Professional Documents
Culture Documents
OV
U
IMM
ERY
OX I I E S
J A X NGL E
OT T U D
CI T Y
A SI
CY T
KIN
ON N
S
O
ST E R
UD AT I
Y TO ELEASE EVALU IMMU
TEST EFFICACY AND
USE A SINGLE HUMAN PBMC DONOR USE MULTIPLE HUMAN PBMC DONORS
to perform dose ranging studies and to recapitulate donor-to-donor variability
to allow lead selection. you’d expect to find in the clinic.
Cancer Cell
Cancer Cell Best of 2020
Best of 2020
H A M A M AT S U C A M E R A S . C O M
Foreword
Cancer Cell provides a high-profile forum for major advances across cancer research and oncology. Cancer
Cell reaches a broad audience of cancer biologists and clinicians to facilitate the translation of biological
discoveries to clinical cancer care. Articles published in Cancer Cell cover a wide range of cancer research,
including clinical trials, translational research, and basic cancer biology studies. We introduced the Best of
Cancer Cell series in 2012 to highlight some of the exciting articles published in the previous year, and we are
pleased to now present the tenth installment, Best of Cancer Cell 2020, which will be presented at the virtual
2021 AACR annual meeting.
For this edition, we selected two reviews and eight full-length articles that strongly captured our readership’s
attention and are among our most-accessed content in 2020. In addition, these papers represent some of the
most exciting areas in cancer research: they allow us to better understand the complexity and heterogeneity
of cancer as well as its progression and response to therapies, and they have strong implications for patient
diagnosis, stratification, and treatment. These articles also showcase a range of cancer models, as well as
cutting-edge technologies and computational algorithms used in cancer research. Finally, we included one of
our series of Voices, in which several experts share techniques used to visualize cancer in the lab and in the
clinic.
We hope that you enjoy reading this collection. We recognize that no single measurement can be accurately
indicative of “the best” research papers over a given period of time. This is especially true for recently published
research, because the community hasn’t had enough time to fully appreciate the relative importance of the
findings. However, we are confident that you will find this collection informative and exciting.
We are grateful for the generosity of our sponsors, who helped to make this reprint collection possible. You can
access the entire Best of Cell Press collection online at www.cell.com/bestof. Visit www.cell.com/cancer-cell
to learn about the latest findings that Cancer Cell has had the privilege to publish and www.cell.com to find
other high-quality cancer-relevant papers published in the full portfolio of Cell Press journals. Please feel free
to contact us at cancer@cell.com to tell us about your latest work or to provide feedback. We look forward to
working with you in 2021 and beyond!
For information for the Best of Cell Press Series, please contact:
Jonathan Christison
Program Director, Best of Cell Press
e: jchristison@cell.com
p: 617-397-2893
t: @CellPressBiz
Cancer Cell ll
OriGene TrueMAB 8.25 X 11_Nature_PRINT.indd 1 12/19/18 2:47 PM
Accelerate Your Antibody Development
and Still Have Peace of Mind
Gain More Insights, Make Better Decisions.
Sartorius accelerates development of breakthrough therapeutics with
groundbreaking, comprehensive solutions for lab filtration, cell/protein
analysis, and more. Our products and technologies provide rapid,
multifactorial results to enable deeper insights and ensure the choice,
quality, and integrity of your biopharmaceutical product.
www.sartorius.com/mab-development
Cancer Cell
Best of 2020
Voices
Visualizing cancer Matthew Krummel, Kenneth Hu, Elisabeth G.E. de Vries,
Jacky G. Goetz, Diana Passaro, Sriram Subramaniam,
Tom Misteli, Melissa Skala, Jose Javier Bravo-Cordero,
and Julie L. Sutcliffe
Reviews
Intratumor heterogeneity: The Rosetta Stone of therapy Andriy Marusyk, Michalina Janiszewska, and Kornelia Polyak
resistance
Engineering CAR-T cells for next-generation cancer Mihe Hong, Justin D. Clubb, and Yvonne Y. Chen
therapy
Articles
Integrated molecular and clinical analysis of 1,000 Scott Ryall, Michal Zapotocky, Kohei Fukuoka, Liana Nobre,
pediatric low-grade gliomas Ana Guerreiro Stucklin, Julie Bennett, Robert Siddaway,
Christopher Li, Sanja Pajovic, Anthony Arnoldo,
Paul E. Kowalski, Monique Johnson, Javal Sheth,
Alvaro Lassaletta, Ruth G. Tatevossian, Wilda Orisme,
Ibrahim Qaddoumi, Lea F. Surrey, Marilyn M. Li,
Angela J. Waanders, Stephen Gilheeney, Marc Rosenblum,
Tejus Bale, Derek S. Tsang, Normand Laperriere,
Abhaya Kulkarni, George M. Ibrahim, James Drake,
Peter Dirks, Michael D. Taylor, James T. Rutka,
Suzanne Laughlin, Manohar Shroff, Mary Shago,
Lili-Naz Hazrati, Colleen D’Arcy, Vijay Ramaswamy,
Ute Bartels, Annie Huang, Eric Bouffet,
Matthias A. Karajannis, Mariarita Santi, David W. Ellison,
Uri Tabori, and Cynthia Hawkins
MYC drives temporal evolution of small cell lung cancer Abbie S. Ireland, Alexi M. Micinski, David W. Kastner,
subtypes by reprogramming neuroendocrine fate Bingqian Guo, Sarah J. Wait, Kyle B. Spainhower,
Christopher C. Conley, Opal S. Chen, Matthew R. Guthrie,
Danny Soltero, Yi Qiao, Xiaomeng Huang,
Szabolcs Tarapcsák, Siddhartha Devarakonda,
Milind D. Chalishazar, Jason Gertz, Justin C. Moser,
Gabor Marth, Sonam Puri, Benjamin L. Witt,
Benjamin T. Spike, and Trudy G. Oliver
ll
Sharpen your senses.
