Integrated Patient-Derived Models Delineate Individualized Therapeutic Vulnerabilities of Pancreatic Cancer

Resource

Graphical Abstract Authors Agnieszka K. Witkiewicz, Uthra Balaji, Cody Eslinger, ..., Gordon B. Mills, Eileen M. O’Reilly, Erik S. Knudsen

Correspondence [email protected] (A.K.W.), [email protected] (E.S.K.)

In Brief Pancreatic cancer is therapy recalcitrant, and new approaches to therapeutic intervention are needed. Witkiewicz et al. used a large panel of patient-derived models to address the ability to use genetics or empirically determined sensitivities to guide treatment. These ﬁndings demonstrate a weakness of the current reliance on genetic analysis and suggest that using functional approaches with patient avatars could be particularly important for navigating the diverse therapeutic sensitivity of pancreatic Highlights cancer. d Only a few genetic events in pancreatic cancer are currently sensitive to therapeutic targeting Accession Numbers GSE84023 d Patient-derived models can serve as the basis for empirically deﬁned drug sensitivities d There are few strong responses to single agents in patient- derived models d Drug combination screens reveal diverse therapeutic sensitivities that are patient selective

Witkiewicz et al., 2016, Cell Reports 16, 2017–2031 August 16, 2016 ª 2016 The Author(s). http://dx.doi.org/10.1016/j.celrep.2016.07.023 Cell Reports Resource

Integrated Patient-Derived Models Delineate Individualized Therapeutic Vulnerabilities of Pancreatic Cancer

Agnieszka K. Witkiewicz,1,2,3,4,* Uthra Balaji,1 Cody Eslinger,1 Elizabeth McMillan,5 William Conway,6 Bruce Posner,7 Gordon B. Mills,8 Eileen M. O’Reilly,9 and Erik S. Knudsen1,2,4,10,* 1McDermott Center for Human Growth and Development, University of Texas Southwestern Medical Center, 6000 Harry Hines Boulevard, Dallas, TX 75390, USA 2Simmons Cancer Center, University of Texas Southwestern Medical Center, 6000 Harry Hines Boulevard, Dallas, TX 75390, USA 3Department of Pathology, University of Arizona, 1501 N. Campbell Street, Tucson, AZ 85724, USA 4University of Arizona Cancer Center, University of Arizona, 1515 N. Campbell Street, Tucson, AZ 85724, USA 5Department of Cell Biology, University of Texas Southwestern Medical Center, 6000 Harry Hines Boulevard, Dallas, TX 75390, USA 6Department of Surgery, Ochsner Medical Center, 1514 Jefferson Highway, Jefferson, LA 70121, USA 7Department of Biochemistry, University of Texas Southwestern Medical Center, 6000 Harry Hines Boulevard, Dallas, TX 75390, USA 8Department of Systems Biology, University of Texas MD Anderson Cancer Center, 1515 Holcombe Boulevard, Houston, TX 77030, USA 9David M. Rubenstein Center for Pancreatic Cancer Research, Memorial Sloan Kettering Cancer Center, 1275 York Street, New York, NY 10065, USA 10Department of Medicine, University of Arizona, 1515 N. Campbell Street, Tucson, AZ 85724, USA *Correspondence: [email protected] (A.K.W.), [email protected] (E.S.K.) http://dx.doi.org/10.1016/j.celrep.2016.07.023

SUMMARY are multiple successes attributed to precision medicine, from the current paradigm for breast cancer treatment stratification Pancreatic ductal adenocarcinoma (PDAC) harbors based on immunohistochemical markers (e.g., estrogen recep- the worst prognosis of any common solid tumor, tor [ER], progesterone receptor [PR], or human epidermal growth and multiple failed clinical trials indicate therapeutic factor receptor 2 [HER2]) to the integrated genetic analysis of recalcitrance. Here, we use exome sequencing lung cancer that reveals multiple targets for therapeutic interven- of patient tumors and find multiple conserved ge- tion (e.g., ALK [anaplastic lymphoma kinase] rearrangements or netic alterations. However, the majority of tumors EGFR [epidermal growth factor receptor] mutations) (Deluche et al., 2015; Lindeman et al., 2013). Based on these and other exhibit no clearly defined therapeutic target. High- successes, multiple clinical trials utilizing genetic information to throughput drug screens using patient-derived cell guide patient treatment are open. lines found rare examples of sensitivity to monother- Pancreatic ductal adenocarcinoma (PDAC) harbors a particu- apy, with most models requiring combination ther- larly poor prognosis, and even after resection, long-term survival apy. Using PDX models, we confirmed the effective- remains poor due to the frequent recurrence as metastatic dis- ness and selectivity of the identified treatment ease (Almhanna and Philip, 2011; Kleger et al., 2014; Paulson responses. Out of more than 500 single and combi- et al., 2013; Yeo et al., 1995). Current systemic treatment of nation drug regimens tested, no single treatment PDAC is dependent on chemotherapy, with minimal success of was effective for the majority of PDAC tumors, and targeted approaches in the clinic. This therapeutic recalcitrance each case had unique sensitivity profiles that could of PDAC is surprising, given substantial pre-clinical investigation not be predicted using genetic analyses. These and multiple provocative findings that would be expected to yield clinical benefit. This disconnect between preclinical testing data indicate a shortcoming of reliance on genetic and clinical outcomes suggests that more relevant models will analysis to predict efficacy of currently available be important for making significant inroads into the treatment agents against PDAC and suggest that sensitivity of PDAC and that some form of patient stratification will be profiling of patient-derived models could inform required to yield improved outcome. Notably, exceptional re- personalized therapy design for PDAC. sponses to therapy do occur in patients with PDAC (Garrido-La- guna et al., 2015), but these represent a very small segment of INTRODUCTION the treated population. To date, pancreatic cancer has largely failed to benefit from A precision approach to cancer medicine is transforming the way the promise of precision therapy. In spite of substantial genetic in which cancer is treated (Aronson and Rehm, 2015; Biankin analyses (Bailey et al., 2016; Collisson et al., 2011; Jones et al., et al., 2015). Conventionally, this approach relies on the use of 2008; Waddell et al., 2015; Witkiewicz et al., 2015), the path for markers to define a treatment strategy for a given disease. There the treatment of the majority of PDAC cases remains obscure.

