Cell Press Selections: Cancer in 3D by Sartorius

Tell your cell’s story with real-time live-cell analysis

With over 1,500 publications and counting, the IncuCyte® live-cell analysis system automatically acquires and analyzes images so researchers can get answers faster – and with presentation-ready images and graphs – get published faster! The IncuCyte S3 system is a flexible assay platform that enables real-time, automated measurements of cell health, movement and function - inside your incubator.

Get started at Essenbio.com/incucyte

© 2018 Essen BioScience, 300 West Morgan Road, Ann Arbor, Michigan, USA 48108. All rights reserved. IncuCyte®, Essen BioScience® and all names of Essen BioScience products are registered trademarks and the property of Essen BioScience unless otherwise specified. Essen Bioscience is a Sartorius Company.

Exploring hallmarks of cancer with real-time live-cell analysis IncuCyte® S3 Live-Cell Analysis System is revolutionizing the field of Oncology research. Scientists around the world are studying cancer cell biology, non-invasively and in real-time, to gain unique insights using advanced cell models. — Cancer Stem Biology — 3D models — Immuno-oncology

Discover more at Essenbio.com/oncology

Apoptosis

Cytotoxicity

Proliferation

Immune-cell killing

Phagocytosis

Angiogenesis

Cell Migration and Invasion

Antibody Internalization

Discover more at Essenbio.com/oncology

Apoptosis

Cytotoxicity

Proliferation

Immune-cell killing

Phagocytosis

Angiogenesis

Cell Migration and Invasion

Antibody Internalization

Measure dynamic interactions between immune cells and cancer cells.

Immune cell activation, chemotaxis and transendothelial migration.

Tumor cell proliferation and metastatic potential.

Discover more at Essenbio.com/immuno-oncology

Immune cell killing of tumor cells and clearance by phagocytosis. © 2018 Essen BioScience, 300 West Morgan Road, Ann Arbor, Michigan, USA 48108. All rights reserved. IncuCyte®, Essen BioScience® and all names of Essen BioScience products are registered trademarks and the property of Essen BioScience unless otherwise specified. Essen Bioscience is a Sartorius Company.

Foreword Converging lines of research have highlighted strong connections between stem cell biology and cancer. Work over the past 20 years has shown that 3D-based methods for growing stem cells provide a powerful platform for asking fundamental questions about their underlying biology, and the advent of organoid technologies has provided in vitro platforms for studying tissue homeostasis and pathology in a dish. Such tools are now being brought to bear in cancer research. In this edition of Cell Press Selections, we feature reviews and research articles that illustrate how the intersections of 3D-based culture methods and stem cell biology can be used to model cancer and inform new opportunities for therapeutic interventions. From understanding basic mechanisms into putative cancer stem cell populations to modeling patient tumors to providing new platforms for drug discovery, the research included here charts recent advances in the quickly emerging area of using 3D cultures to inform our understanding of human disease. We hope this collection will stimulate thought and ideas for developing more physiological methods for studying putative contributions of stem cells to tumorigenesis. To stay abreast in such a rapidly emerging field, we hope that you will visit www.cell.com on a regular basis to keep up with the latest work in developing increasingly realistic human cancer models. Finally, we are grateful for the support of Essen BioScience (now part of Sartorius), who helped to make the publication of this collection possible. Jonathan Saxe Editor, Cell Stem Cell

For more information about Cell Press Selections:

Gordon Sheffield Program Director, Cell Press Selections g.sheffield@cell.com 617-386-2189

Tell your cell’s story with real-time live-cell analysis

Get started at Essenbio.com/incucyte

Cancer in 3D Stem Cell-Based Models of Tumor Biology

Opinions and Forums Personalized Cancer Medicine: An Organoid Approach

Hamidreza Aboulkheyr Es, Leila Montazeri, Amir Reza Aref, Massoud Vosough, and Hossein Baharvand

3D Biomimetic Cultures: The Next Platform for Cell Biology

Christopher S. Chen

Modeling Cancer with Pluripotent Stem Cells

Julian Gingold, Ruoji Zhou, Ihor R. Lemischka, and Dung-Fang Lee

Articles and Resources Prospective Derivation of a Living Organoid Biobank of Colorectal Cancer Patients

Marc van de Wetering, Hayley E. Francies, Joshua M. Francis, Gergana Bounova, Francesco Iorio, Apollo Pronk, Winan van Houdt, Joost van Gorp, Amaro Taylor-Weiner, Lennart Kester, Anne McLaren-Douglas, Joyce Blokker, Sridevi Jaksani, Sina Bartfeld, Richard Volckman, Peter van Sluis, Vivian S.W. Li, Sara Seepo, Chandra Sekhar Pedamallu, Kristian Cibulskis, Scott L. Carter, Aaron McKenna, Michael S. Lawrence, Lee Lichtenstein, Chip Stewart, Jan Koster, Rogier Versteeg, Alexander van Oudenaarden, Julio Saez-Rodriguez, Robert G.J. Vries, Gad Getz, Lodewyk Wessels, Michael R. Stratton, Ultan McDermott, Matthew Meyerson, Mathew J. Garnett, and Hans Clevers

Human Pancreatic Tumor Organoids Reveal Loss of Stem Cell Niche Factor Dependence during Disease Progression

Takashi Seino, Shintaro Kawasaki, Mariko Shimokawa, Hiroki Tamagawa, Kohta Toshimitsu, Masayuki Fujii, Yuki Ohta, Mami Matano, Kosaku Nanki, Kenta Kawasaki, Sirirat Takahashi, Shinya Sugimoto, Eisuke Iwasaki, Junichi Takagi, Takao Itoi, Minoru Kitago, Yuko Kitagawa, Takanori Kanai, and Toshiro Sato

A Biobank of Breast Cancer Explants with Preserved Intra-tumor Heterogeneity to Screen Anticancer Compounds

Alejandra Bruna, Oscar M. Rueda, Wendy Greenwood, Ankita Sati Batra, Maurizio Callari, Rajbir Nath Batra, Katherine Pogrebniak, Jose Sandoval, John W. Cassidy, Ana Tufegdzic-Vidakovic, Stephen-John Sammut, Linda Jones, Elena Provenzano, Richard Baird, Peter Eirew, James Hadfield, Matthew Eldridge, Anne McLaren-Douglas, Andrew Barthorpe, Howard Lightfoot, Mark J. Oâ&#x20AC;&#x2122;Connor, Joe Gray, Javier Cortes, Jose Baselga, Elisabetta Marangoni, Alana L. Welm, Samuel Aparicio, Violeta Serra, Mathew J. Garnett, and Carlos Caldas

(continued on next page)

Inhibition of TRF1 Telomere Protein Impairs Tumor Initiation and Progression in Glioblastoma Mouse Models and Patient-Derived Xenografts

Leire Bejarano, Alberto J. Schuhmacher, Marinela Méndez, Diego Megías, Carmen Blanco-Aparicio, Sonia Martínez, Joaquín Pastor, Massimo Squatrito, and Maria A. Blasco

Secreted Phospholipases A2 Are Intestinal Stem Cell Niche Factors with Distinct Roles in Homeostasis, Inflammation, and Cancer

Matthias Schewe, Patrick F. Franken, Andrea Sacchetti, Mark Schmitt, Rosalie Joosten, René Böttcher, Martin E. van Royen, Louise Jeammet, Christine Payré, Patricia M. Scott, Nancy R. Webb, Michael Gelb, Robert T. Cormier, Gérard Lambeau, and Riccardo Fodde

Organoid Models of Human and Mouse Ductal Pancreatic Cancer

Sylvia F. Boj, Chang-Il Hwang, Lindsey A. Baker, Iok In Christine Chio, Dannielle D. Engle,Vincenzo Corbo, Myrthe Jager, Mariano Ponz-Sarvise, Hervé Tiriac, Mona S. Spector, Ana Gracanin, Tobiloba Oni, Kenneth H. Yu, Ruben van Boxtel, Meritxell Huch, Keith D. Rivera, John P. Wilson, Michael E. Feigin, Daniel Öhlund, Abram HandlySantana, Christine M. Ardito-Abraham, Michael Ludwig, Ela Elyada, Brinda Alagesan, Giulia Biffi, Georgi N. Yordanov, Bethany Delcuze, Brianna Creighton, Kevin Wright, Youngkyu Park, Folkert H.M. Morsink, I. Quintus Molenaar, Inne H. Borel Rinkes, Edwin Cuppen, Yuan Hao, Ying Jin, Isaac J. Nijman, Christine Iacobuzio-Donahue, Steven D. Leach, Darryl J. Pappin, Molly Hammell, David S. Klimstra, Olca Basturk, Ralph H. Hruban, George Johan Offerhaus, Robert G.J. Vries, Hans Clevers, and David A. Tuveson

Glut3 Addiction Is a Druggable Vulnerability for a Molecularly Defined Subpopulation of Glioblastoma

Érika Cosset, Sten Ilmjärv, Valérie Dutoit, Kathryn Elliott, Tami von Schalscha, Maria F. Camargo, Alexander Reiss, Toshiro Moroishi, Laetitia Seguin, German Gomez, Jung-Soon Moo, Olivier Preynat-Seauve, Karl-Heinz Krause, Hervé Chneiweiss, Jann N. Sarkaria, Kun-Liang Guan, Pierre-Yves Dietrich, Sara M. Weis, Paul S. Mischel, and David A. Cheresh

On the cover: 3D culture methods provide a powerful platform for studying stem cells and cancer. This image represents how in vitro 3D technologies, such as spheroid cultures and organoids, can model human organs and disease, breaking human cancer down into tangible and tractable experimental systems. Image by Hayri Er/iStock. Cover design by Yvonne Blanco.

Measure dynamic interactions between immune cells and cancer cells.

Immune cell activation, chemotaxis and transendothelial migration.

Tumor cell proliferation and metastatic potential.

Discover more at Essenbio.com/immuno-oncology

Special Issue: Tissue Engineering

Opinion

Personalized Cancer Medicine: An Organoid Approach Hamidreza Aboulkheyr Es,1 Leila Montazeri,2 Amir Reza Aref,3,4,5 Massoud Vosough,1 and Hossein Baharvand1,6,* Personalized cancer therapy applies speciﬁc treatments to each patient. Using personalized tumor models with similar characteristics to the original tumors may result in more accurate predictions of drug responses in patients. Tumor organoid models have several advantages over pre-existing models, including conserving the molecular and cellular composition of the original tumor. These advantages highlight the tremendous potential of tumor organoids in personalized cancer therapy, particularly preclinical drug screening and predicting patient responses to selected treatment regimens. Here, we highlight the advantages, challenges, and translational potential of tumor organoids in personalized cancer therapy and focus on gene–drug associations, drug response prediction, and treatment selection. Finally, we discuss how microﬂuidic technology can contribute to immunotherapy drug screening in tumor organoids.

Highlights

Importance of Personalized Tumor Models

A tumor organoid, in which cellular and molecular heterogeneity of tumor cells is preserved, has emerged as a promising platform.

Personalized cancer medicine is an approach to tailoring effective therapeutic strategies for each patient according to a tumor’s genomic characterization. There is an urgent demand for research in personalized tumor modeling to conﬁrm the functional aspects of genomic drug response predictions in the preclinical setting. While different tumor models, such as tumor cell lines and patient-derived tumor xenografts, have been proposed, the drawbacks of each model have limited their applications as personalized tumor models.

In conventional approaches to cancer therapy, most patients with a particular type of cancer receive similar ‘one-size-fits-all’ treatments. However, it has recently become clear that certain treatments work well for some patients but do not show promising results in others. Currently, individualized cancer treatments are progressively improving due to better characterization of the molecular and pharmacogenomic features of tumors. This recent approach, called precision or personalized cancer medicine, can be described as a ‘one dose, one patient’ treatment. Each tumor is associated by a heterogeneous tumor microenvironment, which can significantly affect the response to therapy and clinical outcomes. However, the use of gene–drug association (see Glossary) treatment strategies may be limited due to a lack of biological understanding of tumor response to drugs [1]. In other words, detecting mutations (e.g., [354_TD$IF]EGFR or PIK3CA mutations) matched with approved, targeted drugs (e.g., EGFR and PIK3CA inhibitors) does not necessarily mean that the molecular alterations in these pathways are sensitive to the selected therapy [2]. In the field of personalized cancer medicine, the link between functional genomics and pathological data to patient outcome is a major challenge. To address this challenge, different personalized tumor models have been proposed, including cancer cell lines, patient-derived xenografts (PDXs), and 3D culture tumor models, such as organoid culture methods. Here, we discuss the patient-derived tumor organoid as a novel and promising tumor model in personalized cancer therapy. We highlight the potential and challenges of this model system for preclinical drug screening and predicting patient outcomes in comparison with pre-existing 358

https://doi.org/10.1016/j.tibtech.2017.12.005

Recently, numerous studies highlighted the application of tumor organoids in personalized cancer medicine in terms of gene–drug association treatment, the identiﬁcation of new therapies, and prediction of patient outcome.

Department of Stem Cells and Developmental Biology, Cell Science Research Center, Royan Institute for Stem Cell Biology and Technology, ACECR, Tehran, Iran 2 Department of Cell Engineering, Cell Science Research Center, Royan Institute for Stem Cell Biology and Technology, ACECR, Tehran, Iran 3 Belfer Center for Applied Cancer Science, Dana-Farber Cancer Institute, Harvard Medical School, Boston, MA, USA 4 Department of Medicine, Brigham and Women's Hospital and Harvard Medical School, Boston, MA, USA

tumor models. In addition, we point out how other technologies, such as microﬂuidic culture systems, can rise to the challenge of testing immunotherapy drugs on tumor organoids.

Conventional Tumor Models Monolayer and 3D Culture Models The initial results of in vitro tumor modeling were obtained from tumor-derived cell lines due to their high-throughput capacity for pharmaceutical drug screening [3]. However, there is evidence [4] showing the limitations of cancer cell lines, such as the lack of tumor and stromal cells, and interactions between the cells and the extracellular matrix (ECM) [4]. In addition, cancer cell lines lack immune–tumor cell interactions in vitro, whereas immune cells can dramatically alter the efﬁcacy of cancer therapies in patients. Thus, available cancer cell lines have largely failed to model a tumor microenvironment to evaluate new target and chemotherapy medicines [5]. After the emergence of 3D cultures, partial mimicking of the tumor microenvironment allowed researchers to test the functionality of drugs or evaluation of chemoresistance in the presence of cell–cell and cell–ECM interactions in a 3D architecture [6]. Nonetheless, these methods also face several limitations (Box 1) and have so far failed in applications to personalized tumor modeling.

Department of Medical Oncology, Dana Farber Cancer Institute, Boston, MA, USA 6 Department of Developmental Biology, University of Science and Culture, Tehran, Iran *Correspondence: Baharvand@Royaninstitute.org (H. Baharvand).

PDX Models Patient-derived tumor xenografts are generated by transplanting a patient’s tumor cells to an immune-deficient mouse in a subcutaneous and/or orthotopic manner. These models provide promising platforms for cancer research and drug development [9]. They are widely utilized for drug discovery, biomarker detection, and preclinical drug evaluation [10]. Similar to the original tumor, PDX models can fully recapitulate the interactions of tumor cells with their surrounding stromal cells and the ECM, with the exception of interactions with the immune system. These models can mimic physiological and biochemical effects of drugs against an individual patient’s tumors. PDX models have previously shown potential to investigate resistance mechanisms and determine novel treatment approaches [11–13]. Although the development of PDX models has improved the quality of cancer research and translation of in vitro findings to in vivo, their application in precision cancer therapy has been restricted. The choice of effective and successful engraftment methods for different types of tumors is the main limitation of PDX models. For instance, in hormone-sensitive breast cancer, the rate of engraftment is very low compared with triple-negative breast cancer [14]. Time is a critical factor in real-time personalized medicine. Generation of a successful PDX model requires 4–8 months, which results in a time delay between engraftment in mice and scheduled treatment regimens for the patient [10]. Generation of specific subtypes of tumors is another limitation of PDX models. The majority of PDX models have been generated from invasive and metastatic tumors, whereas Box 1. Limitations of Conventional 3D Culture Models Conventional 3D tumor models, known as tumor spheroids, can be generated from single cell types of cancer cell lines (homotypic spheroids) and/or cocultured with other cell types (heterotypic spheroids). These models have a series of challenges that limit their applications as preclinical tumor models. These challenges include variations in 3D culture methods, lack of immune cell interactions in the culture, and inability to fully mimic the tumor microenvironment in terms of cell types and their spatiotemporal architecture [7]. In terms of gene expression profiles, the comparison between original tumors and corresponding cell line-based tumor spheroids has shown significant differences, where common mutations can be observed in cancer cell lines, whereas rare mutations are not preserved. Thus, cell line-based models are unable to imitate the complete genomic background of tumors and, consequently, the drug response of targeted therapy agents [8].

Trends in Biotechnology, April 2018, Vol. 36, No. 4

359

nonmetastatic tumors showed engraftment failure in mice (e.g., in colorectal cancer [15] and hormone-sensitive breast cancer studies) [16,17]. Considering the role of the tumor microenvironment in evaluating the drug response [4,18], replacing human stromal cells with murine cells in PDX models could affect drug screening results and consequently, predictions of the drug response [18]. In addition, in PDX models, the absence of immune system components results in an enhanced engraftment rate. This characteristic of immune-deﬁcient mouse models makes them inappropriate for screening and functional analysis of new immune-therapeutic drugs [10]. Despite marked genetic similarities between patient tumors and related PDX models, a series of changes exist in speciﬁc genes and drug targets. In the context of modeling targeted therapy in PDX models, induced genomic alterations and rearrangements in tumor cells that result from passaging tumor cells into a new mouse limit their application in targeted therapy and gene– drug association studies [19]. Researchers have sought to overcome these challenges in recent years with the advent of new 3D cell culture methods, known as organoid culture. This has opened up new horizons to use reliable tumor models in cancer research and therapeutics, in particular personalized cancer treatment.

Tumor Organoid as a Preclinical Model for Personalized Cancer Therapy Currently, there is no consensus deﬁnition for ‘organoids’ (Box 2). To date, different organoid models have been developed for a variety of normal tissues, including the colon and small intestine [30,31], stomach [32–34], liver [35,36], mammary glands [37], retina, and brain [38]. In cancer research, optimizing various culture conditions (Table 1) has resulted in the development of numerous patient-derived tumor organoids, including colon [30], prostate [21,39], gastric [40], breast [37,41], and pancreatic cancers [42,43], in addition to endometrial/ovary carcinomas, uterine carcinosarcoma, urothelial carcinoma, and renal carcinoma [44]. Organoid models have applications in many research areas, such as biomedical research [45], genomic analysis of various diseases, and therapeutic studies [46–48]. Recently, organoids

Box 2. Patient-Derived Tumor Organoids Organoids are 3D cell cultures that preserve numerous key features of the represented organ. The organoids contain multiple and organ-specific cell types with a spatial architecture similar to that of the corresponding organ. These models can recapitulate some key functions of that organ. Organoids may be generated from one or a few cells derived from primary tissue samples, adult stem cells, or the directed differentiation of pluripotent stem cells [20–22]. Tumor organoids are 3D cultures of cancerous cells that can be derived from tumor tissues for better mimicking the composition of a tumor in the body [20,21,23–27]. The first organoid culture was reported in murine intestinal cells, further developed for other organs, and eventually translated into human cells [28]. These features of organoids made them useful tools for cancer research and therapy for in vitro and clinical studies [29]. Patient-derived tumor organoid culture methods vary depending on the tumor type. A standard and robust technique for primary tumor organoid culture still needs to be developed. Recently, different tissue-specific culture conditions and methods for generation of different tumor organoids have been developed (Table I). The tumor organoid culture is initiated by mechanical and enzymatic digestion of tumor tissue into small pieces, followed by embedding this tissue into a 3D matrix (mostly Matrigel) as a biomimetic scaffold. The cellular architecture of the organoids and its behavior significantly depend on the matrix composition. Matrigel contains laminin, entactin, proteoglycans, and collagen IV. Although it can be enriched with numerous growth factors, reduced growth factor media are mostly used for tumor organoid culture. Paracrine signaling is simulated by a cocktail of different tumor tissue-specific growth factors. The most commonly used growth factors include Wnt3A, R-spondin-1, epidermal growth factor, and bone morphogenetic protein antagonist Noggin. Several additional factors can enhance organoid culture and passage, including Rho-kinase inhibitor Y-27632 and GSK3b-kinase inhibitors. Recently, numerous tissue-specific culture media have been developed

360

Trends in Biotechnology, April 2018, Vol. 36, No. 4

Glossary Afatinib: a tyrosine kinase inhibitor for EGFR and ErbB 2 (HER2). APC: adenomatous polyposis coli, a negative regulator of beta catenin and a modulator of its interaction with E-cadherin. Mutations in the APC gene can result in colon cancer. Biopsy: extraction of small pieces of tissue from a specific site (e.g., tumor tissue). Buparlisib: a targeted therapy drug that inhibits the pan-class I phosphatidylinositol 3-kinase (PI3K) family of lipid kinases. Chemoresistance: resistance of tumor cells to the effects of chemotherapeutics. Chemotherapy: a category of cancer treatment that uses one or more anticancer drugs. Combination therapy: treatment in which a patient is given two or more drugs. EGFR inhibitors: a type of targeted drug therapy that inhibits EGFR on cancer cells. FOLFOX regimen: a chemotherapy regimen composed of folinic acid, 5fluorouracil, and oxaliplatin. Gene–drug association: selection of a drug based on the genomic abnormalities of a tumor. HDAC inhibitors: small molecules that inhibit histone deacetylase. Immunotherapy: a cancer treatment that attempts to stimulate the immune system to destroy tumors. Interferon regulatory factor 8: a protein that plays a key role in regulation of lineage commitment and maturation of myeloid cells. Multicellular tumor spheroids: common methods for a 3D culture of cancer cells. Neoadjuvant chemotherapy: the use of chemotherapy regimens prior to surgery. Off-label drugs: drugs used as treatment that lack FDA approval. Olaparib: a poly (ADP-ribose) polymerase (PARP) inhibitor. Pan-HER inhibitors: small-molecule agents that inhibit HER1, HER2, and HER4 receptors on cancer cells. Patient-derived xenograft (PDX): the transplantation of patient’s tumor cells into immune-deficient mice. Somatic copy number alterations: also called ‘copy number variation’ when a section of the genome is

for different tumor organoid cultures. In an attempt to study the hallmarks of cancer on tumor organoids, numerous in vitro assays have been introduced (Table II).

Table I. Characterization of Different Cancer Modelsa Features

Cell lines

PDXs

Organoids

Success rate of initiation

Expansion

Cancer subtype modeling

â&#x20AC;&#x201C;

Biological stability

Genetic manipulation

â&#x20AC;&#x201C;

High-throughput drug screening

â&#x20AC;&#x201C;

Low-throughput drug screening

Ease of downstream assays

Cost beneďŹ ts

â&#x20AC;&#x201C;

Time consumption for modeling

Ease of maintenance

â&#x20AC;&#x201C;

repeated. The numbers of these repeats play key roles in cancer diagnosis and prognosis. Splenocytes: white blood cells located in the spleen. Targeted therapy: a cancer treatment that employs smallmolecule inhibitors to speciďŹ cally target cancer cells.

Adapted, with permission, from [29]. a , Best; , suitable; , possible; and â&#x20AC;&#x201C;, unsuitable.

Table II. Possible In Vitro Assays in Tumor Organoids to Study the Hallmarks of Cancera

Hallmark of cancer

Possible tumor organoid assays

Evading growth suppressors

Proliferation assays and size measurement

Avoiding immune destruction

Coculture assays with immune cells using microďŹ&#x201A;uidic technology

Replicative immortality

Repeat organoid passaging

Tumor-induced inďŹ&#x201A;ammation

Treatment with inďŹ&#x201A;ammatory cytokines

Invasive and metastasis

96-well trans-well migration and invasion assays

Angiogenesis

Coculture assays with endothelial cells using microďŹ&#x201A;uidic technology

Genome instability and mutation

Whole-genome and/or targeted sequencing.

Resistance to cell death

High-throughput viability assays: MTT, PI

Deregulating cellular energy

Oxygen consumption rate measurement and extracellular acidiďŹ cation rate assays

Sustaining proliferative signaling

Proliferation and viability assays: calcein-AM/PI, CellTiter-Glo

Abbreviations: MTT, 3-(4,5-dimethylthiazol-2-yl)-2,5-diphenyltetrazolium bromide; PI, propidium iodide.

have shown great potential in drug discovery [23], cytotoxicity investigations of new therapeutic compounds [49], in vivo modeling of speciďŹ c cancer metastasis-associated genes [50], and personalized cancer treatments [25,26]. A large body of evidence has provided a proof of concept for applying tumor organoids in personalized cancer therapy, conďŹ rming the genomic and functional resemblances between patient-derived tumor organoids and their original specimens [22,26,27,51]. In contrast to conventional cancer models that require large sample sizes (e.g., PDX models), organoids can be cultured from a small sample size, derived from needle biopsy, with a high success rate for personalized tumor modeling [43]. In patients with resected tumors, multiple organoids can be generated from different areas of the tumor to better mimic tumor heterogeneity [52,53]. Tumor organoids show tremendous potential for modeling speciďŹ c cancer subtypes that have unique genomic mutations [54]. Taken together,

Trends in Biotechnology, April 2018, Vol. 36, No. 4

361

Table 1. Developed Culture Conditions for the Generation of Different Tumor Organoids from Tumor Specimensa Cancer type

Aim of study

Source of organoids

Digestion condition

Culture condition

Refs

Prostate

Tumor modeling

Patient tumor specimens (human)

5 ml of 5 mg/ml collagenase Type II + advanced DMEM/ F12 (ADMEM/F12)

Growth factor-reduced Matrigel + ADMEM/F12

[21]

Prostate

Lineage and cell transition monitoring

Murine and human prostates: Single cells and bulk

Collagenase Type II + ADMEM/F12

Cells were seeded in growth factor-reduced Matrigel (Corning) and overlaid with medium containing the growth factors (EGF, R-spondin-1, Noggin, TGF-b/ALK inhibitor A83-01, dihydrotestosterone, FGF10, FGF2, prostaglandin E2, SB202190, nicotinamide, and DHT)

[55]

Breast

Metastasis at tumor’s leading edge

Human tumor specimens

Collagenase with or without trypsin + DMEM + DNase

Matrigel or 3D collagen-I, human MEM + insulin, EGF, hydrocortisone, and cholera toxin

[37,56]

Breast

Detect drug response of organoids

Human primary tumor

Macrosuspensions of tissue (50–300 mm) by mechanical cutting of the tissue with a scalpel and surgical scissors

Matrigel + DMEM: F12 + EGF + hydrocortisone + insulin (5 mg/ml) + penicillin– streptomycin. Making gel on coverslips

[57]

Renal

Method for renal carcinoma 3D culture

Renal carcinoma specimens

EGM2 + collagenase Type IV (5 mg/ml), 40–50 min with vortexing at 10-min intervals

Renal ECM scaffold + EGM2

[44]

Colon

Genetic diversity of patientderived tumor organoids and the original tumor biopsy

Human tumor specimen

N/A

Basement membrane matrix, growth factor-reduced Matrigel + ADMEM/F-12 Hams, penicillin–streptomycin, HEPES, GlutaMAX, Rspondin-conditioned medium, Noggin-conditioned medium, B27, N-acetyl-cysteine, nicotinamide, EGF, gastrin, TGFb Type I receptor inhibitor A83-01, p38 MAPK inhibitor (p38i) SB202190, prostaglandin E2, and Primocin

[22]

Colon

Organoid biobank, personalized medicine

Human colon tumor

N/A

Human intestinal stem Cell medium (HISC): Basal culture medium with Wntconditioned medium, Rspondin-conditioned medium, Noggin-conditioned medium, B27, N-acetylcysteine, nicotinamide, human EGF, gastrin, A83-01, SB202190, prostaglandin E2, and Primocin

[26]

Colon

Modeling speciﬁc subtype of colon cancer

Human colon and intestinal tumors

Mechanically minced with EDTA cold chelation buffer (distilled water with Na2HPO4, KH2PO4, NaCl, KCl, sucrose, 5 d-sorbitol, dl-dithiothreitol) for 30 min

Advanced Dulbecco’s modiﬁed Eagle medium/F12 with penicillin–streptomycin, HEPES, GlutaMAX, B27, gastrin I, and N-acetylcysteine

[58]

362

Trends in Biotechnology, April 2018, Vol. 36, No. 4

Table 1. (continued)

Cancer type

Aim of study

Source of organoids

Digestion condition

Culture condition

Refs

Pancreatic

PDAC modeling and drug screening

PDAC tumor specimens

Collagenase + PTOM

Matrigel + pancreatic progenitor and tumor organoid medium (PTOM), DMEM, B27, ascorbic acid, insulin, hydrocortisone, FGF2, alltrans retinoic acid, and Y267632 small molecule

[42]

Pancreatic

Organoid model

PDAC tumor specimens

Collagenase II + human complete medium. TrypLE with human CM

Growth factor-reduced Matrigel + human complete medium: ADMEM/F12 medium supplemented with HEPES, GlutaMAX, penicillin/ streptomycin, B27, Primocin, N-acetyl-l-cysteine, Wnt3aconditioned medium, Rspondin-1-conditioned medium, Noggin-conditioned medium or recombinant protein, EGF, gastrin, FGF10, nicotinamide, and A83-01

[43]

Stomach

Genomic-based classiﬁcation of gastric cancer, new driver mutation detection, personalized medicine

Mouse gastric tumor

PBS + EDTA

Growth factor-reduced Matrigel, ADMEM/F12 supplemented with EGF, Rspondin-1, Noggin, Y-27632

[30,40]

Stomach Esophageal

Long-term 3D cultures of human gastric stem cells and bacterial infection study

Human stomach and esophageal tumors

EDTA + cold chelation buffer

Matrigel, ADMEM/F12 supplemented with penicillin/ streptomycin, HEPES, GlutaMAX, B27, Nacetylcysteine. Gastric medium: Basal medium supplemented with EGF, Noggin-conditioned medium, R-spondin-1conditioned medium, Wntconditioned medium, FGF10, gastrin, A-83-01 (TGF-beta inhibitor). Facultative component: Nicotinamide. Additional components: RHOKi (Y27632), IGF, p38 inhibitor (SB202190), GSK3b inhibitor PGE2

[33]

Colon Endometrial Uterine Urothelial Ovary

Personalized tumor modeling

Human patient tumor specimen and PDX

Collagenase IV and trypsin– EDTA

Growth factor-reduced Matrigel, ADMEM with GlutaMAX, B27, penicillin/ streptomycin, Primocin, and tumor type-speciﬁc growth factors

[44]

Abbreviations: ALK, anaplastic lymphoma kinase; DHT, dihydrotestosterone; DMEM/F-12, Dulbecco’s modified Eagle medium/nutrient mixture F-12; ECM, extracellular matrix; EGF, epidermal growth factor; EGM, endothelial cell growth medium; FGF10, fibroblast growth factor-10; FGF2, fibroblast growth factor-2; GSK3b, glycogen synthase kinase 3 beta; HEPES, 4-(2-hydroxyethyl)-1-piperazine-ethane-sulfonic acid; HISC, human intestinal stem cell medium; IGF, insulin growth factor; MAPKs, mitogen-activated protein kinases; MEM, minimum essential medium; PBS, phosphate-buffered saline; PDAC, pancreatic ductal adenocarcinoma; PGE2, prostaglandin 2; PTOM, pancreatic progenitor and tumor organoid medium; RHOKi, rho-associated protein kinase inhibitor; TGFb, transforming growth factor beta.

Trends in Biotechnology, April 2018, Vol. 36, No. 4

363

these key findings suggest that this new model has great potential in personalized cancer therapy, specifically for gene–drug correlation studies, preclinical drug screening of anticancer drugs, and prediction of drug response and patient outcome (Figure 1, Key Figure). For example, Weeber and colleagues [22] developed a tumor organoid culture condition and sequenced 1977 cancer-related genes across 14 colon cancer organoids and corresponding original tumors. They reported 90% preservation of somatic mutations and DNA copy number profile between the developed tumor organoids and biopsies, which resulted in the successful application of organoids in genomic-based personalized medicine [22]. Until now, the most comprehensive and impressive example of the application of tumor organoids in personalized cancer therapy was reported by van de Wetering and colleagues [26], who established an organoid biobank of 20 colon tumors. Whole-exome sequencing analysis revealed that colon tumor organoids preserved the cancer subtypes detected in the tumor samples. In addition, a comparison of somatic copy number alterations in tumor biopsies and related organoids with The Cancer Genome Atlas (TCGA) database showed that genomic alterations found in hypermutated and non-hypermutated clinical samples were represented in tumor organoids [26]. Interestingly, an investigation of drug sensitivities of tumor-derived organoids against the library of 85 therapeutic compounds including chemotherapy and targeted therapy agents resulted in the identification of an effective treatment for each individual patient [45]. This study demonstrated that colon tumor cells that contained TP53 loss of function exhibited resistance to MDM2 inhibitors, and RAS mutants were insensitive to the EGFR inhibitor. In addition to these findings, they suggested a novel treatment approach for the RNF43 mutant colon cancer in which colon tumor organoids with the RNF43 mutation were significantly sensitive to Wnt secretion inhibitors. Gene–drug association plays a key role in personalized cancer therapy and targeted therapy. The study by van de Wetering and colleagues [26] described a strong correlation between gene mutation status and therapeutic response, known as mutation-based drug sensitivity. Because genomics analysis is insufficient to identify effective therapeutic options for the majority of patients with advanced cancer, drug screening on tumor organoids could illuminate the unknowns of the drug response [27]. To determine the clinical application of tumor organoids in personalized cancer therapy, numerous clinical trials on colon and pancreatic tumor organoids have been conducted (NCT03140592). Modeling specific and rare subtypes of cancer by means of genetically engineered organoids can help to identify an effective therapy for a small proportion of patients [24,47,58–60]. In a comparison study between tumor organoids and its counterpart PDX models, 56 tumorderived organoid and 19 PDX models were generated from 769 patients with various cancer types. The matched tumor organoids and generated PDX models showed similar histopathological features to their original tumors, which was further validated via whole-exome sequencing in both models [27]. Genomic analysis of organoids in different passages revealed that the tumor organoids preserved genomic alterations of the original tumor during long-term culture. Interestingly, genome sequencing of large numbers of tumor samples revealed that in 85.8% of cases, somatic alterations in cancer genes were not targetable, whereas 9.6% of cases could be targeted by off-label drugs, and only 0.4% of detected somatic alterations could be targeted by [35_TD$IF]FDA-approved drugs. This observation highlighted the potential for reliable tumor models to identify novel treatment options [27]. The results from screening of a library of 160 drugs, including FDA-approved chemotherapeutics and targeted agents, showed similar drug responses between tumor organoids and PDX models.

364

Trends in Biotechnology, April 2018, Vol. 36, No. 4

Key Figure

Schema of Organoid-Based Personalized Cancer Therapy

130

120

G A T A A A T C T G G T C T T AT T T C C

SSequencing eq qu uencing

Orga Organoid

Gene–drug associaƟon and drug selecƟon

Biopsy resected tumor

Sequencing report

Organoid

Cancer paƟent

Drug screening Tumor organoid culture Applying appropriate therapy

Organoid biobank

Figure 1. In this approach, the procedure begins with sequencing tumor biopsies or dissected samples by using the next-generation sequencing method and continues with culturing patient-derived tumor organoids, which will be histologically and pathologically compared with the primary tumors before they are subjected to (Figure legend continued at the bottom of the next page.) Trends in Biotechnology, April 2018, Vol. 36, No. 4

365

For instance, in endometrial adenocarcinoma, optimal treatment for both organoid and PDX models included the combination of buparlisib and olaparib. In addition, afatinib and other EGFR inhibitors were recorded to be the most effective drugs for advanced stage colon cancer with a mutation in the APC gene. This result was validated in both tumor models [27]. Interestingly, combination therapy screening on tumor organoids suggested that compared with the FOLFOX regimen, a combination of afatinib and histone deacetylase (HDAC) inhibitors led to greater inhibition of growth in colon tumors with the APC mutation [27]. Targeting mutant RAS in patient-derived colorectal cancer organoids identified a relationship between KRASG12D[35_TD$IF] and insensitivity to the combination of MEK and pan-HER inhibitors [25]. The use of a BCl-2 inhibitor in colon organoids overcame resistance to combined therapy of MEK and pan-HER. These results were further confirmed in a PDX model. Therefore, these results suggest that integrating genome data and preclinical drug screening results could improve precision cancer treatment. Presently, the majority of patients are candidates for second-line therapy with different categories of drugs compared with those used as the initial course of therapy. Organoids derived from individual patients screened with first-line therapy could be tested for second-line therapy to ascertain the best possible treatment option. In support of this idea, Skardal and colleagues [61] used liver tumor hybrid organoids as a screening panel for testing different compounds similar to second line of therapy in liver cancer. Their promising results indicated that organoids were suitable for therapeutic testing in vitro [61]. Gao and colleagues [21], in a study of prostate cancer organoids, observed a strong correlation between different therapeutic responses and the genomic profile of individual cancer. Comparing in vitro drug screening results and the patient drug response is another potential application of tumor organoid models. In this context, a large multicenter cohort study on metastatic breast, colon, and nonsmall cell lung cancers (NSCLCs) was conducted by the Foundation Hubrecht Organoid Technology (HUB; Utrecht, The Netherlands, TUMOROID trial: NL49002.031.14). In this study, drug responses in tumor organoids derived from biopsy of the metastatic lesion showed a positive correlation with clinical responses of patients [23]. In addition, the correlation of matched clinical outcome with tumor organoid drug responses has established an in vitro threshold for drug response. However, differences between drug dosages in chemotherapy regimens and those used in in vitro drug screening assays are challenging issues that remain to be solved. The selection of patients for appropriate and effective treatment regimens is another main focus of personalized cancer therapy. Patient-derived tumor organoids show great potential to select patients for specific targeted therapy. In this regard, an ongoing prospective clinical study conducted by [17_TD$IF]The Netherlands Cancer Institute (SENSOR study, NL50400.031.14 EudraCT 2014-003811-13) selects colon and NSCLC patients for targeted treatment by means of screening of a panel of targeted agents on patient-derived tumor organoids. In this study, patients with metastatic colorectal cancer and NSCLC with only one standard line of treatment remaining were considered. Tumor organoids of these patient’s tumors were screened against eight different targeted therapy agents. After identification of the most active agent with the drug screening. In parallel, part of the derived organoids will be preserved as a biobank. To determine effective therapeutic strategies, based on the sequencing results and gene–drug association links, high-throughput drug screening of candidate drugs that include standard chemotherapy and targeted therapy agents can be performed in a replicative process.

366

Trends in Biotechnology, April 2018, Vol. 36, No. 4

highest inhibitory effect, the study offered patients the choice to continue therapy with the identiﬁed agent.

Tumor Organoid Challenges and Possible Solutions to Overcome Them Despite the advantages of tumor organoids in personalized cancer therapy, numerous challenges may hamper the implementation of this approach in a clinical setting. In some cancers, the majority of patients are candidates for neoadjuvant chemotherapy regimens to shrink tumors prior to surgery. These treatment regimens may result in decreased numbers of tumor cells from the biopsy area and possibly preclude organoid generation in preclinical settings. Therefore, optimizing current culture methods to generate tumor organoids from these types of biopsies will promote organoid culture for a wide variety of patients. Challenges include the lack of immune system elements, other key stromal cells, and vasculature factors in tumor organoid cultures; these can restrict functional testing of immunotherapy drugs and/or stromal targeted agents on tumor organoids. However, to overcome this challenge, other technologies such as microfluidics technology (Box 3) may facilitate coculture of tumor organoids with other cell types to model a complex tumor microenvironment [62]. To date, a wide variety of microfluidic devices have been constructed to assess and model the liver, kidneys, lungs, intestines, heart, smooth muscles, bones, blood vessels, and blood–brain barrier in a microarchitecture setting [63]. These microfluidic analogs have the ability to connect with each other to mimic in vivo physiological coupling (called a human-on-a-chip model; Box 3) and to replace animal models in pharmaceutical studies [63]. This has ignited great enthusiasm for the potential of this technology to model immune–tumor cell interactions. Historically, studies on interactions between immune cells and cancer cells in monolayers were critical for discovering tumor-associated antigens [64,65]. Later, coculturing immune cells and multicellular tumor spheroids illustrated new dimensions of tumor immunology [69]. In contrast to 2D cancer–immune interaction, in vitro 3D studies not only demonstrated decreased production of tumor-associated antigens in 3D models, but also highlighted the importance of using 3D tumor models in tumor immunology studies, which can be easily captured in microfluidic systems [64]. However, few studies have been conducted using microfluidic systems to investigate immune–cancer cell interactions [64,70–73]. Agliari and colleagues [73] reported a coculture system with microfluidic devices that investigated the migration of immune cells toward tumor cells by exploiting splenocytes deficient in the transcription factor interferon regulatory factor 8. Recently, Liu and colleagues [74] conducted a study that employed precision medicine for bladder cancer by coculturing different Matrigel-embedded cell types including human bladder cancer cells (T24), fibroblasts (BJ-5Ta), macrophages (Raw 264.7), and human umbilical vein endothelial cells in a microfluidic device. They observed migration of macrophages toward Box 3. Introduction of Microfluidic and Organ-On-Chip Technologies Microfluidics is the science of manipulating fluids in submillimeter channels. This method can control several parameters, such as relative cell and tissue location, fluid flow levels and patterns, mechanical cues, and gradients within a system [63]. The advantages of this technology include (i) smaller sample and material requirements, (ii) enhanced quality of microscopic imaging and quantification of cells, (iii) control over experiments, (iv) low cost of production, and (v) the possibility of a targeted design. These benefits make microfluidic technology a complementary tool for other platforms. A complex tumor microenvironment can be created with the support of this technology and by determining different cell regions [62]. Microfluidic technology can be used to model the metastatic microenvironment, immune–cancer cell interactions, and specific behavior of cancer cells (Figure IA) [64–66]. Organs-on-chips are microfluidic devices that allow the culture of living cells to mimic physiological and functional properties of an organ at micrometer diameters [63]. The human-on-a-chip is an improved method that simulates intracellular relevance and organ interactions, thus providing the possibility of in vitro testing of pharmacodynamics and toxicodynamics of drugs (Figure IB) [67].

Trends in Biotechnology, April 2018, Vol. 36, No. 4

367

(A)

MigraƟng IFN-DCs

Infiltrated IFN-DCs

Immune chamber

Tumor chamber ConnecƟng channels

(B)

Drugs or other compounds can be introduced into the system for tesƟng

Output analysis: cytokines, chemokines, exosomes, etc.

Figure I. Application of Microfluidic Technology in Cancer–Immune Interaction and Organ-On-Chip Concepts. (A) Schematic presentation of microfluidic chip section that shows the migration of interferon-a-conditioned dendritic cells (IFN-DCs) toward the tumor environment by crossing the connecting channels. Reproduced, with permission, from [68]. (B) Graphical presentation of the human-on-chip concept. Microfluidic technology enables the culture of different cells from different organs to approximate a close humanized in vitro model.

368

Trends in Biotechnology, April 2018, Vol. 36, No. 4

Tumor organoid inlet channel Inlet chamber

Inlet chamber

Standard of care drugs: • Chemotherapy drugs • Targeted drugs • Immunotherapy agents • Cell based-immunotherapy: CAR-T cells, DC vaccine, NK cells, etc.

NK cells T cells Treg cells B cells

Outlet chamber

DendriƟc cells Extracellular matrix gel embedding

Figure 2. Schematic of Cocultured Tumor-Derived Organoids, with Immune System Elements Derived from the Peripheral Blood of Patients, in a Microﬂuidic Device. DC, dendritic cell; NK cells, natural killer cells; Treg, T regulatory cells.

cancer cells and screened different chemotherapeutics similar to a neoadjuvant schema in a clinical setting. Their results showed that macrophages in 3D cocultures expressed and released more Arg-1 in the tumor microenvironment, which resulted in alterations in the chemotherapy response [74]. Taken together, in contrast to PDX models, the interaction between tumor organoids and immune cells in microﬂuidic devices may not only overcome the challenge of screening current immunotherapy drugs, but also can predict the effects of new generation of cancer immunotherapy agents including CAR-T cells, engineered natural killer cells, and dendritic cells on tumor organoids in the preclinical setting (Figure 2).

Concluding Remarks and Future Perspectives Personalized cancer medicine is devoted to tailoring the most appropriate drug for an individual patient. To achieve this goal, genomic-based drug response prediction has opened new ways to enable better decision making in oncology. Evaluating the functionality of these predictions is difﬁcult because of the lack of representative patient tumor models that can recapitulate all of the key features of original tumors. Patient-derived tumor organoids show great potential for predicting clinical responses to drugs and for selecting treatments. Although numerous questions need to be addressed (see Outstanding Questions), current clinical studies will improve our understanding of the translational application of tumor organoids to the patient. Here, we have highlighted current studies that assessed the potential application of tumor organoids in precision oncology, in particular, drug response prediction, patient selection, and determination of a novel therapeutic agent. Given that tumor organoids can be generated from

Trends in Biotechnology, April 2018, Vol. 36, No. 4

369

individual cancer patients with close cellular and molecular resemblance to the parental tumor, we believe that tumor organoids will be better, more promising, and more clinically relevant tumor models for personalized cancer therapy compared with pre-existing models. We expect that the use of other technologies, including microﬂuidics, may overcome the challenge of modeling the immunotherapy response on tumor organoids. In this context, further studies should be designed to coculture tumor organoid and immune system elements derived from individual patients in the microﬂuidic system and screen immunotherapy drugs.

Outstanding Questions

[356_TD$IF]Acknowledgments

Can this model be useful for patients who do not have recommended standard treatments?

This study was funded by grants awarded by Royan Institute, the Iran National Science Foundation (INSF, Grant No. 96001316), and Iran Science Elites Federation to H.B.

[357_TD$IF]Disclaimer Statement The authors declare that they have no conﬂicts of interest in this article.

References 1. Friedman, A.A. et al. (2015) Precision medicine for cancer with next-generation functional diagnostics. Nat. Rev. Cancer 15, 747–756

17. Marangoni, E. et al. (2007) A new model of patient tumor-derived breast cancer xenografts for preclinical assays. Clin. Cancer Res. 13, 3989–3998

2. Voest, E.E. and Bernards, R. (2016) DNA-guided precision medicine for cancer: a case of irrational exuberance? Cancer Discov. 6, 130–132

18. Klemm, F. and Joyce, J.A. (2015) Microenvironmental regulation of therapeutic response in cancer. Trends Cell Biol. 25, 198–213

3. Goodspeed, A. et al. (2015) Tumor-derived cell lines as molecular models of cancer pharmacogenomics. Mol. Cancer Res. 14, 3–13 4. McMillin, D.W. et al. (2013) The role of tumour-stromal interactions in modifying drug response: challenges and opportunities. Nat. Rev. Drug Discov. 12, 217–228 5. Wilding, J.L. and Bodmer, W.F. (2014) Cancer cell lines for drug discovery and development. Cancer Res. 74, 2377–2384

19. Reyal, F. et al. (2012) Molecular proﬁling of patient-derived breast cancer xenografts. Breast Cancer Res. 14, R11 20. Lancaster, M.A. and Knoblich, J.A. (2014) Organogenesis in a dish: modeling development and disease using organoid technologies. Science 345, 1247125 21. Gao, D. et al. (2014) Organoid cultures derived from patients with advanced prostate cancer. Cell 159, 176–187

6. Weiswald, L.B. et al. (2015) Spherical cancer models in tumor biology. Neoplasia 17, 1–15

22. Weeber, F. et al. (2015) Preserved genetic diversity in organoids cultured from biopsies of human colorectal cancer metastases. Proc. Natl. Acad. Sci. U. S. A. 112, 13308–13311

7. Bartlett, R. et al. (2014) Personalized in vitro cancer modeling — fantasy or reality? Transl. Oncol. 7, 657–664

23. Weeber, F. et al. (2017) Tumor organoids as a pre-clinical cancer model for drug discovery. Cell Chem. Biol. 24, 1092–1100

8. Stein, W.D. et al. (2004) A Serial Analysis of Gene Expression (SAGE) Database analysis of chemosensitivity: comparing solid tumors with cell lines and comparing solid tumors from different tissue origins. Cancer Res. 64, 2805–2816

24. Clevers, H. and Bender, E. (2015) Q&A: Hans Clevers. Banking on organoids. Nature 521, S15

9. Tentler, J.J. et al. (2012) Patient-derived tumour xenografts as models for oncology drug development. Nat. Rev. Clin. Oncol. 9, 338–350 10. Hidalgo, M. et al. (2014) Patient-derived xenograft models: an emerging platform for translational cancer research. Cancer Discov. 4, 998–1013 11. Leto, S.M. et al. (2015) Sustained inhibition of HER3 and EGFR is necessary to induce regression of HER2-amplified gastrointestinal carcinomas. Clin. Cancer Res. 21, 5519–5531 12. Bertotti, A. et al. (2011) A molecularly annotated platform of patient-derived xenografts (“xenopatients”) identifies HER2 as an effective therapeutic target in cetuximab-resistant colorectal cancer. Cancer Discov. 1, 508–523 13. Hidalgo, M. et al. (2011) A pilot clinical study of treatment guided by personalized tumorgrafts in patients with advanced cancer. Mol. Cancer Ther. 10, 1311–1316 14. Zhang, X. et al. (2013) A renewable tissue resource of phenotypically stable, biologically and ethnically diverse, patient-derived human breast cancer xenograft models. Cancer Res. 73, 4885–4897 15. Tetteh, P.W. et al. (2016) Generation of an inducible colon-specific Cre enzyme mouse line for colon cancer research. Proc. Natl. Acad. Sci. U. S. A. 113, 11859–11864 16. DeRose, Y.S. et al. (2011) Tumor grafts derived from women with breast cancer authentically reflect tumor pathology, growth, metastasis and disease outcomes. Nat. Med. 17, 1514–1520

370

Trends in Biotechnology, April 2018, Vol. 36, No. 4

25. Verissimo, C.S. et al. (2016) Targeting mutant RAS in patientderived colorectal cancer organoids by combinatorial drug screening. Elife 5, e18489 26. van de Wetering, M. et al. (2015) Prospective derivation of a living organoid biobank of colorectal cancer patients. Cell 161, 933– 945 27. Pauli, C. et al. (2017) Personalized in vitro and in vivo cancer models to guide precision medicine. Cancer Discov. 7, 462–477 28. Huch, M. and Koo, B. (2015) Modeling mouse and human development using organoid cultures. Development 142, 3113–3125 29. Sachs, N. and Clevers, H. (2014) Organoid cultures for the analysis of cancer phenotypes. Curr. Opin. Genet. Dev. 24, 68–73 30. Sato, T. et al. (2011) Long-term expansion of epithelial organoids from human colon, adenoma, adenocarcinoma, and Barrett’s epithelium. Gastroenterology 141, 1762–1772 31. Sato, T. et al. (2009) Single Lgr5 stem cells build crypt-villus structures in vitro without a mesenchymal niche. Nature 459, 262–265 32. Barker, N. et al. (2010) Lgr5+ve stem cells drive self-renewal in the stomach and build long-lived gastric units in vitro. Cell Stem Cell 6, 25–36 33. Bartfeld, S. et al. (2015) In vitro expansion of human gastric epithelial stem cells and their responses to bacterial infection. Gastroenterology 148, 126–136.e6 34. Katano, T. et al. (2013) Establishment of a long-term threedimensional primary culture of mouse glandular stomach

Can patient-derived tumor organoid models be a useful approach for clinical management of cancer patients? Can tumor organoids predict the response of chemotherapy regimens and patient outcome?

Can the combination of a microﬂuidic culture system and tumor organoid model become clinical tools for testing the new generation of immunotherapy drugs and personalized immunotherapy of patients?

epithelial cells within the stem cell niche. Biochem. Biophys. Res. Commun. 432, 558–563 35. Huch, M. et al. (2013) In vitro expansion of single Lgr5+ liver stem cells induced by Wnt-driven regeneration. Nature 494, 247–250 36. Huch, M. et al. (2015) Long-term culture of genome-stable bipotent stem cells from adult human liver. Cell 160, 299–312 37. Petersen, O.W. et al. (1992) Interaction with basement membrane serves to rapidly distinguish growth and differentiation pattern of normal and malignant human breast epithelial cells. Proc. Natl. Acad. Sci. U. S. A. 89, 9064–9068 38. Schwarz, J.S. et al. (2015) Value of organoids from comparative epithelia models. Yale J. Biol. Med. 88, 367–374

56. Cheung, K.J. et al. (2013) Collective invasion in breast cancer requires a conserved basal epithelial program. Cell 155, 1639– 1651 57. Walsh, A.J. et al. (2016) Drug response in organoids generated from frozen primary tumor tissues. Sci. Rep. 6, 18889 58. Matano, M. et al. (2015) Modeling colorectal cancer using CRISPR-Cas9–mediated engineering of human intestinal organoids. Nat. Med. 21, 256–262 59. Schwank, G. et al. (2013) Functional repair of CFTR by CRISPR/ Cas9 in intestinal stem cell organoids of cystic ﬁbrosis patients. Cell Stem Cell 13, 653–658

39. Drost, J. et al. (2016) Organoid culture systems for prostate epithelial and cancer tissue. Nat. Protoc. 11, 347–358

60. Schwank, G. et al. (2013) Brief report functional repair of CFTR by CRISPR/Cas9 in intestinal stem cell organoids of cystic ﬁbrosis patients. Stem Cell 13, 653–658

40. Wang, K. et al. (2014) Whole-genome sequencing and comprehensive molecular proﬁling identify new driver mutations in gastric cancer. Nat. Genet. 46, 573–582

61. Skardal, A. et al. (2015) Liver-tumor hybrid organoids for modeling tumor growth and drug response in vitro. Ann. Biomed. Eng. 43, 2361–2373

41. Sachs, N. et al. (2017) A living biobank of breast cancer organoids captures disease heterogeneity. Cell Published online December 5, 2017. http://dx.doi.org/10.1016/j.cell.2017.11.010

62. Yin, X. et al. (2016) Engineering stem cell organoids. Cell Stem Cell 18, 25–38

42. Huang, L. et al. (2015) Ductal pancreatic cancer modeling and drug screening using human pluripotent stem cell- and patientderived tumor organoids. Nat. Med. 21, 1364–1371

63. Bhatia, S.N. and Ingber, D.E. (2014) Microﬂuidic organs-onchips. Nat. Biotechnol. 32, 760–772 64. Boussommier-Calleja, A. et al. (2016) Microﬂuidics: a new tool for modeling cancer–immune interactions. Trends Cancer 2, 6–19

43. Boj, S.F. et al. (2015) Organoid models of human and mouse ductal pancreatic cancer. Cell 160, 324–338

65. Hirt, C. et al. (2014) “In vitro” 3D models of tumor-immune system interaction. Adv. Drug Deliv. Rev. 79, 145–154

44. Batchelder, C.A. et al. (2015) Three dimensional culture of human renal cell carcinoma organoids. PLoS One 10, e0136758

66. Au, S.H. et al. (2016) Clusters of circulating tumor cells traverse capillary-sized vessels. Proc. Natl. Acad. Sci. U. S. A. 113, 4947– 4952

45. Xinaris, C. et al. (2015) Organoid models and applications in biomedical research. Nephron 130, 191–199 46. Dekkers, J.F. et al. (2016) Characterizing responses to CFTRmodulating drugs using rectal organoids derived from subjects with cystic ﬁbrosis. Sci. Transl. Med. 8, 13 47. Dekkers, F. et al. (2012) A functional CFTR assay using primary cystic ﬁbrosis organoids. Pediatr. Pulmonol. 47, 289 48. Freedman, B.S. et al. (2015) Modelling kidney disease with CRISPR-mutant kidney organoids derived from human pluripotent epiblast spheroids. Nat. Commun. 6, 8715 49. Astashkina, A. and Grainger, D.W. (2014) Critical analysis of 3-D organoid in vitro cell culture models for high-throughput drug candidate toxicity assessments. Adv. Drug Deliv. Rev. 69–70 1–18 50. Roper, J. et al. (2017) In vivo genome editing and organoid transplantation models of colorectal cancer and metastasis. Nat. Biotechnol. 35, 569–576

67. Cirit, M. et al. (2016) Human physiome on a chip: merging tissue engineering and systems pharmacology. FASEB J. 30, 938.1 68. Parlato, S. et al. (2017) 3D microfluidic model for evaluating immunotherapy efficacy by tracking dendritic cell behaviour toward tumor cells. Sci. Rep. 7, 1093 69. Feder-Mengus, C. et al. (2008) New dimensions in tumor immunology: what does 3D culture reveal? Trends Mol. Med. 14, 333– 340 70. Hsu, T.-H. et al. (2012) The migration speed of cancer cells influenced by macrophages and myofibroblasts co-cultured in a microfluidic chip. Integr. Biol. (Camb.) 4, 177–182 71. Businaro, L. et al. (2013) Cross talk between cancer and immune cells: exploring complex dynamics in a microfluidic environment. Lab Chip 13, 229–239

51. Gao, D. and Chen, Y. (2015) Organoid development in cancer genome discovery. Curr. Opin. Genet. Dev. 30, 42–48

72. Mattei, F. et al. (2014) A multidisciplinary study using in vivo tumor models and microﬂuidic cell-on-chip approach to explore the cross-talk between cancer and immune cells. J. Immunotoxicol. 11, 337–346

52. Baker, L.A. et al. (2016) Modeling pancreatic cancer with organoids. Trends Cancer 2, 176–190

73. Agliari, E. et al. (2014) Cancer-driven dynamics of immune cells in a microﬂuidic environment. Sci. Rep. 4, 6639

53. Yachida, S. et al. (2010) Distant metastasis occurs late during the genetic evolution of pancreatic cancer. Nature 467, 1114–1117

74. Liu, P. et al. (2015) A bladder cancer microenvironment simulation system based on a microﬂuidic co-culture model. Oncotarget 6, 37695–37705

54. Drost, J. et al. (2017) Use of CRISPR-modiﬁed human stem cell organoids to study the origin of mutational signatures in cancer. Science 358, 234–238 55. Karthaus, W.R. et al. (2014) Identiﬁcation of multipotent luminal progenitor cells in human prostate organoid cultures. Cell 159, 163–175

Trends in Biotechnology, April 2018, Vol. 36, No. 4

371

Therefore, I see a mix of approaches and ways of studying the cell in the future, with investigator-initiated research being in the majority. The need for other approaches, however, is becoming increasingly important. While investigator-driven research has evolved to work well to address speciﬁc hypotheses and problems, the evolution of large-scale team science is still a relatively new frontier and successful models for its effective implementation are still being developed.

Resources i

www.genome.gov/10001772/all-about-the–human-

genome-project-hgp/ 1

Allen Institute for Cell Science, Seattle, WA, USA

*Correspondence: rickh@alleninstitute.org (R. Horwitz). http://dx.doi.org/10.1016/j.tcb.2016.07.007

References 1. Watson, J.D. (1965) Molecular Biology of the Gene, Benjamin 2. Alberts, B. et al. (1983) Molecular Biology of the Cell, Garland Science 3. Liu, Z. et al. (2015) Imaging live-cell dynamics and structure at the single-molecule level. Mol. Cell. 58, 644–659 4. Asano, S. et al. (2016) In situ cryo-electron tomography: a post-reductionist approach to structural biology. J. Mol. Biol. 428, 332–343 5. Nelson, C.M. and Bissell, M.J. (2006) Of extracellular matrix, scaffolds, and signaling: tissue architecture regulates development, homeostasis, and cancer. Annu. Rev. Cell. Dev. Biol. 22, 287–309 6. Engler, A.J. et al. (2009) Multiscale modeling of form and function. Science 324, 208–212 7. Ideker, T. and Lauffenburger, D. (2003) Building with a scaffold: emerging strategies for high- to low-level cellular modeling. Trends Biotechnol. 21, 255–262 8. Doudna, J.A. and Charpentier, E. (2014) Genome editing. The new frontier of genome engineering with CRISPRCas9. Science 346, 1258096 9. Digman, M.A. and Gratton, E. (2012) Scanning image correlation spectroscopy. Bioessays 34, 377–385 10. Caprioli, R.M. (2016) Imaging mass spectrometry: molecular microscopy for the new age of biology and medicine. Proteomics 16, 1607–1612 11. Lubeck, E. and Cai, L. (2012) Single-cell systems biology by super-resolution imaging and combinatorial labeling. Nat. Methods 9, 743–748 12. Trapnell, C. (2015) Deﬁning cell types and states with single-cell genomics. Genome Res. 25, 1491–1498 13. Coulon, A. et al. (2013) Eukaryotic transcriptional dynamics: from single molecules to cell populations. Nat. Rev. Genet. 14, 572–584 14. Tiscornia, G. et al. (2011) Diseases in a dish: modeling human genetic disorders using induced pluripotent cells. Nat. Med. 17, 1570–1576 15. Clevers, H. (2016) Modeling development and disease with organoids. Cell 165, 1586–1597

[(1268)TD.ENIM]7SAR

798

Special Issue: Future of Cell Biology

Forum 3D Biomimetic Cultures: The Next Platform for Cell Biology Christopher S. Chen1,2,* Advances in engineering of cells and culture formats have led to the development of a new generation of 3D cultures that can recapitulate a variety of multicell-type, morphogenetic behaviors that were previously largely observable only in in vivo settings. Ultimately, these systems are likely to be assimilated into and forever change the landscape of biomedical research. As with all sciences, biology advances through our ability to experiment, in this case with living systems. Modern biomedical research essentially relies on two experimental test beds: animals and cultured cells. The knowledge revolution of the past half century that we know as cell biology largely rests on the dissemination of cultured cells[3_TD$IF] –[4_TD$IF] HeLa cells ﬁrst, then other immortalized lines, and now a wide array of primary cells and stem-like cells[3_TD$IF] –[5_TD$IF] as accepted experimental systems to understand cell structure and function. As an apt adjunct to animal systems, which capture the full complexity of biology but with limited ability to quickly isolate detailed mechanisms, experimental manipulation of cells in culture is straightforward and has revolutionized our molecular understanding of cells. However, despite amazing advances our ability to translate cell biological insights has been mitigated because culture on plastic dishes is so different from the in vivo microenvironment. Cells not only change their behavior in this

Trends in Cell Biology, November 2016, Vol. 26, No. 11

non-physiologic environment but are also unable to remodel the matrix and reorganize freely as they would during development and homeostasis. Thus, many key functions are lost or unobservable in cell culture. These limitations compel us to consider whether innovative platforms that allow us to examine cells cultured in more biomimetic contexts can be developed to bridge the gap between traditional cell culture and the whole organism and what impact would such systems[1_TD$IF] have on our biomedical research enterprise (Figure 1). In vivo, local tissue structure deﬁnes the cellular environment, constraining how cells interact with the surrounding extracellular matrix (ECM), neighboring cells, soluble growth factors, and physical forces. These ‘microenvironmental’ cues cooperate to regulate cell behavior. Thus, while it is no surprise that culture on plastic dishes results in decompensated cell signaling, gene expression, phenotype, and function, attempting to fully reconstruct a tissue environment for in vitro applications would be excessive. The real challenge is in identifying which factors to incorporate to appropriately model different in vivo processes in cell culture and then establishing what such systems would and would not be able to recapitulate. In recent years, several ex vivo experimental models have been developed to capture various higher-level behaviors that historically were largely reserved for animal models. Some of these models are methodologically ‘simple’, natural extensions of classical 3D cultures that have been used to generate mammary acini, hanging-drop embryoid bodies, or spheroid cultures, although with remarkable new morphogenetic capabilities. For example, single intestinal stem cells embedded within ECM gels have been shown to give rise to self-organizing structures characteristic of the crypt–villus of the intestine [1]. Similarly, spheroid cultures of neuronal stem cells have been developed to recapitulate the layering and morphogenesis of the developing

recapitulate. It is important to note that, unlike in vivo systems, these models are necessarily and intentionally simplifications to capture a narrow range of behavior, physiology, or time. For example, while intravital recordings of the developing vasculature of the avian ovum or zebrafish can capture vasculogenesis (when endothelial cells assemble to form networks spontaneously), angiogenesis (when existing vessels sprout and branch to form new vessels), or tumor cell trafficking, different biomimetic culture systems have been established to capture each of these events separately (for example, see [11– 13]). Thus, a key feature of these models in their current state of evolution is that they are best adopted when fit for a specific purpose, and expectations that such models would have universal applicability would be unrealistic.

Reduc onist models

Cell–cell

Isolate Validate V lid t

Soluble cues

Biomime c architecture

ic i m te M sla an Tr

In vivo

Matrix

Inform Recombine R bi

Forces

- S ﬀness - Topography - Porosity - Force

Organotypic models Figure 1. In Vivo and In Vitro Models Have Coevolved Synergistically to Provide Distinct Approaches to Understanding Living Systems. New biomimetic models offer the potential to provide a third approach to the ecosystem, reconstituting more complex behaviors in culture.

brain [2]. Other systems, by contrast, involve substantial engineering and incorporation of synthetic materials, prefabricated architectures, and/or microfluidics to model specific biological processes. For example, using a device containing two microfluidic channels separated by a porous elastic membrane Ingber and colleagues were able to model the interface between lung alveolar air, epithelium, capillary endothelium, and blood [3]. Using pumps to control air and blood flow and mechanical actuators to mimic the stretching forces of breathing on the epithelial–endothelial interface, the model has been used to recapitulate injury and inflammation and has inspired a cadre of organ-on-chip efforts from cardiac muscle to liver tissue [4,5]. Incorporation of human cells and human iPS-derived cell types into some of these systems has suggested the possibility that these biomimetic systems have the potential to close the gap between traditional animal models and human physiology and

disease [6,7]. DARPA, the NIH and NCATS, and the popular press have embraced the idea that these systems will ultimately replace preclinical testing of therapeutics in animals [8–10]. How can the research community come together to realize such high expectations, separate reality from hype, and ultimately benefit with a bevy of experimentally tractable systems that model human physiology and disease? Classically, biological experimental systems were used not as models to predict the behavior of other systems but as an end in themselves. Knowledge of anatomical structures in HeLa cells or wing formation in Drosophila was valued for its own sake. By contrast, the 3D biomimetic systems that are now being developed are explicitly valued for their ability to model specific processes, mostly in human biology. It stands to reason that a large part of establishing such models will be defining what the models can or cannot

Thus, key questions remain about how and when different models can or should be used. If minibrains can recapitulate some aspects of neuronal organization, will they show aberrations with known genetically caused brain malformations? Will they respond predictively to neurochemical modulators? Will they predict the neurological side effects of test compounds? Can they model aging? If the lung on chip can model inflammation, can it also recapitulate effects of cystic fibrosis? Will it respond similarly to biomechanical injury? If we take the lessons learned from cell culture and animal models, the key to answering these questions is not to wait for the group that first described these models to test all of these conditions. The only path to establishing these models, continually improving them, or deciding to abandon them is to make the models widely accessible to as many scientists as are willing to study them. There are many reasons why only a handful of cell lines and animal models dominated the research community, but perhaps the foremost were ease of adoption and the ability to share insights and advances among scientists. This poses a major challenge for many of these

Trends in Cell Biology, November 2016, Vol. 26, No. 11

799

role in our basic understanding of life's [(1278)TD.ENIM]SARSpecial Issue: Future of Cell design principles. Analogous to the in vitro Biology reconstitution of subcellular processes, the iterative effort that leads to the syn- Forum thetic reconstitution of multicell-type morphogenetic events will reveal the key components and subsystems necessary to generate such behaviors. Thus, one can only presume that these efforts will lead to a more complete understanding of 1 how cells organize and stabilize within Georg Kustatscher and 1,2, their surroundings and will at a minimum Juri Rappsilber * become a mainstay approach alongside standard reductionist and animal models Proteomic studies find many proDespite these hurdles, the eventual incor- to deepen our understanding of life. teins in unexpected cellular locaporation of these synthetic biomimetic cultions. Can functional components ture systems into biomedical research 1Biomedical Engineering and The Biological Design of organelles be distinguished from laboratories is inevitable. The confluence Center, Boston University, Boston, MA, USA biochemical artefacts or mis2 of technological advances in the engineer- The Wyss Institute for Biologically Inspired Engineering, guided cellular sorting? The clue Harvard University, Boston, MA 02215, USA ing and biological communities appears to might reside in compositional be a virtual perfect storm that will push us *Correspondence: chencs@bu.edu (C.S. Chen). changes that follow biological to continue establishing engineered 3D http://dx.doi.org/10.1016/j.tcb.2016.08.008 challenges and that can be organotypic cultures. On the biological References side, iPSC technologies and stem cell biol- 1. Sato, T. et al. (2009) Single Lgr5 stem cells build crypt–villus decoded by machine learning. structures in vitro without a mesenchymal niche. Nature ogy are coming together to advance 459, 262–265 The Fuzzy Cell access to human cell types and the appli2. Lancaster, M.A. et al. (2013) Cerebral organoids model Textbook views of cellular components, cation of genomic editing technologies human brain development and microcephaly. Nature 501, 373–379 from protein complexes to organelles, foloffers the possibility of both modeling 3. Huh, D. et al. (2010) Reconstituting organ-level lung func- low the paradigm ‘localization = function’. human genetic diseases and mechanistitions on a chip. Science 328, 1662–1668 cally implicating molecular players in these 4. Bhatia, S.N. and Ingber, D.E. (2014) Microfluidic organs- If a protein is found at a cellular location it on-chips. Nat. Biotechnol. 32, 760–772 also functions there. Consequently, the culture systems. On the engineering side, 5. Esch, E.W. et al. (2015) Organs-on-chips at the frontiers of focus of organelle proteomics has been a suite of technologies have been estabdrug discovery. Nat. Rev. Drug Discov. 14, 248–260 lished that can be used to build various 6. Hinson, J.T. et al. (2015) Titin mutations in iPS cells define to get the localization right. For decades sarcomere insufficiency as a cause of dilated cardiomyop- this was attempted by subcellular fractiontypes of system for organ-on-chip appliathy. Science 349, 982–986 cations, including the development of bio- 7. Wang, G. et al. (2014) Modeling the mitochondrial cardio- ation and by sorting out assumed contammyopathy of Barth syndrome with induced pluripotent inants. However, protein location may materials that can begin to mimic and stem cell and heart-on-chip technologies. Nat. Med. 20, have other reasons than function: cellular decouple aspects of the ECM, the appli616–623 cation of [6_TD$IF]microfabrication and nanofabri- 8. Sutherland, M.L. et al. (2013) The National Institutes of components possess an intrinsic, compoHealth Microphysiological Systems Program focuses on sitional ‘fuzziness’. cation tools such as microfluidics to a critical challenge in the drug discovery pipeline. Stem Cell support cell-based systems, advances Res. Ther. 4 (Suppl. 1), I1 of 3D printing and other technologies to 9. National Center for Advancing Translational Sciences. Tis- An often overlooked feature of subcellular sue Chip for Drug Screening. www.ncats.nih.gov/ organization is that it results from affinities organize cells in three dimensions, microstissuechip copy advances to observe living cells in 3D 10. Zhang, S. (2016) Chips that mimic organs could be more and equilibria, in other words is quantitative powerful than animal testing. Wired. Published online and not qualitative. Membranes act as barcontexts, and the use of insights gained by June 7, 2016. http://www.wired.com/2016/06/ riers but also need to be permeable. The tissue engineers to assemble cells and chips-mimic-organs-powerful-animal-testing/ nuclear envelope, for example, is perme11. Nguyen, D-H.T. et al. (2013) Biomimetic model to reconstiECM. The dire need for better models of tute angiogenic sprouting morphogenesis in vitro. Proc. able to proteins smaller than 40 kDa. human physiology and disease than either Natl Acad. Sci. U.S.A. 110, 6712–6717 traditional cell culture or animals also pro- 12. Moya, M.L. et al. (2013) In vitro perfused human capillary However, larger proteins might also make networks. Tissue Eng. Part C Methods 19, 730–737 an uncontrolled entry into the nucleus, vides a pull to advance these systems. 13. Zervantonakis, I.K. et al. (2012) Three-dimensional for example by having some affinity to Last, while ultimately these systems may microfluidic model for tumor cell intravasation and endothelial barrier function. Proc. Natl Acad. Sci. U.S.A. 109, the nuclear import machinery or at the become a primary platform for preclinical 13515–13520 end of mitosis, when the endoplasmic testing, their development will play a major

engineered organotypic models: they do not reproduce themselves; many of the systems are assembled as artisan pieces with many parameters that can affect the model so it can be difﬁcult to teach; many different biomimetic systems or variations would be expected to emerge to highlight different biological events and this customization inherently may limit wider adoption of each speciﬁc system; and it remains unclear which models scientists should congregate around versus leave under-investigated.

