Personalized In Vitro and In Vivo Cancer Models to Guide Precision Medicine

17 Pages·2017·4.38 MB·English

Published OnlineFirst March 22, 2017; DOI: 10.1158/2159-8290.CD-16-1154
RESEARCH ARTICLE
Personalized In Vitro and In Vivo Cancer
Models to Guide Precision Medicine
Chantal Pauli1,2,3, Benjamin D. Hopkins4, Davide Prandi5, Reid Shaw6, Tarcisio Fedrizzi5, Andrea Sboner1,4,7,
Verena Sailer1,2, Michael Augello1,4, Loredana Puca1, Rachele Rosati6, Terra J. McNary1, Yelena Churakova1,
Cynthia Cheung1, Joanna Triscott1, David Pisapia1,2, Rema Rao1,2, Juan Miguel Mosquera1,2, Brian Robinson1,2,
Bishoy M. Faltas1,8, Brooke E. Emerling4, Vijayakrishna K. Gadi9, Brady Bernard6, Olivier Elemento1,4,7,
Himisha Beltran1,8, Francesca Demichelis1,5, Christopher J. Kemp10, Carla Grandori6, Lewis C. Cantley4,
and Mark A. Rubin1,2,4
ABSTRACT Precision medicine is an approach that takes into account the infuence of indi-
viduals’ genes, environment, and lifestyle exposures to tailor interventions. Here,
we describe the development of a robust precision cancer care platform that integrates whole-exome
sequencing with a living biobank that enables high-throughput drug screens on patient-derived tumor
organoids. To date, 56 tumor-derived organoid cultures and 19 patient-derived xenograft (PDX) mod-
els have been established from the 769 patients enrolled in an Institutional Review Board–approved
clinical trial. Because genomics alone was insuffcient to identify therapeutic options for the majority
of patients with advanced disease, we used high-throughput drug screening to discover effective
treatment strategies. Analysis of tumor-derived cells from four cases, two uterine malignancies and
two colon cancers, identifed effective drugs and drug combinations that were subsequently validated
using 3-D cultures and PDX models. This platform thereby promotes the discovery of novel therapeutic
approaches that can be assessed in clinical trials and provides personalized therapeutic options for
individual patients where standard clinical options have been exhausted.
SIGNIFICANCE: Integration of genomic data with drug screening from personalized in vitro and in vivo
cancer models guides precision cancer care and fuels next-generation research. Cancer Discov; 7(5);
1–16. ©2017 AACR.
1Englander Institute for Precision Medicine, Weill Cornell Medicine- Note: Supplementary data for this article are available at Cancer Discovery
New York Presbyterian Hospital, New York, New York. 2Department of Online (http://cancerdiscovery.aacrjournals.org/).
NPaetwh oYlogrky. a3Innds tLitaubtoer aotfo rPya tMhoeldoigciyn ea,n Wd eMilol lCecournlaerl l PMatehdoicloingey, NUenwiv eYrosirtky, C. Pauli and B.D. Hopkins contributed equally to this article.
Hospital Zurich, Zurich, Switzerland. 4Meyer Cancer Center, Weill Cornell Corresponding Author: Mark A. Rubin, Director, Englander Institute for
Medicine, New York, New York. 5Center for Integrative Biology, University Precision Medicine, Weill Cornell Medicine-New York Presbyterian Hospi-
of Trento, Trento, Italy. 6Cure First and SEngine Precision Medicine, Seat- tal, 413 East 69th Street, 14th Floor, New York, NY 10021. Phone: 646-
tle, Washington. 7Institute for Computational Biomedicine, Weill Cornell 962-6164; Fax: [phone]; E-mail:
Published OnlineFirst March 22, 2017; DOI: 10.1158/2159-8290.CD-16-1154
INTRODUCTION across many tumor types may signifcantly affect patient
outcomes in the future.
Precision oncology is an approach to cancer treatment Large-scale drug screens of cell line panels—such as the
that seeks to identify effective therapeutic strategies for NCI60 by the National Cancer Institute or the Cancer Cell
every patient. The Englander Institute for Precision Medi- Line Encyclopedia (CCLE)—have addressed compound sen-
cine (EIPM) initiated its clinical research program in 2013, sitivity in cancer cells to identify mechanisms of growth
using whole-exome sequencing (WES) of metastatic and inhibition and tumor-cell death (4, 5). A more recent study
primary tumors, with prospective follow-up of patients of pharmacogenomic interactions in cancer links genotypes
to identify individualized therapeutic options and to help with cellular phenotypes with the purpose of targeting select
guide clinical decision making (1, 2). The identifcation of cancer subpopulations (6). Unfortunately, for many cancer
mutations that arise during treatment that confer drug types, traditional cell culture methodologies do not ade-
sensitivity is paramount for precision cancer care of patients quately model the biology of the native tumor. The high
with advanced disease (3). However, there remain a signif- failure rate of preclinical compounds in clinical trials clearly
cant number of cases where genomic analysis currently fails demonstrates the limitations of existing preclinical models
to identify effective drugs or applicable clinical trials. Even (7, 8). The accuracy of in vitro drug screens is therefore depen-
when targetable genomic alterations are discovered, patients dent on the optimization of cell culture tools that more
do not always respond to therapy. Strategies to confrm closely mirror patient disease.
therapeutic effcacy or identify additional options would be The taxonomy of cancer has classically been the domain
benefcial to both clinicians and patients. To address this of pathologists using tumor cell morphology to help guide
need, we describe the establishment of a living biobank con- patient care. With the development of molecular markers,
sisting of tumor organoids, which facilitates the integration oncologists can work with pathologists to identify tumor
of genomic data with drug screening of patients’ tumor subgroups. There have been dramatic clinical responses in
samples in an iterative platform to identify effective thera- some of these subgroups with available targeted agents such
peutic regimens for individual patients. Although it may as trastuzumab or imatinib mesylate. The advent of high-
not be feasible to utilize this approach for all patients with throughput sequencing methodologies has enabled consor-
cancer, the integration of genomic with drug-sensitivity data tia—such as The Cancer Genome Atlas (TCGA) and the
MAY 2017 CANCER DISCOVERY | OF2
Downloaded from cancerdiscovery.aacrjournals.org on April 9, 2019. © 2017 American Association for Cancer
Research.