Cell Reports 16, 2017–2031, August 16, 2016 ª 2016 The Author(s). 2017 This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/). AB

F G

(legend on next page) 2018 Cell Reports 16, 2017–2031, August 16, 2016 Pancreatic cancer is driven by KRAS, which is currently consid- derived models provided a unique opportunity to assess the ered a non-actionable target. Additionally, most pancreatic tu- impact of a given allele on therapeutic sensitivities. mors harbor genetic lesions in multiple oncogenic pathways Chromatin remodeling SWI/SNF (switch/sucrose non-ferment- (e.g., MYC amplification) or tumor-suppressive pathways (e.g., able) complex genes (e.g., ARID1A, PBRM1, and SMARCA4) are SMAD4 loss) that make it unclear how targeting a single genetic mutated in up to 20% of PDAC, and recent studies indicated that event would yield therapeutic benefit. Thus, the nature and SWI/SNF deficiency results in sensitivity to EZH2 inhibition (Bitler complexity of tumor genetics in pancreatic cancer represent et al., 2015; Kim et al., 2015). Two cell models exhibited genetic the major impediments to precision treatment. alterations in the ARID1A chromatin remodeling gene, and one Here, we use an integrated collection of patient-derived cell harbored a SMARCA2 deletion (Figure 1D). The models were lines and xenografts that recapitulate the genetics and biology treated with the EZH2 inhibitors GSK126 and GSK343, and sur- of PDAC to identify and validate patient-selective therapeutic vival was determined (Figures 1E and S1). Surprisingly, the sensitivities. This approach bypasses the bottleneck of requiring EMC7310 model that was wild-type for chromatin modifiers genetic targets for therapeutic intervention and revealed multiple was the most sensitive to EZH2 inhibitors, while models harboring unique combinatorial approaches to target individual tumors. mutations in chromatin remodelers had intermediate sensitivities (Figures 1E and S1). This marked sensitivity could be due to the RESULTS presence of high-levels of H3 K27-Me3 in the EMC7310 cell line (Figure 1F) and/or other genetic features of the model (e.g., Pipeline for Genetic and Functional Analysis MYC amplification or RB [retinoblastoma] loss). However, these A pipeline was developed, wherein patients with resectable data underscore the challenge of targeting specific genetic PDAC were consented to the collection of tumor tissue, genetic events occurring in complex cancer genomes and the need to studies, and development of models (Figures 1A and 1B). functionally assess therapeutic sensitivities. Primary tumors, patient-derived cell lines, and PDX models Another case exhibited mutation of the STAG2 gene on the were characterized by exome sequencing and exhibited a X chromosome in the primary tumor and resultant PDX model. high level of genetic conservation with the primary (E.S.K., This event can compromise mitotic fidelity and elicit sensitivity U.B., C.E., C. Moxom, J. Mansour, W.C., E.M.O., and A.K.W., to DNA cross-linking agents (Evers et al., 2014; Solomon unpublished data). Cell models were used to define therapeutic et al., 2011). Consistent with STAG2 deficiency, the PDX ex- sensitivities, and selected treatments were subsequently vali- hibited significant nuclear pleomorphism, mitotic aberrations, dated in the context of the PDX originating from same primary and high baseline DNA damage relative to other PDX models tumor. From the genetic analysis, more than 1,000 tumor-spe- (Figure 1G). Consequently, treatment with mitomycin C signifi- cific genetic events were identified, many of which impacted cantly restricted tumor growth in this model (Figure 1H). This on cancer-relevant pathways that are known to be disrupted in response was marked by extensive DNA damage and evidence PDAC (Figure 1C). of mitotic catastrophe (Figure 1I). Thus, in rare circumstances, the genetics of an individual pancreatic tumor could inform Patient Models Can Inform Genetic-Driven Therapeutic treatment. Sensitivity in PDAC A main precept of precision oncology is that genetic analysis of Patient-Derived Cell Lines Define Selective Therapeutic the tumor will inform therapeutic options. However, from muta- Vulnerabilities tional events identified in the 28 cases, only a handful repre- The relatively low frequency of clear, genetically encoded vulner- sented targets for therapeutic intervention. Additionally, many abilities, combined with the complexity surrounding many of the potentially actionable alleles represented variants of un- genetic variants, makes predicting therapeutic sensitivities in known significance. In this setting, the availability of patient- PDAC difficult. In recognition of this challenge, and to define

Figure 1. Genetic Interrogation of PDAC Models (A) Schematic of the overall pipeline used. Tumor tissue not required for diagnosis was used for characterization of the patient tumor. Parallel tissue was used for development of PDX and cell-line models. Models were utilized to evaluate drug sensitivities that could then inform genetic or empirically defined sensitivities. (B) Summary of the models generated from PDAC cases as used in the study. (C) Overall distribution of genetic events targeting oncogenic pathways in PDAC cases used in this study (green color bar), relative to 109 cases that have previously been published. (D) Summary of models harboring the indicated genetic alterations in chromatin modifiers. (E) The indicated models were treated with increasing doses of GSK126 (0, 10, 20, and 40 mM). Data are from greater than four independent measures, with the mean and SD shown. Data are statistically significant, as determined by unpaired t test. ***p < 0.001. (F) The levels of H3K27-Me3 were determined in the indicated models; total histone levels were utilized for loading control. (G) The presence of aberrant mitosis and basal levels of DNA damage were determined by immunohistochemical analysis with pSer10-Histone H3 (pS10-H3) and gH2AX staining, respectively. Representative images are shown. The scale bar for the hematoxylin and pSer10-Histone H3 staining represents 100 mm, while the scale bar for gH2AX staining represents 50 mm. The EMC93 case has a mutation in STAG2. (H) The EMC93 PDX was randomized for treatment with either vehicle control (n = 4) or mitomycin C (n = 7). Tumor volume was determined by calipers as a function of time. Mice were sacrificed when controls became moribund. The difference in tumor volume was statistically significant as determined by ANOVA. Error bars indicate mean ± SD. (I) Representative images showing confluent necrosis and aberrant nuclei in treated tumors (scale bar, 100 mm). Increased cleaved caspase-3 (CC3; scale bar, 100 mm), gH2AX, and mitotic aberrations were detected by immunohistochemistry (scale bars, 50 mm).