800

Trends in Cell Biology, November 2016, Vol. 26, No. 11

Compositional Dynamics: Deﬁning the Fuzzy Cell

Opinion

Modeling Cancer with Pluripotent Stem Cells Julian Gingold,1 Ruoji Zhou,2,3 Ihor R. Lemischka,4,5,6 and Dung-Fang Lee2,3,7,* The elucidation of cancer pathogenesis has been hindered by limited access to patient samples, tumor heterogeneity, and the lack of reliable model organisms. Characterized by their ability to self-renew indefinitely and differentiate into all adult cell lineages of an organism, pluripotent stem cells (PSCs), including ESCs and induced pluripotent stem cells (iPSCs), provide a powerful and unlimited source to generate differentiated cells that can be used to study disease biology, facilitate drug discovery and development, and provide key insights for developing personalized therapies. This article reviews the recent developments and technologies converting PSCs into clinically relevant model systems for cancer research. Modeling Disease with Pluripotent Stem Cells In 1998, Thomson and colleagues isolated human ESCs (see Glossary) from blastocysts and developed a defined culture system to maintain the cells in vitro [1], opening a new avenue for medical research. Later, in 2006–2007, a breakthrough by the laboratories of Yamanaka and Thomson heralded the development of a new kind of pluripotent cells – induced pluripotent stem cells (iPSCs) [2–4]. Both groups demonstrated that somatic cells (e.g., dermal fibroblasts and peripheral blood) could be reprogrammed to an ES-like cell state using a defined transcriptional factor cocktail (Yamanaka's OCT4, SOX2, KLF4, c-MYC; or Thomson's OCT4, SOX2, NANOG, LIN28) [5]. Over the past decade, subsequent advances facilitated the generation of iPSCs with chemicals, microRNA and modified RNA, or other gene delivery systems (retroviruses, adenoviruses, Sendai virus, transposons, and plasmids) [5]. Applications for iPSCs include regenerative medicine, disease modeling, drug screening, and personalized therapy. The unique combination of pluripotency and self-renewal distinguishes PSCs, including both ESCs and iPSCs, from all other cells (Figure 1A). The unlimited proliferative potential of these undifferentiated cells provides an arbitrarily large source of experimental material, while their pluripotency allows them to be coaxed into forming all adult tissue types. Well-defined protocols, including directed differentiation and organoid cultures have been developed to derive many major target tissues and cell types from PSCs of endodermal (liver, small intestine, stomach, thyroid, and lung), mesodermal (muscle, bone, cartilage, kidney, and blood), or ectodermal (epidermis, retinal, and cerebral tissue) lineages [6–8]. PSCs provide unparalleled advantages as a model system, allowing investigators to study a cell continuously from the moment it differentiates from a multipotent progenitor into a differentiated cell type of interest. The relevant genetic background for the model system can be introduced into PSCs using two primary strategies. In one approach, somatic cells from patients with genetic disorders are used to derive iPSC lines. These patient-derived iPSCs and their derivative differentiated tissues are then used to recapitulate a disease phenotype in vitro or shed light on

Trends in Cancer, September 2016, Vol. 2, No. 9

Trends Modeling cancer using pluripotent stem cells (PSCs) overcomes several disadvantages of current model systems, including limited accessibility of patient samples, tumor heterogeneity, and differences between species. PSCs provide a powerful and unlimited source to generate differentiated cells that can be used to elucidate disease pathogenesis, drug discovery and development, and advance personalized health care. Both patient-derived induced PSCs and engineered ESCs completely phenocopy cancer features, suggesting that PSCs can serve as a useful in vitro human cancer model. Currently evolving methodologies for gene expression manipulation, genome editing, and cell differentiation facilitate the application of PSCs to cancer research.

1 Women's Health Institute, Cleveland Clinic Foundation, Cleveland, OH 44195, USA 2 Department of Integrative Biology and Pharmacology, McGovern Medical School, The University of Texas Health Science Center at Houston, Houston, TX 77030, USA 3 The University of Texas Graduate School of Biomedical Sciences at Houston, Houston, TX 77030, USA 4 Department of Developmental and Regenerative Biology, Icahn School of Medicine at Mount Sinai, New York, NY 10029, USA 5 Department of Pharmacology and System Therapeutics, Icahn School of Medicine at Mount Sinai, New York, NY 10029, USA 6 The Black Family Stem Cell Institute, Icahn School of Medicine at Mount Sinai, New York, NY 10029, USA

http://dx.doi.org/10.1016/j.trecan.2016.07.007 Published by Elsevier Inc.

485

disease-relevant mechanisms [9]. This approach has been applied successfully to study the genetic causes of neurodegeneration [10–12], mental disorder [13], heart disease [14–17], and metabolic disorders [18]. Alternatively, a genetic disease trait can be directly introduced into PSCs. This approach is aided greatly by recent major developments in gene delivery systems such as helper-dependent adenoviral vectors [19], adeno-associated viruses [20], gene manipulation approaches (RNAi [21,22] and piggyBac transposases [23]), and genome-editing tools, such as zinc-ﬁnger nuclease [23–25], transcription activator-like effector nucleases (TALENs) [26,27], and clustered, regularly interspaced, short palindromic repeat/Cas9 (CRISPR/Cas9) [28,29]. These technologies allow introducing alterations (deletions, ampliﬁcations, mutations, or gene fusions) into ESCs or iPSCs of an arbitrary genetic background, allowing studying human monogenic and complex diseases as the pathology develops (Box 1).

7 Center for Stem Cell and Regenerative Medicine, The Brown Foundation Institute of Molecular Medicine for the Prevention of Human Diseases, The University of Texas Health Science Center at Houston, Houston, TX 77030, USA

*Correspondence: dung-fang.lee@uth.tmc.edu (D.-F. Lee).

While the ﬁeld of PSC-derived cancer research remains in its infancy, a number of PSC-derived cell lines have been generated to model disorders with a cancer predisposition (Table 1). Several groups have applied patient-derived iPSCs and/or engineered PSCs to phenocopy cancer features, explore disease mechanisms, and screen potential therapeutic drugs [30–34]. Their experience highlights the potential of human PSCs in cancer studies by overcoming limitations related to availability of patient samples or translation of results from animal models or cell lines with inappropriate genetic backgrounds. Here, we outline the existing PSC cancer models and their potential applications to understanding cancer biology. We discuss how recent developments (e.g., genome-editing and cell differentiation technologies) in PSCs have transformed our understanding of cancer biology and paved the way for new therapeutic strategies. Finally, we review some of the most promising model systems in which we anticipate this powerful technology will be applied.

(A)

(B) ERBB2 p53 SUFU

BRAF

Breast Ovary

ATM

BRCA2 CDH1

CHEK2

KRAS PTEN

Prostate

RB1

Eye cancer

ERBB2

FANC

Ectoderm Brain

CDKN2A

MUTYH

Brain tumor

APC

Thyroid

STK11

MLH1

Skin cancer

PMS2

Skin Endoderm

Liver

SERPINA1

BRAF

Pancreas

PSCs

RB1

IDH1 MYC Pancrea c cancer

TERT

Mesoderm

Thyroid cancer PTEN

Breast cancer

RPL5 Bone cancer

Lung

SMAD4

Gstric cancer Prostate cancer

VHL

IDH1

Car lage

Endoderm

EGFR

Ectoderm

Eye

BRCA1

CDKN2A

Ovarian cancer

PTPN11

WT1 Intes nal cancer

RPS19

Lung cancer Liver cancer

p53

Mesoderm Blood

Stomach

PML

DICER1

Kidney cancer

Car lage tumor

FBXW7 STK11

Bone

Kidney Intes ne

CHEK2

NOTCH1

DICER1

WT1

Blood cancer

KRAS

RET VHL

p53

FANC ABL1

Figure 1. Application of Pluripotent Stem Cells (PSCs) to Study Cancer-Associated Genetic Alterations. (A) PSCs are characterized by their capability to differentiate into all derivative cell types of the three germ layers. PSCs can form blood, kidney, bone, and cartilage cells via the mesoderm; ovary, breast, prostate, thyroid, liver, pancreas, lung, stomach, and intestine cells via the endoderm; and brain, eye, and skin cells via the ectoderm. (B) Loss of tumor suppressor genes, such as p53 mutation; or acquisition of oncogenes, such as ERBB2 ampliﬁcation or ABL1 translocation, results in both hereditary and sporadic cancers in ectodermal, mesodermal, and endodermal tissues.

486

Trends in Cancer, September 2016, Vol. 2, No. 9

Box 1. Tailoring the Pluripotent Stem Cell (PSC) Genome to Model Disease. Advanced genomeediting methodologies have now made it practical to tailor the human genome at the singlenucleotide level. Zinc-finger nucleases (ZFNs), transcription activator-like effector nucleases (TALENs), and clustered, regularly interspaced, short palindromic repeat/Cas9 (CRISPR/Cas9) have revolutionized the genome-editing field by allowing site-directed mutagenesis. A combination of a genomic localization domain with a domain conferring nuclease activity permits the introduction of a directed double-stranded break (DSB) at any site of interest. Site-directed mutagenesis without DSBs has recently become feasible for certain (cytosine to thymidine) ‘base editing’ using a CRISPR/Cas9 fused to cytidine deaminase [92]. By applying any of these approaches, it is now feasible to generate PSCs harboring specific mutations from a PSC line of choice or to correct any genomic alteration in patient-derived induced PSCs (iPSCs). Correcting Mutations in iPSCs Genome editing has been effectively applied to correct the genetic alterations in patient-derived iPSCs for both research and clinical applications. Correction of a mutation in a patient-derived iPSC can be used to generate an isogenic control and demonstrate a mechanistic link between a mutation and its downstream disease state. Importantly, it permits study from a genetic background in which disease penetrance has been established. It also may serve as a promising solution for transplantation-based therapies. For instance, ZFN–piggyBac-mediated correction of SERPINA1 mutations in /1antitrypsin-deficiency patients-derived iPSCs was shown to lead to the rescue of SERPINA1 structure and biological function [23]. TALEN-mediated gene-corrected X-linked severe combined immunodeficiency (SCID-X1) iPSCs demonstrated a rescue of defective hematopoietic differentiation [93]. CRISPR/Cas9–piggyBac-mediated correction of human hemoglobin beta (HBB) mutation in b-thalassemia patient-derived iPSCs restored the expression of HBB expression [94]. Similarly, positive–negative drug selection strategies combined with gene targeting on chromosome 21 have been demonstrated to correct trisomy in Down syndrome iPSCs [81]. While not yet ready for clinical practice, transplantation of differentiated tissue from these engineered iPSC lines back into their patient sources may permit a cure of the underlying disease, particularly in single-gene disorders that can be rescued by only a small amount of functional enzyme. Inducing Mutations in Wild-Type PSCs While iPSC models offer unique advantages for modeling a particular patient's genetic background, the diversity of their genetic origins complicates the integration and comparison of findings across multiple iPSC lines. By contrast, genomeedited PSCs based on extensively characterized lines are likely to be less variable and provide a more useful resource to understand central disease pathogenesis at the genomic scale. For instance, generation of a polycystic kidney disease (PKD)-linked PKD1 or PKD2 knockout in human ESCs results in cyst formation in kidney tubules in an organoid model, recapitulating the human phenotype [70]. Introduction of a long-QT syndrome (LQTS)-associated KCNH2 mutation in human ESCs leads to reduced current conducted by the HERG channel and prolonged action potential duration, recapitulating the LQTS phenotype [95]. In summary, early studies applying genomic editing to PSCs highlight the potential of this system for investigating disease pathogenesis and developing clinical cell therapies.

Modeling Cancer with Pluripotent Stem Cells Over the past 40 years, researchers have used cancer cell lines, patient samples, and small organism models (e.g., fruit ﬂy, zebraﬁsh, and mouse) to study the molecular mechanisms of cancer initiation, progression, and metastasis, but the complexity of the cancer genome and differences among species frequently limit clinical translation. Although there are iPSC models for a number of genetic diseases that predispose to cancer, to date, relatively few of these systems have been used to explore mechanisms of oncogenesis. We discuss several examples of these pioneer models in the following sections. Li–Fraumeni Syndrome Li–Fraumeni syndrome (LFS) is an autosomal dominant inherited cancer syndrome that is characterized by early onset of a variety of tumor types, including soft-tissue sarcoma and osteosarcoma, breast cancer, brain tumors, leukemia, and adrenocortical carcinoma [35]. Our group established a model of LFS using patient-derived iPSCs to delineate mechanisms of mutant p53 in osteosarcoma [30]. In this system, osteoblasts differentiated from LFS iPSCderived mesenchymal stem cells (MSCs) recapitulate osteosarcoma features, including

Glossary Clustered, regularly interspaced, short palindromic repeat/Cas9 (CRISPR/Cas9): a genome-editing methodology based on the bacterial acquired immune system. Functioning in bacteria as a means of resistance to exogenous genetic elements similar to RNA interference in eukaryotic cells, it recognizes and cleaves DNAs based on a target RNA sequence. CRISPR systems depend on CRISPR RNA (crRNA) and trans-activating crRNA (tracrRNA) for sequence-specific silencing. There are three types of CRISPR/Cas9 systems. The type II system used in Cas9 and its variants serves as an RNA-guided DNA nuclease or nickase that cleaves DNA upon crRNA–tracrRNA target recognition. The ease with which constructs targeting desired genomic loci can be generated has made this technology the tool of choice for many seeking to perform genome editing. ESCs: pluripotent stem cells are derived from the inner cell mass of the blastocyst, an embryo at the preimplantation stage. These cells are capable of proliferating and dividing without differentiating for a prolonged period in an in vitro tissue culture environment. Induced pluripotent stem cells: pluripotent stem cells derived from differentiated somatic cells through somatic reprogramming by defined factors (e.g., OCT4, SOX2, KLF4, and c-MYC). Organoid: a collection of multiple organ-specific cells cultured in a 3D system that self-organize into organbud structures. 3D-cultured organoids mimic the microanatomy of organs and are capable of recapitulating specific organ functions, enabling experimental study of otherwise inaccessible tissue. Pluripotent stem cells: cells with equivalent characteristics to the inner cell mass of the blastocyst-stage embryo. Pluripotent stem cells are capable of differentiating into any cell type and give rise to all adult tissues (pluripotency) and extensively replicate without differentiation and/or senescence (self-renewal). Reprogram: the process of converting one specific cell type to another. It includes the conversion of somatic cells (e.g., dermal fibroblasts)

Trends in Cancer, September 2016, Vol. 2, No. 9

487

Table 1. Established PSCs Models of Cancer or Diseases That Predispose to Cancer

Disease

Inheritance

Prevalence

Defected Gene

Cancer Type

Key Model Findings in Cancer

Refs

Alpha-1 antitrypsin deﬁciency

Autosomal co-dominant

1 in 1500 to 3500

SERPINA1

Liver cancer

N/A

[18,23, 77]

Ataxiatelangiectasia

Autosomal recessive

1 in 40 000 to 100 000

ATM

Leukemia and lymphoma

N/A

[78]

Diamond– Blackfan anemia

Autosomal dominant

1 in 5 000 000 to 7 000 000

RPL5, RPL11, RPL35A, RPS7, RPS10, RPS17, RPS19, RPS24, and RPS26

Osteosarcoma

N/A

[79,80]

Diffuse intrinsic pontine gliomas

Sporadic

1 in 250 000 to 500 000

H3.3

Brain tumor

Undifferentiated epigenome structure and a primitive stem cell gene signature in NPCs with H3.3K27 M expression, PDGFRA activation, and p53 loss

[33]

Down syndrome

Sporadic

1 in 800

Chromosome 21

Leukemia

N/A

[11,81]

Dyskeratosis congenita

Autosomal dominant or recessive

1 in 1 000 000

DKC1, TERC, TERT, and TINF2

Leukemia

N/A

[82,83]

Dystrophic epidermolysis bullosa

Autosomal recessive

1 in 150 000 to 1 000 000

COL7A1

Skin cancer

N/A

[84,85]

Fanconi anemia

Autosomal recessive

1 in 160 000

FANC genes, BRCA2, BRIP1, PALB2, and RAD51C

Leukemia

N/A

[86]

Glioblastoma multiforme

Sporadic

1 in 33 000 to 50 000

EGFR, PIK3CA, PTEN, and TP53

Brain tumor

Elevated PAX7 and GBM-associated gene signature in PTEN-deﬁcient NSCs

[34]

Li–Fraumeni syndrome

Autosomal dominant

1 in 20 000

TP53 and CHEK2

Osteosarcoma, breast cancer, brain tumor, and soft-tissue sarcoma

Impaired expression of H19 and osteosarcoma signature in LFS osteoblasts

[30]

del(7q)Myelodysplastic syndrome

Sporadic

1 in 100 000

Chromosome 7q

Leukemia

Impaired myeloid lineage differentiation in del(7q) iPSCs dependent on HIPK2, ATP6V0E2, LUC7L2, and EZH2

[32]

488

Trends in Cancer, September 2016, Vol. 2, No. 9

to pluripotent stem cells and the conversion of one type of somatic cell to another. Transcription activator-like effector nucleases (TALENs): a genome-editing methodology based on fusions of the FokI DNA nuclease and a TAL effector DNA-binding domain derived from Xanthomonas bacteria. Engineered TAL effectors are able to bind to any specific DNA sequence, permitting FokI nuclease activity at a desired DNA location. TAL effectors contain highly conserved 33–34 amino acid repeat domains that recognize a specific base pair. TALENs induce target double-stranded breaks to facilitate homologous recombination and enable customized genome alterations. TALENs are believed to have greater precision for genome editing than zinc-finger nuclease and clustered, regularly interspaced, short palindromic repeat/Cas9. Zinc-finger nuclease: a genomeediting methodology based on a fusion of the FokI restriction enzyme with a Cys2His2 zinc-finger DNAbinding domain. Zinc-finger nucleases were among the first available tools to perform genome editing.

Table 1. (continued) Disease

Inheritance

Prevalence

Defected Gene

Cancer Type

Key Model Findings in Cancer

Refs

Noonan syndrome

Autosomal dominant

1 in 1000 to 2500

PTPN11, BRAF, KRAS, NRAS, RAF1, and SOS1

Leukemia

Proliferation of CD33+ myeloid cells and elevated miR-15a and miR-223 in NS/JMML hematopoietic cells

[31]

Polycythemia vera

Sporadic

1 in 2000

JAK2 and TET2

Leukemia

N/A

[87]

Shwachman– Diamond syndrome

Autosomal recessive

1 in 100 000 to 1 000 000

SBDS

Leukemia

N/A

[11,88]

Werner syndrome

Autosomal recessive

1 in 200 000

WRN

Skin cancer, soft-tissue sarcoma

N/A

[89,90]

Wilms tumor

Sporadic or hereditary

1 in 10 000

WT1

Kidney cancer

N/A

[91]

Abbreviations: GBM, Glioblastoma multiforme; iPSC, induced pluripotent stem cell; LFS, Li–Fraumeni syndrome; NPCs, neural progenitor cells; NS/JMML, Noonan syndrome/juvenile myelomonocytic leukemia; NSC, neural stem cells; PSC, pluripotent stem cell.

defective osteoblastic differentiation and tumorigenic ability. Gene expression in LFS osteoblasts is also similar to the expression profile in primary osteosarcomas, particularly the more aggressive phenotypes. LFS-derived osteoblasts are free of cytogenetic rearrangements, permitting study of early oncogenic mechanisms prior to accumulation of secondary genomic alterations. Expression of the long noncoding RNA H19 had been previously linked to p53 activity [36] and transcriptome analysis suggested impaired expression of H19 in LFS osteoblasts. Further functional studies showed that H19 is essential for normal osteogenesis and inhibition of tumorigenesis. The LFS iPSC disease model uncovered a previously unidentified role of p53 in osteogenic differentiation defects and tumorigenesis. Noonan Syndrome Noonan syndrome (NS) is an autosomal dominant disorder characterized by a wide spectrum of congenital heart abnormalities, short stature, facial dimorphism, and predisposition to hematological malignancies. A subset of NS patients will develop juvenile myelomonocytic leukemia (JMML), an aggressive myelodysplastic and myeloproliferative neoplasm [37]. Mulero-Navarro et al. [31] investigated the molecular mechanisms involved in NS-associated JMML harboring PTPN11 mutations using hematopoietic cells derived from NS/JMML patient-specific iPSCs. These hematopoietic cells recapitulated several JMML characteristics including hypersensitivity to granulocyte-macrophage colony-stimulating factor and increased myeloid population. Comparison of transcriptome profiles of controls and NS/JMML-derived CD33+ myeloid cells confirmed dysregulation of extracellular signal-regulated kinase (ERK) and Janus kinase/signal transducers and activators of transcription signaling (JAK/STAT) and proliferation of NS/JMML CD33+ myeloid cells. Expression levels of miR-15a and miR-223 were also elevated in these cells. Notably, dysregulation of miR-15a and miR-223 is commonly observed in mononuclear cells isolated from JMML patients harboring PTPN11 mutations. Using the NS/JMML iPSC model, Mulero-Navarro et al. [31] demonstrated that inhibition of these miRNAs could restore normal myelopoiesis, providing a novel therapeutic target for PTPN11-mutated JMML.

Trends in Cancer, September 2016, Vol. 2, No. 9

489

Myelodysplastic Syndrome Myelodysplastic syndrome (MDS) is a bone marrow disorder that leads to defective hematopoiesis and a disposition to develop anemia, cytopenia, and leukemia. Sporadic loss of one copy of the long arm of chromosome 5 [del(5q)] and/or chromosome 7 [del(7q)] is a characteristic cytogenetic abnormality in MDS [38]. Kotini et al. [32] established del(7q) MDS iPSCs from patient hematopoietic stem cells with loss of chromosome 7q and iPSCs derived from normal fibroblast isogenic controls. iPSCs with del(7q) recapitulated the phenotype of impaired myeloid lineage differentiation seen in MDS. This defective differentiation potential could be reproduced by engineering hemizygosity of definite 7q segments in normal iPSCs and could be rescued by spontaneous acquisition of an extra chromosome 7. Through phenotype-rescue screening, Kotini et al. [32] identified HIPK2, ATP6V0E2, LUC7L2, and EZH2 as haploinsufficient genes involved in del(7q) MDS-associated hematopoietic defects. Diffuse Intrinsic Pontine Gliomas Diffuse intrinsic pontine gliomas (DIPGs) are rare highly aggressive pediatric brain tumors that arise from glial tissue. Somatic p.Lys27Met substitution in histone 3.3 (H3.3K27M) is commonly detected in patients with DIPGs and is associated with poor survival [39,40]. Funato et al. [33] engineered human ESC-derived neural progenitor cells (NPCs) with heterozygous H3.3K27M mutations. To mimic genetic alterations found in clinical samples of H3.3K27M-mutated DIPGs, ESC-derived NPCs were transduced with lentiviruses also carrying constitutively active PDGFRA (D842V) and p53 small hairpin RNA. In this NPC model, H3.3K27M expression synergized with PDGFRA activation and p53 loss, culminating in neoplastic transformation. Genome-wide analyses of H3.3K27M-transformed NPCs revealed that they maintain both an undifferentiated epigenome structure and a primitive stem cell gene signature, enabling their tumorigenic potential. Using a small-molecule chemical library screen of compounds targeting epigenetic regulators, they identified the MEN1 inhibitor MI-2 as a potential drug for the subset of DIPGs harboring the H3.3K27M mutation. This study demonstrates the potential of PSCs for drug screening. Glioblastoma Multiforme Glioblastoma multiforme (GBM), also known as Grade IV astrocytoma, is a highly malignant brain tumor derived from glial cells. While GBMs are genetically very diverse, mutations in PTEN are common and correlate with increased invasion, drug resistance, and tumor recurrence [41]. Duan et al. [34] engineered PTEN-deficient ESCs using a TALEN-based genome-editing methodology and derived neural stem cells (NSCs) to model GBM. PTEN-deficient NSCs displayed the GBM-associated gene signature and formed intracranial tumors in vivo. Duan et al. [34] found that elevated levels of PAX7 contributed to neoplastic transformation by producing more aggressive phenotypes. Elevated PAX7 expression can be explained directly by PTEN deficiency since PTEN interacts with CREB/CBP and co-occupies the PAX7 promoter. Screening for anticancer compounds in PTEN-deficient NSCs suggested mitomycin C as a potential drug.

Directed Differentiation and Organoids The potential of PSCs in modeling cancer is critically dependent on availability of defined methods to differentiate PSCs into the tissues from which tumors arise. The generation of specific cell or tissue types by directed differentiation and organoids is a fast-growing field in stem cell research [6–8]. Directed Differentiation Directed differentiation is the application of a temporally defined set of external factors or culture conditions in order to produce a cell population enriched for a desired lineage. For example, we have employed a directed differentiation protocol to derive MSCs from p53-mutant iPSCs.

490

Trends in Cancer, September 2016, Vol. 2, No. 9

While the MSCs exhibited no obvious deficits and were free of chromosomal abnormalities, key oncogenic features representative of osteosarcoma emerged after their subsequent differentiation into osteoblasts [30]. PSC differentiation protocols have been established for generation of hepatic cells [42–44], intestinal tissue [45], thyroid follicular cells [46], airway epithelial cells [47–49], renal progenitor cells [50,51], retinal cells [52–55], epithelial stem cells [56], epidermal keratinocytes [57], myeloid cells [58,59], neurons [34,60], MSCs [61], and melanocytes [62], among many others. Application of these protocols to cancer patient-derived iPSCs and engineered ESCs should allow research of cancers of these organs. Directed PSC differentiation protocols pose particular challenges to model cancer. PSCs can form teratomas, a rare tumor type from a normal genetic background. While differentiated cells derived from PSCs do not have this tumorigenic potential, even small contamination of desired differentiated cells with pluripotent cells can give the false appearance of tumorigenic properties. An effective differentiation protocol and purification scheme is therefore essential for any cancer study. Despite these advances, there is a large unmet need for directed differentiation protocols for a broader set of cancer types, including breast, prostate, and ovary. As directed differentiation techniques from PSCs continue to develop, the versatility of using PSCs in cancer modeling will also expand. Organoid Culture Recent advances in 3D culture techniques combined with existing differentiation protocols have enabled the generation of PSC-derived specific tissue or progenitor cells within a self-organized assembly known as an organoid [7,8]. Compared with traditional 2D culture systems, these 3D organoid cultures mimic better their in vivo PSC-derived counterparts, hence positioning the technology as a powerful tool for studying human development and modeling disease. 3D cerebral organoids were derived by differentiation of human PSCs by Lancaster et al [63]. Matrigel droplets containing cerebral organoids were transferred into a spinning bioreactor, enabling a rapid, longer, and more abundant formation of 3D brain tissue. These ‘mini-brain’ systems facilitate the study of human brain development and have been used to model microcephaly, among other neurodevelopmental disorders. With the introduction of appropriate mutations, these brain organoid systems have the potential to enhance our understanding of brain tumor biology. 3D gastrointestinal organoids were also derived from PSCs by Wells et al. [45] and closely mimic the in vivo intestinal epithelium. These organoids contain crypt-like and villus-like structures and consist of differentiated enterocytes, goblet cells, and intestinal stem cells. Another protocol for generating PSC-derived gastric organoids [64] was later published by the same group. These 3D ‘mini-stomach’ systems were used to model Helicobacter pylori infection, shedding light on the pathogenesis of this common disease. Organoids generated from human intestinal stem cells were also used to recapitulate the colorectal adenoma-to-carcinoma transition by Matano et al. [65], highlighting the potential role of 3D organoid culture in studying gastrointestinal oncogenesis. PSC-derived 3D culture and organoid systems have been used to model the liver bud [66,67], lung [47], and pancreas [68] from endoderm; the kidney [69–72] from mesoderm; and the optic cup [73–75] from ectoderm. These methodologies hold substantial potential for investigating the distinct tumor types derived from these now-accessible cell lineages.

Trends in Cancer, September 2016, Vol. 2, No. 9

491

Concluding Remarks

Outstanding Questions

PSC disease models have not only succeeded in replicating disease phenotypes but have also begun to ﬁnd applications in the understanding of disease biology and in the development of novel therapies (see Outstanding Questions). Applications of PSCs to cancer will be advantageous to (i) model disease phenotypes; (ii) elucidate pathological mechanisms; (iii) predict patient survival; (iv) identify potential biomarkers and therapeutic targets; (v) discover haploinsufﬁcient genes by functional mapping of disease-associated chromosomal loss; and (vi) apply drug screening to identify potential compounds to rescue particular disease phenotypes.

Can the prognosis of and potential therapeutic strategies for cancers originating from differentiated lineages (e. g., osteoblasts, myelocytes, or astrocytes) be modeled reliably using patient induced pluripotent stem cell (iPSC)derived cells?

The high levels of genomic alterations already present in cancer cell lines and tumor-derived mouse models make the elucidation of the initial steps of tumor development particularly challenging. Notably, iPSC-derived cells are free of cytogenetic rearrangements [4], allowing the study of early oncogenic mechanisms prior to the accumulation of secondary genomic alterations. A recent investigation comparing the mutational rate of somatic and pluripotent cell lines also demonstrated a tenfold lower mutation rate in iPSCs with each generation compared with somatic cells [76], highlighting the advantages of using genetically stable iPSCs rather than somatic or even more error-prone cancer cells to expand a cell population with a given genetic trait. PSC disease models should be useful in identifying and characterizing the ‘second hit’ during tumor development following a selective and stable introduction of the first one. Nonetheless, newly generated PSC lines should be fully characterized prior to experimental use to rule out chromosomal or genetic abnormalities. Recent progress in genome-editing tools, including optimizations of the TALEN and CRISPR/ Cas9 systems, made the generation of tailored alterations in PSCs from isogenic backgrounds feasible. These powerful tools enable cancer researchers to easily construct a particular model system to investigate the role of specific gene alterations in tumorigenesis in various tissues and organs (Figure 1B). While a number of genetic cancer syndromes exhibit incomplete penetrance, the incidence of disease and the risk factor of developing disease frequently cannot be accurately estimated due to the relatively small number of affected patients. Genome-manipulated ESCs from a wider range of genetic backgrounds may help complement studies from patient-derived iPSCs to clarify the role of genetic backgrounds in cancer predisposition from familial cancer syndromes. Despite the versatility of PSC technology as a model for a number of cancer etiologies, the limited number of available protocols for tissue differentiation remains one of the major impediments precluding the wider application of this approach in cancer research. New developments in directed differentiation protocols and evolving 3D organoid culture techniques not only facilitate disease modeling in neuroscience, cardiology, and regenerative medicine – fields in which PSC models are currently frequently used – but also open up opportunities in cancer research as more tissue types become experimentally available. We hope that new PSC-derived organoid methodologies will fill in the gaps in our ability to produce many currently inaccessible cancer cell types, including breast, prostate, and ovary. In conclusion, PSCs are destined for exciting future applications in the field of cancer biology. We look forward to the wide incorporation of PSC techniques into the toolkit of basic and clinical researchers seeking to efficiently recapitulate disease, characterize alterations stemming from specific cancer-associated mutations or hoping to complement their existing studies with a human model. Acknowledgments We sincerely apologize to authors whose work we could not include due to space limitations. I.R.L is supported by the NIH grant R01 HL119404. D-F.L. is the CPRIT scholar in Cancer Research and supported by NIH Pathway to Independence Award R00 CA181496 and CPRIT Award RR160019.

492

Trends in Cancer, September 2016, Vol. 2, No. 9

Can the generation and manipulation of patient-derived iPSCs become sufficiently standardized to allow for the use of iPSC disease models in the clinical management of familial cancer syndromes? Do iPSCs derived from distinct patients with the same cancer syndrome have similar gene expression profiles? Does individual genetic background significantly affect the gene expression profiles of distinct iPSC-derived tumors? Can PSCs be applied to identify and characterize the ‘second hit’ during tumor development? Are changes in patient-derived iPSCs equivalent to those in genome-manipulated ESCs in familial cancer syndromes? Do in vitro reprogramming and differentiation processes trigger artificial tumor development in PSC disease models? Can patient iPSC-derived cells be used as part of cancer immunotherapy to train the immune system to recognize a cancer signature?

Supplemental Information Supplemental information associated with this article can be found online at http://dx.doi.org/10.1016/j.trecan.2016.07.007.

References 1. Thomson, J.A. et al. (1998) Embryonic stem cell lines derived from human blastocysts. Science 282, 1145–1147

activator-like effector nuclease. Curr. Protoc. Stem Cell Biol. 29, 5A.7.1–5A.18

2. Takahashi, K. and Yamanaka, S. (2006) Induction of pluripotent stem cells from mouse embryonic and adult ﬁbroblast cultures by deﬁned factors. Cell 126, 663–676

27. Ding, Q. et al. (2013) A TALEN genome-editing system for generating human stem cell-based disease models. Cell Stem Cell 12, 238–251

3. Takahashi, K. et al. (2007) Induction of pluripotent stem cells from adult human ﬁbroblasts by deﬁned factors. Cell 131, 861–872

28. Mali, P. et al. (2013) RNA-guided human genome engineering via Cas9. Science 339, 823–826

4. Yu, J. et al. (2007) Induced pluripotent stem cell lines derived from human somatic cells. Science 318, 1917–1920

29. Cong, L. et al. (2013) Multiplex genome engineering using CRISPR/Cas systems. Science 339, 819–823

5. González, F. et al. (2011) Methods for making induced pluripotent stem cells: reprogramming à la carte. Nat. Rev. Genet. 12, 231–242

30. Lee, D.F. et al. (2015) Modeling familial cancer with induced pluripotent stem cells. Cell 161, 240–254

6. Murry, C.E. and Keller, G. (2008) Differentiation of embryonic stem cells to clinically relevant populations: lessons from embryonic development. Cell 132, 661–680 7. Lancaster, M.A. and Knoblich, J.A. (2014) Organogenesis in a dish: modeling development and disease using organoid technologies. Science 345, 1247125 8. Yin, X. et al. (2016) Engineering stem cell organoids. Cell Stem Cell 18, 25–38 9. Grskovic, M. et al. (2011) Induced pluripotent stem cells – opportunities for disease modelling and drug discovery. Nat. Rev. Drug Discov. 10, 915–929 10. Dimos, J.T. et al. (2008) Induced pluripotent stem cells generated from patients with ALS can be differentiated into motor neurons. Science 321, 1218–1221 11. Park, I.H. et al. (2008) Disease-speciﬁc induced pluripotent stem cells. Cell 134, 877–886 12. Yagi, T. et al. (2011) Modeling familial Alzheimer's disease with induced pluripotent stem cells. Hum. Mol. Genet. 20, 4530–4539

31. Mulero-Navarro, S. et al. (2015) Myeloid dysregulation in a human induced pluripotent stem cell model of PTPN11-associated juvenile myelomonocytic leukemia. Cell Rep. 13, 504–515 32. Kotini, A.G. et al. (2015) Functional analysis of a chromosomal deletion associated with myelodysplastic syndromes using isogenic human induced pluripotent stem cells. Nat. Biotechnol. 33, 646–655 33. Funato, K. et al. (2014) Use of human embryonic stem cells to model pediatric gliomas with H3.3K27 M histone mutation. Science 346, 1529–1533 34. Duan, S. et al. (2015) PTEN deﬁciency reprogrammes human neural stem cells towards a glioblastoma stem cell-like phenotype. Nat. Commun. 6, 10068 35. Li, F.P. et al. (1988) A cancer family syndrome in twenty-four kindreds. Cancer Res. 48, 5358–5362 36. Dugimont, T. et al. (1998) The H19 TATA-less promoter is efﬁciently repressed by wild-type tumor suppressor gene product p53. Oncogene 16, 2395–2401 37. Roberts, A.E. et al. (2013) Noonan syndrome. Lancet 381, 333–342

13. Brennand, K.J. et al. (2011) Modelling schizophrenia using human induced pluripotent stem cells. Nature 473, 221–225

38. Lindsley, R.C. and Ebert, B.L. (2013) Molecular pathophysiology of myelodysplastic syndromes. Annu. Rev. Pathol. 8, 21–47

14. Carvajal-Vergara, X. et al. (2010) Patient-speciﬁc induced pluripotent stem-cell-derived models of LEOPARD syndrome. Nature 465, 808–812

39. Wu, G. et al. (2012) Somatic histone H3 alterations in pediatric diffuse intrinsic pontine gliomas and non-brainstem glioblastomas. Nat. Genet. 44, 251–253

15. Pasca, S.P. et al. (2011) Using iPSC-derived neurons to uncover cellular phenotypes associated with Timothy syndrome. Nat. Med. 17, 1657–1662

40. Schwartzentruber, J. et al. (2012) Driver mutations in histone H3.3 and chromatin remodelling genes in paediatric glioblastoma. Nature 482, 226–231

16. Itzhaki, I. et al. (2011) Modelling the long QT syndrome with induced pluripotent stem cells. Nature 471, 225–229

41. Furnari, F.B. et al. (2007) Malignant astrocytic glioma: genetics, biology, and paths to treatment. Genes Dev. 21, 2683–2710

17. Moretti, A. et al. (2010) Patient-speciﬁc induced pluripotent stem-cell models for long-QT syndrome. N. Engl. J. Med. 363, 1397–1409

42. Si-Tayeb, K. et al. (2010) Highly efﬁcient generation of human hepatocyte-like cells from induced pluripotent stem cells. Hepatology 51, 297–305

18. Rashid, S.T. et al. (2010) Modeling inherited metabolic disorders of the liver using human induced pluripotent stem cells. J. Clin. Invest. 120, 3127–3136 19. Liu, G.H. et al. (2011) Targeted gene correction of laminopathyassociated LMNA mutations in patient-specific iPSCs. Cell Stem Cell 8, 688–694 20. Smith-Arica, J.R. et al. (2003) Infection efficiency of human and mouse embryonic stem cells using adenoviral and adeno-associated viral vectors. Cloning Stem Cells 5, 51–62 21. Ivanova, N. et al. (2006) Dissecting self-renewal in stem cells with RNA interference. Nature 442, 533–538 22. Lee, D.F. et al. (2012) Regulation of embryonic and induced pluripotency by aurora kinase-p53 signaling. Cell Stem Cell 11, 179–194 23. Yusa, K. et al. (2011) Targeted gene correction of /1-antitrypsin deficiency in induced pluripotent stem cells. Nature 478, 391–394 24. Zou, J. et al. (2009) Gene targeting of a disease-related gene in human induced pluripotent stem and embryonic stem cells. Cell Stem Cell 5, 97–110 25. Hockemeyer, D. et al. (2009) Efficient targeting of expressed and silent genes in human ESCs and iPSCs using zinc-finger nucleases. Nat. Biotechnol. 27, 851–857 26. Luo, Y. et al. (2014) Generation of GFP reporter human induced pluripotent stem cells using AAVS1 safe harbor transcription

43. Goldman, O. et al. (2013) KDR identiﬁes a conserved human and murine hepatic progenitor and instructs early liver development. Cell Stem Cell 12, 748–760 44. Hannan, N.R. et al. (2013) Production of hepatocyte-like cells from human pluripotent stem cells. Nat. Protoc. 8, 430–437 45. Spence, J.R. et al. (2011) Directed differentiation of human pluripotent stem cells into intestinal tissue in vitro. Nature 470, 105–109 46. Antonica, F. et al. (2012) Generation of functional thyroid from embryonic stem cells. Nature 491, 66–71 47. Dye, B.R. et al. (2015) In vitro generation of human pluripotent stem cell derived lung organoids. Elife 4, e05098 48. Huang, S.X. et al. (2014) Efﬁcient generation of lung and airway epithelial cells from human pluripotent stem cells. Nat. Biotechnol. 32, 84–91 49. Huang, S.X. et al. (2015) The in vitro generation of lung and airway progenitor cells from human pluripotent stem cells. Nat. Protoc. 10, 413–425 50. Xia, Y. et al. (2013) Directed differentiation of human pluripotent cells to ureteric bud kidney progenitor-like cells. Nat. Cell. Biol. 15, 1507–1515 51. Takasato, M. et al. (2014) Directing human embryonic stem cell differentiation towards a renal lineage generates a self-organizing kidney. Nat. Cell Biol. 16, 118–126

Trends in Cancer, September 2016, Vol. 2, No. 9

493

52. Osakada, F. et al. (2009) Stepwise differentiation of pluripotent stem cells into retinal cells. Nat. Protoc. 4, 811–824 53. Nakano, T. et al. (2012) Self-formation of optic cups and storable stratified neural retina from human ESCs. Cell Stem Cell 10, 771–785 54. Mellough, C.B. et al. (2012) Efficient stage-specific differentiation of human pluripotent stem cells toward retinal photoreceptor cells. Stem Cells 30, 673–686 55. Lamba, D.A. et al. (2006) Efficient generation of retinal progenitor cells from human embryonic stem cells. Proc. Natl. Acad. Sci. U.S. A. 103, 12769–12774 56. Yang, R. et al. (2014) Generation of folliculogenic human epithelial stem cells from induced pluripotent stem cells. Nat. Commun. 5, 3071 57. Kidwai, F.K. et al. (2014) Differentiation of epidermal keratinocytes from human embryonic stem cells. Methods Mol. Biol. 1195, 13–22

75. Kuwahara, A. et al. (2015) Generation of a ciliary margin-like stem cell niche from self-organizing human retinal tissue. Nat. Commun. 6, 6286 76. Rouhani, F.J. et al. (2016) Mutational history of a human cell lineage from somatic to induced pluripotent stem cells. PLoS Genet. 12, e1005932 77. Somers, A. et al. (2010) Generation of transgene-free lung disease-speciﬁc human induced pluripotent stem cells using a single excisable lentiviral stem cell cassette. Stem Cells 28, 1728–1740 78. Fukawatase, Y. et al. (2014) Ataxia telangiectasia derived iPS cells show preserved x-ray sensitivity and decreased chromosomal instability. Sci. Rep. 4, 5421 79. Garçon, L. et al. (2013) Ribosomal and hematopoietic defects in induced pluripotent stem cells derived from Diamond Blackfan anemia patients. Blood 122, 912–921 80. Ge, J. et al. (2015) Dysregulation of the transforming growth factor b pathway in induced pluripotent stem cells generated from patients with Diamond Blackfan anemia. PLoS ONE 10, e0134878

58. Grigoriadis, A.E. et al. (2010) Directed differentiation of hematopoietic precursors and functional osteoclasts from human ES and iPS cells. Blood 115, 2769–2776

81. Li, L.B. et al. (2012) Trisomy correction in Down syndrome induced pluripotent stem cells. Cell Stem Cell 11, 615–619

59. Kennedy, M. et al. (2007) Development of the hemangioblast deﬁnes the onset of hematopoiesis in human ES cell differentiation cultures. Blood 109, 2679–2687

82. Agarwal, S. et al. (2010) Telomere elongation in induced pluripotent stem cells from dyskeratosis congenita patients. Nature 464, 292–296

60. Chambers, S.M. et al. (2009) Highly efﬁcient neural conversion of human ES and iPS cells by dual inhibition of SMAD signaling. Nat. Biotechnol. 27, 275–280

83. Batista, L.F. et al. (2011) Telomere shortening and loss of selfrenewal in dyskeratosis congenita induced pluripotent stem cells. Nature 474, 399–402

61. Lian, Q. et al. (2007) Derivation of clinically compliant MSCs from CD105+, CD24– differentiated human ESCs. Stem Cells 25, 425–436

84. Itoh, M. et al. (2011) Generation of keratinocytes from normal and recessive dystrophic epidermolysis bullosa-induced pluripotent stem cells. Proc. Natl. Acad. Sci. U.S.A. 108, 8797–8802

62. Fang, D. et al. (2006) Deﬁning the conditions for the generation of melanocytes from human embryonic stem cells. Stem Cells 24, 1668–1677

85. Tolar, J. et al. (2011) Induced pluripotent stem cells from individuals with recessive dystrophic epidermolysis bullosa. J. Invest. Dermatol. 131, 848–856

63. Lancaster, M.A. et al. (2013) Cerebral organoids model human brain development and microcephaly. Nature 501, 373–379

86. Raya, A. et al. (2009) Disease-corrected haematopoietic progenitors from Fanconi anaemia induced pluripotent stem cells. Nature 460, 53–59

64. McCracken, K.W. et al. (2014) Modelling human development and disease in pluripotent stem-cell-derived gastric organoids. Nature 516, 400–404 65. Matano, M. et al. (2015) Modeling colorectal cancer using CRISPR-Cas9-mediated engineering of human intestinal organoids. Nat. Med. 21, 256–262

87. Ye, Z. et al. (2009) Human-induced pluripotent stem cells from blood cells of healthy donors and patients with acquired blood disorders. Blood 114, 5473–5480

66. Takebe, T. et al. (2013) Vascularized and functional human liver from an iPSC-derived organ bud transplant. Nature 499, 481–484

88. Tulpule, A. et al. (2013) Pluripotent stem cell models of Shwachman-Diamond syndrome reveal a common mechanism for pancreatic and hematopoietic dysfunction. Cell Stem Cell 12, 727–736

67. Takebe, T. et al. (2014) Generation of a vascularized and functional human liver from an iPSC-derived organ bud transplant. Nat. Protoc. 9, 396–409

89. Cheung, H.H. et al. (2014) Telomerase protects werner syndrome lineage-speciﬁc stem cells from premature aging. Stem Cell Reports 2, 534–546

68. Boj, S.F. et al. (2015) Organoid models of human and mouse ductal pancreatic cancer. Cell 160, 324–338

90. Shimamoto, A. et al. (2014) Reprogramming suppresses premature senescence phenotypes of Werner syndrome cells and maintains chromosomal stability over long-term culture. PLoS ONE 9, e112900

69. Xia, Y. et al. (2014) The generation of kidney organoids by differentiation of human pluripotent cells to ureteric bud progenitor-like cells. Nat. Protoc. 9, 2693–2704 70. Freedman, B.S. et al. (2015) Modelling kidney disease with CRISPR-mutant kidney organoids derived from human pluripotent epiblast spheroids. Nat. Commun. 6, 8715 71. Morizane, R. et al. (2015) Nephron organoids derived from human pluripotent stem cells model kidney development and injury. Nat. Biotechnol. 33, 1193–1200 72. Takasato, M. et al. (2015) Kidney organoids from human iPS cells contain multiple lineages and model human nephrogenesis. Nature 526, 564–568 73. Eiraku, M. et al. (2011) Self-organizing optic-cup morphogenesis in three-dimensional culture. Nature 472, 51–56 74. Völkner, M. et al. (2016) Retinal organoids from pluripotent stem cells efﬁciently recapitulate retinogenesis. Stem Cell Reports 6, 525–538

494

Trends in Cancer, September 2016, Vol. 2, No. 9

91. Thatava, T. et al. (2011) Successful disease-speciﬁc induced pluripotent stem cell generation from patients with kidney transplantation. Stem Cell Res. Ther. 2, 48 92. Komor, A.C. et al. (2016) Programmable editing of a target base in genomic DNA without double-stranded DNA cleavage. Nature 533, 420–424 93. Menon, T. et al. (2015) Lymphoid regeneration from genecorrected SCID-X1 subject-derived iPSCs. Cell Stem Cell 16, 367–372 94. Xie, F. et al. (2014) Seamless gene correction of b-thalassemia mutations in patient-speciﬁc iPSCs using CRISPR/Cas9 and piggyBac. Genome. Res. 24, 1526–1533 95. Bellin, M. et al. (2013) Isogenic human pluripotent stem cell pairs reveal the role of a KCNH2 mutation in long-QT syndrome. EMBO J. 32, 3161–3175

Resource Prospective Derivation of a Living Organoid Biobank of Colorectal Cancer Patients Marc van de Wetering,1,2,12 Hayley E. Francies,3,12 Joshua M. Francis,4,5,12 Gergana Bounova,6 Francesco Iorio,7 Apollo Pronk,8 Winan van Houdt,8 Joost van Gorp,9 Amaro Taylor-Weiner,4 Lennart Kester,1 Anne McLaren-Douglas,3 Joyce Blokker,1,2 Sridevi Jaksani,1,2 Sina Bartfeld,1 Richard Volckman,10 Peter van Sluis,10 Vivian S.W. Li,11 Sara Seepo,4 Chandra Sekhar Pedamallu,4,5 Kristian Cibulskis,4 Scott L. Carter,4,5 Aaron McKenna,4 Michael S. Lawrence,4 Lee Lichtenstein,4 Chip Stewart,4 Jan Koster,10 Rogier Versteeg,10 Alexander van Oudenaarden,1 Julio Saez-Rodriguez,7 Robert G.J. Vries,1,2 Gad Getz,4,5 Lodewyk Wessels,6 Michael R. Stratton,3 Ultan McDermott,3 Matthew Meyerson,4,5 Mathew J. Garnett,3,* and Hans Clevers1,2,* 1Hubrecht Institute, Royal Netherlands Academy of Arts and Sciences (KNAW), Cancer Genomics and University Medical Center, 3584 CT Utrecht, the Netherlands 2Foundation Hubrecht Organoid Technology (HUB), 3584 CT Utrecht, the Netherlands 3Wellcome Trust Sanger Institute, Wellcome Trust Genome Campus, Hinxton, Cambridgeshire CB10 1SA, UK 4The Broad Institute of MIT and Harvard, Cambridge, MA 02142, USA 5Department of Medical Oncology, Dana-Farber Cancer Institute, Boston, MA 02115, USA 6Computational Cancer Biology, Netherlands Cancer Institute, 1066 CX Amsterdam, the Netherlands 7European Molecular Biology Laboratory, European Bioinformatics Institute, Wellcome Trust Genome Campus, Cambridge CB10 1SA, UK 8Department of Surgery, Diakonessenhuis, 3582 KE Utrecht, the Netherlands 9Department of Pathology, Diakonessenhuis, 3582 KE Utrecht, the Netherlands 10Department of Oncogenomics, Academic Medical Center, University of Amsterdam, 1105 AZ Amsterdam, the Netherlands 11Division of Stem Cell Biology and Developmental Genetics, MRC National Institute for Medical Research, The Ridgeway, Mill Hill, London NW7 1AA, UK 12Co-first author *Correspondence: mg12@sanger.ac.uk (M.J.G.), h.clevers@hubrecht.eu (H.C.) http://dx.doi.org/10.1016/j.cell.2015.03.053

SUMMARY

In Rspondin-based 3D cultures, Lgr5 stem cells from multiple organs form ever-expanding epithelial organoids that retain their tissue identity. We report the establishment of tumor organoid cultures from 20 consecutive colorectal carcinoma (CRC) patients. For most, organoids were also generated from adjacent normal tissue. Organoids closely recapitulate several properties of the original tumor. The spectrum of genetic changes within the ‘‘living biobank’’ agrees well with previous large-scale mutational analyses of CRC. Gene expression analysis indicates that the major CRC molecular subtypes are represented. Tumor organoids are amenable to highthroughput drug screens allowing detection of gene-drug associations. As an example, a single organoid culture was exquisitely sensitive to Wnt secretion (porcupine) inhibitors and carried a mutation in the negative Wnt feedback regulator RNF43, rather than in APC. Organoid technology may fill the gap between cancer genetics and patient trials, complement cell-line- and xenograft-based drug studies, and allow personalized therapy design. INTRODUCTION Colorectal carcinoma (CRC) represents one of the major forms of cancer. Seminal studies have revealed a series of molecular

pathways that are critical to the pathogenesis of CRC, including WNT, RAS-MAPK, PI3K, P53, TGF-b, and DNA mismatch repair (Fearon, 2011; Fearon and Vogelstein, 1990). Large-scale sequencing analyses have dramatically extended the list of recurrently mutated genes and chromosomal translocations (Garraway and Lander, 2013; Vogelstein et al., 2013). CRC cases are characterized by either microsatellite instability (MSI) (associated with a hyper-mutator phenotype), or as microsatellite-stable (MSS) but chromosomally unstable (CIN) (Lengauer et al., 1997). The absolute number and combination of genetic alterations in CRC confounds our ability to unravel the functional contribution of each of these potential cancer genes. Thus, while genome changes in tumors of individual patients can be assessed in great detail and at low cost, these data are difficult to interpret in terms of prognosis, drug response, or patient outcome, necessitating model systems for analysis of genotype-to-phenotype correlations. Self-renewal of the intestinal epithelium is driven by Lgr5 stem cells located in crypts (Barker et al., 2007). We have recently developed a long-term culture system that maintains basic crypt physiology (Sato et al., 2009). Wnt signals are required for the maintenance of active crypt stem cells (Korinek et al., 1998; Kuhnert et al., 2004; Pinto et al., 2003). Indeed, the Wnt agonist R-spondin1 induces dramatic crypt hyperplasia in vivo (Kim et al., 2005). R-spondin-1 is the ligand for Lgr5 (Carmon et al., 2011; de Lau et al., 2011). Epidermal growth factor (EGF) signaling is associated with intestinal proliferation (Wong et al., 2012), while transgenic expression of Noggin induces a dramatic increase in crypt numbers (Haramis et al., 2004). The combination of R-spondin-1, EGF, and Noggin in Cell 161, 933–945, May 7, 2015 ª2015 Elsevier Inc. 933

Basement Membrane Extract (BME) sustains ever-expanding small intestinal organoids, which display all hallmarks of the original tissue in terms of architecture, cell-type composition, and self-renewal dynamics. We adapted the culture condition for long-term expansion of human colonic epithelium and primary colonic adenocarcinoma, by adding nicotinamide, A83-01 (Alk inhibitor), Prostaglandin E2, and the p38 inhibitor SB202190 (Sato et al., 2011). Of note, a 2D culture method for cells from normal and malignant primary tissue has been described by Liu et al. (2012). Here, we explore organoid technology to routinely establish and phenotypically annotate ‘‘paired organoids’’ derived from adjacent tumor and healthy epithelium from CRC patients. RESULTS Establishment of a Living CRC Biobank Surgically resected tissue was obtained from previously untreated CRC patients. Tissue from rectal cancer patients was excluded because they routinely undergo irradiation before surgery. For multiple tissues, we observe that normal tissuederived organoids outcompete tumor organoids under the optimized culture conditions, presumably due to genomic instability and resulting apoptosis in the latter. Combination of Wnt3A and the Wnt amplifier R-spondin1 is essential to grow organoids from normal epithelium. Over 90% of CRC cases harbor mutations that aberrantly activate the Wnt signaling pathway (Cancer Genome Atlas Network, 2012), so we exploited the Wnt-dependency of normal colonic stem cells to selectively expand tumor organoids. A total of 22 tumor organoid cultures and 19 normal-adjacent organoid cultures were derived from 20 patients (P19 and P24 each carried two primary tumors separated by >10 cm; Figure 1A). We successfully generated organoid cultures from 22 of 27 tumor samples. For one, we never observed growth. Four were lost due to bacterial/yeast infection. Since then, we have added next-generation antibiotics (see Experimental Procedures) and currently observe an 90% success rate. The number of primary tumor organoids varied between patient samples, with some tumors rendering thousands of primary organoids whereas others yielded only 10–20 primary organoids. This difference in derivation likely reflects the heterogeneous composition of tumors, with proliferative areas intermingled with regions of differentiated cells, stromal cells or necrosis. The growth rate of the organoids from patients 5 and 27 decreased over time, which prohibited their inclusion in the drug screen. All other organoids could be readily expanded and frozen to create a master cell bank. Upon thawing, cell survival was typically >80%. Unlike healthy tissue-derived organoids, tumor-derived organoids presented with a range of patient-specific morphologies, ranging from thin-walled cystic structures to compact organoids devoid of a lumen. H&E staining on primary tumors and the corresponding organoids revealed that the ‘‘cystic versus solid’’-organization of the epithelium was generally preserved. Yet, marker expression analysis (KI67, OLFM4, KRT 20, Alcian blue) revealed heterogeneity both between patients and individual organoids within each culture (Figure 1B; Data S1). 934 Cell 161, 933–945, May 7, 2015 ª2015 Elsevier Inc.

Genomic Characterization of Tumor-Derived Organoids Genomic DNA was isolated from tumor and matched normal organoid cultures for whole-exome sequencing in order to identify tumor-specific somatic mutations (Cancer Genome Atlas Network, 2012). Genomic DNA from the corresponding biopsy specimens were available for comparative analysis for 16 of these cases (Table S1A). The mutation rates per Mb varied widely for different tumor organoids (range 2.0–77.9), with a median value of 3.7 in the tumor organoids, similar to the median rate of 3.6 in the biopsy samples (Figure 2A; Table S1B). Mutations were predominantly CpG to T transitions, consistent with results from large-scale CRC sequencing (Figures S1A and S1B; Table S1C). Of the 22 tumor organoids, six displayed hypermutation (>10 mutations/Mb): P7, P10 and the organoids from the two patients with two tumors each (P19a and P19b, P24a and P24b). Interestingly, the P19a and P19b tumors share TP53 R273C and BRAF V600E alterations, suggesting they arose from the same somatically altered progenitor cell but then diverged to acquire independent secondary alterations (Figures S1C and S1D). In contrast, the P24a and P24b tumors share 80% (469/590) of somatic alterations but then have discordant driving alterations in APC and TP53, indicating that the hypermutator phenotype may have been present prior to the acquisition of growth promoting mutations (Figures S1E and S1F). The frequency of hypermutated organoid cultures in our patient panel (20%; 4 of 20) agreed with the reported frequency in a much larger cohort of clinical samples and display comparable somatic copy number alterations (SCNAs) (Figure 2B; Table S1D) (Bass et al., 2011; Cancer Genome Atlas Network, 2012). The successful derivation of both hypermutated and non-hypermutated organoids implies an absence of culture-based bias. Somatic variants within the coding regions in organoid cultures were highly concordant with the corresponding biopsy specimen for both hypermutated and non-hypermutated patients (median = 0.88 frequency of concordance, range 0.62–1.00) (Figure 3A; Table S1E). Indeed, combined analysis of SCNAs and single nucleotide variants (SNVs) to infer Cancer Cell Fractions (CCF) (Carter et al., 2012; Landau et al., 2013) in the biopsy and tumor organoids, revealed that the common CRC driver mutations were maintained in culture. In 13 out of 14 organoid-biopsy pairs tested, tumor subclones sharing common CRC drivers were detected in the biopsy. In 50% of the organoids, a dominant subclone from the biopsy was present, likely representing sampling during derivation but it could also indicate loss in culture (Figures S2A and S2B; Tables S1F and S1G). Transcriptome analysis of single organoids showed subtle differences in gene expression within an organoid culture, confirming their heterogeneous composition. The differences in overall gene expression were more pronounced in the organoids derived from the hypermutant tumors (Figure S2C). Discordant mutations were assessed for their likely biological significance in cancer, based on Cancer Gene Census and data reported from the PanCancer analysis of 5,000 whole exomes (Futreal et al., 2004; Lawrence et al., 2014). Only 4% (27/679) of discordant mutations found in organoids affected cancerrelated genes, including a third hit to APC, which was already biallelically inactivated in P14, SMAD4 mutation in P16, and POLE mutation in P19b (Table S1H). Cancer-significant genes

Figure 1. Derivation of Organoids from Primary Tissue (A) Overview of the procedure. A total of 22 tumor organoids and 19 normal control organoids were derived and analyzed by exome-sequencing, RNA expression analysis and high-throughput drug screening. To determine the concordance between tumor organoids and primary tumor, DNA from the primary tumor was also isolated. (B) Organoids architecture resembles primary tumor epithelium. H&E staining of primary tumor and the tumor organoids derived of these. A feature of most organoids is the presence of one or more lumens, resembling the tubular structures of the primary tumor (e.g., P8 and P19b). Tumors devoid of lumen give rise to compact organoids without lumen (P19a). Scale bar, 100 mM. See also Data S1.

that were discordant in the biopsy represented 4.4% (12/271) (Table S1H). The discordant mutations had a mean allelic frequency of 10.3% and 34.1% for the biopsy and organoids, respectively. This could represent the enrichment or depletion of a sub-clonal population in the organoid culture present within

the original tumor, as well as acquisition of additional mutations during derivation or propagation. The most commonly altered genes in CRC (Bass et al., 2011; Cancer Genome Atlas Network, 2012; Lawrence et al., 2014) were well represented in the organoid cultures (Figure 3B; Cell 161, 933â&#x20AC;&#x201C;945, May 7, 2015 ÂŞ2015 Elsevier Inc. 935

Figure 2. CRC Subtypes Are Present in Organoid Cultures (A) Whole exome sequencing of the tumor and corresponding biopsy, when available, revealed the presence of hypermutated (>10 mutations/Mb) and non-hypermutated subtypes within the organoids. Comparable rates of mutations were observed in the tumor organoid (O) and tumor biopsy (B). Organoids without corresponding biopsy are indicated in with red (O). (B) Comparison of somatic copy-number alterations found in the biopsies and corresponding organoids (Biop/Org) and TCGA CRC in hypermutated and non-hypermutated samples. See also Figure S1 and Tables S1A–S1D.

Tables S1I and S1J). Inactivating alterations to the tumor suppressors APC, TP53, FBXW7, and SMAD4, as well as activating mutations in KRAS (codon 12 and 146) and PIK3CA (codon 545 and 1047) were observed. Activating mutations in BRAF and TGFBR1/2 mutations were observed in the hypermutated organoids, consistent with previous reports for primary CRC (Cancer Genome Atlas Network, 2012). Mutations of genes in DNA mismatch repair (MMR)-associated pathways are associated with a hypermutated phenotype (Boland and Goel, 2010). Consistent with their classification as hypermutated CRC cases (Cancer Genome Atlas Network, 2012), missense mutations were present in MSH3 in P7, and POLE mutations were detected in P10, P19a, and P19b. We did not observe mutations in MMR-associated genes in P24a and P24b and expression analysis showed normal levels of the pertinent genes. The culprit for hyper mutability thus remains to be identified for P24. The limited cohort size did not allow a statistical analysis for somatic copy number alterations to identify significant regions of amplification and deletions. However, manual inspection of the top regions identified by TCGA did reveal the presence of ERBB2-, MYC-, and IGF2-amplified organoids, as well as a reported gain of 13q in the non-hypermutated group (Figure 3C) In aggregate, these analyses demonstrate that organoid cultures faithfully capture the genomic features of the primary tumor from which they derive and much of the genomic diversity of CRC. Most CRC cases carry activating mutations in the WNT pathway: inactivation mutations in APC, FBXW7, AXIN2, and FAM123B, or activating mutations in CTNNB1 (Cancer Genome Atlas Network, 2012). Gene fusions involving the Wnt-agonistic RSPO2 and RSPO3 genes have been observed in 5%–10% of CRC (Seshagiri et al., 2012). RNF43 encodes a negative regulator of the Wnt pathway, which serves to remove the Wnt receptor FZ in a negative feedback loop (Hao et al., 2012; Koo et al., 2012, de Lau et al., 2014). Recent sequencing efforts of gastric, ovary, and pancreatic neoplasias identified RNF43 mutations (Jiao et al., 2014; Ryland et al., 2013; Wang et al., 2014), and RNF43 mutations have been observed in 936 Cell 161, 933–945, May 7, 2015 ª2015 Elsevier Inc.

CRC (Giannakis et al., 2014; Ivanov et al., 2007; Koo et al., 2012) We found APC alterations in all but four of the organoids (P11, P19a/b, P28). Western blotting revealed P11 to express a truncated APC protein, pointing to a mutational event not covered by our exome-sequencing (Figure S3). The wtAPC organoid P28 carries an activating mutation in CTNNB1 (T41A). In both P19a and P19b, we detected RNF43 mutations: frameshifts at aa positions 659 and 355, respectively. Only the latter is predicted to affect protein function. RNA Analysis of Normal and Tumor-Derived Organoids Organoid cultures consist purely of epithelial cells. Therefore, the system allows for direct gene expression analysis without a contamination from mesenchyme, blood vessels, immune cells, etc. Normal colon-derived and tumor-derived organoids were plated under identical conditions in complete medium (+Wnt). After 3 days, RNA was analyzed using Affymetrix single transcript arrays. Figure 4A shows the correlation heatmap of the organoid samples. Normal colon-derived organoids clustered tightly together, while the tumor-derived organoids exhibited much more heterogeneity. Next, we searched for genes differentially expressed between normal and tumor organoids. Normal colon-derived organoids (Figure 4B) expressed genes of differentiated cells (e.g., the goblet cell markers MUC1 and MUC4 and the colonocyte marker CA2). Genes enriched in tumor organoids included cancer-associated genes such as PROX1, BAMBI, and PTCH1 and the Wnt target gene APCDD1 (Takahashi et al., 2002). Several CRC classifications have been proposed based on RNA expression. We combined expression data from organoid samples and TCGA tissue samples and classified these in subtypes using the gene signatures by Sadanandam et al. (2013). Figure 4C displays the subtyping of the 22 organoid samples and 431 TCGA RNA sequencing (RNA-seq) tumor tissue samples. The heatmap shows the normalized scores of genes by samples, both sorted by subtype (see Experimental Procedures). Organoid samples were spread across the subtypes, with the transit-amplifying (TA) subtype being most frequently represented. The enterocyte subtype was not represented. In addition, the RNA expression data allowed expression analysis

Figure 3. Genomic Alterations Found in CRC Are Represented in Organoid Cultures (A) Concordance of somatic mutations detected in organoid and corresponding biopsies. Bar graph represents the proportion of coding alterations that are concordant between the biopsy and the corresponding organoid culture and those that are found only in organoid or biopsy specimen. N/A indicates cases in which exome-sequencing was not performed on the corresponding biopsy. (B) Overview of the mutations found in the tumor organoids. The hash-mark in each box represents each allele and whether it was subject to deletion, mutation, frame-shift alteration, nonsense mutation or splice site mutation. Those alterations present in >10% of cases are compared to the percentage of cases reported by the TCGA CRC. *Indicates discordant mutations targeting the same gene between the two sites in P19 and P24. See also Tables S1I and S1J. (C) Somatic copy-number alterations in organoids among commonly amplified genes identified in TCGA CRC. See also Figures S2 and S3 and Tables S1D–S1J.

of individual genes in organoids. MLH1 expression was absent from two tumor organoids from patient 19 as well as from patient 7 (that is also mutant in MSH3) (Figure S4). In the two tumor organoids from P24, we did not detect expression changes in MLH1 or any other MSI-associated gene. Effect of Porcupine Inhibitor on RNF43 Mutant Organoids Unlike most other WNT pathway mutations, RNF43 mutations yield a cell that is hypersensitive to—yet still dependent on— secreted WNT. Array data confirmed the expression of several WNTs by the organoids (Figure S5A). The O-acyltransferase Porcupine is required for the secretion of WNTs and its inhibition prevents autocrine/paracrine activation of the pathway (Kadowaki et al., 1996). The small molecule porcupine inhibitor IWP2 (Chen et al., 2009) was tested on a small panel of the tumor organoids and strongly affected the RNF43 mutant P19b organoid (Figure 5A). This observation implied that porcupine inhibition may be evaluated for treatment of the small subset of cancer patients mutant in RNF43. Organoid Proof-of-Concept Drug Screen Prompted by this, we developed a robotized drug sensitivity screen in 3D-organoid culture and correlated drug sensitivity with genomic features to identify molecular signatures associated with altered drug response. Organoid cultures were gently disrupted and plated on BME-coated 384-well plates in a 2%

BME solution. Organoids were left overnight before being drugged and left for 6 days before measuring cell number using CellTiter-Glo reagent. Drug sensitivity was represented by the half-maximal inhibitory concentration (IC50), the slope of the dose-response curve, and area under the dose-response curve (AUC). A bespoke 83 compound library was assembled for screening, including drugs in clinical use (n = 25), chemotherapeutics (n = 10), drugs previously investigated in or currently undergoing studies in clinical trials (n = 29), and experimental compounds to a diverse range of cancer targets (n = 29) (Table S2A). The library included the anti-EGFR antibody cetuximab, used clinically for KRAS/NRAS/BRAF wild-type CRC, as well as oxaliplatin and 5-FU, first line chemotherapeutics for CRC treatment. In total, 19 of 20 tumor organoids (from 18 different patients) were successfully screened in experimental triplicate, generating >5,000 measurements of organoid-drug interactions (Table S2B). We incorporated a number of controls into the assay design. The median Z factor score, a measure of assay plate quality, across all screening plates was 0.62 (n = 119; upper and lower quartile = 0.85 and 0.3, respectively), consistent with an experimentally robust assay. We did observe some unexplained organoid-specific variation in assay plate quality. Dose-response measurements were performed in experimental triplicate or duplicate (on separate plates) and replicate AUC values were highly correlated (Pearson correlation [Rp] > 0.87) (Figure 5B). Cell 161, 933–945, May 7, 2015 ª2015 Elsevier Inc. 937

Figure 4. RNA Expression Analysis (A) Correlation heat map of normal organoids versus tumor organoids based on 2,186 genes (the top 10% of genes in terms of SD). The normal organoids are very highly correlated with each other, whereas the tumor samples exhibit more heterogeneity. The colors represent pairwise Pearson correlations after the expression values have been logged and mean-centered for every gene. The hierarchical clustering is based on one minus correlation distance. The affix N = normal, T = tumor. (B) MA plot of logged normal versus tumor gene expression. p values are computed with the R package limma, by comparing normal versus tumor gene expression. Cancer-associated genes (e.g., APCDD1, PROX1, and PTCH1) are shown in the top half. (C) CRC molecular subtypes are represented by the organoid panel. Genes by samples heat map of normalized gene expression of 22 organoid samples and 431 TCGA RNA-seq tumor tissue samples, organized by subtype. Within each subtype, samples are sorted by their mean gene expression for the signature genes associated with that specific subtype. See also Figure S4.

938 Cell 161, 933â&#x20AC;&#x201C;945, May 7, 2015 ÂŞ2015 Elsevier Inc.

Figure 5. Development of a High-Throughput Drug Screening Assay Utilizing Organoid Models (A) Autocrine/paracrine WNT signaling in P19b. A small panel of tumor organoids was incubated with increasing amounts of the Porcupine inhibitor IWP2. Growth of the RNF43 mutant P19b was inhibited, indicative of dependency on autocrine/paracrine WNT signaling. Error bars indicate the SD of triplicate measurements. See also Figure S5. (B) Scatterplot of (1-AUC) values for all technical replicates of drug screening data. Plots show the correlation between the three different technical replicates and each data point represents the (1-AUC) value for an individual organoid. (C) Scatterplots of the correlation in (1-AUC) values for three compounds (GDC0941, obatoclax mesylate, and trametinib) screened twice during every screening run. Values are the mean of three technical replicates.