Published OnlineFirst March 22, 2017; DOI: 10.1158/2159-8290.CD-16-1154
HCRAESER ELCIT RA Pauli et al.
International Cancer Genome Consortium (ICGC)—to gen- genomic region. In our cohort, there were FDA-approved
erate large datasets across a broad range of cancer types, drugs identifed for 0.4% (3/737) of patients. Based on an
providing insights into the genomic landscape of cancer and expanded list of targeted therapies available at My Cancer
identifying new potential therapeutic targets (9–13). However, Genome (30), 9.6% (71/737) of the analyzed patients had
an understanding of the functionality of these alterations potentially targetable cancer gene alterations (e.g., EGFR
and the infuence they have on treatment response remains p.L858R; BRAF p.V600E; ERBB2 amplifcation), though with-
limited due to a paucity of personalized preclinical mod- out current FDA-approved drug indication (Fig. 1B and C).
els (14, 15). Through the establishment of personal tumor It is important to note that these numbers refect a select
organoids and the implementation of high-throughput drug population of patients with advanced cancer, the majority
screens, our platform pairs drug-sensitivity information with of whom have failed prior therapies or lack standard-of-care
detailed genomic profles. This allows for the generation of options, and are therefore not indicative of patients with
direct correlative associations between the cancer genome and cancer in general.
the outcome of drug treatment. The most frequently mutated cancer genes with single-
Organoid technology is used in research as an intermedi- nucleotide variants (SNV)/indels in our cohort (≥5%) were
ate model between cancer cell lines in vitro and xenografts as TP53 (37.4%), APC (11.3%), NOTCH1 (8.4%), EGFR (6.7%),
shown for colorectal, pancreatic, and prostate cancers (16–21). KMT2D (6.6%), ARID1A (6.6%), TET2 (6.3%), KRAS (6.2%),
This technique differs from traditional cell culture by main- CREBBP (5.5%), and PIK3CA (5.0%; Fig. 1D). The most com-
taining cancer cells in three-dimensional (3-D) cultures. Can- mon somatic copy-number aberrations (SCNA; ≥18%) were
cer cells that are grown in 3-D retain cell–cell and cell–matrix seen in CDKN2A (25.3%), RB1 (24.9%), WRN (24.3%), PCM1
interactions that more closely resemble those of the original (22.4%), PTEN (22.0%), CDKN2B (21.9%), LCP1 (20.6%), AR
tumor compared with cells grown in two dimensions on plastic (20.2%), FGFR1 (19.9%), WHSC1L1 (19.8%), MYC (18.5%),
(22–29). Utilizing our newly established 3-D patient organoid BMPR1A (18.4%), and TP53 (18.0%; Fig. 1E). Of the 4.02%
culture system, personalized high-throughput drug screening (32/769) of cases sequenced, mutations were found only
coupled with genomic analysis from patient-derived tumor in genes with unknown clinical or biological signifcance.
samples offers a unique opportunity to stratify and identify Together, these data suggest that WES—although highly
effective cancer therapies for individual patients. By adding a informative for some cancers with targetable mutations (e.g.,
drug screening component into our precision medicine plat- BRAF and EGFR)—is insuffcient to nominate therapeutic
form, we are able to (i) compare the response of individual alternatives in many advanced cancer types.
tumors to specifc drugs in order to provide individualized
Patietn eD- ir ev d uT rom agrO ion sd a dn eX argon tf s
recommendations to help guide patient care; (ii) assess how
a s lo T s rof Per ic is no aC ecn r aC er
individual tumors adapt in response to therapies and better
understand the context in which these agents are effcacious; To complement the genomic information and to provide
(iii) determine the next course of action for cases where stan- therapeutic options for patients, we integrated personalized
dard clinical options have already been exhausted; (iv) create PDTO drug screens and PDX generation into our plat-
a database that relates drug sensitivity to tumor genetics form (Fig. 2). Fresh tumor tissue biopsies or formalin-fxed,
to nominate potential therapeutic strategies even when only paraffn-embedded (FFPE) material was used for sequencing
genomic data are available. Herein, we describe a precision (Fig. 3A). Fresh tissue was snap-frozen for sequencing, and
oncology approach that combines WES, patient-derived tumor PDTOs were generated from cases with suffcient available
organoids (PDTO), high-throughput drug screening, and material (Supplementary Fig. S1A). To date, 145 specimens
patient-derived xenografts (PDX). We further outline how this have been collected, representing 18 different tumor types
platform can discover novel treatment strategies in a clinically derived from patients with metastatic solid tumors of epi-
relevant timeframe and lead to innovative clinical trials. thelial and mesenchymal origin. These include metastatic
and primary tumors originating from the prostate (n = 52),
bladder/ureter (n = 24), kidney (n = 10), colon/rectum
RESULTS
(n = 10), brain (n = 9), pancreas (n = 7), breast (n = 6), stomach
SE W s I usn I ic � f etn t o edI tn i y f lC iin ac ll yaT egr talb e
and esophagus (n = 6), soft tissue (n = 6), small intestine (n
lA tear ti sno rof aM yn avdA ecn d aC ecn r epy T s
= 3), lung (n = 2), liver (n = 2), adrenal gland (n = 2), uterus
Our institute established the EXaCT-1 Test, a WES-based (n = 2), ovary (n = 1), appendix (n = 1), and thyroid (n = 1),
precision medicine platform designed to inform therapeutic and cancer of unknown primary (CUP; n = 1; Supplementary
decision-making for patients with cancer (1, 2). To date, the Fig. S1B). Tumor organoids were successfully established
EIPM has sequenced and analyzed 769 tumor–normal pairs from 56 specimens (38.6%), including 43 of 120 tissue biop-
from an array of different primary and metastatic tumor sies and 13 of 32 surgical resection specimens (Fig. 3B). We
sites from 501 patients, the majority of whom had advanced defned successful establishment of PDTO cultures as when
disease (Fig. 1A). WES identifed alterations involving known they contain viable cells that form spheroid-like structures
cancer genes in 95.8% (737/769) of the analyzed specimens. and can be propagated after the initial processing for at least
Here, cancer genes are defned according to the updated fve passages. These specimens were characterized, stored
list from the COSMIC cancer gene census (cancer.sanger. in our living biobank, and used for functional studies. Cell
ac.uk/census). The data presented report cases with broad viability was assessed in the frst 10 established cultures
genomic structural variations (i.e., amplifcations or larger- at passages 2–4, and in 9 out of 10 cases, >90% of cells were
scale deletions) that include all genes contained within the viable (Supplementary Fig. S1C). Tumor organoids were
OF3 | CANCER DISCOVERY MAY 2017 www.aacrjournals.org
Downloaded from cancerdiscovery.aacrjournals.org on April 9, 2019. © 2017 American Association for Cancer
Research.