FINE SURGICAL INSTRUMENTS FOR RESEARCH™
Dendritic cell paucity leads to dysfunctional immune Samarth Hegde, Varintra E. Krisnawan, Brett H. Herzog,
surveillance in pancreatic cancer Chong Zuo, Marcus A. Breden, Brett L. Knolhoff,
Graham D. Hogg, Jack P. Tang, John M. Baer, Cedric Mpoy,
Kyung Bae Lee, Katherine A. Alexander, Buck E. Rogers,
Kenneth M. Murphy, William G. Hawkins, Ryan C. Fields,
Carl J. DeSelm, Julie K. Schwarz, and David G. DeNardo
Treatment-induced tumor dormancy through Kari J. Kurppa, Yao Liu, Ciric To, Tinghu Zhang,
YAP-mediated transcriptional reprogramming Mengyang Fan, Amir Vajdi, Erik H. Knelson, Yingtian Xie,
of the apoptotic pathway Klothilda Lim, Paloma Cejas, Andrew Portell,
Patrick H. Lizotte, Scott B. Ficarro, Shuai Li, Ting Chen,
Heidi M. Haikala, Haiyun Wang, Magda Bahcall,
Yang Gao, Sophia Shalhout, Steffen Boettcher,
Bo Hee Shin, Tran Thai, Margaret K. Wilkens,
Michelle L. Tillgren, Mierzhati Mushajiang, Man Xu,
Jihyun Choi, Arrien A. Bertram, Benjamin L. Ebert,
Rameen Beroukhim, Pratiti Bandopadhayay, Mark M. Awad,
Prafulla C. Gokhale, Paul T. Kirschmeier, Jarrod A. Marto,
Fernando D. Camargo, Rizwan Haq, Cloud P. Paweletz,
Kwok-Kin Wong, David A. Barbie, Henry W. Long,
Nathanael S. Gray, and Pasi A. Jänne
Predicting drug response and synergy using a deep Brent M. Kuenzi, Jisoo Park, Samson H. Fong,
learning model of human cancer cells Kyle S. Sanchez, John Lee, Jason F. Kreisberg, Jianzhu Ma,
and Trey Ideker
Conserved interferon-γ signaling drives clinical response Catherine S. Grasso, Jennifer Tsoi, Mykola Onyshchenko,
to immune checkpoint blockade therapy in melanoma Gabriel Abril-Rodriguez, Petra Ross-Macdonald,
Megan Wind-Rotolo, Ameya Champhekar, Egmidio Medina,
Davis Y. Torrejon, Daniel Sanghoon Shin, Phuong Tran,
Yeon Joo Kim, Cristina Puig-Saus, Katie Campbell,
Agustin Vega-Crespo, Michael Quist, Christophe Martignier,
Jason J. Luke, Jedd D. Wolchok, Douglas B. Johnson,
Bartosz Chmielowski, F. Stephen Hodi, Shailender Bhatia,
William Sharfman, Walter J. Urba, Craig L. Slingluff,
Jr., Adi Diab, John B.A.G. Haanen, Salvador Martin Algarra,
Drew M. Pardoll, Valsamo Anagnostou, Suzanne L. Topalian,
Victor E. Velculescu, Daniel E. Speiser, Anusha Kalbasi,
and Antoni Ribas
See sequencing in a new light
Explore the NextSeq™ 2000
ZZZLOOXPLQDFRP1H[W6HT
For Research Use Only. Not for use in diagnostic procedures.
© 2020 Illumina, Inc. All rights reserved.
The PD-1/PD-L1-checkpoint restrains T cell immunity Floris Dammeijer, Mandy van Gulijk, Evalyn E. Mulder,
in tumor-draining lymph nodes Melanie Lukkes, Larissa Klaase, Thierry van den Bosch,
Menno van Nimwegen, Sai Ping Lau, Kitty Latupeirissa,
Sjoerd Schetters, Yvette van Kooyk, Louis Boon,
Antien Moyaart, Yvonne M. Mueller, Peter D. Katsikis,
Alexander M. Eggermont, Heleen Vroman,
Ralph Stadhouders, Rudi W. Hendriks, Jan von der Thüsen,
Dirk J. Grünhagen, Cornelis Verhoef, Thorbald van Hall,
and Joachim G. Aerts
On the cover: Featured on the cover is a cross-section of visually striking images published on
the cover of Cancer Cell in 2020. Images are courtesy of (from left to right): Kristen E. Labbe,
Catríona M. Dowling, Hua Zhang, and Mahnoor Arshad (volume 37, issue 1); Bianca Dunn (volume
37, issue 6); Samantha Walker (volume 38, issue 2); Dylan Mortimer (volume 38, issue 1); and
Himanshu Brahmbhatt, Jennifer MacDiarmid, and Martin Hale (volume 37, issue 3).