Furthermore, the compounds trametinib, GDC0941, and obatoclax mesylate were screened twice independently on separate assay plates and a good correlation was observed between the experimentally determined AUC values (Rp = 0.79, 0.71, and 0.76, respectively) (Figure 5C). As a first validation, the only tumor organoid in the panel that was sensitive to the Porcupine inhibitor LGK974 was P19b (Figure S5B), confirming the observations made with IWP2 (Figure 5A). The clustering of compounds based on their IC50 values demonstrated a diverse range of sensitivities across the organoids and identified three major sub-groups (Figure 6A). One group was associated with sensitivity to a majority of the compounds (organoids P8, P7, and P19a), in contrast to the cluster (P31, P11) exhibiting insensitivity. The remaining organoids had intermediate sensitivity. Interestingly, the multifocal tumors P19a and P19b, derived from the same patient and

both carrying the BRAF V600E mutation, differed in their overall drug response profile. We observed clustering of drugs that inhibit the IGF1R and PI3K-AKT signaling pathways (Figure 6A), and compounds with similar nominal targets had comparable activity across the organoid collection. For example, a similar sensitivity pattern was observed for the PI3K inhibitors GDC0941 and BYL719 (a-selective), the IGF1R inhibitors OSI-906 and BMS-536924, EGFR inhibitors cetuximab and gefitinib, and the BRAF inhibitors dabrafenib and PLX4720 (Figure 6B). All but one of the organoids displayed a lack of sensitivity to BRAF inhibition. P19a, a BRAF V600E mutant organoid, displayed partial sensitivity to dabrafenib with an IC50 of 0.5 mM, comparable to IC50 values of BRAF V600E colorectal cancer cell lines (range 0.004–2.55 mM; average 0.96 mM). To identify genetic correlates between individual oncogenic mutations and drug response, we performed a multivariate Cell 161, 933–945, May 7, 2015 ª2015 Elsevier Inc. 939

(legend on next page)

940 Cell 161, 933–945, May 7, 2015 ª2015 Elsevier Inc.

analysis of variance (MANOVA) incorporating IC50 values and slopes of the corresponding dose-response curves, with MSIstatus as a covariate. Complete drug sensitivity and genomic data sets were available for 18 organoids and used for this analysis. The analysis included 16 genes identified as mutated, amplified, or deleted in CRC (referred to as mutant genes) as described by Lawrence et al. (2014) (Table S3). The MANOVA identified a subset (12 of 864, 1%) of gene-drug associations as statistically significant (p < 0.005, incorporating a 30% false discovery rate [FDR]) (Table S4). These results were further filtered based on the magnitude of the effect size on the IC50 values of wild-type versus mutant cell line populations (effect size >2; Cohen’s D), and correlations identified due to a singlet outlier organoids were removed. This resulted in the identification of one high confidence gene-drug association already reported in the literature (Vassilev et al., 2004). Loss-of-function mutations of the tumor suppressor TP53 were associated with resistance to nutlin-3a (p = 0.0018), an inhibitor of MDM2 (Figure 7A). Of the four organoids that were wild-type for TP53 by DNA sequencing, only P18 was (unexpectedly) insensitive to nutlin-3a. However, immunohistochemistry of p53 in P18 revealed the protein to be stabilized, indicative of functional inactivation of the p53 pathway (Figure 7B). We could also readily detect resistance to the anti-EGFR inhibitors cetuximab and BIBW2992 (afatinib) in the setting of KRAS mutant organoids (p = 0.008/FDR 37% and p = 0.029/ FDR 54%, respectively), although these associations were below statistical significance when considering an FDR <30% (Figures 7C and S6). Of the KRAS wild-type organoids, a subset 2/10 was insensitive to cetuximab, including P19b that has a BRAF mutation, a known mediator of cetuximab resistance (Di Nicolantonio et al., 2008). For the remaining organoid, further mechanisms beyond mutated KRAS/NRAS/BRAF are likely to be involved in cetuximab resistance (De Roock et al., 2010; Vecchione, 2014). We also identified a number of compounds with differential activity in the absence of an apparent genetic biomarker (Figure 7D). For example, a subset of organoids was exquisitely sensitive to the AKT1/2 inhibitor MK2206. Similarly, we observed distinct subsets of organoids that are exquisitely sensitive to the pan-ERBB inhibitor AZD8931 and the chemotherapeutic gemcitabine. We also performed a validation screen with 11 of the original 83 compounds across the organoid panel and compared the measured responses (Figure S7; Table S5). We observed positive correlation for all compounds and nine exhibited good to fair reproducibility as indicated by an Rp of 0.5 or greater (Figures 7E and 7F). Variation within the assay was likely due to inherent technical noise, biological variation, and sensitivity to outlier data points due to the small number of organoids. In summary, the successful application of organoids in a systematic and unbiased high-throughput drug screen to

identify clinically relevant biomarkers demonstrates the feasibility and utility of organoid technology for investigating the molecular basis of drug response. Furthermore, the identification of putative novel molecular markers has opened avenues for further investigation of drug sensitivity in CRC. The current analysis is still constrained by the relatively small number of patients. The derivation of a significantly larger organoid collection would increase the representation of rare genotypes and the statistical power to detect molecular markers of drug response. DISCUSSION Cancer cell lines have served for many years as the workhorse model in cancer research. Recent studies have exploited highthroughput screening of large panels of cancer cell lines to identify drug-sensitivity patterns and to correlate drug sensitivity to genomic alterations (Barretina et al., 2012; Garnett et al., 2012). From these high-throughput cell-line-based studies, a picture emerges of a complex network of biological factors that affect sensitivity to the majority of cancer drugs. For instance, no direct relationship may exist between sensitivity to a certain drug and a single genomic alteration. Instead, difficult-to-find, complex interactions between multiple genomic alterations may determine drug sensitivity outcome. Thus, with currently available insights, it remains a challenge to develop algorithms that accurately predict the drug sensitivity of a patient’s tumor based on the spectrum of genomic alterations present, in the context of the unique genetic background. Two approaches to determine directly the drug sensitivity in a patient-derived sample have been quite widely exploited, namely the short-term culture of tumor sections (Centenera et al., 2013), and xeno-transplantation of the tumor into immunodeficient mice (Jin et al., 2010; Tentler et al., 2012). Short-term culture allows for in vitro screening at a reasonably large scale, but is constrained by the limited proliferative capacity of the cultures. Xenotransplantation allows for in vivo screening but is resource-intensive due to the need for large mouse colonies. It thus appears of interest to develop additional technologies that allow the combination of sequencing and high-throughput drug screening in patient-derived samples. Here, we demonstrate that the organoid culture platform can be exploited for genomic and functional studies at the level of the individual patient at a scale that cannot be achieved by existing approaches. Our organoid drug screening assay generates reproducible high quality drug sensitivity data, positive correlation of biological replicates, and reproducible activity of compounds inhibiting the same target. By connecting genetic and drug sensitivity data, we were able to confirm the activity of cetuximab in a subset of KRAS wild-type organoids reflecting observations made in the clinic (De Roock et al., 2010) as well as Nutlin-3a effectiveness in TP53 wild-type organoids. Furthermore, we describe

Figure 6. Heatmap of IC50s of All 85 Compounds against 19 Colorectal Cancer Organoids (A) Organoids have been clustered based on their IC50 values across the drug panel. The drug names and their nominal target(s) are provided in the bottom panel. (B) Drugs with the same nominal targets have similar activity profiles across the organoid panel. (1-AUC) values are plotted for inhibitor of PI3K (GDC0941 and BYL719), IGF1R (OSI-906 and BMS-536924), EGFR (cetuximab and gefitinib), and BRAF (PLX4720 and dabrafenib). See also Tables S2A and S2B.

Cell 161, 933–945, May 7, 2015 ª2015 Elsevier Inc. 941

Figure 7. Gene-Drug Associations and Differential Drug Sensitivity Profiles of Interest (A) Association of TP53 mutational status with nutlin-3a response. Viability response curves of the altered (blue) and wild-type organoids (gray) as well as scatter plots of cell line IC50 (mM) values are shown. IC50 values are on a natural logarithmic scale. Each circle represents one cell line, red bars indicate geometric means of IC50 values and black bold bars indicate median log IC50 values. Box top/low bounds indicate upper/lower quartiles, and whiskers (indicated by the dashed lines) extend to extreme values (minimal and maximal) excluding outliers (i.e., whose value is more than 3/2 times the upper quartile and less than 3/2 times the lower quartile). Purple bar positions on the y axis indicate means +/ log IC50 SD. (B) Immunohistochemical staining showing stabilization of TP53 in organoid P18. Scale bar, 100 mM. (C) Association of KRAS status and cetuximab response. Colors and symbols coding is the same as (A). (D) Dose-response curves after 6 days treatment with MK2206, AZD8931, and gemcitabine. (E) Reproducibility of drug response profiles for 11 drugs. The Pearson correlation score of (1-AUC) values from the primary screen compared to (1-AUC) values from validation screens are used for comparison. The validation screen was performed twice (run 1 and 2) with >1 month elapsed between each screen. NA, data unavailable for this drug. (F) The correlation of 1-AUC values from the primary and validation screens for AZD8931, gemcitabine, and nutlin-3a. See also Figures S6 and S7 and Tables S3, S4, and S5.

942 Cell 161, 933â&#x20AC;&#x201C;945, May 7, 2015 ÂŞ2015 Elsevier Inc.

the differential activity of a handful of clinical and preclinical compounds (gemcitabine, MK2206, and AZD8941). Tumors are composed of a mixture of sub-clones that coevolve through a Darwinian selection process. This cellular heterogeneity and phenotypic variation allows the emergence of a complex clonal architecture, which underpins important features such as drug resistance and metastatic potential (Burrell et al., 2013). Our CCF analysis of clonal structure determined that almost all of the biopsies were polyclonal at the time of resection, and this is reflected to varying extent in the corresponding organoid culture. The ability to capture sub-clonal populations in in vitro organoid culture should enable more predictive modeling of patient responses to therapy. In many respects, the clonal selection and heterogeneity observed in organoids is similar to PDX models of cancers (Eirew et al., 2015). For both models, understanding the factors that affect tumor heterogeneity and evolution, and how heterogeneity impacts on drug response, will be important to fully exploit their potential for predicting patient responses. We perceive patient-derived organoids to be used to directly test drug sensitivity of the tumor in a personalized treatment approach. For this, we envision organoids to be tested against a limited number of clinically approved drugs within weeks after derivation. While building this pilot biobank, we observed that normal epithelial tissue always yield good numbers of organoids within weeks, while significant differences in ‘‘take rates’’ were observed between patients’ tumor organoids. Crucial for this approach to be effective, is to decrease the time needed to derive and expand the organoids. In conclusion, tumor organoids may fill the gap between cancer genetics and patient trials, complement cell-line- and xenograft-based drug studies, and allow personalized therapy design. EXPERIMENTAL PROCEDURES Human Tissues Colonic tissues were obtained from The Diakonessen Hospital Utrecht with informed consent and the study was approved by the ethical committee. All patients were diagnosed with colorectal cancer. From the resected colon segment, normal as well as tumor tissue was isolated. The isolation of healthy crypts and tumor epithelium was performed essentially as described by Sato et al. (2011). Organoid Culture Healthy tissue-derived organoids were cultured in Human Intestinal Stem Cell medium (HISC). The composition of HISC is: Basal culture medium with 50% Wnt conditioned medium, 20% R-Spondin conditioned medium, 10% Noggin conditioned medium, 13 B27, 1,25 mM n-Acetyl Cysteine, 10 mM Nicotinamide, 50 ng/ml human EGF, 10 nM Gastrin, 500 nM A83-01, 3 uM SB202190, 10 nM Prostaglandin E2, and 100 mg/ml Primocin (Vivogen). Tumor organoids were cultured in HICS minus Wnt. See the Extended Experimental Procedures for a detailed description. Whole-Exome Sequencing and Copy-Number Analysis For each sample, 250 ng of DNA was sheared and subject to whole-exome sequencing using the Agilent v2 capture probe set and sequenced by HiSeq2500 using 76 base pair reads, as previously described (Fisher et al., 2011; Imielinski et al., 2012). A median 9.6 Gb of unique sequence was generated for each sample (Table S1A). Sequence data were locally realigned to improve sensitivity and reduce alignment artifacts prior to identification of mutations, insertions, and deletions

as previously described (Cibulskis et al., 2013; DePristo et al., 2011; Ojesina et al., 2014). Somatic copy-number analysis was performed using segmented copynumber profiles generated from whole-exome sequencing using the SegSeq algorithm (Table S1D) (Chiang et al., 2009). The procedure is described in detail in the Extended Experimental Procedures. Organoid Data Processing RNA from 22 organoid tumor samples and 15 paired normal samples was hybridized on Affymetrix Human Gene 2.0 ST arrays. The raw CEL files were processed with Affymetrix Power Tools using the Hg19 genome build and NetAffx annotation dating from 09-30-2012. Between-array normalization was performed using rma-sketch, within APT. This resulted in an intensity matrix of 21,681 genes by 37 samples. For analysis of individual genes, data were analyzed using the R2 web application, which is freely available at http://r2.amc.nl. To subtype the samples, we used the gene signature published by Sadanandam et al. (2013). The procedure is described in detail in the Extended Experimental Procedures. Organoid Viability Assays Eight microliters of 7 mg/ml BME was dispensed in to 384-well microplates and allowed to polymerize. Organoids were mechanically dissociated by pipetting before being resuspended in 2% BME/growth media (15–20,000 organoids/ml) and dispensed into drug wells. The following day a 5-point 4-fold dilution series of each compound was dispensed using liquid handling robotics and cell viability assayed using CellTiter-Glo (Promega) following 6 days of drug incubation. All screening plates were subjected to stringent quality control measures and a Z factor score comparing negative and positive control wells calculated. Dose-response curves were fitted to the luminescent signal intensities utilizing a method previously described (Garnett et al., 2012). Further information of the compounds used, data-fitting algorithm, and validation screen can be found in the Extended Experimental Procedures. Systematic Multivariate Analysis of Variance We excluded from the analysis drugs with no IC50 values falling within the range of tested concentrations. For each of the remaining drugs, we assembled an 18 3 2 matrix Y composed by two vectors of length n = 18, containing IC50 values and dose-response curve slopes b, respectively, obtained by treating 18 organoids with the drug under consideration. A multivariate analysis of variance (MANOVA) model was then fitted to this drug response data matrix with factors including the microsatellite stability status of the organoids and the status (altered or wild-type) of 16 genomic features (Extended Experimental Procedures). Significance and effect size scores were obtained for each of the genomic-feature/drug pairs. Q values were subsequently obtained by correcting the MANOVA p values for multiple hypotheses testing, and a threshold of 30% of positive false discovery rate, IC50, and effect size >2 (as quantified by the Cohen’s D) was used to identify significant associations. ACCESSION NUMBERS The accession number for the healthy and tumor organoid array data reported in this paper is GEO: GSE64392. The accession number for the single organoid RNA-seq data is GEO: GSE65253. SUPPLEMENTAL INFORMATION Supplemental Information includes Extended Experimental Procedures, seven figures, five tables, and one data file and can be found with this article online at http://dx.doi.org/10.1016/j.cell.2015.03.053. AUTHOR CONTRIBUTIONS M.v.d.W. derived, maintained, and analyzed organoid cultures. H.E.F. developed, performed, and analyzed the organoid drug screen. J.M.F. analyzed

Cell 161, 933–945, May 7, 2015 ª2015 Elsevier Inc. 943

sequencing data. G.B. analyzed RNA expression data. R.G.J.V. organized ethical approval. F.I. performed analyses and statistical inferences on the drug screening data supervised by J.S.R. A.P. and W.v.H. performed surgery. J.v.G. isolated tumor and normal tissue from resected material. A.T.M. performed cancer cell fraction analysis. L.K. performed single organoid transcriptomics supervised by A.v.O. A.M. assisted in drug screening. J.B. performed immunostainings and assisted in culturing organoids together with S.J. and S.B. P.v.d.S. and R.V. processed and analyzed RNA, supervised by R.V. V.S.W.L. performed APC western analysis. S.S. processed DNA samples for exome sequencing. C.S.P., K.C., S.L.C., A.M., M.S.L., L.L., and C.S. helped in processing and analyzing sequencing data. G.G. and M.M. supervised the sequencing and analysis. L.W. supervised the RNA analysis. M.R.S., U.M., M.G., and H.C. participated in the development of the project concept. M.G. aided in drug data analysis. M.v.d.W. and H.E.F. participated in data analysis and project design. M.v.d.W., H.E.F., J.M.F., M.M., M.J.G., and H.C. wrote the manuscript.

ACKNOWLEDGMENTS Thanks to the Broad Institute Genomics Platform for processing and generating the sequence data, the cell line screening, and analysis team at the Sanger Institute, Liliane Wijnaendts and Joost Oudejans for help with tumor dissections, Harry Begthel and Jeroen Korving for histology, Sepideh Darakhshan for help with organoids, and the members of the M.M., M.J.G., and H.C. laboratories for support. M.v.d.W. is supported by Stichting Virtutis Opus and Stichting Vrienden van het Hubrecht. R.G.J.V., S.J., and J.B. are supported by a Grant from Alpe dHuzes/KWF. This work was supported with a grant from the Dutch Cancer Society to M.S. (H1/2014-6919). The organoids are available for academic research upon evaluation of a research proposal by the Medical Ethical Comity. M.M. received a commercial research grant from Bayer and has ownership interest (including patents) in and is a consultant/advisory board member for Foundation Medicine. Received: October 29, 2014 Revised: January 20, 2015 Accepted: March 18, 2015 Published: May 7, 2015 REFERENCES Barker, N., van Es, J.H., Kuipers, J., Kujala, P., van den Born, M., Cozijnsen, M., Haegebarth, A., Korving, J., Begthel, H., Peters, P.J., and Clevers, H. (2007). Identification of stem cells in small intestine and colon by marker gene Lgr5. Nature 449, 1003–1007. Barretina, J., Caponigro, G., Stransky, N., Venkatesan, K., Margolin, A.A., Kim, S., Wilson, C.J., Lehár, J., Kryukov, G.V., Sonkin, D., et al. (2012). The Cancer Cell Line Encyclopedia enables predictive modelling of anticancer drug sensitivity. Nature 483, 603–607. Bass, A.J., Lawrence, M.S., Brace, L.E., Ramos, A.H., Drier, Y., Cibulskis, K., Sougnez, C., Voet, D., Saksena, G., Sivachenko, A., et al. (2011). Genomic sequencing of colorectal adenocarcinomas identifies a recurrent VTI1ATCF7L2 fusion. Nat. Genet. 43, 964–968. Boland, C.R., and Goel, A. (2010). Microsatellite instability in colorectal cancer. Gastroenterology 138, 2073–2087. Burrell, R.A., McGranahan, N., Bartek, J., and Swanton, C. (2013). The causes and consequences of genetic heterogeneity in cancer evolution. Nature 501, 338–345. Cancer Genome Atlas Network (2012). Comprehensive molecular characterization of human colon and rectal cancer. Nature 487, 330–337. Carmon, K.S., Gong, X., Lin, Q., Thomas, A., and Liu, Q. (2011). R-spondins function as ligands of the orphan receptors LGR4 and LGR5 to regulate Wnt/beta-catenin signaling. Proc. Natl. Acad. Sci. USA 108, 11452–11457. Carter, S.L., Cibulskis, K., Helman, E., McKenna, A., Shen, H., Zack, T., Laird, P.W., Onofrio, R.C., Winckler, W., Weir, B.A., et al. (2012). Absolute quantifica-

944 Cell 161, 933–945, May 7, 2015 ª2015 Elsevier Inc.

tion of somatic DNA alterations in human cancer. Nat. Biotechnol. 30, 413–421. Centenera, M.M., Raj, G.V., Knudsen, K.E., Tilley, W.D., and Butler, L.M. (2013). Ex vivo culture of human prostate tissue and drug development. Nat. Rev. Urol. 10, 483–487. Chen, B., Dodge, M.E., Tang, W., Lu, J., Ma, Z., Fan, C.-W., Wei, S., Hao, W., Kilgore, J., Williams, N.S., et al. (2009). Small molecule-mediated disruption of Wnt-dependent signaling in tissue regeneration and cancer. Nat. Chem. Biol. 5, 100–107. Chiang, D.Y., Getz, G., Jaffe, D.B., O’Kelly, M.J.T., Zhao, X., Carter, S.L., Russ, C., Nusbaum, C., Meyerson, M., and Lander, E.S. (2009). High-resolution mapping of copy-number alterations with massively parallel sequencing. Nat. Methods 6, 99–103. Cibulskis, K., Lawrence, M.S., Carter, S.L., Sivachenko, A., Jaffe, D., Sougnez, C., Gabriel, S., Meyerson, M., Lander, E.S., and Getz, G. (2013). Sensitive detection of somatic point mutations in impure and heterogeneous cancer samples. Nat. Biotechnol. 31, 213–219. de Lau, W., Barker, N., Low, T.Y., Koo, B.-K., Li, V.S.W., Teunissen, H., Kujala, P., Haegebarth, A., Peters, P.J., van de Wetering, M., et al. (2011). Lgr5 homologues associate with Wnt receptors and mediate R-spondin signalling. Nature 476, 293–297. de Lau, W., Peng, W.C., Gros, P., and Clevers, H. (2014). The R-spondin/Lgr5/ Rnf43 module: regulator of Wnt signal strength. Genes Dev. 28, 305–316. De Roock, W., Claes, B., Bernasconi, D., De Schutter, J., Biesmans, B., Fountzilas, G., Kalogeras, K.T., Kotoula, V., Papamichael, D., Laurent-Puig, P., et al. (2010). Effects of KRAS, BRAF, NRAS, and PIK3CA mutations on the efficacy of cetuximab plus chemotherapy in chemotherapy-refractory metastatic colorectal cancer: a retrospective consortium analysis. Lancet Oncol. 11, 753–762. DePristo, M.A., Banks, E., Poplin, R., Garimella, K.V., Maguire, J.R., Hartl, C., Philippakis, A.A., del Angel, G., Rivas, M.A., Hanna, M., et al. (2011). A framework for variation discovery and genotyping using next-generation DNA sequencing data. Nat. Genet. 43, 491–498. Di Nicolantonio, F., Martini, M., Molinari, F., Sartore-Bianchi, A., Arena, S., Saletti, P., De Dosso, S., Mazzucchelli, L., Frattini, M., Siena, S., and Bardelli, A. (2008). Wild-type BRAF is required for response to panitumumab or cetuximab in metastatic colorectal cancer. J. Clin. Oncol. 26, 5705–5712. Eirew, P., Steif, A., Khattra, J., Ha, G., Yap, D., Farahani, H., Gelmon, K., Chia, S., Mar, C., Wan, A., et al. (2015). Dynamics of genomic clones in breast cancer patient xenografts at single-cell resolution. Nature 518, 422–426. Fearon, E.R. (2011). Molecular genetics of colorectal cancer. Annu. Rev. Pathol. 6, 479–507. Fearon, E.R., and Vogelstein, B. (1990). A genetic model for colorectal tumorigenesis. Cell 61, 759–767. Fisher, S., Barry, A., Abreu, J., Minie, B., Nolan, J., Delorey, T.M., Young, G., Fennell, T.J., Allen, A., Ambrogio, L., et al. (2011). A scalable, fully automated process for construction of sequence-ready human exome targeted capture libraries. Genome Biol. 12, R1. Futreal, P.A., Coin, L., Marshall, M., Down, T., Hubbard, T., Wooster, R., Rahman, N., and Stratton, M.R. (2004). A census of human cancer genes. Nat. Rev. Cancer 4, 177–183. Garnett, M.J., Edelman, E.J., Heidorn, S.J., Greenman, C.D., Dastur, A., Lau, K.W., Greninger, P., Thompson, I.R., Luo, X., Soares, J., et al. (2012). Systematic identification of genomic markers of drug sensitivity in cancer cells. Nature 483, 570–575. Garraway, L.A., and Lander, E.S. (2013). Lessons from the cancer genome. Cell 153, 17–37. Giannakis, M., Hodis, E., Jasmine Mu, X., Yamauchi, M., Rosenbluh, J., Cibulskis, K., Saksena, G., Lawrence, M.S., Qian, Z.R., Nishihara, R., et al. (2014). RNF43 is frequently mutated in colorectal and endometrial cancers. Nat. Genet. 46, 1264–1266.

Hao, H.-X., Xie, Y., Zhang, Y., Charlat, O., Oster, E., Avello, M., Lei, H., Mickanin, C., Liu, D., Ruffner, H., et al. (2012). ZNRF3 promotes Wnt receptor turnover in an R-spondin-sensitive manner. Nature 485, 195–200. Haramis, A.-P.G., Begthel, H., van den Born, M., van Es, J., Jonkheer, S., Offerhaus, G.J.A., and Clevers, H. (2004). De novo crypt formation and juvenile polyposis on BMP inhibition in mouse intestine. Science 303, 1684–1686. Imielinski, M., Berger, A.H., Hammerman, P.S., Hernandez, B., Pugh, T.J., Hodis, E., Cho, J., Suh, J., Capelletti, M., Sivachenko, A., et al. (2012). Mapping the hallmarks of lung adenocarcinoma with massively parallel sequencing. Cell 150, 1107–1120. Ivanov, I., Lo, K.C., Hawthorn, L., Cowell, J.K., and Ionov, Y. (2007). Identifying candidate colon cancer tumor suppressor genes using inhibition of nonsensemediated mRNA decay in colon cancer cells. Oncogene 26, 2873–2884. Jiao, Y., Yonescu, R., Offerhaus, G.J.A., Klimstra, D.S., Maitra, A., Eshleman, J.R., Herman, J.G., Poh, W., Pelosof, L., Wolfgang, C.L., et al. (2014). Whole-exome sequencing of pancreatic neoplasms with acinar differentiation. J. Pathol. 232, 428–435. Jin, K., Teng, L., Shen, Y., He, K., Xu, Z., and Li, G. (2010). Patient-derived human tumour tissue xenografts in immunodeficient mice: a systematic review. Clin. Transl. Oncol. 12, 473–480. Kadowaki, T., Wilder, E., Klingensmith, J., Zachary, K., and Perrimon, N. (1996). The segment polarity gene porcupine encodes a putative multitransmembrane protein involved in Wingless processing. Genes Dev. 10, 3116– 3128. Kim, K.-A., Kakitani, M., Zhao, J., Oshima, T., Tang, T., Binnerts, M., Liu, Y., Boyle, B., Park, E., Emtage, P., et al. (2005). Mitogenic influence of human R-spondin1 on the intestinal epithelium. Science 309, 1256–1259. Koo, B.-K., Spit, M., Jordens, I., Low, T.Y., Stange, D.E., van de Wetering, M., van Es, J.H., Mohammed, S., Heck, A.J.R., Maurice, M.M., and Clevers, H. (2012). Tumour suppressor RNF43 is a stem-cell E3 ligase that induces endocytosis of Wnt receptors. Nature 488, 665–669. Korinek, V., Barker, N., Moerer, P., van Donselaar, E., Huls, G., Peters, P.J., and Clevers, H. (1998). Depletion of epithelial stem-cell compartments in the small intestine of mice lacking Tcf-4. Nat. Genet. 19, 379–383. Kuhnert, F., Davis, C.R., Wang, H.-T., Chu, P., Lee, M., Yuan, J., Nusse, R., and Kuo, C.J. (2004). Essential requirement for Wnt signaling in proliferation of adult small intestine and colon revealed by adenoviral expression of Dickkopf-1. Proc. Natl. Acad. Sci. USA 101, 266–271. Landau, D.A., Carter, S.L., Stojanov, P., McKenna, A., Stevenson, K., Lawrence, M.S., Sougnez, C., Stewart, C., Sivachenko, A., Wang, L., et al. (2013). Evolution and impact of subclonal mutations in chronic lymphocytic leukemia. Cell 152, 714–726. Lawrence, M.S., Stojanov, P., Mermel, C.H., Robinson, J.T., Garraway, L.A., Golub, T.R., Meyerson, M., Gabriel, S.B., Lander, E.S., and Getz, G. (2014). Discovery and saturation analysis of cancer genes across 21 tumour types. Nature 505, 495–501.

Ojesina, A.I., Lichtenstein, L., Freeman, S.S., Pedamallu, C.S., Imaz-Rosshandler, I., Pugh, T.J., Cherniack, A.D., Ambrogio, L., Cibulskis, K., Bertelsen, B., et al. (2014). Landscape of genomic alterations in cervical carcinomas. Nature 506, 371–375. Pinto, D., Gregorieff, A., Begthel, H., and Clevers, H. (2003). Canonical Wnt signals are essential for homeostasis of the intestinal epithelium. Genes Dev. 17, 1709–1713. Ryland, G.L., Hunter, S.M., Doyle, M.A., Rowley, S.M., Christie, M., Allan, P.E., Bowtell, D.D.L., Gorringe, K.L., and Campbell, I.G.; Australian Ovarian Cancer Study Group (2013). RNF43 is a tumour suppressor gene mutated in mucinous tumours of the ovary. J. Pathol. 229, 469–476. Sadanandam, A., Lyssiotis, C.A., Homicsko, K., Collisson, E.A., Gibb, W.J., Wullschleger, S., Ostos, L.C.G., Lannon, W.A., Grotzinger, C., Del Rio, M., et al. (2013). A colorectal cancer classification system that associates cellular phenotype and responses to therapy. Nat. Med. 19, 619–625. Sato, T., Vries, R.G., Snippert, H.J., van de Wetering, M., Barker, N., Stange, D.E., van Es, J.H., Abo, A., Kujala, P., Peters, P.J., and Clevers, H. (2009). Single Lgr5 stem cells build crypt-villus structures in vitro without a mesenchymal niche. Nature 459, 262–265. Sato, T., Stange, D.E., Ferrante, M., Vries, R.G.J., Van Es, J.H., Van den Brink, S., Van Houdt, W.J., Pronk, A., Van Gorp, J., Siersema, P.D., and Clevers, H. (2011). Long-term expansion of epithelial organoids from human colon, adenoma, adenocarcinoma, and Barrett’s epithelium. Gastroenterology 141, 1762–1772. Seshagiri, S., Stawiski, E.W., Durinck, S., Modrusan, Z., Storm, E.E., Conboy, C.B., Chaudhuri, S., Guan, Y., Janakiraman, V., Jaiswal, B.S., et al. (2012). Recurrent R-spondin fusions in colon cancer. Nature 488, 660–664. Takahashi, M., Fujita, M., Furukawa, Y., Hamamoto, R., Shimokawa, T., Miwa, N., Ogawa, M., and Nakamura, Y. (2002). Isolation of a novel human gene, APCDD1, as a direct target of the beta-Catenin/T-cell factor 4 complex with probable involvement in colorectal carcinogenesis. Cancer Res. 62, 5651–5656. Tentler, J.J., Tan, A.C., Weekes, C.D., Jimeno, A., Leong, S., Pitts, T.M., Arcaroli, J.J., Messersmith, W.A., and Eckhardt, S.G. (2012). Patient-derived tumour xenografts as models for oncology drug development. Nat. Rev. Clin. Oncol. 9, 338–350. Vassilev, L.T., Vu, B.T., Graves, B., Carvajal, D., Podlaski, F., Filipovic, Z., Kong, N., Kammlott, U., Lukacs, C., Klein, C., et al. (2004). In vivo activation of the p53 pathway by small-molecule antagonists of MDM2. Science 303, 844–848. Vecchione, L. (2014). Optimization of anti-EGFR treatment of advanced colorectal cancer. Curr. Colorectal Cancer Rep. 10, 263–271. Vogelstein, B., Papadopoulos, N., Velculescu, V.E., Zhou, S., Diaz, L.A., Jr., and Kinzler, K.W. (2013). Cancer genome landscapes. Science 339, 1546– 1558.

Lengauer, C., Kinzler, K.W., and Vogelstein, B. (1997). Genetic instability in colorectal cancers. Nature 386, 623–627.

Wang, K., Yuen, S.T., Xu, J., Lee, S.P., Yan, H.H.N., Shi, S.T., Siu, H.C., Deng, S., Chu, K.M., Law, S., et al. (2014). Whole-genome sequencing and comprehensive molecular profiling identify new driver mutations in gastric cancer. Nat. Genet. 46, 573–582.

Liu, X., Ory, V., Chapman, S., Yuan, H., Albanese, C., Kallakury, B., Timofeeva, O.A., Nealon, C., Dakic, A., Simic, V., et al. (2012). ROCK inhibitor and feeder cells induce the conditional reprogramming of epithelial cells. Am. J. Pathol. 180, 599–607.

Wong, V.W.Y., Stange, D.E., Page, M.E., Buczacki, S., Wabik, A., Itami, S., van de Wetering, M., Poulsom, R., Wright, N.A., Trotter, M.W.B., et al. (2012). Lrig1 controls intestinal stem-cell homeostasis by negative regulation of ErbB signalling. Nat. Cell Biol. 14, 401–408.

Cell 161, 933–945, May 7, 2015 ª2015 Elsevier Inc. 945

Cell Stem Cell

Resource Human Pancreatic Tumor Organoids Reveal Loss of Stem Cell Niche Factor Dependence during Disease Progression Takashi Seino,1 Shintaro Kawasaki,1 Mariko Shimokawa,1 Hiroki Tamagawa,1 Kohta Toshimitsu,1 Masayuki Fujii,1 Yuki Ohta,1 Mami Matano,1 Kosaku Nanki,1 Kenta Kawasaki,1 Sirirat Takahashi,1 Shinya Sugimoto,1 Eisuke Iwasaki,1 Junichi Takagi,3 Takao Itoi,4 Minoru Kitago,2 Yuko Kitagawa,2 Takanori Kanai,1 and Toshiro Sato1,5,* 1Department

of Gastroenterology, Keio University School of Medicine, Tokyo 160-8582, Japan of Surgery, Keio University School of Medicine, Tokyo 160-8582, Japan 3Laboratory of Protein Synthesis and Expression, Institute for Protein Research, Osaka University, Suita 565-0871, Japan 4Department of Gastroenterology, Tokyo Medical University, Tokyo 160-0023, Japan 5Lead Contact *Correspondence: t.sato@keio.jp https://doi.org/10.1016/j.stem.2017.12.009 2Department

SUMMARY

Despite recent efforts to dissect the inter-tumor heterogeneity of pancreatic ductal adenocarcinoma (PDAC) by determining prognosis-predictive gene expression signatures for specific subtypes, their functional differences remain elusive. Here, we established a pancreatic tumor organoid library encompassing 39 patient-derived PDACs and identified 3 functional subtypes based on their stem cell niche factor dependencies on Wnt and R-spondin. A Wnt-non-producing subtype required Wnt from cancer-associated fibroblasts, whereas a Wntproducing subtype autonomously secreted Wnt ligands and an R-spondin-independent subtype grew in the absence of Wnt and R-spondin. Transcriptome analysis of PDAC organoids revealed gene-expression signatures that associated Wnt niche subtypes with GATA6-dependent gene expression subtypes, which were functionally supported by genetic perturbation of GATA6. Furthermore, CRISPR-Cas9-based genome editing of PDAC driver genes (KRAS, CDKN2A, SMAD4, and TP53) demonstrated non-genetic acquisition of Wnt niche independence during pancreas tumorigenesis. Collectively, our results reveal functional heterogeneity of Wnt niche independency in PDAC that is non-genetically formed through tumor progression.

INTRODUCTION Pancreatic ductal adenocarcinoma (PDAC) is a devastating disease that has an extremely poor prognosis, with a median survival of <1 year and a 5-year overall survival rate of <9% (National Cancer Institute). Whereas PDACs are generally chemo-resistant, a fraction of patients benefit from current ther454 Cell Stem Cell 22, 454â&#x20AC;&#x201C;467, March 1, 2018 ÂŞ 2017 Elsevier Inc.

apeutics, underscoring the importance of understanding interpatient tumor heterogeneity in depth and stratifying PDACs to predict clinical behaviors. Recent gene-expression-based classification identified 2 major prognosis-predicting molecular subtypes in PDAC, namely the classical and quasi-mesenchymal (QM) subtypes (Bailey et al., 2016; Collisson et al., 2011; Moffitt et al., 2015; Noll et al., 2016). The classical subtype was characterized by differentiated duct cell marker expression and favorable prognosis, whereas the QM subtype was characterized by aggressive clinical behavior and by gene silencing of definitive endoderm specification genes, including GATA6, FOXA2, and HNF4A. Similar QM-like subtypes regulated by GATA6 expression were also reported as squamous (Bailey et al., 2016) and basal-like (Moffitt et al., 2015) subtypes. Despite the robust identification of gene expression subtypes, whether these subtypes reflect genetically distinct cell-of-origin or tumor-progression statuses has remained elusive, owing to a paucity of functional assay systems for human PDAC. Functional analyses of PDAC have mainly relied on genetically engineered mice, cell lines, and patient-derived tumor xenograft models. The genetic tractability of mouse models and cell lines has contributed to the understanding of pancreas tumorigenesis, yet its relevance to the clinical traits of human PDAC, including histological and gene expression subtypes, remains unknown (Hwang et al., 2016; Hruban et al., 2006). Xenograft models can be efficiently generated from clinical samples with preserved histological and molecular subtypes, whereas the labor-intensive and genetically intractable natures of such models have limited large-scale analyses and prospective genetic approaches for PDAC. The organoid culture system has recently emerged as a technology that can propagate epithelial tissues as 3D structures using artificial stem cell niche environments (Sato et al., 2009). A defined niche factor combination of R-spondin, epidermal growth factor (EGF), fibroblast growth factor 10 (FGF10), and Noggin (a BMP4 inhibitor) promoted the expansion of normal mouse pancreatic duct cells (Huch et al., 2013) and was later applied to patient-derived PDAC organoids (Boj et al., 2015). To date, a dozen lines of PDAC organoids have been generated in which the preservation of histological

Figure 1. Establishment of the PTOL Using Niche-Based Selection (A) Overview of the procedure used to develop the PTOL. Pancreatic tumor organoids and CAFs were established from surgically resected-, FNA-, or punctured ascites specimens. (B) Strategy to enrich PDAC organoids. Tumor organoids were selected using the indicated culture conditions. Tumor organoids were judged as originating from PDAC based on the retention of growth capacity under at least one of the indicated selection conditions. (C) Representative images before and after eliminating contaminating normal organoids by EGF removal. Pancreatic tumors (top) occasionally include both normal (black arrowheads) and PDAC organoids (white arrowheads), and PDAC organoids were selectively expanded in culture in the absence of EGF (bottom). Of note, this selection medium did not include Nutlin3 or BMP4. (D) Niche factor requirement (top), genetic mutations (middle), and chromosomal alterations (bottom) of PDAC and normal-like organoids. The mutations are indicated as follows: protein-damaging mutations or oncogenic mutations (black); deletions (light blue); and wild-type (light gray). Squares and triangles indicate homozygosity and heterozygosity, respectively. Gene alterations with heterozygous deletion-only patterns are omitted (specified later in Figure S2). Hotspot missense mutations in KRAS (codons 12, 13, and 61) and GNAS (codon 201) are demonstrated. *KRAS mutation was not detected in the whole-exome sequencing due to a low coverage but confirmed by Sanger sequencing. The bar plots on the right depict the alteration frequency of each gene in the PTOL (black) and the TCGA database (white; data derived from cBioPortal). Copy number alteration status is shown: gain (red) and loss (blue). The overall chromosomal status of each sample is shown in the right panel. See also Figure S1 and Tables S1, S2, and S5.

traits and genetic-mutation profiles of the parental tumors were confirmed. Despite these advances, large-scale transcriptome analysis of human PDAC organoids has not been conducted, posing a bottleneck in connecting biological behavior with gene expression subtypes (Boj et al., 2015; Huang et al., 2015). In this study, by refining organoid culture conditions, we established 39 lines of PDAC organoids and performed comprehensive molecular characterization, which illuminated different modes of Wnt/R-spondin niche dependency in association with gene expression subtypes. Furthermore, CRISPR-Cas9-mediated engineering of pancreas organoids demonstrated the stepwise tumorigenesis of PDAC with progressive acquisition of niche independency.

RESULTS Establishment of a Human Pancreatic Tumor Organoid Library Using surgical, fine-needle aspiration (FNA) and ascites specimens, we established organoids from pancreatic tumors as well as normal organoids when adjacent normal pancreatic tissues were available (Figure 1A). Using previously published culture conditions (Huch et al., 2013), normal pancreas duct organoids ceased proliferation within 2 or 3 months (Boj et al., 2015). We found that the replacement of serum-stabilized Wnt3A-conditioned medium with serum-free Afamin-stabilized Wnt3A (Mihara et al., 2016) enabled stable culture for over Cell Stem Cell 22, 454â&#x20AC;&#x201C;467, March 1, 2018 455

8 months. Notably, the addition of serum triggered senescence, suggesting that serum in standard Wnt3A-conditioned medium is detrimental for the long-term maintenance of human pancreas organoids (Figure S1A). Using this culture protocol, we established a pancreatic tumor organoid library (PTOL) consisting of 49 organoid lines generated from patient-derived pancreatic tumors (Table S1). In our initial experiments, we often observed the outgrowth of normal pancreas organoids from PDAC specimens. The selective growth of ‘‘contaminating’’ normal organoids has been previously reported during the establishment of PDAC and prostate cancer organoids (Boj et al., 2015; Gao et al., 2014). Considering the high prevalence of KRAS mutations that occur among PDACs, the organoids were first placed in an EGF-depleted condition to enrich the KRAS mutant organoids (Figure 1B). In contrast to the standard Wnt-conditioned medium that could activate EGFR and other cognate receptors by serum-derived growth factors (Wang et al., 2013), EGF removal from the serum-free medium efficiently selected KRAS mutant organoids. The efficient enrichment of PDAC organoids can be readily visualized when spherical PDAC organoids grew along with normal cystic organoids; after EGF-based selection, KRAS mutant spherical PDAC organoids dominated whereas cystic organoids with wild-type KRAS disappeared (Figures 1C, S1B, and S1C). KRAS mutant spherical organoids, but not normal cystic organoids, displayed phospho-ERK expression in the absence of EGF, confirming the ligand-independent activation of Ras signaling in PDAC organoids (Figure S1D). Determination of the Cancer Origin in Established PDAC Organoids Organoids that were susceptible to EGF removal were alternatively treated with Nutlin3 (an MDM2 inhibitor) or with Noggin removal/BMP4 to select potentially existing TP53 or SMAD4 mutant organoids, respectively (Figure 1B). Together with the aforementioned EGF-based selection, the niche-based selection was used to diagnose 39 out of 49 organoids in the PTOL as PDAC organoids. The remaining 10 organoids exhibited strict niche dependencies indistinguishable from those of normal pancreas organoids and were referred to as normal-like (NL) organoids. To determine the accuracy of niche-factor-based diagnosis, we analyzed the genetic status of the established organoids using whole-exome sequencing and comparative genomic-hybridization microarray analyses. Consistent with previous large-scale deep-sequencing analyses (Waddell et al., 2015), PDAC organoids harbored common driver-gene alterations at the expected frequencies: KRAS (36/38); CDKN2A (31/38); TP53 (29/38); and SMAD4 (14/38; Table S2). GNAS hotspot mutations (GNASR201H) were detected in two organoid lines (2/49), one of which was derived from intraductal papillary mucinous neoplasm (IPMN). Of note, no authenticated IPMN cell line with a GNAS mutation has been derived to date (Furukawa et al., 2011). We did not detect recurrent driver-gene mutations in normal-like organoids, corroborating the accurate selection of PDAC organoids (Figure 1D). Whereas detailed copy number analyses of PDACs have been hampered by the low tumor content of clinical PDAC samples (Shain et al., 2012; Witkiewicz et al., 2015), the purely epithelial composition of PDAC organoids enabled accurate assessment 456 Cell Stem Cell 22, 454–467, March 1, 2018

of copy number alterations. NL organoids invariably showed euploidy, reinforcing their non-cancer origins, whereas most PDAC organoids acquired well-defined chromosomal alterations, namely losses of 6p, 9p, 17p, and 18q (Su et al., 1998). Of note, four PDAC lines exhibited euploidy or near euploidy, of which three harbored KRAS mutations, confirming their cancer origin (Figure 1D). The remaining near-euploidy line (PC8) was devoid of driver-gene mutations, including those in KRAS, and its tumorigenic mechanism was unclear. Upon xenografting into orthotopic pancreata of NOD.CgPrkdcscidIl2rgtm1Sug/Jic (NOG) mice, PDAC organoids (including the wild-type KRAS organoid) formed tumors resembling their parental PDACs, whereas NL organoids did not successfully engraft (Figure S1F). Taken together, these data indicate that niche-based selection efficiently propagated PDAC organoids with the accurate exclusion of potentially contaminating normal organoids. Identification of Three Functional PDAC Organoid Subtypes with Distinct Wnt Niche Requirements Once PDAC organoids were established, each organoid was subjected to successive niche-based treatments to determine minimally essential niche factors, which demonstrated that driver-gene alterations largely dictated the requirements for the corresponding niche factors. Specifically, sensitivity to EGF removal, Noggin removal/BMP4 treatment, A83-01 removal/ transforming growth factor b1 (TGF-b1) treatment, and Nutlin3 treatment were associated with KRAS, SMAD4, TGFBR2, and TP53 mutations/in-del alterations, respectively (Figures S2A– S2H). In contrast to these mutation-driven adaptations, we noted that Wnt/R-spondin dependency was mostly unrelated to Wntsignaling mutations in PDAC organoids (Figures 2A and 2B). Akin to the niche dependency of normal pancreas organoids, 14 PDAC organoids required both Wnt3A and R-spondin for their growth. Interestingly, the remaining R-spondin-dependent PDAC organoids grew in the absence of Wnt3A. Because R-spondin is known to potentiate Wnt signaling through stabilization of Wnt receptors (Koo et al., 2012), this phenotype suggested that R-spondin-dependent PDAC organoids harness either exogenously supplied or endogenously produced Wnt ligands. Therefore, we divided R-spondin-dependent PDAC organoids into two subtypes, namely Wnt-non-secreting (W ) and Wnt-secreting (W+) PDAC organoids, based on their requirements for exogenous Wnt3A. To validate the Wnt-producing capacity of W+ PDAC organoids, we tested the effect of a porcupine inhibitor (Porcn-i; C59) (Proffitt et al., 2013), which abrogates the production of biologically active Wnt ligands. Notably, Porcn-i treatment suppressed the growth of W+ PDAC organoids in parallel with the reduction of their Wnt target gene-expression levels, and these effects were reversed by the supplementation with exogenous Wnt3A (Figures 2C and 2D). Furthermore, W+ PDAC organoids exhibited higher expression of Wnt target genes than W PDAC organoids in the absence of Wnt3A (Figure S3A). The potent niche function of Wnt ligands secreted from W+ PDAC organoids was further validated by their growth-promoting effects on co-cultured W PDAC organoids (Figure S3B). These results collectively indicated that W+ PDAC organoids autonomously create their own Wnt niche.

Figure 2. PDAC Organoids Exhibit Distinct Wnt Niche Requirements (A) Representative images depicting Wnt and R-spondin dependency of PDAC organoids. (B) R-spondin/Wnt dependency profiles (green) and Wnt pathway mutations (black) in PDAC organoids and NL organoids. Triangles indicate heterozygosity. Bars on the top represent organoid subtypes as follows: gray, normal like (NL); pink, W ; red, W+; and purple, WRi. (C) The potent effect of a Porcn-i on W+ PDAC organoid growth. Supplementation of exogenous Wnt3A blocks this negative effect. (D) qPCR analyses of Wnt target genes (LGR5 and AXIN2) in 3 Wnt-independent PDAC organoid lines cultured under the indicated conditions (n = 3 per each line per for each condition). mRNA expression levels relative to b-actin (log10 value) are shown. Upon Porcn-i treatment, LGR5 and AXIN2 mRNA expression levels were significantly reduced (p < 0.05; Student’s t test). (E) Schematic representation of the proposed Wnt niche subtypes. NL and Wnt-non-producing (W ) organoids require for exogenous Wnt and R-spondin ligand; Wnt-producing organoids are independent of exogenous Wnt ligand but rely on R-spondin; and Wnt and R-spondin-independent organoids (WRi) have no requirement for Wnt signal activation. (A and C) The values shown indicate organoid areas relative to the optimal growth conditions (the mean ± SEM; n = 4 per each line for each condition). See also Figure S3 and Tables S2, S5, and S6.

We next characterized six Wnt and R-spondin-independent (WRi) PDAC organoids. To determine whether WRi PDAC organoids require Wnt-signal activation for their growth, we tested the effect of ICG001, a small-molecule inhibitor that blocks downstream Wnt-b-catenin signaling (Emami et al., 2004). Upon ICG001 treatment, both W and W+ PDAC organoids irreversibly terminated their proliferation in line with their require-

ment for Wnt-signal activation. In contrast, the growth of WRi PDAC organoids was maintained, suggesting that Wnt-signal activation itself was not essential for maintaining these organoids (Figure S3C). Though WRi PDAC organoids also tolerated Porcn-i treatment corroborating their dispensability of Wntsignal activation, some WRi PDAC organoids were responsive to the Porcn-i treatment, suggesting their partial dependency Cell Stem Cell 22, 454–467, March 1, 2018 457

on self-producing Wnt ligands (Figure S3D). In sum, our results revealed 3 functional subtypes of PDAC organoids that displayed unique requirements for Wnt and R-spondin niche environments (Figure 2E). Cancer-Associated Fibroblasts Provide Wnt Ligands for PDACs W PDAC cells critically depend on exogenous Wnt ligands for their survival and growth, yet the source of the Wnt ligands remains obscure. Because PDAC is characterized by abundant stromal cell infiltration, we inferred that these stromal cells could support the growth of PDACs through Wnt production. To determine the functional role of stromal Wnt ligands, we established patient-derived cancer-associated fibroblasts (CAFs) (Figure 1A). Immunostaining of fibroblast activation protein alpha (FAP) and a smooth muscle actin (aSMA), characteristic markers for CAFs in PDAC (Öhlund et al., 2017), confirmed that the established CAFs were of stromal origin (Figure S4A). In contrast to the quiescence induction in CAFs embedded in Matrigel (Öhlund et al., 2017), CAFs in a collagen type I Matrigel mixture showed proliferation potency. To investigate whether these CAFs can functionally support the growth of W PDAC organoids, we generated single stroma-attached organoids by aggregating dissociated PDAC cells and CAFs (Figures 3A and 3B). Interestingly, this physical stroma attachment enabled W PDAC organoids to grow without exogenous Wnt3A, and as expected, Porcn-i treatment abrogated this growth-promoting effect (Figures 3C and 3D). Consistent with the short-range Wnt gradient in intestinal organoids (Farin et al., 2016), this growth-promoting effect on W PDAC organoids was not observed in conditioned medium from CAFs or when CAFs and W PDAC organoids were co-cultured without physical attachment (Figure S4B). These results demonstrated that the juxtacrine interaction with PDAC cells was critical for CAFs to support the growth of W PDACs. To further validate the pro-tumorigenic effects of CAFs on PDAC organoids in vivo, PDAC organoids were subcutaneously transplanted, either alone or with CAFs. In the absence of CAFs, W+ and WRi PDAC organoids efficiently engrafted, whereas two out of the three examined W PDAC organoid lines were poorly tumorigenic, suggesting the requirement for Wnt niche during tumor formation. Interestingly, when interfaced with CAFs prior to transplantation, these organoids successfully formed subcutaneous tumors (Figures 3E and 3F). We observed eventual replacement of transplanted CAFs with host-derived fibroblasts (Figure S4C), suggesting that the effect of co-transplanted CAFs was limited to the initial phase of xenotransplantation. Indeed, co-transplantation with CAFs increased the engraftment rate of W PDAC organoids but did not enhance the tumorigenic growth of xenograft-competent organoids (Figure 3F). To investigate whether the pro-tumorigenic effect of CAFs was mediated by their stromal Wnt production, we next treated xenografts with a Porcn-i (Figure S4D). Whereas the therapeutic effect of Porcn-i has only been observed in RNF43 mutant PDAC cell lines in previous studies (Jiang et al., 2013), Porcn-i treatment significantly reduced the growth of xenografts from two independent RNF43-wild-type W PDAC organoids (Figure S4E). Conversely, Porcn-i treatment did not affect the growth of WRi PDAC organoids in vivo. These results demonstrated that the CAF-dependent growth of PDACs was driven by stromal Wnt 458 Cell Stem Cell 22, 454–467, March 1, 2018

niche environments and that Wnt-targeting therapeutics could be potent against wild-type W PDACs, regardless of the RNF43 mutation status. Epithelial Wnt Ligand Expression Is Associated with the Clinical Progression of PDACs In contrast to W PDAC organoids, W+ PDAC organoids harnessed their self-produced Wnt ligands. To determine which Wnt ligands were expressed in PDAC organoids, their expression levels were assessed by microarray analyses. An unbiased criterion was set to select Wnt ligands that were expressed at significant levels in at least one W+ PDAC organoid line (>5 SD expression from the mean expression levels of NL organoids; Figure 4A), which nominated 6 Wnt ligands from 19 Wnt-ligand family members. To determine whether any of these Wnt ligands could serve as an epithelial Wnt niche, we overexpressed each Wnt ligand in NL organoids and examined their potential for driving niche function. In this functional assay, 4 Wnt ligands (WNT3, WNT7A, WNT7B, and WNT10A) were found to substitute for Wnt3A and, thus, were designated as ‘‘epithelial’’ Wnt ligands (Figures 4A and S5A). Importantly, the expression levels of these epithelial Wnt genes were significantly higher in WRi and W+ PDAC organoids than in W PDAC organoids, indicating the association between epithelial Wnt-ligand expression and Wnt niche independency (Figure 4B). To determine whether epithelial Wnt ligands are expressed in clinical specimens, we analyzed their expression levels in whole PDAC tissues using a publicly available transcriptome dataset. Hierarchical clustering of Wnt-ligand gene-expression levels aggregated the epithelial Wnt-ligand genes in a single cluster (Figure S5B). WNT2, WNT2B, WNT4, and WNT5A were expressed in PDAC tissues but were rarely detected in organoids, suggesting their stromal origins. Real-time qPCR analyses of Wnt ligand gene-expression levels in organoids and CAFs confirmed these tissue-specific expression patterns (Figure 4C). Of note, the functional assay revealed that only WNT2 and WNT2B served as potent niche factors among stromal Wnt ligands (Figure 4A). To determine the differential expression of epithelial Wnts in patients, we next performed in situ hybridization of the clinical specimens for the epithelial Wnt genes. WNT7B and WNT10A were markedly expressed in the epithelial component of W+ PDACs, whereas no or subtle expression of these genes was observed in W PDACs and adjacent normal pancreas tissues (Figure 4D). Stromal expression of WNT2B was detected in close proximity to PDAC tissue (Figure 4D), consistent with the shortrange activity of Wnt ligands. These results suggested that the expression of epithelial Wnts could serve as a surrogate marker to define Wnt-producing PDACs. In addition, the high expression of epithelial Wnts in clinical PDACs was associated with significantly poor survival and metastatic progression (Figures 4E and S5C). These results demonstrated that W+ PDACs cell autonomously activated their own Wnt signaling by expressing epithelial Wnts, which also predicts aggressive clinical behaviors. Wnt Niche Dependency Is Associated with Gene Expression Subtypes To explore the mechanisms underlying each functional PDAC subtype, we performed transcriptome analyses of the PTOL.

Figure 3. CAFs Support W– PDAC Growth via Wnt Ligand Secretion (A) Overview of the PDAC organoid-CAF co-culture system. (B) Time course images of a hybrid organoid during formation. (C and D) EdU incorporation (red) in PDAC organoids (PC8) cultured in the above protocol. EdU was pulsed 1 hr before the analyses. The aggregated organoids without CAFs ceased their proliferation in the absence of Wnt3A supplementation (C). When co-cultured with CAFs, PDAC organoids started to incorporate EdU. EdU incorporation was decreased by Porcn-i treatment and was maintained by exogenous Wnt3A supplementation in W PDAC organoids (PC8) (D). Similar result was obtained with another line of W PDAC organoids (PC11). The scale bar represents 100 mm. (E) Schematic representation of the transplantation of a co-cultured organoid (left). Representative images of co-cultured organoids (PC11) xenotransplanted into the subcutaneous space of nude mice (right) are shown. (Top) bright field is shown; (bottom) GFP fluorescence is shown. (F) Tumor volumes of the organoids transplanted with or without CAFs. W , W+, and WRi PDAC organoids were subcutaneously xenotransplanted, either alone or after CAF attachment. Tumor sizes were calculated from GFP-positive regions. n = 6–12 per each group. Each dot indicates the size of an individual tumor. The take rates of xenografts are indicated below. Bars represent mean volumes of the grafts. *p < 0.05; **p < 0.01; N.S., not significant; determined by Wilcoxon’s rank sum test. See also Figure S4.

An unbiased projection of global gene expression with t-distributed stochastic neighbor embedding (tSNE) analysis illustrated linearly connected gene expression clusters corresponding to the Wnt niche subtypes (Figure 5A). Notably, the linear trajectory

was directed from normal organoids toward W , W+, and WRi subtypes, suggesting serial transition of gene expression signature in line with acquisition of Wnt niche independency. To gain insights into the transcriptional programs regulating this Cell Stem Cell 22, 454–467, March 1, 2018 459

Figure 4. Expression of Epithelial Wnts Defines the Wnt-Producing PDAC Subtype (A) Identification of the potent epithelial Wnt ligands. Dot plots depicting expression levels of Wnt ligands in PDAC and NL organoids. Dots represent individual organoids (gray, NL; pink, W PDAC; red, W+ PDAC; purple, WRi PDAC). The dashed line indicates the threshold of 5 SD used to determine differentially expressed Wnt ligands. The niche potency of each Wnt ligand was evaluated by measuring the growth of NL organoids overexpressing each ligand (right). Ratios relative to the organoid area in the presence of Wnt3A are demonstrated as bars. NE, not examined. (B) Boxplot showing the epithelial Wnt score in NL samples and the indicated PDAC subtypes. *p < 0.05; Studentâ&#x20AC;&#x2122;s t test. (C) Heatmap showing mRNA expression (Z score) of the indicated Wnt ligands, as determined by qPCR analysis. WRi (PC1), W+ (PC3) PDAC organoids, W (PC7 and PC8) PDAC organoids, and CAFs were analyzed. ND, not detected by qPCR. The raw values are shown in Table S6. (D) In situ hybridization for WNT10A, WNT7B, and WNT2B in parental PDACs or normal pancreas tissues. Nuclear counter staining, hematoxylin. The scale bar represents 100 mm. Insets, magnified views. (E) Kaplan-Meier plots showing overall survival based on data deposited in the ICGC Data Portal (left) or reported by Moffitt et al. (2015; right). Tumor samples were stratified based on the epithelial Wnt score (sum of the Z score, calculated using the mean and SD of all samples with survival information). p value; log rank test. See also Figure S5 and Tables S5 and S6.

process, we generated gene sets consisting of differentially expressed genes between Wnt-dependent (NL and W ) and Wnt-independent (W+ and WRi) organoids (Table S3). Interestingly, the total expression of each gene set, Wnt-independent or dependent score, either monotonously increased or decreased along with the Wnt niche independency statuses, suggesting the presence of transcriptional programs regulating Wnt niche dependency (Figures 5B and S6A). Previous reports have demonstrated distinct molecular PDAC subtypes based on gene-expression patterns, and thus, we next sought to investigate whether the Wnt niche subtypes reflected these gene expression subtypes. When the ratio of basal/QM gene expression levels to those of classical genes were calculated, PDAC organoids showed incremental up-regulation of these indices 460 Cell Stem Cell 22, 454â&#x20AC;&#x201C;467, March 1, 2018

along with the loss of Wnt niche dependency, i.e., in the sequence of NL, W , W+, and WRi PDAC organoids (Figures 5B and S6B). These results implicated common transcriptional programs in operating both Wnt niche subtypes and gene expression subtypes in PDACs. The gene expression subtypes correlated with GATA6 expression, and a recent report (Bailey et al., 2016) showed the epigenetic silencing of GATA6 in the QM-like subtype, suggesting the pivotal role of GATA6 expression in determining the gene expression subtype. Consistently, GATA6 expression was associated with the extent of Wnt niche independency (Figure 5C). Furthermore, GATA6 expression was regulated by DNA methylation, as a methylation microarray revealed progressive DNA methylation of GATA6 in parallel with Wnt niche subtypes

Figure 5. Transcriptome Analyses of the PTOL Revealed an Association of Wnt Niche Subtypes with GATA6 Expression and the Gene Expression Subtypes (A) A tSNE plot of the PTOL showing trajectory formation along with Wnt niche independency, in the sequence of NL (gray), W (pink), W+ (red), and WRi (purple) PDAC organoids. (B) Boxplots of gene expression score (sum of Z score) depicted gradual increase of Wnt-niche-independent genes or basal to classical subtype gene signatures along with Wnt niche independency. Basal and classical gene sets were retrieved from data generated by Moffitt et al. (2015). (C) Boxplots of GATA6 expression (Z score) depicted gradual decreases of GATA6 expression along with Wnt niche independency. (D) Hierarchical clustering of differentially methylated genes among the W , W+, and WRi subtypes. Methylation values are shown as M values. The gene lists and corresponding M values are shown in Table S4. (E) Plots showing significant correlations between gene expression (log2 values) and gene methylations (M values) in indicated probes for GATA6 and FOXA2. Pearson’s correlation coefficients and adjusted p values are indicated. (B and C) *p < 0.05; **p < 0.005; determined by Student’s t test. See also Figure S6 and Table S5.

(Figures 5D and 5E; Table S4). We also observed similar trends in other endoderm lineage-specific genes, FOXA2 and HNF4A (Figures 5E and S6C). Conversely, W PDAC organoids exhibited higher methylation of WNT10A than the other subtypes (Figures 5D and S6C). These results suggested the existence of GATA6-dependent epigenetic regulation of Wnt niche independency. GATA6 Expression Regulates Epithelial Wnt Expression in W+ PDAC Organoids To investigate whether GATA6 expression levels are functionally relevant in differentiating molecular subtypes and Wnt production capacities, we further performed GATA6 short hairpin RNA (shRNA)-based knockdown (KD) and CRISPR-Cas9-based knockout (KO) experiments with W PDAC organoids. Of note, two GATA6 translational isotypes exist, namely GATA6L (long

form) and GATA6S (short form) (Brewer et al., 1999). Although GATA6L was reported to have a higher transactivation potential than GATA6S, the functional difference between these isoforms has remained unexplored in pancreas tissues (Brewer et al., 1999). As our sgRNAs targeted only GATA6L, we observed a complete loss of GATA6L, but GATA6S expression remained detectable in the KO experiments. Nevertheless, we confirmed the reduction of total GATA6 expression at the protein level in both GATA6-KD and GATA6L-KO lines (Figure 6A). Furthermore, both lines consistently showed the reduction of FOXA2, a GATA6 target gene, indicating successful GATA6 loss of function (Figure 6B). Importantly, upon the downregulation of GATA6, W PDAC organoids acquired Wnt self-activation capacity along with WNT7B upregulation (Figures 6C and 6D). The gene set enrichment analysis (GSEA) revealed a significant enrichment of the signature genes associated with the Wnt-independent Cell Stem Cell 22, 454–467, March 1, 2018 461

Figure 6. GATA6 Expression Regulated Wnt Dependency of PDAC Organoids (A) Western blotting analysis showing GATA6 protein levels in W PDAC (PC24) with GATA6L knockout (-KO) and GATA6 knockdown (-KD) organoids. The upper and the lower band depict the long and the short isoform of GATA6, respectively. The expression levels of NL (PC4) and W+ PDAC (PC3) organoids were included as references. (B and C) qPCR analysis of FOXA2 (B) and WNT7B (C) in parental W PDAC (PC24), GATA6-KD, and GATA6L-KO organoids. mRNA expression relative to b-actin is shown. n = 3 per genotype. The expression levels of W+ (PC3) and WRi (PC1) PDAC organoids were included as references. (D) GATA6-KD and GATA6L-KO enabled organoids to expand in Wnt3A-depleted culture medium. +Wnt, culture medium including Wnt3A; Wnt, culture medium without Wnt3A. The values shown indicate organoid area relative to the +Wnt condition (the mean ± SEM; n = 4 per each line for each condition). (E) Immunohistochemistry of GATA6 (top, brown) and in situ hybridization of WNT10A and WNT7B (bottom, red) in the clinical specimens of W+ PDACs (PC3 and PC14). Nuclear counterstaining, hematoxylin (blue). The scale bar represents 100 mm. (F) GATA6-high PDACs expressed higher levels of epithelial Wnt ligands (epithelial Wnt score) than GATA6-low PDACs in the ICGC (left) and GSE71729 (reported by Moffitt et al., 2015; right) datasets. The expression value was indicated as the Z score. The GATA6-high and GATA6-low statuses were determined by Z score of 0. *p < 0.005; Student’s t test. See also Figure S6 and Tables S3, S6, and S7.

subtype in GATA6-engineered organoids (Figure S6E). In contrast, the overexpression of GATA6 rendered W+ PDAC organoids dependent on exogenous Wnt ligands in conjunction with reduced WNT7B expression (Figures S6F–S6H). These results suggested the presence of a GATA6-dependent transcriptional program that regulates Wnt dependency in PDAC. We also investigated the clinical relevance of GATA6-regulated expression of Wnt ligands. In the parental specimens of two independent W+ PDAC lines, GATA6 immunostaining presented a heterogeneous pattern with predominant GATA6negative compartments and GATA6-expressing (GATA6+) subpopulations. Interestingly, these GATA6+ lesions were devoid of WNT10A or WNT7B mRNA expression, corroborating the negative regulation of Wnt ligands by GATA6 in clinical specimens (Figure 6E). To confirm this relationship on a larger scale, we analyzed gene expression correlations between GATA6 462 Cell Stem Cell 22, 454–467, March 1, 2018

and epithelial Wnt ligands using public datasets. Notably, GATA6 expression inversely correlated with the expression of epithelial Wnt ligands in two independent datasets (Figure 6F). These results collectively indicate the presence of GATA6-regulated Wnt niche dependency in patient PDACs. Transformation of Human Pancreatic Organoids into W– PDAC by Driver Gene Engineering The observations above demonstrated that patient-derived PDAC organoids acquired niche independency through driver gene mutations and GATA6-mediated transcriptional reprogramming. To verify this functionality in a prospective manner, we generated genetically engineered pancreas organoids using CRISPR-Cas9. Our previously reported electroporation and selective culture protocol (Matano et al., 2015) enabled efficient introduction of TP53 or SMAD4 in-del mutations to organoids

Figure 7. CRISPR-Cas9-Engineered Human Pancreas Organoids Demonstrated Mutation-Independent Adaptation to Wnt-free Environments (A) Strategy for generating CRISPR-Cas9 genomeedited organoids carrying driver gene mutations in KRAS (K), TP53 (T), CDKN2A (C), and/or SMAD4 (S). (B) KT or KCTS organoids were subcutaneously transplanted to nude mice. Representative histological images of xenografted tumors are shown. Arrows indicate tumor budding formation. The scale bar represents 100 mm. (C) Various engineered organoids were cultured with or without Wnt3A after a passage; +Wnt P1 (left) and Wnt P1 (middle), respectively. When the organoids can be passaged in the absence of Wnt3A, the image of PDAC organoids in passage 4 is shown ( Wnt P4, right). The schematic representation is shown in Figure S7G. After removing Wnt3A from culture medium, parental wild-type, KC, and KT organoids ceased their proliferation and became extinct within 1â&#x20AC;&#x201C;3 weeks. KCT and KCTS organoids stopped their proliferation but survived in the Wnt3A removed culture medium (middle). After multiple passages, KCT and KCTS organoids grew without exogenous Wnt3A (right). (D and E) qPCR analysis of WNT7B and GATA6 expression in KCT (D) and KCTS (E) organoids grown in the presence or absence of exogenous Wnt3A for four passages. (F) GSEA analysis revealed that adapted KCTS organoids (versus control KCTS organoids without adaptation) were positively and negatively enriched for Wnt-independent and dependent gene signatures, respectively. (G) Schematic representation of PDAC progression. GATA6 expression regulated both Wnt niche subtype and gene expression (GE) subtype. EGF, Noggin, and A83-01 (ENA) niche requirements were regulated by driver gene mutations, whereas Wnt and R-spondin (WR) niche dependency was associated with GATA6 expression. *p < 0.05; **p < 0.01; Studentâ&#x20AC;&#x2122;s t test. See also Figure S7 and Tables S3, S6, and S7.