Published OnlineFirst March 22, 2017; DOI: 10.1158/2159-8290.CD-16-1154
Personalized Cancer Models to Guide Precision Medicine CRAESER H CIT RA EL
A Specimen distribution based on tumor origin B EXaCT-1 overview: detected genomic
alterations
Prostate Somatic alterations
Bladder in currently not targetable
Kidney cancer genes
Blood/bone marrow 85.8% Somatic alterations in
25% 34%
Ovary targetable cancer genes
34%
Colon/rectum (potential off-label drug use)
2%
Lung
Somatic alterations in
2%
Pancreas
cancer genes with FDA-
3%
3% 9% Breast approved drugs
3% 9% Small intestine
Somatic alterations with
3%
Brain
3% unknown clinical and
4% Others (e.g., soft tissue/ 9.6%
bone, uterus, stomach, 4.2% biological significance
0.4%
esophagus, etc.)
C Cancer gene alterations per case sequenced
300
Number of cancer gene alterations per case sequenced (gray)
200
100
0 200 400 600 800
0
* * *
Number of cancer gene alterations per case that could be targeted by FDA-approved drugs in
green* (0.4%) and that could be clinically actionable by potential off-label use in orange (9.6%).
D Most common SNVs/indels in cancer genes E Most common SCNA in cancer genes
TP53 37.4% CDKN2A 25.3%
APC 11.3% RB1 24.9%
NOTCH1 8.4% WRN 24.3%
EGFR 6.7% PCM1 22.4%
KMT2D 6.6% PTEN 22.0%
ARID1A 6.6% CDKN2B 21.9%
TET2 6.3% LCP1 20.6%
KRAS 6.2% AR 20.2%
CREBBP 5.5% FGFR1 19.9%
PIK3CA 5.0% WHSC1L1 19.8%
RB1 4.9% MYC 18.5%
ATM 4.9% BMPR1A 18.4%
ALK 4.9% TP53 18.0%
KIAA1549 4.7% MALT1 17.8%
BRCA2 4.6% LHFP 17.8%
NF1 4.1% BRCA2 16.9%
PTCH1 3.9% BCL2 16.9%
SMARCA4 3.8% USP6 16.7%
NOTCH2 3.8% COX6C 16.7%
IDH1 3.7% CBFA2T3 16.4%
0 5 10 15 20 25 30 35 40 % 0 5 10 15 20 25 30 %
Figure 1.  WES detects a limited number of clinically targetable alterations in patients with advanced cancer. ,A Overview of the sites of origin of speci-
men collected from patients and run through the EXaCT-1 test. The majority of these samples were taken from metastatic sites of patients with advanced
disease. ,B WES has been performed on a total of 769 specimens. Data presented here also include large-scale deletions (> 50 genes), and each gene is indi-
vidually included in the analysis. In 85.8% (660/769) of the cases sequenced, somatic alterations in currently not targetable cancer genes were detected.
In total, three cases (0.4%), two gastrointestinal stromal tumors (GIST) with an activating KIT mutation and a clear cell renal cell carcinoma with a BRAF
mutation, have FDA-approved drugs available. In 9.6% (71/737) of these, there are somatic alterations in cancer genes that could be clinically actionable
by off-label use of approved drugs; however, clinical effcacy has not been proven. In 4.2% (32/737) of the cases, we did not detect any somatic altera-
tions in known cancer genes. C , Bar graph showing above the x -axis the number of alterations in cancer genes detected in each case (gray). Below the x-axis
we show in green (*) the three cases that have FDA-approved drugs available and in orange the cases that have clinically actionable gene alterations by
potential off-label use of FDA-approved drugs. ,D List of the 20 most relevant SNVs/indels in cancer genes detected in our cohort. Cancer genes (red) that
have FDA-approved drugs available for non–small cell lung cancer (crizotinib, erlotinib, geftinib) and for ovarian cancer (olaparib). ,E The 20 most common
SCNA-detected genes that have FDA-approved drugs available (green) for ovarian cancer (olaparib) and for chronic lymphocytic leukemia (venetoclax).
MAY 2017 CANCER DISCOVERY | OF4
Downloaded from cancerdiscovery.aacrjournals.org on April 9, 2019. © 2017 American Association for Cancer
Research.
Published OnlineFirst March 22, 2017; DOI: 10.1158/2159-8290.CD-16-1154
HCRAESER ELCIT RA Pauli et al.