APRIL 10-15
MAY 17-21
We look forward to
seeing you in 2021!
★ Become a Member!
Join the AACR and receive a discount on registration.
AACR.ORG ★ #AACR21
AACR.ORG/AACR2021
2001015Yy
WITH BD AS YOUR PARTNER IN IMMUNO-ONCOLOGY FROM BENCH TO BEDSIDE. BD is your partner in immuno-oncology
from discovery research to clinical applications, with a range of high-performance solutions designed to give you high-quality,
reliable research data for even the most complex experiments. More solutions. More answers. More data you can trust. Discover
more with BD at your side. Discover the new BD.
For Research Use Only. Not for use in diagnostic or therapeutic procedures.
BD, San Jose, CA, 95131, U.S.
BD and the BD Logo are trademarks of Becton, Dickinson and
Company or its affiliates. © 2019 BD. All rights reserved. (0619)
Recommend Trends in Cancer
to your librarian today!
With your recommendation, your library can review the opportunity
for ongoing access to seminal articles in Trends in Cancer.
Voices
Visualizing Cancer
Imaging has had a profound impact on our ability to understand and treat cancer. We invited some experts to
discuss imaging approaches that can be used in various aspects of cancer research, from investigating the
complexity and diversity of cancer cells and their environments to guiding clinical decision-making.
Cancer Cell 38, December 14, 2020 ª 2020 Elsevier Inc. 753
ll
Voices
Review
*Correspondence: kornelia_polyak@dfci.harvard.edu
https://doi.org/10.1016/j.ccell.2020.03.007
Advances in our understanding of molecular mechanisms of tumorigenesis have translated into knowledge-
based therapies directed against specific oncogenic signaling targets. These therapies often induce dra-
matic responses in susceptible tumors. Unfortunately, most advanced cancers, including those with robust
initial responses, eventually acquire resistance to targeted therapies and relapse. Even though immune-
based therapies are more likely to achieve complete cures, acquired resistance remains an obstacle to their
success as well. Acquired resistance is the direct consequence of pre-existing intratumor heterogeneity and
ongoing diversification during therapy, which enables some tumor cells to survive treatment and facilitates
the development of new therapy-resistant phenotypes. In this review, we discuss the sources of intratumor
heterogeneity and approaches to capture and account for it during clinical decision making. Finally, we
outline potential strategies to improve therapeutic outcomes by directly targeting intratumor heterogeneity.
Sources of Intratumor Heterogeneity external or internal mutagen exposure. These include UV-related
Cellular phenotypic heterogeneity within tumors is a complex, mutagenesis in skin cancers; tobacco-related mutagenesis in
multifactorial phenomenon, which integrates genetic, epige- oral, lung, and bladder cancers; reactive oxygen species-
netic, and environmental inputs (Figure 1). induced mutagenesis in a wide range of malignancies, and de-
Genetic Heterogeneity fects in DNA repair mechanisms, which have been detected in
Genetic heterogeneity is the most studied and best understood certain tumors (e.g., deficiencies in DNA mismatch repair and
aspect of intratumor heterogeneity (ITH), although our under- APOBEC pathways) (Alexandrov et al., 2013; Fox et al., 2013;
standing is still far from being sufficiently complete (McGranahan Kandoth et al., 2013).
and Swanton, 2017). Despite mounting challenges, gene and Small-scale genetic mutations remain best understood and
mutation-centric focus remains at the core of molecular easiest to study, but they represent only a tip of the iceberg of
oncology, and the majority of cancer researchers would prob- genetic heterogeneity. The vast majority of spontaneous human
ably still agree with the famous statement by Bert Vogelstein: cancers display aneuploidy, which is linked with chromosomal
‘‘The revolution in cancer research can be summed up in a single instability (CIN)––an increased rate of genomic mutational errors
sentence: cancer is, in essence, a genetic disease’’ (Vogelstein involving loss, gains, and translocations of large fragments of
and Kinzler, 2004). Discovering tumor-associated genetic muta- genomic DNA (Bakhoum and Cantley, 2018). Aneuploid cells
tions and interrogating their functional and clinical impact has commonly emerge following whole-genome doubling (WGD)
been the major focus of cancer research over the last few due to mitotic failure leading to tetraploidization (Dewhurst
decades, and the recent revolution in DNA sequencing technol- et al., 2014). WGD is an early event commonly observed across
ogies enabled an outpour of studies documenting startling intra- most human cancer types and it is associated with poor prog-
tumor genetic heterogeneity. nosis (Bielski et al., 2018). Compared with point mutations, these
Even though eukaryotic cells replicate their DNA with large-scale genomic events are much more likely to impact
astounding fidelity, the mechanism is not entirely error free. cellular phenotypes. Although reduced cell fitness is the most
Every time a cell divides, a few mutational errors in the form of common outcome of large-scale genomic changes, CIN enables
nucleotide substitutions and small deletions are introduced much faster rates of genomic diversification, thus speeding up
even in the absence of internal and external mutagens (Zhang evolution during both tumor development and cancer therapies
and Vijg, 2018). Owing to the constant turnover of tumor cells (Tang and Amon, 2013). At the same time aneuploid cells also in-
and the large size of tumor cell populations, some of these fluence their microenvironment and may trigger a specific anti-
stochastic mutational hits inevitably affect genes with known tumor immune response, which in turn may facilitate immune
cancer relevance, leading to the activation of oncogenes and evasion through upregulation of immune suppressive mecha-
inactivation of tumor suppressors. Whereas the sufficiency of nisms (Senovilla et al., 2012).
these baseline replication error rates to account for carcinogen- The functional relevance of genomic heterogeneity in tumor
esis remains a subject of debate (Beckman and Loeb, 2006; progression and therapy resistance remains poorly understood.