Cell Reports 16, 2017–2031, August 16, 2016 2019 A B

E F

(legend on next page)

2020 Cell Reports 16, 2017–2031, August 16, 2016 therapeutic vulnerabilities in the majority of tumors that do not response analysis, we found that the EMC828 model was sensi- harbor actionable genetic events, a drug sensitivity screen with tive to low-dose MEK inhibition (Figure 3A). In general, resistance a library of 305 agents (Data S1) was used against the patient- to MEK inhibitors is associated with compensatory activation of derived cell lines we developed and the established PDAC cell AKT signaling (Mirzoeva et al., 2009; Pettazzoni et al., 2015). The lines. The drug panel included compounds that are in clinical exceptional sensitivity to MEK inhibitor was associated with a use or advanced development and target multiple aspects of tu- failure to engage the upregulation of AKT, and AKT activity was mor biology (e.g., cell cycle, apoptosis, growth factor signaling actually suppressed upon treatment with MEK inhibition in this pathways). For each cell line, early and late passage cultures model (Figure 3B). RNA sequencing revealed that the excep- were evaluated and showed marked conservation of drug sensi- tional response to MEK inhibition was associated with an tivities (Figure 2A; Figure S2). Cell lines developed directly from enrichment of genes involved in cell adhesion, epithelial versus the primary tumor or from PDX established from the same pri- mesenchymal differentiation, and KRAS dependence signature mary tumor clustered together, indicating preservation of thera- (Singh et al., 2009). This signature was significantly enriched in peutic responsiveness across derivative models (Figure 2B). 59 the EMC828 model versus the less sensitive cell lines (Figure 3C). patient-derived cell-line models were screened in total at a The sensitivity to MEK inhibition was associated with both inhibi- dose range of 100 nM–1 mM, and area under the curve (AUC) tion of cell-cycle progression and the induction of apoptosis that was calculated per drug per cell line (range = 0.08–4.95) (Data translated into overall suppression of viability in clonal assays S1). Among the 305 drugs screened, 76 drugs with an AUC and 3D organoid cultures (Figures 3D–3F). Since MEK inhibition <1.5 in at least one model were identified as hits and were clus- typically exerts a cytostatic effect, these data reinforced the tered based on Euclidian distance (Figure 2C). The data showed concept that this select tumor would be highly sensitive to that similar targeted agents cluster together, indicating that there MEK inhibition. The PDX model developed from the same pri- are intrinsic sensitivities to specific pathways (e.g., MEK [MAPK/ mary tumor was subjected to treatment with the MEK inhibitor ERK kinase] or EGFR) inhibition or chemotherapy. Unsupervised AZD6244 (Figures 3G and 3H). Treatment resulted in marked clustering of models and affinity propagative clustering showed suppression of tumor growth. Importantly, this response was se- that established cancer cell lines had distinct sensitivities lective to the model predicted to be sensitive, as AZD6244 had compared to primary models (Figures 2C and S2). In general, minimal effect on tumor growth in another PDX (i.e., EMCT2). established cell line models were significantly more sensitive The response to AZD6244 in vivo was characterized by suppres- to chemotherapy agents (i.e., gemcitabine, pemetrexed, and sion of ERK1/2 phosphorylation and Ki67 in the sensitive model docetaxel), and one targeted agent (mubritinib) was selectively EMC828, but not the comparator tumor EMCT2 (Figure 3I). effective in these cell lines (Figure 2D). Across the panel, Consistent with these data, MEK treatment of xenografts re- primary patient-derived cell lines exhibited a highly variable res- sulted in the suppression of RB phosphorylation and cell-cycle- ponse to treatment with selected models markedly inhibited by regulated proteins, as determined by reverse-phase protein targeted therapies. For example, one of the models was excep- arrays (RPPAs) (Figure S3). Less dramatic single-agent re- tionally sensitive to multiple MEK inhibitors (Figures 2E and 2F), sponses observed in cell culture with multiple targeted agents while others had selective sensitivity to EGFR or tyrosine kinase (IMD-0354, ABT-737, and CHIR125) were associated with inhibitors (Figure S2). essentially no therapeutic response in corresponding PDX models (Figure S3). These findings suggest that exceptional re- Patient Models Define Exceptional Responses to sponses in cell culture are required in order to have therapeutic Therapeutic Agents impact on PDAC tumors in vivo and recapitulate the known ther- Identifying exceptional responses to therapy has become one of apeutic recalcitrance of the disease. the critical approaches in defining targeted means to treat PDAC and other therapy-recalcitrant diseases. Although MEK inhibi- Uncovering Effective Combination Therapies by tors had limited single-agent antitumor activity in clinical trials High-Throughput Screening conducted in unselected PDAC patients, there has been evi- Combination therapies have the potential to increase the fre- dence for exceptional response in patients with metastatic quency and duration of therapeutic response. However, rational disease (Garrido-Laguna et al., 2015). Through detailed dose- design of combination therapies remains difficult, due to

Figure 2. PDAC Cell Lines Reveal Disparate Sensitivity to Cancer Drugs (A) Early- and late-passage cell lines were subjected to drug screening. Regression correlation analysis revealed that the drug sensitivities across passages remained highly correlated. (B) Spearman correlation analysis of multiple cell lines derived from the same model showed that distinct models harbored highly correlated responses to cancer drugs relative to other established pancreatic cancer models. (C) Heatmap of primary cell lines (gray) and established cell lines (black) clustered based on AUC. The agent color bar is associated with different drug classes as indicated. (D) Drug sensitivities that were statistically different between patient-derived and established lines are noted in box-and-whisker plots, with statistical signiﬁcance determined by t test. Error bars indicate mean ± SD. **p < 0.01; ***p < 0.001. (E) The AUC of drug response to the indicated MEK inhibitors was used to cluster models based on Euclidian distance. The EMC828 cell line exhibited a unique sensitivity to these agents. (F) Dose-response analysis from EMC828 versus EMCT2 cell line models treated at 100 nM, 250 nM, and 1 mM. The mean surviving fraction and SD are shown (n = 3). The difference in response between EMC828 and EMCT2 was statistically signiﬁcant as determined by unpaired t test. **p < 0.01; ***p < 0.001.