Sequencing Sequencing report
Tumor organoids
Drug screening
Patient-derived xenograft In vivo drug validation
Figure 2.  Personalized models to guide precision medicine in advanced cancer. Illustration of our precision medicine program depicting the workfow,
beginning with sequencing using the EXaCT-1 WES test (top), continuing with the establishment of PDTOs, which are compared with the primary tumor
sample through histology and sequencing before they are subjected to drug screening (middle row; arrow with dotted line suggests unestablished path-
way), and utilized to generate PDXs where potential drugs are validated in mice (bottom). The sequencing data are available in our internal cbio-portal
and reported back to the referring physician. Tumor organoid cultures are prepared from fresh patient tumor samples as personalized in vitro models.
After the initial characterization, targeted or high-throughput single and combination drug screens can be performed in an iterative process in order to
nominate therapeutic strategies that are further evaluated in personalized in vivo models.
characterized using cytology and histology as previously analysis of 1,062 putative cancer genes showed a median of
described (ref. 31; Supplementary Fig. S1D–S1F). 86% concordance (see Methods) when comparing PDTOs and
Tumor organoids with a passage number < 20 were used PDXs to the native tumor tissues (Fig. 4). Minor differences
for pharmacologic screens and the establishment of PDXs. observed are either due to subclonality in the native tumor
To date, we have generated 19 PDXs out of 22 attempts (subclones not represented in the PDTO) or due to the pro-
(86.4%) from PDTOs representing colorectal, pancreas, pros- gression of the PDTO/PDX. Similarly, SNV analysis showed
tate, urothelial, uterine, kidney, and lung cancers, and sar- excellent concordance between native tumors, matching
coma (Fig. 3C). The matched PDTOs and PDXs had similar PDTOs (Supplementary Fig. S2D; Supplementary Table S1)
histopathology to the parental tumors from which they were and PDXs upon considering mouse DNA reads (Supplemen-
derived (Fig. 3D). The WES data of PDTOs and PDXs from tary Fig. S2E–S2H; Methods).
9 solid tumor types in 15 patients (Supplementary Table S1)
Ex Vivo urD g ercS ein gn imoN an te s
were analyzed to test concordance at base-pair resolution
ehT ar ep uti c aC idn ad tes
with the native tumor. We evaluated tumor purity and ploidy
using the CLONET computational framework (CLONal- Tumor organoids, derived from biopsies and surgical speci-
ity Estimate in Tumors; Supplementary Fig. S2A and S2B; mens of four patients with cancer, were subjected to high-
ref. 32). As expected, PDTO and PDX samples demonstrate throughput drug dose-response screens. The four organoid
high purity in all but one case, PDX WCM236_P1, where we samples were derived from patients with a uterine carci-
detected mouse DNA admixed with the human tumor tissue nosarcoma (patient A, stage IIIB: PIK3CA p.Q546H; PTEN
DNA (see Methods). Ploidy and genomic burden (33) pro- p.K6fs*4), an endometrial adenocarcinoma (patient B, IIIC2:
fles (Supplementary Fig. S2B and S2C) of PDTOs and PDXs PIK3CA p.H1047R; PTEN p.K267fs*9; CTNNB1 p.D32G),
also matched patient tumor data. Allele-specifc copy-number a stage IV colorectal cancer with a clinically relevant KRAS
OF5 | CANCER DISCOVERY MAY 2017 www.aacrjournals.org
Downloaded from cancerdiscovery.aacrjournals.org on April 9, 2019. © 2017 American Association for Cancer
Research.
Published OnlineFirst March 22, 2017; DOI: 10.1158/2159-8290.CD-16-1154
Personalized Cancer Models to Guide Precision Medicine CRAESER H CIT RA EL
A Total tumor samples n = 769 B Tumor organoid etablishment from fresh tissue
Fresh tissue
55
(342/769) 10/52
FFPE 50 Tissue submitted for
50.9% 44.5%
(392/769) organoid establishment
45
Blood, Tumor organoids
bone marrow established
40
(35/769)
4.6%
35
C PDX establishment from tumor organoids
30
7/7
7 25 8/24
6
PDX established
20
6/10
5 PDX failed
15
4
8/10
3/3 2/3 10 5/9
3 4/6
1/6 3/6 5/7
2/2 1/2 2/2
2 5 1/2 2/3 2/2
1/1
1/1 1/1 0/1
1 0
0
D Native tumor specimens and their derived tumor organoid and xenograft
Endometrial Pancreatic ductal Colorectal Uterine Urothelial Renal cell
adenocarcinoma adenocarcinoma cancer carcinosarcoma carcinoma carcinoma
Figure 3.  Development of preclinical models for the guidance of precision medicine. ,A Of the 769 samples that have been run through the sequencing
program, 50.9% were FFPE specimens, 44.5% were freshly collected tissue specimens, and a minority of 4.6% were from patients with hematologic malig-
nancies sent to us as DNA specimens. ,B 152/342 of the freshly collected specimens had a tissue biopsy or resection specimen to attempt the development
of tumor organoids. Of these 152, 56 (36.8%) patient-derived organoids were successfully initiated from numerous tissues, including prostate (10/52), blad-
der/ureter (8/24), kidney (6/10), breast (4/6), colon/rectum (8/10), esophagus (1/6), soft tissue (3/6), brain (5/9), pancreas (5/7), lung (1/2), small intestine
(2/3), ovary (1/1), and uterus (2/2). C , 22 of these patient-derived organoid models were subsequently injected into mice of which 19 organoid lines from
colorectal cancer (CRC; n = 7), pancreatic ductal adenocarcinoma (PDAC; n = 3), uterine cancers (n = 2), neuroendocrine prostate cancer (NEPC; n = 2), renal
cell carcinoma (RCC; n = 2), urothelial cancer (n = 1), lung adenocarcinoma (n = 1), and sarcoma (n = 1) successfully engrafted. ,D Histology of primary tumor
samples, tumor organoids, and xenografts from six different solid tumors: endometrial adenocarcinoma, pancreatic ductal adenocarcinoma, colorectal cancer,
uterine carcinosarcoma, urothelial carcinoma, and renal cell carcinoma. Tumor organoid gross morphology (row 2) shows tumor type–specifc structures such
as the formation of lumina as seen for pancreas and colon. Tumor organoid and PDX histology shows conservation of the histopathologic features of the native
tumors. Hematoxylin and eosin (H&E) stain from native tumor tissue, scale bar, 200 μm; tumor organoids bright feld view in vitro, scale bar, 10 μm; H&E stain
from tumor organoids, scale bar, 20 μm (urothelial carcinoma 200 μm); H&E stain from PDXs, scale bar, 20 μm.