Tomasetti and Vogelstein, 2015; Tomlinson et al., 1996), many The prevalent approach is to reduce the impact of large-scale
tumors display increased mutation rates, due to either increased gains to changes in copy numbers (and related expression
Cancer Cell 37, April 13, 2020 ª 2020 Elsevier Inc. 471
Cancer Cell
Review
Figure 1. Sources of Intratumor
heterogeneity
Intratumor heterogeneity represents integration of
inputs from genetic, phenotypic, and microenvi-
ronmental heterogeneity, in turn increasing the
odds of both pre-existence of tolerant and resis-
tant subpopulations, and the ability to evolve new
adaptations.
Review
tumor evolution, which then enable the design of improved (Gaiti et al., 2019) have demonstrated that inferring tumor evolu-
therapeutic strategies for heterogeneous tumors. tion based on genetic and DNA methylation patterns largely
Epigenetic Heterogeneity overlaps. Furthermore, quantitative measures of ITH in Barrett’s
Studies on tumor heterogeneity have primarily focused on esophagus based on DNA methylation and genetic mutations
cancer cell genomes even though selection forces act on were both predictive of the progression to esophageal carci-
phenotypes rather than genotypes. Despite the critical impor- noma, implying that diversity is an inherent tumor characteristic
tance of cancer-associated mutations on clinically relevant and heritable epigenetic changes have similar impact on tumor
phenotypic features, such as responses to growth signaling, evolution as genetic ones (Merlo et al., 2010).
proliferation, and death, epigenetic (we use the term in a broader ITH for DNA methylation has consistently been observed in
sense, covering all non-genetic phenotypic determinants) mech- regulatory regions that affect the transcription of genes relevant
anisms have a greater impact on tumor cell phenotypes (Flava- to the disease process. For example, in prostate cancer high ITH
han et al., 2017). For example, the gene expression and epige- is observed for enhancers bound by the androgen receptor
netic profiles of genetically abnormal CD44+CD24– progenitor- (Brocks et al., 2014), a ligand-dependent transcription factor
like cancer cells isolated from primary breast tumors shows driving the expression of many genes essential for the survival
closer resemblance to the profiles of normal CD44+CD24– mam- and proliferation of prostate tumor cells. Similarly, in CLL binding
mary epithelial cells than to that of more differentiated sites for transcription factors with known functional relevance,
CD44 CD24+ cancer cells from the same tumor; the same is such as NFKB1 and MYBL1 displayed lower epimutation poten-
true for CD44 CD24+ cells (Shipitsin et al., 2007). These data tially due to the protection of these sites from epimutations by
suggest that differentiation state-related epigenetic programs transcription factor binding (Gaiti et al., 2019). Advances in tech-
have a dominant impact in shaping phenotypes compared with nologies have enabled the single-cell multiomic profiling of tu-
cancer-related genetic aberrations. Similar observations have mors that allows the direct analyses between genetic and epige-
been made in brain tumors, which represent one of the best ex- netic alterations and gene expression profiles. A recent study
amples of cancer stem cell-driven malignancies (Liau et al., performing single-cell RNA sequencing and single-cell assay
2017; Neftel et al., 2019; Patel et al., 2014). However, more differ- for transposase-accessible chromatin sequencing (ATAC-seq)
entiated cancer cells can give rise to stem cell-like cancer cells profiling of mixed-phenotype acute leukemias have demon-
through epithelial-to-mesenchymal transition (EMT) triggered strated common malignant signatures despite extensive hetero-
by certain microenvironmental signals, such as hypoxia (Mohlin geneity across patients and within individual cases, and identi-
et al., 2017) or interaction with stromal cells (Karnoub et al., fied the RUNX1 transcription factor as a key regulator of these
2007). Thus, tumor cells display high epigenetic and phenotypic programs (Granja et al., 2019).
plasticity. This phenotypic plasticity is likely to be key for the phe- Because tumor cells display clear parallels with differentiation
nomenon of drug-tolerant persistence, as well as the ability of hierarchy-related phenotypic heterogeneity in normal tissues,
cells to acquire resistance through stable non-genetic changes the cancer stem cell framework has been, and still remains, a
in gene expression, as we will elaborate below. popular paradigm to conceptualize the ITH. However, pheno-
Although epigenetic heterogeneity could be key in acquisition typic ITH cannot be reduced to differentiation hierarchies, as tu-
of traits of profound clinical importance, such as therapeutic mor cells display far greater cell-to-cell variability compared with
resistance (Shaffer et al., 2017) or metastatic dissemination their differentiation state normal counterparts (Jenkinson et al.,
(Roe et al., 2017), accounting for non-genetic heterogeneity is 2017; Landau et al., 2014). This increased phenotypic variability
far more challenging since, in contrast to genetic heterogeneity, likely reflects the impact of intratumor genetic heterogeneity, as
the phenotypes of tumor cells can be highly plastic. Epigeneti- well as greater variability in contextual signals provided by aber-
cally defined phenotypic traits range from essentially hard-wired rant and less-structured (compared with normal tissues) tumor
ones, which can be conceptualized as epi-mutations, such as microenvironments. In addition to increased plasticity and the
silencing of key tumor suppressor genes, mediated by DNA hy- diversifying inputs of genetic and microenvironmental heteroge-
permethylation on one end of the spectrum, to noise-driven cell- neity, this increased cell-to-cell variability might also be a
to-cell differences that dissipate within a few cell divisions on the consequence of global epigenetic changes, generally observed
other end of the spectrum (Marusyk et al., 2012). Whereas these in cancer cell epigenomes (such as global hypomethylation of
two extremes are conceptually the easiest to deal with, the CpG islands) (Feinberg, 2007), and can be summarized as higher
majority of non-genetic phenotypic heterogeneity, reflecting entropy of the cancer epigenome (Jenkinson et al., 2017).