Cell Reports 16, 2017–2031, August 16, 2016 2021 AB C

D E

(legend on next page) 2022 Cell Reports 16, 2017–2031, August 16, 2016 inadequate understanding of individual targets, drug interac- tential biomarkers, which underscores the challenge associated tions, and a paucity of biomarkers. Empirical approaches to with developing predictive markers for combination therapies. uncovering combination therapy sensitivities have proved suc- cessful in defining new treatment strategies (Chen et al., 2012; Tumor-Selective Trametinib Combinatorial Sensitivity Crystal et al., 2014; Vora et al., 2014). To test this approach in Translates to Disease Control In Vivo PDAC models, 14 clinically relevant drugs were evaluated in all Among the agents tested, the MEK inhibitor trametinib exhibited pairwise combinations across 11 different cell models at three potent combination activity with several agents (i.e., dasatinib, doses, yielding a total of 5,880 sensitivity measures (Figure 4A; docetaxel, and everolimus) and served as the backbone for a trial Data S2). These analyses defined a unique pattern of sensitivity of cell-line-encoded differential combinatorial sensitivities (Fig- or ‘‘barcode’’ for each tumor model (Figures 4A and S4). In some ure 5A). Trametinib, in combination with dasatinib, was effective cases, low-dose responses to combination therapy increased in several models, including EMC7310 and EMC29 (Figure 5B). In existing monotherapy sensitivity. For example EMC7310 was these models, trametinib alone induced compensatory tyrosine sensitive to dasatinib, but this sensitivity was augmented by phosphorylation events that were blocked by dasatinib and multiple additional agents. However, there were many combina- associated with increased apoptosis (Figures 5C–5E). These tions that could not be predicted from single-agent sensitivities, effects translated into control of tumorigenic growth in the and, surprisingly, PI3K (phosphatidylinositol 3-kinase) inhibitors corresponding PDX that, based on cell-line data, would be (BYL719 and GDC0941) had little positive impact in combina- predicted to be sensitive to this combination (Figure 5F). In tion with any other agent (Figure 4B; Data S2). To evaluate contrast, another PDX model that would be predicted to be resis- combinatorial potency, multi-drug dose-response analyses tant to this combination failed to respond (Figure S5). Analysis of were conducted and uncovered model-selective synergistic sensitive PDX tissues confirmed suppression of p-ERK1/2, drug interactions (Figures 4C and S4; Data File S2). These data p-Src, tyrosine phosphorylation, and the induction of apoptosis underscore the diversity of therapeutic response and indicate (Figure 5G). that, in spite of using multiple combinations, there were no uni- The combination of trametinib with docetaxel was the most versal vulnerabilities in PDAC; however, subsets of cases were potent treatment identified in cell-line screening, with synergistic sensitive to select combination treatments. interaction observed in several models (Figure 6A). In xenograft The robust responses to select combinations did not asso- assays, this combination potently suppressed tumor growth for ciate with common genomic lesions (e.g., KRAS, TP53, more than 30 days in two PDX models (Figures 6B and S6). In SMAD4) present in models (data not shown), nor did they corre- contrast, single-agent treatments with docetaxel or trametinib spond to previously described predictive markers. For example, were insufficient to prevent rapid disease progression (Figures the ratio of MCL1/NOXA implicated in resistance to ABT737 6B and S6). Analysis of PDX tissue showed that MEK inhibition could not predict response to ABT737 in combination with generally limited proliferation and that docetaxel elicited cell chemotherapy (Geserick et al., 2014). Based on these results, death, while the combination yielded a dual response (Figure 6C). gene expression features were explored relative to the response To determine whether this therapeutic sensitivity was selective, to select combination treatments. Analysis of the top upregu- the same treatment was deployed against two PDXs models lated and downregulated genes revealed a relatively broad that would be predicted, based on the cell-line data, to be resis- spectrum of deregulated biological processes. The elastic-net tant. In these models, there was veritably no response to the method was used to more stringently define genes that were treatment (Figures 6D and 6E). These data indicate that thera- positively or inversely correlated with the combinations tested. peutic sensitivities identified in the primary cell lines translate These combined approaches yielded few genes that were repro- into responses in PDX derived from the same primary tumor. ducibly associated with combinatorial sensitivities (Figure S4). To further interrogate the idea that sensitivities are model and Unfortunately, data from clinical studies using the tested combi- combination specific, the EMC519 PDX (resistant to trametinib + nations are not available to validate the performance of the po- docetaxel and trametinib + dasatinib combinations but sensitive

Figure 3. Exceptional Response to MEK Inhibition (A) Dose-response analysis of EMC828 versus EMC29 and EMC1222 cell-line models exposed to AZD6244. Average and SD are shown from three experiments. (B) Immunoblot analysis of the indicated proteins in the models treated with 0, 0.25, and 1 mM AZD6244. (C) The published KRAS dependence signature was subjected to supervised clustering based on the exceptional response versus resistance. (D) The suppression of bromodeoxyuridine (BrdU) incorporation and the induction of apoptosis were determined for the indicated cell lines, with increasing concentrations of AZD6244 (0, 0.25, and 1 mM). Mean and SD are shown from three independent experiments. p value relative to vehicle control was determined by unpaired t test; ***p < 0.001. (E) Representative crystal violet staining for cell viability in the indicated cell lines exposed to AZD6244. CTRL, control. (F) Representative micrographs of the EMC828 model grown as an organoid in 3D culture. Images were taken at 43; scale bars, 1 mm. Cells grown in 3D and 2D from three independent cultures were subjected to increasing doses of AZD6244, and viability was determined. Mean and SD are shown. (G) The EMC828 xenograft was randomized to treatment with vehicle (n = 5) or AZD6244 (n = 5) until the control arm became moribund. The mean and SEM is shown; statistical significance was determined by ANOVA. (H) Tumor weight from the indicated PDX models treated with vehicle or AZD6244 was determined at sacrifice. Statistical analysis was determined by unpaired t test. Error bars indicate SD. (I) Representative immunohistochemical images of pERK1/2 and Ki67. Scale bars, 100 mm. Quantification of the staining and the average and SD from more than ten high-power fields are shown. Statistical analysis was by unpaired t test. ***p < 0.001.

Cell Reports 16, 2017–2031, August 16, 2016 2023 Figure 4. Combination Treatments (A) Representation of drug sensitivity to combination treatment. The sensitivity of all treatments was graphed, and the color bar indicates fractional survival relative to DMSO control. (B) Heatmap clustering the AUC values of effective combinations in at least one of the cell lines. Cell lines and combinations were clustered based on Euclidean distance. (C) Representative dose-response analysis of the trametinib and dasatinib combination in a model that yielded synergistic toxicity. Value shown in the heatmap is percent survival minus 100 (i.e., 0 indicates no effect on survival; À100 indicates complete lack of survival).