(not shown). Nevertheless, KRASG12V mutation knockin clones could not be obtained, presumably due to the relatively low electroporation efficiency. To improve the genome-editing efficiency, we used a â&#x20AC;&#x2DC;â&#x20AC;&#x2DC;cold shockâ&#x20AC;&#x2122;â&#x20AC;&#x2122; method (Doyon et al., 2011), where organoids were cultured at 30 C for 3 days after electroporation. This method substantially improved the electroporation efficiency, enabling KRASG12V (K) mutation knockin. KRASG12V mutant organoids exhibited slow proliferation, whereas simultaneous engineering of CDKN2A (C) allowed efficient establishment of KC organoids. KCT and KCTS organoids were generated by subsequently introducing TP53 (T) and/or SMAD4 (S) mutations, respectively (Figures 7A and S7Aâ&#x20AC;&#x201C;S7D). Transcriptome analyses confirmed robust associations between

driver gene mutations and the alteration of their target gene expression in both PDAC organoids and the engineered organoids, corroborating the validity of the introduced mutations (Figure S7E). To determine the tumorigenic potential of the engineered organoids, KC, KT, and KCTS organoids were xenotransplanted into the subcutaneous space of immunodeficient mice. KC organoids failed to engraft. KT organoids formed subcutaneous tumors only when co-transplanted with CAFs and showed benign histopathology reminiscent of pancreatic intraepithelial neoplasia (PanIN) (Figure 7B). In contrast, KCTS organoids engrafted without CAFs and exhibited severe nuclear atypia, structural deformations, and tumor budding phenotypes corresponding to the PDAC histology, which indicated that the quadruple mutations induced histological transformation (Figure 7B). Of note, KCTS tumors exhibited heterogeneous histological appearance comprising PanIN-like lesions and invasive PDACs, implicating Cell Stem Cell 22, 454â&#x20AC;&#x201C;467, March 1, 2018 463

the histological progression after the acquisition of KCTS mutations (Figure 7B). The four major driver gene mutations endowed engineered organoids with mutation-specific niche independency (Figure S7F). In contrast, the engineered organoids lost proliferative potential after Wnt removal, indicating that the driver gene mutations did not confer Wnt-niche-independent growth (Figure 7C). Interestingly, 1–3 weeks after the Wnt removal, KC and KT organoids became extinct, whereas KCT and KCTS organoids survived and slowly expanded in the Wnt-removed culture condition (Figures 7C and S7G). This result suggested that the presence of CDKN2A and TP53 mutations allowed organoids to circumvent the apoptosis/senescence responses induced by the Wnt removal. Consistently, though the proportion of PDAC organoids with four-driver gene mutations was comparable among Wnt niche subtypes, TP53 mutation frequency was significantly higher in W+/WRi PDAC organoids than in W PDAC organoids, suggesting the role of TP53 mutations in Wnt niche adaptive response in patient PDACs (Figures S7H and S7I). This phenotype was dependent on endogenous Wnt production, as indicated by the sensitivity to Porcn-i treatment (Figure S7J). Furthermore, gene expression analyses showed GATA6 downregulation and acquisition of the Wnt-independent PDAC gene signature during the adaptation to a Wnt-free culture condition (Figures 7D–7F). Taken together, our results suggested that niche independency was mainly acquired through driver gene mutations, whereas the Wnt niche independency was predominantly regulated by epigenetic mechanisms, highlighting a unique nicheadaptation process during pancreas tumorigenesis (Figure 7G). DISCUSSION In this study, by optimizing the culture method, we established an organoid library comprising 39 PDAC organoid lines. The use of serum-free Afamin-stabilized Wnt3A enabled stable propagation of human normal pancreas organoids and exclusion of contaminating normal organoids. Extensive characterization of the genotype-phenotype correlation in PDAC organoids identified 3 novel Wnt niche subtypes that could not be defined by genetic mutations. Of note, two of these subtypes (the W+ and W subtypes) exhibited stringent dependency on Wnt3A and/or R-spondin, indicating that these subtypes were not previously derived as cell lines. Whereas an antecedent study showed consistent Wnt and R-spondin dependency in PDAC organoids (Boj et al., 2015), our larger scale PDAC organoid library revealed various Wnt-niche dependencies among human PDACs. The development of organoid culture systems is based on the identification of defined niche factors that fuel the self-renewal of epithelial stem cells (Sato and Clevers, 2013). Although the existence of pancreatic stem cells in the adult normal pancreas remains controversial, Wnt signal activation is essential during development and in organoid culture, underscoring the niche functions of Wnt ligands and R-spondin (Kopp et al., 2016). We previously proposed that niche factor signaling pathways coincided with recurrently dysregulated pathways in colorectal cancers (Sato and Clevers, 2013). Contrary to this notion, Wnt pathway genes were rarely mutated in PDAC despite its essentiality in niche signaling. Given the fact that b-catenin mutations retarded KRAS-mediated pancreas tumorigenesis in a genetic 464 Cell Stem Cell 22, 454–467, March 1, 2018

mouse model (Heiser et al., 2008), Wnt signal mutations might serve as tumor suppressors in the pancreas. Thus, pancreas duct cells may favor Wnt-signaling activation by ligand stimulation during pancreas tumorigenesis. The expression of Wnt ligands was observed in pancreas stromal cells, but whether and how the stromal Wnt ligands contribute to the PDAC ecosystem have not been determined (Moffitt et al., 2015). In this study, we established a PDAC organoid-CAF co-culture assay and revealed that CAFs transmit a pro-tumorigenic niche signal to PDACs through the juxtacrine production of stromal Wnt ligands. Furthermore, we also found that CAF-derived Wnt niche signaling could be targeted by Porcn-i treatment, the therapeutic effects of which have previously been observed exclusively in RNF43 mutant PDAC cell lines (Jiang et al., 2013). These results suggested the broad application of Wnt-targeting therapy for PDACs based on an organoid-centered, non-genetic screening system. Besides Wnt production, CAFs have pleiotropic effects on pancreas tumorigenesis (Apte et al., 2012; Sousa et al., 2016). Recently, a similar co-culture assay was established and used to identify inflammatory CAFs (iCAFs) that promoted tumorigenesis through inflammatory cytokine production (Öhlund et al., 2017). Notably, the Wnt-producing CAFs resided in mutual proximity to the PDAC cells, whereas the iCAFs were scattered in areas distant from PDAC cells, underscoring the distinct pro-tumorigenic functions between these CAF subpopulations. In contrast, other recent studies revealed an anti-tumorigenic role of CAFs in pancreas tumorigenesis (Özdemir et al., 2014; Rhim et al., 2014). The conflicting results have been partly reconciled by the heterotypic subtypes identified in CAFs and PDACs (Moffitt et al., 2015; Öhlund et al., 2017). Indeed, in our study, the pro-tumorigenic effect of CAFs varied depending on the Wnt niche subtypes of host PDAC organoids, supporting the context-dependent role of CAFs in pancreas tumorigenesis. The tumor heterogeneity of human PDAC was molecularly dissected by the gene expression subtypes. Interestingly, these gene expression subtypes pertained to the Wnt niche subtypes. The epigenetic regulation of GATA6 identified in both taxonomies suggested the existence of a common GATA6-dependent transcriptional program driving the subtype heterogeneity. In gene expression classification studies, the subtypes were formulated based on snapshots of the gene expression signatures derived from crude tumor samples, and thus, it has not been defined whether the identified subtypes inherently represent distinct tumorigenic pathways or transitive statuses during tumor progression. In contrast, the organoid platform is amenable to functional assays and prospective genetic engineering, by which we substantiated the GATA6-mediated subtype switching in PDAC organoids. Furthermore, KCTS organoids showed adaptive responses to Wnt-free environments, coinciding with the GATA6dependent subtype conversion. These results suggested that Wnt niche subtypes reflected dynamic reprogramming during tumor progression rather than an intrinsic property of a given tumorigenic pathway. The GATA6-mediated subtype switching was not unprecedented, as GATA6 downregulation also contributed to squamous subtype commitment and malignant progression in a lung cancer model (Cheung et al., 2013). Of note, KCTS organoids did not progress to the QM subtype and remained dependent on R-spondin, suggesting that four-driver mutations

were insufficient to confer on the engineered organoids the full spectrum of PDAC phenotypes, including WRi phenotype. Recently, transcriptional reprogramming by another endodermrelated transcription factor, FOXA1, was reported to induce metastatic progression in PDAC (Roe et al., 2017). Though how these endodermal-specific genes coordinate the niche requirements and progression of PDACs remains unexplored, the prevalent methylation of these genes in QM and WRi PDAC subtypes indicates that epigenetic regulations drive these processes. Further studies are warranted to understand the mechanistic role of these transcription factors in the malignant progression of PDAC. Decades of pathologic and genetic research on human pancreas tumorigenesis have indicated that PDAC originates from PanIN lesions (Hruban et al., 2000), but its prospective demonstration has not been achieved in human pancreas. In this study, using CRISPR-Cas9-based KO and mutation knockin, we generated isogenic KCTS organoids from normal pancreas organoids, which faithfully replicated the pancreas carcinogenic process. Recently, Lee et al. (2017) also generated similar engineered pancreas organoids using CRISPR-Cas9 system. In contrast to our KCTS organoid-derived tumors that were histologically compatible with PDAC, the pancreas organoids engineered by Lee et al. (2017) only formed tumors corresponding to PanINs. This discrepancy might be associated with the different methods used to introduce KRAS mutations (KRAS overexpression by Lee et al., 2017 versus KRAS knockin in our study). The discordant outcomes between these two genetic approaches has been previously observed in mouse and human pancreas cells, underscoring the ability of the KRAS knockin strategy in faithful PDAC disease modeling (Arena et al., 2007; Brembeck et al., 2003; Hingorani et al., 2003; Konishi et al., 2007). It should also be noted that, unlike the xenografts generated in orthotopic pancreas by Lee et al. (2017), we xenografted organoids into subcutaneous sites, in which KCTS organoids displayed tumor budding formation. Therefore, our KCTS organoids provided the first isogenic PDAC model that recapitulated the PanIN-PDAC sequence from human normal pancreas organoids. Of note, in mouse models, pancreas acinar cells have been proposed to engender PDACs through acinar-to-ductal metaplasia (Kopp et al., 2012). As the current organoid culture cannot propagate or reproduce acinarto-ductal metaplasia (Huch et al., 2013), it is premature to either exclude or assert the possibility of tumorigenesis originating directly from human pancreas acinar cells. In conclusion, we established a PTOL encompassing a panoply of patient-derived PDACs and engineered organoids, followed by comprehensive molecular and functional analyses. This resource can bridge the gap between cancer genotypes and biological phenotypes that have previously been elusive, due to a lack of genetically tractable patient-derived PDAC models. Our results provide novel insights into PDAC tumorigenesis and Wnt-based therapeutic strategies against PDACs. STAR+METHODS Detailed methods are provided in the online version of this paper and include the following: d d

KEY RESOURCES TABLE CONTACT FOR REAGENT AND RESOURCE SHARING

d d

EXPERIMENTAL MODEL AND SUBJECT DETAILS B Establishment of human pancreas organoids B Mice METHOD DETAILS B Co-culture of PDAC Organoids with CAFs B Xenotransplantation of Organoids B Gene Engineering of Organoids B Immunohistochemistry and in situ Hybridization B Whole-Exome Sequence Analysis B Copy-Number Analysis B Methylation Analysis B Real-Time Quantitative PCR B Western blot Analysis B Gene-Expression Microarray Analysis B Data Analysis for Publicly Available PDAC Specimens QUANTIFICATION AND STATISTICAL ANALYSIS DATA AND SOFTWARE AVAILABILITY

SUPPLEMENTAL INFORMATION Supplemental Information includes seven figures and seven tables and can be found with this article online at https://doi.org/10.1016/j.stem.2017.12.009. ACKNOWLEDGMENTS This work was supported by the Project for Cancer Research and Therapeutic Evolution (P-CREATE) from the Japan Agency for Medical Research and Development (AMED), a Grant-in-Aid for Scientific Research on Innovative Areas Stem Cell Ageing and Disease, and Grants-in-Aid for Scientific Research funded by the Ministry of Education, Culture, Sports, Science and Technology of Japan. Y.O., M.F., K.K., and S.S. were supported by the Japan Society for the Promotion of Science Research Fellowships for Young Scientists. We also thank the Collaborative Research Resources, School of Medicine, Keio University for the technical assistance provided. The R-spondinproducing cell line was a kind gift from C. Kuo (Stanford University). AUTHOR CONTRIBUTIONS Conceptualization, T. Seino and T. Sato; Methodology, T. Seino, S.K., and T. Sato; Investigation, T. Seino, S.K., M.S., Y.O., M.M., K.N., M.F., H.T., K.K., S.T., S.S., and T. Sato; Data Curation, M.S. and K.T.; Writing – Original Draft, T. Seino and T. Sato; Writing – Review & Editing, T. Seino, M.S., M.F., and T. Sato; Funding Acquisition, T. Sato; Resources, E.I., J.T., T.I., M.K., Y.K., and T.K. DECLARATION OF INTERESTS The authors declare no competing interests. Received: May 18, 2017 Revised: October 30, 2017 Accepted: December 14, 2017 Published: January 11, 2018 REFERENCES Apte, M.V., Pirola, R.C., and Wilson, J.S. (2012). Pancreatic stellate cells: a starring role in normal and diseased pancreas. Front. Physiol. 3, 344. Arena, S., Isella, C., Martini, M., de Marco, A., Medico, E., and Bardelli, A. (2007). Knock-in of oncogenic Kras does not transform mouse somatic cells but triggers a transcriptional response that classifies human cancers. Cancer Res. 67, 8468–8476. Bailey, P., Chang, D.K., Nones, K., Johns, A.L., Patch, A.M., Gingras, M.C., Miller, D.K., Christ, A.N., Bruxner, T.J., Quinn, M.C., et al.; Australian

Cell Stem Cell 22, 454–467, March 1, 2018 465

Pancreatic Cancer Genome Initiative (2016). Genomic analyses identify molecular subtypes of pancreatic cancer. Nature 531, 47–52.

Hruban, R.H., Goggins, M., Parsons, J., and Kern, S.E. (2000). Progression model for pancreatic cancer. Clin. Cancer Res. 6, 2969–2972.

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 human and mouse ductal pancreatic cancer. Cell 160, 324–338.

Hruban, R.H., Adsay, N.V., Albores-Saavedra, J., Anver, M.R., Biankin, A.V., Boivin, G.P., Furth, E.E., Furukawa, T., Klein, A., Klimstra, D.S., et al. (2006). Pathology of genetically engineered mouse models of pancreatic exocrine cancer: consensus report and recommendations. Cancer Res. 66, 95–106.

Brembeck, F.H., Schreiber, F.S., Deramaudt, T.B., Craig, L., Rhoades, B., Swain, G., Grippo, P., Stoffers, D.A., Silberg, D.G., and Rustgi, A.K. (2003). The mutant K-ras oncogene causes pancreatic periductal lymphocytic infiltration and gastric mucous neck cell hyperplasia in transgenic mice. Cancer Res. 63, 2005–2009.

Huang, L., Holtzinger, A., Jagan, I., BeGora, M., Lohse, I., Ngai, N., Nostro, C., Wang, R., Muthuswamy, L.B., Crawford, H.C., et al. (2015). Ductal pancreatic cancer modeling and drug screening using human pluripotent stem cell- and patient-derived tumor organoids. Nat. Med. 21, 1364–1371.

Brewer, A., Gove, C., Davies, A., McNulty, C., Barrow, D., Koutsourakis, M., Farzaneh, F., Pizzey, J., Bomford, A., and Patient, R. (1999). The human and mouse GATA-6 genes utilize two promoters and two initiation codons. J. Biol. Chem. 274, 38004–38016.

Huch, M., Bonfanti, P., Boj, S.F., Sato, T., Loomans, C.J., van de Wetering, M., Sojoodi, M., Li, V.S., Schuijers, J., Gracanin, A., et al. (2013). Unlimited in vitro expansion of adult bi-potent pancreas progenitors through the Lgr5/R-spondin axis. EMBO J. 32, 2708–2721.

Cheung, W.K., Zhao, M., Liu, Z., Stevens, L.E., Cao, P.D., Fang, J.E., Westbrook, T.F., and Nguyen, D.X. (2013). Control of alveolar differentiation by the lineage transcription factors GATA6 and HOPX inhibits lung adenocarcinoma metastasis. Cancer Cell 23, 725–738.

Hwang, C.I., Boj, S.F., Clevers, H., and Tuveson, D.A. (2016). Preclinical models of pancreatic ductal adenocarcinoma. J. Pathol. 238, 197–204.

Collisson, E.A., Sadanandam, A., Olson, P., Gibb, W.J., Truitt, M., Gu, S., Cooc, J., Weinkle, J., Kim, G.E., Jakkula, L., et al. (2011). Subtypes of pancreatic ductal adenocarcinoma and their differing responses to therapy. Nat. Med. 17, 500–503. Commo, F., Guinney, J., Ferté, C., Bot, B., Lefebvre, C., Soria, J.C., and André, F. (2016). rCGH: a comprehensive array-based genomic profile platform for precision medicine. Bioinformatics 32, 1402–1404. Doyon, Y., Vo, T.D., Mendel, M.C., Greenberg, S.G., Wang, J., Xia, D.F., Miller, J.C., Urnov, F.D., Gregory, P.D., and Holmes, M.C. (2011). Enhancing zincfinger-nuclease activity with improved obligate heterodimeric architectures. Nat. Methods 8, 74–79. Emami, K.H., Nguyen, C., Ma, H., Kim, D.H., Jeong, K.W., Eguchi, M., Moon, R.T., Teo, J.L., Kim, H.Y., Moon, S.H., et al. (2004). A small molecule inhibitor of beta-catenin/CREB-binding protein transcription [corrected]. Proc. Natl. Acad. Sci. USA 101, 12682–12687. Farin, H.F., Jordens, I., Mosa, M.H., Basak, O., Korving, J., Tauriello, D.V., de Punder, K., Angers, S., Peters, P.J., Maurice, M.M., and Clevers, H. (2016). Visualization of a short-range Wnt gradient in the intestinal stem-cell niche. Nature 530, 340–343. Fujii, M., Matano, M., Nanki, K., and Sato, T. (2015). Efficient genetic engineering of human intestinal organoids using electroporation. Nat. Protoc. 10, 1474–1485. Fujii, M., Shimokawa, M., Date, S., Takano, A., Matano, M., Nanki, K., Ohta, Y., Toshimitsu, K., Nakazato, Y., Kawasaki, K., et al. (2016). A colorectal tumor organoid library demonstrates progressive loss of niche factor requirements during tumorigenesis. Cell Stem Cell 18, 827–838. Furukawa, T., Kuboki, Y., Tanji, E., Yoshida, S., Hatori, T., Yamamoto, M., Shibata, N., Shimizu, K., Kamatani, N., and Shiratori, K. (2011). Whole-exome sequencing uncovers frequent GNAS mutations in intraductal papillary mucinous neoplasms of the pancreas. Sci. Rep. 1, 161. Gao, D., Vela, I., Sboner, A., Iaquinta, P.J., Karthaus, W.R., Gopalan, A., Dowling, C., Wanjala, J.N., Undvall, E.A., Arora, V.K., et al. (2014). Organoid cultures derived from patients with advanced prostate cancer. Cell 159, 176–187. Heiser, P.W., Cano, D.A., Landsman, L., Kim, G.E., Kench, J.G., Klimstra, D.S., Taketo, M.M., Biankin, A.V., and Hebrok, M. (2008). Stabilization of beta-catenin induces pancreas tumor formation. Gastroenterology 135, 1288–1300. Higasa, K., Miyake, N., Yoshimura, J., Okamura, K., Niihori, T., Saitsu, H., Doi, K., Shimizu, M., Nakabayashi, K., Aoki, Y., et al. (2016). Human genetic variation database, a reference database of genetic variations in the Japanese population. J. Hum. Genet. 61, 547–553. Hingorani, S.R., Petricoin, E.F., Maitra, A., Rajapakse, V., King, C., Jacobetz, M.A., Ross, S., Conrads, T.P., Veenstra, T.D., Hitt, B.A., et al. (2003). Preinvasive and invasive ductal pancreatic cancer and its early detection in the mouse. Cancer Cell 4, 437–450.

466 Cell Stem Cell 22, 454–467, March 1, 2018

Ito, M., Hiramatsu, H., Kobayashi, K., Suzue, K., Kawahata, M., Hioki, K., Ueyama, Y., Koyanagi, Y., Sugamura, K., Tsuji, K., et al. (2002). NOD/SCID/ gamma(c)(null) mouse: an excellent recipient mouse model for engraftment of human cells. Blood 100, 3175–3182. Jiang, X., Hao, H.X., Growney, J.D., Woolfenden, S., Bottiglio, C., Ng, N., Lu, B., Hsieh, M.H., Bagdasarian, L., Meyer, R., et al. (2013). Inactivating mutations of RNF43 confer Wnt dependency in pancreatic ductal adenocarcinoma. Proc. Natl. Acad. Sci. USA 110, 12649–12654. Johnson, W.E., Li, C., and Rabinovic, A. (2007). Adjusting batch effects in microarray expression data using empirical Bayes methods. Biostatistics 8, 118–127. Konishi, H., Karakas, B., Abukhdeir, A.M., Lauring, J., Gustin, J.P., Garay, J.P., Konishi, Y., Gallmeier, E., Bachman, K.E., and Park, B.H. (2007). Knock-in of mutant K-ras in nontumorigenic human epithelial cells as a new model for studying K-ras mediated transformation. Cancer Res. 67, 8460–8467. Koo, B.K., Spit, M., Jordens, I., Low, T.Y., Stange, D.E., van de Wetering, M., van Es, J.H., Mohammed, S., Heck, A.J., Maurice, M.M., and Clevers, H. (2012). Tumour suppressor RNF43 is a stem-cell E3 ligase that induces endocytosis of Wnt receptors. Nature 488, 665–669. Kopp, J.L., von Figura, G., Mayes, E., Liu, F.F., Dubois, C.L., Morris, J.P., 4th, Pan, F.C., Akiyama, H., Wright, C.V., Jensen, K., et al. (2012). Identification of Sox9-dependent acinar-to-ductal reprogramming as the principal mechanism for initiation of pancreatic ductal adenocarcinoma. Cancer Cell 22, 737–750. Kopp, J.L., Grompe, M., and Sander, M. (2016). Stem cells versus plasticity in liver and pancreas regeneration. Nat. Cell Biol. 18, 238–245. Lee, J., Snyder, E.R., Liu, Y., Gu, X., Wang, J., Flowers, B.M., Kim, Y.J., Park, S., Szot, G.L., Hruban, R.H., et al. (2017). Reconstituting development of pancreatic intraepithelial neoplasia from primary human pancreas duct cells. Nat. Commun. 8, 14686. Matano, M., Date, S., Shimokawa, M., Takano, A., Fujii, M., Ohta, Y., Watanabe, T., Kanai, T., and Sato, T. (2015). Modeling colorectal cancer using CRISPR-Cas9-mediated engineering of human intestinal organoids. Nat. Med. 21, 256–262. Mihara, E., Hirai, H., Yamamoto, H., Tamura-Kawakami, K., Matano, M., Kikuchi, A., Sato, T., and Takagi, J. (2016). Active and water-soluble form of lipidated Wnt protein is maintained by a serum glycoprotein afamin/a-albumin. eLife 5, e11621. Moffitt, R.A., Marayati, R., Flate, E.L., Volmar, K.E., Loeza, S.G., Hoadley, K.A., Rashid, N.U., Williams, L.A., Eaton, S.C., Chung, A.H., et al. (2015). Virtual microdissection identifies distinct tumor- and stroma-specific subtypes of pancreatic ductal adenocarcinoma. Nat. Genet. 47, 1168–1178. National Cancer Institute. Cancer Stat Facts: Pancreas Cancer. https://seer. cancer.gov/statfacts/html/pancreas.html. Noll, E.M., Eisen, C., Stenzinger, A., Espinet, E., Muckenhuber, A., Klein, C., Vogel, V., Klaus, B., Nadler, W., Rösli, C., et al. (2016). CYP3A5 mediates basal and acquired therapy resistance in different subtypes of pancreatic ductal adenocarcinoma. Nat. Med. 22, 278–287.

Öhlund, D., Handly-Santana, A., Biffi, G., Elyada, E., Almeida, A.S., PonzSarvise, M., Corbo, V., Oni, T.E., Hearn, S.A., Lee, E.J., et al. (2017). Distinct populations of inflammatory fibroblasts and myofibroblasts in pancreatic cancer. J. Exp. Med. 214, 579–596. Ootani, A., Li, X., Sangiorgi, E., Ho, Q.T., Ueno, H., Toda, S., Sugihara, H., Fujimoto, K., Weissman, I.L., Capecchi, M.R., and Kuo, C.J. (2009). Sustained in vitro intestinal epithelial culture within a Wnt-dependent stem cell niche. Nat. Med. 15, 701–706. Özdemir, B.C., Pentcheva-Hoang, T., Carstens, J.L., Zheng, X., Wu, C.C., Simpson, T.R., Laklai, H., Sugimoto, H., Kahlert, C., Novitskiy, S.V., et al. (2014). Depletion of carcinoma-associated fibroblasts and fibrosis induces immunosuppression and accelerates pancreas cancer with reduced survival. Cancer Cell 25, 719–734. Proffitt, K.D., Madan, B., Ke, Z., Pendharkar, V., Ding, L., Lee, M.A., Hannoush, R.N., and Virshup, D.M. (2013). Pharmacological inhibition of the Wnt acyltransferase PORCN prevents growth of WNT-driven mammary cancer. Cancer Res. 73, 502–507. Rhim, A.D., Oberstein, P.E., Thomas, D.H., Mirek, E.T., Palermo, C.F., Sastra, S.A., Dekleva, E.N., Saunders, T., Becerra, C.P., Tattersall, I.W., et al. (2014). Stromal elements act to restrain, rather than support, pancreatic ductal adenocarcinoma. Cancer Cell 25, 735–747. Roe, J.S., Hwang, C.I., Somerville, T.D.D., Milazzo, J.P., Lee, E.J., Da Silva, B., Maiorino, L., Tiriac, H., Young, C.M., Miyabayashi, K., et al. (2017). Enhancer reprogramming promotes pancreatic cancer metastasis. Cell 170, 875– 888.e20. Sato, T., and Clevers, H. (2013). Growing self-organizing mini-guts from a single intestinal stem cell: mechanism and applications. Science 340, 1190–1194. Sato, T., Vries, R.G., Snippert, H.J., van de Wetering, M., Barker, N., Stange, D.E., van Es, J.H., Abo, A., Kujala, P., Peters, P.J., and Clevers, H. (2009).

Single Lgr5 stem cells build crypt-villus structures in vitro without a mesenchymal niche. Nature 459, 262–265. Shain, A.H., Giacomini, C.P., Matsukuma, K., Karikari, C.A., Bashyam, M.D., Hidalgo, M., Maitra, A., and Pollack, J.R. (2012). Convergent structural alterations define SWItch/Sucrose NonFermentable (SWI/SNF) chromatin remodeler as a central tumor suppressive complex in pancreatic cancer. Proc. Natl. Acad. Sci. USA 109, E252–E259. Shimokawa, M., Ohta, Y., Nishikori, S., Matano, M., Takano, A., Fujii, M., Date, S., Sugimoto, S., Kanai, T., and Sato, T. (2017). Visualization and targeting of LGR5+ human colon cancer stem cells. Nature 545, 187–192. Sousa, C.M., Biancur, D.E., Wang, X., Halbrook, C.J., Sherman, M.H., Zhang, L., Kremer, D., Hwang, R.F., Witkiewicz, A.K., Ying, H., et al. (2016). Pancreatic stellate cells support tumour metabolism through autophagic alanine secretion. Nature 536, 479–483. Su, G.H., Hilgers, W., Shekher, M.C., Tang, D.J., Yeo, C.J., Hruban, R.H., and Kern, S.E. (1998). Alterations in pancreatic, biliary, and breast carcinomas support MKK4 as a genetically targeted tumor suppressor gene. Cancer Res. 58, 2339–2342. Waddell, N., Pajic, M., Patch, A.M., Chang, D.K., Kassahn, K.S., Bailey, P., Johns, A.L., Miller, D., Nones, K., Quek, K., et al.; Australian Pancreatic Cancer Genome Initiative (2015). Whole genomes redefine the mutational landscape of pancreatic cancer. Nature 518, 495–501. Wang, M., Maier, P., Wenz, F., Giordano, F.A., and Herskind, C. (2013). Mitogenic signalling in the absence of epidermal growth factor receptor activation in a human glioblastoma cell line. J. Neurooncol. 115, 323–331. Witkiewicz, A.K., McMillan, E.A., Balaji, U., Baek, G., Lin, W.C., Mansour, J., Mollaee, M., Wagner, K.U., Koduru, P., Yopp, A., et al. (2015). Whole-exome sequencing of pancreatic cancer defines genetic diversity and therapeutic targets. Nat. Commun. 6, 6744.

Cell Stem Cell 22, 454–467, March 1, 2018 467

Resource A Biobank of Breast Cancer Explants with Preserved Intra-tumor Heterogeneity to Screen Anticancer Compounds Alejandra Bruna,1,11 Oscar M. Rueda,1,11 Wendy Greenwood,1 Ankita Sati Batra,1 Maurizio Callari,1 Rajbir Nath Batra,1 Katherine Pogrebniak,1 Jose Sandoval,1 John W. Cassidy,1 Ana Tufegdzic-Vidakovic,1 Stephen-John Sammut,1 Linda Jones,1,2 Elena Provenzano,2 Richard Baird,1,2 Peter Eirew,3 James Hadfield,1 Matthew Eldridge,1 Anne McLaren-Douglas,4 Andrew Barthorpe,4 Howard Lightfoot,4 Mark J. O’Connor,5 Joe Gray,6 Javier Cortes,7 Jose Baselga,8 Elisabetta Marangoni,9 Alana L. Welm,10 Samuel Aparicio,3 Violeta Serra,7 Mathew J. Garnett,4 and Carlos Caldas1,2,12,* 1Department of Oncology and Cancer Research UK Cambridge Institute, Li Ka Shing Centre, University of Cambridge, Cambridge CB2 0RE, UK 2Cambridge Breast Unit, NIHR Cambridge Biomedical Research Centre and Cambridge Experimental Cancer Medicine Centre at Cambridge University Hospitals NHS Foundation Trust, Cambridge CB2 2QQ, UK 3Department of Molecular Oncology, British Columbia Cancer Research Centre, Vancouver, BC V5Z 1L3, Canada 4Wellcome Trust Sanger Institute, Wellcome Trust Genome Campus, Hinxton, Cambridgeshire CB10 1SA, UK 5DNA Damage Response Biology Area, Oncology IMED, AstraZeneca, Alderley Park, Macclesfield SK10 4TG, UK 6OHSU Knight Cancer Institute, Oregon Health & Science University, 3181 SW Sam Jackson Park Road, Portland, OR 97239, USA 7Vall d’Hebron Institute of Oncology, 08035 Barcelona, Spain 8Human Oncology and Pathogenesis Program, Department of Medicine, Memorial Sloan Kettering Cancer Center, NY 10065, USA 9Translational Research Department, Institut Curie, 26 rue d’Ulm, Paris 75005, France 10Huntsman Cancer Institute, Salt Lake City, UT 84112, USA 11Co-first author 12Lead Contact *Correspondence: carlos.caldas@cruk.cam.ac.uk http://dx.doi.org/10.1016/j.cell.2016.08.041

SUMMARY

The inter- and intra-tumor heterogeneity of breast cancer needs to be adequately captured in pre-clinical models. We have created a large collection of breast cancer patient-derived tumor xenografts (PDTXs), in which the morphological and molecular characteristics of the originating tumor are preserved through passaging in the mouse. An integrated platform combining in vivo maintenance of these PDTXs along with short-term cultures of PDTX-derived tumor cells (PDTCs) was optimized. Remarkably, the intra-tumor genomic clonal architecture present in the originating breast cancers was mostly preserved upon serial passaging in xenografts and in shortterm cultured PDTCs. We assessed drug responses in PDTCs on a high-throughput platform and validated several ex vivo responses in vivo. The biobank represents a powerful resource for pre-clinical breast cancer pharmacogenomic studies (http://caldaslab. cruk.cam.ac.uk/bcape), including identification of biomarkers of response or resistance. INTRODUCTION Molecular stratification is the first step toward precision cancer medicine (Aparicio and Caldas, 2013). Recently, we reported

(Curtis et al., 2012; Dawson et al., 2013; Dvinge et al., 2013) and validated (Ali et al., 2014) a genome driver-based molecular taxonomy of breast cancer. Modeling this diverse inter-tumor heterogeneity of breast cancer is challenging and requires generation of explant models representing the ten identified integrative clusters (IntClust). Cancer cell lines have been extensively used for drug development and biomarker discovery (Heiser et al., 2012) but are successful at predicting clinical responses in only a handful of examples (Kim et al., 2015; Sharma et al., 2010). The modest clinical predictive value of cancer cell lines results from their recognized shortcomings: limited capacity to recapitulate interand intra-tumor heterogeneity and adaptation to growth in artificial conditions. These limitations are significant because both tumor subtype and cancer genome evolution, resulting in intra-tumor heterogeneity, remain the main challenges to successful cancer treatment. The increasing understanding of cancer biology has led to the availability of targeted therapies. These drugs typically explore oncogene addiction or synthetic lethality (Kaelin, 2005; Luo et al., 2009; Torti and Trusolino, 2011). Unfortunately, the inherent heterogeneity of cancer means that either primary or acquired resistance nearly always occurs. Successful early drug development hence requires molecular stratification and characterization of intra-tumor heterogeneity. Patient-derived tumor xenografts (PDTXs) have emerged as powerful pre-clinical models to recapitulate the diversity of human tumors (Cassidy et al., 2015). The greatest promise of PDTXs is their potential to improve the rates of attrition in cancer

260 Cell 167, 260–274, September 22, 2016 ª 2016 The Authors. Published by Elsevier Inc. This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/4.0/).

drug development (Aparicio et al., 2015; Gao et al., 2015; Hidalgo et al., 2014; Tentler et al., 2012). However, generalized use of PDTXs in high-throughput drug studies is unrealistic, for both cost and animal welfare reasons. Moreover, it has not been clear whether PDTXs retain the heterogeneity of the original tumor. Here, we demonstrate molecularly characterized PDTXs and their matched PDTX-derived tumor cells (PDTCs) in shortterm culture do retain this heterogeneity and may be used as a platform for cancer drug screening with the potential to uncover molecular mechanisms of therapy response. RESULTS Generation of Breast Cancer PDTXs Representing Most Breast Cancer Clinical and Molecular Subtypes We have established a large bank (n = 83) of live human breast cancer explants by implantation of tumor samples in highly immunodeficient mice (NOD.Cg-Prkdcscid Il2rgtm1Wjl/SzJ or NSGs; see STAR Methods). Comprehensive clinical information on the patients and originating cancer sample implanted to generate PDTXs can be found in Table S1. To date, PDTXs have been successfully established from both primary (n = 46) and metastatic (n = 37) sites, and more than 50% (n = 50) are from ER+ disease (Table S1). The PDTX growth rates upon initial engraftment and after subsequent re-implantation were variable across models, remained mostly stable upon serial engraftment, and tended to be faster in explants originated from ER tumors (Figure 1A shows data for 31 models). Importantly, all established models tested to date could be flash frozen and subsequently successfully engrafted, ensuring the persistence of the living biobank. In order to be classified into one of the IntClust using the method we described (Ali et al., 2014), the PDTXs were subject to copy number profiling, by shallow whole-genome sequencing (‘‘sWGS’’), and expression profiling, by microarrays (‘‘RNAexp’’). The copy number profiles of PDTXs classified into each IntClust were similar to those reported in primary tumors (Curtis et al., 2012; Figure S1A). The goodness of fit scores of IntClust assignment were computed (for IntClusts with more than one xenograft model: IntClust1; IntClust3; IntClust4; IntClust5; IntClust6; IntClust9; and IntClust10), and with one exception (IntClust3), these scores were very similar to classifying primary tumors (Figure S1B). Copy number aberrations (CNAs) in known breast cancer driver genes (Curtis et al., 2012) present in the PDTXs included gains/amplifications of MYC (78%), CCNE1 (34%), ZNF703 (25%), CCND1 (31%), MDM2 (25%), and ERBB2 (9%) and deletions of PTEN (41%), PPP2R2A (72%), CDKN2A (47%), and CDKN2B (47%). This CNA frequency distribution is different from that seen in a breast cancer clinical population (METABRIC dataset) and reflects both the origin of the PDTXs (around 45% were from metastatic biopsies) and the disproportionate engraftment of triple-negative basal-like cancers (IntClust10) and more-aggressive subtypes of ER+ tumors (IntClust1 and IntClust9; Figure 1B). In contrast, we observed lower engraftment of ER+ tumors from better-outcome subtypes (IntClust3, IntClust4, IntClust7, and IntClust8; Figure 1B; Table S1).

Downstream analysis of mRNA expression data using the gene set variation analysis (GSVA) approach (Hänzelmann et al., 2013), a method to estimate pathway activity, also showed that the diversity of activity scores in cancer-related pathways (Molecular Signatures Database; http://software.broadinstitute. org/gsea/msigdb; Liberzon et al., 2011) in PDTXs was similar to that observed in the breast cancer clinical population (Figure S1C). Furthermore, in matched pairs, the activity of breastcancer-related pathways (e.g., PTEN, Tp53, BRCA1, Her2, and Cyclin D1) in the PDTXs was correlated with and predicted the activity scores in the originating breast cancer samples (Figure S1D). The subtype distribution of engrafted PDTXs was also reflected by the mutation frequencies identified using wholeexome sequencing (‘‘WES’’). The most-commonly mutated genes in ER breast cancers (Cancer Genome Atlas Network, 2012) were mutated at similar frequencies in ER PDTXs (Figure 1C). In contrast, frequencies of mutated genes in ER+ PDTXs mirrored those found in more-aggressive subtypes of ER+ tumors (Pereira et al., 2016). As an example, PIK3CA mutations were found in only 27% of ER+ PDTX models (Figure 1C), compared to 46% and 38% in the METABRIC and The Cancer Genome Atlas (TCGA) cohorts, respectively. In summary, these data show that we have successfully generated a living biobank of breast cancer xenografts, representing the clinical and molecular diversity of the disease. PDTXs Retain Their Original Histological and Molecular Features through Passaging Histologically, PDTXs (23 models analyzed) showed similar morphology to the originating tumor; tubule formation and associated stroma were present in the xenograft, as seen in the matched patient cancer sample (Figure S2A). Histological review of multiple PDTX passages (Table S2) revealed that tumor tissue morphology remained stable with serial engraftment. Analysis of immunohistochemistry for epithelial markers (CK5, CK8, CK14, CK18, E-cadherin, and epithelial specific antigen) and for clinical biomarkers (ER, PR, Her2, Ki67, and p53) showed these features were similar in matched pairs of PDTX model and originating breast cancer sample and were consistently retained with passaging (Figure S2A and Table S2 for summary of the data). The PDTX samples were comprehensively molecularly characterized at several passages using sWGS (for CNAs), WES (for single nucleotide variations [SNVs]), reduced-representation bisulfite sequencing (‘‘RRBS’’) (for DNA methylation), and RNAexp (for global expression and pathway activity profiling). The analysis of sequencing data from PDTX samples is complicated by the presence of a variable and unknown amount of mouse cells. To address this, a serial dilution series of control samples with known mixtures of human and mouse DNA was created to develop a robust computational pipeline to discriminate human and mouse reads with an accuracy >99.9% (see Figure S2B and STAR Methods for details). This pipeline identified three spontaneous mouse tumors arising at or near the implantation site, which were discarded from further experiments. Post-filtered aligned data from this pipeline were used for somatic copy number and mutation calling (see STAR Methods).

Cell 167, 260–274, September 22, 2016 261

Figure 1. Derivation of an Extensively Annotated Breast Cancer PDTX-PDTC Biobank Representing Breast Cancer Subtypes (A) Timeline of engraftment for established PDTX models (n = 31; ER+ in red; ER in blue). Each square represents a time point of engraftment. Average ER+ and ER re-implantation time is shown on furthermost right panel. Model IDs are color coded according to integrative cluster (IntClust). (B) Bar plots showing the IntClust distribution of PDTX models (n = 40; shadowed) and for comparison primary breast cancers from METABRIC (n = 1,980; dense). (C) Distribution of somatic mutations in tumors from the TCGA cohort (n = 495) and PDTX models (n = 30), stratified by ER status. See also Figure S1.

262 Cell 167, 260â&#x20AC;&#x201C;274, September 22, 2016

We used these data to determine how implantation, serial passaging, and replicate engraftment affected gene expression, cancer pathway activation scores, allelic fractions of somatic mutations, CNAs, and DNA methylation. This analysis, done also for comparison in reference sets (different tumor samples and technical and biological replicates), revealed a high degree of correlation in matched sample pairs for all data types (Figure 2A). The biological relevance of the models was evidenced by similar gene expression profiles between the originating tumor and the PDTX (RNAexp n = 44; r = 0.95; interquartile range [IQR] 0.94–0.97), oncogenic pathway scores (r = 0.67; IQR 0.44–0.79), CNA profiles (sWGS n = 45; r = 0.91; IQR 0.84– 0.93), SNV allelic fractions (WES n = 82; r = 0.81; IQR 0.74– 0.88), and DNA methylation profiles (RRBS n = 8; r = 0.82; IQR 0.79–0.84). The biological robustness of the models across serial passaging was also remarkable, with retention of gene expression profiles (RNAexp n = 217; r = 0.98; IQR 0.97–0.99), oncogenic pathway scores (r = 0.87; IQR 0.80–0.93), CNA profiles (sWGS n = 109; r = 0.97; IQR 0.92–0.98), SNV allelic fractions (WES n = 201; r = 0.92; IQR 0.87–0.95), and DNA methylation profiles (RRBS n = 37; r = 0.82; IQR 0.78–0.90; Figure 2A). Mutational signatures (Alexandrov et al., 2013; Rosenthal et al., 2016) in matched PDTXs across serial passages and the originating sample were also conserved (Figure 2B). Representative examples across the molecular data types for individual PDTX models are shown in Figure 2B. In summary, the comprehensive characterization of histopathological characteristics, somatic genomic aberrations (CNAs and SNVs), methylation profiles, and gene expression of the biobank of human breast cancer explants confirms these models have a remarkable level of multi-dimensional molecular resemblance with their matched cancer of origin, significantly extending the observations we and others had previously reported (DeRose et al., 2011; Eirew et al., 2015; Li et al., 2013; Marangoni et al., 2007). Our findings robustly demonstrate these multidimensional molecular features are conserved through serial engraftment in the mouse. Mouse Stromal Composition of PDTXs Remains Stable through Passaging Breast cancer PDTXs retain similar architecture to the originating tumor and, through passaging, this remains stable. This occurs despite mouse stroma replacing the human stroma (DeRose et al., 2011; Hidalgo et al., 2014). We used the custom sequencing analysis pipeline described above (Figure S2B; STAR Methods) to deconvolute the proportion of mouse DNA sequences in PDTX samples as a surrogate for mouse stromal cell content. Out of the 94 xenograft samples examined, only five had more than 40% mouse cells. Replicates obtained from these five had lower mouse stromal content, reflecting intra-PDTX heterogeneity (Figure S2C). The data from multiple PDTX models also showed the proportion of mouse content does not change significantly across passages (Figure S2C). Two independent methods were used to validate these observations: fluorescence-activated cell sorting (FACS) of PDTX-derived single-cell suspensions with an MHC-class I anti-mouse H-2Kb/H-2Db antibody and fluorescence in situ

hybridization (FISH) with mouse and human centromeric probes in tissue sections (Lawson et al., 2015; Li et al., 2013; Figure S2D). In summary, these data show the mouse stroma contribution to the xenografts is stable across serial passaging. Intra-tumor Heterogeneity and Clonal Architecture Are Maintained in PDTXs Human breast cancers are composed of clones differing in mutation content (Aparicio and Caldas, 2013), resulting in intra-tumor genomic heterogeneity. This intra-tumor heterogeneity, although variable across tumors, is already present at diagnosis (Shah et al., 2012) and evolves dynamically in space and time (Ding et al., 2010; Murtaza et al., 2015; Shah et al., 2009). WES data were used to interrogate both intra-tumor heterogeneity and clonal architecture in matched originating tumor, initial engrafted, and serially passaged xenografts. Quantification of intra-tumor heterogeneity using the mutantallele tumor heterogeneity (MATH) method (Mroz and Rocco, 2013) revealed that the originating patient tumor samples had a range of scores (from low to high), as expected given their diverse IntClust subtype. The heterogeneity scores in multiple passages of matched PDTXs were similar, demonstrating explants preserve intra-tumor heterogeneity (Figure S3A). Clonal architecture in individual samples and clonal dynamics upon engraftment and across serial passaging were assessed on 104 samples from 22 models using PyClone (Roth et al., 2014), as we recently described (Eirew et al., 2015). PyClone identified 190 clonal clusters across the samples analyzed, but only 38 clonal clusters (20%) had significant changes in cellular prevalence estimates (Table S3 for extended information from PyClone analysis in all models tested). Clonal selection was seen upon initial engraftment (average change in clonal prevalence 0.21) but minimal through serial transplantation (average change in clonal prevalence 0.07; Figure S3B). We next asked whether clonal clusters showing engraftment-associated dynamics were enriched for cancer drivers. Recently, our group used a ratiometric method (Vogelstein et al., 2013) to identify 40 breast cancer mutation driver genes in 2,433 breast cancers (Pereira et al., 2016). Remarkably, in only 4 of the 38 clonal clusters that changed significantly after engraftment or during passaging could we identify a mutation driver: BAP1 in STG139 (cluster 12); KDM6A in HCI004 (cluster 3); MAP3K1 in STG143 (cluster 3); and PIK3CA in HCI008 (cluster 2; Table S3). These data strongly suggest that most of the clonal dynamics within xenografts are not associated with known driver genes. Figure 3A shows examples both of individual clonal cluster plots and of variant allele frequency distributions for individual genes within these clusters. Figure S3C shows all individual clonal cluster plots generated from the 22 models analyzed to illustrate the full diversity of clonal architectures observed in the PDTX biobank. We analyzed in detail the clonal architecture of two cases for which we had both primary and subsequent metastasis samples: STG139 and a lung metastasis 12 months later, STG139M, and AB521 and a liver metastasis 8 months later,

Cell 167, 260–274, September 22, 2016 263

Figure 2. PDTXs Closely Match Originating Patient Cancer Samples (A) Heatmap of Pearson correlation scores across molecular data types (different sample sizes described in the main text). (B) Panels with individual examples for five types of molecular data. (Left panel: top) CNA plots for AB551 (originating sample [T], PDTX, and PDTC) are shown; (bottom) scatterplot of methylated CpGs (from RRBS data) in AB521M is shown. (Right panel: top) Scatterplots of pathway activity scores in AB521M are shown. (Middle) Scatterplots of variant allelic fractions in STG139 are shown. (Bottom) Mutational profiles in AB551 are shown.

264 Cell 167, 260â&#x20AC;&#x201C;274, September 22, 2016

(legend on next page)

Cell 167, 260â&#x20AC;&#x201C;274, September 22, 2016 265

AB521M. For the first patient, we generated a xenograft from the primary tumor (STG139-X) and a xenograft from the lung metastasis (STG139M-X), and for the second patient, we generated a xenograft from the liver metastasis (AB521M-X). Figure 3B shows the clonal cluster plots and a heatmap of variant allele frequencies derived from WES data for sample sets from both patients. Clonal clusters shared by both the originating primary tumor and metastatic samples (STG139: clonal clusters 1, 7, and 8; AB521: clonal clusters 2, 8, and 10) had stable cellular prevalence across passaging. These clusters contained around 80% of all SNVs detected in these two cases (Figure 3B) and included not only the stem or truncal cluster but significantly also included sub-clonal and even very minor clusters (estimated cellular prevalence <5%). Metastasis-only clonal clusters (STG139: cluster 4; AB521: cluster 7) were also preserved upon serial passaging. Finally, there were clusters detected only in engrafted-derived samples (clonal clusters 3, 11, and 12 in STG139 and 6 in AB521, respectively). Although these results need a degree of cautious interpretation (only two cases analyzed), it demonstrates that both originating tumor and xenografts contain multiple clones, and the dynamics of clones in the patient (by comparing primary and metastasis biopsies) and in the mouse (by comparing passages) have both similarities and differences. Detailed analyses of clonal dynamics in more matched primary metastatic samples and their derived PDTXs, with mirrored treatment regimes, will be extremely informative toward understanding the mechanisms that are operative in tumor clonal ecosystems (Heppner, 1984; Tabassum and Polyak, 2015). From one large breast cancer brain metastasis (CAMBMT1), we obtained five spatially separate biopsies, which were implanted into five different NSG mice. WES data from all five biopsies showed similar clonal architectures (Figure 3C, left panel; Table S3). This case allowed us to compare the clonal architecture of the five xenografted samples, revealing remarkable similarity, despite some variation in the originating cellular prevalence in the separate biopsies (see, for example, variant allele frequencies of GATA3, OTOGL, and BTD; Figure 3C, right panel). These near identical clonal dynamics upon engraftment strongly suggest deterministic mechanisms operate on clonal selection and validate our previous hypothesis that specific mutations act as genetic markers of fitness and dictate evolutionary trajectories (Eirew et al., 2015). In summary, these data show PDTXs constitute a pre-clinical model that captures the most-clinically relevant feature in human cancer: heterogeneous genomic architecture that dynamically evolves. Moreover, the data also indicate that the clonal dy-

namics of the derived and serially passaged explants are not stochastic. Generation of Short-Term Cultures of PDTCs The PDTXs described constitute a living biobank of breast cancer explants that retain through passaging the inter- and intra-tumor heterogeneity encountered in the clinic. We therefore developed a method to enable the use of this valuable resource for high-content pre-clinical drug screening, similar to the approach widely used with cell lines (Barretina et al., 2012; Garnett et al., 2012). The method involved optimizing short ex vivo culture of cells isolated from the PDTXs (named PDTCs). These short-term PDTC cultures were successfully generated from all models where attempted (n = 27, at least two different passages from each; Figure S4A; see STAR Methods). Sequencing data confirmed that the PDTCs had a proportion of mouse-derived cells similar to that found in the originating PDTX (Figure S4B). Cell proliferation, cell viability, and cell divisions (measured by PKH26 assay) were analyzed in the cultures and showed the expected variability across models, reflecting the diversity of the originating cancer (Figures S4C and S4D). PDTCs derived from ten of the PDTX models were extensively characterized using WES, sWGS, and RNAexp. Analysis of these data showed the short ex-vivo-cultured PDTCs retained the molecular features of the originating PDTX (Figures 2A and 2B), including similar clonal architecture (Figures 3A, S3B, and S3C; Table S3). The average absolute change in clonal cluster cellular prevalence in matched PDTC-PDTX pairs was 0.08 (Figure S3B, left panel). In summary, PDTCs can be systematically and consistently generated from PDTXs and retain their genomic features, making them an excellent model system for high-throughput drug screens. High-Throughput Drug Screening in PDTC Models We tested the use of PDTCs as a pre-clinical drug-screening platform with an approach similar to that which we previously reported for cell lines and organoids (Garnett et al., 2012; van de Wetering et al., 2015). A selection of 22 different PDTX models were plated as PDTCs and 24 hr later screened with 108 compounds, representing a total of 6,634 drug tests performed (see STAR Methods). The compounds used were either approved cancer treatments or drugs targeting key cancer pathways (Table S4). The effect of drug treatment on cell viability was determined by CellTiter-Glo (CTG) (Garnett et al., 2012; van de Wetering et al., 2015) and drug responses represented by

Figure 3. Clonal Architecture and Clonal Dynamics of Breast Cancer PDTXs (A) Example plots of AB551 (left panel), HCI002 (middle panel), and STG282 (right panel). (Left graph) The mean cellular prevalence estimates of mutation clusters in originating patient samples (T) and subsequent xenograft passages (Xn; n for passage number) or PDTCs (XnCy; y for days in culture) are shown. PyClone was used to infer clusters and cellular prevalence using WES data. Line widths indicate the number of SNVs comprising each mutation cluster (numbers in brackets adjacent to each plot). Asterisks indicate clonal clusters with significant changes in cellular prevalence. (Right graph) Plots of distribution of variant allele frequency for selected genes within clusters. (B) PyClone plots (as in A) and cellular prevalence heatmap plots for STG139 and AB521 samples. (C) PyClone plot and plots of distribution of variant allele frequency for selected genes within clusters (as in A) of five spatially separated biopsies and their matched xenografts in CAMBMT1. See also Figure S3.

266 Cell 167, 260â&#x20AC;&#x201C;274, September 22, 2016

(legend on next page)

Cell 167, 260â&#x20AC;&#x201C;274, September 22, 2016 267

(1) the half-maximal inhibitory concentration (IC50), (2) the doseresponse curve, and (3) the area under the dose response curve (AUC). In total, 2,550 drug-PDTC combinations were tested, with a range of 5â&#x20AC;&#x201C;20 (mean = 16) PDTC models screened per drug. For most models, the drug treatment was performed in at least three technical replicates (same model, same passage, and same mouse) and in two or three biological replicates (same model, different passage). One significant limitation of these analyses is that these measurements (IC50 and AUC) did not account for cell division rates across the different PDTC models. Growth rate inhibition metrics have recently been shown to provide more-reliable measurements of sensitivity to cancer drugs (Hafner et al., 2016). Nevertheless, we have been able to make several observations that attest to the value of the drug screening results obtained despite this caveat. First, the observed AUC values across all drugs and models tested were highly correlated across technical (Pearson correlation of 0.94) and biological replicates (Pearson correlation of 0.78; Figure 4A). These results are highly similar to those we previously reported in established cell lines or tumor organoids (Garnett et al., 2012; van de Wetering et al., 2015). To further verify the robustness of these in vitro drug response data, we tested in eight PDTC models a set of 19 drugs using CyQUANT and Sytox endpoint assays, in addition to CTG (see STAR Methods). The results of these experiments revealed highly correlated drug responses independently of the assay used (Figure S4E; Table S5). Second, analysis of AUC data for compounds targeting the same pathway or with similar mechanism of action showed highly correlated response profiles. One example with inhibitors of the PI3K-AKT-mTOR pathway (NVP-BEZ235/dactolisib, AZD8055, GDC0941/pictilisib, AKT inhibitor, and MK-2206) is shown in Figure 4B. Another example with compounds targeting homologous recombination repair defects (PARP inhibitor BMN673/talazoparib and cisplatin, a DNA cross-linking agent) is shown in Figure 4C. Distinct PDTC models can sometimes share the same IC50 and AUC values for a compound and have very different doseresponse curves. Hence, a new method, based on the pattern of the slope of the dose-response curve, was developed to classify drug sensitivity patterns into eight groups (see STAR Methods). Figure S5A shows for each compound the proportion of drug responses classified into each of the eight drug sensitivity patterns across all models tested with that drug. Clustering of drug sensitivity patterns (Figure S5B) confirmed the high reproducibility and biological robustness of the data: different passages of the same model and compounds with similar mechanisms of action and target specificities clustered together.

Third, we explored whether the combined analysis of PDTC drug responses and molecular data recapitulated known mechanisms of drug sensitivity and resistance. For example, sensitivity to the EGFR/ERBB2 inhibitor BIBW2992 (afatinib) was seen in two of the three Her2+ models tested (Figure S6A). Sensitivity to PARP inhibition (Drew et al., 2011) was seen in a model with somatic BRCA1 promoter methylation and consequent lack of expression (STG201) and in a model from a patient with a germline-truncating BRCA1 mutation (VHIO124; Figures S6B and S6C; Table S6, and Figure 6 for ex vivo and in vivo data, respectively). Interestingly, two models from BRCA1 germline mutation carriers were resistant to PARP inhibitors, and these had inactivating mutations of 53BP1 (STG316: c.134+3A > C) and MAD2L2 (VHIO179: c.66_67delAG; Table S6; Figure 6). Resistance to PARP inhibitors due to loss of non-homologous end-joining (NHEJ) has been previously reported for both 53BP1 (Bouwman et al., 2010; Bunting et al., 2010; Chapman et al., 2012) and MAD2L2 (Boersma et al., 2015; Xu et al., 2015). These data therefore further demonstrate breast cancer explants recapitulate known mechanisms of both drug sensitivity and resistance. Finally, we explored multiple layers of molecular data in the context of PI3K pathway inhibition. In Figure 4D, we present a schematic of the PI3K-AKT-mTOR pathway to illustrate the complexity of the associations. Sensitivity to the PI3Ka inhibitor GDC00941 (pictilisib) was seen in models with PIK3CA-activating mutations (3/15), PTEN loss (5/15), INPP4B loss (2/15), high p-AKT levels (4/15), or a combination of these features (Table S6). The difference in response (measured by AUC) to pathway inhibitors was compared in the presence or absence of a biomarker in the pathway (based on expression, SNVs, CNAs, or promoter methylation). This showed models with mutant versus wild-type PIK3CA responded better to PI3Ka and AKT inhibitors, but not to mTOR and PI3Kb inhibitors. We did a similar analysis for JQ1, a BET inhibitor recently tested in breast cancer models (Shu et al., 2016). We tested 19 models, and seven were JQ1 sensitive, including 4/7 ER+ (IntClust1 [3] and IntClust10 [1]) and 3/12 ER (IntClust10 [2] and IntClust9 [1]; Table S6). These data highlight the heterogeneous nature of single biomarker/drug-response associations in breast cancer and suggest integrative analysis of molecular and drug response data are more informative. Further improvements are expected in the future using new drug-response metrics that are insensitive to cell division rates. Use of PDTCs to Test Drug-Drug Combinations Combination therapy is increasingly being used as an approach to combat development of resistance in cancer treatment. To

Figure 4. High-Throughput Drug Screening Using PDTCs (A) AUCs scatterplots showing reproducibility of PDTC drug testing. (Left panel) AUCs of technical replicates (n = 6,325; same sample, same compound) are shown. (Right plot) AUCs of biological replicates (n = 1,341; same model, different passages, same compound) are shown. r, Pearson correlation. (B) AUC scatterplots of all drugs targeting PI3K/AKT/mTOR pathway (n = 34 passages from 20 models). Red indicates Pearson correlation > 0.5. (C) AUC scatterplot for cisplatin and BMN-673 treatment across models tested (n = 15). (D) Illustration of the PI3K pathway with panels depicting difference in the AUC in models (n = 15) with versus without molecular alteration in pathway member. (Left panels) Inhibitors of PI3K alpha and PI3Kbeta are shown. (Right panels) Inhibitors of AKT1 and mTOR are shown. See also Figures S4, S5, and S6.

268 Cell 167, 260â&#x20AC;&#x201C;274, September 22, 2016

(legend on next page)

Cell 167, 260â&#x20AC;&#x201C;274, September 22, 2016 269

test the use of PDTC models in high-throughput drug-drug combination assays, we designed a 5 3 5 matrix with standard of care chemotherapy agents (cisplatin and paclitaxel) and six clinically relevant targeted compounds (Figure S7A). Singleagent drug responses in these drug-drug combination assays were highly correlated with those obtained from the 108 individual compound screen (Pearson correlation 0.84), further confirming the robust and reproducible performance of our PDTX/PDTC platform (Figure S7B). The Bliss model (see STAR Methods), an approach that does not require precise estimates of IC50s, was used to compute synergy and antagonism. The performance of the Bliss model was validated by showing in a Her2-positive model (HCI008) synergy of an Hsp90 inhibitor (17-AAG or tanespimycin) in combination with paclitaxel, which has been previously reported (Modi et al., 2011; Figure 5A). The testing of pairwise combinations using the six targeted compounds (Figure S7A) confirmed the rationally predicted synergistic effects of combining an IGFR1/ INSR1 inhibitor (BMS-754807) with a dual PI3K/mTOR inhibitor (NVP-BEZ235) or an EGFR inhibitor (gefitinib/Iressa; Figure 5B). In summary, these data show that PDTCs can be successfully used to test drug-drug combinations. PDTC Testing Predicts In Vivo Drug Responses in PDTXs We next tested whether PDTC drug responses ex vivo predict responses in vivo in a series of pre-clinical trials using PDTXs as xenopatients. This step was crucial to validate the utility of the platform reported here, given that PDTXs have recently been shown to predict human clinical trial drug responses (Gao et al., 2015). We selected 40 ex vivo PDTC drug responses tested in eight different models for in vivo validation. Significantly, even though different compounds with the same specificity had sometimes to be used in PDTXs (for formulation or bioavailability reasons), 33 out of 40 (82.5%) ex vivo drug responses were recapitulated in vivo (Figure 6 for examples; Table S7 for details on all ex vivo and in vivo drug tests performed). This included validation of responses in vivo for PI3K-AKT-mTOR pathway, ER, PARP, Wee1, and IGFR1 inhibitors (Figure 6A; Table S7). We also validated in vivo the predicted synergistic combinations of PI3K plus IGFR1/INSR1 inhibitors (Table S7). Overall, these results show the value of PDTCs as predictive drug response models prior to in vivo testing using PDTXs. DISCUSSION The use of PDTXs in pre-clinical cancer drug development has become widespread (Crystal et al., 2014; Marangoni and Po-

upon, 2014; Messersmith et al., 2009). The data available (DeRose et al., 2011; Eirew et al., 2015; Messersmith et al., 2009), to which we add extensively here, show PDTXs share most molecular and architectural features with their originating patient tumor sample. A recently published large study of 1,000 PDTX models adds a further crucial piece of evidence supporting their potential utility by showing the use of xenograft models to predict human clinical trial drug responses (Gao et al., 2015). The dataset presented here shows the unique value of a living biobank of breast cancer explants that preserve intratumor heterogeneity as a platform for drug screening, including the demonstration of reproducible drug responses across different xenograft passages. A significant limitation of PDTXs as a pre-clinical platform is the fact that in vivo studies are not well suited for high-throughput drug screening. The PDTX/PDTC platform presented here overcomes this limitation, and we have demonstrated its use for both high-throughput single and drug-drug combination studies. The platform has remarkably good reproducibility and selectivity, similar to that observed in analogous studies using cell lines or organoids (Garnett et al., 2012; van de Wetering et al., 2015). The demonstration that compounds affecting the same pathway or target and those with similar mechanisms of action shared the same drug responses across models testifies to its biological robustness. We independently tested a set of drug responses in a selection of models with a DNA-based method, showing very good correlation with CTG results (which is based on ATP levels), as others have recently reported (Haverty et al., 2016). Further refinement of the in vitro screening will come from introducing growth rate inhibition metrics (Hafner et al., 2016). The in vivo validation of 33 out of 40 in-vitro-predicted drug responses tested suggests that, in the future, PDTCs can be used as a drug-screening platform prior to downstream testing with the 1X1X1 PDTX clinical trial design (Gao et al., 2015). Crucially, we found that PDTXs and PDTCs are communities of clones of varying complexity and that these explants display intra-tumor heterogeneity similar to that that is found in the clinical population. The preservation of clonal communities within heterogeneous tumors in pre-clinical models has recently re-emerged as key to improving therapeutic strategies (Heppner, 1984; Tabassum and Polyak, 2015). This feature uniquely positions PDTXs as a human pre-clinical model to study breast cancer biology and drug responses. The framework we developed of ex vivo PDTC drug screening followed by in vivo PDTX response validation is a cost-effective pipeline for pre-clinical drug development. The extensive

Figure 5. Drug-Drug Combination Studies in PDTCs (A) Synergism of paclitaxel in combination with 17-AAG. (Top panel) Bliss independence model residuals for paclitaxel combinations are shown. The 95% percentile of these differences (in percentage) is plotted. For each drug combination, the expected response is compared to the observed response in all the dose ranges in the combination. (Middle panel) Boxplots of distribution of residuals (Bliss independence model) for paclitaxel and 17-AGG combination in each PDTC model tested are shown. (Bottom panel) Detailed analysis for HCI008 (from top to bottom: single drug curves, bivariate isotonic fit for the combination, and residuals of the Bliss model for each dose combination) is shown. Red shades, synergistic effects; blue shades, antagonistic effects. (B) Synergism of IGF-1R/IR inhibitor (BMS-754807) with PI3K/mTOR inhibitor (NVP-BEZ235). Panels are the same as in A (bottom panel: detailed analysis for STG201).

270 Cell 167, 260â&#x20AC;&#x201C;274, September 22, 2016

Figure 6. Validation of Ex Vivo PDTC Drug Responses with In Vivo PDTX Testing Representative sensitive (gray panel) and resistant (pink panel) drug responses in several models. (Left plots) PDTC ex vivo dose response is shown. (Right plots) PDTX in vivo tumor growth curves are shown (sample sizes are indicated in the plot; average values and error bars representing SDs are shown). See also Figure S7.

detailed STAR Methods accompanying this report, including both processed and raw molecular profiling and drug sensitivity information, constitutes a publicly available dataset that we will continue to expand with more models and further drug testing. We will provide viable xenograft fragments to academic collaborators and will also make models available to the wider community through licensing. The extensive data generated already represent a valuable resource, which can be easily browsed in

a purpose-built public web portal (http://caldaslab.cruk.cam. ac.uk/bcape). We are using the PDTX/PDTC platform to study mechanisms of drug resistance, to unravel clonal dynamics in response to therapeutic perturbation, and to perform genomewide perturbations with small hairpin RNA (shRNA) and CRISPR-CAS libraries (Marcotte et al., 2016; Shalem et al., 2015), and these newly generated data will be continually deposited into the public domain.

Cell 167, 260â&#x20AC;&#x201C;274, September 22, 2016 271

STAR+METHODS Detailed methods are provided in the online version of this paper and include the following: d d d

d d d d d d d d

KEY RESOURCES TABLE CONTACT FOR REAGENT AND RESOURCE SHARING EXPERIMENTAL MODEL AND SUBJECT DETAILS B Generation and Maintenance of a Living Biobank of Human Breast Cancer Explants B Generation of Viable PDTX-Derived Tumor Cells B Sample Nomenclature B Sample labeling METHOD DETAILS B Histopathological Review B PKH26 Assay B Centrosome FISH B Flow Cytometric Analysis of Xenograft Mouse Stromal Cell Content B Cell Viability Assays B Treatment of PDTXs In Vivo B Experimental Design QUANTIFICATION AND STATISTICAL ANALYSIS B Details and Number of Samples Analyzed COMPUTATIONAL PIPELINE FOR DISCRIMINATING MOUSE AND HUMAN SEQUENCES WES ANALYSIS sWGS MICROARRAY EXPRESSION ANALYSIS METHYLATION REDUCED REPRESENTATION BISULFITE SEQUENCING ANALYSIS ANALYSIS OF HIGH-THROUGHPUT DRUG SCREENING USING PDTCS DATA AND SOFTWARE AVAILABILITY B Software B Data Resources

ber 260791). M.C. has received funding from the European Union’s Horizon 2020 research and innovation program under the Marie Sk1odowska-Curie grant agreement no. 660060 and was supported by the Department of Experimental Oncology and Molecular Medicine, Fondazione IRCCS Istituto Nazionale dei Tumori, Milan, Italy. R.N.B. is supported by the Wellcome Trust PhD Programme in Mathematical Genomics and Medicine. S-J.S. is supported by the Wellcome Trust PhD Programme for Clinicians in Cambridge. A.Bruna, O.M.R., E.M., V.S., and C.C. are members of the EurOPDX Consortium. We are very grateful for the generosity of all the patients that donated samples for implantation. We are also deeply indebted to all the staff (surgeons, pathologists, oncologists, theatre staff, and other ancillary personnel) at the Cambridge Breast Unit, Cambridge University Hospital NHS Foundation Trust, for facilitating the timely collection of samples. We thank the Cancer Research UK Cambridge Institute Genomics, Bioinformatics, Histopathology, Flow Cytometry, Biological Resource, and Bio-repository Core Facilities for support during the execution of this project. Received: February 5, 2016 Revised: June 21, 2016 Accepted: August 18, 2016 Published: September 15, 2016 REFERENCES Abecasis, G.R., Auton, A., Brooks, L.D., DePristo, M.A., Durbin, R.M., Handsaker, R.E., Kang, H.M., Marth, G.T., and McVean, G.A.; 1000 Genomes Project Consortium (2012). An integrated map of genetic variation from 1,092 human genomes. Nature 491, 56–65. Adzhubei, I., Jordan, D.M., and Sunyaev, S.R. (2013). Predicting functional effect of human missense mutations using PolyPhen-2. Curr. Protoc. Hum. Genet. Chapter 7, Unit7.20. Alexandrov, L.B., Nik-Zainal, S., Wedge, D.C., Aparicio, S.A., Behjati, S., Biankin, A.V., Bignell, G.R., Bolli, N., Borg, A., Børresen-Dale, A.L., et al.; Australian Pancreatic Cancer Genome Initiative; ICGC Breast Cancer Consortium; ICGC MMML-Seq Consortium; ICGC PedBrain (2013). Signatures of mutational processes in human cancer. Nature 500, 415–421. Ali, H.R., Rueda, O.M., Chin, S.F., Curtis, C., Dunning, M.J., Aparicio, S.A., and Caldas, C. (2014). Genome-driven integrated classification of breast cancer validated in over 7,500 samples. Genome Biol. 15, 431. Aparicio, S., and Caldas, C. (2013). The implications of clonal genome evolution for cancer medicine. N. Engl. J. Med. 368, 842–851.

SUPPLEMENTAL INFORMATION

Aparicio, S., Hidalgo, M., and Kung, A.L. (2015). Examining the utility of patient-derived xenograft mouse models. Nat. Rev. Cancer 15, 311–316.

Supplemental Information includes seven figures and seven tables and can be found with this article online at http://dx.doi.org/10.1016/j.cell.2016.08.041.

Bailey, J.A., Gu, Z., Clark, R.A., Reinert, K., Samonte, R.V., Schwartz, S., Adams, M.D., Myers, E.W., Li, P.W., and Eichler, E.E. (2002). Recent segmental duplications in the human genome. Science 297, 1003–1007.

AUTHOR CONTRIBUTIONS

Barretina, J., Caponigro, G., Stransky, N., Venkatesan, K., Margolin, A.A., Kim, S., Wilson, C.J., Lehár, J., Kryukov, G.V., Sonkin, D., et al. (2012). The Cancer Cell Line Encyclopedia enables predictive modelling of anticancer drug sensitivity. Nature 483, 603–607.

Conceptualization, A. Bruna, O.M.R., M.J.G., and C.C.; Methodology, A. Bruna, W.G., A.S.B., J.S., J.W.C., A.T.-V., A.M.-D., A. Barthorpe, H.L., M.J.O., E.M., A.L.W., V.S., S.A., M.J.G., and C.C.; Software, O.M.R., M.C., R.N.B., K.P., P.E., and M.E.; Validation, A. Bruna, O.M.R., M.J.G., and C.C.; Formal Analysis, O.M.R., M.C., R.N.B., K.P., P.E., and S.A.; Investigation, A. Bruna, W.G., A.S.B., J.S., J.W.C., A.T.-V., J.H., A.M.-D., A. Barthorpe, H.L., M.J.O., and V.S.; Resources, S.-J.S., L.J., E.P., R.B., J.G., J.C., J.B., E.M., A.L.W., and V.S.; Data Curation, A. Bruna, O.M.R., W.G., A.S.B., M.J.G., and C.C.; Writing–Original Draft, A. Bruna and C.C.; Writing–Review & Editing, A. Bruna, O.M.R., P.E., S.A., M.J.G., and C.C.; Visualization, A. Bruna, O.M.R., M.E., and C.C.; Supervision, A. Bruna, O.M.R., M.J.G., and C.C.; Project Administration, A. Bruna, O.M.R., M.J.G., and C.C.; Funding Acquisition, C.C. ACKNOWLEDGMENTS This research was supported with funding from Cancer Research UK and from the European Union to the EUROCAN Network of Excellence (FP7; grant num-

272 Cell 167, 260–274, September 22, 2016

Bliss, C.I. (1939). The toxicity of poisons applied jointly. Ann. Appl. Biol. 26, 585–615. Boersma, V., Moatti, N., Segura-Bayona, S., Peuscher, M.H., van der Torre, J., Wevers, B.A., Orthwein, A., Durocher, D., and Jacobs, J.J. (2015). MAD2L2 controls DNA repair at telomeres and DNA breaks by inhibiting 50 end resection. Nature 521, 537–540. Bouwman, P., Aly, A., Escandell, J.M., Pieterse, M., Bartkova, J., van der Gulden, H., Hiddingh, S., Thanasoula, M., Kulkarni, A., Yang, Q., et al. (2010). 53BP1 loss rescues BRCA1 deficiency and is associated with triple-negative and BRCA-mutated breast cancers. Nat. Struct. Mol. Biol. 17, 688–695. Bunting, S.F., Callén, E., Wong, N., Chen, H.T., Polato, F., Gunn, A., Bothmer, A., Feldhahn, N., Fernandez-Capetillo, O., Cao, L., et al. (2010). 53BP1 inhibits homologous recombination in Brca1-deficient cells by blocking resection of DNA breaks. Cell 141, 243–254.

Cancer Genome Atlas Network (2012). Comprehensive molecular portraits of human breast tumours. Nature 490, 61–70. Cassidy, J.W., Caldas, C., and Bruna, A. (2015). Maintaining tumor heterogeneity in patient-derived tumor xenografts. Cancer Res. 75, 2963–2968. Chapman, J.R., Sossick, A.J., Boulton, S.J., and Jackson, S.P. (2012). BRCA1-associated exclusion of 53BP1 from DNA damage sites underlies temporal control of DNA repair. J. Cell Sci. 125, 3529–3534. Crystal, A.S., Shaw, A.T., Sequist, L.V., Friboulet, L., Niederst, M.J., Lockerman, E.L., Frias, R.L., Gainor, J.F., Amzallag, A., Greninger, P., et al. (2014). Patient-derived models of acquired resistance can identify effective drug combinations for cancer. Science 346, 1480–1486. Curtis, C., Shah, S.P., Chin, S.F., Turashvili, G., Rueda, O.M., Dunning, M.J., Speed, D., Lynch, A.G., Samarajiwa, S., Yuan, Y., et al.; METABRIC Group (2012). The genomic and transcriptomic architecture of 2,000 breast tumours reveals novel subgroups. Nature 486, 346–352. Dawson, S.J., Rueda, O.M., Aparicio, S., and Caldas, C. (2013). A new genome-driven integrated classification of breast cancer and its implications. EMBO J. 32, 617–628. DePristo, M.A., Banks, E., Poplin, R., Garimella, K.V., Maguire, J.R., Hartl, C., Philippakis, A.A., del Angel, G., Rivas, M.A., Hanna, M., et al. (2011). A framework for variation discovery and genotyping using next-generation DNA sequencing data. Nat. Genet. 43, 491–498. DeRose, Y.S., Wang, G., Lin, Y.C., Bernard, P.S., Buys, S.S., Ebbert, M.T., Factor, R., Matsen, C., Milash, B.A., Nelson, E., et al. (2011). Tumor grafts derived from women with breast cancer authentically reflect tumor pathology, growth, metastasis and disease outcomes. Nat. Med. 17, 1514–1520. Ding, L., Ellis, M.J., Li, S., Larson, D.E., Chen, K., Wallis, J.W., Harris, C.C., McLellan, M.D., Fulton, R.S., Fulton, L.L., et al. (2010). Genome remodelling in a basal-like breast cancer metastasis and xenograft. Nature 464, 999–1005. Dong, C., Wei, P., Jian, X., Gibbs, R., Boerwinkle, E., Wang, K., and Liu, X. (2015). Comparison and integration of deleteriousness prediction methods for nonsynonymous SNVs in whole exome sequencing studies. Hum. Mol. Genet. 24, 2125–2137. Drew, Y., Mulligan, E.A., Vong, W.T., Thomas, H.D., Kahn, S., Kyle, S., Mukhopadhyay, A., Los, G., Hostomsky, Z., Plummer, E.R., et al. (2011). Therapeutic potential of poly(ADP-ribose) polymerase inhibitor AG014699 in human cancers with mutated or methylated BRCA1 or BRCA2. J. Natl. Cancer Inst. 103, 334–346. Dunning, M.J., Smith, M.L., Ritchie, M.E., and Tavaré, S. (2007). beadarray: R classes and methods for Illumina bead-based data. Bioinformatics 23, 2183– 2184. Dvinge, H., Git, A., Gräf, S., Salmon-Divon, M., Curtis, C., Sottoriva, A., Zhao, Y., Hirst, M., Armisen, J., Miska, E.A., et al. (2013). The shaping and functional consequences of the microRNA landscape in breast cancer. Nature 497, 378–382. Eirew, P., Steif, A., Khattra, J., Ha, G., Yap, D., Farahani, H., Gelmon, K., Chia, S., Mar, C., Wan, A., et al. (2015). Dynamics of genomic clones in breast cancer patient xenografts at single-cell resolution. Nature 518, 422–426. Fraley, C., and Raftery, A.E. (2002). Model-based clustering, discriminant analysis, and density estimation. J. Am. Stat. Assoc. 97, 611–631. Gao, H., Korn, J.M., Ferretti, S., Monahan, J.E., Wang, Y., Singh, M., Zhang, C., Schnell, C., Yang, G., Zhang, Y., et al. (2015). High-throughput screening using patient-derived tumor xenografts to predict clinical trial drug response. Nat. Med. 21, 1318–1325. Garnett, M.J., Edelman, E.J., Heidorn, S.J., Greenman, C.D., Dastur, A., Lau, K.W., Greninger, P., Thompson, I.R., Luo, X., Soares, J., et al. (2012). Systematic identification of genomic markers of drug sensitivity in cancer cells. Nature 483, 570–575. Greco, W.R., Bravo, G., and Parsons, J.C. (1995). The search for synergy: a critical review from a response surface perspective. Pharmacol. Rev. 47, 331–385.