integration of microenvironmental inputs and stochastic events Microenvironmental Heterogeneity
and differentiation status, falls in the gray zone in terms of Although heterogeneity of genotypes within tumors translates
heritability. into diversity of phenotypes, phenotypes are not simple func-
Studies assessing epigenetic heterogeneity within tumors tions of genotypes. Indeed, all normal cells share nearly identical
have been largely focusing on DNA methylation, since this is wild-type genomes, which encode highly diverse phenotypic
technically less challenging to measure even at single-cell level manifestations. Phenotypic diversity in normal cells reflects
than chromatin modification (Mazor et al., 2016). Because of developmental processes triggered by responses to microenvi-
the reversible nature of epigenetic modifications, it was not clear ronmental cues. Whereas tumor cells often display aberrant re-
if they can be used to define subclones, track tumor evolution, sponses to microenvironmental signals (e.g., suppression of
and assess intratumor subclonal and cellular heterogeneity. death from anoikis), they still retain a large repertoire of normal
However, studies in prostate cancer (Brocks et al., 2014), glioma responses. An extreme example is the ability of some tumor cells
(Mazor et al., 2015), and chronic lymphocytic leukemia (CLL) to become a functional part of normal tissues, when embedded
Review
Figure 2. Microenvironmental
Heterogeneity
In normal tissues (such as normal breast depicted
here), most epithelial cells experience similar
concentrations of nutrients, oxygen, and growth
factors. However, structural disorganization of
epithelia, stroma, and vasculature leads to signifi-
cant inequality in concentrations of these factors
(one factor is illustrated). Furthermore, differences
in diffusion and consumption rates of different
factors combinatorially increase variability in
microenvironmental cues directly influencing
phenotypic heterogeneity and creating distinct
selection forces.
Review
example, in non-small cell lung cancer (NSCLC) assessing muta- heterogeneity for HER2, and several clinical trials are ongoing
tions and TCR diversity in different regions of the same tumor (NCT02326974) to determine if this could potentially improve
demonstrated that the abundance of expanded intratumor ubiq- the clinical management of patients. Preliminary analysis of the
uitous TCRs is associated with the number of nonsynonymous data suggests that patients with higher pre-treatment cellular
mutations (Joshi et al., 2019). At the same time the number of heterogeneity for ERBB2 amplification are less likely to have a
expanded regional TCRs correlated with the number of regional complete pathologic response to neoadjuvant pertuzumab and
nonsynonymous mutations. T-DM1 treatment (Filho et al., 2019).
In serous ovarian cancer metastases, higher epithelial CD8+ Quantitative assessment of cell-to-cell variation in the expres-
T cell infiltration was associated with lower tumor ITH presum- sion of a therapeutic target at the protein level is much harder to
ably due to immune-mediated elimination of certain subclones evaluate, as it is usually measured using immunostaining and
present as minor subpopulations (Zhang et al., 2018). Subclonal manually scored by pathologists. Scoring discordance between
neoepitopes have been shown to be more immunogenic than individuals and variation between different batches of staining
clonal ones (Jimenez-Sanchez et al., 2017), potentially explain- contribute to limited reporting of the observed heterogeneity.
ing these results. In some cases, immune evasion was due to However, technological advances in multiplex immunostaining
HLA (human leukocyte antigen) loss of heterozygosity in the tu- techniques, digital pathology, and machine learning are paving
mor cells coupled with upregulation of immune checkpoint inhib- the way to more accurate reporting of ITH.
itors. Interestingly, while tumor regions with high CD8+ and CD4+ Decreased expression of the target is only one of the many
T cell densities had high TCR diversity, this was not associated mechanisms by which tumors evade therapy. Many known resis-
with tumor cell ITH implying that T cell and tumor cell cellular tance mechanisms are related to genetic alterations that are iden-
heterogeneity are related but distinct features. tified based on unbiased genome-wide analyses. Several compu-
Thus, phenotypes of tumor cells are shaped by an integration of tational approaches have been developed to use next-generation
genetic, epigenetic, and microenvironmental inputs (Figure 1). sequencing read frequency to assess ITH and infer clonal evolu-
Whereas we have introduced these inputs separately, they are tion of a tumor. Multiregional sequencing of renal carcinoma, glio-
highly interrelated. For example, microenvironmental inflammation blastoma, and lung cancer have all showed significant divergence
has been linked with increased CIN and increased epigenetic plas- between distinct areas of the same tumor (Gerlinger et al., 2012;
ticity (Colotta et al., 2009; Grivennikov et al., 2010; Korkaya et al., Thrane et al., 2018; Yates et al., 2015; Zhang et al., 2014).
