturov

Development of methodology: E. Banina, I.G. Serebriiskii, K. Paz, V. Khazak, I. Astsaturov

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): N. Beglyarova, E. Bobrov, E. Lysenko, N. Skobeleva, L. Gabitova, M. Pressman, J.P. Hoffman, K. Paz, D. Behrens, V. Khazak, S.A. Jablonski, L.M. Weiner, I. Astsaturov

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): N. Beglyarova, E. Banina, Y. Zhou, R. Mukhamadeeva, G. Andrianov, E. Bobrov, K. Paz, V. Khazak, S.A. Jablonski, E.A. Golemis, L.M. Weiner, I. Astsaturov

Writing, review, and/or revision of the manuscript: I.G. Serebriiskii, K. Paz, E.A. Golemis, L.M. Weiner, I. Astsaturov

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): D. Restifo, V. Khazak, I. Astsaturov

Study supervision: I. Astsaturov

Other (research conducting): N. Beglyarova

We are grateful to the FCCC Cell Culture Facility and Jin Fang for technical assistance with derivation of pancreatic carcinoma cell culture. The authors' work has been supported by the Fox Chase Cancer Center Genomics Facility (Drs. Yuesheng Li and Emmanuelle Nicolas). Champions Oncology Inc. contributed to the data of chemotherapy sensitivity of pancreatic cancer patient-derived xenografts presented in Fig. 2A and Supplementary Fig. S2.

In Memoriam of Dr. Elena Gitelson, a physician and pancreatic cancer patient.

This work was supported by NIH core grant CA-06927, by the Pew Charitable Fund, and by generous gifts from Concetta Greenberg, the Kalargheros family, and from Donald Geary to Fox Chase Cancer Center. Some of the authors were supported by NIH R01 CA188430, K22 CA160725, R21 CA164205, a career development award from Genentech, by Tobacco Settlement funding from the State of Pennsylvania (I. Astsaturov), by a grant from the Bucks County Board of Associates (D. Restifo, L. Gabitova, and I. Astsaturov), R01 CA63366 (E.A. Golemis), by the Program of Competitive Growth of Kazan Federal University (L. Gabitova and I.G. Serebriiskii), and by the grant 15-15-20032 of the Russian Science Foundation (I.G. Serebriiskii, R. Mukhamadeeva, G. Andrianov, and I. Astsaturov) for the bioinformatics analyses of drug sensitivity profiles and RNA expression.

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.