Hafner, M., Niepel, M., Chung, M., and Sorger, P.K. (2016). Growth rate inhibition metrics correct for confounders in measuring sensitivity to cancer drugs. Nat. Methods 13, 521–527. Haibe-Kains, B., Desmedt, C., Loi, S., Culhane, A.C., Bontempi, G., Quackenbush, J., and Sotiriou, C. (2012). A three-gene model to robustly identify breast cancer molecular subtypes. J. Natl. Cancer Inst. 104, 311–325. Hänzelmann, S., Castelo, R., and Guinney, J. (2013). GSVA: gene set variation analysis for microarray and RNA-seq data. BMC Bioinformatics 14, 7. Haverty, P.M., Lin, E., Tan, J., Yu, Y., Lam, B., Lianoglou, S., Neve, R.M., Martin, S., Settleman, J., Yauch, R.L., and Bourgon, R. (2016). Reproducible pharmacogenomic profiling of cancer cell line panels. Nature 533, 333–337. Heiser, L.M., Sadanandam, A., Kuo, W.L., Benz, S.C., Goldstein, T.C., Ng, S., Gibb, W.J., Wang, N.J., Ziyad, S., Tong, F., et al. (2012). Subtype and pathway specific responses to anticancer compounds in breast cancer. Proc. Natl. Acad. Sci. USA 109, 2724–2729. Heppner, G.H. (1984). Tumor heterogeneity. Cancer Res. 44, 2259–2265. Hidalgo, M., Amant, F., Biankin, A.V., Budinská, E., Byrne, A.T., Caldas, C., Clarke, R.B., de Jong, S., Jonkers, J., Mælandsmo, G.M., et al. (2014). Patient-derived xenograft models: an emerging platform for translational cancer research. Cancer Discov. 4, 998–1013. Huber, W., Carey, V.J., Gentleman, R., Anders, S., Carlson, M., Carvalho, B.S., Bravo, H.C., Davis, S., Gatto, L., Girke, T., et al. (2015). Orchestrating highthroughput genomic analysis with Bioconductor. Nat. Methods 12, 115–121. Jurasinski, G., Koebsch, F., Guenther, A., and Beetz, S. (2014). flux: Flux rate calculation from dynamic closed chamber measurements. Kaelin, W.G., Jr. (2005). The concept of synthetic lethality in the context of anticancer therapy. Nat. Rev. Cancer 5, 689–698. Kim, H.S., Sung, Y.J., and Paik, S. (2015). Cancer cell line panels empower genomics-based discovery of precision cancer medicine. Yonsei Med. J. 56, 1186–1198. Krueger, F., and Andrews, S.R. (2011). Bismark: a flexible aligner and methylation caller for Bisulfite-Seq applications. Bioinformatics 27, 1571–1572. Kuilman, T., Velds, A., Kemper, K., Ranzani, M., Bombardelli, L., Hoogstraat, M., Nevedomskaya, E., Xu, G., de Ruiter, J., Lolkema, M.P., et al. (2015). CopywriteR: DNA copy number detection from off-target sequence data. Genome Biol. 16, 49. Kumar, P., Henikoff, S., and Ng, P.C. (2009). Predicting the effects of coding non-synonymous variants on protein function using the SIFT algorithm. Nat. Protoc. 4, 1073–1081. Lawson, D.A., Bhakta, N.R., Kessenbrock, K., Prummel, K.D., Yu, Y., Takai, K., Zhou, A., Eyob, H., Balakrishnan, S., Wang, C.Y., et al. (2015). Single-cell analysis reveals a stem-cell program in human metastatic breast cancer cells. Nature 526, 131–135. Li, H., and Durbin, R. (2009). Fast and accurate short read alignment with Burrows-Wheeler transform. Bioinformatics 25, 1754–1760. Li, H., Handsaker, B., Wysoker, A., Fennell, T., Ruan, J., Homer, N., Marth, G., Abecasis, G., and Durbin, R.; 1000 Genome Project Data Processing Subgroup (2009). The Sequence Alignment/Map format and SAMtools. Bioinformatics 25, 2078–2079. Li, S., Shen, D., Shao, J., Crowder, R., Liu, W., Prat, A., He, X., Liu, S., Hoog, J., Lu, C., et al. (2013). Endocrine-therapy-resistant ESR1 variants revealed by genomic characterization of breast-cancer-derived xenografts. Cell Rep. 4, 1116–1130. Liberzon, A., Subramanian, A., Pinchback, R., Thorvaldsdóttir, H., Tamayo, P., and Mesirov, J.P. (2011). Molecular signatures database (MSigDB) 3.0. Bioinformatics 27, 1739–1740. Luo, J., Solimini, N.L., and Elledge, S.J. (2009). Principles of cancer therapy: oncogene and non-oncogene addiction. Cell 136, 823–837. Marangoni, E., and Poupon, M.F. (2014). Patient-derived tumour xenografts as models for breast cancer drug development. Curr. Opin. Oncol. 26, 556–561. Marangoni, E., Vincent-Salomon, A., Auger, N., Degeorges, A., Assayag, F., de Cremoux, P., de Plater, L., Guyader, C., De Pinieux, G., Judde, J.G., et al.

Cell 167, 260–274, September 22, 2016 273

(2007). A new model of patient tumor-derived breast cancer xenografts for preclinical assays. Clin. Cancer Res. 13, 3989–3998.

in a lobular breast tumour profiled at single nucleotide resolution. Nature 461, 809–813.

Marcotte, R., Sayad, A., Brown, K.R., Sanchez-Garcia, F., Reimand, J., Haider, M., Virtanen, C., Bradner, J.E., Bader, G.D., Mills, G.B., et al. (2016). Functional genomic landscape of human breast cancer drivers, vulnerabilities, and resistance. Cell 164, 293–309.

Shah, S.P., Roth, A., Goya, R., Oloumi, A., Ha, G., Zhao, Y., Turashvili, G., Ding, J., Tse, K., Haffari, G., et al. (2012). The clonal and mutational evolution spectrum of primary triple-negative breast cancers. Nature 486, 395–399.

Messersmith, W.A., Rajeshkumar, N.V., Tan, A.C., Wang, X.F., Diesl, V., Choe, S.E., Follettie, M., Coughlin, C., Boschelli, F., Garcia-Garcia, E., et al. (2009). Efficacy and pharmacodynamic effects of bosutinib (SKI-606), a Src/Abl inhibitor, in freshly generated human pancreas cancer xenografts. Mol. Cancer Ther. 8, 1484–1493.

Shalem, O., Sanjana, N.E., and Zhang, F. (2015). High-throughput functional genomics using CRISPR-Cas9. Nat. Rev. Genet. 16, 299–311. Sharma, S.V., Haber, D.A., and Settleman, J. (2010). Cell line-based platforms to evaluate the therapeutic efficacy of candidate anticancer agents. Nat. Rev. Cancer 10, 241–253.

Meyer, M.C., Wu, J., and Opsomer, J.D. (2014). isotonic.pen: Penalized Isotonic Regression in one and two dimensions.

Sherry, S.T., Ward, M.H., Kholodov, M., Baker, J., Phan, L., Smigielski, E.M., and Sirotkin, K. (2001). dbSNP: the NCBI database of genetic variation. Nucleic Acids Res. 29, 308–311.

Modi, S., Stopeck, A., Linden, H., Solit, D., Chandarlapaty, S., Rosen, N., D’Andrea, G., Dickler, M., Moynahan, M.E., Sugarman, S., et al. (2011). HSP90 inhibition is effective in breast cancer: a phase II trial of tanespimycin (17-AAG) plus trastuzumab in patients with HER2-positive metastatic breast cancer progressing on trastuzumab. Clin. Cancer Res. 17, 5132–5139.

Shu, S., Lin, C.Y., He, H.H., Witwicki, R.M., Tabassum, D.P., Roberts, J.M., Janiszewska, M., Huh, S.J., Liang, Y., Ryan, J., et al. (2016). Response and resistance to BET bromodomain inhibitors in triple-negative breast cancer. Nature 529, 413–417.

Mroz, E.A., and Rocco, J.W. (2013). MATH, a novel measure of intratumor genetic heterogeneity, is high in poor-outcome classes of head and neck squamous cell carcinoma. Oral Oncol. 49, 211–215. Murtaza, M., Dawson, S.J., Pogrebniak, K., Rueda, O.M., Provenzano, E., Grant, J., Chin, S.F., Tsui, D.W., Marass, F., Gale, D., et al. (2015). Multifocal clonal evolution characterized using circulating tumour DNA in a case of metastatic breast cancer. Nat. Commun. 6, 8760. Obenchain, V., Lawrence, M., Carey, V., Gogarten, S., Shannon, P., and Morgan, M. (2014). VariantAnnotation: a Bioconductor package for exploration and annotation of genetic variants. Bioinformatics 30, 2076–2078. Olshen, A.B., Venkatraman, E.S., Lucito, R., and Wigler, M. (2004). Circular binary segmentation for the analysis of array-based DNA copy number data. Biostatistics 5, 557–572. Parker, J.S., Mullins, M., Cheang, M.C.U., Leung, S., Voduc, D., Vickery, T., Davies, S., Fauron, C., He, X., Hu, Z., et al. (2009). Supervised risk predictor of breast cancer based on intrinsic subtypes. J. Clin. Oncol. 27, 1160–1167. Pereira, B., Chin, S.F., Rueda, O.M., Vollan, H.K., Provenzano, E., Bardwell, H.A., Pugh, M., Jones, L., Russell, R., Sammut, S.J., et al. (2016). Erratum: The somatic mutation profiles of 2,433 breast cancers refine their genomic and transcriptomic landscapes. Nat. Commun. 7, 11908. R Core Team (2016). R: A language and environment for statistical computing. Rosenthal, R., McGranahan, N., Herrero, J., Taylor, B.S., and Swanton, C. (2016). DeconstructSigs: delineating mutational processes in single tumors distinguishes DNA repair deficiencies and patterns of carcinoma evolution. Genome Biol. 17, 31. Roth, A., Khattra, J., Yap, D., Wan, A., Laks, E., Biele, J., Ha, G., Aparicio, S., Bouchard-Côté, A., and Shah, S.P. (2014). PyClone: statistical inference of clonal population structure in cancer. Nat. Methods 11, 396–398. Scheinin, I., Sie, D., Bengtsson, H., van de Wiel, M.A., Olshen, A.B., van Thuijl, H.F., van Essen, H.F., Eijk, P.P., Rustenburg, F., Meijer, G.A., et al. (2014). DNA copy number analysis of fresh and formalin-fixed specimens by shallow whole-genome sequencing with identification and exclusion of problematic regions in the genome assembly. Genome Res. 24, 2022–2032. Schwarz, J.M., Rödelsperger, C., Schuelke, M., and Seelow, D. (2010). MutationTaster evaluates disease-causing potential of sequence alterations. Nat. Methods 7, 575–576. Shah, S.P., Morin, R.D., Khattra, J., Prentice, L., Pugh, T., Burleigh, A., Delaney, A., Gelmon, K., Guliany, R., Senz, J., et al. (2009). Mutational evolution

274 Cell 167, 260–274, September 22, 2016

Subramanian, A., Tamayo, P., Mootha, V.K., Mukherjee, S., Ebert, B.L., Gillette, M.A., Paulovich, A., Pomeroy, S.L., Golub, T.R., Lander, E.S., and Mesirov, J.P. (2005). Gene set enrichment analysis: a knowledge-based approach for interpreting genome-wide expression profiles. Proc. Natl. Acad. Sci. USA 102, 15545–15550. Tabassum, D.P., and Polyak, K. (2015). Tumorigenesis: it takes a village. Nat. Rev. Cancer 15, 473–483. Tamayo, P., Cho, Y.-J., Tsherniak, A., Greulich, H., Ambrogio, L., Schoutenvan Meeteren, N., Zhou, T., Buxton, A., Kool, M., Meyerson, M., et al. (2011). Predicting relapse in patients with medulloblastoma by integrating evidence from clinical and genomic features. J. Clin. Oncol. 29, 1415–1423. Tentler, J.J., Tan, A.C., Weekes, C.D., Jimeno, A., Leong, S., Pitts, T.M., Arcaroli, J.J., Messersmith, W.A., and Eckhardt, S.G. (2012). Patient-derived tumour xenografts as models for oncology drug development. Nat. Rev. Clin. Oncol. 9, 338–350. Torti, D., and Trusolino, L. (2011). Oncogene addiction as a foundational rationale for targeted anti-cancer therapy: promises and perils. EMBO Mol. Med. 3, 623–636. van de Wetering, M., Francies, H.E., Francis, J.M., Bounova, G., Iorio, F., Pronk, A., van Houdt, W., van Gorp, J., Taylor-Weiner, A., Kester, L., et al. (2015). Prospective derivation of a living organoid biobank of colorectal cancer patients. Cell 161, 933–945. Van Loo, P., Nordgard, S.H., Lingjærde, O.C., Russnes, H.G., Rye, I.H., Sun, W., Weigman, V.J., Marynen, P., Zetterberg, A., Naume, B., et al. (2010). Allele-specific copy number analysis of tumors. Proc. Natl. Acad. Sci. USA 107, 16910–16915. Vogelstein, B., Papadopoulos, N., Velculescu, V.E., Zhou, S., Diaz, L.A., Jr., and Kinzler, K.W. (2013). Cancer genome landscapes. Science 339, 1546– 1558. Wang, K., Li, M., and Hakonarson, H. (2010). ANNOVAR: functional annotation of genetic variants from high-throughput sequencing data. Nucleic Acids Res. 38, e164. Wood, S.N. (2004). Stable and efficient multiple smoothing parameter estimation for generalized additive models. J. Am. Stat. Assoc. 99, 673–686. Xu, G., Chapman, J.R., Brandsma, I., Yuan, J., Mistrik, M., Bouwman, P., Bartkova, J., Gogola, E., Warmerdam, D., Barazas, M., et al. (2015). REV7 counteracts DNA double-strand break resection and affects PARP inhibition. Nature 521, 541–544.

Cell Career Network just got better!

Additional reach ch via exte extended network Post a job within 5 minutes via self service More intuitive feel and easier navigation New vacancy packages and pricing

Your recruitment process just became even easier. Get started today!

careers.cell.com/recruiters

Cancer Cell

Article Inhibition of TRF1 Telomere Protein Impairs Tumor Initiation and Progression in Glioblastoma Mouse Models and Patient-Derived Xenografts Leire Bejarano,1 Alberto J. Schuhmacher,2 Marinela Méndez,1 Diego Megı́as,3 Carmen Blanco-Aparicio,4 Sonia Martı́nez,4 Joaquı́n Pastor,4 Massimo Squatrito,2 and Maria A. Blasco1,5,* 1Telomeres and Telomerase Group, Molecular Oncology Program, Spanish National Cancer Research Centre (CNIO), Melchor Fernández Almagro 3, Madrid, 28029, Spain 2Seve-Ballesteros Foundation Brain Tumor Group, Cancer Cell Biology Program, Spanish National Cancer Centre (CNIO), Melchor Fernández Almagro 3, Madrid, 28029, Spain 3Confocal Microscopy Unit, Biotechnology Program, Spanish National Cancer Research Centre (CNIO), Madrid, 28029 Spain 4Experimental Therapeutics Program, Spanish National Cancer Centre (CNIO), Melchor Fernández Almagro 3, Madrid, 28029, Spain 5Lead Contact *Correspondence: mblasco@cnio.es https://doi.org/10.1016/j.ccell.2017.10.006

SUMMARY

Glioblastoma multiforme (GBM) is a deadly and common brain tumor. Poor prognosis is linked to high proliferation and cell heterogeneity, including glioma stem cells (GSCs). Telomere genes are frequently mutated. The telomere binding protein TRF1 is essential for telomere protection, and for adult and pluripotent stem cells. Here, we find TRF1 upregulation in mouse and human GBM. Brain-specific Trf1 genetic deletion in GBM mouse models inhibited GBM initiation and progression, increasing survival. Trf1 deletion increased telomeric DNA damage and reduced proliferation and stemness. TRF1 chemical inhibitors mimicked these effects in human GBM cells and also blocked tumor sphere formation and tumor growth in xenografts from patient-derived primary GSCs. Thus, targeting telomeres throughout TRF1 inhibition is an effective therapeutic strategy for GBM.

INTRODUCTION Malignant gliomas represent the majority of primary CNS neoplasms. The most frequent and aggressive glioma is glioblastoma multiforme (GBM) (Louis et al., 2007). Despite all the advances in the molecular characterization of glioblastoma, the median survival has not improved in the last 50 years, being only 14–16 months (Wen and Kesari, 2008). GBM is known for the high proliferative and infiltrative nature (Chen et al., 2012). GBMs are also highly heterogeneous tumors (Soeda et al., 2015). Cells within the tumor present different expression profiles and may have different responses to radioand chemotherapy (Bhat et al., 2013). In addition, several studies

suggest the existence of a small fraction of cells within the bulk of the tumor with stem-like properties, also termed glioma stem-like cells. These cells are able to recapitulate the original tumor after injection into the brain of immunodeficient mice (Singh et al., 2004). They exhibit radio- and chemoresistant properties, which might explain GBM recurrence after treatment (Bao et al., 2006). This complexity highlights the need for new effective treatments. Telomeres are heterochromatic structures at the end of chromosomes essential for chromosome stability (De Lange, 2005). Mammalian telomeres are formed by tandem repeats of the TTAGGG sequence bound by the so-called shelterin complex, formed by TRF1, TRF2, POT1, TPP1, TIN2, and RAP1

Significance Glioblastoma is an incurable tumor mainly owing to its high tumor-initiating cell potential. Telomere maintenance is among the most frequent alterations in human glioblastoma, but if telomeres are good targets to cease GBM growth remains unclear. This study demonstrates that disrupting telomeric capping through direct inhibition of the shelterin protein TRF1 is a promising strategy for treating GBM. We demonstrate that inhibition of TRF1 effectively blocks GBM in both murine and human GBM models. We further demonstrate the striking effectiveness of TRF1 inhibition in impairing the growth of glioma stem cells. These results have a potential impact in cancer treatment as current therapies are unable to kill glioma stem cells and patients die owing to the recurrence of the tumors. 590 Cancer Cell 32, 590–607, November 13, 2017 ª 2017 Elsevier Inc.

Figure 1. Trf1 Is Upregulated in Mouse and Human GBM (A) Representative images (left) and quantification (right) of percentage of cells with high TRF1 expression determined by immunofluorescence. Scale bars, 10 mm. (B) Western blot images (left) and quantification (right) of TRF1 protein levels in the indicated cells. (C) Generation of mouse models of GBM by overexpression of PDGFB or PDGFA or by knock down of Nf1 and p53 in Nestin-expressing cells (i.e., glial progenitors). (D) TRF1 mRNA levels by qRT-PCR in GBM subtypes compared with non-tumor areas. (E) Quantification of nuclear TRF1 fluorescence in tumor and non-tumor areas and representative images. Scale bars, 5 mm. (legend continued on next page)

Cancer Cell 32, 590â&#x20AC;&#x201C;607, November 13, 2017 591

(Liu et al., 2004). With each cell division, telomeres shorten due to the incomplete replication of chromosome ends, a phenomenon known as the ‘‘end-replication’’ problem (Harley et al., 1990). This telomere shortening can be compensated through the de novo addition of telomeric repeats by telomerase, a reverse transcriptase composed of a catalytic subunit (TERT) and an RNA component (Terc), used as template for the synthesis of TTAGGG repeats (Greider and Blackburn, 1985). In normal adult cells, however, telomerase is not usually expressed, and telomeres progressively shorten associated to organismal aging (Martı́nez and Blasco, 2011; Harley et al., 1990). Telomere maintenance above a minimum length is essential to sustain the indefinite proliferation potential of cancer cells, thus telomeres are considered as potential anti-cancer targets (Hanahan and Weinberg, 2011; Kim et al., 1994). More than 90% of human tumors aberrantly express telomerase (Kim et al., 1994; Shay and Bacchetti, 1997), while the remaining telomerase-negative tumors activate an alternative mechanism to elongate telomeres based on recombination between telomeric sequences, known as ALT (Bryan et al., 1997). The promoter of the TERT gene is mutated in 58%–84% of human primary GBMs (Nonoguchi et al., 2013; Boldrini et al., 2006), while pediatric GBMs frequently display an ALT phenotype associated with ATRX mutations (Heaphy et al., 2011). These facts highlight the importance of telomere maintenance in glioblastoma and pinpoint telomeres as promising targets. In this regard, most studies have focused on telomerase inhibition. However, telomerase-deficient mice are only cancer resistant when telomeres reach a critically short length (Gonzalez-Suarez et al., 2000) and this effect is lost in the absence of the Trp53 tumor suppressor gene, which is commonly mutated in cancer (Chin et al., 1999). Clinical trials with telomerase inhibitors have only shown therapeutic benefits in a few myeloid malignancies but have largely failed in solid tumors (El Fassi et al., 2015; Middleton et al., 2014; Parkhurst et al., 2004), maybe as a consequence of telomere length heterogeneity within tumors, which may hamper the effective killing of all tumor cells including the tumor-initiating populations with stem cell-like properties. Several studies suggest that inhibiting TRF1 could represent an alternative to telomerase inhibitors to target telomeres independently of telomere length. TRF1 directly binds TTAGGG telomere repeats where it is essential for telomere protection (De Lange, 2005; Martı́nez and Blasco, 2011). Trf1 deletion in vivo induces a persistent DNA damage response (DDR) at telomeres, which is sufficient to block cell division and induce senescence and/or apoptosis in different mouse tissues (Martı́nez et al., 2009; Beier et al., 2012; Schneider et al., 2013). Interestingly, TRF1 is overexpressed in adult stem cell compartments and in pluripotent stem cells, where it is essential to maintain tissue homeostasis and pluripotency, respectively (Schneider et al., 2013; Boué et al., 2010). TRF1 is also overexpressed in several cancer types such as renal cell carcinoma (Pal et al., 2015) and gastrointestinal tumors (Hu et al., 2010). Furthermore, dominant nega-

tive mutations in the TRF1-interacting protein POT1 have been found in different tumor types including familiar glioblastoma cases, again highlighting the importance of telomeres in GBM (Bainbridge et al., 2015; Ramsay et al., 2013; Calvete et al., 2015; Robles-Espinoza et al., 2014). Finally, we recently reported that induction of telomere uncapping by Trf1 genetic depletion or chemical inhibition can effectively block the growth of aggressive and rapidly growing lung tumors in Trp53-deficient KrasG12V mice, in a manner that is independent of telomere length (Garcı́a-Beccaria et al., 2015), thus further supporting that TRF1 could be a good anti-cancer target for aggressive tumors. Here, we set to address whether TRF1 inhibition blocks GBM growth in both in vivo mouse models and human xenograft models, and to address whether these effects occur independently of telomere length. RESULTS TRF1 Is Overexpressed in Both Human and Mouse GBM We first addressed whether TRF1 expression is altered in human and mouse GBM. For this, we analyzed TRF1 protein levels by immunofluorescence in a total of 30 normal human brains, 7 astrocytomas, and 14 GBMs. The percentage of cells presenting high TRF1 levels was highest in GBMs, followed by astrocytomas, while TRF1 was almost undetectable in normal brain tissue (Figure 1A). We validated these results by determining TRF1 total protein levels by western blot in three independent human GBM cell lines and in two patient-derived primary glioblastoma stem cell (GSC) cultures. Again, TRF1 protein expression was significantly increased in all three human GBM cell lines (U251, U87, and T98G) and in the two patient-derived primary GSCs cultures (h543 and h676) compared with primary astrocytes (Figure 1B). TRF2 and RAP1 protein levels were also found upregulated in the patient-derived primary GSCs compared with normal astrocytes (Figure S1A). This upregulation of different shelterin proteins in human GBM cells compared with normal astrocytes, seemed to occur at the post-transcriptional level as we did not find significant differences in the mRNA levels of different shelterins when using qRT-PCR (Figure S1B). Next, we studied TRF1 expression in various mouse models of GBM generated by specifically targeting Nestin-expressing cells with RCAS vectors carrying different oncogenic insults in NestinTva transgenic mice. Neural stem cells (NSCs) express Nestin, and, thus, these cells are targeted by the different oncogenic insults. In particular, we generated mice with different GBM subtypes by either overexpressing PDGFB or PDGFA in a Cdkn2a null background, or by knocking down Nf1 and p53 in a wild-type background (Figures 1C and S1C–S1F). PDGFA overexpression results in proneural-like GBMs, while PDGFB and sh-Nf1 sh-p53 induced glioblastomas with a mesenchymal signature (Ozawa et al., 2014) (Figures S1C–S1F). TRF1 mRNA levels were significantly upregulated in the three mouse GBM

(F) Western blot for TRF1 protein levels in PDGFB tumors (T) and non-tumor (NT) areas. (G) Telomere Q-FISH analysis of tumor and non-tumor areas. Scale bars, 5 mm. Data are represented as the mean ± SD with the exception of 1A, which is the mean ± SEM. n represents the number of independent human samples in (A), the number of biological replicates in (B), and the number of mice in (D–G). Statistical analysis: unpaired t test. *p < 0.05, **p < 0.01, ***p < 0.001. See also Figures S1 and S2.

592 Cancer Cell 32, 590–607, November 13, 2017

Figure 2. Trf1 Deletion Impairs Tumor Initiation in Various GBM Subtypes (A) GBM are induced by PDGFB overexpression simultaneously with Cre expression to delete the Trf1lox allele, specifically in Nestin-expressing cells. (B) Schematic representation of the experimental procedures. (C) Survival curves of mice with the indicated genotypes. Histological analysis is performed 45 days after tumor induction in a different cohort of mice. (D) Percentage of mice with GBM 45 days after tumor induction. (E) Representative images (left) and quantification (right) of tumor area by H&E at 45 days after tumor induction. Scale bars, 500 mm (left) and 200 mm (right). (F) Representative images (left) and quantification (right) of Ki67-positive cells per field at 45 days after tumor induction. Scale bar, 100 mm. (legend continued on next page)

Cancer Cell 32, 590â&#x20AC;&#x201C;607, November 13, 2017 593

subtypes, with highest levels in PDGFA- and PDGFB-induced GBM (Figure 1D). The TRF1 protein was also upregulated in all GBM subtypes, again with PDGFB- and PDGFA-induced tumors showing the highest TRF1 levels (Figure 1E). TRF1 protein upregulation was validated by western blot in PDGFB-induced tumors compared with normal tissue (Figure 1F). Next, we addressed whether other shelterin components, namely TRF2, RAP1, POT1, TPP1, and TIN2 (Liu et al., 2004), were also upregulated in mouse GBM models. TPP1, POT1, and TIN2 mRNA levels were slightly higher in the three GBM subtypes by qRT-PCR, but the differences did not reach statistical significance with the exception of TIN2 upregulation in PDGFB-induced GBM and TPP1 upregulation in PDGFAinduced tumors (Figure S1G). RAP1 was significantly downregulated in PDGFB and PDGFA-induced tumors and TRF2 did not change in any of the GBM subtypes (Figure S1G). To assess whether higher TRF1 levels were the consequence of longer telomeres in tumors compared with normal tissue, we measured telomere length by quantitative telomere FISH (Q-FISH) in the different mouse GBM subtypes. We did not find any significant differences in telomere length between tumors and non-tumoral tissue in any of the GBM subtypes (Figure 1G). This is in agreement with previous findings showing that high TRF1 levels in pluripotent and adult stem cells are uncoupled from telomere length (Marion et al., 2009; Tejera et al., 2010; Schneider et al., 2013). We next studied whether TRF1 levels correlated with the well-known stem cell markers SOX2, NESTIN, CD133, and MYC, using the Gliovis data portal for analysis of GBM expression datasets (Bowman et al., 2017). We found a positive correlation between TRF1 levels and all the stem cell markers with the exception of MYC (Figures S2A– S2D), although TRF1 was positively correlated with the MYC modulator USP13 (Fang et al., 2017) (Figure S2E). These findings suggest that TRF1 overexpression in GBM is not the simple consequence of longer telomeres in these tumors, but instead may reflect their high cancer stem cell nature as TRF1 is upregulated in stem cells and pluripotent stem cells (Lathia et al., 2015; Schneider et al., 2013). In summary, all three mouse GBM subtypes upregulate TRF1 in a manner that is independent of telomere length. Trf1 Genetic Deletion Impairs Tumor Initiation in Different Mouse GBM Subtypes We next set to genetically validate TRF1 as a potential anti-cancer target in the GBM subtypes studied here. We first studied the impact of Trf1 abrogation in the PDGFB-induced GBM model, as it showed the highest TRF1 overexpression. To this end, we mice with Trf1 inducible crossed Nestin-Tva;Cdkn2a / knockout mice (Martı́nez et al., 2009) to obtain Trf1lox/lox;NestinTva;Cdkn2a / or Trf1+/+;Nestin-Tva;Cdkn2a / mice (Figure 2A). Then, we injected, intracranially into the subventricular zone (SVZ) of adult mice (4.5–6 weeks), the RCAS-PDGFB DF-1-producing cells together with RCAS-Cre-producing cells to simultaneously delete Trf1 in a 1:3 ratio (the number of RCAS-Cre-producing cells

was three times higher than the RCAS-PDGFB-producing cells). This results in overexpression of PDGFB in glial progenitors simultaneously with Cre-mediated Trf1 excision specifically in these cells (Figures 2A, 2B, and S1D), allowing assessment of the impact of Trf1 abrogation on tumor initiation. We validated Trf1 excision by injecting RCAS-Cre-producing DF-1 cells into the brain of 2 days old pups, as the percentage of Nestin-expressing cells is higher in newborns (Mignone et al., 2004). PCR, qRT-PCR, and immunofluorescence analysis of the brain 2 days after injection confirmed TRF1 downregulation in Trf1lox/lox mice compared with controls (Figures S3A–S3C). In the setting of Cdkn2a deficiency, tumors started to appear 4–5 weeks after intracranial injection (Ozawa et al., 2014) (Figure 2B). Strikingly, even in this setting of fast-growing tumors, mice with Trf1-deleted brains showed an increased survival of 80% compared with the Trf1+/+ controls (Figure 2C). At time of death, Trf1-deleted tumors were histological indistinguishable from Trf1+/+ tumors and showed normal TRF1 mRNA and protein levels, indicating that they were escapers (Figures S3D–S3F). Telomere Q-FISH analysis also revealed that all the tumors had the same telomere length (Figure S3G). The fact that no tumors lacking TRF1 expression were found suggests that TRF1 is essential for PDGFB-induced GBM initiation. To study the cellular and molecular effects of Trf1 deletion in GBM initiation, we killed the mice at an earlier time point before they started dying from GBM (i.e., 45 days after tumor induction). At this time point, 91% of Trf1+/+ mice showed brain tumors compared with only 6% of Trf1lox/lox mice (Figure 2D). Histological analysis revealed a significant difference in tumor size between both genotypes, with Trf1lox/lox tumors being undetectable by H&E staining in most of the cases (Figure 2E). Immunohistochemistry analysis of the human astrocyte (HA) tag, which marks PDGFB-expressing cells in this model, also confirmed smaller Trf1lox/lox tumors compared with controls (Figure S3H). Finally, Trf1lox/lox brains showed very few proliferating cells compared with a high proliferation index in Trf1+/+ tumors (Figure 2F), indicating impaired tumor growth by Trf1 deletion. Similar results were obtained in the PDGFA-induced mouse model of GBM, where tumors grow more slowly than in the PDGFB model. In this case, Trf1lox/lox mice showed a highly significant 65% increase in survival compared with Trf1+/+ mice (Figures 2A and 2G). In particular, by day 150 after tumor induction around 75% of Trf1+/+ mice had already died from GBM, while only 10% of Trf1lox/lox mice were affected (Figure 2H). Trf1 Deficiency Leads to Telomeric Damage and Reduced Stemness in Primary NSCs To study how Trf1 deletion impairs GBM initiation, we addressed the cellular and molecular effects of abrogating Trf1 specifically on isolated NSCs. NSCs express Nestin, and, thus, these cells are the targets of the RCAS vectors when we perform the intracranial injections into the SVZ of NestinTva mice. To this end, we established an in vitro system using

(G) Survival curve of mice of the indicated genotypes injected with PDGFA-producing DF-1 cells. (H) Percentage of mice affected with GBM 150 days after tumor induction; same cohort as (G). Data are represented as mean ± SD. n represents the number of mice. Statistical analysis: unpaired t test, log rank test, and chi-square test. **p < 0.01, ***p < 0.001. See also Figure S3.

594 Cancer Cell 32, 590–607, November 13, 2017

Figure 3. Trf1 Abrogation Induces DNA Damage and Reduced Stemness in NSCs (A) NSCs are obtained by digestion of brain of 2-day-old pups from the indicated genotypes with papain. Trf1 allele is depicted by Cre-mediated excision. Scale bar, 100 mm. (B) Representative images (left) and quantification (right) of TRF1 protein levels determined by TRF1 immunofluorescence in Trf1+/+ and Trf1lox/lox NSC. Scale bars, 5 mm. (C) TRF1 mRNA levels by qRT-PCR in Trf1+/+ and Trf1lox/lox NSC. (D) Representative images (left) and quantification (right) of gH2AX nuclear intensity in Trf1+/+ and Trf1lox/lox NSCs 5 days after the last RCAS-Cre infection. Scale bars, 5 mm. (legend continued on next page)

Cancer Cell 32, 590â&#x20AC;&#x201C;607, November 13, 2017 595

primary NSCs isolated from Trf1+/+ and Trf1lox/lox Nestin-Tva Cdkn2a / newborn brains (2 days old) (Figure 3A). To induce Trf1 deletion, we transduced two independent lines of Trf1+/+ and eight independent lines of Trf1lox/lox primary NSC with the RCAS-Cre virus. We confirmed downregulation of both TRF1 mRNA and protein by qRT-PCR and by immunofluorescence, respectively, in Trf1lox/lox primary NSC compared with the Trf1+/+ controls (Figures 3B and 3C). Trf1 deletion has been previously shown to induce a persistent DDR at telomeres in both fibroblasts and epithelial cells (Martı́nez et al., 2009). To address whether Trf1 deletion also leads to DNA damage in NSCs, we quantified 53BP1 and gH2AX levels by immunofluorescence in Trf1-deficient NSC compared with Trf1+/+ controls. We found significantly higher gH2AX and 53BP1 nuclear fluorescence intensities in Trf1lox/lox NSC compared with Trf1+/+ controls (Figures 3D and 3E). In addition, double immunofluorescence staining of gH2AX and the telomeric protein RAP1 showed increased DNA damage specifically located at telomeres (the so-called telomere induced foci or TIFs) in the case of Trf1-deleted NSC compared with Trf1+/+ controls (Figure 3F), indicating that telomeres were being uncapped as the consequence of Trf1 deletion, leading to a DDR at telomeres. As TRF1 expression is upregulated and it is essential for both adult stem cells and pluripotent stem cells (Schneider et al., 2013), we next studied the impact of Trf1 deletion on the stem potential of NSC. To this end, we performed a neurosphere formation assay by disaggregating NSCs from Trf1+/+ and Trf1lox/lox brains and plating single cells in serial dilutions. Both the number and diameter of secondary neurospheres were significantly decreased in Trf1-deficient NSC compared with Trf1+/+ controls (Figures 3G and 3H). This was accompanied by a significant reduction of the Ki67 proliferation marker (Figure 3I) and a reduction of percentage of Nestin-positive cells (Figure 3J). In summary, Trf1 deletion in NSC induces a DDR at telomeres and reduces stemness in NSCs, probably reducing the tumor-initiating potential of these cells upon oncogenic transformation. Therapeutic Effects of Trf1 Abrogation in Already Established GBMs To validate TRF1 as a therapeutic target that could be translated into human patients, we next generated additional mouse models in which we could first induce the different GBM subtypes and then delete Trf1 once the tumors were established. To this end, we crossed Trf1lox/lox;Nestin-Tva;Cdkn2a / or Trf1+/+;Nestin-Tva;Cdkn2a / mice with hUBC-CreERT2 mice to obtain Trf1lox/lox;Nestin-Tva;Cdkn2a / ;hUBC-CreERT2 and Trf1+/+;Nestin-Tva;Cdkn2a / ;hUBC-CreERT2, which allowed

for ubiquitous Cre-induced Trf1 deletion upon tamoxifen administration once the tumors had formed. We first induced tumors with PDGFB, and 2.5 weeks after tumor induction we started the treatment with tamoxifen to delete Trf1 in the whole organism (Figures 4A and 4B). At this time point tumors had already started to form (Figure S4A). Trf1lox/lox-deleted mice showed a 33% increase in survival compared with Trf1+/+ mice (Figure 4C), suggesting therapeutic effectiveness of Trf1 deletion in reducing GBM tumor growth once the tumors were already established. Furthermore, we found that 25% of the GBM tumors appearing in Trf1-deleted mice were escapers as they showed normal TRF1 expression, and were excluded from further analyses (Figures S4B and S4C), again highlighting the potent anti-tumorigenic effect of Trf1 deletion. To better study the effects of Trf1 abrogation in already established tumors, we killed the mice at an earlier time point (32 days after tumor induction). We found a 50% decrease in TRF1 protein fluorescence in Trf1lox/lox tumors compared with Trf1+/+ controls (Figure 4D). Similarly to the tumor initiation models, Trf1 deletion did not cause any significant change in telomere length in these tumors (Figure S4D), further confirming that the therapeutic effects of Trf1 deletion consist in direct telomere uncapping in GBM tumor cells independently of their telomere length. Also, we did not find any changes in the mRNA levels of other shelterin components upon TRF1 deletion as determined by qRT-PCR (Figure S4E). Histopathological analysis showed that Trf1-deleted tumors were significantly smaller compared with the controls (Figure 4E). Trf1-deleted tumors also showed significantly less Ki67-positive cells (Figure 4F), indicating lower proliferation. In addition, we found significantly increased numbers of cells with DNA damage (gH2AX-positive cells) in the Trf1-deficient tumors compared with the controls (Figure 4G). This damage was located at telomeres, as indicated by a significantly increased percentage of cells with more than one TIF in Trf1-deficient tumors compared with the controls (Figure 4H). Increased telomeric damage was also accompanied by a significant increase in downstream cell-cycle inhibitors p21 and p53 and in the apoptosis marker AC3 (Figure 4I). However, we did not see significant changes in phospho-RPA32 (Figure S4F), indicating that DNA damage is probably independent of the ATR/Chk1 pathway. In summary, Trf1 deletion in already-formed GBM tumors leads to impaired proliferation and increased telomeric DNA damage. Similar results were obtained when Trf1 was deleted in already established tumors that were induced using PDGFA as an independent oncogenic insult. In this case, mice were fed with

(E) Representative images (left) and quantification (right) of 53BP1 nuclear intensity in Trf1+/+ and Trf1lox/lox NSC 5 days after the last RCAS-Cre infection. Scale bars, 5 mm. (F) Representative images (left) and quantification (right) of percentage of cells presenting three or more gH2AX and RAP1 colocalizing foci (TIFs) 5 days after the last RCAS-Cre infection. White arrowheads: colocalization of gH2AX and RAP1. Scale bars, 5 mm. (G) Representative images (left) and of number (right) of neurospheres from Trf1+/+ and Trf1lox/lox NSC 1 week after plating. Scale bars, 100 mm. (H) Quantification of neurosphere diameter from Trf1+/+ and Trf1lox/lox NSC 1 week after plating. (I) Representative images (left) and percentage (right) of Ki67-positive cells in Trf1+/+ and Trf1lox/lox NSCs. Scale bars, 20 mm. (J) Representative images (left) and percentage (right) of Nestin-expressing cells. Scale bars, 50 mm. Data are represented as mean ± SD, with the exception of (H), which is mean ± SEM. n represents independent NSC lines, with the exception of (H), which is the number of neurospheres. Statistical analysis: unpaired t test. *p < 0.05, **p < 0.01, ***p < 0.001.

596 Cancer Cell 32, 590–607, November 13, 2017

(legend on next page)

Cancer Cell 32, 590â&#x20AC;&#x201C;607, November 13, 2017 597

tamoxifen 5–6 weeks after tumor induction, whenever the first mice started to die from tumors. Again, Trf1 deletion post-tumor induction in Trf1lox/lox mice resulted in a significant increase in survival compared with Trf1+/+ mice (Figures 4A and 4J). As in the PDGFB model, Trf1 deletion was determined by PCR and TRF1 expression was determined by immunofluorescence in all tumors to eliminate all the tumors that did not delete Trf1 for further analyses (Figures S4G and S4H). In summary, Trf1 deletion effectively blocks tumor progression in two independent (PDGFB and PDGFA) GBM mouse models concomitant with induction of telomere DNA damage. Trf1-Deficient GSCs Show Decreased Stemness and Tumorigenicity To study the effects of Trf1 abrogation specifically in isolated GSCs from already-formed GBM tumors, we established an in vitro system by isolating GSC from Trf1lox/lox;Nestin-Tva; Cdkn2a / ;hUBC-CreERT2 and Trf1+/+;Nestin-Tva;Cdkn2a / ; hUBC-CreERT2 mice after tumor induction with PDGFB (Figure 5A). After 15 days of in vitro treatment with tamoxifen to delete Trf1, PCR analysis of the Trf1 locus showed a population of Trf1lox GSC deleted for Trf1 (Figure 5B). These Trf1-deleted cells showed a significant reduction in both number of neurospheres and diameter of the neurospheres compared with the controls (Figures 5C and 5D). Next, we studied whether Trf1 abrogation affected the tumorigenic potential of these cells. Previous studies have shown that GSCs have the ability to form secondary tumors after orthotopic injection into the brain of syngeneic mice (Jiang et al., 2011). Thus, we injected Trf1+/+ and Trf1lox/lox GSC into the brain of syngeneic mice fed with tamoxifen to induce Trf1 deletion (Figure 5E). Mice injected with Trf1lox/lox GSC showed a significant increase in survival compared with the controls (Figure 5F), indicating that Trf1 deletion significantly decreased the ability of GSC to form secondary GBM tumors. Moreover, 120 days after GSC injection, 71% of mice injected with Trf1+/+ GSC had died owing to secondary tumors compared with only 10% of the mice injected with Trf1lox/lox GSC (Figure 5G). Postmortem histological analysis showed that the secondary tumors were histologically similar to the parental PDGFB-induced tumors, from which cells were extracted (Figure 5H); thus, Trf1 abrogation in GSCs strongly reduces their stemness and tumor forming potential.

Brain-Specific or Whole-Body Trf1 Deletion in Mice Does Not Significantly Impair Memory, Neuromuscular, and Olfactory Functions Previous studies showed that Trf1 whole-body deletion in adult mice was compatible with mouse viability, although highly proliferative compartments such as the skin and the bone marrow showed decreased cellularity (Garcı́a-Beccaria et al., 2015). Similarly, deletion of Terf2, which encodes another essential shelterin component in adulthood, does not lead to brain dysfunction (Lobanova et al., 2017). To validate TRF1 as a safe target in the treatment of GBM, we set to address whether brain-specific or whole-body Trf1 deletion would affect the brain functions of mice. We first checked TRF1 expression in the normal brain. In agreement with increased TRF1 expression in adult stem cells in mice (Schneider et al., 2013), we found significant TRF1 expression in the SVZ compared with cerebral cortex (Figure 6A). The SVZ is one of the main areas of adult neurogenesis, characterized by the expression of the stem cell marker Nestin (Faiz et al., 2015). Next, as the percentage of Nestin-positive cells is higher in newborns (Mignone et al., 2004), we injected RCASCre virus-producing DF-1 cells to induce Trf1 deletion into the brain of Trf1lox/lox;Cdkn2a / and Trf1+/+;Cdkn2a / newborns (2 days old) (Figure 6B) and let these mice reach adulthood to assess any neurological effects of TRF1 deletion (see Figures S3A–S3C for brain TRF1 levels). When mice reached adulthood, around 2.5 months after RCAS-Cre-producing DF-1 cell injection, PCR analysis confirmed that four out of the six mice still showed Trf1 deletion in the brain (Figure S5). Thus, decreased TRF1 levels specifically in the brain are maintained to adulthood without resulting in decreased mouse viability. To address whether Trf1 depletion in newborn brains affected adult brain function, we performed different tests to measure cognitive and olfactory capacities, memory, coordination, and balance. Olfactory capacities were measured by using the so-called buried food test, in which mice were fasted 24 hr and then moved to a new cage with a buried food pellet (Yang and Crawley, 2009) (Figure 6C). Mice of both genotypes were able to find the food pellet with a 100% success rate (Figure 6D). The time used to find the pellet was also similar in Trf1+/+ and Trf1lox/lox mice (Figure 6E). Next, we evaluated the memory skills by using the object recognition test (Bernardes de Jesus et al., 2012). We first trained the mice by placing them in a box with two identical objects (A and A), and then we changed one of

Figure 4. Trf1 Deletion Delays Tumor Progression in PDGFB-Driven GBM (A) Schematic representation of tumors induced by PDGFB overexpression and Trf1lox allele deletion generated by tamoxifen treatment after tumors are formed. (B) PDGFB-producing cells are injected to induce tumors. At 2.5 weeks after tumor induction mice are treated with tamoxifen. Mice start dying from GBM at week 4 after treatment. (C) Survival curves of the indicated genotypes. Histological analysis is performed 32 days after tumor induction in a different cohort of mice. (D) Representative images (left) and quantification (right) of TRF1 nuclear fluorescence intensity at 32 days after tumor induction. Scale bars, 5 mm. (E) Representative images (left) and quantification (right) of tumor areas by H&E at 32 days after tumor induction. Scale bars, 1 mm (left) and 100 mm (right). (F) Representative images (left) and number (right) of Ki67-positive cells per field in tumors at 32 days after tumor induction. Scale bars, 100 mm. (G) Representative images (left) and number (right) of gH2AX-positive cells per field in Trf1+/+ and Trf1lox/lox tumors at 32 days after tumor induction. Scale bars, 20 mm. (H) Representative images (left) and percentage (right) of cells presenting one or more 53BP1 and telomere colocalizing foci (TIFs). White arrowheads: colocalization of 53BP1 and telomeres. Scale bars, 5 mm. (I) Representative images (left) and percentage (right) of p53-, p21-, and AC3-positive cells. Scale bars, 50 mm. (J) Survival curves of mice of the indicated genotypes injected with PDGFA-producing cells. Data are represented as mean ± SD. n represents the number of mice. Statistical analysis: unpaired t test and log rank test. *p < 0.05, **p < 0.01, ***p < 0.001. See also Figure S4.

598 Cancer Cell 32, 590–607, November 13, 2017

Figure 5. Trf1-Deficient Glioma Stem Cells Show Reduced Stemness and Reduced Tumor Forming Potential (A) Trf1+/+ and Trf1lox/lox GSCs are obtained by tumor digested with papain. The Trf1lox allele is generated by tamoxifen treatment. Scale bar, 100 mm. (B) Analysis of Trf1 excision by PCR. (C) Representative images (left) and number (right) of neurospheres from Trf1+/+ and Trf1lox/lox GSC. Scale bars, 100 mm. (D) Quantification of neurosphere diameter from Trf1+/+ and Trf1lox/lox GSC. (E) GSCs are orthotopically injected in syngeneic mice fed with tamoxifen. Scale bar, 100 mm. (F) Survival curves of mice injected with Trf1+/+ and Trf1lox/lox GSCs. (G) Percentage of mice affected by the injection of Trf1+/+ and Trf1lox/lox GSCs. (H) Representative image of Trf1+/+ tumor histology. Scale bars, 500 mm (left) and 100 mm (right). Data are represented as mean ± SD with the exception of (D), which is the mean ± SEM. n represents the number of fields in (C), the number of spheres in (D), and the number of mice in (F and G). Statistical analysis: unpaired t test and log rank test. **p < 0.01, ***p < 0.001.

Cancer Cell 32, 590–607, November 13, 2017 599

Figure 6. Trf1 Brain-Specific Deletion in Healthy Mice Does Not Compromise Brain Function of Organism Viability (A) Representative images (left) and quantification (right) of TRF1 nuclear fluorescence in the SVZ compared with the surrounding cerebral cortex. Scale bars, 50 mm (left) and 10 mm (right). (B) Trf1 deletion is induced by Cre-mediated recombination in 2-day-old newborns. (C) After 24 hr fasting, mice are moved to a cage with a buried food pellet and both the success and the time to find the pellet are measured. (D and E) Percentage of success finding the pellet (D) and time needed to find the pellet (E). (F) Mice were trained in a box with two identical objects (A). The test day one of the object was changed (B). (legend continued on next page)

600 Cancer Cell 32, 590â&#x20AC;&#x201C;607, November 13, 2017

the objects on the day of the test (A and B) (Figure 6F). By calculating the time spent with the new object B and by dividing this by the time spent with (A + B), which is an indication of the memory skills, we observed no significant differences between genotypes (Figure 6G). To evaluate coordination and balance we performed two independent tests, the rotarod and the tight rope (Bernardes de Jesus et al., 2012; Tomás-Loba et al., 2008). In the rotarod test we measured the time mice could stay on the rod. In the tightrope test, we evaluated the ability of the mice to stay on the rope without falling, and we considered the test a ‘‘success’’ if mice were able to stay on the rope for more than 1 min. No significant differences were found between Trf1+/+ and Trf1lox/lox mice in any of these two tests (Figures 6H and 6I). In parallel, to address the effects of whole-body Trf1 deletion in adult mice, we fed 10 week old Trf1lox/lox;Cdkn2a / ;hUBCCreERT2 and Trf1+/+;Cdkn2a / ;hUBC-CreERT2 mice tamoxifen (Figure 6J). These mice with whole-body Trf1 deletion present a normal survival up to at least 7 months of age (Figure 6K). After 2 months of continuous treatment, we performed the buried food test, the object recognition test, and the rotarod and tightrope tests, and found no significant differences between Trf1 wild-type and Trf1-deleted mice (Figures 6L–6P), demonstrating that Trf1 deletion in the brain does not affect cognitive, olfactory, memory, or neuromuscular abilities of the mice. TRF1 Inhibition Reduces Stemness and Proliferation and Increases Telomeric Aberrations and DNA Damage in Human GBM Cells We previously reported the discovery of small molecules that can inhibit TRF1 telomeric foci (Garcı́a-Beccaria et al., 2015). To address whether they show therapeutic effects in GBM, we first treated U251 human GBM cells with the TRF1 inhibitors, ETP-47228, ETP-47037, and ETP-50946 (Figure 7A). Immunofluorescence analysis to quantify TRF1 foci intensity revealed that the three compounds effectively reduced TRF1 nuclear foci fluorescence by approximately 50% compared with DMSO-treated cells (Figure 7B). We also confirmed reduced TRF1 protein levels by western blot analysis (Figure 7C). Similar to TRF1 genetic depletion, TRF1 chemical inhibitors significantly reduced proliferation (Figure 7D) and induced DNA damage, as determined by 53BP1 levels (Figure S6A). To assess whether the DNA damage was specifically located at telomeres, we determined the abundance of TIFs. To this end, we performed a double immunofluorescence of gH2AX with the telomeric protein RAP1, which showed that the percentage of cells with two or more TIFs was significantly increased upon treatment with the TRF1 chemical inhibitors (Figure 7E). Next, we set to address whether TRF1 chemical inhibitors also reduced stemness of human GBM cells. To this end, we cultured

U251 cells with NSC medium to obtain a suspension culture enriched in stem cells, and performed a sphere formation assay upon treatment with ETP-47228, ETP-47037, and ETP-50946, or DMSO, for 7 days. Treated cells showed a strong reduction in both number and diameter of neurospheres compared with the controls (Figures 7F and S6B). To discard possible off-target effects of TRF1 chemical inhibitors, we knocked down TRF1 by small hairpin RNA in the U251 GBM cell line (Figures S6C and S6D). Similar to chemical inhibition, TRF1 knocked down cells showed decreased proliferation (Figure S6E) and increased DNA damage markers gH2AX and 53BP1 (Figure S6F). We also observed an increase in the so-called multitelomeric signals (Figure S6G), a telomere aberration previously associated to loss of TRF1 (Martı́nez et al., 2009; Sfeir et al., 2009). Also, similar to TRF1 chemical inhibition, the number and diameter of the neurospheres was significantly decreased in TRF1 knocked down cells compared with the controls (Figures S6H and S6I). Thus, TRF1 genetic inhibition mimics the effects shown by TRF1 chemical inhibition in human GBM cells. TRF1 Chemical Inhibition Synergizes with g-Irradiation and Temozolomide to Reduce Proliferation of Human GBM Cells The standard of care for GBM patients consists of surgical resection combined with radiation and chemotherapy, as well as adjuvant chemotherapy. Unfortunately, frequent recurrences after treatment are observed owing to their radio- and chemoresistant properties (Bhat et al., 2013; Bao et al., 2006). Dysfunctional telomeres have been shown to lead to increased radiosensitivity, most likely as the consequence of telomere uncapping (Goytisolo et al., 2000; Alt et al., 2000). In addition, low levels of telomerase expression are shown to correlate with a higher sensitivity to temozolomide (TMZ), indicating that telomeres may play a role in TMZ resistance (Kanzawa et al., 2003). As both genetic and chemical TRF1 inhibition significantly impairs GBM proliferation and stemness concomitant with induction of a DDR at telomeres, we set to study the combined effects of simultaneous TRF1 inhibition and g-irradiation or TMZ treatments in human U251 GBM cells. Upon irradiation, glioma cells predominantly arrest in the G2/M phase (Badie et al., 1999) (Figure 7G). Combined TRF1 chemical inhibition and g-irradiation (6 Gys) synergistically increased the percentage of G2 arrested cells (Figure 7G). These effects were also recapitulated in TRF1 knocked down cells (Figure S7A). We also observed a synergistic effect of TRF1 inhibitors with irradiation in increasing DNA damage, as determined by gH2AX levels and colocalization of gH2AX with the telomeric protein RAP1 (Figures 7H and S7B). Combined TRF1 inhibition and TMZ treatment for 3 days also synergistically reduced cell viability

(G) Quantification of time spent with B/(A + B). (H) Time spend in the rotarod. (I) Percentage success in the tightrope. (J) Trf1 whole-body deletion is induced by tamoxifen diet from the age of 10 weeks. (K) Survival curves of mice of the indicated genotypes. (L and M) Percentage of success finding the pellet (L) and time needed to find the pellet (M). (N) Quantification of time spent with B/(A + B). (O) Time spend in the rotarod. (P) Percentage success in the tightrope. Data are represented as mean ± SD. n represents the number of mice. Statistical analysis: unpaired t test. ns, not significant. ***p < 0.001. See also Figure S5.

Cancer Cell 32, 590–607, November 13, 2017 601

(legend on next page)

602 Cancer Cell 32, 590â&#x20AC;&#x201C;607, November 13, 2017

in U251 GBM cells (Figure 7I), thus suggesting that TRF1 inhibition synergizes with current standard treatments for GBM. TRF1 Chemical Inhibition Reduces Stemness and Xenograft Tumor Growth of Patient-Derived Primary GSCs To validate the TRF1 chemical inhibitors in a more relevant clinical setting, we treated two independent patient-derived primary GSCs (h543 and h676) cultures with the TRF1 inhibitors. Again, treatment of h543 and h676 cells with the three inhibitors revealed a significant reduction in both the number and diameter of spheres compared with the untreated controls (Figures 8A, S8A, and S8B). As the TRF1 inhibitors cannot cross the blood-brain barrier, we next injected h676 and h543 cells subcutaneously into nude mice and treated them with orally administered TRF1 inhibitors. One week after GSC injection, mice received oral administration of the vehicle as placebo or ETP-47037 5 days/week, every week until human endpoint, and tumors were continuously followed up by caliper measurements (Figure 8B). Xenografts from h676 GSCs and those treated with ETP-47037 showed a drastic reduction in tumor area compared with vehicle-treated mice at all time points after treatment until the placebo group reached the human endpoint (Figure 8C). Xenografts from h543 GSCs showed slower tumor growth, but again ETP-47037-treated mice showed decreased tumor growth compared with the vehicle, which was maintained until vehicle-treated mice were killed owing to very large tumors (Figure 8D). Postmortem tumor analysis of xenografts from h676 GSCs revealed a striking decrease in tumor size and tumor weight in the ETP-47037-treated tumors compared with those treated with the placebo (Figure 8E). TRF1 immunofluorescence analysis confirmed that tumors treated with ETP-47037 showed an 80% reduction in TRF1 protein levels (Figure 8F). This TRF1 reduction was accompanied by a decrease in the proliferation marker Ki67 (Figure 8G) and an increase in the DNA damage marker gH2AX (Figure 8H). In addition, full histological analysis revealed that, while placebo-treated tumors showed high cellularity, open chromatin, active nucleus, and mitotic cells, ETP-47037-treated tumors were characterized by low cellularity, compacted chromatin, fragmented DNA, apoptotic bodies, and necrotic areas (Figure S8C).

Of relevance, we did not observe any signs of sickness or morbidity in the ETP-47037-treated cohorts compared with the placebo group, in agreement with a therapeutic window for TRF1 inhibition (Garcı́a-Beccaria et al., 2015). Histological analysis of the mice did not reveal any deleterious effects of ETP-47037 treatment in the highly proliferative tissues, including the skin, intestine, and bone marrow (Figure S8D). Only one mouse out of four showed a mild widening of lymphatic vessels in the intestine (Table S1). In summary, TRF1 chemical inhibitors effectively reduce number and diameter of neurospheres in vitro in two independent patient-derived primary GSC cultures. Furthermore, oral administration of TRF1 inhibitors to patientderived xenograft models using primary GSCs, drastically impairs tumor growth. DISCUSSION GBM remains an incurable tumor, with a mean survival of less than 2 years from diagnosis (Wen and Kesari, 2008). Recent efforts to understand the genetic origin of GBM have identified telomere maintenance genes (i.e., telomerase) among the most frequently mutated in GBM (Nonoguchi et al., 2013; Boldrini et al., 2006; Heaphy et al., 2011). Telomere maintenance above a minimum length is thought to be necessary for indefinite cancer cell growth (Hanahan and Weinberg, 2011), thus leading to the proposal that inhibition of telomerase may be an anticancer strategy (Kim et al., 1994). However, both telomerase abrogation in mouse cancer models and human clinical trials with telomerase inhibitors have shown limited benefit, as telomerase inhibition only affects cell viability when telomeres are short, and tumors are heterogeneous in terms of telomere length (Middleton et al., 2014; Parkhurst et al., 2004; Gonzalez-Suarez et al., 2000; Chin et al., 1999). In the particular case of GBM, telomere targeting has also focused on direct (Marian et al., 2010) or indirect telomerase inhibition (Hasegawa et al., 2016). In this study, we investigated an alternative approach to target telomeres by targeting the telomere protective protein TRF1, with which we expected to induce telomere uncapping in every tumor cell, independently of its telomere length. In addition, the fact that TRF1 is enriched in adult stem cell compartments and pluripotent stem cells, and that it is essential for maintenance of tissue homeostasis and pluripotency (Schneider et al., 2013;

Figure 7. TRF1 Chemical Inhibitors Induce DNA Damage and Reduce Stemness in GBM Human Cells (A) Structure of the chemical compounds ETP-47228, ETP-47037, and ETP-50946. (B) Representative images (left) and quantification (right) of TRF1 nuclear fluorescence of U251 cells treated with the indicated compounds. Scale bars, 5 mm. (C) Western blot images (left) and TRF1 protein levels (right) of U251 cells treated with the indicated compounds. (D) Cell numbers at 24 and 48 hr of U251 cells treated with the indicated compounds. (E) Representative images (left) and percentage (right) of cells presenting two or more gH2AX and RAP1 colocalizing foci (TIFs). White arrowheads: colocalization of gH2AX and RAP1. Scale bars, 10 mm. (F) Representative images (left) and quantification (right) of the number of neurospheres formed by U251 cells treated with the indicated compounds. Scale bars, 100 mm. (G) Percentage of U251 cells in G2 phase upon 6 Gy irradiation and treated with the indicated compounds. (H) Representative images (left) and percentage (right) of cells presenting two or more gH2AX and RAP1 colocalizing foci (TIFs) upon 6 Gy irradiation and treated with the indicated compounds. DMSO represents IRR alone. White arrowheads: colocalization of gH2AX and RAP1. Scale bars, 10 mm. (I) Cell viability measured by an MTT assay in the U251 human cell line treated with the indicated compounds and no temozolomide, temozolomide 500 mM, or temozolomide 1,000 mM for 3 days. Data are represented as mean ± SD. n represents the number of biological replicates. Statistical analysis: unpaired t test. *p < 0.05, **p < 0.01, ***p < 0.001. See also Figures S6 and S7.

Cancer Cell 32, 590–607, November 13, 2017 603

(legend on next page)

604 Cancer Cell 32, 590â&#x20AC;&#x201C;607, November 13, 2017

Boué et al., 2010), suggests that targeting TRF1 could also impair tumor-initiating capabilities in GBM (Chen et al., 2012). In line with a role for TRF1 in GBM growth, we found here that TRF1 is overexpressed in both human GBM models (tumor cell lines and patient-derived primary GSCs) and in several GBM mouse models in a manner that is independent of telomere length. We found that TRF1 is also overexpressed in the NSC compartments in the mouse, such as the subventricular zone, which is enriched in Nestin-expressing NSCs, and that TRF1 expression also correlates with the well-known stem cell markers SOX2, CD133, and Nestin in human GBM samples, further reinforcing the notion that TRF1 is upregulated in adult stem cell compartments and required for maintenance of tissue homeostasis (Schneider et al., 2013). Importantly, we demonstrate here TRF1 is an effective target in GBM by using both genetic ablation mouse models and chemical inhibitors. Genetic deletion of TRF1 effectively blocks both GBM initiation and progression of already established GBM tumors in various mouse models of GBM resulting in a striking increase in survival. As predicted, the therapeutic effect of TRF1 inhibition occurred in a telomere length-independent manner, overcoming the potential problem of telomere length heterogeneity within tumors and the inability to kill all tumor cells, including the tumorinitiating populations. Indeed, Trf1 deletion reduced the stemness of both NSCs and glioma stem-like cells, at the same time that induced a DDR at telomeres. Of note, in our experimental setting, around 25% of the tumors were escapers and future experiments warrant the study of these potential resistance mechanisms. TRF1 chemical inhibition recapitulated the findings observed with TRF1 genetic deletion, including decreased TRF1 protein levels, induction of telomere DNA damage located at telomeres, and decreased proliferation and stemness of glioma cells. In particular, TRF1 chemical inhibitors showed a potent blocking effect in sphere formation, thus demonstrating the ability of these inhibitors to block stemness in GBM. Importantly, oral administration of TRF1 chemical inhibitors drastically reduced tumor growth in vivo in xenograft mouse models from patient-derived primary GSCs. We find these observations of potentially clinical relevance, as the recurrence of GBM after the current treatments is due to the resistance of the GSCs and their capability to recapitulate the original tumor. Any potential anti-cancer target, however, must also fulfill the important requisite of not showing deleterious effects in healthy tissues or compromising organism viability.

In this regard, here we demonstrate that Trf1 genetic deletion in the brain does not affect the cognitive or neuromuscular abilities of mice. Similarly, we did not defect any signs of sickness or morbidity in the xenograft models treated with TRF1 chemical inhibitors compared with the placebo group, supporting a therapeutic window for TRF1 inhibition. A recent report also showed that deletion of the essential shelterin component TRF2 in the brain does not lead to any brain dysfunction (Lobanova et al., 2017). In summary, we demonstrate here the effectiveness of targeting TRF1 in different mouse glioblastoma subtypes, as well as in patient-derived human xenograft models, in the absence of any detectable deleterious effects for brain function or organismal viability. STAR+METHODS Detailed methods are provided in the online version of this paper and include the following: d d d

KEY RESOURCES TABLE CONTACT FOR REAGENT AND RESOURCE SHARING EXPERIMENTAL MODEL AND SUBJECT DETAILS B Mice B Cell Culture and Transfection B Neural Stem Cell (NSC) and Glioma Stem Cell (GSC) Isolation and Culture METHODS DETAILS B Generation of Mouse Models with Brain Tumors B Neurosphere Formation Assays B Intracranial Cell Transplantation into Syngeneic Mice B Xenografts Experiments B Telomere Length Analyses on Tissue Sections B Cognitive Tests B Immunofluorescence Analyses in Cells and Tissue Sections B Immunohistochemistry Analyses in Tissue Sections B Real-Time qPCR B PCR B Western-Blots B TRF1 Chemical Inhibitors B Irradiation and Temozolomide B Tissue Micro Array (TMA) QUANTIFICATION AND STATISTICAL ANALYSIS

Figure 8. TRF1 Chemical Inhibitors Reduce Stemness and Xenograft Tumor Growth in Patient-Derived Primary GSCs (A) Representative images (left) and quantification (right) of number of neurospheres formed by h543 and h676 GSCs treated with the indicated compounds. Scale bars, 100 mm. (B) Xenograft mouse models from patient-derived primary GSCs are generated by subcutaneous injection of GSCs into nude mice. One week after injection, mice are treated either with ETP-47037 or with vehicle as placebo. (C) Representative image of tumors (left) and longitudinal tumor growth follow-up (right) in ETP-47037- or vehicle-treated xenograft models with h676 GSCs (right). (D) Representative image of tumors (left) and longitudinal tumor growth follow-up (right) in ETP-47037- or vehicle-treated mice injected with h543 GSCs and representative image of tumors. (E) Representative image of tumors (left) and tumor weight (right) in ETP-47037- or vehicle-treated mice injected with h676 GSCs at postmortem. (F) TRF1 nuclear fluorescence in ETP-47037- or vehicle-treated tumors. Scale bars, 5 mm. (G) Representative images (left) and percentage (right) of Ki67-positive cells per field in ETP-47037- or vehicle-treated tumors. Scale bars, 50 mm. (H) Representative images (left) and percentage (right) of gH2AX-positive cells per field in in ETP-47037- or vehicle-treated tumors. Scale bars, 50 mm. Data are represented as mean ± SD. n represents the number of biological replicates in (A) and the number of tumors in (C–H). Statistical analysis: unpaired t test. *p < 0.05, **p < 0.01, ***p < 0.001. See also Figure S8 and Table S1.

Cancer Cell 32, 590–607, November 13, 2017 605

SUPPLEMENTAL INFORMATION Supplemental Information includes eight figure and two tables and can be found with this article online at https://doi.org/10.1016/j.ccell.2017.10.006. AUTHOR CONTRIBUTIONS Investigation, L.B.; Writing – Original Draft, M.A.B. and L.B.; Writing – Review & Editing, M.A.B. and L.B.; Resources, A.J.S., M.S., M.M., S.M., C.B.A., J.P., and D.M.; Supervision, M.A.B. ACKNOWLEDGMENTS We thank R. Serrano for mice handling, J.M. Flores for histopathology, M. Valiente for the human astrocyte (HA) cell line, and the Comparative Pathology and Biobank Units at CNIO. MAB laboratory is funded by SAF201345111-R from MINECO, Fundación Botı´n, and Banco Santander, Worldwide Cancer Research 16-1177. L.B. is a fellow of the La Caixa-Severo Ochoa International PhD Program. Received: April 19, 2017 Revised: July 28, 2017 Accepted: October 7, 2017 Published: November 13, 2017 REFERENCES Alt, F.W., Chang, S., Weiler, S.R., Ganesan, S., Chaudhuri, J., Zhu, C., Artandi, S.E., Rudolph, K.L., Gottlieb, G.J., Chin, L., et al. (2000). Telomere dysfunction impairs DNA repair and enhances sensitivity to ionizing radiation. Nat. Genet. 26, 85–88. Badie, B., Goh, C.S., Klaver, J., Herweijer, H., and Boothman, D.A. (1999). Combined radiation and p53 gene therapy of malignant glioma cells. Cancer Gene Ther. 6, 155–162. Bainbridge, M.N., Armstrong, G.N., Gramatges, M.M., Bertuch, A.A., Jhangiani, S.N., Doddapaneni, H., Lewis, L., Tombrello, J., Tsavachidis, S., Liu, Y., et al. (2015). Germline mutations in shelterin complex genes are associated with familial glioma. J. Natl. Cancer Inst. 107, 384. Bao, S., Wu, Q., McLendon, R.E., Hao, Y., Shi, Q., Hjelmeland, A.B., Dewhirst, M.W., Bigner, D.D., and Rich, J.N. (2006). Glioma stem cells promote radioresistance by preferential activation of the DNA damage response. Nature 444, 756–760. Beier, F., Foronda, M., Martinez, P., and Blasco, M.A. (2012). Conditional TRF1 knockout in the hematopoietic compartment leads to bone marrow failure and recapitulates clinical features of dyskeratosis congenita. Blood 120, 2990–3000. Bernardes de Jesus, B., Vera, E., Schneeberger, K., Tejera, A.M., Ayuso, E., Bosch, F., and Blasco, M.A. (2012). Telomerase gene therapy in adult and old mice delays aging and increases longevity without increasing cancer. EMBO Mol. Med. 4, 691–704.

Calvete, O., Martinez, P., Garcia-Pavia, P., Benitez-Buelga, C., PaumardHernández, B., Fernandez, V., Dominguez, F., Salas, C., Romero-Laorden, N., Garcia-Donas, J., et al. (2015). A mutation in the POT1 gene is responsible for cardiac angiosarcoma in TP53-negative Li-Fraumeni-like families. Nat. Commun. 6, 8383. Carro, A., Perez-Martinez, M., Soriano, J., Pisano, D.G., and Megias, D. (2015). iMSRC: converting a standard automated microscope into an intelligent screening platform. Sci. Rep. 5, 10502. Chen, J., McKay, R.M., and Parada, L.F. (2012). Malignant glioma: lessons from genomics, mouse models, and stem cells. Cell 149, 36–47. Chin, L., Artandi, S.E., Shen, Q., Tam, A., Lee, S.L., Gottlieb, G.J., Greider, C.W., and DePinho, R.A. (1999). p53 deficiency rescues the adverse effects of telomere loss and cooperates with telomere dysfunction to accelerate carcinogenesis. Cell 97, 527–538. De Lange, T. (2005). Shelterin: the protein complex that shapes and safeguards human telomeres. Genes Dev. 19, 2100–2110. El Fassi, D. (2015). Telomerase inhibitor imetelstat in essential thrombocythemia and myelofibrosis. N. Engl. J. Med. 373, 2579–2581. Faiz, M., Sachewsky, N., Gascón, S., Bang, K.W., Morshead, C.M., and Nagy, A. (2015). Adult neural stem cells from the subventricular zone give rise to reactive astrocytes in the cortex after stroke. Cell Stem Cell 17, 624–634. Fang, X., Zhou, W., Wu, Q., Huang, Z., Shi, Y., Yang, K., Chen, C., Xie, Q., Mack, S.C., Wang, X., et al. (2017). Deubiquitinase USP13 maintains glioblastoma stem cells by antagonizing FBXL14-mediated Myc ubiquitination. J. Exp. Med. 214, 245–267. Garcı́a-Beccaria, M., Martı́nez, P., Méndez-Pertuz, M., Martı́nez, S., BlancoAparicio, C., Cañamero, M., Mulero, F., Ambrogio, C., Flores, J.M., Megias, D., et al. (2015). Therapeutic inhibition of TRF1 impairs the growth of p53-deficient K-RasG12V-induced lung cancer by induction of telomeric DNA damage. EMBO Mol. Med. 7, 930–949. Gonzalez-Suarez, E., Samper, E., Flores, J.M., and Blasco, M.A. (2000). Telomerase-deficient mice with short telomeres are resistant to skin tumorigenesis. Nat. Genet. 26, 114–117. Goytisolo, F.A., Samper, E., Martı́n-Caballero, J., Finnon, P., Herrera, E., Flores, J.M., Bouffler, S.D., and Blasco, M.A. (2000). Short telomeres result in organismal hypersensitivity to ionizing radiation in mammals. J. Exp. Med. 192, 1625–1636. Greider, C.W., and Blackburn, E.H. (1985). Identification of a specific telomere terminal transferase activity in tetrahymena extracts. Cell 43, 405–413. Hambardzumyan, D., Amankulor, N.M., Helmy, K.Y., Becher, O.J., and Holland, E.C. (2009). Modeling adult gliomas using RCAS/t-va technology. Transl. Oncol. 2, 89–95. Hanahan, D., and Weinberg, R.A. (2011). Hallmarks of cancer: the next generation. Cell 144, 646–674. Harley, C.B., Futcher, A.B., and Greider, C.W. (1990). Telomeres shorten during ageing of human fibroblasts. Nature 345, 458–460.

Bhat, K.P.L., Balasubramaniyan, V., Vaillant, B., Ezhilarasan, R., Hummelink, K., Hollingsworth, F., Wani, K., Heathcock, L., James, J.D., Goodman, L.D., et al. (2013). Mesenchymal differentiation mediated by NF-kB promotes radiation resistance in glioblastoma. Cancer Cell 24, 331–346.

Hasegawa, D., Okabe, S., Okamoto, K., Nakano, I., Shin-ya, K., and Seimiya, H. (2016). G-quadruplex ligand-induced DNA damage response coupled with telomere dysfunction and replication stress in glioma stem cells. Biochem. Biophys. Res. Commun. 471, 75–81.

Boldrini, L., Pistolesi, S., Gisfredi, S., Ursino, S., Ali, G., Pieracci, N., Basolo, F., Parenti, G., and Fontanini, G. (2006). Telomerase activity and hTERT mRNA expression in glial tumors. Int. J. Oncol. 28, 1555–1560.

Heaphy, C.M., de Wilde, R.F., Jiao, Y., Klein, A.P., Edil, B.H., Shi, C., Bettegowda, C., Rodriguez, F.J., Eberhart, C.G., Hebbar, S., et al. (2011). Altered telomeres in tumors with ATRX and DAXX mutations. Science 333, 425.