2011). On the other hand, CIN has been linked to the induction of Although evolutionary trajectories can be drawn between distinct
inflammatory signaling through the cGAS-STING anti-viral clones at different sites, clinical benefit of these analyses remains
pathway (Bakhoum and Cantley, 2018), thus modulating microen- to be determined. The relatively high cost of exome-sequencing
vironments, including immune responses. The resulting pheno- makes it challenging to perform these analyses for all patients. A
typic ITH translates to heterogeneity in the sensitivity of tumor cells large clinical trial (NCT01888601) was initiated in 2014 for NSCLC
to anti-cancer drugs and cytotoxic immunity. In addition, some of patients, aiming at multiregional and longitudinal sampling of
the heterogeneous features of the tumor microenvironment can primary and recurrent tumors to determine the impact of ITH on
directly modulate therapeutic responses as described below. clinical outcomes. Initial results from this study show that ITH for
copy number alterations and genomic instability are associated
Quantitative Assessment of ITH with increased risk of recurrence or death (Jamal-Hanjani et al.,
ITH in Tissue Samples 2017). Moreover, immune cell heterogeneity within tumors points
Even though all types of ITH have been linked to poor patient out- to a correlation between T cell clonality and tumor antigen diver-
comes and therapeutic resistance in many different tumor types, sity in different regions of the same tumor (Reuben et al., 2017).
assessing heterogeneity in human tissue samples is still a major Advances leading to decrease of sequencing costs and stream-
challenge, which partially explains why ITH is still not commonly lined bioinformatic analysis will be required for the more wide-
used to guide clinical decisions. One of the best examples for spread use of genome-wide approaches for routine ITH assess-
this is ITH for ERBB2 (encoding HER2) amplification in HER2+ ment for patient prognostication.
breast cancer driven by oncogenic HER2 signaling. HER2-tar- Regional sampling of heterogeneous tumors remains one of
geted therapies have been very successful for HER2+ tumors, the largest challenges in clinical diagnostics. In certain tumor
but they are only effective when the tumor cells uniformly ex- types, such as lung or breast cancer, biopsies from several
press and are dependent on HER2. However, in many cases distinct regions may be taken. However, in glioblastoma and
HER2 expression and ERBB2 amplification is not homogeneous pancreatic cancer, additional sample collection is associated
within tumors and this has been associated with shorter disease- with higher risk for the patient. Multiregional sequencing has
free survival (Rye et al., 2018). Assessing HER2 status by immu- invariably showed regional heterogeneity in different tumor
nohistochemistry and fluorescence in situ hybridization (FISH) types, raising many questions. How representative is a single
has long been used in the clinic for the identification of patients biopsy? How many biopsies would we need to collect to cover
who will most likely benefit from HER2-targeted therapies. How- the extent of heterogeneity of a large highly aggressive tumor?
ever, due to the increasing recognition of ITH, HER2 heterogene- An important tool in addressing these seemingly impossible is-
ity, defined as HER2 positivity by FISH in >5% and <50% of sues is ‘‘liquid biopsy’’: sampling blood and other body fluids.
tumor cells, has also been incorporated into the report. Consid- ITH in Liquid Biopsies
ering that HER2 status is determined by automated counting of Blood drawn from cancer patients contains circulating tumor
FISH signal in 50 single cells, this assay could relatively easily cells (CTCs) and circulating tumor DNA (ctDNA) (Rossi and Igna-
be expanded into a more quantitative assessment of cellular tiadis, 2019). Liquid biopsies can easily be repeated weekly or
Review
Figure 3. Evolution of Therapeutic
Resistance
Heterogeneous primary tumors likely contain
subpopulations of cancer cells with pre-existing
partial or full resistance to therapy due to genetic or
epigenetic mechanisms. Genetically, resistant
cancer cells can also outgrow from epigenetically
plastic persisters in part due to therapy-induced
alterations. Interaction with the microenvironment
also contributes to phenotypic heterogeneity and
associated therapeutic resistance within tumors.
In most advanced-stage tumors, both genetic and
epigenetic resistance is likely to be present, lead-
ing to multiple different resistance mechanisms.
Review
with mutations in epidermal growth factor receptor (EGFR) and cancer types, including breast and brain tumors (Hinohara
anaplastic lymphoma kinase (ALK), or melanomas with mutant et al., 2018; Hong et al., 2019; Knoechel et al., 2014; Liau
BRAF (Zhang et al., 2009). Despite the often-spectacular initial et al., 2017; Risom et al., 2018; Sharma et al., 2010; Shu et al.,
tumor responses, and continuous development of more efficient 2016). At least in some cases, phenotypic transition toward
inhibitors, resistance emerges with near inevitability in drug tolerance might be induced by the treatment (Goldman
advanced, metastatic cancers. In many cases of clinical relapse, et al., 2015; Pisco and Huang, 2015), although the ability to un-
resistance is associated with specific genetic mutations, such as dergo phenotypic transition is likely to be limited to only certain
genomic amplification of the therapy target, point mutations that subpopulations of tumor cells (Hong et al., 2019; Shaffer
allosterically reduce the ability of the drug to block enzymatic et al., 2017).