MAY 2017 CANCER DISCOVERY | OF6
Downloaded from cancerdiscovery.aacrjournals.org on April 9, 2019. © 2017 American Association for Cancer
Research.
Cases
Xenograft Tumor organoid Native tumor
CRC
PDAC
NEPC
Uterine
cancer
Lung
adenoca.
Urothelial
cancer
Sarcoma
RCC
Breast
Biopsies
Prostate
Bladder/ureter
Kidney
Breast
Colon/rectum
Stomach/
esophagus
Soft tissue
Brain
Pancreas
Lung
Small intestine
Ovary
Uterus
Published OnlineFirst March 22, 2017; DOI: 10.1158/2159-8290.CD-16-1154
HCRAESER ELCIT RA Pauli et al.
Tissue type Copy number status Tumor type
T: Native tumor Homo del Loss of het Colorectal cancer Neuroendocrine tumor Serous carcinoma of the ovary
O: Tumor organoid Hemi del Gain Endometrial adenocarcinoma Pancreatic adenocarcinoma Urothelial carcinoma
P: Organoid-derived PDX Wild-type Amp Leiomyosarcoma Renal cancer Uterine carcinosarcoma
Chromosome
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22
P1
WCMC236
T1
P1
O3
O2
WCMC748
O1
T2
T1
O1
WCMC334
O1
WCMC244
T1
T1
WCMC331
O1
O1
WCMC602
T3
P1
T1
WCMC526
T2
O3
WCMC773 T1
O2
P1
WCMC389
O1
O2
O1
WCMC610
T1
O1
WCMC609 O1
TO1
T1
P2
T1
T1
WCMC601
P1
P1
O2
WCMC715
O2
O1
O1
T4
T3
T2
T1
WCMC616
P1
O3
O2
PO1
TO15
O4
WCMC392 O3
O2
O1
T1
Figure 4.  Allele-specifc copy-number heat map showing the genomic characterization of patient-derived in vitro and in vivo models. Allele-specifc
copy number of 1,062 putative cancer genes (columns) derived from 57 whole-exome tumor tissue samples (rows) from 15 patients. Six copy-number
states are represented: homozygous (Homo del) and hemizygous deletions (Hemi del), wild-type, low-level gain (Gain), and high-level amplifcation (Amp),
loss of heterozygosity (Loss of het). Relevant biomarkers are highlighted at the bottom. Left column annotation reports the tissue type (N, native tumor;
O, tumor organoid; P, organoid-derived PDX). The different numbers for T denote different locations of the native tumor samples sequenced. Different O
numbers indicate tumor organoids in culture over time (O1, passage 5; O2, passage 10; O3, passage 15; O4, passage 20; O5, passage 30). P1 is the initial
xenograft derived from tumor organoid passage < 20 and P2 is an orthotopic xenograft (WCMC601 intrauterine). Right column annotation reports the
tumor type of the patient.
p.G13D and TP53 p.R282W mutation (patient C), and a stage agents under clinical development (Supplementary Table S2).
IV colorectal cancer with an APC mutation p.G857X and an These compounds target clinically relevant molecules and
APC frameshift insertion p.E1536_fs (patient D). The drug signaling pathways, including PI3K, RTKs, CDK, MEK,
libraries used consisted of a total of ∼160 drugs (119 NCI HDAC, IKK/NFKB, mTOR, PKC, HH, EGFR, CHK, PARP,
Library + 120 SEngine Library with 70% overlap), includ- epigenetic modifers, the proteasome, and apoptosis. Each
ing current FDA-approved chemotherapeutics and targeted drug was tested with a minimum of 6 to a maximum of 8
OF7 | CANCER DISCOVERY MAY 2017 www.aacrjournals.org
Downloaded from cancerdiscovery.aacrjournals.org on April 9, 2019. © 2017 American Association for Cancer
Research.