2024 Cell Reports 16, 2017–2031, August 16, 2016 A B

C DE

F G

Figure 5. Combinatorial Sensitivity to Dasatinib with Trametinib (A) Combination treatment response barcode from a cell line that is sensitive to dasatinib and trametinib combination. (B) Representative dose-response analysis of the trametinib-dasatinib combination in models that yielded synergistic toxicity. Value shown in the heatmap is percent survival minus 100 (i.e., 0 indicates no effect on survival; À100 indicates complete lack of survival), and the combination index for two cell lines is shown. (C) Representative immunoblot analysis of the indicated proteins in the sensitive model exposed to the indicated agents: trametinib, 100 nM; dasatinib, 100 nM; combination, 50 nM of each agent. (D) Induction of cell death as a function of drug treatment from three independent cultures. The average and SD are shown; statistical analysis was by t test comparing the combination to single-agent dasatinib treatment. **p < 0.01; ***p < 0.001. (E) Representative 43 micrographs of cells on plates treated with the indicated drugs (scale bars, 1 mm). (F) The PDX was randomized to treatment with vehicle control (n = 5) or the combination of trametinib and dasatinib (n = 5). The average tumor volume and SEM are plotted. Mice were sacrificed when the control arm became moribund; statistical significance was determined by ANOVA. (G) Representative staining of the indicated markers by immunohistochemistry. Scale bars, 100 mm. to trametinib + everolimus in cell-line screens) was treated with therapy. As shown, the BCL2 inhibitor ABT737 and the check- trametinib and everolimus (Figure 6F). This combination yielded point kinase inhibitor AZD7762 resulted in model-specific measurable control of the PDX tumor growth (Figure 6F). augmentation of the response to either gemcitabine or docetaxel Together, these data suggest that cell models could be used (Figures 7A–7C). ABT737 was particularly potent when com- to infer drug sensitivities to direct the combinatorial treatments. bined with gemcitabine or docetaxel in the EMC29 and EMC3226 lines, respectively (Figures 7B and 7C). In contrast, Model-Guided Chemotherapy Regimens AZD7762 impacted on a separate collection of cell lines. In the analysis of the combination drug screening data, it was These data suggest underlying cellular vulnerabilities to apparent that select targeted agents cooperate with chemo- apoptotic or cell-cycle checkpoint inhibitors. The response to

Cell Reports 16, 2017–2031, August 16, 2016 2025 A B

D E

(legend on next page)

2026 Cell Reports 16, 2017–2031, August 16, 2016 the combination of ABT737 and gemcitabine was synergistic in class of drugs were not uniformly present in cases harboring mu- two cell models (Figure 7D) and, in PDX, translated to stable dis- tations in chromatin-remodeling genes. This finding suggests ease that was associated with a modest effect of therapy on that, although tumors acquire genetic alterations in specific proliferation but a robust induction of apoptosis (Figure 7E). genes, the implicated pathway may not be functionally inactive Similarly, the combination of AZD7762 with gemcitabine yielded or therapeutically actionable. Therefore, annotated patient- disease control in the PDX model, and the response was accom- derived models provide a unique test bed for interrogating spe- panied by extensive DNA damage (Figure 7F). Together, these cific therapeutic dependencies in a genetically tractable system. data similarly illustrate that combinations with chemotherapy can be effective but must be directed. Empirical Definition of Therapeutic Sensitivities and Clinical Relevance DISCUSSION Cell lines offer the advantage of the ability to conduct high- throughput approaches to interrogate many therapeutic agents. PDAC has a particularly poor prognosis, and even with new tar- A large number of failed clinical trials have demonstrated the dif- geted therapies and chemotherapy, the survival is poor. Here, ficulty in treating PDAC. Based on the data herein, the paucity of we show that patient-derived models can be developed and clinical success is, most probably, due to the diverse therapeutic used to investigate therapeutic sensitivities determined by ge- sensitivity of individual PDAC cases, suggesting that, with an un- netic features of the disease and to identify empirical therapeutic selected patient population, it will be veritably impossible to vulnerabilities. These data reveal several key points that are of demonstrate clinical benefit. Additionally, very few models ex- prime relevance to pancreatic cancer and tumor biology in hibited an exceptional response to single agents across the general. breadth of a library encompassing 305 agents. We could identify only one tumor that was particularly sensitive to MEK inhibition The Challenges of Using Genetic Analysis to Inform and another model that was sensitive to EGFR and tyrosine ki- Treatment in PDAC nase inhibitors. Precision oncology is dependent on the existence of known vul- In contrast to the limited activity of single agents, combination nerabilities encoded by high-potency genetic events and drugs screens yielded responses at low-dose concentrations in the capable of exploiting these vulnerabilities. At present, the reper- majority of models. Specific combinations were effective across toire of actionable genetic events in PDAC is limited. Rare BRAF several models, indicating that, by potentially screening more V600E mutations are identified in PDAC and could represent the models, therapeutic sensitivity clades of PDAC will emerge. In basis for targeted inhibition, as our group and others have previ- the pharmacological screens performed in this study, MEK inhi- ously published (Collisson et al., 2012; Witkiewicz et al., 2015). bition, coupled with MTOR, docetaxel, or tyrosine kinase inhibi- Similarly, germline BRCA deficiency is the basis for ongoing tors, was effective in 30% of models tested. Resistance to poly(ADP-ribose) polymerase (PARP) inhibitor clinical trials MEK inhibitors occurs through several mechanisms, including (Lowery et al., 2011). As shown here, out of 28 cases, only one upregulation of oncogenic bypass signaling pathways such as genetic event was identified that yielded sensitivity to a thera- AKT, tyrosine kinase, or MTOR (mammalian target of rapamycin) peutic strategy. In this case, existence of the matched model signaling. In the clinic, the MEK and MTOR inhibitors (e.g., allowed us to confirm the biological relevance of the STAG2 mu- NCT02583542) are being tested. An intriguing finding from the tation by showing sensitivity of the model to a DNA cross-linking drug screen was sensitivity of a subset of models to combined agent. Therefore, annotated patient-derived models provide a MEK and docetaxel inhibition. This combination has been substrate upon which to functionally dissect the significance of observed to synergistically enhance apoptosis and inhibit tumor novel and potentially actionable genetic events that occur within growth in human xenograft tumor models (Balko et al., 2012; a tumor. McDaid et al., 2005) and is currently being tested in a phase III Another challenge of genomics-driven personalized medicine study in patients with KRAS-mutated, advanced non-small-cell is assessing the effect of specific molecular aberrations on ther- lung adenocarcinoma (Ja¨ nne et al., 2016). Interestingly, in the apeutic response in the context of complex genetic changes models tested herein, there was limited sensitivity imparted present in individual tumors. For example, KRAS has been pro- through the combination of gemcitabine and MEK inhibition. posed to modify therapeutic dependency to EZH2 inhibitors This potentially explains why the combination of MEK inhibitor (Kim et al., 2015), and in the models tested, responses to this and gemcitabine tested in the clinic did not show improved

Figure 6. Patient-Selective Targeting of Trametinib Combinations (A) Dose-response analysis of combination treatment, indicating the synergistic interaction of trametinib and docetaxel in two models. Value shown in the heatmap is percent survival minus 100 (i.e., 0 indicates no effect on survival; À100 indicates complete lack of survival). (B) The PDX was randomized to treatment with single-agent trametinib (n = 4) or docetaxel (n = 4), or the combination (n = 5). Tumor volume was measured, and the mean and SEM are shown. Mice were sacriﬁced when the single-agent mice became moribund or at the end of 33 days of treatment. (C) Representative immunohistochemistry and quantiﬁcation. The average and SD are shown from at least three tumors. Statistical analysis was by unpaired t test. **p < 0.01; ***p < 0.001, relative to the vehicle control. Scale bars, 100 mm. (D and E) Two independent PDX models that were predicted to be resistant to trametinib + docetaxel failed to respond to treatment. p > 0.05. Error bars indicate SEM. (F) Dose-response analysis of combination treatment, indicating the synergistic interaction of trametinib and everolimus from a model resistant to trametinib and docetaxel. Response of the PDX to trametinib and everolimus treatment (n = 5 per arm). Tumor volume was measured, and the mean and SEM is shown.