activity, or amplification/mutation of other genes, enabling tumor Drug tolerance is often thought as a distinct, well-defined
cells to sustain oncogenic signaling (Lovly and Shaw, 2014). phenotypic state that might be defined by similar molecular un-
Most of these mutational changes are considered to be sufficient derpinnings across multiple cancer types, such as dependency
to confer full resistance, underlying clinical relapse. In this case, on certain metabolic pathways (Hangauer et al., 2017). However,
probability of pre-existing or de novo resistance could be our lineage tracing-based studies revealed that tolerance toward
inferred using mathematical models, involving target population ALK inhibitors develops from heterogeneous pre-existent sub-
size (numbers of tumor cells capable of self-renewal), mutational populations which differ in their ability to survive or grow under
probability, and proliferation/death rates (Bozic et al., 2013; Foo different inhibitors (Vander Velde et al., 2019). Likewise, func-
and Michor, 2014; Wodarz and Komarova, 2005). Assessing tional phenotypic heterogeneity was recently implied in
tumors at different stages of progression and building chemo-resistance in pancreatic cancers (Seth et al., 2019), sug-
mathematical models based on these data could also be used gesting that tolerance might reflect ITH in general, rather than a
to predict probable tumor evolutionary paths toward resistance well-defined phenotypic state. At the same time, certain
(Angelova et al., 2018; Khan et al., 2018). The reality, however, epigenetic states may confer ‘‘universal’’ drug resistance. For
might be more complicated, as development of resistance might example, high expression and activity of the KDM5 family of his-
be multifactorial (Hong et al., 2019; Shaffer et al., 2017; Vander tone demethylases has been linked to multiple different types of
Velde et al., 2019), in which case pre-existing diversity and therapeutic resistance in multiple different cancers. In lung can-
ongoing diversification, which fuel natural selection, rather than cer, high KDM5A expression is associated with EGFR inhibitor
presence of specific mutations, might be essential for the evolu- resistance (Sharma et al., 2010); in melanoma, quiescent cells
tion of resistance (Merlo et al., 2006). The relationship between required for tumor propagation that also play a role in resistance
genetic diversity and evolution is more complicated for copy to chemotherapeutic agents and BRAF inhibitors have high
number variations due to CIN: while providing a powerful source KDM5B levels (Roesch et al., 2013). A recent study in breast can-
of diversification, the majority of large-scale chromosomal cer investigating mechanisms by which high KDM5B activity
changes are disadvantageous (Tang and Amon, 2013). Both leads to endocrine resistance determined that higher KDM5B
experimental studies and analysis of clinical data support the levels are associated with higher cell-to-cell transcriptomic het-
existence of a ‘‘sweet spot’’ of CIN-related diversity, where erogeneity (Hinohara et al., 2018). Thus, mutant and abnormally
intermediate levels, which, balance evolvability with the ability expressed epigenetic regulators may influence tumor progres-
to maintain fitness of tumor cell populations, are most dangerous sion and therapeutic responses via their effect on cellular epige-
(Andor et al., 2016; Godek et al., 2016). This constraint most netic and phenotypic heterogeneity and, therefore, may repre-
likely does not apply to copy number changes mediated by sent good combination agents even if their single-agent
ecDNA, which avoids fitness cost of large-scale chromosomal efficacy is modest.
changes, while providing a highly dynamic focused diversifica- Whereas drug tolerance is likely responsible for the ability of
tion mechanism, enabling populations of tumor cells to quickly tumor cells to survive therapy, leading to minimal residual dis-
find optimum under therapy-imposed selection pressures ease, clinical relapse requires acquisition of stronger resistance
(Nathanson et al., 2014). phenotypes, capable of maintaining net positive tumor growth in
Despite the clear evidence of genetic mechanisms of resis- the presence of treatment. Bona fide resistance can develop
tance, in many cases no known genetic drivers could be identi- from drug-tolerant cells through acquisition of new genetic mu-
fied. Whereas it is formally possible that resistance in these tations, thus requiring new genetic diversification (Figure 3).
cases could be attributed to uncharacterized mutations, com- Notably, acquisition of EGFR T790M gatekeeper mutations by
monality of long remission times, and a growing body of exper- tolerant cells leads to more robust tumor cell phenotypes. These
imental evidence suggests that resistance can arise through cells are less sensitive to T790M targeting third-generation
epigenetic mechanisms, involving semi-stable changes in gene EGFR inhibitors compared with T790M mutations on the back-
expression mediated by chromatin remodeling (Hinohara et al., ground of therapy naive cells (Hata et al., 2016). By maintaining
2018; Knoechel et al., 2014; Liau et al., 2017; Risom et al., a reservoir of viable tumor cells, tolerance enables acquisition
2018; Sharma et al., 2010; Shu et al., 2016). A seminal paper of multiple distinct resistance mechanisms, thus supporting
by Sharma et al. (2010) implied stochastic, reversible epigenetic more heterogeneous relapse, which reduces chances of sec-
transition into a distinct drug-tolerant phenotypic state, enabling ond-line therapy success. Similarly, in colorectal carcinomas
tumor cells to survive therapy, but incapable of supporting treatment with EGFR and BRAF inhibitors downregulates DNA
robust net positive growth rates to drive tumor relapse. A number repair pathways and upregulates error-prone polymerases in
of subsequent studies demonstrated the relevance of this phe- drug-tolerant persister cells, therefore increasing mutation rates
nomenon toward multiple therapeutic modalities across many and the likelihood of resistance (Russo et al., 2019).