Patient ID
Tissue type
Tumor type
ABL2
ALK
PDGFRA
PDGFRB
ROS1
CDK6
BRAF
ABL1
RET
KRAS
CDK4
FLT3
BRCA2
ERBB2
BRCA1
BCL2
JAK3
Published OnlineFirst March 22, 2017; DOI: 10.1158/2159-8290.CD-16-1154
Personalized Cancer Models to Guide Precision Medicine CRAESER H CIT RA EL
different concentrations selected to cover the therapeutic single agent for patient B (Fig. 5D and E), it was selected as
range (Supplementary Table S2). As proof of principle, a modifying agent for a subsequent combinatorial screen
+
screening of patient-derived examples of ER breast cancer because of its effect within the range of therapeutic doses
and BRAF-mutant melanoma with the drug library identi- and patient B’s PIK3CA mutation. The results of the combi-
fed tamoxifen and BRAF as well as MEK inhibitor as top- nation screen indicated that buparlisib sensitized cells from
scoring drugs, respectively, indicating the high-throughput patient B to two HDAC inhibitors, vorinostat and belinostat,
screens can identify targeted agents that are clinically vali- and to a PARP inhibitor, olaparib (Fig. 6D–F; Supplementary
dated (Supplementary Fig. S3A–S3H). To directly compare Table S5; Supplementary Fig. S5B). The combination screen
drug responses in liquid versus semisolid culture conditions for patient C showed that the MEK inhibitor trametinib sen-
with Matrigel, we tested the same 120-drug library with sitized the KRAS-mutant colon cancer cells to multiple drugs
tumor cells from patient B using both conditions. There was including celecoxib, nilotinib, vorinostat, belinostat, and
a high degree of concordance between the two conditions afatinib (Fig. 6G–I; Supplementary Table S6; Supplementary
with a Spearman rank correlation coeffcient (Spearman Rho) Fig. S5C), indicating sensitization to multiple FDA-approved
of 0.77 (Supplementary Fig. S3I). Only ∼5% of drugs tested agents, which on their own had no or little effect. In contrast,
showed preferential activity when assayed in 2-D versus 3-D. for the KRAS WT colon cancer cells from patient D, a sen-
Unless noted, the results below were obtained from culture sitizing screen with the EGFR inhibitor afatinib, which was
of organoids in 2-D (for the colon cancer cases, collagen was the top-scoring single agent, showed an enhanced response
used as a substratum to improve cell viability). Selected top- to HDAC and IGF1R inhibitors (Fig. 6J–L; Supplementary
scoring drugs were subsequently validated in 3-D. Table S7; Supplementary Fig. S5D). This result is consistent
Tumor cells from patients A and B with uterine malig- with a colorectal cancer model showing that EGFR inhibitors
nancies demonstrated in vitro responses to PI3K inhibitors, synergize with HDAC inhibitors (35). Thus, we were able to
consistent with PIK3CA mutations in both samples, single- identify unique combinations for each patient.
agent chemotherapeutics (e.g., purine synthesis inhibitors
D-3 and P X D Models aV lidate iS nlg e egA nts and
such as fudarabine), and HDAC inhibitors (Fig. 5A–F; Sup-
Coibm nations dI entie� d In Vitro and Proiv de
plementary Table S3–S5; Supplementary Fig. S4A–S4F). The
aS ef t y nI of ram tion
KRAS-mutant colorectal tumor cells from patient C were
resistant to most agents but showed an exceptional response In order to validate high-throughput drug screen results
to the MEK inhibitor trametinib. The KRAS WT colorectal and prioritize agents for in vivo studies, we tested top-scoring
tumor cells from patient D demonstrated sensitivity to EGFR single agents and combinations in 3-D PDTOs for all four
inhibitors, and in particular afatinib (Fig. 5G–L; Supplemen- patients. Dose-response cell survival assays in the PTDOs
tary Table S7; Supplementary Fig. S4G–S4J), consistent with confrmed the specifcity and sensitivity of both single agents
known sensitivity of KRAS WT colon cancer to inhibition of and combinations for all four patients (Supplementary Fig.
EGFR and approval of cetuximab (34). S5E–S5H). Drug combinations were selected for further vali-
Development of resistance to both chemotherapies and dation using PDX models based on the drug screen effcacy
targeted agents is very frequent in clinical practice and is as well as safety considerations. Drug combinations that had
a major cause of mortality. Anticipating this, we set out to already been tested in clinical trials, such as buparlisib and
identify optimal combination therapies for all four tumor olaparib, were given preference over novel combinations, as
organoid lines using combination drug screens with selected they had a greater potential to be of immediate clinical utility.
compounds identifed from single-agent screens and/or com- We also compared response with standard-care therapies that
pounds indicated by the genetic data (Fig. 6A–L). Each of the the patients received. We studied PDX models from patient
selected compounds was used at sublethal concentrations B’s uterine cancer and patient D’s colorectal cancer because
(∼IC30) and tested in combination with the entire 120-drug organoids from these patients exhibited optimal growth kin-
library. Drugs that showed strong curve shifts in combi- etics after successful engraftment.
nation relative to single agents (top 15% by way of AUC fold Patient B had a radical hysterectomy and is currently receiv-
change) and uniquely sensitive to the patient sample (nega- ing the chemotherapy combination carbotaxol (carboplatin
tive AUC Z-score) were selected as candidate drug combi- at 50 mg/kg and paclitaxel 20 mg/kg), given once a week,
nations. For further prioritization, we manually curated the and shows stable disease. This treatment was compared with
drugs to identify the most promising combinations (e.g., daily buparlisib at 50 mg/kg, vorinostat at 200 mg/kg, and
novelty, clinical utility, and targeted therapies). For cells olaparib at 50 mg/kg as single agents, as well as buparlisib in
from patient A, the top drug in the buparlisib combination combination with vorinostat or olaparib. Greater inhibition
screen was the HDAC inhibitor vorinostat (Supplementary of tumor growth was observed in mice treated with buparlisib
Table S3). An additional combination screen was performed as a monotherapy and in combination with vorinostat or
using the HDAC inhibitor vorinostat (which was a top drug olaparib as compared with mice treated with carbotaxol, vori-
in the single-agent screen for this patient) as the sensitizing nostat, or olaparib administered as monotherapies (Fig. 7A
agent (Supplementary Table S4). This second combination and B). These results demonstrate good concordance of both
screen confrmed the fndings of the buparlisib-sensitizing single agents and drug combinations identifed in vitro with
screen by demonstrating that vorinostat further sensitized in vivo tumor response. Also, targeted agents identifed through
patient A’s carcinosarcoma cells to PI3K inhibitors, including drug screens performed better than existing standard-of-care
buparlisib (Fig. 6A–C; Supplementary Table S4; Supplemen- therapies. However, directly comparing PDX response to
tary Fig. S5A). Although buparlisib was not the top-scoring patient response would be needed to establish clinical utility.
MAY 2017 CANCER DISCOVERY | OF8
Downloaded from cancerdiscovery.aacrjournals.org on April 9, 2019. © 2017 American Association for Cancer
Research.