Cell Reports 16, 2017–2031, August 16, 2016 2027 AB

Figure 7. Differential Cooperation of Apoptotic and Cell-Cycle Checkpoint Inhibitors with Chemotherapy (A) Heatmap showing the AUC of single and combination treatments with ABT737 or AZD7762 and chemotherapy. Cell lines and drugs were clustered based on Euclidian distance. (B) Relative survival of the indicated cell lines treated with increasing concentrations (25, 84, and 250 nM) of gemcitabine or docetaxel, as measured in triplicate from two independent experiments. (legend continued on next page)

2028 Cell Reports 16, 2017–2031, August 16, 2016 efficacy over gemcitabine alone (Infante et al., 2014). Another In spite of these challenges, progressively more effort is promising strategy that emerged from this study involves using going into the development of patient-derived models for CHK or BCL2 inhibitors as agents that drive enhanced sensitivity guidance of disease treatment (Aparicio et al., 2015; Boj to chemotherapy. Together, the data suggest that the majority of et al., 2015; Crystal et al., 2014; van de Wetering et al., PDAC tumors have intrinsic therapeutic sensitivities, but the 2015). Several ongoing trials use PDX models to direct a limited challenge is to prospectively identify effective treatment. repertoire of agents (e.g., NCT02312245, NCT02720796, and ERCAVATAR2015). Given the experience here, PDAC cell lines Patient-Derived Model-Based Approach to Precision would provide the opportunity to rapidly interrogate a larger Medicine portfolio of combinations that could be used to guide patient This study supports a path for guiding patient treatment based care and provide a novel approach to precision medicine. on the integration of genetic and empirically determined sensitiv- Validation of this approach would require the establishment ities of the patient’s tumor (Figure S7). In reference to defined ge- of challenging multi-arm or N-of-1 clinical trials. However, netic susceptibilities, the models provide a means to interrogate considering the dire outcome for PDAC patients and the the voracity of specific drug targets. Parallel unbiased screening long-lasting difficulty in developing effective treatments, this enables the discovery of sensitivities that could be exploited in non-canonical approach might be particularly impactful in the clinic. The model-guided treatment must be optimized, al- pancreatic cancer. lowing for the generation of data in a time frame compatible with clinical decision making and appropriate validation. In the EXPERIMENTAL PROCEDURES present study, the majority of models were developed, cell lines were drug screened, and select hits were validated in PDX Cell Culture and PDX models within a 10- to 12-month window (Figure S7). This chro- Tumor tissue acquisition was performed under an Institutional Review Board (IRB) protocol approved at the University of Texas Southwestern Medical nology would allow time to inform frontline therapy for recurrent Center and Ochsner Clinic. Informed written consent was obtained from all disease for most patients who were surgically resected and patients. PDX models were developed and utilized in accordance with Institu- treated with a standard of care where the median time to recur- tional Animal Care and Use Committee approved protocols at UT South- rence is approximately 14 months (Saif, 2013). Although most western Medical Center. Primary cell models were established and cultured models were generated from surgically resected specimens, on collagen-coated tissue culture plates in supplemented KSF media. Cells two of the models (EMC3226 and EMC62) were established were passaged by trypsinization and used at early passage (p < 5) and late passage (p > 20) for the analysis of drug sensitivity. Established cell lines from primary tumor biopsies, indicating that this approach could were from the ATCC and cultured using published methodology. The detailed be used with only a limited amount of tumor tissue available. In description of these models will be published elsewhere, but the description the context of inoperable pancreatic cancer, application of of the derivation approach is provided in the Supplemental Experimental data from a cell-line screen without in vivo validation in PDX Procedures. would permit the generation of sensitivity data in the time frame compatible with treatment. We acknowledge that model-guided Drug Treatments treatment is also not without significant logistical hurdles, Cell models were subjected to drug screening with libraries, combination treat- including the availability of drugs for patient treatment, clinically ments, and single agents, as summarized in the Supplemental Experimental Procedures. The treatment of PDX models was in accordance with institutional relevant time frames, patient-performance status, toxicity of animal care and use committee (IACUC) protocols at the University of Texas combination regiments, and quality metrics related to model Southwestern Medical Center and is summarized in the Supplemental Exper- development and therapeutic response evaluation. Additionally, imental Procedures. it will be very important to monitor ex vivo genetic and pheno- typic divergence with passage and try to understand the features Immunohistochemistry and Model Analysis of tumor heterogeneity that could undermine the efficacy of Immunohistochemistry was performed on a DAKO stainer using conditions as using models to direct treatment. As shown here, drug sensitiv- described in the Supplemental Experimental Procedures. Immunoblotting, immunofluorescence, RPPA analysis, flow cytometry, and other methods ities remained stable with passage in cell culture and, impor- were performed using standard procedures. The specific features of the tantly, were confirmed in PDX models, suggesting that the experimentation are provided in the Supplemental Experimental Procedures. dominant genetic drivers and related therapeutic sensitivities RNA sequencing was performed on an Illumina instrument with paired-end are conserved. reads.

(C) Relative survival of the indicated cell lines treated with increasing concentrations (25, 84, and 250 nM) of the indicated drug combinations as measured in triplicate from two independent experiments. (D) Summary of the cooperation between ABT737 and gemcitabine in a sensitive model. Dose-response analysis of combination treatments, indicating selective synergistic interaction in the EMC29 model and related cooperative index. Value shown in the heatmap is percent survival minus 100 (0 indicates no effect on survival; À100 indicates complete lack of survival). (E) The PDX was randomized to treatment with vehicle control (n = 3) or gemcitabine and ABT737 (n = 8), and tumor volume was determined. The average tumor volume and SEM are plotted. Mice were sacrificed when the control arm became moribund. Statistical significance was determined by ANOVA. Representative staining and quantitation of the indicated markers by immunohistochemistry. The average and SD are shown. Scale bars, 100 mm. Statistical analysis was by unpaired t test. *p < 0.05; **p < 0.01. (F) A PDX model sensitive to AZD7762 and gemcitabine was treated singly or in combination (n = 5 per arm). Tumor volume was measured as a function of time, and statistical significance was determined by ANOVA. Representative gH2AX staining is shown (scale bars, 50 mm). Error bars indicate SD in (B) and (C) and SEM in (F).