Boué, S., Paramonov, I., Barrero, M.J., and Izpisúa Belmonte, J.C. (2010). Analysis of human and mouse reprogramming of somatic cells to induced pluripotent stem cells. what is in the plate? PLoS One 5, 1–14. Bowman, R.L., Wang, Q., Carro, A., Verhaak, R.G., and Squatrito, M. (2017). GlioVis data portal for visualization and analysis of brain tumor expression datasets. Neuro Oncol. 19, 139–141. Bryan, T.M., Englezou, A., Dalla-Pozza, L., Dunham, M.A., and Reddel, R.R. (1997). Evidence for an alternative mechanism for maintaining telomere length in human tumors and tumor-derived cell lines. Nat. Med. 3, 1271–1274.

606 Cancer Cell 32, 590–607, November 13, 2017

Holland, E.C., Hively, W.P., DePinho, R.A., and Varmus, H.E. (1998). A constitutively active epidermal growth factor receptor cooperates with disruption of G1 cell-cycle arrest pathways to induce glioma-like lesions in mice. Genes Dev. 12, 3675–3685. Hu, H., Zhang, Y., Zou, M., Yang, S., and Liang, X.Q. (2010). Expression of TRF1, TRF2, TIN2, TERT, KU70, and BRCA1 proteins is associated with telomere shortening and may contribute to multistage carcinogenesis of gastric cancer. J. Cancer Res. Clin. Oncol. 136, 1407–1414.

Jiang, Y., Boije, M., Westermark, B., and Uhrbom, L. (2011). PDGF-B can sustain self-renewal and tumorigenicity of experimental glioma-derived cancerinitiating cells by preventing oligodendrocyte differentiation. Neoplasia 13, 492–503.

Pal, D., Sharma, U., Singh, S.K., Kakkar, N., and Prasad, R. (2015). Overexpression of telomere binding factors (TRF1 & TRF2) in renal cell carcinoma and their inhibition by using SiRNA induce apoptosis, reduce cell proliferation and migration in vitro. PLoS One 10, e0115651.

Kanzawa, T., Germano, I.M., Kondo, Y., Ito, H., Kyo, S., and Kondo, S. (2003). Inhibition of telomerase activity in malignant glioma cells correlates with their sensitivity to temozolomide. Br. J. Cancer 89, 922–929.

Parkhurst, M.R., Riley, J.P., Igarashi, T., Li, Y., Robbins, P.F., and Rosenberg, S.A. (2004). Immunization of patients with the hTERT:540-548 peptide induces peptide-reactive T lymphocytes that do not recognize tumors endogenously expressing telomerase. Clin. Cancer Res. 10, 4688–4698.

Kim, N., Piatyszek, M.A., Prowse, K.R., Harley, C.B., West, M.D., Ho, P.L., Coviello, G.M., Wright, W.E., Weinrich, S.L., and Shay, J.W. (1994). Specific association of human telomerase activity with immortal cells and cancer. Science 266, 2011–2015. Lathia, J.D., Mack, S.C., Mulkearns-Hubert, E.E., Valentim, C.L., and Rich, J.N. (2015). Cancer stem cells in glioblastoma. Genes Dev. 29, 1203–1217. Liu, D., O’Connor, M.S., Qin, J., and Songyang, Z. (2004). Telosome, a mammalian telomere-associated complex formed by multiple telomeric proteins. J. Biol. Chem. 279, 51338–51342. Lobanova, A., She, R., Pieraut, S., Clapp, C., Maximov, A., and Denchi, E.L. (2017). Different requirements of functional telomeres in neural stem cells and terminally differentiated neurons. Genes Dev. 31, 639–647. Louis, D.N., Ohgaki, H., Wiestler, O.D., Cavenee, W.K., Burger, P.C., Jouvet, A., Scheithauer, B.W., and Kleihues, P. (2007). The 2007 WHO Classification of Tumours of the Central Nervous System. Acta Neuropathol. 114, 97, 114, 547. Marian, C.O., Cho, S.K., McEllin, B.M., Maher, E.A., Hatanpaa, K.J., Madden, C.J., Mickey, B.E., Wright, W.E., Shay, J.W., and Bachoo, R.M. (2010). The telomerase antagonist, imetelstat, efficiently targets glioblastoma tumor-initiating cells leading to decreased proliferation and tumor growth. Clin. Cancer Res. 16, 154–163. Marion, R.M., Strati, K., Li, H., Tejera, A., Schoeftner, S., Ortega, S., Serrano, M., and Blasco, M.A. (2009). Telomeres acquire embryonic stem cell characteristics in induced pluripotent stem cells. Cell Stem Cell 4, 141–154. Martı́nez, P., and Blasco, M.A. (2011). Telomeric and extra-telomeric roles for telomerase and the telomere-binding proteins. Nat. Rev. Cancer 11, 161–176. Martı́nez, P., Thanasoula, M., Muñoz, P., Liao, C., Tejera, A., McNees, C., Flores, J.M., Fernández-Capetillo, O., Tarsounas, M., and Blasco, M.A. (2009). Increased telomere fragility and fusions resulting from TRF1 deficiency lead to degenerative pathologies and increased cancer in mice. Genes Dev. 23, 2060–2075. Middleton, G., Silcocks, P., Cox, T., Valle, J., Wadsley, J., Propper, D., Coxon, F., Ross, P., Madhusudan, S., Roques, T., et al. (2014). Gemcitabine and capecitabine with or without telomerase peptide vaccine GV1001 in patients with locally advanced or metastatic pancreatic cancer (TeloVac): an open-label, randomised, phase 3 trial. Lancet Oncol. 15, 829–840. Mignone, J.L., Kukekov, V., Chiang, A.S., Steindler, D., and Enikolopov, G. (2004). Neural stem and progenitor cells in nestin-GFP transgenic mice. J. Comp. Neurol. 469, 311–324. Nonoguchi, N., Ohta, T., Oh, J.E., Kim, Y.H., Kleihues, P., and Ohgaki, H. (2013). TERT promoter mutations in primary and secondary glioblastomas. Acta Neuropathol. 126, 931–937. Ozawa, T., Riester, M., Cheng, Y.K., Huse, J.T., Squatrito, M., Helmy, K., Charles, N., Michor, F., and Holland, E.C. (2014). Most human non-GCIMP glioblastoma subtypes evolve from a common proneural-like precursor glioma. Cancer Cell 26, 288–300.

Ramsay, A.J., Quesada, V., Foronda, M., Conde, L., Martı́nez-Trillos, A., Villamor, N., Rodrı́guez, D., Kwarciak, A., Garabaya, C., Gallardo, M., et al. (2013). POT1 mutations cause telomere dysfunction in chronic lymphocytic leukemia. Nat. Genet. 45, 526–530. Robles-Espinoza, C.D., Harland, M., Ramsay, A.J., Aoude, L.G., Quesada, V., Ding, Z., Pooley, K.A., Pritchard, A.L., Tiffen, J.C., Petljak, M., et al. (2014). POT1 loss-of-function variants predispose to familial melanoma. Nat. Genet. 46, 478–481. Rohle, D., Popovici-Muller, J., Palaskas, N., Turcan, S., Grommes, C., Campos, C., Tsoi, J., Clark, O., Oldrini, B., Komisopoulou, E., et al. (2013). An inhibitor of mutant IDH1 delays growth and promotes differentiation of glioma cells. Science 340, 626–630. Ruzankina, Y., Pinzon-Guzman, C., Asare, A., Ong, T., Pontano, L., Cotsarelis, G., Zediak, V.P., Velez, M., Bhandoola, A., and Brown, E.J. (2007). Deletion of the developmentally essential gene ATR in adult mice leads to age-related phenotypes and stem cell loss. Cell Stem Cell 1, 113–126. Schneider, R.P., Garrobo, I., Foronda, M., Palacios, J.A., Marión, R.M., Flores, I., Ortega, S., and Blasco, M.A. (2013). TRF1 is a stem cell marker and is essential for the generation of induced pluripotent stem cells. Nat. Commun. 4, 1946. Serrano, M., Lee, H., Chin, L., Cordon-Cardo, C., Beach, D., and DePinho, R.A. (1996). Role of the INK4a locus in tumor suppression and cell mortality. Cell 85, 27–37. Sfeir, A., Kosiyatrakul, S.T., Hockemeyer, D., MacRae, S.L., Karlseder, J., Schildkraut, C.L., and de Lange, T. (2009). Mammalian telomeres resemble fragile sites and require TRF1 for efficient replication. Cell 138, 90–103. Shay, J.W., and Bacchetti, S. (1997). A survey of telomerase activity in human cancer. Eur. J. Cancer 33, 787–791. Singh, S.K., Hawkins, C., Clarke, I.D., Squire, J.A., Bayani, J., Hide, T., Henkelman, R.M., Cusimano, M.D., and Dirks, P.B. (2004). Identification of human brain tumour initiating cells. Nature 432, 396–401. Soeda, A., Hara, A., Kunisada, T., Yoshimura, S., Iwama, T., and Park, D.M. (2015). The evidence of glioblastoma heterogeneity. Sci. Rep. 5, 7979. Tejera, A.M., Stagno d’Alcontres, M., Thanasoula, M., Marion, R.M., Martinez, P., Liao, C., Flores, J.M., Tarsounas, M., and Blasco, M.A. (2010). TPP1 is required for TERT recruitment, telomere elongation during nuclear reprogramming, and normal skin development in mice. Dev. Cell 18, 775–789. Tomás-Loba, A., Flores, I., Fernández-Marcos, P.J., Cayuela, M.L., Maraver, A., Tejera, A., Borrás, C., Matheu, A., Klatt, P., Flores, J.M., et al. (2008). Telomerase reverse transcriptase delays aging in cancer-resistant mice. Cell 135, 609–622. Wen, P.Y., and Kesari, S. (2008). Malignant gliomas in adults. N. Engl. J. Med. 359, 492–507. Yang, M., and Crawley, J.N. (2009). Simple behavioral assessment of mouse olfaction. Curr. Protoc. Neurosci. Chapter 8. Unit 8.24.

Cancer Cell 32, 590–607, November 13, 2017 607

Cell Stem Cell

Article Secreted Phospholipases A2 Are Intestinal Stem Cell Niche Factors with Distinct Roles in Homeostasis, Inflammation, and Cancer Matthias Schewe,1 Patrick F. Franken,1 Andrea Sacchetti,1 Mark Schmitt,1 Rosalie Joosten,1 René Böttcher,2 Martin E. van Royen,1,3 Louise Jeammet,4 Christine Payré,4 Patricia M. Scott,5 Nancy R. Webb,6 Michael Gelb,7 Robert T. Cormier,5 Gérard Lambeau,4 and Riccardo Fodde1,* 1Department

of Pathology of Urology 3Erasmus Optical Imaging Centre Erasmus MC Cancer Institute, Rotterdam 3000CA, The Netherlands 4Institute of Molecular and Cellular Pharmacology, Centre National de la Recherche Scientifique and University of Nice Sophia Antipolis, Valbonne 06560, France 5Department of Biomedical Sciences, University of Minnesota Medical School Duluth, Duluth, MN 55812-3031, USA 6Department of Pharmacology and Nutritional Sciences, University of Kentucky, Lexington, KY 40506-9983, USA 7Department of Chemistry, University of Washington, Seattle, WA 98195-1700, USA *Correspondence: r.fodde@erasmusmc.nl http://dx.doi.org/10.1016/j.stem.2016.05.023 2Department

SUMMARY

The intestinal stem cell niche provides cues that actively maintain gut homeostasis. Dysregulation of these cues may compromise intestinal regeneration upon tissue insult and/or promote tumor growth. Here, we identify secreted phospholipases A2 (sPLA2s) as stem cell niche factors with contextdependent functions in the digestive tract. We show that group IIA sPLA2, a known genetic modifier of mouse intestinal tumorigenesis, is expressed by Paneth cells in the small intestine, while group X sPLA2 is expressed by Paneth/goblet-like cells in the colon. During homeostasis, group IIA/X sPLA2s inhibit Wnt signaling through intracellular activation of Yap1. However, upon inflammation they are secreted into the intestinal lumen, where they promote prostaglandin synthesis and Wnt signaling. Genetic ablation of both sPLA2s improves recovery from inflammation but increases colon cancer susceptibility due to release of their homeostatic Wnt-inhibitory role. This ‘‘trade-off’’ effect suggests sPLA2s have important functions as genetic modifiers of inflammation and colon cancer.

INTRODUCTION In 1993, by taking advantage of an inbred mouse model predisposed to multiple intestinal neoplasia (Min), Dove and collaborators identified the Mom-1 (modifier of Min) locus as a major modifier of intestinal tumor multiplicity driven by mutations in the Apc (adenomatous polyposis coli) tumor suppressor gene (Dietrich et al., 1993). Subsequently, Pla2g2a, encoding the secreted type IIA phospholipase A2 (sPLA2-IIA), was identified as the 38 Cell Stem Cell 19, 38–51, July 7, 2016 ª 2016 Elsevier Inc.

main responsible gene within the Mom-1 locus (MacPhee et al., 1995). sPLA2-IIA and other sPLA2 members produce biologically active lipid mediators such as lysophosphatidic acid (LPA) and arachidonic acid (AA), the latter a substrate for prostaglandin E2 (PGE2) biosynthesis (Lambeau and Gelb, 2008). Notably, Mom-1-susceptible strains (e.g., C57BL/6J) carry a single nucleotide insertion in the Pla2g2a gene, resulting in a null allele, whereas Mom-1-resistant strains (e.g., FVB/N, BALB/c, CBA) are Pla2g2a proficient (MacPhee et al., 1995). More definitive evidence for the identity of the Pla2g2a gene as Mom-1 came from a transgenic mouse overexpressing Pla2g2a in the C57BL/6J inbred genetic background that, in the presence of the ApcMin mutation, causes a significant reduction in tumor multiplicity, though without fully recapitulating the decrease in tumor multiplicity observed in resistant Mom-1 strains (Cormier et al., 1997, 2000). Somatic inactivation of the human APC gene is a rate-limiting and initiating event in the vast majority of sporadic colon cancer cases (Fodde et al., 2001). Based on the above mouse studies, PLA2G2A, i.e., the human ortholog of Mom-1/Pla2g2a, was thought to represent a modifier of bowel cancer risk in the general population. However, studies failed to find either somatic or germline PLA2G2A alterations in sporadic colon tumors or hereditary colon cancer patients, respectively (Riggins et al., 1995; Spirio et al., 1996). This may be due to the specific expression of the PLA2G2A/Pla2g2a gene in Paneth cells, a secretory lineage almost exclusively found in the small intestine and only to a lesser extent in the distal colon (Cormier et al., 2000; Mounier et al., 2008). Accordingly, Apc mutant mice mainly develop multiple upper GI (gastrointestinal) adenomas with very infrequent colonic lesions. Therefore, other members of the sPLA2 multigene family may act as colon cancer genetic modifiers in man. The Paneth-cell-specific expression pattern of Pla2g2a is of interest in view of the ‘‘niche’’ role played by these secretory cells in supporting Lgr5+ intestinal stem cells (ISCs) in the intestinal crypt (Sato et al., 2011). Moreover, both Paneth cells and their

secretory precursors were shown to represent infrequently dividing stem-like cells capable of re-entering the cell cycle upon tissue insults, thus contributing to the regenerative response (Buczacki et al., 2013; Roth et al., 2012). Therefore, we hypothesized that sPLA2s may represent ISC niche factors with specific functional roles in homeostasis, inflammation, and cancer. RESULTS Pla2g2a Expression Modulates ISC Function and Paneth Cell Differentiation To test the hypothesis that Pla2g2a is an important stem cell niche factor, we employed the ‘‘mini-gut’’ assay (Sato et al., 2011) as a readout of stem cell function in small intestinal crypts from Mom-1-sensitive (Mom1S or Pla2g2a / ; C57BL/6J and 129/SW) or Mom-1-resistant (Mom1R or Pla2g2a+/+; FVB/N, BALB/c, CBA) inbred mouse lines (MacPhee et al., 1995). The multiplicity of organoids derived from Mom1S strains is significantly higher when compared with Mom1R mice (Figure 1A). However, these inbred strains differ at several loci other than Pla2g2a, and thus differences in organoid formation cannot be attributed directly to the Mom-1 locus. Therefore, we employed a transgenic mouse model in which expression of the Pla2g2a gene is restored in the C57BL/6J background (Tg-Pla2g2a) (Cormier et al., 1997) at slightly higher levels ( 1.5- to 2-fold at the protein level) compared to Pla2g2a-proficient inbred strains such as FVB/N (Figure S1, available online). There was a significant decrease in organoid formation in both male and female transgenic mice compared with Pla2g2a / (C57BL/6J) littermates (Figure 1B), confirming that expression of the Pla2g2a gene negatively affects the ability of isolated crypts to form intestinal organoids. The overall organoid morphology did not differ among strains (data not shown). The Pla2g2a gene is mainly, yet not exclusively, expressed in the small intestine and in particular by Paneth cells (Figure S1A). As Pla2g2a inhibits organoid growth, we investigated the histopathology of the digestive tract of Tg-Pla2g2a mice. Notably, immunohistochemical (IHC) analysis revealed an almost complete absence of lysozyme-positive Paneth cells in the transgenic mice (Figure 1C). In addition, the proliferating (Ki67+) compartment, normally limited to the transient amplifying (TA) progenitors in control C57BL/6J mice, was extended to the lower crypt of Tg-Pla2g2a mice, where post-mitotic Paneth cells are usually located (Figure 1D). However, granulated Paneth-like cells were present in the crypt bottom of transgenic animals (Figure S2A). qRT-PCR (Figure 1E) and IHC (Figure S2B) analysis of additional lineage-specific markers confirmed the absence of mature Paneth cells in the Tg-Pla2g2a model. Furthermore, IHC (Figures S3A and S3B) and fluorescence-activated cell sorting (FACS) (Figures S3C and S3D) analyses of lineage-specific markers confirmed the accumulation of the Lgr5+/ChgA+ (chromogranin A) secretory (Paneth and goblet cells) precursor lineage (Buczacki et al., 2013) in transgenic animals. Notably, the overall FACS pattern obtained from C57BL/6J and Tg-Pla2g2a mice did not differ with the CD24/SSC marker combination (data not shown). Thus, the results point to a reduction in fully mature Paneth cells rather than the complete ablation of this lineage in the transgenic mice, possibly due to a maturation defect.

Overall, Pla2g2a may modulate ISC function by inhibiting Paneth cell differentiation and full maturation. Intracellular Pla2g2a Inhibits Wnt Signaling by Increasing Yap1 Phosphorylation The above experiments were carried out under homeostatic conditions, i.e., in mice kept under normal conditions and fed ad libitum. As Pla2g2a encodes the sPLA2-IIA protein whose expression and secretion are highly enhanced during inflammation but are thought to be kept at a basal level during homeostasis, we investigated whether the observed inhibitory effects on intestinal organoid growth are exerted by the intracellular pool or by the basal secreted fraction of sPLA2-IIA. The organoid assay was repeated on crypts from FVB (Pla2g2a+/+), C57BL/ 6J (Pla2g2a / ), and Tg-Pla2g2a mice in the presence of the soluble (non-membrane-bound) sPLA2-IIA receptor Pla2r1 to scavenge any secreted Pla2g2a protein (Rouault et al., 2007). The lack of effect on organoid multiplicity (Figure S3E) indicated that the inhibitory effect exerted by Pla2g2a is likely due to its intracellular (non-secreted) fraction. The canonical Wnt signaling plays a role in both ISC maintenance and in Paneth cell maturation (van Es et al., 2005a). As such, it may underlie the organoid formation and Paneth cell maturation defects observed in Pla2g2a-expressing transgenic mice. Therefore, we analyzed the effects of PLA2G2A expression on Wnt signaling in human colon cancer cell lines with constitutive pathway activity due to mutations in the APC or b-catenin genes. Transient expression of PLA2G2A caused significant downregulation of Wnt signaling activity in all three lines (Figure 2A). To elucidate the mechanisms underlying Pla2g2a-driven inhibition of Wnt signaling, we analyzed previously published intestinal expression profiling studies comparing Tg-Pla2g2a mice to C57BL/6J controls (Fijneman et al., 2008, 2009) and found upregulation of Yap1 (yes-associated protein 1) in intestinal cells from Tg-Pla2g2a mice (2.63-fold; p value 0.0072). Although Yap1 has oncogenic and growth-stimulating capacities, it also has unexpected growth-suppressive functions; while nuclear Yap1 promotes growth, its phosphorylation by Hippo kinases prevents nuclear translocation and restricts Wnt signaling (Azzolin et al., 2014; Barry et al., 2013; Camargo et al., 2007). We FACS purified Paneth cells from Lgr5–EGFP–ires–CreERT2 reporter mice (Barker et al., 2007) on Tg-Pla2g2a and C57BL/6J backgrounds, as previously described (Roth et al., 2012). Yap1 was specifically upregulated in Paneth cells from Tg-Pla2g2a mice (Figure 2B), reflecting the Paneth-specific Pla2g2a expression pattern in those animals (Figure S1B). In contrast, Pla2g2a expression levels did not differ between Lgr5+ stem cells from control and transgenic animals. Moreover, Yap1 was expressed at higher levels in Paneth cells from FVB/N (Pla2g2a+/+) mice (Figure 2B). This was confirmed by IHC and IF analysis: whereas in wildtype (C57BL/6J) crypts only the slender Lgr5+ stem cells stain positive for Yap1, clear Yap1 cytoplasmic staining was observed in the (immature) Paneth cells from Tg-Pla2g2a mice (Figures 2C and S4C). FACS analysis of wild-type (C57BL/6J) and TgPla2g2a mice confirmed increased YAP1 phosphorylation in Paneth cells from the transgenic animals (Figures S4A and S4B), consistent with decreased organoid formation in these mice. Cell Stem Cell 19, 38–51, July 7, 2016 39

Figure 1. Pla2g2a Expression Modulates ISC Function and Paneth Cell Differentiation (A) Organoid assays performed with crypts from indicated Mom-1-sensitive or Mom-1-resistant strains. Differences are statistically significant (p < 0.001) except for TgPla2g2a versus CBA/J and FVB/N versus BALB/c. (B) Organoid assays performed on C57BL/6J (Pla2g2a / ) and Tg-Pla2g2a male and female mice (n = 5, *p < 0.001). (C) Lysozyme IHC analysis of duodenal sections of C57BL/6J (Pla2g2a / ) and Tg-Pla2g2a mice. (D) Ki67 IHC analysis of small intestinal crypts (duodenum and ileum) from Pla2g2a / (C57BL/6J) and Tg-Pla2g2a mice. (E) Heatmap representation qRT-PCR expression analysis of Paneth- and stem-cell-specific genes from FACS-purified CD24/SSC lineages.

We then confirmed the central role of Paneth cells and of Yap1 downstream of Pla2g2a in the regulation of the ISC niche during homeostasis. Organoid formation was assessed by sorting and co-incubating Lgr5+ stem cells from control (C57BL/6J; Pla2g2a / ) mice with Paneth cells from Tg-Pla2g2a mice, and vice versa (Figure 3A). A substantial decrease in organoid numbers was observed only when Pla2g2a-expressing Paneth 40 Cell Stem Cell 19, 38â&#x20AC;&#x201C;51, July 7, 2016

cells from transgenic mice were mixed with control Lgr5+ stem cells (Figure 3B), demonstrating that inhibitory effects of Pla2g2a expression on organoid formation are largely mediated by the immature Paneth cells. Live cell time-lapse imaging of the re-association of sorted single Lgr5+ stem cells (EGFP+; green) and single Paneth cells from wild-type C57BL/6J mice confirmed that physical contact between the stem cells and Paneth cells

Figure 2. Pla2g2a Expression Modulates Wnt Signaling through Yap1 Phosphorylation (A) TOP-Flash luciferase reporter analysis of Wnt signaling activity in the colon cancer cell lines SW480, HCT116, and HT29 upon transient transfection with a PLA2G2A expression vector in the presence/absence of Wnt3a conditioned medium (n = 3, *p < 0.05). (B) Yap1 qRT-PCR analysis in Paneth cells (CD24hiSSChi), secretory precursors (CD24medSSClo), and stem cells (Lgr5+CD24medSSClo) sorted by FACS from C57BL/6J (Pla2g2a / ) and Tg-Pla2g2a mice in the Lgr5EGFP-IRES-creERT2 (Lgr5+) background (left). CD24hiSSChi Paneth cells were also sorted from Pla2g2a+/+ (FVB/N) mice and compared with the corresponding subpopulations sorted from Pla2g2a / (C57BL/6J) and Tg-Pla2g2a mice (right) (n = 3, **p < 0.001). (C) Phospho-Yap1 IHC analysis of small intestinal crypts from C57BL/6J (Pla2g2a / ; left panel) and Tg-Pla2g2a mice (right panel). The black and white arrows indicate an Lgr5+ stem cell (positive for Yap1) and a Paneth cell (negative for Yap1), respectively.

is required for organoid formation (Sato et al., 2011). Multiple cell clusters formed and occasionally fused to form larger structures (Movie S1). However, when Paneth cells from Tg-Pla2g2a mice were employed, their interactions with the EGFP+ Lgr5+ stem cells were significantly decreased, as was the corresponding number of organoids (Movie S2). The relationship between Pla2g2a and Yap1 was further confirmed by small interfering RNA (siRNA)-mediated knockdown in Paneth cells purified from C57BL6/J, Tg-Pla2g2a, and Pla2g2a-proficient FVB/N animals. Yap1 is expressed at higher levels in the FVB/N Paneth cells compared with Paneth cells from Tg-Pla2g2a mice (Figure 2B) and, as expected, with those from the Pla2g2a-deficient strain C57BL/6J. Accordingly, siRNA-driven knockdown of Pla2g2a or Yap1 in Paneth cells from Tg-Pla2g2a and FVB/N mice effectively rescued the inhibitory effects of Plag2g2a on organoid formation (Figure 3C). Similarly, siRNA-driven YAP1 knockdown in human colon

cancer cells fully rescued the Wnt-inhibitory effects of transient PLA2G2A expression (Figure 3D). Overall, these data indicate that intracellular Pla2g2a expression inhibits Wnt signaling in Paneth cells, through increased Yap1 expression and phosphorylation, to negatively regulate the ability of ISCs to form organoids. Consistently, transgenic Pla2g2a expression in the C57BL/ 6J inbred genetic background blocks Paneth cell maturation, leading to the accumulation of secretory precursors (see Graphical Abstract). Secreted Pla2g2a Enhances ISC Function upon Inflammation Pla2g2a expression and secretion is increased upon inflammation (Lambeau and Gelb, 2008), and its concentration is expected to increase in the intestinal lumen. To mimic and assess the effects of secreted Pla2g2a on the ISC niche, the organoid assays were repeated from C57BL/6J mice with recombinant Cell Stem Cell 19, 38â&#x20AC;&#x201C;51, July 7, 2016 41

Figure 3. Reconstitution Organoid Assays Highlight the Role of Pla2g2a in Inhibiting Wnt Signaling through Yap1 (A) Flowchart of the reconstitution organoid assay. Single-cell suspensions were obtained from small intestinal crypts isolated from C57BL/6J (Pla2g2a / ) and Tg-Pla2g2a mice previously bred with Lgr5-EGFP reporter animals (Lgr5EGFP-IRES-creERT2). Purified Lgr5+ stem cells (Lgr5+CD24medSSClo) and Paneth cells (CD24hiSSChi) were mixed and plated in matrigel for organoid growth. (B) Results of the reconstitution organoid assay with Paneth cells (PCs) and Lgr5+ stem cells (SCs) from C57BL/6J (B6) and Tg-Pla2g2a (Tg) mice previously bred with Lgr5-EGFP reporter animals (Lgr5EGFP-IRES-creERT2) (n = 3, *p < 0.05; **p < 0.001). (C) Organoid formation following reconstitution of Lgr5+stem cells (SCs) and Paneth cells (PCs) sorted from Pla2g2a / (B6), Tg-Pla2g2a (Tg), and FVB/N (FVB) mice. When indicated, PCs were pretreated with siRNA oligonucleotides directed against Pla2g2a, Yap1, or a ‘‘scrambled’’ (SCR) control sequence (n = 3, *p < 0.05; **p < 0.001). (1) B6-PCs + B6-SCs, (2) Tg-PCs + Tg-SCs, (3) Tg-PCs [siRNA-Pla2g2a] + B6-SCs, (4) Tg-PCs [siRNA-Yap1] + B6-SCs, (5) Tg-PCs [siRNASCR] + B6-SCs, (6) Tg-PCs, (7) B6-PCs + B6-SCs, (8) FVB-PCs + B6-SCs, (9) FVB-PCs [siRNA-Pla2g2a] + B6-SCs, (10) FVB-PCs [siRNA-Yap1] + B6-SCs, (11) FVB-PCs [siRNA-SCR] + B6-SCs, and (12) FVB-PCs. (D) TOP-Flash reporter analysis of Wnt signaling activity in the colon cancer cell line SW480 upon transient transfection with a PLA2G2A expression vector in the presence/absence of siRNA-driven YAP1 downregulation (n = 3, *p < 0.05).

Pla2g2a protein added to the culture medium. A >2-fold increase (p < 0.001) was observed in organoid formation upon addition of recombinant sPLA2-IIA even at low concentrations (i.e., in the nM range; Figure 4A). Repeating the experiment in the presence of soluble Pla2r1 receptor, which sequesters the recombinant sPLA2-IIA protein with high affinity (Rouault et al., 2007), completely prevented these stimulatory effects (Figure 4A). We then utilized an in vivo model of inflammation to confirm increased Pla2g2a secretion exerts a positive effect on ISC function. Inflammation was induced in C57BL/6J (Pla2g2a / ), FVB/ N (Pla2g2a+/+), and Tg-Pla2g2a mice by adding 3% dextran sodium sulfate (DSS) to their drinking water for a week. This results 42 Cell Stem Cell 19, 38–51, July 7, 2016

in loose stools, fecal bleeding, infiltration of the mucosa with granulocytes, and significant body weight loss (Wirtz et al., 2007). The DSS treatment causes an inflammatory response throughout the entire GI tract (pan-gastroenteritis) (Yazbeck et al., 2011) and is therefore suited to study the role of Pla2g2a secretion in the small intestine. Notably, the overall disease activity index (DAI) (Cooper et al., 1993) following DSS treatment, here employed as a general indicator of the detrimental effects of inflammation, was moderately improved in Tg-Pla2g2a compared to C57BL/6J mice (DAI = 5.2 and 8.1, respectively; Table S1). The subtle nature of this effect was confirmed by the observation that histologic parameters (i.e., CD3- and

Figure 4. Secreted Pla2g2a Enhances ISC Function through PGE2 Synthesis and Wnt Signaling upon Inflammation (A) C57BL/6J organoid formation in the absence/presence of recombinant Pla2g2a and soluble Pla2r1 receptor (n = 5, *p < 0.001). (B) Quantified organoid formation following DSS treatment of C57BL/6J (Pla2g2a / ), Tg-Pla2g2a, and FVB/N (Pla2g2a+/+) mice, in the absence/presence of soluble Pla2r1 (n = 3, *p < 0.001; **p < 0.05). (C) Results of organoid assays performed on Pla2g2a / (C57BL/6J) mice in the presence of agonists (AA, PGE2) and antagonists (NSAIDs; ketoprofen and diclofenac), the latter in the presence of recombinant Pla2g2a (n = 5, *p < 0.001). (D) Results of organoid assays from C57BL/6J versus Pla2r1 / mice in the absence/presence of recombinant Pla2g2a (n = 5, *p < 0.001). (E) qRT-PCR expression analysis of Pla2r1 in sorted Lgr5+, Paneth, and p16 (Paneth-like) cells of the small (SI) and large (LI) intestine. Control qRT-PCR analyses were performed to validate the quality and identity of the sorted stem (Lgr5), Paneth (Lyz), and p16 (Pla2g10) cells. (F) Left panel (2–9): organoid numbers following reconstitution of Paneth cells (PCs) (2–5) or Lgr5+ stem cells (SCs) (6–9) sorted from the small intestine of Lgr5reporter C57BL/6J (B6) mice pretreated with recombinant Pla2g2a. When indicated, sorted cells were also treated with siRNA oligonucleotides specific for Pla2r1 and Cox2, or scrambled sequences (SCR) (n = 3, *p < 0.05; **p < 0.001). (1) B6-PCs + B6-SCs (negative control; no rec. Pla2g2a), (2) B6-PCs + B6-SCs, (3) B6-PCs [siRNA-SCR] + B6-SCs, (4) B6-PCs [siRNA-Pla2R1] + B6-SCs, (5) B6-PCs [siRNA-Cox2] + B6-SCs, (6) B6-PCs + B6-SCs, (7) B6-PCs + B6-SCs [siRNA-SCR], (8) B6-PCs + B6-SCs [siRNA-Pla2R1], and (9) B6-PCs + B6-SCs [siRNA-Cox2]. Right panel (10–12): organoid numbers upon reconstitution of B6-PCs with B6-SCs in (legend continued on next page)

Cell Stem Cell 19, 38–51, July 7, 2016 43

MUC1-positive cells, presence of infiltrating cells and ulcers, and destruction of tissue architecture) did not differ between the two genotypes (data not shown). We also confirmed a substantial increase in Pla2g2a secretion following DSS-induced inflammation by ELISA on stool samples from Tg-Pla2g2a and C57BL/6J mice (Figure S5A, left). Consistently, a striking increase in organoid multiplicity was observed upon DSS-induced inflammation from the Pla2g2aproficient strains (Tg-Pla2g2a and FVB/N mice) when compared to Pla2g2a-deficient C57BL/6J mice (Figure 4B). Notably, both Tg-Pla2g2a and FVB/N were characterized by a low organoidforming capacity under homeostatic conditions (Figures 1A and 1B). The soluble Pla2r1 scavenger inhibited this increase, confirming that Pla2g2a secretion underlies this measure of enhanced ISC function (Figure 4B). This conclusion was confirmed by treating organoids with the pro-inflammatory cytokine interferon-gamma (IFN-g) to trigger Paneth cell degranulation and release of Pla2g2a into the culture medium (Farin et al., 2014), which significantly increased organoid multiplicity from Tg-Pla2g2a, but not C57BL/6J, mice. sPLA2-IIA proteins play important roles in prostaglandin synthesis. sPLA2 proteins hydrolyze phospholipids to produce AA that is converted by cyclooxygenases (Cox1 and Cox2) into prostaglandin E2 (PGE2) (Lambeau and Gelb, 2008). Accordingly, a >100-fold increase in PGE2 synthesis was observed in intestinal organoids from C57BL/6J (Pla2g2a / ) upon addition of recombinant Pla2g2a to the culture medium (Figure S5C). To test whether the observed positive effect exerted by secreted Pla2g2a on ISC function is due to increased PGE2 production, we repeated the organoid assays in the presence of prostaglandin biosynthesis agonists (AA, PGE2) and antagonists (NSAIDs; ketoprofen and diclofenac). Both AA and PGE2 had positive effects on organoid multiplicity similar to recombinant Pla2g2a, whereas either NSAID inhibited the Pla2g2a-dependent increase in number of organoids obtained from C57BL/6J animals (Figure 4C). Additionally, changes in organoid morphology into enterospheres, spherical structures primarily composed of stem cells and that lack budding and branching structures (Stelzner et al., 2012), were noted upon addition of Pla2g2a, AA, and PGE2, and in organoids obtained by DSS-treated mice (Figure S5D). Conversion of organoids to enterospheres is regulated by Wnt signaling (Mustata et al., 2013). Consistent with these results and previous reports of synergism between PGE2 and the canonical Wnt pathway in hematopoietic stem cells (Goessling et al., 2009) and colon carcinoma cells (Castellone et al., 2005; Shao et al., 2005), PGE2 enhanced Wnt activity in SW480 and HCT116 colon cancer cell lines (Figure S5E). To elucidate how exogenous Pla2g2a triggers PGE2 biosynthesis, we tested the organoid-forming capacity of Pla2g2a receptor type 1 (Pla2r1) knockout mice in the presence/absence of recombinant Pla2g2a. While no change in organoid number was seen between wild-type (C57BL/6J) and Pla2r1 / mice in the absence of exogenous Pla2g2a, the latter did not respond to recombinant Pla2g2a (Figure 4D). Hence, Pla2r1 mediates

the positive effects of exogenous Pla2g2a on ISC function. We then determined that within the intestine, Pla2r1 expression was exclusively detected in Paneth cells (Figure 4E). To functionally validate this observation, we performed organoid reconstitution assays with Paneth cells pretreated with siRNA oligonucleotides against Pla2r1. Knockdown of Pla2r1 in Paneth cells, though not in Lgr5+ stem cells, abolished the positive effect of recombinant Pla2g2a on organoid multiplicity (Figure 4F, left). The cytosolic phospholipase A2 group IVa (Pla2g4a or cPLA2a) is activated by Pla2r1 to catalyze AA biosynthesis (Lambeau and Gelb, 2008). The specific cPLA2a inhibitor pyrrophenone suppressed the increase in organoid multiplicity observed by adding exogenous Pla2g2a (Figure 4G, left); also, organoids grown from Pla2g4a knockout mice (Bonventre et al., 1997) were insensitive to the stimulatory effects of exogenous Pla2g2a (Figure 4G, right). Additional reconstitution assays with sorted Paneth cells treated with PGE2 (Figure 4F, right) and anti-Cox2 siRNA oligonucleotides (Figure 4F, left) confirmed that prostaglandin biosynthesis plays a central role in mediating effects of secreted Pla2g2a on ISC function. In sum, these results show that secreted Pla2g2a positively regulates ISC proliferation through the Pla2r1 receptor, which activates Pla2g4a/cPLA2a. This triggers the release of AA, which is then converted by Cox1/2 into PGE2 to synergistically activate the canonical Wnt pathway. Pla2g10 Is Expressed by the Paneth-like Secretory Cell Lineage in the Mouse Colon Similarly to Paneth cells in supporting Lgr5+ ISCs in the small intestine, we hypothesized that a secretory cell lineage with similar niche functions may be present in the colon and could physically associate with Lgr5+ CBCs to promote organoid formation. A c-Kit+ subpopulation of secretory epithelial cells from the colon was recently shown to promote organoid formation from Lgr5+ ISCs (Rothenberg et al., 2012). We performed FACS analysis of the colonic epithelium from Lgr5-EGFP reporter mice with CD24 and c-Kit markers and defined four distinct sorting gates (p11, CD24med/cKitlo; p15, CD24med/cKitmed; p16, CD24hi/cKithi; p18, CD24med/cKithi) (Figure 5A). Within these gates, Lgr5+ cells were present either as single cells (mostly in p11) or in physical association (as doublets or triplets) with potential niche cells. Reconstitution organoid assays were then performed by incubating colonic Lgr5+ stem cells (p11; Lgr5+/CD24med/cKitlo) with the same number of (GFP ) single cells from each of the p15, p16, and p18 subpopulations. Lgr5+ doublets and triplets were excluded by sorting, as previously described (Roth et al., 2012). The p16 subpopulation (c-Kit+/CD24hi) unequivocally promotes organoid formation compared with all other subpopulations, each of which gave rise to no or very few organoids (Figure 5B). Confocal microscopy confirmed that only the p16 subpopulation contained doublets and triplets of smaller Lgr5+ stem cells and larger Lgr5-negative cells (these clusters were subsequently excluded during sorting; Figures 5C and S6). Sorted p16 cells expressed a mixed signature of

the presence of 1 mM PGE2 for 30 min. (10) Both B6-SCs and B6-PCs pretreated with PGE2, (11) only PCs pretreated with PGE2, and (12) only PCs pretreated with PGE2. (G) Organoid assays from C57BL/6J (Pla2g2a / ) mice in the presence/absence of pyrrophenone (PP) and recombinant Pla2g2a (left panel), and from C57BL/6J versus Pla2g4a / in the absence/presence of recombinant Pla2g2a (right panel) (n = 5, *p < 0.001).

44 Cell Stem Cell 19, 38â&#x20AC;&#x201C;51, July 7, 2016

Figure 5. Pla2g10 Is Expressed by the Paneth-like Secretory Cell Lineage in the Mouse Colon (A) CD24/cKit/SSC FACS sorting strategy employed for colonic epithelial cells from Lgr5-EGFP reporter animals (Lgr5EGFP-IRES-creERT2). Upper panel, whole colon; lower panel, GFP+ (Lgr5+) fraction. Sorting gates were defined as follows: p11 (CD24med/cKitlo), p15 (CD24med/cKitmed), p16 (CD24hi/cKithi), and p18 (CD24med/cKithi). (B) Organoid formation assay upon reconstitution of 1,000 cells from the p15/16/18 gates (after depletion of GFP-positive doublets by side scatter) with 1,000 Lgr5+ stem cells (p11) (n = 3, *p < 0.001). (C) Confocal microscopy images of doublets constituted by smaller Lgr5+ (GFP+) stem cells and larger Lgr5-negative and c-Kit+ (red) cells found in p16. (D) Heatmap representation of the results of the qRT-PCR expression analysis of sorted p16, p15, and p11 cells for Paneth- and goblet-specific genes.

Paneth- (Mmp7, Dll1, Dll4, Ephb3) and goblet-cell-specific genes (Muc2, Spdef, Spink4) (Figure 5D). Notably, another member of the sPLA2 family, namely Pla2g10 encoding the mouse group X secreted phospholipase A2 (sPLA2-X or mGX-sPLA2), was expressed at a significantly higher level in p16 than in the other populations (Figure 5D). Accordingly, previous reports indicated that sPLA2-X is expressed in the colon (Mounier et al., 2008). Pla2g10 Has Opposing Roles in Regulating Wnt Signaling and ISC Function in the Colon during Homeostasis and Inflammation We then assessed whether Pla2g10 controls stem cell function in the colon. We performed organoid assays from colonic crypts derived from Pla2g10 / knockout mice in the C57BL/6J background (therefore null for Pla2g2a and Pla2g10) (Shridas et al., 2010) and Pla2g10+/+ controls. Genetic ablation of Pla2g10 significantly increased the number of colon organoids (Figure 6A). We then assessed whether the negative effect of

Pla2g10 expression on colonic stem cell proliferation resulted from Wnt signaling inhibition, similarly to Pla2g2a in the small intestine. Transient PLA2G10 expression inhibited TOPflash Wnt reporter activity in colon cancer lines with constitutive Wnt activation (Figure 6B). Consistently, stable PLA2G10 expression in colon cancer cells significantly reduced CSC multiplicity as shown by ‘‘megacolony’’ formation (Yeung et al., 2010) (Figure S7B). Plag2g10 expression in p16 cells mediates these inhibitory effects. Pre-treating p16 cells (c-Kit+/CD24hi) with Pla2g10 siRNA before reconstituting them with Lgr5+ stem cells (p11; CD24med/cKitlo) from the colonic epithelium of Lgr5EGFP reporter mice effectively increased colon organoid formation (Figure 6C), confirming the results obtained with the Pla2g10-deficient mice. Similar to the small intestine, siRNAmediated Yap1 downregulation rescued the capacity of p16 cells from C57BL/6J mice to form colon organoids when reconstituted with Lgr5+ stem cells (Figure 6C) and suppressed the inhibitory effect of PLA2G10 expression on Wnt in human colon cancer cells (Figure 6D). Hence, under homeostatic conditions Pla2g10 negatively regulates Wnt signaling and stem cell proliferation in the colon via Yap1, analogous to Pla2g2a in the small intestine. The parallels between Pla2g2a and Pla2g10 in regulating stem cell function extend beyond homeostasis. Addition of recombinant sPLA2-X protein in organoid reconstitution assays to mimic the effects of its secretion significantly enhanced organoid formation (Figure 7A); this effect was negated by addition of soluble Pla2r1. The catalytic mGX sPLA2 mutant H48Q also enhanced organoid multiplicity (Figure 7A), indicating that the effects of sPLA2-X are mediated primarily via Pla2r1, which binds Cell Stem Cell 19, 38–51, July 7, 2016 45

Figure 6. Pla2g10, Analogous to Pla2g2a in the Small Intestine, Has Opposing Roles in Regulating Wnt and ISC Function in the Colon during Homeostasis and Inflammation (A) Colon organoid assay of C57BL/6J (Pla2g2a / /Pla2g10+/+) and Pla2g10-knockout (Pla2g2a / /Pla2g10 / ) mice (n = 3, *p < 0.001). (B) TOP-Flash reporter assay of Wnt activity in colon cancer cell lines SW480 and HCT116 upon transient transfection with a PLA2G10 expression plasmid (n = 3, *p < 0.001). (C) Organoid formation following reconstitution of colonic Lgr5+ stem cells (SCs) with Pla2g10-proficient (B6) and Pla2g10- deficient (Pla2g10KO) Paneth-like cells (p16). When indicated, p16 cells were pretreated with siRNAs against Pla2g10, Yap1, or a ‘‘scrambled’’ (SCR) control sequence (n = 3, *p < 0.05; **p < 0.001). (1) B6-p16, (2) B6p16 + B6-SCs, (3) B6-p16 [siRNA-SCR] + B6-SCs, (4) B6-p16 [siRNA-Pla2g10] + B6-SCs, (5) B6-p16 [siRNA-Yap1] + B6-SCs, (6) Pla2g10KO-p16, (7) Pla2g10KO-p16 + B6-SCs, (8) Pla2g10KO-p16 [siRNA-SCR] + B6-SCs, (9) Pla2g10KO-p16 [siRNA-Pla2g10] + B6-SCs, and (10) Pla2g10KOp16 [siRNA-Yap1] + B6-SCs. (D) TOP-Flash Wnt reporter assay in the colon cancer cell line SW480 upon transient transfection with a PLA2G10-expression plasmid and siRNAdriven YAP1 downregulation (n = 3, *p < 0.05).

sPLA2-X with high affinity in the mouse colon (Rouault et al., 2007). As observed for Pla2g2a, PGE2-driven Wnt stimulation contributes to this stimulatory effect, supported by the significant increase in prostaglandin level in the culture medium of colonic organoids from C57BL/6J mice upon addition of recombinant sPLA2-X (Figure S5C). As expected, this increase was entirely suppressed by the addition of NSAIDs (ketoprofen and diclofenac, inhibitors of cyclooxygenase activity) to the culture medium together with the recombinant Pla2g2a (Figure S5C). Accordingly, the positive effects on organoid formation upon addition of recombinant Pla2g10 were inhibited by an antibody directed against mouse Pla2r1 (Rouault et al., 2007), the two NSAIDs, and the cPLA2a inhibitor pyrrophenone (Lambeau and Gelb, 2008), thus confirming that the pathway through which the secreted Pla2g10 stimulates PGE2 synthesis is analogous to that observed for secreted Pla2g2a (Figure 7B). As it was observed for Paneth cells in the small intestine, Pla2r1 expression was exclusively found in p16 and not in Lgr5+ stem cells of the colon (Figure 4E). Moreover, knockdown of Pla2r1 in the same p16 cells, though not in Lgr5+ cells, abolishes the positive effect brought about by the recombinant Pla2g10 on organoid multiplicity in reconstitution assays (Figure S7A). Additional reconstitution assays performed with sorted p16 cells treated with PGE2 and with siRNAs directed against Cox2 confirmed the central role of the prostaglandin biosynthesis pathway in underlining the positive effect of secreted Pla2g10 on colonic stem cell function (Figure S7A). In vivo validation for these observations was provided by the significantly increased organoid for46 Cell Stem Cell 19, 38–51, July 7, 2016

mation capacity upon DSS treatment of Pla2g10 knockout mice when compared with C57BL/6J (Pla2g10+/+) (Figure 7D). Accordingly, Pla2g10 is secreted upon inflammation, as shown by TR-FIAs analysis of stool samples from DSS-treated C57BL/6J mice (Eerola et al., 2006) (Figure S5A, right). Overall, these results confirm that the previously shown opposite roles of Pla2g2a as intracellular and secreted stem cell niche factor in Paneth cells of the small intestine are conserved by the Pla2g10 gene in p16 (Paneth-like) secretory cells of the colon. Pla2g2a and Pla2g10 Are Genetic Modifiers of IBD and Colon Cancer: The ‘‘Trade-Off’’ Effect In view of the C57BL/6J (Pla2g2a / ) inbred background of the Pla2g10 / mice, a more thorough genetic analysis was needed to address the distinct roles of the two phospholipases as modifiers of intestinal bowel inflammation and colon cancer. To this aim, we first administered 3% DSS in the drinking water for a week to mice of all four Pla2 genotypes (Pla2g2a / / Pla2g10+/+ [C57BL/6J], Tg-Pla2g2a/Pla2g10+/+ [Tg-Pla2g2a], Tg-Pla2g2a/Pla2g10 / , and Pla2g2a / /Pla2g10 / ). The DAI (Table S1) (Cooper et al., 1993) and in particular the loss of weight (Figure 7C) show that genetic ablation of Pla2g10 resulted in an improved response to the DSS-induced colitis both in the presence or absence of a functional Pla2g2a allele. As also observed in the small intestine, immune-histopathologic parameters in the colon (number of colonic ulcers, presence of inflammatory cells) did not show any major difference among the four genotypes (data not shown). Supporting these data, the organoid formation capacity upon DSS treatment of Pla2g2a / /Pla2g10 / mice was significantly

Figure 7. Pla2g2a and Pla2g10 Are Genetic Modifiers of IBD and Colon Cancer (A) Results of colon organoid assay of C57BL/6J (Pla2g2a / /Pla2g10+/+) and Pla2g10-knockout (Pla2g2a / /Pla2g10 / ) mice in the presence/absence of recombinant mGX sPLA2 (both wild-type and the H48Q catalytic mutant proteins were employed), and of the recombinant (soluble) Pla2r1 receptor (n = 3, *p < 0.001; **p < 0.05). (B) Results of colon organoid assays performed on C57BL/6J mice in the presence of recombinant mGX sPLA2 alone or in combination with an antibody directed against the mouse Pla2r1 receptor (Rouault et al., 2007), pyrrophenone, a specific cPLA2a inhibitor (Lambeau and Gelb, 2008), and NSAIDs (ketoprofen and diclofenac) (n = 3, *p < 0.001). (C) Quantification of the response to DSS-induced inflammation in mice of the four Pla2g2a/Pl2g10 genotypes as measured by percentage of weight loss (compared with starting weight). See also Table S1 for more details. (D) Results of colon organoid assay of C57BL/6J (Pla2g2a / /Pla2g10+/+) and Pla2g10-knockout (Pla2g2a / /Pla2g10 / ) mice upon DSS-driven inflammation in the presence/absence of the soluble Pla2r1 receptor (n = 3, *p < 0.001). (E) Quantification of the response to DSS-induced inflammation and varespladib administration (from day 8 onward) in mice of the four Pla2g2a/Pl2g10 genotypes as measured by percentage of weight loss (compared with starting weight). Varespladib treatment of C57BL/6J animals prevented their recovery, and the mice had to be euthanized at day 10. See also Table S2 for more details. (F) Incidence of colonic tumors in Pla2g2a / /Pla2g10+/+ (C57BL/6J), Pla2g2a / /Pla2g10 /+, and Pla2g2a / /Pla2g10 / mice upon AOM-DSS treatment (n = 10, *p < 0.001).

increased when compared with C57BL/6J (Pla2g2a / / Pla2g10+/+) mice (Figure 7D). In the same set of experiments, addition of the recombinant Pla2r1 soluble receptor to the culture medium reduced organoid multiplicity from C57BL/6J mice, whereas it had, as expected, no effect on organoids from Pla2g2a / /Pla2g10 / mice (Figure 7D). This indicates that the increase in organoid numbers observed in double

knockout (DKO) mice upon inflammation cannot be replicated in DSS-treated Pla2g2a / /Pla2g10+/+ animals by scavenging the secreted group X PLA2 with the soluble Pla2r1 receptor. To further investigate the role of the secreted phospholipases in response to inflammation, we repeated the DSS treatment of Pla2g2a / /Pla2g10 / and C57BL/6J (Pla2g2a / /Pla2g10+/+) mice as described above, followed by administration of Cell Stem Cell 19, 38â&#x20AC;&#x201C;51, July 7, 2016 47

varespladib, a non-permeable inhibitor of group IIA, V, and X sPLA2 (Smart et al., 2006). As shown in Figure 7E and Table S2, loss of body weight and the overall DAI scores were significantly improved in DKO mice when compared with the C57BL/ 6J controls independently of varespladib or vector administration. However, while vehicle-treated C57BL/6J animals started recovering body weight from day 8 onward, treatment with varespladib prevented their recovery and led to severe deterioration (i.e., increased loss of weight, diarrhea, gross fecal bleeding, and lethargic state) of these mice, which had to be euthanized at day 10 (Figure 7E). Hence, group X sPLA2 inhibition in the absence of sPLA2-IIA prevents recovery from acute inflammation, whereas null mutations at both Pla2g2a and Pla2g10 genes improve the overall response to the inflammatory insult, analogous to the effect of Pla2r1 on organoid multiplicity after DSS treatment. As presented above, several reports failed to demonstrate that the human PLA2G2A gene represents a major modifier of colon cancer in man (Riggins et al., 1995; Spirio et al., 1996). In view of these reports, we asked whether the combined genetic ablation of both Pla2g2a and Pla2g10 confers susceptibility to colon cancer. To this aim, we applied the well-established AOM-DSS protocol to elicit colon tumors in Pla2g2a / /Pla2g10+/ and Pla2g2a / /Pla2g10 / mice when compared with C57BL/6J (Pla2g2a / /Pla2g10+/+) controls (Neufert et al., 2007; Tanaka et al., 2003). Notably, the C57BL/6J strain has been reported to be resistant to this protocol (Suzuki et al., 2006). Upon a single AOM injection followed by 1 week with 3% DSS in the drinking water, Pla2g2a / /Pla2g10+/ and Pla2g2a / /Pla2g10 / animals developed a significantly increased number of colonic tumors when compared with Pla2g2a / /Pla2g10+/+ littermates (Figure 7F). Histological and IHC analysis of the colonic tumors obtained in Pla2g10 / mice revealed carcinoma in situ features with b-catenin nuclear translocation (data not shown), as also shown in other AOM-DSS rodent models of colon carcinogenesis. However, no histopathological differences were observed between tumors found in Pla2g10 / and those arising in Pla2g10+/ and Pla2g10+/+ mice (data not shown). Hence, the combined deficiency of Pla2g2a and Pla2g10 increases susceptibility to colon cancer in the context of DSS-induced colitis. Hence, it appears that while complete genetic depletion of both Pla2g2a and Pla2g10 protects against the detrimental effects of the DSS-induced intestinal inflammation, the trade-off for this protective effect is the pronouncedly increased predisposition to multifocal colon cancers. DISCUSSION Collectively, the data presented in our study show that the Pla2g2a and Pla2g10 genes, encoding for the group IIA and group X sPLA2s, play regulatory roles as stem cell niche factors in both the small intestine (in Paneth cells) and in the large bowel (in secretory Paneth-like cells of the colon). Notably, their stem cell regulatory functions are context dependent: both inhibit Wnt signaling from within the intracellular compartment during homeostasis, thus controlling maturation and the stem-cell-supporting function of these secretory lineages. Upon inflammation, however, they are secreted into the intestinal lumen and stimulate Wnt in autocrine fashion through PGE2 biosynthesis, thus contributing to the regenerative response 48 Cell Stem Cell 19, 38â&#x20AC;&#x201C;51, July 7, 2016

(Zhang et al., 2015) (see Graphical Abstract). It should be noted that although Wnt has been the major object of our investigations, it is likely that other signaling pathways play important roles downstream of the A2 phospholipases. For example, the observed cell-cell defect (see Movie S1) is likely to result from the decreased expression of the Notch ligands Dll1 and Dll4 in Paneth cells from Tg-Pla2g2a mice (Figure 1E) as it has been shown that Lgr5+ CBCs require Notch signals, which are governed by direct cell-cell contacts (Fre et al., 2005; van Es et al., 2005b). Based on the above, the human PLA2G2A and PLA2G10 genes and other members of the here-identified pathways are likely to represent important genetic modifiers of inflammatory bowel disease (IBD) and colon cancer in man. In a recently published GWAS association study of IBD (Liu et al., 2015), 38 risk loci were identified, including three genes, namely PTGS2 (COX2), PLA2G4A, and PLA2R1, which are integral members of the prostaglandin biosynthesis pathway through which sPLA2s positively affect ISC function during inflammation. In the context of inflammation, group IIA and X sPLA2 are secreted into the intestinal lumen, where they promote PGE2 synthesis through interaction with the Pla2r1 receptor, activation of cPLA2a (Pla2g4a), and AA synthesis from membrane phospholipids. AA represents the main substrate of Cox2 for production of PGE2. As previously shown and here confirmed, prostaglandin exerts stimulatory effects on Wnt signaling (Castellone et al., 2005; Goessling et al., 2009; Shao et al., 2005). Accordingly, inhibition of the prostaglandin-degrading enzyme 15-PGDH (15-hydroxyprostaglandin dehydrogenase) increases regenerative capacity in a broad spectrum of tissues, including bone marrow and the colon (Zhang et al., 2015). Based on the model arising from the above data, expression of the group IIA and X secreted phospholipases is expected to improve the overall response to inflammation, whereas null or hypomorphic alleles at these genes are predicted to increase susceptibility to the detrimental effects of IBD. This was in part confirmed in the TgPla2g2a mice compared to the Pla2g2a-deficient C57BL/6J inbred strain. In the presence of a functional Pla2g10 gene, Pla2g2a secretion upon DSS-treatment improves stem cell function and the overall response to the deleterious effects of inflammation (Table S1). Surprisingly, the most improved response to DSS-induced inflammation was observed in Pla2g2a / /Pla2g10 / mice, i.e., in the absence of the secreted group IIA and X phospholipases (Figure 7C; Table S1). Two distinct results obtained here may point to a likely explanation for this apparently contradictory observation. First, addition of the Pla2r1 soluble receptor to the culture medium negatively affected organoid multiplicity from DSS-treated C57BL/6J mice (Figure 7D). This shows that the high organoid multiplicity observed in DKO mice upon inflammation cannot be replicated by scavenging the secreted Pla2g10 in DSS-treated C57BL/6J (Pla2g2a / /Pla2g10+/+) mice. This ex vivo result was then confirmed in vivo by administrating varespladib, a nonpermeable inhibitor of both Pla2g2a and Pla2g10 (Smart et al., 2006), to C57BL/6J mice after DSS administration. Whereas vehicle-treated C57BL/6J animals started recovering only a few days after DSS treatment, varespladib administration prevented recovery and severely compromised the overall

health status of these animals (Figure 7E; Table S2). Hence, scavenging of group X sPLA2 (in the absence of sPLA2-IIA) fails to reproduce the improved response to the inflammatory insult observed in DKO mice. This suggests that the underlying cause of the protective effect to the detrimental effects of DSS observed in DKO mice may reside in the functional role that both phospholipases play in the ISC niche during homeostasis when they exert a regulatory function on niche cells by inhibiting Wnt. Genetic ablation of Pla2g10 and Pla2g2a relieves this intracellular inhibitory effect, thus enhancing Wnt and positively affecting ISC function and tissue regeneration. Accordingly, the expression of known Wnt targets in sorted p16 (Paneth-like) and total colon cells from mice of all four genotypes was assayed. As shown in Figure S7C, during homeostasis specific Wnt targets (i.e., Axin2, Jun, Klf4, Efnb2, Sox9) are upregulated in Paneth-like cells of the colon in a gradient from Pla2g10/ Pla2g2a-proficient mice to the DKO animals. This effect was not seen when whole crypts were analyzed (Figure S7C). The observed increase in Wnt activity upon genetic ablation of both genes is also reflected by the highest organoid formation efficiency obtained from the colon of DKO mice both in homeostasis (Figure 6A) and upon inflammation (Figure 7D). These data support the upregulation of Wnt signaling in niche cells of the colon upon Pla2g10 and Pla2g2a genetic ablation and provide a plausible explanation for their improved response to acute inflammation. Among the Wnt targets found to be upregulated in colonic Paneth-like cells of DKO mice, Sox9 is noteworthy as it encodes for an established transcription factor in Paneth cell differentiation (Bastide et al., 2007; Mori-Akiyama et al., 2007). Moreover, Sox9 expression was also associated with the appearance of ectopic Paneth cells in the colon upon inducible Apc loss and constitutive Wnt activation (Feng et al., 2013). Hence, SOX9 upregulation is likely to underlie Paneth cell metaplasia in IBD patients with improved response to the inflammatory insult (Paterson and Watson, 1961). Increased Wnt signaling in colonic p16 cells triggers Sox9 upregulation and the consequent maturation toward the Paneth cell lineage. Metaplastic Paneth cells provide additional niche support and thereby improve the regenerative response to inflammation. Whereas the observed Wnt activation, elicited by the genetic ablation of both Pla2g2a and Pla2g10, can explain the improved regenerative response to the detrimental effects of inflammation, it is also likely to enhance susceptibility to colon cancer. Indeed, Pla2g2a / /Pla2g10 / mice revealed a striking multifocal colon cancer phenotype upon AOM/DSS treatment. To assess whether the increased colon cancer susceptibility of the DKO mice was dependent on inflammation, we bred them with Apc1638N/+, a model characterized by a mild upper GI tumor phenotype but no colon tumors (Fodde et al., 1994). Notwithstanding the limited number of animals here analyzed (Apc1638N/+/Pla2g2a / /Pla2g10 /+, n = 5, 28 weeks old; Apc1638N/+/Pla2g2a / /Pla2g10 / , n = 3, 22 weeks old; Apc1638N/+/Pla2g2a / /Pla2g10+/+, n = 3, 28 weeks old), heteroand homozygous Pla2g10 mutant mice were shown to develop a significantly increased multiplicity of aberrant crypt foci (average per mouse was 1.8 and 4.0, respectively, versus 0.7 in controls) and tumors (average per mouse was 1.6 and 3.7, respectively, versus 0 in controls) in the colon. Hence, the predisposition to

colon cancer conferred by Pla2g10 seems to be independent of inflammation. Overall, although hypomorphic alleles at the PLA2G2A and PLA2G10 genes and at other loci encoding for members of the same biochemical pathways may protect the individual from the detrimental effects of inflammation of the bowel, the tradeoff of this initial beneficial effect may lie in an increased susceptibility to colon cancer. In the near future, it would be of interest to establish whether Paneth cell metaplasia in IBD patients is associated with improved response to inflammation in combination with increased cancer risk. In conclusion, the observed ‘‘yin-yang’’ mode of action of two sPLA2s in intestinal homeostasis and inflammation can have profound consequences for an individual’s response and susceptibility to inflammation and to both sporadic (i.e., non-inflammatory) and IBD-related colon cancer. Finally, in view of the abundance of sPLA2s in several mammalian tissues, it will be of interest to assess whether their newly identified role as key ISC niche factors in health and disease holds true for a broader spectrum of organs. EXPERIMENTAL PROCEDURES Organoid Assays To form organoids ex vivo, two distinct sources of cells were employed, namely whole crypts and single-cell suspensions from small intestine or colon. Whole crypt cultures were performed as previously described (Sato et al., 2009). All organoid assays were performed in several biological replicates (i.e., with independent animals of the same genotype and/or experimental conditions) defined by ‘‘n,’’ whereas three technical replicates were implemented for each biological replicate. Organoids were counted at day 5 in most of the experiments unless stated otherwise. AOM/DSS-Induced Colon Carcinogenesis To induce colon cancer in the context of inflammation, a modification of the well-established AOM/DSS protocol was employed (Neufert et al., 2007). In short, mice were administered a single azoxymethane (AOM) injection (intraperitoneal; 10 mg/kg), followed by 1 week with the inflammatory agent DSS (MP Biomedicals) in the drinking water (2.5%). Mice were then euthanized 16 weeks after AOM injection or at the first appearance of signs of discomfort. All procedures were performed in agreement with local animal welfare laws and guidelines. DAI Scores and Varespladib Treatment To establish DAI scores upon DSS-induced inflammation, we employed the method originally described by Cooper et al. (1993), with minor modifications. Mice were administered 3% DSS in the drinking water for a week, during which they were monitored for a number of parameters including stool consistency, blood loss, appearance, and percent weight loss. For the treatment with varespladib, mice were first administered 3% DSS for 1 week, followed by varespladib treatment at day 8 in the drinking water (60 mM). The vehicle in which the varespladib is dissolved is 5% DMSO, 5% ETOH, and 30% PEG 300 in water. sPLA2 Quantification Time-resolved fluoroimmunoassays (TR-FIAs) for Pla2g2a were performed as described previously with minor modifications (Eerola et al., 2006). More comprehensive Experimental Procedures can be found in the Supplemental Experimental Procedures. SUPPLEMENTAL INFORMATION Supplemental Information includes Supplemental Experimental Procedures, seven figures, two tables, and two movies and can be found with this article online at http://dx.doi.org/10.1016/j.stem.2016.05.023.

Cell Stem Cell 19, 38–51, July 7, 2016 49

AUTHOR CONTRIBUTIONS Experimental contributions were as follows: organoid assays (M. Schewe and P.F.F.); qRT-PCR analysis (M. Schewe and R.B.); siRNA and TOP-Flash assays (M. Schewe); colon cancer cell line culture and transfections (M. Schewe); immunohistochemistry (P.F.F.); FACS analysis and sorting (A.S.); western blot analysis (M. Schmitt); mouse colony management, AOM/DSS treatments, and phenotypic analysis (M. Schewe and R.J.); transgenic mice (R.T.C., P.M.S., M.G., C.P., G.L., and N.R.W.); recombinant proteins (C.P., M.G., and G.L.); ELISA assays (L.J. and C.P.); and live imaging of organoid formation (M.E.vR. and M. Schewe). The project was conceived by R.F. and M. Schewe with substantial contributions from R.T.C. and G.L. The manuscript was written by R.F. with substantial contributions from M. Schewe, G.L., and R.T.C. All authors contributed to the final version of the manuscript. ACKNOWLEDGMENTS The authors are grateful to Mr. Frank van der Panne for his help with the artwork, and to Dr. Len Augenlicht, Dr. Hans Clevers, and Dr. Pekka Katajisto for critical reading of the manuscript. This study was made possible by funding from the Dutch Cancer Society (KWF; EMCR 2012-5473) and from the Netherlands Institute of Regenerative Medicine (NIRM; http://www. regeneratieve-geneeskunde.nl/) to R.F., and from CNRS and the Fondation ARC pour la Recherche sur le Cancer to G.L. Received: August 31, 2015 Revised: February 10, 2016 Accepted: May 19, 2016 Published: June 9, 2016 REFERENCES Azzolin, L., Panciera, T., Soligo, S., Enzo, E., Bicciato, S., Dupont, S., Bresolin, S., Frasson, C., Basso, G., Guzzardo, V., et al. (2014). YAP/TAZ incorporation in the b-catenin destruction complex orchestrates the Wnt response. Cell 158, 157–170. Barker, N., van Es, J.H., Kuipers, J., Kujala, P., van den Born, M., Cozijnsen, M., Haegebarth, A., Korving, J., Begthel, H., Peters, P.J., and Clevers, H. (2007). Identification of stem cells in small intestine and colon by marker gene Lgr5. Nature 449, 1003–1007. Barry, E.R., Morikawa, T., Butler, B.L., Shrestha, K., de la Rosa, R., Yan, K.S., Fuchs, C.S., Magness, S.T., Smits, R., Ogino, S., et al. (2013). Restriction of intestinal stem cell expansion and the regenerative response by YAP. Nature 493, 106–110. Bastide, P., Darido, C., Pannequin, J., Kist, R., Robine, S., Marty-Double, C., Bibeau, F., Scherer, G., Joubert, D., Hollande, F., et al. (2007). Sox9 regulates cell proliferation and is required for Paneth cell differentiation in the intestinal epithelium. J. Cell Biol. 178, 635–648. Bonventre, J.V., Huang, Z., Taheri, M.R., O’Leary, E., Li, E., Moskowitz, M.A., and Sapirstein, A. (1997). Reduced fertility and postischaemic brain injury in mice deficient in cytosolic phospholipase A2. Nature 390, 622–625. Buczacki, S.J., Zecchini, H.I., Nicholson, A.M., Russell, R., Vermeulen, L., Kemp, R., and Winton, D.J. (2013). Intestinal label-retaining cells are secretory precursors expressing Lgr5. Nature 495, 65–69. Camargo, F.D., Gokhale, S., Johnnidis, J.B., Fu, D., Bell, G.W., Jaenisch, R., and Brummelkamp, T.R. (2007). YAP1 increases organ size and expands undifferentiated progenitor cells. Curr. Biol. 17, 2054–2060. Castellone, M.D., Teramoto, H., Williams, B.O., Druey, K.M., and Gutkind, J.S. (2005). Prostaglandin E2 promotes colon cancer cell growth through a Gs-axin-beta-catenin signaling axis. Science 310, 1504–1510. Cooper, H.S., Murthy, S.N., Shah, R.S., and Sedergran, D.J. (1993). Clinicopathologic study of dextran sulfate sodium experimental murine colitis. Lab. Invest. 69, 238–249. Cormier, R.T., Hong, K.H., Halberg, R.B., Hawkins, T.L., Richardson, P., Mulherkar, R., Dove, W.F., and Lander, E.S. (1997). Secretory phospholipase Pla2g2a confers resistance to intestinal tumorigenesis. Nat. Genet. 17, 88–91.

50 Cell Stem Cell 19, 38–51, July 7, 2016

Cormier, R.T., Bilger, A., Lillich, A.J., Halberg, R.B., Hong, K.H., Gould, K.A., Borenstein, N., Lander, E.S., and Dove, W.F. (2000). The Mom1AKR intestinal tumor resistance region consists of Pla2g2a and a locus distal to D4Mit64. Oncogene 19, 3182–3192. Dietrich, W.F., Lander, E.S., Smith, J.S., Moser, A.R., Gould, K.A., Luongo, C., Borenstein, N., and Dove, W. (1993). Genetic identification of Mom-1, a major modifier locus affecting Min-induced intestinal neoplasia in the mouse. Cell 75, 631–639. Eerola, L.I., Surrel, F., Nevalainen, T.J., Gelb, M.H., Lambeau, G., and Laine, V.J. (2006). Analysis of expression of secreted phospholipases A2 in mouse tissues at protein and mRNA levels. Biochim. Biophys. Acta 1761, 745–756. Farin, H.F., Karthaus, W.R., Kujala, P., Rakhshandehroo, M., Schwank, G., Vries, R.G., Kalkhoven, E., Nieuwenhuis, E.E., and Clevers, H. (2014). Paneth cell extrusion and release of antimicrobial products is directly controlled by immune cell-derived IFN-g. J. Exp. Med. 211, 1393–1405. Feng, Y., Sentani, K., Wiese, A., Sands, E., Green, M., Bommer, G.T., Cho, K.R., and Fearon, E.R. (2013). Sox9 induction, ectopic Paneth cells, and mitotic spindle axis defects in mouse colon adenomatous epithelium arising from conditional biallelic Apc inactivation. Am. J. Pathol. 183, 493–503. Fijneman, R.J., Peham, J.R., van de Wiel, M.A., Meijer, G.A., Matise, I., Velcich, A., and Cormier, R.T. (2008). Expression of Pla2g2a prevents carcinogenesis in Muc2-deficient mice. Cancer Sci. 99, 2113–2119. Fijneman, R.J., Bade, L.K., Peham, J.R., van de Wiel, M.A., van Hinsbergh, V.W., Meijer, G.A., O’Sullivan, M.G., and Cormier, R.T. (2009). Pla2g2a attenuates colon tumorigenesis in azoxymethane-treated C57BL/6 mice; expression studies reveal Pla2g2a target genes and pathways. Cell. Oncol. 31, 345–356. Fodde, R., Edelmann, W., Yang, K., van Leeuwen, C., Carlson, C., Renault, B., Breukel, C., Alt, E., Lipkin, M., Khan, P.M., et al. (1994). A targeted chain-termination mutation in the mouse Apc gene results in multiple intestinal tumors. Proc. Natl. Acad. Sci. USA 91, 8969–8973. Fodde, R., Smits, R., and Clevers, H. (2001). APC, signal transduction and genetic instability in colorectal cancer. Nat. Rev. Cancer 1, 55–67. Fre, S., Huyghe, M., Mourikis, P., Robine, S., Louvard, D., and ArtavanisTsakonas, S. (2005). Notch signals control the fate of immature progenitor cells in the intestine. Nature 435, 964–968. Goessling, W., North, T.E., Loewer, S., Lord, A.M., Lee, S., Stoick-Cooper, C.L., Weidinger, G., Puder, M., Daley, G.Q., Moon, R.T., and Zon, L.I. (2009). Genetic interaction of PGE2 and Wnt signaling regulates developmental specification of stem cells and regeneration. Cell 136, 1136–1147. Lambeau, G., and Gelb, M.H. (2008). Biochemistry and physiology of mammalian secreted phospholipases A2. Annu. Rev. Biochem. 77, 495–520. Liu, J.Z., van Sommeren, S., Huang, H., Ng, S.C., Alberts, R., Takahashi, A., Ripke, S., Lee, J.C., Jostins, L., Shah, T., et al.; International Multiple Sclerosis Genetics Consortium; International IBD Genetics Consortium (2015). Association analyses identify 38 susceptibility loci for inflammatory bowel disease and highlight shared genetic risk across populations. Nat. Genet. 47, 979–986. MacPhee, M., Chepenik, K.P., Liddell, R.A., Nelson, K.K., Siracusa, L.D., and Buchberg, A.M. (1995). The secretory phospholipase A2 gene is a candidate for the Mom1 locus, a major modifier of ApcMin-induced intestinal neoplasia. Cell 81, 957–966. Mori-Akiyama, Y., van den Born, M., van Es, J.H., Hamilton, S.R., Adams, H.P., Zhang, J., Clevers, H., and de Crombrugghe, B. (2007). SOX9 is required for the differentiation of paneth cells in the intestinal epithelium. Gastroenterology 133, 539–546. Mounier, C.M., Wendum, D., Greenspan, E., Fléjou, J.F., Rosenberg, D.W., and Lambeau, G. (2008). Distinct expression pattern of the full set of secreted phospholipases A2 in human colorectal adenocarcinomas: sPLA2-III as a biomarker candidate. Br. J. Cancer 98, 587–595. Mustata, R.C., Vasile, G., Fernandez-Vallone, V., Strollo, S., Lefort, A., Libert, F., Monteyne, D., Pérez-Morga, D., Vassart, G., and Garcia, M.I. (2013). Identification of Lgr5-independent spheroid-generating progenitors of the mouse fetal intestinal epithelium. Cell Rep. 5, 421–432.