Published OnlineFirst March 22, 2017; DOI: 10.1158/2159-8290.CD-16-1154
HCRAESER ELCIT RA Pauli et al.
Patient A; uterine carcinosarcoma
Genomic alteration detected: PIK3CAmut, PTENmut
A B C
1 High-throughput single compound screen Tumor organoid drug validation
1 1.2
Buparlisib [IC50 ~ 0.32 µmol/L]
0 GDC0980 [IC50 ~ 0.025 µmol/L]
0.8 Vorinostat [IC50 ~ 0.28 µmol/L]
−1 Belinostat [IC50 ~ 0.20 µmol/L]
Paclitaxel [IC50 ~ 0.0009 µmol/L]
0.4 Carboplatin [IC50 ~ 3.63 µmol/L]
−2
Buparlisib
Vorinostat
−3 0.0
0.0 0.2 0.4 0.6 0.8 1.0 1.2 1.4 −6 −4 −2 0 2
0
Normalized AUC log[Agonist], µmol/L
Patient B; endometrial adenocarcinoma
Genomic alteration detected: PIK3CAmut, PTENmut, CTNNB1mut
D E High-throughput single compound screen F Tumor organoid drug validation
1
1 1.2
Buparlisib [IC50 ~ 1.541 µmol/L]
0
0.8 GDC0980 [IC50 ~ 0.4 µmol/L]
Vorinostat [IC50 ~ 0.55 µmol/L]
−1 Buparlisib Belinostat [IC50 ~ 0.20 µmol/L]
Vorinostat Paclitaxel [IC
50 ~ 0.0017 µmol/L]
0.4
−2 Carboplatin [IC50 undefined]
−3 0.0
0.0 0.2 0.4 0.6 0.8 1.0 1.2 1.4 −6 −4 −2 0 2
0
Normalized AUC log[Agonist], µmol/L
Patient C; colorectal cancer
Genomic alteration detected: KRASmut, TP53mut, APC mut
Tumor organoid drug validation
G 1 H High-throughput single compound screen I
1.2
1
Trametinib [IC50 ~ 0.006 µmol/L]
0 Afatinib Afatinib [IC50 undefined]
0.8
Oxaliplatin [IC50 ~ 1.06 µmol/L]
−1 5 FU [IC50 ~ 0.1 µmol/L]
Trametinib
0.4
−2
−3 0.0
0.0 0.2 0.4 0.6 0.8 1.0 1.2 1.4 −6 −4 −2 0 2
0
Normalized AUC log[Agonist], µmol/L
Patient D; colorectal cancer
Genomic alteration detected: APCmut, APC frameshift insertion
J K High-throughput single compound screen L Tumor organoid drug validation
1
1 1.2
0 Trametinib [IC
50 undefined]
Trametinib 0.8 Oxaliplatin [IC50 ~ 3.2 µmol/L]
−1 5 FU [IC50 ~ 0.04 µmol/L]
0.4
−2 Afatinib
−3 0.0
0.0 0.2 0.4 0.6 0.8 1.0 1.2 1.4 −6 −4 −2 0 2
0
Normalized AUC log[Agonist], µmol/L
Figure 5.  High-throughput drug screen and validation to nominate patient-specifc therapeutic options. Once organoids have been established and
validated the cells can be utilized to identify patient-specifc responses to therapeutic agents through the use of selected or high-throughput drug
screens. A, D, G, and J, Heat maps of the drug screen results depicting the relative sensitivity of the patients’ tumor cells from most resistant (red) to
most sensitive (blue). Black dots indicate agents that were selected for validation and further studies. B, E, H, and K, Graphs of the response of patient’s
tumor cells to each compound in the library as a Z-score (AUC) compared with other primary cells screened with the same library (N(SEngine) = 43, N(NCI) = 10).
C, F, I, and L, The in vitro validation of selected drugs in the 3-D system. Patient A: The patient’s tumor cells were generally resistant to many agents.
Selective enrichment was seen for targeted agent classes such as PI3K (AZD8482, buparlisib, GDC-0980, idelalisib, taselisib, PIK-75, NVP-BGT226) and
HDAC inhibitors (vorinostat, belinostat). Drug sensitivity was validated using the 3-D Matrigel system and compared with the patient’s actual treatment
paclitaxel and carboplatin as single agents. Patient B: The high-throughput drug screen demonstrated that cells from patient B were responsive to a
broad array of chemotherapeutic drugs, including antimetabolites methotrexate and fudarabine phosphate. Mitoxantron and paclitaxel as part of the
patient’s actual treatment and topotecan were also effective in these cells. The cells showed sensitivity to several classes of targeted agents, including
inhibitors of PI3K (AZD8482, buparlisib, GDC-0980, idelalisib, taselisib, and PIK-75) and HDAC (vorinostat and belinostat). Patient C: Tumor cells showed
resistance to most chemotherapeutics and targeted agents (as indicated in the heat map); high sensitivity was seen for the targeted agent trametinib,
a MEK inhibitor. Drug sensitivity was validated in our 3-D Matrigel system also using oxaliplatin and 5-FU in comparison with what the patient initially
received. Patient D: This patient harbors an APC mutation and a frameshift deletion. The tumor cells were sensitive to a small number of drugs, including
EGFR inhibitors, particularly for afatinib, and showed sensitivity to neratinib. Drug sensitivity was validated in our 3-D Matrigel system using oxaliplatin
and 5-FU in comparison with what the patient was initially treated with.
OF9 | CANCER DISCOVERY MAY 2017 www.aacrjournals.org
Downloaded from cancerdiscovery.aacrjournals.org on April 9, 2019. © 2017 American Association for Cancer
Research.