Cell Reports 16, 2017–2031, August 16, 2016 2029 ACCESSION NUMBERS Boj, S.F., Hwang, C.I., Baker, L.A., Chio, I.I., Engle, D.D., Corbo, V., Jager, M., Ponz-Sarvise, M., Tiriac, H., Spector, M.S., et al. (2015). Organoid models of The accession number for the expression values reported in this paper is GEO: human and mouse ductal pancreatic cancer. Cell 160, 324–338. GSE84023. Chen, Z., Cheng, K., Walton, Z., Wang, Y., Ebi, H., Shimamura, T., Liu, Y., Tup- per, T., Ouyang, J., Li, J., et al. (2012). A murine lung cancer co-clinical trial SUPPLEMENTAL INFORMATION identifies genetic modifiers of therapeutic response. Nature 483, 613–617. Collisson, E.A., Sadanandam, A., Olson, P., Gibb, W.J., Truitt, M., Gu, S., Supplemental Information includes Supplemental Experimental Procedures, Cooc, J., Weinkle, J., Kim, G.E., Jakkula, L., et al. (2011). Subtypes of pancre- seven figures, and two data files and can be found with this article online at atic ductal adenocarcinoma and their differing responses to therapy. Nat. http://dx.doi.org/10.1016/j.celrep.2016.07.023. Med. 17, 500–503. Collisson, E.A., Trejo, C.L., Silva, J.M., Gu, S., Korkola, J.E., Heiser, L.M., AUTHOR CONTRIBUTIONS Charles, R.P., Rabinovich, B.A., Hann, B., Dankort, D., et al. (2012). A central role for RAF/MEK/ERK signaling in the genesis of pancreatic ductal adeno- Conceptualization, A.K.W. and E.S.K.; Methodology, E.S.K., A.K.W., C.E., and carcinoma. Cancer Discov. 2, 685–693. B.P.; Software: U.B. and E.M.; Formal Analysis: U.B., C.E., and B.P.; Investi- Crystal, A.S., Shaw, A.T., Sequist, L.V., Friboulet, L., Niederst, M.J., Locker- gation: A.K.W., U.B., C.E., E.M., W.C., B.P., G.B.M., and E.S.K.; Resources: man, E.L., Frias, R.L., Gainor, J.F., Amzallag, A., Greninger, P., et al. (2014). A.K.W., B.P., W.C., G.B.M., E.M.O., and E.S.K.; Data Curation: U.B.; Writing: Patient-derived models of acquired resistance can identify effective drug com- E.S.K., A.K.W., and E.M.O.; Visualization: U.B., C.E., E.S.K., and A.K.W.; Su- binations for cancer. Science 346, 1480–1486. pervision: E.S.K., B.P., and A.K.W.; Funding Acquisition: E.S.K. and A.K.W.; Jointly directed the research and are fully responsible for this manuscript: Deluche, E., Onesti, E., and Andre, F. (2015). Precision medicine for metastatic A.K.W. and E.S.K. breast cancer. Am. Soc. Clin. Oncol. Educ. Book 2015, e2–e7. Evers, L., Perez-Mancera, P.A., Lenkiewicz, E., Tang, N., Aust, D., Kno¨ sel, T., ACKNOWLEDGMENTS Rummele,€ P., Holley, T., Kassner, M., Aziz, M., et al. (2014). STAG2 is a clinically relevant tumor suppressor in pancreatic ductal adenocarcinoma. The authors thank members of the E.S.K. and A.K.W. laboratories for thought- Genome Med. 6,9. provoking discussion, technical assistance, and help with manuscript prepa- Garrido-Laguna, I., Tometich, D., Hu, N., Ying, J., Geiersbach, K., Whisenant, ration. Nicholas Borja and Eboni Holloman assisted with select animal studies. J., Wang, K., Ross, J.S., and Sharma, S. (2015). N of 1 case reports of excep- Ngoc Haoi assisted with informatics and figure development. The Tissue Man- tional responders accrued from pancreatic cancer patients enrolled in first-in- agement Shared Resource of the Simmons Cancer Center (Cheryl Lewis, man- man studies from 2002 through 2012. Oncoscience 2, 285–293. ager) assisted with the acquisition of the tissue and coordinating histological Geserick, P., Wang, J., Feoktistova, M., and Leverkus, M. (2014). The ratio of and immunohistochemical analysis of tumor specimens. The High-Throughput Mcl-1 and Noxa determines ABT737 resistance in squamous cell carcinoma of Screening core and Tissue Management Shared Resource are supported by the skin. Cell Death Dis. 5, e1412. the Cancer Center Support grant of the Simmons Cancer Center (P30 CA142543). The research is supported by grants to A.K.W. and E.S.K. from Infante, J.R., Somer, B.G., Park, J.O., Li, C.P., Scheulen, M.E., Kasubhai, S.M., the NIH (CA142543-05S2). Oh, D.Y., Liu, Y., Redhu, S., Steplewski, K., and Le, N. (2014). A randomised, double-blind, placebo-controlled trial of trametinib, an oral MEK inhibitor, in Received: May 18, 2016 combination with gemcitabine for patients with untreated metastatic adeno- Revised: June 28, 2016 carcinoma of the pancreas. Eur. J. Cancer 50, 2072–2081. Accepted: July 9, 2016 Ja¨ nne, P.A., Mann, H., and Ghiorghiu, D. (2016). Study design and rationale for Published: August 4, 2016 a randomized, placebo-controlled, double-blind study to assess the efficacy and safety of selumetinib in combination with docetaxel as second-line treat- REFERENCES ment in patients with KRAS-mutant advanced non-small cell lung cancer (SELECT-1). Clin. Lung Cancer 17, e1–e4. Almhanna, K., and Philip, P.A. (2011). Defining new paradigms for the treat- Jones, S., Zhang, X., Parsons, D.W., Lin, J.C., Leary, R.J., Angenendt, P., Man- ment of pancreatic cancer. Curr. Treat. Options Oncol. 12, 111–125. koo, P., Carter, H., Kamiyama, H., Jimeno, A., et al. (2008). Core signaling Aparicio, S., Hidalgo, M., and Kung, A.L. (2015). Examining the utility of pa- pathways in human pancreatic cancers revealed by global genomic analyses. tient-derived xenograft mouse models. Nat. Rev. Cancer 15, 311–316. Science 321, 1801–1806. Aronson, S.J., and Rehm, H.L. (2015). Building the foundation for genomics in Kim, K.H., Kim, W., Howard, T.P., Vazquez, F., Tsherniak, A., Wu, J.N., Wang, precision medicine. Nature 526, 336–342. W., Haswell, J.R., Walensky, L.D., Hahn, W.C., et al. (2015). SWI/SNF-mutant Bailey, P., Chang, D.K., Nones, K., Johns, A.L., Patch, A.M., Gingras, M.C., cancers depend on catalytic and non-catalytic activity of EZH2. Nat. Med. 21, Miller, D.K., Christ, A.N., Bruxner, T.J., Quinn, M.C., et al.; Australian Pancre- 1491–1496. atic Cancer Genome Initiative (2016). Genomic analyses identify molecular Kleger, A., Perkhofer, L., and Seufferlein, T. (2014). Smarter drugs emerging in subtypes of pancreatic cancer. Nature 531, 47–52. pancreatic cancer therapy. Ann. Oncol. 25, 1260–1270. Balko, J.M., Cook, R.S., Vaught, D.B., Kuba, M.G., Miller, T.W., Bhola, N.E., Lindeman, N.I., Cagle, P.T., Beasley, M.B., Chitale, D.A., Dacic, S., Giaccone, Sanders, M.E., Granja-Ingram, N.M., Smith, J.J., Meszoely, I.M., et al. G., Jenkins, R.B., Kwiatkowski, D.J., Saldivar, J.S., Squire, J., et al.; College of (2012). Profiling of residual breast cancers after neoadjuvant chemotherapy American Pathologists International Association for the Study of Lung Cancer identifies DUSP4 deficiency as a mechanism of drug resistance. Nat. Med. and Association for Molecular Pathology (2013). Molecular testing guideline for 18, 1052–1059. selection of lung cancer patients for EGFR and ALK tyrosine kinase inhibitors: Biankin, A.V., Piantadosi, S., and Hollingsworth, S.J. (2015). Patient-centric tri- guideline from the College of American Pathologists, International Association als for therapeutic development in precision oncology. Nature 526, 361–370. for the Study of Lung Cancer, and Association for Molecular Pathology. J. Mol. Bitler, B.G., Aird, K.M., Garipov, A., Li, H., Amatangelo, M., Kossenkov, A.V., Diagn. 15, 415–453. Schultz, D.C., Liu, Q., Shih, I.-M., Conejo-Garcia, J.R., et al. (2015). Synthetic Lowery, M.A., Kelsen, D.P., Stadler, Z.K., Yu, K.H., Janjigian, Y.Y., Ludwig, E., lethality by targeting EZH2 methyltransferase activity in ARID1A-mutated can- D’Adamo, D.R., Salo-Mullen, E., Robson, M.E., Allen, P.J., et al. (2011). An cers. Nat. Med. 21, 231–238. emerging entity: pancreatic adenocarcinoma associated with a known