Neufert, C., Becker, C., and Neurath, M.F. (2007). An inducible mouse model of colon carcinogenesis for the analysis of sporadic and inflammation-driven tumor progression. Nat. Protoc. 2, 1998–2004. Paterson, J.C., and Watson, S.H. (1961). Paneth cell metaplasia in ulcerative colitis. Am. J. Pathol. 38, 243–249. Riggins, G.J., Markowitz, S., Wilson, J.K., Vogelstein, B., and Kinzler, K.W. (1995). Absence of secretory phospholipase A2 gene alterations in human colorectal cancer. Cancer Res. 55, 5184–5186. Roth, S., Franken, P., Sacchetti, A., Kremer, A., Anderson, K., Sansom, O., and Fodde, R. (2012). Paneth cells in intestinal homeostasis and tissue injury. PLoS ONE 7, e38965. Rothenberg, M.E., Nusse, Y., Kalisky, T., Lee, J.J., Dalerba, P., Scheeren, F., Lobo, N., Kulkarni, S., Sim, S., Qian, D., et al. (2012). Identification of a cKit(+) colonic crypt base secretory cell that supports Lgr5(+) stem cells in mice. Gastroenterology 142, 1195–1205.e6. Rouault, M., Le Calvez, C., Boilard, E., Surrel, F., Singer, A., Ghomashchi, F., Bezzine, S., Scarzello, S., Bollinger, J., Gelb, M.H., and Lambeau, G. (2007). Recombinant production and properties of binding of the full set of mouse secreted phospholipases A2 to the mouse M-type receptor. Biochemistry 46, 1647–1662. Sato, T., Vries, R.G., Snippert, H.J., van de Wetering, M., Barker, N., Stange, D.E., van Es, J.H., Abo, A., Kujala, P., Peters, P.J., and Clevers, H. (2009). Single Lgr5 stem cells build crypt-villus structures in vitro without a mesenchymal niche. Nature 459, 262–265. Sato, T., van Es, J.H., Snippert, H.J., Stange, D.E., Vries, R.G., van den Born, M., Barker, N., Shroyer, N.F., van de Wetering, M., and Clevers, H. (2011). Paneth cells constitute the niche for Lgr5 stem cells in intestinal crypts. Nature 469, 415–418. Shao, J., Jung, C., Liu, C., and Sheng, H. (2005). Prostaglandin E2 Stimulates the beta-catenin/T cell factor-dependent transcription in colon cancer. J. Biol. Chem. 280, 26565–26572. Shridas, P., Bailey, W.M., Boyanovsky, B.B., Oslund, R.C., Gelb, M.H., and Webb, N.R. (2010). Group X secretory phospholipase A2 regulates the expression of steroidogenic acute regulatory protein (StAR) in mouse adrenal glands. J. Biol. Chem. 285, 20031–20039. Smart, B.P., Oslund, R.C., Walsh, L.A., and Gelb, M.H. (2006). The first potent inhibitor of mammalian group X secreted phospholipase A2: elucidation of sites for enhanced binding. J. Med. Chem. 49, 2858–2860.

Spirio, L.N., Kutchera, W., Winstead, M.V., Pearson, B., Kaplan, C., Robertson, M., Lawrence, E., Burt, R.W., Tischfield, J.A., Leppert, M.F., et al. (1996). Three secretory phospholipase A(2) genes that map to human chromosome 1P35-36 are not mutated in individuals with attenuated adenomatous polyposis coli. Cancer Res. 56, 955–958. Stelzner, M., Helmrath, M., Dunn, J.C., Henning, S.J., Houchen, C.W., Kuo, C., Lynch, J., Li, L., Magness, S.T., Martin, M.G., et al.; NIH Intestinal Stem Cell Consortium (2012). A nomenclature for intestinal in vitro cultures. Am. J. Physiol. Gastrointest. Liver Physiol. 302, G1359–G1363. Suzuki, R., Kohno, H., Sugie, S., Nakagama, H., and Tanaka, T. (2006). Strain differences in the susceptibility to azoxymethane and dextran sodium sulfateinduced colon carcinogenesis in mice. Carcinogenesis 27, 162–169. Tanaka, T., Kohno, H., Suzuki, R., Yamada, Y., Sugie, S., and Mori, H. (2003). A novel inflammation-related mouse colon carcinogenesis model induced by azoxymethane and dextran sodium sulfate. Cancer Sci. 94, 965–973. van Es, J.H., Jay, P., Gregorieff, A., van Gijn, M.E., Jonkheer, S., Hatzis, P., Thiele, A., van den Born, M., Begthel, H., Brabletz, T., et al. (2005a). Wnt signalling induces maturation of Paneth cells in intestinal crypts. Nat. Cell Biol. 7, 381–386. van Es, J.H., van Gijn, M.E., Riccio, O., van den Born, M., Vooijs, M., Begthel, H., Cozijnsen, M., Robine, S., Winton, D.J., Radtke, F., and Clevers, H. (2005b). Notch/gamma-secretase inhibition turns proliferative cells in intestinal crypts and adenomas into goblet cells. Nature 435, 959–963. Wirtz, S., Neufert, C., Weigmann, B., and Neurath, M.F. (2007). Chemically induced mouse models of intestinal inflammation. Nat. Protoc. 2, 541–546. Yazbeck, R., Howarth, G.S., Butler, R.N., Geier, M.S., and Abbott, C.A. (2011). Biochemical and histological changes in the small intestine of mice with dextran sulfate sodium colitis. J. Cell. Physiol. 226, 3219–3224. Yeung, T.M., Gandhi, S.C., Wilding, J.L., Muschel, R., and Bodmer, W.F. (2010). Cancer stem cells from colorectal cancer-derived cell lines. Proc. Natl. Acad. Sci. USA 107, 3722–3727. Zhang, Y., Desai, A., Yang, S.Y., Bae, K.B., Antczak, M.I., Fink, S.P., Tiwari, S., Willis, J.E., Williams, N.S., Dawson, D.M., et al. (2015). Tissue regeneration. Inhibition of the prostaglandin-degrading enzyme 15-PGDH potentiates tissue regeneration. Science 348, aaa2340.

Cell Stem Cell 19, 38–51, July 7, 2016 51

Resource Organoid Models of Human and Mouse Ductal Pancreatic Cancer Sylvia F. Boj,1,2,14 Chang-Il Hwang,3,4,14 Lindsey A. Baker,3,4,14 Iok In Christine Chio,3,4,14 Dannielle D. Engle,3,4,14 Vincenzo Corbo,3,4,14 Myrthe Jager,1,14 Mariano Ponz-Sarvise,3,4 Hervé Tiriac,3,4 Mona S. Spector,3,4 Ana Gracanin,1,2 Tobiloba Oni,3,4,5 Kenneth H. Yu,3,4,6,7 Ruben van Boxtel,1 Meritxell Huch,1,15 Keith D. Rivera,3 John P. Wilson,3 Michael E. Feigin,3,4 Daniel Öhlund,3,4 Abram Handly-Santana,4,8 Christine M. Ardito-Abraham,3,4 Michael Ludwig,3,4 Ela Elyada,3,4 Brinda Alagesan,3,4,9 Giulia Biffi,3,4 Georgi N. Yordanov,4,8 Bethany Delcuze,3,4 Brianna Creighton,3,4 Kevin Wright,3,4 Youngkyu Park,3,4 Folkert H.M. Morsink,10 I. Quintus Molenaar,11 Inne H. Borel Rinkes,11 Edwin Cuppen,1 Yuan Hao,3 Ying Jin,3 Isaac J. Nijman,1 Christine Iacobuzio-Donahue,6 Steven D. Leach,6 Darryl J. Pappin,3 Molly Hammell,3 David S. Klimstra,12 Olca Basturk,12 Ralph H. Hruban,13 George Johan Offerhaus,10 Robert G.J. Vries,1,2 Hans Clevers,1,* and David A. Tuveson3,4,6,* 1Hubrecht Institute, Royal Netherlands Academy of Arts and Sciences (KNAW), University Medical Centre Utrecht and CancerGenomics.nl, 3584 CT Utrecht, the Netherlands 2foundation Hubrecht Organoid Technology (HUB), 3584 CT Utrecht, the Netherlands 3Cold Spring Harbor Laboratory, Cold Spring Harbor, NY 11724, USA 4Lustgarten Foundation Pancreatic Cancer Research Laboratory, Cold Spring Harbor, NY 11724, USA 5Graduate Program in Molecular and Cellular Biology, Stony Brook University, Stony Brook, NY 11794, USA 6Rubenstein Center for Pancreatic Cancer Research, Memorial Sloan Kettering Cancer Center, New York, NY 10065, USA 7Weill Medical College at Cornell University, New York, NY 10065, USA 8Watson School of Biological Sciences, Cold Spring Harbor Laboratory, Cold Spring Harbor, NY 11724, USA 9Graduate Program in Genetics, Stony Brook University, Stony Brook, NY 11794, USA 10Department of Pathology, University Medical Centre Utrecht, 3584 CX Utrecht, the Netherlands 11Department of Surgery, University Medical Center Utrecht, 3584 CX Utrecht, the Netherlands 12Department of Pathology, Memorial Sloan Kettering Cancer Center, New York, NY 10065, USA 13The Sol Goldman Pancreatic Cancer Research Center, Johns Hopkins University School of Medicine, Baltimore, MD 21231, USA 14Co-first author 15Present address: Gurdon Institute-University of Cambridge, Tennis Court Road, Cambridge CB2 1QN, UK *Correspondence: h.clevers@hubrecht.eu (H.C.), dtuveson@cshl.edu (D.A.T.) http://dx.doi.org/10.1016/j.cell.2014.12.021

SUMMARY

Pancreatic cancer is one of the most lethal malignancies due to its late diagnosis and limited response to treatment. Tractable methods to identify and interrogate pathways involved in pancreatic tumorigenesis are urgently needed. We established organoid models from normal and neoplastic murine and human pancreas tissues. Pancreatic organoids can be rapidly generated from resected tumors and biopsies, survive cryopreservation, and exhibit ductal- and disease-stage-specific characteristics. Orthotopically transplanted neoplastic organoids recapitulate the full spectrum of tumor development by forming early-grade neoplasms that progress to locally invasive and metastatic carcinomas. Due to their ability to be genetically manipulated, organoids are a platform to probe genetic cooperation. Comprehensive transcriptional and proteomic analyses of murine pancreatic organoids revealed genes and pathways altered during disease progression. The confirmation of many of these protein changes in human tissues demonstrates that organoids are a facile model 324 Cell 160, 324–338, January 15, 2015 ª2015 Elsevier Inc.

system to discover characteristics of this deadly malignancy. INTRODUCTION Mortality due to pancreatic cancer is projected to surpass that of breast and colorectal cancer by 2030 in the United States (Rahib et al., 2014; Siegel et al., 2013). This dire scenario reflects an aging population, the improvement of outcomes for breast and colorectal cancer patients, the advanced stage at which most patients with pancreatic cancer are diagnosed, and the lack of durable treatment responses in pancreatic cancer patients. Indeed, effective therapeutic strategies for patients with pancreatic ductal adenocarcinoma (PDA) have been difficult to identify (Abbruzzese and Hess, 2014). The therapeutic resistance of PDA has been explored in a variety of cell culture and animal model systems, with clinically actionable findings encountered only occasionally (Villarroel et al., 2011). Patient-derived xenografts (PDXs) have yielded insights into PDA, but their generation requires a large amount of tissue, and they take multiple months to establish (Kim et al., 2009; Rubio-Viqueira et al., 2006). Genetically engineered mouse models (GEMMs) of PDA have also been generated as a parallel system for fundamental biological investigation and preclinical studies (Pérez-Mancera et al., 2012). These GEMMs

accurately mimic the pathophysiological features of human PDA, including disease initiation from preinvasive pancreatic intraepithelial neoplasms (PanINs) (Hingorani et al., 2003; Pérez-Mancera et al., 2012) and were used to discover that PDA possesses a deficient vasculature that impairs drug delivery (Erkan et al., 2009; Jacobetz et al., 2013; Koong et al., 2000; Olive et al., 2009; Provenzano et al., 2012). Although GEMMs have informed PDA therapeutic development (Beatty et al., 2011; Frese et al., 2012; Neesse et al., 2014), they are expensive and time consuming (Pérez-Mancera et al., 2012). In addition, both human PDA and GEMMs exhibit an extensive stromal component that decreases the neoplastic cellularity, making it difficult to isolate and characterize the epithelium-derived malignant cells in pancreatic neoplastic tissues. To study neoplastic cells, dissociated human tumors are often grown in two-dimensional (2D) culture conditions (Sharma et al., 2010), which do not support growth of untransformed, nonneoplastic pancreatic cells. Three-dimensional (3D) culture strategies have been developed to study normal, untransformed cells but so far have only allowed minimal propagation (Agbunag and Bar-Sagi, 2004; Lee et al., 2013; Means et al., 2005; Rovira et al., 2010; Seaberg et al., 2004). A comprehensive 3D cell culture model of murine and human PDA progression would facilitate investigation of genetic drivers, therapeutic targets, and diagnostics for PDA. To address this deficiency, we sought to generate normal and neoplastic pancreatic organoids by modifying approaches we previously pioneered to culture intestinal (Sato et al., 2009), gastric (Barker et al., 2010), colon carcinoma (Sato et al., 2011), hepatic (Huch et al., 2013b), pancreatic (Huch et al., 2013a), and prostatic organoids (Gao et al., 2014; Karthaus et al., 2014). We developed 3D organoids from normal and malignant murine pancreatic tissues and used this model system to investigate PDA pathogenesis. Pancreatic organoids derived from wild-type mice and PDA GEMMs accurately recapitulate physiologically relevant aspects of disease progression in vitro. Following orthotopic transplantation, organoids from wild-type mouse normal pancreata are capable of regenerating normal ductal architecture, unlike other 3D model systems. We further developed methods to generate pancreatic organoids from normal and diseased human tissues, as well as from endoscopic needle biopsies. Following transplantation, organoids derived from murine and human PDA generate lesions reminiscent of PanIN and progress to invasive PDA. Finally, we demonstrate the utility of organoids to identify molecular pathways that correlate with disease progression and that represent therapeutic and diagnostic opportunities. RESULTS Murine Pancreatic Ductal Organoids Expressing Oncogenic Kras Recapitulate Features of PanINs Recently, we derived continuously proliferating, normal pancreatic organoids from adult murine ductal cells (Huch et al., 2013a). We optimized this approach to generate models of PDA progression. We manually isolated small intralobular ducts and established organoid cultures from C57Bl/6 mouse normal pancreata and pancreatic tissues that contained low-grade murine PanIN

(mPanIN-1a/b) from Kras+/LSL-G12D; Pdx1-Cre (‘‘KC’’) mice (Figure 1A). KC mice develop a spectrum of preinvasive ductal lesions that mirror human PanINs and, upon aging, stochastically develop primary and metastatic PDA (Hingorani et al., 2003). Ducts from KC pancreata were often larger and exhibited higher grades of dysplasia compared to those from wild-type mice (Figure 1A). After 1–3 days in culture, organoid growth was observed from isolated ducts (Figure 1A). We created a collection of 10 murine normal (mN) and 9 PanIN (mP) organoid cultures that we have continuously propagated for over 20 passages and successfully cryopreserved (Table S1A available online). mP organoids exhibited recombination of the conditional KrasLSL-G12D allele and higher levels of Kras-GTP when compared to mN organoids (Figure 1B). To determine the contribution of different pancreatic lineages to the organoids, we evaluated the expression of pancreatic lineage markers in these cultures. Genes associated with the ductal lineage (Ck19 and Sox9) (Cleveland et al., 2012) were enriched in the mN and mP organoids compared to total pancreatic tissues, which contain relatively few ductal cells (Figure 1C). In addition, the mP organoids upregulated genes indicative of a PanIN disease state (Muc5ac, Muc6, and Tff1) relative to mN, with no difference in Klf4 (Figure 1D) (Prasad et al., 2005). GFP-transduced mN and mP organoids were orthotopically transplanted into syngeneic C57Bl/6 or Nu/Nu mice. mN organoids quickly formed ductal structures comprised of simple cuboidal cells that persisted for up to 1 month (n = 9/27 transplants) but were not observed after 2 months (n = 0/13 transplants) (Figure 1E and Table S1B). In comparison, mP organoids formed small cysts lined with a single layer of simple cuboidal ductal cells interspersed with mucin-containing columnar epithelial cells. Although we could not demonstrate that the mP transplants were contiguous with the native ductal system, they resembled preinvasive mPanIN (Figure S1C). These dysplastic epithelial cells persisted for 2 months or longer (n = 16/18 transplants), were GFP and Ck19 positive, expressed the mPanIN-associated mucin Muc5ac, and stained prominently with Alcian blue (Figure 1E and Table S1C). In addition, when compared to mN transplants, mP transplants had increased proliferation and a robust stromal response, which are characteristics of autochthonous mPanIN tissue (Figures S1A–S1C). The ability of transplanted mP organoids to form lesions with many of the features of mPanINs demonstrates the utility of this system as a model for early pancreatic neoplasia. Multiple cellular origins have been proposed for the development of PDA, with the pancreatic acinar cell hypothesized to be a major contributor to PDA initiation (De La O et al., 2008; Gidekel Friedlander et al., 2009; Guerra et al., 2003; Habbe et al., 2008; Kopp et al., 2012; Morris et al., 2010; Sawey et al., 2007). However, recent studies have suggested that transformation of pancreatic ductal cells can also give rise to PDA (Pylayeva-Gupta et al., 2012; Ray et al., 2011; von Figura et al., 2014). Acinar cells isolated from wild-type pancreata are unable to form organoids in our conditions (Huch et al., 2013a). Therefore, our pancreatic ductal organoid system offers a unique opportunity to determine whether ductal cells can give rise to mPanIN. To assess whether expression of oncogenic Kras in pancreatic ductal organoids is sufficient to induce mPanIN formation in vivo, we derived Cell 160, 324–338, January 15, 2015 ª2015 Elsevier Inc. 325

Figure 1. Oncogenic KrasG12D Expression in Pancreatic Ductal Organoids Is Sufficient to Induce Preinvasive Neoplasms (A) Hematoxylin and eosin (H&E) staining of murine pancreatic tissue used to prepare organoids (top). Arrows indicate mouse normal or PanIN ductal structures. Ducts embedded in Matrigel immediately following isolation (middle) and organoids 3 days postisolation (bottom). Arrowheads mark isolated ducts and growing organoids. Scale bars, 50 mm. (B) Immunoblots for Kras, pan Ras, Kras-GTP by RBD-GST pull-down, and Tubulin in mN and mPanIN (mP) organoids. PCR confirmation of Cre-mediated recombination of the KrasLSL-G12D allele (bottom). (C) qRT-PCR of ductal (Pdx1, Ck19, Sox9, and Hnf6), acinar (Ptf1a, Cpa1, and Amy), and endocrine (Ngn3, Chga, and Ins2) lineage markers in mN and mP organoids. Means of three biological replicates are shown. Error bars indicate SEMs. Values were normalized to mouse normal pancreas. (D) qRT-PCR of genes indicative of PanIN lesions (Muc5ac, Muc6, Tff1, and Klf4) in mN and mP organoids. Values were normalized to mN organoids. Means of three biological replicates are shown. Error bars indicate SEMs. **p < 0.01 by two-tailed Student’s t test. (E) H&E, Alcian blue staining, and immunohistochemistry (IHC) of orthotopic, syngeneic transplants of GFP-transduced mN and mP organoids. Scale bars, 200 mm. (F) Immunoblots for Kras, pan Ras, Kras-GTP by RBD-GST pull-down, and tubulin in Kras+/LSL-G12D organoids transduced with adenoviral-Cre (Ad-Cre) or adenoviral-blank (Ad-Bl). PCR confirmation of Cre-mediated recombination of the KrasLSL-G12D allele (bottom). (G) H&E, Alcian blue staining, and IHC of orthotopic syngeneic transplants of organoids transduced with Ad-Bl (Kras+/LSL-G12D; R26LSL-YFP) and Ad-Cre (Kras+/G12D; R26YFP) 2 weeks posttransplant. Scale bars, 200 mm. See also Figure S1 and Table S1.

326 Cell 160, 324–338, January 15, 2015 ª2015 Elsevier Inc.

organoids from ducts harboring the conditional KrasLSL-G12D allele (Hingorani et al., 2003). Following activation of Kras by adenoviral-Cre (Ad-Cre) infection, KrasG12D organoids maintained expression of genes specific to ductal cells and not acinar or endocrine lineages (Figures S1D and S1E). Recombination of the KrasLSL-G12D allele was confirmed by PCR, and levels of GTP-bound Kras were increased relative to control-infected organoids (Figure 1F). In addition, expression of KrasG12D resulted in the upregulation of genes associated with human PanIN (Figure S1F). The KrasG12D-expressing organoids demonstrated increased proliferation relative to control organoids (Figure S1G). Finally, KrasG12D organoids formed mPanIN-like structures with columnar cell morphology when implanted orthotopically into syngeneic mice (Figure 1G). This morphology contrasted with the normal-appearing ductal architecture formed by transplanting Kras+/LSL-G12D organoids or wild-type mN (Figures 1E and 1G). The ability of mPanIN-like structures to develop from KrasG12Dexpressing ductal organoids following transplantation demonstrates that ductal cells are also competent to form mPanINs. Tumor-Derived Organoids Provide a Model for Murine PDA Progression We prepared pancreatic ductal organoids from multiple murine primary tumors (mT) and metastases (mM) from KC and Kras+/LSL-G12D; Trp53+/LSL-R172H; Pdx1-Cre (‘‘KPC’’) mice, which develop mPDA more rapidly than KC mice (Figures 2A and Table S2A) (Hingorani et al., 2005). mT and mM organoids exhibited recombination of the Kras LSL-G12D allele, as well as increased levels of Kras-GTP and Kras protein (Figure 2B). mT and mM organoids had increased levels of S6 phosphorylation, but not of Erk or Akt phosphorylation (Figure 2B). Orthotopic transplantation of mT organoids initially generated low- and high-grade lesions that resembled mPanIN (Figure 2C and Table S2B). Over longer periods of time (1–6 months), transplants developed into invasive primary and metastatic mPDA (Figure 2C and Table S2B). mT organoids engrafted with a similar efficiency upon orthotopic transplantation in Nu/Nu mice (91.7%) compared to C57Bl/6 mice (85%), but disease progression was accelerated in Nu/Nu hosts (Table S2B). Although most mT organoid transplants required several months to progress from early mPanIN-like lesions to invasive and metastatic cancer (Figure 2C and Table S2B), mM organoids rapidly formed invasive mPDA within 1 month (Table S2C). The ability of organoid transplants to reproduce the discrete stages of disease progression contrasts with the rapid formation of advanced mPDA following transplantation of 2D cell lines (Figures S2A–S2C) (Olive et al., 2009). Tumors derived from transplanted mT and mM organoids exhibited prominent stromal responses and resembled autochthonous tumors from KPC mice (Figure S2A) (Olive et al., 2009). This stromal response is often absent in tumors formed from 2D cell lines (Figure S2A) (Olive et al., 2009). Low vascular density and high vessel-to-tumor distance were also observed, demonstrating the close resemblance of the organoid transplantation models to autochthonous mPDA, in contrast to transplanted 2D cell lines (Figures S2A–S2C) (Olive et al., 2009). Loss of heterozygosity (LOH) for Trp53 has been reported as a common feature of mPDA based on studies of 2D cell lines (Hin-

gorani et al., 2005). Therefore, we assayed for Trp53 LOH in our murine 3D organoids. All mT organoids prepared from KPC tumors maintained expression of p16, did not exhibit Trp53 LOH, and maintained a stable karyotype, whereas most mM organoids lost the wild-type Trp53 allele and were aneuploid (Figures 2D, 2E, and S2D). We generated 2D cell lines from mT and mM organoids but found that mN and mP organoids were unable to propagate in 2D. mT1 was derived from a KC mouse PDA, lacks the mutant Trp53 allele, and was also unable to propagate in 2D. All mT-derived 2D cell lines exhibited Trp53 LOH and were aneuploid (Figures 2E and S2D). To determine whether organoids are suitable for genetic cooperation experiments, shRNAs targeting p53 and p16/p19 were introduced into mP organoids (Figure S2E). Although the proliferation of mP organoids increased upon knockdown of either p53 or p16/p19 (Figure S2G), only p53 knockdown enabled 2D growth and colony formation (Figure S2F; data not shown). Also, only p53 knockdown promoted progression of mP organoid transplants to invasive carcinoma within 3 months (Figure S2H). This contrasts with a previous report that Kras mutation and biallelic loss of p16/p19 promoted mPDA (Aguirre et al., 2003; Bardeesy et al., 2006) and may reflect differences in the genetic system or the initiating cellular compartment. Nevertheless, the cooperation between p53 depletion and oncogenic Kras demonstrates that organoids are a facile system to evaluate genetic mediators of PDA progression. Human Pancreatic Organoids Model PanIN to PDA Progression We modified our culture conditions to support the propagation of human normal and malignant pancreatic tissues. Isolation of ductal fragments was not always feasible because some normal pancreatic tissue samples were predigested in preparation for islet transplantation. Therefore, we directly embedded digested material into Matrigel. This approach achieved an isolation efficiency of 75%–80% for human normal (hN) organoids (Figures 3A and S3 and Table S3). hN organoids require transforming growth factor b (TGF-b) pathway inhibitors (A83-01 and Noggin), R-Spondin1 and Wnt3a-conditioned media, EGF, and PGE2 for propagation (Figures 3B and 3C). Unlike mN organoids, which have unlimited propagation in culture, hN organoids ceased proliferating after 20 passages or 6 months but could be cryopreserved. We adapted the methods described above to accommodate the extensive desmoplastic reaction in freshly resected PDA specimens and generated human tumor-derived organoids (hT) (Figures 3A and S3 and Table S3). hT organoids could be passaged indefinitely and cryopreserved (Figure 3C). The establishment of hT organoids had efficiencies of 75% (n = 3/4) and 83% (n = 5/6) in the Netherlands and USA, respectively (Table S3). The first specimen that failed to generate an organoid culture was obtained from a patient that had undergone neo-adjuvant chemotherapy, and histologic examination of this specimen revealed extensive necrosis. The second specimen that did not generate an organoid culture was predominantly composed of stromal cells, without sufficient viable tumor cells to establish a culture. Although the hN organoids had a simple, cuboidal morphology, the hT organoids had differing degrees of Cell 160, 324–338, January 15, 2015 ª2015 Elsevier Inc. 327

Figure 2. Modeling Murine PDA Progression with Tumor- and Metastasis-Derived Organoids (A) H&E staining of murine tissue from which tumor and metastasis organoids were derived (top). Arrowhead indicates metastasis. Scale bars, 50 mm. Digested murine tissues embedded in Matrigel immediately following isolation (middle) and organoids 3 days postisolation (bottom). Scale bars, 200 mm. (B) Immunoblots of selected signaling effectors, Kras-GTP and Ras-GTP by RBD-GST pull-down, and tubulin. PCR confirmation of KrasLSL-G12D recombination in mP, mT, and mM organoids (bottom). (C) H&E staining of tumors and metastases (Met) derived from mT organoid orthotopic transplants. Scale bars, 200 mm (top) and 50 mm (bottom). (D) Loss of heterozygosity of the wild-type Trp53 allele determined by PCR (top) and immunoblot analysis of Trp53, Smad4, p16, and Tubulin. mM3L, derived from a liver metastasis. (E) Karyotypes of organoids and monolayer (2D) cell lines. See also Figure S2 and Table S2.

328 Cell 160, 324â&#x20AC;&#x201C;338, January 15, 2015 ÂŞ2015 Elsevier Inc.

Figure 3. Human Pancreatic Ductal Organoids Recapitulate Features of Normal and Neoplastic Ducts (A) Representative images (top) and H&E staining (middle) of human organoid cultures established from normal tissues (hN1-2), resected primary tumors (hT1-2), a resected metastatic lung lesion (hM1), and a fine-needle aspiration biopsy of a metastatic lesion (hFNA2). H&E staining of the resected tissues from which the organoids were derived (bottom). Scale bars, 500 mm (top), 250 mm (middle), and 500 mm (bottom). (B) Representative images of hN and hT organoids cultured for 2 weeks (1 passage) in human complete media or in human complete media lacking the indicated factors. Scale bars, 500 mm. (C) Number of passages hN and hT organoids could be propagated in the absence of the indicated factors. (D) Targeted sequencing analysis of human organoids. Genes altered in more than one sample and/or known to be mutated in PDA are shown. If multiple mutations were found in a gene, only one mutation per gene is shown. Color key for the type of genetic alterations is shown. Met indicates organoids derived from metastatic samples. See also Figure S3 and Tables S3 and S4.

Cell 160, 324â&#x20AC;&#x201C;338, January 15, 2015 ÂŞ2015 Elsevier Inc. 329

dysplastic tall columnar cells, resembling low-grade PanINs (Figures 3A). hT organoids tolerated the withdrawal of certain growth factors from the media (Figures 3B and 3C). 85% of pancreatic cancer patients are ineligible for surgical resection of their tumors (Ryan et al., 2014). Therefore, we determined whether hT organoids could be generated from the limited amount of cellular material provided by endoscopic biopsies using fine needle aspirations (FNA). Initial attempts to generate organoids from FNA biopsies were hampered by loss of cellular material during digestion. Upon optimization of these conditions, human FNA biopsy organoids (hFNA) were generated from two specimens that were not dissociated prior to suspension in Matrigel (Figures 3A and S3 and Table S3). This approach is broadly applicable to PDA patients and enables serial sampling. Targeted sequencing of 2,000 cancer-associated genes was performed on hN and hT organoids. As expected, no mutations were detected in the hN organoid cultures. These analyses identified oncogenic KRAS mutations in the majority of tumorderived samples (n = 8), as well as mutations in TP53 (n = 7), SMAD4 (n = 5), and CDKN2A (n = 4) (Figure 3D and Table S4). We also noted amplification of known oncogenes, such as MYC (n = 4), and loss of tumor suppressors, including TGFBR2 (n = 3) and DCC (n = 5). Importantly, the same KRAS mutations observed in several hT organoids were confirmed in the primary PDA from which they were derived (Table S4). The allele frequency of oncogenic KRAS variants in hT1–hT5 and hFNA2 ranged from 50–100%. In contrast, the KRASG12V allele frequency in hFNA1 was only 1% (Table S4), which may result from coexistence of wild-type ductal cells. Although KRAS mutations were not detected in hT8 (Figure 3D and Table S4), the presence of mutations in known PDA genes (ARID1A and MLL3) suggests that hT8 contains malignant cells (Table S4). To further characterize the cell types present in primary PDA organoids, we evaluated the expression of pancreatic lineage markers. hN and hT organoids expressed markers of ductal cells, but not other pancreatic lineages (Figure 4A). The karyotypes of hT organoids were highly aneuploid, whereas the hN organoids were predominantly and stably diploid (Figure 4B). The PDA-associated biomarker CA19-9 (Makovitzky, 1986) was also elevated in hT relative to hN organoids (Figure 4C). The hN and hT organoids are therefore reflective of normal and neoplastic human pancreatic ductal cells and offer a model system to explore pancreatic cancer biology in the more genetically complex background of human cancer. Following orthotopic transplantation into Nu/Nu mice, hN organoids produced normal ductal structures at low efficiency (n = 2/23), whereas hT organoids efficiently generated a spectrum of low- and high-grade, extraductal PanIN-like lesions within 1 month (n = 9/12) (Figures 4D and S4A and Table S4D). The hT-derived transplants initially formed well-defined hollow lesions lined by a single layer of columnar epithelial cells with apical mucin and basally located, relatively uniform nuclei. The nuclei were small and lacked the pleomorphism and hyperchromasia often seen in invasive PDA. These lesions progressed over several months to infiltrative carcinoma comprised of poorly defined and invasive glands (Figures 4D and S4A and Table S4). A prominent desmoplastic reaction was present in hTderived PanIN-like structures and PDA, including the deposition 330 Cell 160, 324–338, January 15, 2015 ª2015 Elsevier Inc.

of a collagen-rich stroma and the recruitment of aSMA-positive cells (Figure S4B). The mutation or loss of TP53 or SMAD4 in hT1 and hT2 was also detected by IHC in these tumors (Figure S4C and Table S4). Overall, hT organoids represent a transplantable model of human pancreatic cancer progression. Gene Expression Analysis of Murine Pancreatic Ductal Organoids Implicates Candidate Genes in PDA Progression The mouse organoids were prepared from syngeneic mice, offering the ability to discern gene expression changes in organoids and determine whether these changes correlate with PDA progression. We harvested RNA from mN (n = 7), mP (n = 6), and mT (n = 6) organoids and generated strand-specific RNAsequencing (RNA-seq) libraries. Sequences were mapped to the mm9 version of the mouse genome, and relative transcript abundances (transcripts per million) of 29,777 mouse genes were determined (Table S5). Principal component analysis revealed that mN organoids were distinct from mP and mT organoids (Figure 5A and Table S5). Genes whose levels differed significantly among mN, mP, and mT organoids were identified. 772 genes were found downregulated and 863 genes upregulated in mP relative to mN organoids (Figure 5B and Table S5). When mT organoids were compared to mN organoids, 2,721 genes were downregulated and 2,695 were upregulated. In addition, 823 genes were downregulated and 640 genes were upregulated in mT relative to mP organoids. Distinct patterns of gene expression were found in the data set (Figure 5C). The majority of genes differentially expressed in mP relative to mN organoids changed in a similar manner in mT relative to mN organoids (Figure 5D). However, a much larger cohort of genes changed in expression in mT relative to mN than in mP relative to mN organoids (Figure 5D), suggesting that mP organoids represent an intermediate state between mN and mT organoids. The glycosyltransferase Gcnt3 and putative protein disulfide isomerase Agr2 were among the most upregulated genes in both mP and mT organoids and have been demonstrated to be elevated in human PDA (Figure 5E) (Dumartin et al., 2011; Zhao et al., 2014). The most upregulated gene in both mP and mT relative to mN organoids was the acyl-CoA synthetase Acsm3 (Figure 5E). RNA-seq results were confirmed by qRT-PCR for 35 out of 40 genes (Table S5), including the upregulation of Agr2, Acsm3, Gcnt1, Gcnt3, and Ugdh and the downregulation of Ptprd in mP and mT organoids (Figure 5F and Table S5). Among the genes upregulated in mP and mT relative to mN organoids, Gcnt1, Gcnt3, Acsm3, Agr2, Syt16, Nt5e, and Ugdh were upregulated following the Ad-Cre-induced expression of oncogenic KrasG12D, suggesting that these genes are activated downstream of mutant KrasG12D (Figure S5A). To determine whether organoid RNA-seq profiles resembled gene expression patterns in vivo, we compared our organoid RNA-seq data to a published transcription profile of murine pancreatic tumors upon KrasG12D inactivation (Ying et al., 2012). Genes differentially expressed upon inactivation of oncogenic Kras overlapped significantly with those up or downregulated in mP or mT relative to mN organoids (Figure S5B). These analyses demonstrate the ability of the organoid system to identify molecular alterations associated with PDA progression.

Figure 4. Molecular Characterization and Orthotopic Transplantation of Human Organoids (A) qRT-PCR of pancreas lineage markers in hN (n = 3) and hT (n = 4) organoids. Mean expression levels were normalized to total pancreas. Error bars indicate SEMs. (B) Karyotyping of human organoids (2 hN, 2hT) at the indicated passages (P). (C) CA19-9 and actin levels in hN, hT, or hM organoids. The solid line indicates noncongruent lanes. (D) H&E, Alcian blue staining, and IHC of orthotopic hT2 transplants and the primary tumor. Scale bars, 200 mm (top two panels) and 50 mm (bottom two panels). See also Figure S4 and Table S4D.

Proteomic Alterations in Murine Pancreatic Ductal Organoids Predict Pathways Associated with PDA Progression As an orthogonal method to investigate molecular alterations in murine pancreatic organoids, we characterized the global proteomes of mN (n = 5), mP (n = 4), and mT (n = 5) organoids. Protein lysates were processed using amine-reactive isobaric tags for relative and absolute quantification (iTRAQ) mass spectrometry (Wiese et al., 2007). Samples were run in four 8-plex experiments and merged using an approach that normalizes the data to common samples included across all experiments (Extended Experimental Procedures). Upon merging, 6,051 unique protein

isoforms were quantified in all samples. We applied linear regression modeling on the normalized intensity peak values and identified 710 protein isoform expression changes between mN and mP organoids (Figure 6A). 1,047 protein isoforms changed expression between mN and mT organoids, and 63 differentially expressed proteins were identified between mP and mT (Figure 6A). The relatively small number of protein expression changes identified between mP and mT organoids reflects their biological similarity (Figure S6A). mN organoids showed unique proteomic profiles from their mP and mT counterparts (Figures 6B and 6C). To compare the proteomic and RNA-seq data, we collapsed the unique protein Cell 160, 324â&#x20AC;&#x201C;338, January 15, 2015 ÂŞ2015 Elsevier Inc. 331

Figure 5. Gene Expression Analysis of Murine Organoids Reveals Genetic Changes Correlated with Pancreatic Cancer Progression (A) Principal component analysis of gene expression data for mN, mP, and mT organoids. (B) The number of genes differentially expressed (DESeq adjusted p value < 0.05) among mN (n = 7), mP (n = 6), and mT (n = 6) organoids. (C) Heatmap showing relative expression levels using Z score normalization among mN, mP, and mT organoids. Color key of Z score is shown. (D) Venn diagrams show overlap of genes significantly differentially expressed in mP and mT relative to mN organoids. The p values for overlaps were determined by two-tailed Fisher’s exact test. (E) Genes with the largest fold changes in mP or mT relative to mN organoids. (F) qRT-PCR validation of mN, mP and mT organoid gene expression changes. Values were normalized to mean levels in mN organoids. n = 8 mN, 7 mP, and 8 mT organoid cultures. Error bars indicate SEMs. *p < 0.05, **p < 0.01, ***p < 0.001, and ns, not significant by two-tailed Student’s t test. See also Figure S5 and Table S5.

isoforms into their corresponding 4,155 genes. Some protein expression changes (e.g., 123/150 for downregulated and 96/ 151 for upregulated mP proteins) did not reflect corresponding transcriptional changes, indicating that protein stability may play a role in cancer progression, particularly in mP organoids (Figure 6D). Nonetheless, the proteomic data validated many of the expression changes identified by RNA-seq (Figure 6D), including upregulation of Gcnt3, Agr2, and Ugdh (Table S6). Additionally, of the 1,599 genes whose expression levels changed in mT relative to mN organoids that were measured by mass spectrometry, 301 (19%) showed corresponding protein changes (Figure 6D). Gene Set Enrichment Analysis (GSEA) on the RNA-seq and proteomic data (Subramanian et al., 2005) revealed elevated 332 Cell 160, 324–338, January 15, 2015 ª2015 Elsevier Inc.

expression of genes and proteins involved in glutathione metabolism and biological oxidations in mP relative to mN organoids (Figures 6E, S6B, and S6C and Table S7), which is consistent with elevations in reactive oxygen species metabolism previously reported in KrasG12D cells (DeNicola et al., 2011; Ying et al., 2012). Enrichment of proteins involved in glutathione metabolism was also found in mT relative to mN organoids (Table S7). Additionally, we identified a significant positive enrichment of proteins involved in the steroid biosynthesis, cholesterol biosynthesis, one carbon pool by folate, and pyrimidine metabolism pathways (Figures 6E, S6B, and S6C and Table S7), which is consistent with an earlier report (Ying et al., 2012). Similar pathways were enriched in mP relative to mN organoids (cholesterol biosynthesis, one carbon pool by folate, and pyrimidine

Figure 6. Proteomic Profiling of Murine Organoids Uncovers Molecular Pathways Linked to Pancreatic Cancer Progression (A) Protein expression changes by iTRAQ proteomic analysis of murine organoids. Both unique protein isoforms and protein isoforms encoded by the same gene are included (adjusted p value < 0.1 by linear regression analysis). (B) Heatmap of unique protein isoforms that differ (adjusted p value < 0.05) among mN, mP, and mT organoids. Color key of the Z score is shown. (C) Venn diagrams showing overlaps between proteins differentially expressed (p < 0.05) in mP and mT relative to mN organoids. p values for overlaps were determined by two-tailed Fisher’s exact test. (D) Venn diagrams showing overlaps between genes and proteins found differentially expressed by RNA-seq and proteomic analyses (adjusted p < 0.05). p values for the overlaps were determined by two-tailed Fisher’s exact test. (E) Molecular pathways found enriched by GSEA analysis of RNA-seq and proteomic data. Normalized enrichment scores (NESs), p and q values are shown. (F) Heatmap showing relative gene expression levels of nucleoporins in mN, mP, and mT organoids determined by RNA-seq. Color key of the Z score is shown. See also Figure S6 and Tables S6 and S7.

metabolism) (Figures S6B and S6C and Table S7), whereas fatty acid metabolism and TCA cycle/respiratory electron transport pathways were downregulated (Figure S6C and Table S7). The increase in anabolic and decrease in catabolic pathways suggest that complex alterations in fatty acid and nucleotide metabolism occur during PDA progression. Interestingly, we also found broad upregulation of the nucleoporin family at both the RNA and protein levels in the mT relative to mN organoids (Figures 6E and 6F and Table S6). The individual nucleoporins NUP214, NUP153, and NUPL1 were previously identified in shRNA dropout screens in PDA cell lines (Cheung et al., 2011; Shain et al., 2013). Furthermore, amplification of

NUP153 was detected in one human PDA cancer cell line, and elevation of NUP88 was detected in human primary PDA (Cheung et al., 2011; Gould et al., 2000; Shain et al., 2013). This systematic analysis of molecular alterations in pancreatic organoids implicates nuclear transport as a pathway correlated with pancreatic cancer progression. In Vivo Mouse and Human Validation of Candidates Associated with PDA Progression in Organoids To demonstrate that the mouse organoid culture system represents a biological resource for the accurate discovery of genes associated with PDA progression, we selected 16 genes Cell 160, 324–338, January 15, 2015 ª2015 Elsevier Inc. 333

Figure 7. Increased Levels of ACSM3, NT5E, and GCNT3 Correlate with Mouse and Human PDA Progression (A) IHC analysis of 14 candidate genes in mouse adjacent normal ducts, mPanlN and mPDA. Differential expression is indicated as (negative), + (weak), ++ (moderate), or +++ (strong). Only the ductal component of the normal pancreas was scored. (B) IHC analysis of Acsm3, Nt5e, and Gcnt3 in mouse normal ducts, mPanlN and mPDA tissues. Arrow indicates adjacent normal ducts in mPanlN tissues. Arrowhead indicates mPanlN or mPDA. Scale bars, 50 mm. (C) IHC analysis of seven candidate genes in human normal pancreas, hT orthotopic transplants, and PDA tissues. Differential expression is indicated as (negative), + (weak), ++ (moderate), or +++ (strong). Only the ductal component of the normal pancreas was scored. (D) IHC analysis of ACSM3, NT5E, and GCNT3 in human normal pancreas and PDA tissues. Arrow indicates normal ducts, and arrowhead indicates PDA. Scale bars, 50 mm. See also Figure S7.

upregulated in mT organoids for validation in primary tissue specimens by IHC and immunofluorescence (IF) (Figure 7A). These 16 genes included enzymes, membrane proteins, structural proteins, and secreted ligands, which could represent candidate biomarkers and therapeutic targets. Of the 14 antibodies that generated a detectable signal on murine pancreatic tissue sections, 13 antibodies confirmed the increased expres334 Cell 160, 324â&#x20AC;&#x201C;338, January 15, 2015 ÂŞ2015 Elsevier Inc.

sion of the candidate protein in mPanIN and mPDA lesions in concordance with the RNA-seq and proteomic data (Figures 7A, 7B, and S7A). 11 of the 13 candidate antibodies were compatible for evaluation in human tissues, and 7 of these candidates were upregulated in human PDA when compared to normal pancreatic ductal tissues (Figures 7C, 7D, and S7B). The high expression of many of these markers was recapitulated

in orthotopic transplants of hT organoids into Nu/Nu mice (Figure 7C). These results indicate that the organoid culture system accurately models PDA progression and can serve as a resource for the discovery and genetic dissection of pathways driving human pancreatic tumorigenesis.

the University of Illinois at Chicago and University of Miami Miller School of Medicine. All human experiments were approved by the ethical committees of the University Medical Centre Utrecht or the IRBs of MSKCC, MDACC, WCMC, and CSHL. Written informed consent from the donors for research use of tissue in this study was obtained prior to acquisition of the specimen. Samples were confirmed to be tumor or normal based on pathological assessment.

DISCUSSION We have established pancreatic organoids as a tractable and transplantable system to probe the molecular and cellular properties of neoplastic progression in mice and humans. In contrast to prior reports (Agbunag and Bar-Sagi, 2004; Rovira et al., 2010; Seaberg et al., 2004), our culture conditions prevent the rapid exhaustion of normal ductal cells in vitro and generate a normal ductal architecture following orthotopic transplantation. Importantly, the ability to passage and transplant both normal and neoplastic ductal cells enables a detailed analysis of molecular pathways and cellular biology that is not possible when neonatal pancreatic fragments are propagated in air-liquid interfaces or when induced pluripotent cells are employed (Agbunag and Bar-Sagi, 2004; Kim et al., 2013; Li et al., 2014). Our finding that nucleoporins are broadly upregulated in the neoplastic murine organoids, coupled with the known associations of nucleoporins to cell proliferation and cell transformation, presents a class of proteins to investigate in pancreatic cancer progression (Gould et al., 2000; KoĚ&#x2C6;hler and Hurt, 2010). Furthermore, the ability to systematically characterize human pancreatic cancer organoids that lack KRAS mutations, such as hT8, will reveal driver genes for PDA. Finally, because organoids can be readily established from small patient biopsies, they should hasten the development of personalized approaches for pancreatic cancer patients.

Human Pancreatic Tumor and Normal Organoid Culture Tumor tissue was minced and digested with collagenase II (5 mg/ml, GIBCO) in human complete medium (see below) at 37 C for a maximum of 16 hr. The material was further digested with TrypLE (GIBCO) for 15 min at 37 C, embedded in GFR Matrigel, and cultured in human complete medium (AdDMEM/F12 medium supplemented with HEPES [13, Invitrogen], Glutamax [13, Invitrogen], penicillin/streptomycin [13, Invitrogen], B27 [13, Invitrogen], Primocin [1 mg/ml, InvivoGen], N-acetyl-L-cysteine [1 mM, Sigma], Wnt3a-conditioned medium [50% v/v], RSPO1-conditioned medium [10% v/v, Calvin Kuo], Noggin-conditioned medium [10% v/v] or recombinant protein [0.1 mg/ml, Peprotech], epidermal growth factor [EGF, 50 ng/ml, Peprotech], Gastrin [10 nM, Sigma], fibroblast growth factor 10 [FGF10, 100 ng/ml, Preprotech], Nicotinamide [10 mM, Sigma], and A8301 [0.5 mM, Tocris]). Normal samples were processed as above, except that the collagenase digestion was done for a maximum of 2 hr in the presence of soybean trypsin inhibitor (1 mg/ml, Sigma). Following digestion, cells were embedded in GFR Matrigel and cultured in human complete medium with the addition of PGE2 (1 mM, Tocris). Additional experimental details and methods can be found in the Extended Experimental Procedures. ACCESSION NUMBERS All RNA-seq data are available at Gene Expression Omnibus (GEO) under accession number GSE63348. The proteomic raw data are available at PeptideAtlas under accession number PASS00625. The targeted DNAsequencing data are available at EMBL European Nucleotide Archive under the accession number ERP006373.

EXPERIMENTAL PROCEDURES

SUPPLEMENTAL INFORMATION

Animals Trp53+/LSL-R172H, Kras+/LSL-G12D, and Pdx1-Cre strains in C57Bl/6 background were interbred to obtain Pdx1-Cre; Kras+/LSL-G12D (KC) and Pdx1Cre; Kras+/LSL-G12D; Trp53+/LSL-R172H (KPC) mice (Hingorani et al., 2005). The R26LSL-YFP strain was interbred to get the desired genotype. C57Bl/6 and athymic Nu/Nu mice were purchased from Charles River Laboratory and Jackson Laboratory. All animal experiments were conducted in accordance with procedures approved by the IACUC at Cold Spring Harbor Laboratory (CSHL).

Supplemental Information includes Extended Experimental Procedures, seven figures, and seven tables and can be found with this article online at http://dx. doi.org/10.1016/j.cell.2014.12.021.

Murine Pancreatic Ductal Organoid Culture Detailed procedures to isolate normal pancreatic ducts have been described previously (Huch et al., 2013a). In brief, normal and preneoplastic pancreatic ducts were manually picked after enzymatic digestion of pancreas with 0.012% (w/v) collagenase XI (Sigma) and 0.012% (w/v) dispase (GIBCO) in DMEM media containing 1% FBS (GIBCO) and were seeded in growth-factor-reduced (GFR) Matrigel (BD). For tumors and metastases, bulk tissues were minced and digested overnight with collagenase XI and dispase and embedded in GFR Matrigel. Human Specimens Pancreatic cancer tissues and adjacent normal pancreas were obtained from patients undergoing surgical resection at the University Medical Centre Utrecht Hospital, Memorial Sloan-Kettering Cancer Center (MSKCC), MD Anderson Cancer Center (MDACC), and Weill Cornell Medical College (WCMC). Normal pancreatic tissue was also obtained from islet transplant programs at

AUTHOR CONTRIBUTIONS S.F.B. initiated the project, developed the methods for isolating mouse and human organoids, and characterized human organoids (Figures 1A, 3, 4B, 4D, and S4C and Tables S3 and S4). C.-I.H. developed transplantation models for organoids and performed shRNA knockdown and histological and karyotypic analyses (Figures 1A, 1E, 1G, 2A, 2Câ&#x20AC;&#x201C;2E, 4D, 7, S1Aâ&#x20AC;&#x201C;S1C, S2Aâ&#x20AC;&#x201C;S2F, S2H, S4A, S4B, S7A, and S7B and Tables S1, S2, and S4). L.A.B. performed RNA-seq on mouse organoids and analyzed RNA-seq and proteomic data (Figures 5, 6, S5B, and S6 and Tables S5, S6, and S7). I.I.C.C. conducted proteomic evaluation of mouse organoids and analyzed proteomic data (Figures 6 and S6C). D.D.E. developed mouse organoid methods and evaluated CA19-9 levels in human organoids (Figures 1A, 2A, 4C, S3, and S6 and Table S6). V.C. developed human organoid methods, performed molecular analyses of organoids, and prepared material for DNA-sequencing and sequencing of Kras (Figures 1C, 1D, 3A, 3D, 4A, S1Dâ&#x20AC;&#x201C;S1F, and S5A and Tables S3, S4, and S5). M.J. performed and analyzed the DNA sequencing of human organoids (Figure 3D and Table S4). Mouse and human organoid preparation and characterization was performed by M.P.-S., H.T., M.S.S., T.O., D.OĚ&#x2C6;., A.H.-S., C.M.A.-A., M.L., E.E., B.A., M.E.F., G.N.Y., G.B., B.D., B.C., K.W., K.H.Y., Y.P., M. Huch, A.G., F.H.M.M., and S.D.L. Sequencing analyses were performed by Y.H., Y.J., M. Hammell, I.J.N., E.C., and R.v.B. Pathological analyses were

Cell 160, 324â&#x20AC;&#x201C;338, January 15, 2015 ÂŞ2015 Elsevier Inc. 335

performed by G.J.O., R.H.H., D.S.K., O.B., and C.I.-D. Surgical resections and tissue dissection were performed by I.Q.M. and I.H.B.R. Proteomic development was performed by D.J.P., K.D.R., and J.P.W. Overall study management was conducted by D.A.T., H.C., and R.G.J.V. S.F.B., D.D.E., L.A.B., M.E.F., C.H., H.T., V.C., M.P.S., R.G.J.V., H.C., I.I.C.C., and D.A.T. contributed to manuscript writing. ACKNOWLEDGMENTS We thank Peter Kapitein and Jan Schuurman from Inspire 2 Live for helping to establish the collaboration between D.A.T. and H.C. We also thank H. Begthel and J. Korving for technical assistance. This work was performed with assistance from the CSHL Proteomic, Histology, DNA Sequencing, Antibody, and Bioinformatics Shared Resources, which are supported by the Cancer Center Support Grant 5P30CA045508. D.A.T. is a distinguished scholar of the Lustgarten Foundation and Director of the Lustgarten Foundation-designated Laboratory of Pancreatic Cancer Research. D.A.T. is also supported by the Cold Spring Harbor Laboratory Association, the Carcinoid Foundation, PCUK, and the David Rubinstein Center for Pancreatic Cancer Research at MSKCC. In addition, we are grateful for support from the following: Stand Up to Cancer/KWF (H.C.), the STARR foundation (I7-A718 for D.A.T.), DOD (W81XWH-13-PRCRP-IA for D.A.T.), the Sol Goldman Pancreatic Cancer Research Center (R.H.H.), the Italian Ministry of Health (FIRB - RBAP10AHJ for V.C.), Sociedad Española de Oncologı́a Médica (SEOM for M.P.S.), Louis Morin Charitable Trust (M.E.F.), the Swedish Research Council (537-2013-7277 for D.Ö.), The Kempe Foundations (JCK-1301 for D.Ö.) and the Swedish Society of Medicine (SLS-326921, SLS-250831 for D.Ö.), the Damon Runyon Cancer Research Foundation (DRG-2165-13 for I.I.C.C.), the Human Frontiers Science Program (LT000403/2014 for E.E.), the Weizmann Institute of Science Women in Science Award (E.E.), the American Cancer Society (PF-13-317-01-CSM for C.M.A.A.), the Hearst Foundation (A.H.S.), and the NIH (5P30CA45508-26, 5P50CA101955-07, 1U10CA18094401, 5U01CA168409-3, and 1R01CA190092-01 for D.A.T.; CA62924 for R.H.H.; CA134292 for S.D.L.; 5T32CA148056 for L.A.B. and D.D.E.; and CA101955 UAB/UMN SPORE for L.A.B.). In addition, S.F.B. and M.H. are supported by KWF/PF-HUBR 2007-3956, A.G is supported by EU/232814StemCellMark, and R.G.J.V. is supported by GenomiCs.nl (CGC). M.J., R.B., and E.C. are supported by the CancerGenomics.nl (NWO Gravitation) program. Ralph Hruban receives royalty payments from Myriad Genetics for the PalB2 inventions. Hans Clevers and Meritxell Huch have patents pending and granted on the organoid technology. Received: August 1, 2014 Revised: November 24, 2014 Accepted: December 10, 2014 Published: December 31, 2014 REFERENCES Abbruzzese, J.L., and Hess, K.R. (2014). New option for the initial management of metastatic pancreatic cancer? J. Clin. Oncol. 32, 2405–2407. Agbunag, C., and Bar-Sagi, D. (2004). Oncogenic K-ras drives cell cycle progression and phenotypic conversion of primary pancreatic duct epithelial cells. Cancer Res. 64, 5659–5663. Aguirre, A.J., Bardeesy, N., Sinha, M., Lopez, L., Tuveson, D.A., Horner, J., Redston, M.S., and DePinho, R.A. (2003). Activated Kras and Ink4a/Arf deficiency cooperate to produce metastatic pancreatic ductal adenocarcinoma. Genes Dev. 17, 3112–3126. Bardeesy, N., Aguirre, A.J., Chu, G.C., Cheng, K.H., Lopez, L.V., Hezel, A.F., Feng, B., Brennan, C., Weissleder, R., Mahmood, U., et al. (2006). Both p16(Ink4a) and the p19(Arf)-p53 pathway constrain progression of pancreatic adenocarcinoma in the mouse. Proc. Natl. Acad. Sci. USA 103, 5947–5952. Barker, N., Huch, M., Kujala, P., van de Wetering, M., Snippert, H.J., van Es, J.H., Sato, T., Stange, D.E., Begthel, H., van den Born, M., et al. (2010). Lgr5(+ve) stem cells drive self-renewal in the stomach and build long-lived gastric units in vitro. Cell Stem Cell 6, 25–36.

336 Cell 160, 324–338, January 15, 2015 ª2015 Elsevier Inc.

Beatty, G.L., Chiorean, E.G., Fishman, M.P., Saboury, B., Teitelbaum, U.R., Sun, W., Huhn, R.D., Song, W., Li, D., Sharp, L.L., et al. (2011). CD40 agonists alter tumor stroma and show efficacy against pancreatic carcinoma in mice and humans. Science 331, 1612–1616. Cheung, H.W., Cowley, G.S., Weir, B.A., Boehm, J.S., Rusin, S., Scott, J.A., East, A., Ali, L.D., Lizotte, P.H., Wong, T.C., et al. (2011). Systematic investigation of genetic vulnerabilities across cancer cell lines reveals lineage-specific dependencies in ovarian cancer. Proc. Natl. Acad. Sci. USA 108, 12372– 12377. Cleveland, M.H., Sawyer, J.M., Afelik, S., Jensen, J., and Leach, S.D. (2012). Exocrine ontogenies: on the development of pancreatic acinar, ductal and centroacinar cells. Semin. Cell Dev. Biol. 23, 711–719. De La O, J.P., Emerson, L.L., Goodman, J.L., Froebe, S.C., Illum, B.E., Curtis, A.B., and Murtaugh, L.C. (2008). Notch and Kras reprogram pancreatic acinar cells to ductal intraepithelial neoplasia. Proc. Natl. Acad. Sci. USA 105, 18907– 18912. DeNicola, G.M., Karreth, F.A., Humpton, T.J., Gopinathan, A., Wei, C., Frese, K., Mangal, D., Yu, K.H., Yeo, C.J., Calhoun, E.S., et al. (2011). Oncogeneinduced Nrf2 transcription promotes ROS detoxification and tumorigenesis. Nature 475, 106–109. Dumartin, L., Whiteman, H.J., Weeks, M.E., Hariharan, D., Dmitrovic, B., Iacobuzio-Donahue, C.A., Brentnall, T.A., Bronner, M.P., Feakins, R.M., Timms, J.F., et al. (2011). AGR2 is a novel surface antigen that promotes the dissemination of pancreatic cancer cells through regulation of cathepsins B and D. Cancer Res. 71, 7091–7102. Erkan, M., Reiser-Erkan, C., Michalski, C.W., Deucker, S., Sauliunaite, D., Streit, S., Esposito, I., Friess, H., and Kleeff, J. (2009). Cancer-stellate cell interactions perpetuate the hypoxia-fibrosis cycle in pancreatic ductal adenocarcinoma. Neoplasia 11, 497–508. Frese, K.K., Neesse, A., Cook, N., Bapiro, T.E., Lolkema, M.P., Jodrell, D.I., and Tuveson, D.A. (2012). nab-Paclitaxel potentiates gemcitabine activity by reducing cytidine deaminase levels in a mouse model of pancreatic cancer. Cancer Discov. 2, 260–269. Gao, D., Vela, I., Sboner, A., Iaquinta, P.J., Karthaus, W.R., Gopalan, A., Dowling, C., Wanjala, J.N., Undvall, E.A., Arora, V.K., et al. (2014). Organoid cultures derived from patients with advanced prostate cancer. Cell 159, 176–187. Gidekel Friedlander, S.Y., Chu, G.C., Snyder, E.L., Girnius, N., Dibelius, G., Crowley, D., Vasile, E., DePinho, R.A., and Jacks, T. (2009). Context-dependent transformation of adult pancreatic cells by oncogenic K-Ras. Cancer Cell 16, 379–389. Gould, V.E., Martinez, N., Orucevic, A., Schneider, J., and Alonso, A. (2000). A novel, nuclear pore-associated, widely distributed molecule overexpressed in oncogenesis and development. Am. J. Pathol. 157, 1605–1613. Guerra, C., Mijimolle, N., Dhawahir, A., Dubus, P., Barradas, M., Serrano, M., Campuzano, V., and Barbacid, M. (2003). Tumor induction by an endogenous K-ras oncogene is highly dependent on cellular context. Cancer Cell 4, 111–120. Habbe, N., Shi, G., Meguid, R.A., Fendrich, V., Esni, F., Chen, H., Feldmann, G., Stoffers, D.A., Konieczny, S.F., Leach, S.D., and Maitra, A. (2008). Spontaneous induction of murine pancreatic intraepithelial neoplasia (mPanIN) by acinar cell targeting of oncogenic Kras in adult mice. Proc. Natl. Acad. Sci. USA 105, 18913–18918. Hingorani, S.R., Petricoin, E.F., Maitra, A., Rajapakse, V., King, C., Jacobetz, M.A., Ross, S., Conrads, T.P., Veenstra, T.D., Hitt, B.A., et al. (2003). Preinvasive and invasive ductal pancreatic cancer and its early detection in the mouse. Cancer Cell 4, 437–450. Hingorani, S.R., Wang, L., Multani, A.S., Combs, C., Deramaudt, T.B., Hruban, R.H., Rustgi, A.K., Chang, S., and Tuveson, D.A. (2005). Trp53R172H and KrasG12D cooperate to promote chromosomal instability and widely metastatic pancreatic ductal adenocarcinoma in mice. Cancer Cell 7, 469–483. Huch, M., Bonfanti, P., Boj, S.F., Sato, T., Loomans, C.J., van de Wetering, M., Sojoodi, M., Li, V.S., Schuijers, J., Gracanin, A., et al. (2013a). Unlimited in vitro

expansion of adult bi-potent pancreas progenitors through the Lgr5/R-spondin axis. EMBO J. 32, 2708–2721. Huch, M., Dorrell, C., Boj, S.F., van Es, J.H., Li, V.S., van de Wetering, M., Sato, T., Hamer, K., Sasaki, N., Finegold, M.J., et al. (2013b). In vitro expansion of single Lgr5+ liver stem cells induced by Wnt-driven regeneration. Nature 494, 247–250. Jacobetz, M.A., Chan, D.S., Neesse, A., Bapiro, T.E., Cook, N., Frese, K.K., Feig, C., Nakagawa, T., Caldwell, M.E., Zecchini, H.I., et al. (2013). Hyaluronan impairs vascular function and drug delivery in a mouse model of pancreatic cancer. Gut 62, 112–120.

profiles in pancreatic intraepithelial neoplasia reflect the effects of Hedgehog signaling on pancreatic ductal epithelial cells. Cancer Res. 65, 1619– 1626. Provenzano, P.P., Cuevas, C., Chang, A.E., Goel, V.K., Von Hoff, D.D., and Hingorani, S.R. (2012). Enzymatic targeting of the stroma ablates physical barriers to treatment of pancreatic ductal adenocarcinoma. Cancer Cell 21, 418–429. Pylayeva-Gupta, Y., Lee, K.E., Hajdu, C.H., Miller, G., and Bar-Sagi, D. (2012). Oncogenic Kras-induced GM-CSF production promotes the development of pancreatic neoplasia. Cancer Cell 21, 836–847.

Karthaus, W.R., Iaquinta, P.J., Drost, J., Gracanin, A., van Boxtel, R., Wongvipat, J., Dowling, C.M., Gao, D., Begthel, H., Sachs, N., et al. (2014). Identification of multipotent luminal progenitor cells in human prostate organoid cultures. Cell 159, 163–175.

Rahib, L., Smith, B.D., Aizenberg, R., Rosenzweig, A.B., Fleshman, J.M., and Matrisian, L.M. (2014). Projecting cancer incidence and deaths to 2030: the unexpected burden of thyroid, liver, and pancreas cancers in the United States. Cancer Res. 74, 2913–2921.

Kim, M.P., Evans, D.B., Wang, H., Abbruzzese, J.L., Fleming, J.B., and Gallick, G.E. (2009). Generation of orthotopic and heterotopic human pancreatic cancer xenografts in immunodeficient mice. Nat. Protoc. 4, 1670–1680.

Ray, K.C., Bell, K.M., Yan, J., Gu, G., Chung, C.H., Washington, M.K., and Means, A.L. (2011). Epithelial tissues have varying degrees of susceptibility to Kras(G12D)-initiated tumorigenesis in a mouse model. PLoS ONE 6, e16786.

Kim, J., Hoffman, J.P., Alpaugh, R.K., Rhim, A.D., Reichert, M., Stanger, B.Z., Furth, E.E., Sepulveda, A.R., Yuan, C.X., Won, K.J., et al. (2013). An iPSC line from human pancreatic ductal adenocarcinoma undergoes early to invasive stages of pancreatic cancer progression. Cell Rep. 3, 2088–2099. Köhler, A., and Hurt, E. (2010). Gene regulation by nucleoporins and links to cancer. Mol. Cell 38, 6–15. Koong, A.C., Mehta, V.K., Le, Q.T., Fisher, G.A., Terris, D.J., Brown, J.M., Bastidas, A.J., and Vierra, M. (2000). Pancreatic tumors show high levels of hypoxia. Int. J. Radiat. Oncol. Biol. Phys. 48, 919–922. Kopp, J.L., von Figura, G., Mayes, E., Liu, F.F., Dubois, C.L., Morris, J.P., 4th, Pan, F.C., Akiyama, H., Wright, C.V., Jensen, K., et al. (2012). Identification of Sox9-dependent acinar-to-ductal reprogramming as the principal mechanism for initiation of pancreatic ductal adenocarcinoma. Cancer Cell 22, 737–750. Lee, J., Sugiyama, T., Liu, Y., Wang, J., Gu, X., Lei, J., Markmann, J.F., Miyazaki, S., Miyazaki, J., Szot, G.L., et al. (2013). Expansion and conversion of human pancreatic ductal cells into insulin-secreting endocrine cells. eLife 2, e00940. Li, X., Nadauld, L., Ootani, A., Corney, D.C., Pai, R.K., Gevaert, O., Cantrell, M.A., Rack, P.G., Neal, J.T., Chan, C.W., et al. (2014). Oncogenic transformation of diverse gastrointestinal tissues in primary organoid culture. Nat. Med. 20, 769–777. Makovitzky, J. (1986). The distribution and localization of the monoclonal antibody-defined antigen 19-9 (CA19-9) in chronic pancreatitis and pancreatic carcinoma. An immunohistochemical study. Virchows Arch. B Cell Pathol. Incl. Mol. Pathol. 51, 535–544. Means, A.L., Meszoely, I.M., Suzuki, K., Miyamoto, Y., Rustgi, A.K., Coffey, R.J., Jr., Wright, C.V.E., Stoffers, D.A., and Leach, S.D. (2005). Pancreatic epithelial plasticity mediated by acinar cell transdifferentiation and generation of nestin-positive intermediates. Development 132, 3767–3776. Morris, J.P., 4th, Cano, D.A., Sekine, S., Wang, S.C., and Hebrok, M. (2010). Beta-catenin blocks Kras-dependent reprogramming of acini into pancreatic cancer precursor lesions in mice. J. Clin. Invest. 120, 508–520. Neesse, A., Frese, K.K., Chan, D.S., Bapiro, T.E., Howat, W.J., Richards, F.M., Ellenrieder, V., Jodrell, D.I., and Tuveson, D.A. (2014). SPARC independent drug delivery and antitumour effects of nab-paclitaxel in genetically engineered mice. Gut 63, 974–983. Olive, K.P., Jacobetz, M.A., Davidson, C.J., Gopinathan, A., McIntyre, D., Honess, D., Madhu, B., Goldgraben, M.A., Caldwell, M.E., Allard, D., et al. (2009). Inhibition of Hedgehog signaling enhances delivery of chemotherapy in a mouse model of pancreatic cancer. Science 324, 1457–1461. Pérez-Mancera, P.A., Guerra, C., Barbacid, M., and Tuveson, D.A. (2012). What we have learned about pancreatic cancer from mouse models. Gastroenterology 142, 1079–1092. Prasad, N.B., Biankin, A.V., Fukushima, N., Maitra, A., Dhara, S., Elkahloun, A.G., Hruban, R.H., Goggins, M., and Leach, S.D. (2005). Gene expression

Rovira, M., Scott, S.G., Liss, A.S., Jensen, J., Thayer, S.P., and Leach, S.D. (2010). Isolation and characterization of centroacinar/terminal ductal progenitor cells in adult mouse pancreas. Proc. Natl. Acad. Sci. USA 107, 75–80. Rubio-Viqueira, B., Jimeno, A., Cusatis, G., Zhang, X., Iacobuzio-Donahue, C., Karikari, C., Shi, C., Danenberg, K., Danenberg, P.V., Kuramochi, H., et al. (2006). An in vivo platform for translational drug development in pancreatic cancer. Clin. Cancer Res. 12, 4652–4661. Ryan, D.P., Hong, T.S., and Bardeesy, N. (2014). Pancreatic adenocarcinoma. N. Engl. J. Med. 371, 1039–1049. Sato, T., Vries, R.G., Snippert, H.J., van de Wetering, M., Barker, N., Stange, D.E., van Es, J.H., Abo, A., Kujala, P., Peters, P.J., and Clevers, H. (2009). Single Lgr5 stem cells build crypt-villus structures in vitro without a mesenchymal niche. Nature 459, 262–265. Sato, T., Stange, D.E., Ferrante, M., Vries, R.G., Van Es, J.H., Van den Brink, S., Van Houdt, W.J., Pronk, A., Van Gorp, J., Siersema, P.D., and Clevers, H. (2011). Long-term expansion of epithelial organoids from human colon, adenoma, adenocarcinoma, and Barrett’s epithelium. Gastroenterology 141, 1762–1772. Sawey, E.T., Johnson, J.A., and Crawford, H.C. (2007). Matrix metalloproteinase 7 controls pancreatic acinar cell transdifferentiation by activating the Notch signaling pathway. Proc. Natl. Acad. Sci. USA 104, 19327– 19332. Seaberg, R.M., Smukler, S.R., Kieffer, T.J., Enikolopov, G., Asghar, Z., Wheeler, M.B., Korbutt, G., and van der Kooy, D. (2004). Clonal identification of multipotent precursors from adult mouse pancreas that generate neural and pancreatic lineages. Nat. Biotechnol. 22, 1115–1124. Shain, A.H., Salari, K., Giacomini, C.P., and Pollack, J.R. (2013). Integrative genomic and functional profiling of the pancreatic cancer genome. BMC Genomics 14, 624. Sharma, S.V., Haber, D.A., and Settleman, J. (2010). Cell line-based platforms to evaluate the therapeutic efficacy of candidate anticancer agents. Nat. Rev. Cancer 10, 241–253. Siegel, R., Naishadham, D., and Jemal, A. (2013). Cancer statistics, 2013. CA Cancer J. Clin. 63, 11–30. Subramanian, A., Tamayo, P., Mootha, V.K., Mukherjee, S., Ebert, B.L., Gillette, M.A., Paulovich, A., Pomeroy, S.L., Golub, T.R., Lander, E.S., and Mesirov, J.P. (2005). Gene set enrichment analysis: a knowledge-based approach for interpreting genome-wide expression profiles. Proc. Natl. Acad. Sci. USA 102, 15545–15550. Villarroel, M.C., Rajeshkumar, N.V., Garrido-Laguna, I., De Jesus-Acosta, A., Jones, S., Maitra, A., Hruban, R.H., Eshleman, J.R., Klein, A., Laheru, D., et al. (2011). Personalizing cancer treatment in the age of global genomic analyses: PALB2 gene mutations and the response to DNA damaging agents in pancreatic cancer. Mol. Cancer Ther. 10, 3–8.