Single compound screen Single compound screen Single compound screen Single compound screen
Sensitivity Resistance
Sensitivity Resistance Resistance Resistance
Sensitivity
Sensitivity
Normalized AUC Normalized AUC
Normalized AUC
Normalized AUC
AUC Z-score AUC Z-score AUC Z-score AUC Z-score
% Cell viability % Cell viability
% Cell viability % Cell viability
Afatinib [IC50 ~ 0.004 µmol/L]
Published OnlineFirst March 22, 2017; DOI: 10.1158/2159-8290.CD-16-1154
Personalized Cancer Models to Guide Precision Medicine CRAESER H CIT RA EL
Patient A; uterine carcinosarcoma
Genomic alteration detected: PIK3CAmut, PTENmut
A 1 B High-throughput combination compound screen C Tumor organoid drug validation
3 1.2 Buparlisib
****
2 Vorinostat
1 IC30 Vorinostat-buparlisib
0.8
0
−1
Buparlisib
0.4
−2
−3
−4 0.0
0 1 2 3 4 −4 −2 −0 2
0
AUC fold change log[Agonist], µmol/L
Patient B; endometrial adenocarcinoma
Genomic alteration detected: PIK3CAmut, PTENdel, CTNNB1mut
D E High-throughput combination compound screen F Tumor organoid drug validation
1
3 1.2
**** Olaparib
2 Buparlisib
IC30 Buparlisib-olaparib
1 0.8
0
−1 0.4
� � Vorinostat
Olaparib
−2
−3 0.0
0 1 2 3 4 5 −4 −2 −0 2
0
AUC fold change log[Agonist], µmol/L
Patient C; colorectal cancer
Genomic alteration detected: KRAS mut, TP53 mut, APCdel
G 1 H High-throughput combination compound screen I Tumor organoid drug validation
3 1.2
Celecoxib
****
2 Trametinib
1 0.8 IC30 trametinib-celecoxib
0
− � Vorinostat 1 Afat�inib 0.4
� Celecoxib
− 2
− 3 0.0
0 1 2 3 4 5 6 7 −4 −2 −0 2
0 A U log[Agonist], µmol/L C f o l d c h a n g e
Patient D; colorectal cancer
Genomic alteration detected: APC mut, APC frameshift insertion
J 1 K High-throughput combination compound screen L Tumor organoid drug validation
3 1.2 Vorinostat
****
Afatinib
2
IC30 afatinib-vorinostat
1 0.8
0 T r a m e t i n i b
� �
�
– B 1 0.4 M S − 7 5 4 8 0 7
– 2
– 3 0.0
0.0 0.5 1.0 1.5 2.0 2.5 3.0 −4 −2 −0 2
0 A U log[Agonist], µmol/L C f o l d c h a n g e
Figure 6.  High-throughput combination drug screening to nominate potent drug combinations. Using a combination of the genomics (EXaCT-1)
and drug sensitivities from the primary drug screen, secondary drug screens were performed on the patient-derived tumor organoids from the same
four cases. ,A ,D G , and ,J Heat maps of the single therapy (left) and combination drug (right) screens depicting the relative sensitivity of the patients’
tumor cells from most resistant (red) to most sensitive (blue). Black dots indicate agents that were selected for validation. ,B ,E ,H and ,K Graphs of the
response of patient’s tumor cells to each compound in the library as a Z-score in the presence of the combination treatments (y-axis) and the fold change
of the IC50 identifed with each compound in combination as compared with the agent as a monotherapy. C , , F ,I and ,L The in vitro validation of selected
drugs in our 3-D Matrigel system. Patient A: A sensitizing drug screen using a PI3K inhibitor (buparlisib), as the investigational most advanced selec-
tive inhibitor of p110α/β/δ/γ did show an enhanced drug effect for HDAC inhibitors (vorinostat and belinostat; Supplementary Table S3). However, a
high-throughput combination drug screen using vorinostat (at its IC30) showed that many of the investigational PI3K/AKT pathway drugs were enhanced
compared with monotherapeutic use (Supplementary Table S4). A signifcant difference using vorinostat (IC30) with buparlisib in combination was seen
compared with buparlisib alone (2-way ANOVA: ****, P < 0.0001). Vorinostat also enhanced the effects of EGFR inhibitors in this patient’s tumor cells
(Supplementary Table S2). Patient B: Sensitizing screen with buparlisib for patient B enhanced the drug effects of drugs such as HDAC inhibitors and
olaparib (Supplementary Table S5). For the validation of the olaparib and buparlisib combination, we extended the drug assay up to 6 days and noticed a
signifcant difference compared with olaparib as monotherapy (2-way ANOVA: ****, P < 0.0001). Patient C: The combination drug screen using trametinib
has increased the effcacy of several compounds (Supplementary Table S6). As an example, celecoxib did not have an effect on cell survival as a single
compound activity but showed a signifcant effect in combination (2-way ANOVA: ****, P < 0.0001). Patient D: Afatinib-sensitizing drug screen for patient
D showed enhanced effects for HDAC and IGF1R inhibitors, which were confrmed in our 3-D Matrigel system (Supplementary Table S7). A signifcant
difference using vorinostat with afatinib (IC30) in combination was seen compared with vorinostat alone (2-way ANOVA: ****, P < 0.0001).
MAY 2017 CANCER DISCOVERY | OF10
Downloaded from cancerdiscovery.aacrjournals.org on April 9, 2019. © 2017 American Association for Cancer
Research.
Combination compound Combination compound Combination compound Combination compound
screen with afatinib screen with trametinib screen with buparlisib screen with vorinostat
Sensitivity Resistance Sensitivity Resistance Sensitivity Resistance
Resistance
Sensitivity
N o N r o
N m r o N a m r o l a m r i l a m z i l a e z i l d e z i A d e z U A d e C U A d C U A C U C
Combination AUC Z-score
C o C m o Combination AUC bZ-score m i b n i a n t a i t o i n o A n U A C U Z C - Z s - c s o c r o e r e
% C e l l v i a b i l i t y
% Cell viability % Cell viability % Cell viability
V o r i n o s t a t