2030 Cell Reports 16, 2017–2031, August 16, 2016 BRCA mutation: clinical descriptors, treatment implications, and future direc- Solomon, D.A., Kim, T., Diaz-Martinez, L.A., Fair, J., Elkahloun, A.G., Harris, tions. Oncologist 16, 1397–1402. B.T., Toretsky, J.A., Rosenberg, S.A., Shukla, N., Ladanyi, M., et al. (2011). McDaid, H.M., Lopez-Barcons, L., Grossman, A., Lia, M., Keller, S., Pe´ rez- Mutational inactivation of STAG2 causes aneuploidy in human cancer. Sci- Soler, R., and Horwitz, S.B. (2005). Enhancement of the therapeutic efficacy ence 333, 1039–1043. of taxol by the mitogen-activated protein kinase kinase inhibitor CI-1040 in van de Wetering, M., Francies, H.E., Francis, J.M., Bounova, G., Iorio, F., nude mice bearing human heterotransplants. Cancer Res. 65, 2854–2860. Pronk, A., van Houdt, W., van Gorp, J., Taylor-Weiner, A., Kester, L., et al. Mirzoeva, O.K., Das, D., Heiser, L.M., Bhattacharya, S., Siwak, D., Gendel- (2015). Prospective derivation of a living organoid biobank of colorectal cancer man, R., Bayani, N., Wang, N.J., Neve, R.M., Guan, Y., et al. (2009). Basal sub- patients. Cell 161, 933–945. type and MAPK/ERK kinase (MEK)-phosphoinositide 3-kinase feedback Vora, S.R., Juric, D., Kim, N., Mino-Kenudson, M., Huynh, T., Costa, C., Lock- signaling determine susceptibility of breast cancer cells to MEK inhibition. erman, E.L., Pollack, S.F., Liu, M., Li, X., et al. (2014). CDK 4/6 inhibitors sensi- Cancer Res. 69, 565–572. tize PIK3CA mutant breast cancer to PI3K inhibitors. Cancer Cell 26, 136–149. Paulson, A.S., Tran Cao, H.S., Tempero, M.A., and Lowy, A.M. (2013). Thera- Waddell, N., Pajic, M., Patch, A.M., Chang, D.K., Kassahn, K.S., Bailey, P., peutic advances in pancreatic cancer. Gastroenterology 144, 1316–1326. Johns, A.L., Miller, D., Nones, K., Quek, K., et al.; Australian Pancreatic Cancer Pettazzoni, P., Viale, A., Shah, P., Carugo, A., Ying, H., Wang, H., Genovese, Genome Initiative (2015). Whole genomes redefine the mutational landscape G., Seth, S., Minelli, R., Green, T., et al. (2015). Genetic events that limit the ef- of pancreatic cancer. Nature 518, 495–501. ficacy of MEK and RTK inhibitor therapies in a mouse model of KRAS-driven Witkiewicz, A.K., McMillan, E.A., Balaji, U., Baek, G., Lin, W.C., Mansour, J., pancreatic cancer. Cancer Res. 75, 1091–1101. Mollaee, M., Wagner, K.U., Koduru, P., Yopp, A., et al. (2015). Whole-exome Saif, M.W. (2013). Advancements in the management of pancreatic cancer: sequencing of pancreatic cancer defines genetic diversity and therapeutic tar- 2013. JOP 14, 112–118. gets. Nat. Commun. 6, 6744. Singh, A., Greninger, P., Rhodes, D., Koopman, L., Violette, S., Bardeesy, N., Yeo, C.J., Cameron, J.L., Lillemoe, K.D., Sitzmann, J.V., Hruban, R.H., and Settleman, J. (2009). A gene expression signature associated with ‘‘K-Ras Goodman, S.N., Dooley, W.C., Coleman, J., and Pitt, H.A. (1995). Pancreatico- addiction’’ reveals regulators of EMT and tumor cell survival. Cancer Cell 15, duodenectomy for cancer of the head of the pancreas. 201 patients. Ann. 489–500. Surg. 221, 721–731, discussion 731–733.

Cell Reports 16, 2017–2031, August 16, 2016 2031