Cell 160, 324–338, January 15, 2015 ª2015 Elsevier Inc. 337

von Figura, G., Fukuda, A., Roy, N., Liku, M.E., Morris Iv, J.P., Kim, G.E., Russ, H.A., Firpo, M.A., Mulvihill, S.J., Dawson, D.W., et al. (2014). The chromatin regulator Brg1 suppresses formation of intraductal papillary mucinous neoplasm and pancreatic ductal adenocarcinoma. Nat. Cell Biol. 16, 255–267.

Ying, H., Kimmelman, A.C., Lyssiotis, C.A., Hua, S., Chu, G.C., Fletcher-Sananikone, E., Locasale, J.W., Son, J., Zhang, H., Coloff, J.L., et al. (2012). Oncogenic Kras maintains pancreatic tumors through regulation of anabolic glucose metabolism. Cell 149, 656–670.

Wiese, S., Reidegeld, K.A., Meyer, H.E., and Warscheid, B. (2007). Protein labeling by iTRAQ: a new tool for quantitative mass spectrometry in proteome research. Proteomics 7, 340–350.

Zhao, L.L., Zhang, T., Liu, B.R., Liu, T.F., Tao, N., and Zhuang, L.W. (2014). Construction of pancreatic cancer double-factor regulatory network based on chip data on the transcriptional level. Mol. Biol. Rep. 41, 2875–2883.

338 Cell 160, 324–338, January 15, 2015 ª2015 Elsevier Inc.

Go figure With author-narrated animations

Introducing Figure360, an author-narrated animation of select figures in Cell Press primary research and review journals. A short, digestible synopsis puts the figure in context and helps you zoom in on the most important take-home message in a matter of minutes. Why go it alone when the author can help you figure it out in less than half the time? Check it out at www.cell.com/figure360

Cancer Cell

Article Glut3 Addiction Is a Druggable Vulnerability for a Molecularly Defined Subpopulation of Glioblastoma €rv,2 Valérie Dutoit,3 Kathryn Elliott,1 Tami von Schalscha,1 Maria F. Camargo,1 Érika Cosset,1,9,* Sten Ilmja Alexander Reiss,1 Toshiro Moroishi,4,10 Laetitia Seguin,1 German Gomez,1 Jung-Soon Moo,4,11 Olivier Preynat-Seauve,5 Karl-Heinz Krause,2 Hervé Chneiweiss,6 Jann N. Sarkaria,7 Kun-Liang Guan,4 Pierre-Yves Dietrich,3 Sara M. Weis,1 Paul S. Mischel,8 and David A. Cheresh1,12,* 1Department of Pathology, Moores Cancer Center, Sanford Consortium for Regenerative Medicine, University of California San Diego, La Jolla, CA 92037, USA 2Department of Pathology and Immunology, Medical School, University of Geneva, Geneva, Vital-IT Group, SIB Swiss Institute of Bioinformatics, Lausanne, Switzerland 3Laboratory of Tumor Immunology, Centre of Oncology, Geneva University Hospitals, University of Geneva, Geneva, Switzerland 4Department of Pharmacology, Moores Cancer Center, Sanford Consortium for Regenerative Medicine, University of California San Diego, La Jolla, CA 92037, USA 5Division of Hematology, Departments of Internal Medicine and Human Protein Science, Faculty of Medicine, University of Geneva, Geneva, Switzerland 6INSERM U1310, Sorbonne Universités, Paris, France 7Department of Radiation Oncology, Mayo Clinic, Rochester, MN 55905, USA 8Ludwig Institute for Cancer Research, Department of Pathology, Moores Cancer Center, University of California San Diego, La Jolla, CA 92093, USA 9Present address: Department of Internal Medicine Specialities, Centre of Translational Onco-Hematology, University of Geneva, Geneva, Switzerland 10Present address: Department of Molecular Enzymology, Faculty of Life Sciences, Kumamoto University, Kumamoto 860-8556, Japan 11Present address: School of Medicine, Ajou University, Yeongtong-gu, Suwon 16499, Korea 12Lead Contact *Correspondence: erika.cosset@unige.ch (É.C.), dcheresh@ucsd.edu (D.A.C.) https://doi.org/10.1016/j.ccell.2017.10.016

SUMMARY

While molecular subtypes of glioblastoma (GBM) are defined using gene expression and mutation profiles, we identify a unique subpopulation based on addiction to the high-affinity glucose transporter, Glut3. Although Glut3 is a known driver of a cancer stem cell phenotype, direct targeting is complicated by its expression in neurons. Using established GBM lines and patient-derived stem cells, we identify a subset of tumors within the ‘‘proneural’’ and ‘‘classical’’ subtypes that are addicted to aberrant signaling from integrin avb3, which activates a PAK4-YAP/TAZ signaling axis to enhance Glut3 expression. This defined subpopulation of GBM is highly sensitive to agents that disrupt this pathway, including the integrin antagonist cilengitide, providing a targeted therapeutic strategy for this unique subset of GBM tumors.

INTRODUCTION Glioblastoma multiforme (GBM) represents high-grade gliomas and remains the most frequent and deadliest primary brain tumor

in adults. Despite major research efforts and some clinical progress, GBM ultimately becomes resistant to all current forms of treatment, helping to explain why the overall survival rate has not dramatically changed over the past 20 years (Stupp et al.,

Significance While GBM tumors are highly aggressive and therapy resistant, individual tumors achieve this state via distinct molecular pathways. Here, we define a unique biological subpopulation addicted to an integrin avb3-mediated pathway that enhances glucose uptake, making tumors highly sensitive to a variety of agents that disrupt this advantage. Interestingly, avb3 expression alone is not sufficient to define this population, as only a subset of avb3-expressing GBM tumors are addicted to this pathway. Our findings may explain why the integrin antagonist cilengitide in a clinical trial had a benefit in some patients, but not others. By revealing a direct link between aberrant integrin expression and altered glucose metabolism, this work identifies a context-dependent druggable vulnerability that can be exploited for GBM therapy.

856 Cancer Cell 32, 856–868, December 11, 2017 Published by Elsevier Inc.

2005). By identifying distinct gene expression profiles, GBMs have been stratified by various gene signature profiles into four molecular subtypes (classical, neural, proneural, and mesenchymal) with specific driver mutations, prognoses, and response to therapy (Brennan et al., 2013; Freije et al., 2004; Noushmehr et al., 2010; Nutt et al., 2003; Phillips et al., 2006; Verhaak et al., 2010). However, this advance in knowledge has yet to reveal new druggable targets and development of new therapeutic strategies to have an impact on disease progression and/or outcome. GBMs typically contain cancer stem cells (CSCs) that are associated with both tumor progression and resistance to therapeutic intervention (Lathia et al., 2015). GBM CSCs not only possess self-renewing and tumor-initiating properties but they are able to survive in a nutrient-deficient microenvironment, giving them a particular advantage in the brain. In fact, Flavahan and colleagues revealed that CSCs thrive in part by upregulating the high-affinity glucose transporter Glut3, enabling these cells to survive glucose deprivation (Flavahan et al., 2013). Understanding how Glut3 expression is regulated or how to target it therapeutically would therefore provide an opportunity to attack the most aggressive and drug-resistant cells within the tumor. Integrins are ab heterodimers composed of an extracellular domain, transmembrane domain, and a short cytoplasmic tail. Noncovalent association between ab subunits defines their specificity for particular components of the extracellular matrix, such as vitronectin, fibronectin, or laminin (Desgrosellier and Cheresh, 2010; Weis and Cheresh, 2011). By modulating cellmatrix adhesion, integrins affect diverse aspects of cancer cell behavior, including invasion, proliferation, survival, and the promotion of angiogenesis (Desgrosellier and Cheresh, 2010). In GBM, expression of avb3 and its ligand vitronectin are both linked to tumor progression and invasive behavior at the tumor margin in the brain of patients with GBM (Gladson and Cheresh, 1991). This prompted development of cilengitide, a cyclic peptide antagonist capable of targeting the ligand binding site of avb3. Despite encouraging phase I/II results showing a durable response to cilengitide for some patients (Nabors et al., 2007; Reardon et al., 2008), phase III CENTRIC and phase II CORE trials failed to meet overall survival endpoints (Stupp et al., 2014). In a follow-up study, immunohistological analysis of tissues obtained during the CORE trial revealed that higher avb3 levels were associated with improved survival in patients treated with cilengitide (Weller et al., 2016). Because this was not the case for the CENTRIC trial, it is still not clear how to identify patients who may benefit from this drug. Here, we consider how to identify those GBMs that are particularly sensitive to avb3 antagonists, including cilengitide. RESULTS Integrin b3 mRNA Expression Correlates with Poor Survival and Expression of Genes Involved in Glucose Metabolism We previously reported that expression of integrin avb3 increases with glioma grade (Gladson and Cheresh, 1991). To further investigate the clinical relevance of integrin expression in gliomas, we considered the correlation between integrin expression and patient survival for several datasets containing

both low-grade gliomas (LGG) and GBM samples. Our analysis reveals ITGB3 (b3) as the only b subunit whose mRNA expression correlates with poor survival in all gliomas (low and high grades) for both the TCGA GBM-LGG dataset (p = 0.03) (Figure 1A) and the Freije dataset (Freije et al., 2004) (p < 0.0001) (Figure 1B). Considering that expression of the integrin b subunit is a rate-limiting determinant of integrin heterodimer formation (Cheresh, 1987), and because b3 pairs exclusively with the av subunit in GBM cells, this finding is consistent with our previous report of enriched integrin avb3 protein expression in GBM compared with low-grade astroglial-derived tumors (Gladson and Cheresh, 1991). Taken together, Kaplan-Meier curves from multiple datasets consistently show ITGB3 to be a strong prognostic factor associated with poor survival (Table S1). By generating a hierarchical cluster and stratifying patients into two groups according to median survival, we identify a b3high subset of TCGA GBM-LGG samples within the shortersurvival group (Figure 1A). We reasoned that understanding how integrin b3 contributes to the aggressive phenotype for this subpopulation would enable the design of a targeted therapy approach to exploit the vulnerabilities of this subset. To consider how high integrin b3 expression may lead to poor survival in GBM, we compared gene expression profiles between b3high versus b3low samples in GBM patients from the Freije dataset. We find genes involved in glucose metabolism (ALDOC, PFKM, and SLC2A3 [Glut3]) as one of the main family of genes correlated with b3 expression (Figure 1C and Tables 1 and S2). As for integrin b3, Kaplan-Meier analysis indicates that poor survival correlates with expression of SLC2A3 (Glut3) (p = 0.0021), ALDOC (p = 0.0065), and PFKM (p = 0.00032) in glioma patients (Figure 1D). To further validate the clinical relevance of this profile, we generated Kaplan-Meier curves from the ‘‘Lee’’ and ‘‘TCGA’’ datasets. Whereas ALDOC and PFKM do not consistently correlate with patient outcome, we find that SLC2A3 (Glut3) expression tracks with poor survival for all datasets (Table S3). Moreover, analysis of multiple datasets using a multiple experiment matrix (MEM) reveals ITGB3 and SLC2A3 (Glut3) as co-expressed genes not only in GBM (Figure 1E) but also in other cancer types (Figure S1A). Targeting b3 Strongly Inhibits Glut3 Expression to Decrease Cell Survival and Anchorage Independence We next considered whether the ability of integrin avb3 to promote an aggressive GBM phenotype might be linked to Glut3-mediated cell survival and glucose uptake. For three established GBM cell lines, short hairpin RNA (shRNA)-mediated knockdown of integrin b3 strongly inhibits Glut3 expression (Figures 2A and 2B), glucose uptake (Figure 2C), and lactate production (Figure 2D). In fact, the effect of b3 knockdown on cell survival is accentuated under low glucose conditions (Figure S2A). Moreover, we observed that knockdown of either b3 or Glut3 decreases anchorage independence (Figure 2E) and tumorsphere formation (Figure 2F), properties associated with CSCs. To determine whether highly efficient glucose uptake provides a competitive advantage for b3+ cells, we co-cultured b3+ (GFP ) and b3 (GFP+) cells under standard (4.5 g/L) or low (0.4 g/L) glucose conditions and monitored their ratio using flow cytometry. Indeed, there are significantly more viable b3+ cells present after 1 week of glucose restriction compared with cells for which Cancer Cell 32, 856–868, December 11, 2017 857

Shorter survival

Longer survival

TCGA GBM-LGG dataset

Figure 1. b3 Levels Correlate with Poor Survival and Expression of Genes Involved in Glucose Metabolism

Correlation of gene expression with glioma survival

Risk p-value ITGB5 0.61 ITGB2 0.59 ITGB4 0.06 ITGB8 0.62 ITGB1 0.48 ITGB7 0.84 ITGB6 0.15 ITGB3 0.03 *

Colors & Histogram

Values

β3high subpopulation within the shorter-survival group

Freije dataset ITGB3 (β3)

Probability of Survival

0.0 0.2 0.4 0.6 0.8 1.0

C immune system process metabolic process multicellular organismal process developmental process response to stimulus biological regulation localization cellular process cellular organization/biogenesis

Low High

p< 0.0001 0

* 0

500

1000

1500

2000

2500

20 40 60 Percentage of Enrichment

Time (days) Freije dataset

Freije dataset

SLC2A3 (Glut3)

0.0 0.2 0.4 0.6 0.8 1.0

Probability of Survival

500

Freije dataset

ALDOC

PFKM

Low

High

p = 0.0021 0

p = 0.0065 1000

1500

2000

2500 0

500

p = 0.00032 1000

1500

2000

2500 0

500

Time (days)

SLC2A3 (Glut3) ITGB3 (β3)

either b3 or Glut3 had been knocked down (Figure 2G). More importantly, knockdown of either b3 or Glut3 significantly delays the orthotopic growth of GBM tumors in mice (Figures 2H and S2B). Collectively, these results indicate that b3 and Glut3 promote the survival of GBM cells and their tumorigenic capacity in the brain. We previously reported that knockdown of b3 induced a senescent phenotype in GBM cells (Franovic et al., 2015). Here, we show that Glut3 knockdown also induces multiple markers of senescence in vitro, including b-galactosidase (SA-b-gal) activity, G0/G1 cell-cycle arrest, and gH2AX (Figures 2I, 2J, and S2C–S2E). In vivo, cells with knockdown of either b3 or Glut3 show increased SA-b-galactosidase activity within subcutaneous xenografts (Figure 2K). In contrast, knockdown of the Glut1 or Glut6 glucose transporters does not induce a senescent phenotype (Figure S2F). We therefore asked whether ectopic expression of Glut3 is sufficient to drive GBM growth in the 858 Cancer Cell 32, 856–868, December 11, 2017

(A) Hierarchical clustering of integrin b subunit expression correlated to a risk score predicting patient survival for the TCGA GBM-LGG dataset. *p < 0.05. (B) Kaplan-Meier analysis of the Freije dataset for ITGB3 (b3) expression (n = 42 b3 low, n = 43 b3 high; p < 0.0001). Low, low-risk group; high, highrisk group. (C) Functional annotation clustering (series GEO: GSE4412, Freije dataset) of gene set enrichment analysis based on b3high versus b3low expression. Graph shows the percent enrichment for each family of genes. *p < 0.05. (D) Kaplan-Meier analysis of the Freije dataset for SLC2A3 (Glut3), ALDOC, and PFKM expression. SLC2A3 (p = 0.002); ALDOC (p = 0.0065); PFKM (p = 0.0003). (E) b3 and Glut3 expression are significantly correlated across a range of GBM datasets according to the MEM output. See also Figure S1 and Tables S1–S3.

1000

1500

2000

2500

absence of b3. Indeed, ectopic Glut3 ‘‘rescues’’ the effects of b3 knockdown on two- and three-dimensional and growth and prevents the senescent phenotype in vitro and within tumors in vivo (Figures 2K, 2O, S2G, and S2H), suggesting that the regulation of Glut3 expression may largely account for the impact of integrin avb3 on GBM progression.

Integrin avb3 Modulates Glut3 Expression through the PAK4-YAP/ TAZ Axis To understand how integrin avb3 regulates Glut3 expression in GBM cells, we considered transcriptional regulators that correlate with b3 expression. We identified ‘‘cell signaling’’ as an important family of genes associated with b3 expression (Table 1) and found the transcriptional coactivator WWTR1 (WW domain-containing transcription regulator 1, also known as TAZ) as the top transcription factor in our list of genes (Table 1). Along with its paralog Yes-associated protein (YAP), YAP/TAZ affects a wide variety of cellular functions, including epithelial-mesenchymal transition, cell growth, organ development, metabolism, and stress responses (Moroishi et al., 2015). Of note, the Kaplan-Meier curves generated from the Freije (Figure 3A) dataset reveals that WWTR1 (TAZ) expression correlates with poor survival (p = 0.02). Moreover, we find that b3 knockdown leads to a marked decrease of YAP/TAZ expression (Figures 3B and 3C). Consistent with previous reports of Glut3 as a YAP-regulated gene (Wang et al., 2015), we find that YAP/TAZ knockdown decreases Glut3 expression (Figures 3D, 3E, and S3A), and this also induces senescence as demonstrated by SA-b-galactosidase

Table 1. List of Genes Differentially Expressed Based on b3high versus b3low Expression for the Freije Dataset Rank

Gene Names

Rank

Gene Names

Rank

Gene Names

Rank

Gene Names

Rank

Gene Names

THRA

SPATA2

ABAT

Rank 91

BHLHE40

121

CAMK2G

151

LIMCH1

ALDOC

SUSD4

CAMTA1

CYP46A1

122

OGFOD3

152

RAB6B

FXYD6

TCEAL2

BEX4

FAM127A

123

COL6A1

153

ARHGEF4

ITPR3

FAM192A

PIK3R1

ITGA4

124

ICAM1

154

LAMB1

FTO

RAB27A

MAPT

PAAF1

125

RBMS1

155

APC

NRXN1

ABHD10

FDFT1

DZIP3

126

NRP1

156

NFE2L3

ITGB3

GORASP1

TJP2

NRP2

127

GMFB

157

C1ORF61

CLASP2

C14ORF132

RUFY3

TRAPPC2L

128

CPD

158

LOX

AKTIP

NCOA1

NDUFS1

ERBB4

129

MYO18A

159

B3GAT1

WWTR1

KIF5C

HEXB

100

ADD3

130

ABAT

160

TSSC1

THRA

CLCN6

LAMC1

101

ANKRD46

131

LAMA4

161

NRXN1

SHC1

SERPINE1

GTDC1

102

NRP2

132

GNAZ

162

UGGT1

OMG

NDRG2

KDELR3

103

PFN2

133

KCNQ2

163

ICAM1

SVIL

TTYH1

ATP9A

104

BLCAP

134

ZNF189

164

SLC6A1

TSPYL4

ACO2

MAPT

105

NAP1L3

135

TUBB4A

165

SLC2A3

OSMR

DESI1

SC5D

106

PKIA

136

KIF5C

166

HMGCS1

BCAT1

PMAIP1

SLC9A6

107

PFKM

137

CA12

167

CEP68

KIF1B

APBA2

ALDH5A1

108

TPM4

138

ATP8A1

168

BCR

CTNND2

ADGRB3

WEE1

109

NOL12

139

TMEM35B

169

EDEM1

TNFRSF10B

NCAN

SLC22A17

110

MAPT

140

RBMS1

170

ABHD6

IQGAP1

NRXN2

CTIF

111

COL6A1

141

ASRGL1

171

NGRN

GABARAPL2

SLC20A1

RTN3

112

FDFT1

142

PIK3R1

172

DOPEY1

IQGAP1

PRKACB

SDC1

113

MR1

143

MARCKSL1

173

NCOA1

NRXN2

NTM

ADAM22

114

HIP1R

144

THTPA

174

BEX1

WASF3

NUDT3

ADGRE5

115

145

ANXA2P2

175

APC

ITPR3

PLCB1

SCN3A

116

NCALD

146

CHSY1

176

RUNX1

FUT9

MMP14

PTBP2

117

KIF21B

147

COL6A1

177

KCNB1

SHC1

CLIP3

RAB27A

118

SEC24A

148

MAP1A

178

PPP2R2B

CLASP2

VMP1

TNFRSF12A

119

GDF15

149

WASF1

179

LOX

PEA15

ADD1

APBA2

120

GRIA2

150

ACACA

180

PHLPP1

Only the top 180 genes are shown, ranked from 1 to 180, and only genes with adjusted p value <0.05 have been considered for analysis. See also Table S2.

activity (Figure 3F). Furthermore, ectopic expression of YAP can rescue colony-forming ability in b3-knockdown cells (Figures 3G and S3B). Since we recently implicated PAK4 as a mediator of b3 function (Franovic et al., 2015), we considered whether this kinase may also be required for b3-mediated regulation of YAP/TAZ expression. Indeed, inhibition of PAK4 activity using the PAK4 kinase inhibitor PF-03758309 or knockdown of PAK4 expression using shRNA led to a decrease of YAP/TAZ (and Glut3) expression (Figures S3C–S3F and 3H). Moreover, knockdown of PAK4 (like YAP/TAZ) induced markers of senescence, including SA-b-gal and G0/G1 cell-cycle arrest (Figures 3I and 3J). Whereas a critical role for Glut3 in GBM has recently been reported (Flavahan et al., 2013), there have so far been no therapeutic agents capable of targeting its function. By understanding how Glut3 expression is regulated in GBM cells, our findings highlight multiple strategies to therapeutically target this signaling axis in cells that are addicted to Glut3 for survival.

Integrin avb3 Is Required for Glut3 Expression in Patient-Derived Gliomaspheres To further examine the link between b3 and Glut3 in models that reflect the genetic heterogeneity of human glioblastoma, we derived glioblastoma stem cells (GSCs) from 12 GBM patients and confirmed tumorigenicity, multipotency capacity, and expression of stem cell markers (Figures S4A–S4C). For this panel, a third of the GSCs models show high integrin b3 expression (Figure 4A), and this correlates with positive expression of Glut3 (Figure 4A). Similarly, histological analysis of a GBM tissue array confirms that a subset of GBM specimens show high expression of both b3 and Glut3 (Figures S4D and S4E). For the b3-positive GSC models (Ge479, Ge518, and Ge269), knockdown of b3 decreases Glut3 expression (Figures 4B, S4F, and S4G), whereas ectopic expression of b3 in the b3-negative GBM6 model induces both Glut3 and YAP expression (Figure S4H). While only a subset of the GSC panel shows this phenotype, all of the established GBM cell lines examined contain high levels of both Cancer Cell 32, 856–868, December 11, 2017 859

LN229

shβ3

LN18

U87MG

β3

LN229

LN18

Glut3

shβ3

B mRNA expression (fold change vs Ctrl)

U87MG

Protein expression (fold change vs Ctrl)

U87MG 1

LN229

Figure 2. The Impact of Integrin avb3 on GBM Is Attributed to Its Regulation of Glut3 Expression

(A) Immunoblots show expression of indicated proteins for U87MG, LN229, and LN18 GBM cells infected by shRNA control (Ctrl) or *** *** *** ** shb3. Graph shows the fold change of protein *** *** 0.1 *** * expression determined by densitometry Glut3 β3 β3 Glut3 0.1 *** analysis. (ITGB3) (SLC2A3) C1.2 D E (B) mRNA expression was determined by qPCR in shCtrl shβ3 shCtrl shβ3 120 1.2 shCtrl siGlut3 shβ3 U87MG, LN229, and LN18 infected with shRNA 1 1 100 * control (shCtrl) or shb3. * * 0.8 0.8 80 ** (C) Relative glucose uptake in U87MG, LN229, *** *** *** * 0.6 *** 0.6 60 and LN18 cells with b3 knockdown compared ** 0.4 0.4 40 with control (shCtrl). 0.2 0.2 20 (D) Bars represent the relative lactate production 0 0 0 in U87MG and LN229 cells with b3 knockdown U87MG LN229 U87MG LN229 0.4 g/L 0.8 g/L 4.5 g/L compared with control (shCtrl). Glucose concentration F 120 GG H (E) Effect of b3 and Glut3 knockdown on 120 anchorage-independent growth of U87MG under *** 100 100 high (4.5 g/L) or low (0.4 or 0.8 g/L) glucose 80 *** 80 * ** conditions. *** 60 60 ** (F) Effect of b3 and Glut3 knockdown on tumor40 40 sphere formation of U87MG under low glucose 20 20 conditions (0.4 g/L). 0 0 shCtrl shβ3 shGlut3 (G) Flow cytometry was used to quantify b3+ shCtrl shβ3 shGlut3 versus b3 as well as Glut3+ versus Glut3 in a growth competition assay under low glucose J I β-gal+ K G0/G1 S G2/M conditions (0.4 g/L). 1.0 100 * * 11.5 12.2 (H) Effect of b3 and Glut3 knockdown on tumor 16.3 6.9 6.0 0.8 80 8.5 growth in vivo: U87MG shCtrl and U87MG b3 and 0.6 Glut3 shRNA (n = 15 mice per group). 60 * * (I) Graph represents the fold change of b-galac0.4 40 80.5 78.5 71.5 tosidase-positive cells versus the total cell num0.2 20 ber. Inverted microscopy images of acidic senescence-associated b-galactosidase staining 0.0 0 siCtrl siβ3 siGlut3 siGlut3 siCtrl siβ3 in U87MG shCtrl and U87MG b3 and Glut3 shRNA (n = 5 fields counted per group). M N O L 120 120 (J) Cell-cycle analysis showing the percentage of 1.0 ns ns 100 100 cells in G0/G1, S, and G2/M in U87MG cells with 0.8 b3 and Glut3 knockdown. 80 80 0.6 (K) Images show acidic senescence-associated 60 60 b-galactosidase staining, a marker of senes0.4 40 40 cence, in mice implanted with U87MG 0.2 ns 20 20 ns shCtrl, shb3, shGlut3, or shb3 with ectopic 0 0.0 0 expression of Glut3. Scale bars: 100 mm (top shCtrl shCtrl shβ3 shCtrl shβ3 shβ3 Glut3+ Glut3+ Glut3+ left), 25 mm (top right). Arrows show senescent cells. (L) Flow cytometry was used to quantify U87MG shCtrl (GFP ) versus U87MG shb3-Glut3+ (GFP+) in a growth competition assay. (M) Effect of ectopic expression of Glut3 on anchorage-independent growth of U87MG b3 shRNA cells. (N) Graph represents the fold change of b-galactosidase positive cells versus the total cell number. Inverted microscopy images of acidic senescence-associated b-galactosidase staining in U87MG b3 shRNA overexpressing Glut3 compared with U87MG shCtrl (n = 5 fields counted per group). (O) Effect of ectopic expression of Glut3 on tumor growth in vivo: U87MG shCtrl versus U87MG shb3 Glut3+. (n = 15 mice per group). This experiment was performed at the same time as the in vivo experiment shown in Figure 2H. Data are represented as means (n = 3–5) ± SEM (*p < 0.05, **p < 0.01, and ***p < 0.001). See also Figure S2. β3

U87MG colony formation (% of Ctrl)

Relative lactate production

β3

% of U87MG clones

shGlut3

shβ3

shβ3 Glut3+

% Cells per Phase

shCtrl

% of U87MG viable cells

% U87MG spheres

β-gal+ cells (fold increase vs total cells)

% of viable cells

Ctrl

β-gal+ cells (fold increase vs total cells)

Ctrl

Relative glucose uptake

β-actin shRNA: Ctrl β3

avb3 and Glut3 (Figures 2A and S4I), highlighting the inability of cultured cell lines to accurately reflect the heterogeneity of GBM in this context. Patient-Derived Gliomaspheres Show Heterogeneity in Glut3 ‘‘Addiction’’ In contrast to the established GBM cell lines that are uniformly addicted to both avb3 and Glut3, we find that not all of the avb3+/Glut3+ patient-derived GSC models are dependent on glucose and/or Glut3 expression for survival. While the pa860 Cancer Cell 32, 856–868, December 11, 2017

tient-derived cells Ge479 and GBM39 are highly sensitive to glucose deprivation, others (Ge269 and Ge518) show less sensitivity or appear glucose indifferent (Ge738 and GBM6), as demonstrated by their equivalent viability under low or high glucose conditions (Figure 4C). Importantly, Glut3 knockdown decreases the survival of the glucose-addicted GSC models Ge479 and GBM39, while Ge269 and Ge518 are only moderately dependent on glucose and not dependent on Glut3 (Figures 4D and S4G). For the glucose-addicted model, Ge479, b3, and Glut3 knockdown induces the same pattern

Low High

p< 0.002 500

1000 1500 Time (days)

2000

2500 shβ3

U87MG

LN229

LN18

β3 YAP β-actin shRNA: Ctrl

β3

Ctrl

β3 Ctrl

β3

Protein expression (fold change vs Ctrl)

U87MG

LN18

** ** *

0.1 YAP

β3

YAP Glut3 β-actin shRNA: Ctrl

LN229

YT Ctrl

U251

YT Ctrl

Protein expression (fold change vs Ctrl)

U87MG

* Glut3

0.8 0.6

shCtrl

0.4 0.2

100 80 60 40 20 0

J 1.0

% Cells per Phase

0.8

100

0.6 0.4 0.2 0.0

G0/G1

shYAPTAZ

***

9.3

66.75

*** *** *** *** ***

YAP TAZ Glut3 (YAP) (WWTR1) (SLC2A3) shPAK4

G2/M

LN229

0.1

10.9 11.6

20.5

***

TAZ (WWTR1)

U87MG

74.6

LN229

0.1

shCtrl shβ3 YAP+

H 120

shYAPTAZ

*** 0.1

Protein expression (fold change vs Ctrl)

% of U87MG clones

F β-gal+ cells (fold increase vs total cells)

YAP

***

E U251

* *

LN229

YAP (YAP)

0.1

β-gal+ cells (fold increase)

LN229

U87MG

***

shYAPTAZ

shβ3

LN229

mRNA expression (fold change vs Ctrl)

0.0 0.2 0.4 0.6 0.8 1.0

Probability of Survival

Figure 3. b3 Modulates Glut3 Expression through the PAK4-YAP/TAZ Axis

Freije dataset WWTR1 (TAZ)

* * β3

PAK4

* * YAP

Glut3

(A) Kaplan-Meier analysis of the Freije dataset for WWTR1 (TAZ) expression (n = 42 for b3 low and n = 43 for b3 high; p = 0.02). Low, low-risk group; high, high-risk group. (B) Immunoblots show the effect of b3 knockdown on protein expression of YAP and b3. Bars represent the fold change of protein expression determined by densitometry analysis. Data are represented as mean (n = 3–5) ± SEM (*p < 0.05, **p < 0.01, and ***p < 0.001). (C) Graph shows the effect of b3 knockdown on mRNA expression of YAP and TAZ determined by qRT-PCR, displayed as fold change for gene expression normalized to sh control in U87MG (n = 3), LN229 (n = 3), and U251 (n = 2). (D) Immunoblots show the effect of YAP/TAZ knockdown on Glut3 protein expression, and the graph shows the fold increase determined by densitometry analysis. U87MG (n = 3), LN229 (n = 3), and U251 (n = 2). (E) Graph shows the effect of YAP/TAZ knockdown on mRNA expression for SLC2A3 (Glut3), YAP, and WWTR1 (TAZ) determined by qRT-PCR, displayed as fold change of gene expression normalized to sh control in U87MG (n = 3) and LN229 (n = 2). (F) Acidic senescence-associated b-galactosidase staining in U87MG shCtrl versus YAP/TAZ shRNA (n = 3). Arrows show senescent cells. Scale bar, 50 mm. (G) Effect of ectopic expression of YAP on U87MG b3 shRNA on anchorage-independent growth in U87MG (n = 3). (H) Graph shows the fold change of protein expression in U87MG (n = 2) and LN229 (n = 2) determined by densitometry analysis. (I) Acidic senescence-associated b-galactosidase staining in U87MG shCtrl and PAK4 siRNA (n = 3). (J) Cell-cycle analysis showing the percentage of cells in G0/G1, S, and G2/M in U87MG cells with PAK4 siRNA (n = 3). Data are represented as means (n = 2–5) ± SEM (*p < 0.05,**p < 0.01, and ***p < 0.001). See also Figure S3.

0 siCtrl

siPAK4

U87MG

siCtrl

siPAK4 U87MG

of gene expression (increased ALDOC and a trend toward increased HK3), which is in line with the differential gene expression analysis (Figure 4E). The apparent dichotomy in avb3/Glut3 expression versus addiction prompted us to consider how the two groups of GSC models may differ in terms of molecular subtype. Indeed, the Glut3-addicted GSC models Ge479 and GBM39 express genes consistent with a ‘‘proneural-classical’’ GBM subtype (EGFR, GLI1, NES, DLL3, OLIG2), while the Glut3-independent GSC models Ge269 and Ge518 express markers indicating the mesenchymal GBM subtype (CHI3L1 (YKL40), LOX, CD44, and RELB) (Figure S5A). Altogether, our results indicate that within the population of GSCs defined by dual high expression of both avb3 and Glut3, only a subset of these tumors (i.e., those with proneural-classical markers) depend on Glut3 for survival.

The Mesenchymal Subtype of GBM Is Enriched for Glycolytic Genes but Is Insensitive to Antagonists of the avb3/PAK4/YAP/TAZ Pathway GBM cells avidly take up glucose and are highly metabolically active. This particularity has been exploited clinically by positron emission tomography combined with an intravenous injection of 18 F-fluorodeoxy-glucose (FDG), a glucose analog. However, not all GBM subtypes avidly take up FDG, suggesting metabolic heterogeneity, which is not clearly understood. To investigate how avb3 might have an impact on the metabolic landscape of GBM, we performed an enrichment analysis of GBM patients with high versus low expression of genes involved in the glycolytic/gluconeogenesis pathway. As previously reported (Bhat et al., 2011; Mao et al., 2013), we found the mesenchymal subtype to be significantly enriched for several genes involved in the glycolytic pathway, including HK3, LDHA, PFKL, PGK1, Cancer Cell 32, 856–868, December 11, 2017 861

Classical

970.2 Classical

904

Neural

Proneural 898

Mesenchymal

885

Proneural 738

835

Mesenchymal

Proneural

Mesenchymal

Classical 688

518

479

269

Proneural 6

Classical

Ge479

Molecular subtype

YAP Glut3 β-actin shRNA: Ctrl

Patient-derived stem cell models

β3

β-actin

cell viability (% vs 2.5g/L of glucose)

8 GSCs β3/Glut3/TAZlow 6,688,738,835,885,898,904,970.2

Ctrl

Glucose 120

2.5g/l

D ns

100 80

***

60 40

***

*** **

Ctrl

β3

Ge518

Ge269

** 0.1

*** ***

***

* ***

***

Glut3

β3

2.0

0.45g/l

β3

0.01

Cell viability (fold vs. Ctrl)

4 GSCs β3/Glut3/TAZhigh 518,269,479,39

Protein expression (fold vs. Ctrl)

TAZ

β3

Ge479 αvβ3 pathway markers

Glut3

Figure 4. Integrin avb3 Is Required for Glut3 Expression in Patient-Derived Glioma Spheres that Show Heterogeneity in Glut3 ‘‘Addiction’’

Ge269

Ge518

β3

siCtrl # 1

YAP

siGlut3 # 1

siGlut3 # 2 ns

ns 1.5

ns ns

1.0

* *

0.5 20

(A) Representative immunoblots show expression of b3, Glut3, and TAZ in GSCs with a schematic representing the decision tree for selecting GSCs based on b3/Glut3 expression (n = 2). (B) Immunoblots show the effect of b3 knockdown on expression of indicated proteins in Ge479 (n = 3). Graph represents the fold change of protein expression relative to sh control determined by densitometry analysis. (C) Effect of glucose concentration on cell viability measured by CellTiter-Glo in GSCs (n = 3–5). (D) Effect of Glut3 knockdown on cell viability measured by CellTiter-Glo in GSCs (n = 3–4). (E) Expression of glycolytic, pentose phosphate, and mitochondrial oxidative phosphorylation (OXPHOS)-related genes were determined by qRT-PCR after b3 or Glut3 knockdown in Ge479 (n = 3). Bars show the fold change of gene expression normalized to sh control. Data are represented as means (n = 2–5) ± SEM (*p < 0.05, **p < 0.01, and ***p < 0.001). See also Figure S4.

* 0.0

0 Ge479 GBM39 Ge269 Ge518 Ge738 GBM6

Ge479

GBM39

Addicted

Ge269

Ge518

Non-addicted

We then considered how the Glut3 addiction status of a given tumor might 4 be predicted using molecular profiling. Ge479 To do this, we identified samples from siGlut3 siβ3 * the Freije dataset with high expression * of Glut3 by comparing gene expression 1 profiles of Glut3high versus Glut3low samples in GBM patients. For the Glut3high * subset, we asked which genes tracked * with Glut3 in terms of patient survival. * * 0.25 This generated a list of Glut3/survivalGlycolysis Pathway Pentose Phosphate Pathway OXPHOS associated genes predicted to identify the Glut3-addicted phenotype (Table S4). As a validation, we asked if this profile could differentiate between the GLUT3, GLUT5, and GLUT10, with a trend toward enrichment for proneural/classical Glut3-addicted GSC models (GBM39 and HK2, ENO1, PFKM, GAPDH, and ALDOA, and significantly low Ge479) and the mesenchymal Glut3-non-addicted GSC models expression of ALDOC, PFKP, and LDHB (Figures 5A and S5B). (Ge269 and Ge518). Out of a 96-gene panel, a 19-gene subset Kaplan-Meier analysis confirms the clinical relevance for several (Figure 5D and Table S4) allowed us to distinguish between of these genes (Figures 5B and S5C–S5F). Despite the highly mesenchymal (LOX, THBS1, and DCN) and proneural/classical glycolytic expression signature of the mesenchymal subtype in subtypes (DLL3, OLIG2, CDK17, and MAP2). Therefore, assessthe Freije dataset and the enrichment of b3, Glut3, YAP, and ing GBM molecular subtype or using this gene expression panel TAZ (Figure 5C), we find that the Glut3 nonaddicted models could provide a means to identify which avb3/Glut3high tumors show a mesenchymal-like signature (Figure S5A). It is possible are addicted to Glut3. We hypothesized that avb3/Glut3high, Glut3-addicted GSCs that the abundance of glycolytic genes can compensate for the role of Glut3, thus explaining its nonessential role in tumors of (GBM39 and Ge479) would be highly sensitive to agents that this subtype. Alternatively, the mesenchymal subtype may disrupt the b3-PAK4-YAP/TAZ axis. To test this, we evaluated depend on metabolic pathways, other than the glycolytic GSC survival in the presence of the av integrin antagonists cilenpathway, for survival. Together, these findings suggest that gitide (a cyclic peptide that inhibits av integrins) or LM609 (a agents targeting the avb3/PAK4/YAP/TAZ/Glut3 signaling axis function-blocking monoclonal antibody specific for human but would be most effective for avb3/Glut3high tumors that show not rodent integrin avb3) (Figure 6A). Indeed, we found that markers defining a proneural/classical, but not mesenchymal, sensitivity to integrin blockade does not exclusively depend on avb3/Glut3 expression or mutation status but rather on Glut3 subtype. mRNA expression (fold vs. Ctrl)

862 Cancer Cell 32, 856–868, December 11, 2017

Figure 5. The Mesenchymal Subtype of GBM Is Enriched for Genes Involved in the Glycolytic Pathway and Correspond to a Glut3 Nonaddicted Genetic Signature

D Freije dataset

Mesenchymal

Frequency

Other sub

(A) Enrichment analysis of glycolytic genes for the Freije dataset. Compared with other subtypes (Other sub), the mesenchymal subtype showed high expression of SLC2A3 (Glut3), HK3, PFKP, PGK1, LDHA, SLC2A5 (Glut5), and SLC2A10 (Glut10), and no or low expression of LDHB, PFKP, and ALDOC. (B) Kaplan-Meier analysis of the Freije dataset for PGK1 expression (n = 42 for b3 low and n = 43 for b3 high; p = 0.00000007). (C) Enrichment analysis for ITGB3 (b3), SLC2A3 (Glut3, also found in Figure 5A), YAP and WWTR1 (TAZ). Other sub = Other subtypes. (D) Glut3-addicted versus Glut3-non-addicted samples are identified using 96 signature genes. mRNA was determined by qRT-PCR (n = 2) and Bio-Rad software has been used for analysis. Only the most significant genes are shown. See also Figure S5 and Table S7.

20 15 10 5 0

High expression

Low expression

Freije dataset

B Probability of Survival

0.0 0.2 0.4 0.6 0.8 1.0

PGK1 Low High

p< 0.0001 0

500

1000

1500

2000

2500

Time (days)

Freije dataset

Frequency

Mesenchymal

Color key

Other sub

15 10 5 0

addiction status, which appears to be linked to a proneural/classical-like subtype (Table S5). In contrast, GSC with low b3/Glut3 expression (Ge738, GBM6, Ge904, Ge970.2, Ge835, and Ge885) consistently show either a moderate or a significant enhancement of viability when treated with cilengitide or LM609 (Figure 6A). Similar to blockade of avb3 directly, inhibitors of YAP or PAK4 reduce in vitro viability of the Glut3-addicted proneural-like Ge479 GSC but not the Glut3-non-addicted mesenchymal-like Ge518 model (Figure 6B). To validate our hypothesis and test the ability of our signature to predict sensitivity to avb3 antagonists, we analyzed the available gene expression data for 41 models from the Mayo Clinic Brain Tumor Patient-Derived Xenograft National Resource. Based on their expression of genes associated with the Glut3-addicted versus -nonaddicted signature we generated, we predicted that eight of the models ( 20%) should be sensitive based on their high expression of b3/Glut3 and the Glut3-addicted signature. We therefore obtained three models predicted to be addicted, two nonaddicted, and two with b3/Glut3-low to directly test sensitivity to the avb3 antagonists cilengitide and LM609 (Figures 6C, S6A, and Table S6). Similar to Ge479 and GBM39, we find sensitivity to integrin blockade for GBM14, 85, and 64, which we predicted to be Glut3 addicted (Figures 6C and 6D). Consistently, GBM150 and GBM59 with Glut3-nonaddicted signatures are not affected by the integrin antagonists. Like the other GSC with

Clas Pro Mes Mes

low b3/Glut3 expression, GBM26 and GBM12 show no effect or a moderate enhancement of viability upon cilengitide or LM609 treatment (Figures 6C and 6D). Based on gene expression alone, we were able to predict whether a given GBM PDX model would be sensitive or insensitive to avb3 blockade for this collection of samples. Our success with a modest sample size suggests promise for expanding this strategy to clinical testing. Notably, we also find that ectopic expression of b3 in a GSC model with low b3/Glut3 (GBM6) is not sufficient to sensitize the tumor cells to integrin blockade, while b3 knockdown in the Glut3-addicted Ge479 model abolishes their sensitivity (Figures 6E and 6F). More importantly, systemic treatment with the integrin antagonist cilengitide dramatically prolongs the survival of mice bearing Ge479, but not Ge518, orthotopic tumors (Figures 6G and S6B), further linking Glut3 addiction to a differential selectivity to avb3 blockade in vivo. Altogether, our results identify a molecularly defined subset of GBM tumors that are highly sensitive to inhibition of the b3-PAK4-YAP/TAZ axis by virtue of their Glut3 addiction (Figure 7). Glut3

addicted

Glut3

non-addicted

DISCUSSION Previous studies have linked avb3 expression to GBM progression (Gladson and Cheresh, 1991). Here, we reveal that integrin avb3-mediated activation of PAK4 is required for Glut3 expression in GBM cells, which in some patients leads to Glut3 addiction and sensitivity to avb3 antagonists. Although all established GBM cell lines we examined express avb3 as a biomarker predicting both Glut3 addiction and sensitivity to inhibitors of avb3 integrin, PAK4, or YAP/TAZ, we find this holds true for only a subset of patient-derived gliomasphere models that may more accurately represent the genetic heterogeneity of GBM. Cancer Cell 32, 856â&#x20AC;&#x201C;868, December 11, 2017 863

2.0

Addicted

Non-addicted

β3 / Glut3 high

β3 / Glut3 low

B β3 / Glut3 low

479

904

970.2

738

269

518

835

885

ProN

Clas

ProN

Clas

ProN

Mes

*** *

1.0

* 0.5

***

Cell viability (% vs Ctrl)

Addicted

Non-addicted

Ge479

Ge518

1 0.8

ns ns

0.6 0.4

PAK4 inhibitor (50nM)

0.2

Cilengitide (0.5 μM)

YAP inhibitor (1μM)

LM609 (10 μg/ml)

* * 0.3

β3 / Glut3 high

NGS, GBM specimen Glut3 addicted signature GBM14, 64, 85

β3/Glut3 low GBM26, 12

2.0 Cell viability (% vs Ctrl)

Glut3 non-addicted signature GBM150, 59

2.0 ns

1.0 ns 0.5

Cilengitide (0.5 μM) LM609 (10 μg/ml)

0.3

150

ProN

Clas

Neu

Mes

ns ns

1.0

** *

0.5

ns *

*** ***

Cilengitide (0.5 μM) LM609 (10 μg/ml)

GBM6 Parental pLENTI β3 2.0

ns ns ns

ns 1.0

0.5 Cilengitide (0.5 μM) LM609 (10 μg/ml)

0.3

Addicted Ge479

Non-addicted Ge518

0.0 0

Time (days)

0.8 0.6 0.4 0.2

***

Vehicle Cilengitide

0.0

0.2

0.4

0.6

0.8

Fraction survival

1.0

Fraction survival

Cell viability (% vs Ctrl)

shPAK4 shYAPTAZ ns

β3 / Glut3 high

Clas

0.3

Ge479 shβ3

Parental

β3 / Glut3 low

Cell viability (Cilengitide, LM609)

Non-addicted

Addicted

Time (days)

Figure 6. The Proneural/Classical Subtype of GBM Is Sensitive to Antagonists of avb3, YAP, and PAK4 (A) Effect of LM609 (avb3 function-blocking antibody) and cilengitide (cyclic peptide antagonist of av integrins including avb3 and avb5) on cell viability measured by CellTiter-Glo in GSCs (n = 3–5). (B) Effect of YAP inhibitor (Verteporfin) or PAK4 inhibitor (PF-03758309) on cell viability measured by CellTiter-Glo in GSCs (n = 3–5). (C) Schematic depicting a Mayo Clinic sample request. Samples were requested based on their Glut3-addicted versus nonaddicted signature and then analyzed for cell viability in the presence of cilengitide and LM609. (legend continued on next page)

864 Cancer Cell 32, 856–868, December 11, 2017

Patient-derived GSCs

β3/Glut3low excluded GBM6 Ge738 Ge970.2 Ge885 Ge904 GBM12 GBM26

β3/Glut3high

Glut3addicted

Glut3non-addicted

Proneural and Classical subtype gene signature

Mesenchymal subtype gene signature

Sensitive to αvβ3/PAK4/YAP inhibitors

Insensitive to αvβ3/PAK4/YAP inhibitors

Ge479 GBM39 GBM14 GBM64 GBM85

Ge518 Ge269 Ge835 GBM59 GBM150

Figure 7. Schematic Depicting the Proposed Model of Glut3 Addiction in GBM In contrast to established GBM cell lines that are uniformly b3/Glut3high and Glut3 addicted, patient-derived GSC models show heterogeneity in expression of b3/Glut3. Importantly, the population of b3/Glut3high GSC models can be further separated into Glut3-addicted versus Glut3-nonaddicted subsets based on a gene signature and/or molecular subtype. Only the b3/Glut3high GSC models with proneural/classical subtype markers are sensitive to inhibitors that target elements of the avb3/PAK4/YAP pathway.

Indeed, dual expression of avb3/Glut3 drives addiction to this pathway only for GBM tumors with expression of proneural-classical subtype markers. In contrast, elements of this pathway are not critical for the growth and viability of patient-derived gliomaspheres that show a gene signature consistent with the mesenchymal GBM subtype. Thus, our findings provide a possible explanation for the failure of cilengitide to meet its primary survival endpoint in phase II/III trials, and we predict that patients with avb3-positive proneural-classical subtype tumors might be the best candidates for this drug. Integrin avb3 as a Target for GBM Therapy While a number of integrins contribute to the growth and progression of a wide array of cancers (Desgrosellier and Cheresh, 2010; Desgrosellier et al., 2014; Seguin et al., 2014), we find that avb3 expression is significantly linked to glioblastoma progres-

sion. This is consistent with our previous studies showing avb3 protein expression on the most advanced form of this disease, and most highly expressed on those cells at the tumor margin (Gladson and Cheresh, 1991). However, despite promising activity in phase I (Nabors et al., 2007) and II (Reardon et al., 2008) trials, the av integrin antagonist cilengitide failed to produce a significant overall survival benefit in the phase III CENTRIC trial (Stupp et al., 2014), and further clinical development of cilengitide for GBM has been halted (Mason, 2015). A number of factors may have contributed to the clinical failure of cilengitide, including the stability and pharmacokinetic properties of the drug, its combination with alkylating agents, and use in highly aggressive, drug-resistant cancer (Paolillo et al., 2016). However, in this era of precision medicine, it may be important to select a more focused GBM patient population. While higher levels of avb3 were associated with a modest survival benefit in the phase II CORE trial, avb3 expression did not correlate with outcome for the phase III CENTRIC trial (Weller et al., 2016). These findings, along with our new data, suggest that profiling avb3 expression alone is not sufficient to predict sensitivity to this drug. Instead, we have linked cilengitide sensitivity with the ability of avb3 to drive Glut3 addiction since orthotopic GBM tumors with this dependence show a significant survival benefit compared with tumors not addicted to Glut3. Understanding Why Certain Tumors Are Addicted to avb3, Glucose, and Glut3 Using loss-/gain-of-function approaches, we have determined that integrin avb3 is required for expression of the high-affinity glucose transporter, Glut3, in a PAK4 and YAP/TAZ-dependent manner. In turn, Glut3 appears to be a critical mediator of avb3 addiction in GBM, as ectopic Glut3 expression can completely rescue the orthotopic tumor growth capacity of b3-knockdown cells by allowing them to avoid senescence. While normal astrocytes do not express Glut3, its expression level correlates to astrocytoma grade (Boado et al., 1994). Previous studies have reported a correlation between glucose level/uptake and poor survival (Patronas et al., 1985), and Flavahan and colleagues reported that brain-tumor-initiating cells express Glut3, allowing them to outcompete nontumor cells for glucose within the glucose-limited tumor environment (Flavahan et al., 2013). Recently, Birsoy and collaborators reported that certain glucose-sensitive cell lines do not increase oxygen consumption upon glucose limitation, and gene expression analysis revealed that these lines have low Glut3 and Glut1 expression (Birsoy et al., 2014). A recent single-cell RNA-seq study highlighted the strong heterogeneity in GBM specimens that was not previously well appreciated (Patel et al., 2014). Indeed, among all five tumors analyzed, the authors have shown individual cells

(D) Effect of LM609 (avb3 function-blocking antibody) and cilengitide (cyclic peptide antagonist of av integrins including avb3 and avb5) on Mayo Clinic GSC cell viability measured by CellTiter-Glo in GSCs (n = 3–5, except n = 2 for GBM150 and GBM85). (E) Effect of LM609 (avb3 function-blocking antibody) and cilengitide (cyclic peptide antagonist of av integrins including avb3 and avb5) on the cell viability of Ge479 knockdown for b3, PAK4, and YAP/TAZ measured by CellTiter-Glo in GSCs (n = 3–5). For Ge479 parental, the same data are displayed Figure 6A. (F) Effect of LM609 (avb3 function-blocking antibody) and cilengitide (cyclic peptide antagonist of av integrins including avb3 and avb5) on the cell viability of GBM6 with ectopic expression of b3 measured by CellTiter-Glo in GSCs (n = 3–5). For GBM6 parental, the same data are displayed Figure 6A. (G) Effect of cilengitide on tumor growth. Mice bearing orthotopic Ge518 (Glut3-nonaddicted) and Ge479 (Glut3-addicted) brain tumors were treated with vehicle or cilengitide (25mg/kg; 8 mice per group). Data are represented as means (n = 3–5) ± SEM (*p < 0.05, **p < 0.01, and ***p < 0.001). See also Figure S6.

Cancer Cell 32, 856–868, December 11, 2017 865

corresponding to different GBM subtypes. Together, these studies suggest a complicated heterogeneity and metabolic landscape among individual GBM tumors that may not only explain clinical trial failures but also highlight the need to better understand GBM heterogeneity in order to design appropriate therapeutic regimens. Furthermore, the impact of intratumoral heterogeneity for ITGB3 expression on clinical outcomes represents a potential limitation of our study, as tumors with high overall ITGB3 expression may contain a subpopulation of cells with low ITGB3 and GLUT3. Despite the functional advantages offered by Glut3 expression, we find that only a subpopulation of our patient-derived GSC models actually depends on glucose/Glut3 for survival. In contrast, all long-term established GBM cultured cell lines express a high level of Glut3 and are addicted to this transporter for survival. As such, these well-established GBM cell lines may somehow enrich for this phenotype, providing a poor reflection of its frequency within patient tumors. The fact that only 15% of our patient-derived GSC models appear to be avb3/Glut3 addicted suggests a similar portion of patients might thus be sensitive to avb3 antagonists. In this respect, our study reinforces the need to carefully consider whether biomarkers and drug sensitivity established using cell-based models will relate to the heterogeneity of GBM.

which suppress avb3-mediated Glut3 expression in GBM cells. While the importance of YAP/TAZ in GBM aggressiveness has been reported, our new findings provide some insights into its regulation, signaling, and function within a molecularly defined GBM subpopulation. Aside from cilengitide, there are a number of avb3-targeted strategies in development for GBM, including GLPG0187, a small-molecule antagonist of multiple integrins, including avb3, avb5, avb6, and a5b1 (Cirkel et al., 2016), as well as approaches that use RGD peptides for avb3-targeted delivery of radionuclides (Jin et al., 2017), small interfering RNA (He et al., 2017), and chemotherapy-loaded nanoparticles or nanogels (Chen et al., 2017; Fang et al., 2017). Considering that Glut3 addiction is also a feature of GBM CSCs (Flavahan et al., 2013), targeting this phenotype with an avb3 antagonist has the potential to eradicate the most aggressive and drug-resistant subpopulation within the tumor. STAR+METHODS Detailed methods are provided in the online version of this paper and include the following: d d d

Identification of Glucose/Glut3-Addicted Tumors While we are able to determine glucose/Glut3 addiction status using cell viability assays, we can also identify these cells based on a genetic signature. Indeed, we find that avb3-positive glucose/Glut3-addicted versus nonaddicted tumors can be differentiated in terms of a molecular GBM subtype. Specifically, the glucose/Glut3-addicted tumors represent a subpopulation within the proneural and classical subgroups and can be further delineated based on their stem cell behavior. In contrast, a subpopulation of tumors in the mesenchymal group tend to be positive for avb3/Glut3, yet surprisingly are not addicted to Glut3 and remain insensitive to avb3 antagonists. Thus, we estimate that 10%–15% of GBM patients may show very significant responses to agents targeting avb3/Glut3. Indeed, a number of individual patients showed very significant, durable, yet unexplained responses to cilengitide (Nabors et al., 2007; Reardon et al., 2008). In the mesenchymal subtype, we found an abundance of glycolytic genes, and we found that all mesenchymal patient-derived cells were nonaddicted to Glut3. Thus, the role of Glut3 may be negligible when other glycolytic genes are highly expressed. Or, this subtype might be addicted to another glycolytic gene product, as suggested by Mao et al. (2013). At present, it is unclear why certain GBM tumors are, and/or become, addicted to Glut3, while others can circumvent this dependence. Broader Implications for GBM Therapeutics We report that among avb3/Glut3-expressing tumors, only a subpopulation is ‘‘addicted’’ to glucose/Glut3. Not only does this phenotype render them particularly sensitive to avb3 integrin inhibitors (including the av integrin-targeting cyclic peptide cilengitide or the monoclonal avb3 antibody LM609) but we show that such tumors are also sensitive to inhibitors of PAK4 or YAP/TAZ, 866 Cancer Cell 32, 856–868, December 11, 2017

KEY RESOURCES TABLE CONTACT FOR REAGENT AND RESOURCE SHARING EXPERIMENTAL MODEL AND SUBJECT DETAILS B Human GBM Cell Lines and Patient-Derived Models B Orthotopic Brain Tumor Xenograft Model METHOD DETAILS B Cell Culture Experiments B Protein and mRNA Analysis B Analysis of Microarray Data QUANTIFICATION AND STATISITCAL ANALYSIS

SUPPLEMENTAL INFORMATION Supplemental Information includes six figures and eight tables and can be found with this article online at https://doi.org/10.1016/j.ccell.2017.10.016. AUTHOR CONTRIBUTIONS Conceptualization, E.C., P.M., S.M.W., O.P.S., H.C., V.D., P.-Y.D., D.A.C., and K.-H.K.; Methodology, E.C.; Software, S.I.; Validation, E.C., S.I., K.E., and T.V.S.; Formal Analysis, E.C. and S.I.; Investigation, E.C., K.E., T.V.S., A.R., M.F.C., G.G., J.-S.M., K.-L.G, T.M., L.S., and J.N.S.; Resources, V.D., P.-Y.D., J.N.S., and P.M.; Writing – Original Draft, E.C., S.M.W., and D.A.C.; Writing – Review & Editing, E.C., S.M.W., and D.A.C.; Supervision, E.C.; Funding Acquisition, E.C., S.M.W., J.N.S., and D.A.C. ACKNOWLEDGMENTS We thank B. Walsh and M. Hall for technical support. We also thank M. Yebra, M. Gozo, B. Walsh, J. Desgrosellier, T. Rakhshandehroo, J. Wawrzyniak, H. Wettersten, and members of Sarkaria and Chneiweiss lab for helpful discussions and collaboration as well as members of Dietrich and Dutoit lab for collaboration. E.C. was supported by The Fonds National Suisse, D.A.C. was supported by NCI-CA45726. The authors also thank the Mayo SPORE in Brain Cancer (CA108961) for financial support. Received: June 9, 2017 Revised: August 31, 2017 Accepted: October 29, 2017 Published: November 30, 2017

REFERENCES Adler, P., Kolde, R., Kull, M., Tkachenko, A., Peterson, H., Reimand, J., and Vilo, J. (2009). Mining for coexpression across hundreds of datasets using novel rank aggregation and visualization methods. Genome Biol. 10, R139. Aguirre-Gamboa, R., Gomez-Rueda, H., Martinez-Ledesma, E., MartinezTorteya, A., Chacolla-Huaringa, R., Rodriguez-Barrientos, A., Tamez-Pena, J.G., and Trevino, V. (2013). SurvExpress: an online biomarker validation tool and database for cancer gene expression data using survival analysis. PLoS One 8, e74250. Bhat, K.P., Salazar, K.L., Balasubramaniyan, V., Wani, K., Heathcock, L., Hollingsworth, F., James, J.D., Gumin, J., Diefes, K.L., Kim, S.H., et al. (2011). The transcriptional coactivator TAZ regulates mesenchymal differentiation in malignant glioma. Genes Dev. 25, 2594–2609. Birsoy, K., Possemato, R., Lorbeer, F.K., Bayraktar, E.C., Thiru, P., Yucel, B., Wang, T., Chen, W.W., Clish, C.B., and Sabatini, D.M. (2014). Metabolic determinants of cancer cell sensitivity to glucose limitation and biguanides. Nature 508, 108–112.

Gladson, C.L., and Cheresh, D.A. (1991). Glioblastoma expression of vitronectin and the alpha v beta 3 integrin. Adhesion mechanism for transformed glial cells. J. Clin. Invest. 88, 1924–1932. He, S., Cen, B., Liao, L., Wang, Z., Qin, Y., Wu, Z., Liao, W., Zhang, Z., and Ji, A. (2017). A tumor-targeting cRGD-EGFR siRNA conjugate and its anti-tumor effect on glioblastoma in vitro and in vivo. Drug Deliv. 24, 471–481. Jin, Z.H., Furukawa, T., Ohya, T., Degardin, M., Sugyo, A., Tsuji, A.B., Fujibayashi, Y., Zhang, M.R., Higashi, T., Boturyn, D., et al. (2017). 67CuRadiolabeling of a multimeric RGD peptide for alphaVbeta3 integrin-targeted radionuclide therapy: stability, therapeutic efficacy, and safety studies in mice. Nucl. Med. Commun. 38, 347–355. Lathia, J.D., Mack, S.C., Mulkearns-Hubert, E.E., Valentim, C.L., and Rich, J.N. (2015). Cancer stem cells in glioblastoma. Genes Dev. 29, 1203–1217. Lee, Y., Scheck, A.C., Cloughesy, T.F., Lai, A., Dong, J., Farooqi, H.K., Liau, L.M., Horvath, S., Mischel, P.S., and Nelson, S.F. (2008). Gene expression analysis of glioblastomas identifies the major molecular basis for the prognostic benefit of younger age. BMC Med. Genomics 1, 52.

Boado, R.J., Black, K.L., and Pardridge, W.M. (1994). Gene expression of GLUT3 and GLUT1 glucose transporters in human brain tumors. Brain Res. Mol. Brain Res. 27, 51–57.

Mao, P., Joshi, K., Li, J., Kim, S.H., Li, P., Santana-Santos, L., Luthra, S., Chandran, U.R., Benos, P.V., Smith, L., et al. (2013). Mesenchymal glioma stem cells are maintained by activated glycolytic metabolism involving aldehyde dehydrogenase 1A3. Proc. Natl. Acad. Sci. USA 110, 8644–8649.

Brennan, C.W., Verhaak, R.G., McKenna, A., Campos, B., Noushmehr, H., Salama, S.R., Zheng, S., Chakravarty, D., Sanborn, J.Z., Berman, S.H., et al. (2013). The somatic genomic landscape of glioblastoma. Cell 155, 462–477.

Mason, W.P. (2015). End of the road: confounding results of the CORE trial terminate the arduous journey of cilengitide for glioblastoma. Neuro Oncol. 17, 634–635.

Cancer Genome Atlas Research Network. (2008). Comprehensive genomic characterization defines human glioblastoma genes and core pathways. Nature 455, 1061–1068.

Mi, H., Poudel, S., Muruganujan, A., Casagrande, J.T., and Thomas, P.D. (2016). PANTHER version 10: expanded protein families and functions, and analysis tools. Nucleic Acids Res. 44, D336–D342.

Chen, W., Zou, Y., Zhong, Z., and Haag, R. (2017). Cyclo(RGD)-decorated reduction-responsive nanogels mediate targeted chemotherapy of integrin overexpressing human glioblastoma in vivo. Small 13, https://doi.org/10. 1002/smll.201601997.

Moroishi, T., Hansen, C.G., and Guan, K.-L. (2015). The emerging roles of YAP and TAZ in cancer. Nat. Rev. Cancer 15, 73–79.

Cheresh, D.A. (1987). Human endothelial cells synthesize and express an ArgGly-Asp-directed adhesion receptor involved in attachment to fibrinogen and von Willebrand factor. Proc. Natl. Acad. Sci. USA 84, 6471–6475. Cirkel, G.A., Kerklaan, B.M., Vanhoutte, F., Van der Aa, A., Lorenzon, G., Namour, F., Pujuguet, P., Darquenne, S., de Vos, F.Y., et al. (2016). A dose escalating phase I study of GLPG0187, a broad spectrum integrin receptor antagonist, in adult patients with progressive high-grade glioma and other advanced solid malignancies. Invest. New Drugs 34, 184–192. Cosset, E., Petty, T., Dutoit, V., Tirefort, D., Otten-Hernandez, P., Farinelli, L., Dietrich, P.Y., and Preynat-Seauve, O. (2016). Human tissue engineering allows the identification of active miRNA regulators of glioblastoma aggressiveness. Biomaterials 107, 74–87. Desgrosellier, J.S., and Cheresh, D.A. (2010). Integrins in cancer: biological implications and therapeutic opportunities. Nat. Rev. Cancer 10, 9–22. Desgrosellier, J.S., Lesperance, J., Seguin, L., Gozo, M., Kato, S., Franovic, A., Yebra, M., Shattil, S.J., and Cheresh, D.A. (2014). Integrin alphavbeta3 drives slug activation and stemness in the pregnant and neoplastic mammary gland. Dev. Cell 30, 295–308. Fang, Y., Jiang, Y., Zou, Y., Meng, F., Zhang, J., Deng, C., Sun, H., and Zhong, Z. (2017). Targeted glioma chemotherapy by cyclic RGD peptide-functionalized reversibly core-crosslinked multifunctional poly(ethylene glycol)-b-poly (epsilon-caprolactone) micelles. Acta Biomater. 50, 396–406. Flavahan, W.A., Wu, Q., Hitomi, M., Rahim, N., Kim, Y., Sloan, A.E., Weil, R.J., Nakano, I., Sarkaria, J.N., Stringer, B.W., et al. (2013). Brain tumor initiating cells adapt to restricted nutrition through preferential glucose uptake. Nat. Neurosci. 16, 1373–1382. Franovic, A., Elliott, K.C., Seguin, L., Camargo, M.F., Weis, S.M., and Cheresh, D.A. (2015). Glioblastomas require integrin alphavbeta3/PAK4 signaling to escape senescence. Cancer Res. 75, 4466–4473. Freije, W.A., Castro-Vargas, F.E., Fang, Z., Horvath, S., Cloughesy, T., Liau, L.M., Mischel, P.S., and Nelson, S.F. (2004). Gene expression profiling of gliomas strongly predicts survival. Cancer Res. 64, 6503–6510.

Nabors, L.B., Mikkelsen, T., Rosenfeld, S.S., Hochberg, F., Akella, N.S., Fisher, J.D., Cloud, G.A., Zhang, Y., Carson, K., Wittemer, S.M., et al. (2007). Phase I and correlative biology study of cilengitide in patients with recurrent malignant glioma. J. Clin. Oncol. 25, 1651–1657. Noushmehr, H., Weisenberger, D.J., Diefes, K., Phillips, H.S., Pujara, K., Berman, B.P., Pan, F., Pelloski, C.E., Sulman, E.P., Bhat, K.P., et al. (2010). Identification of a CpG island methylator phenotype that defines a distinct subgroup of glioma. Cancer Cell 17, 510–522. Nutt, C.L., Mani, D.R., Betensky, R.A., Tamayo, P., Cairncross, J.G., Ladd, C., Pohl, U., Hartmann, C., McLaughlin, M.E., Batchelor, T.T., et al. (2003). Gene expression-based classification of malignant gliomas correlates better with survival than histological classification. Cancer Res. 63, 1602–1607. Paolillo, M., Serra, M., and Schinelli, S. (2016). Integrins in glioblastoma: still an attractive target? Pharmacol. Res. 113, 55–61. Parsons, D.W., Jones, S., Zhang, X., Lin, J.C., Leary, R.J., Angenendt, P., Mankoo, P., Carter, H., Siu, I.M., Gallia, G.L., et al. (2008). An integrated genomic analysis of human glioblastoma multiforme. Science 321, 1807–1812. Patel, A.P., Tirosh, I., Trombetta, J.J., Shalek, A.K., Gillespie, S.M., Wakimoto, H., Cahill, D.P., Nahed, B.V., Curry, W.T., Martuza, R.L., et al. (2014). Singlecell RNA-seq highlights intratumoral heterogeneity in primary glioblastoma. Science 344, 1396–1401. Patronas, N.J., Di Chiro, G., Kufta, C., Bairamian, D., Kornblith, P.L., Simon, R., and Larson, S.M. (1985). Prediction of survival in glioma patients by means of positron emission tomography. J. Neurosurg. 62, 816–822. Phillips, H.S., Kharbanda, S., Chen, R., Forrest, W.F., Soriano, R.H., Wu, T.D., Misra, A., Nigro, J.M., Colman, H., Soroceanu, L., et al. (2006). Molecular subclasses of high-grade glioma predict prognosis, delineate a pattern of disease progression, and resemble stages in neurogenesis. Cancer Cell 9, 157–173. Reardon, D.A., Fink, K.L., Mikkelsen, T., Cloughesy, T.F., O’Neill, A., Plotkin, S., Glantz, M., Ravin, P., Raizer, J.J., Rich, K.M., et al. (2008). Randomized phase II study of cilengitide, an integrin-targeting arginine-glycine-aspartic acid peptide, in recurrent glioblastoma multiforme. J. Clin. Oncol. 26, 5610–5617.

Cancer Cell 32, 856–868, December 11, 2017 867

Ritchie, M.E., Phipson, B., Wu, D., Hu, Y., Law, C.W., Shi, W., and Smyth, G.K. (2015). Limma powers differential expression analyses for RNA-sequencing and microarray studies. Nucleic Acids Res. 43, e47. Seguin, L., Kato, S., Franovic, A., Camargo, M.F., Lesperance, J., Elliott, K.C., Yebra, M., Mielgo, A., Lowy, A.M., Husain, H., et al. (2014). An integrin beta(3)KRAS-RalB complex drives tumour stemness and resistance to EGFR inhibition. Nat. Cell Biol. 16, 457–468. Stupp, R., Mason, W.P., van den Bent, M.J., Weller, M., Fisher, B., Taphoorn, M.J.B., Belanger, K., Brandes, A.A., Marosi, C., Bogdahn, U., et al. (2005). Radiotherapy plus concomitant and adjuvant temozolomide for glioblastoma. N. Engl. J. Med. 352, 987–996. Stupp, R., Hegi, M.E., Gorlia, T., Erridge, S.C., Perry, J., Hong, Y.K., Aldape, K.D., Lhermitte, B., Pietsch, T., Grujicic, D., et al. (2014). Cilengitide combined with standard treatment for patients with newly diagnosed glioblastoma with methylated MGMT promoter (CENTRIC EORTC 26071-22072 study): a multicentre, randomised, open-label, phase 3 trial. Lancet Oncol. 15, 1100–1108. Sun, L., Hui, A.M., Su, Q., Vortmeyer, A., Kotliarov, Y., Pastorino, S., Passaniti, A., Menon, J., Walling, J., Bailey, R., et al. (2006). Neuronal and gli-

868 Cancer Cell 32, 856–868, December 11, 2017

oma-derived stem cell factor induces angiogenesis within the brain. Cancer Cell 9, 287–300. Verhaak, R.G., Hoadley, K.A., Purdom, E., Wang, V., Qi, Y., Wilkerson, M.D., Miller, C.R., Ding, L., Golub, T., Mesirov, J.P., et al. (2010). Integrated genomic analysis identifies clinically relevant subtypes of glioblastoma characterized by abnormalities in PDGFRA, IDH1, EGFR, and NF1. Cancer Cell 17, 98–110. Wang, W., Xiao, Z.-D., Li, X., Aziz, K.E., Gan, B., Johnson, R.L., and Chen, J. (2015). AMPK modulates Hippo pathway activity to regulate energy homeostasis. Nat. Cell Biol. 17, 490–499. Weis, S.M., and Cheresh, D.A. (2011). Tumor angiogenesis: molecular pathways and therapeutic targets. Nat. Med. 17, 1359–1370. Weller, M., Nabors, L.B., Gorlia, T., Leske, H., Rushing, E., Bady, P., Hicking, C., Perry, J., Hong, Y.K., Roth, P., et al. (2016). Cilengitide in newly diagnosed glioblastoma: biomarker expression and outcome. Oncotarget 7, 15018–15032. Yan, H., Parsons, D.W., Jin, G., McLendon, R., Rasheed, B.A., Yuan, W., Kos, I., Batinic-Haberle, I., Jones, S., Riggins, G.J., et al. (2009). IDH1 and IDH2 mutations in gliomas. N. Engl. J. Med. 360, 765–773.

OUR NETWORK IS YOUR NETWORK With Cell Press Webinars, our network is your network! Cell Press Webinars give you access to hot topics in emerging research and the application of new technology.

Need-to-know topics, editorially curated World-class presenters, experts in their field

Watch essential, need-to-know webcasts via live streaming or on demand from the comfort and convenience of your office, lab, or home.

Moderated by Cell Press editors

Tap into our network today! Visit www.cell.com/webinars

Webinars

ALL THE COMFORTS WITHOUT THE CREATURES

Connect with the best and the brightest at Cell Symposia Cell Symposia are designed with you in mind. Short, compact programs are held in convenient, central locationsâ&#x20AC;&#x201C;often in close proximity to scientific centers of excellence.

Make sure the time you spend away from the lab is quality time, register for a Cell Symposia. Our editors curate symposia on a broad range of topics. See for yourself at www.cell.com/symposia

Itâ&#x20AC;&#x2122;s happening at Cell Symposia!

Tell your cell’s story with real-time live-cell analysis

Get started at Essenbio.com/incucyte