Review
On the other hand, full resistance can also develop from vironmental diversification remains poorly explored, its inputs
weakly resistant subpopulations through non-genetic mecha- likely shape both genetic and epigenetic diversification.
nisms. A case in point is a recent study on acquisition of resis- Sufficiently complete understanding of resistance cannot be
tance to BRAF inhibitors, where transient expression heteroge- accomplished exclusively with mainstream reductionistic meth-
neity in resistance-associated genes, such as AXL and EGFR, odological approaches, but instead requires integrative studies
enabled survival of subpopulations of tumor cells, but develop- and development of new tools.
ment of fully resistant phenotypes involved acquisition of multi- Heterogeneity in Therapy Resistance––Immunotherapy
ple, partially coordinated gene expression changes (Shaffer Immune surveillance is one of the key microenvironmental
et al., 2017). A similar phenomenon has been described in breast constrains that all tumors must overcome in order to progress
and lung cancers (Hong et al., 2019; Vander Velde et al., 2019). and grow. Immunoediting refers to the process by which the im-
These studies shed a new light on the association between mune system impacts tumor evolution and it is classified into
‘‘stemness’’ and resistance. Although EMT and stemness are three phases: elimination, equilibrium, and escape (Dunn et al.,
often considered to be proximal causes of therapy resistance 2004). Studies in experimental models and patients with
(Holohan et al., 2013; Singh and Settleman, 2010), phenotypic compromised immune system have demonstrated that most
plasticity per se might be the true culprit, by enabling cancer neoplastic cells are eliminated by the immune system before
cells to access resistance phenotypes by rewiring of gene regu- they become clinically relevant. However, tumor cells can have
latory networks (Pisco and Huang, 2015). variable expression of neoantigens and may display differences
Whereas the majority of studies focus on cell-autonomous in signaling pathways related to immunity, thus ITH for these
resistance mechanisms, a growing body of evidence suggests traits will limit the efficacy of antitumor immune responses. As
that cell-to-cell and microenvironmental interactions could a consequence, all clinically relevant tumors have already
dramatically impact drug sensitivity. For example, growth factors escaped immune surveillance by various mechanisms.
that can be produced by both tumor cells and normal cells within Immunotherapy is thought to be relatively unaffected by ITH,
tumor microenvironment have been shown to partially or since tumor neoantigens are usually not functional cancer
completely de-sensitize tumor cells to many types of tyrosine drivers, thus they are not expected to be impacted by cancer
kinase inhibitors across numerous cancer types (Wilson et al., therapy. However, recent data demonstrate that ITH influences
2012). Similarly, drug sensitivity could be dramatically reduced the success of immunotherapy as well. Acquired resistance to
by signaling, mediated by interaction of tumor cells with the immunotherapy can develop due to ITH for neoantigens, antigen
ECM (Hirata et al., 2015). Whereas in typical in vitro studies, presentation, and interferon signaling (Havel et al., 2019; Kalbasi
every tumor cell can be equally impacted, microenvironmental and Ribas, 2019). Tumor mutational burden (TMB), defined as
heterogeneity in vivo likely leads to heterogeneity in environmen- the number of somatic mutations within tumors, is one of the
tally mediated therapy resistance, where only some of the tumor strongest predictors of response to immune checkpoint inhibi-
cells can evade the impact of therapy (Marusyk et al., 2016). tors, with high TMB in general associated with better response
Although the impact of microenvironmentally mediated resis- (Ribas and Wolchok, 2018). However, higher TMB can also in-
tance on relapse remains unclear (as tumor cells remain sensitive crease ITH and recent data in both experimental models and
outside of the borders of protective niches), the phenomenon patients show that tumors with high subclonal ITH are less likely
likely contributes to residual disease, providing reservoir of tu- to respond to immune checkpoint inhibitors than more homoge-
mor cells capable of developing cell-autonomous resistance neous tumors (Anagnostou et al., 2017; Gejman et al., 2018;
mechanisms (Hirata and Sahai, 2017; Meads et al., 2009). The McGranahan et al., 2016; Milo et al., 2018; Wolf et al., 2019).
link between microenvironmental heterogeneity is not limited to Resistance to immunotherapy due to ITH could be due to multi-
paracrine signals, cell-to-cell, and cell-ECM contact. Given the ple mechanisms. Higher ITH is associated with higher risk of hav-
well-documented impact of hypoxia, acidity, and nutrient status ing a resistant clone. At the same time, subclonal mutations may
on therapy sensitivity (Gillies et al., 2012), metabolic microenvi- not trigger such an effective immune response as clonal ones
ronmental heterogeneity likely provides an important modulator (Anagnostou et al., 2017; Gejman et al., 2018; McGranahan
of tumor responses to therapies, a notion supported by in silico et al., 2016; Milo et al., 2018; Wolf et al., 2019), although there
modeling (Robertson-Tessi et al., 2015). are also data suggesting that subclonal neoantigens are more
A standard approach for a biomedical research paper entails immunogenic (Jimenez-Sanchez et al., 2017). Thus, ITH has
reducing a phenomenon, such as therapy resistance, to a single prognostic and predictive value for immunotherapy as well.
mechanism. However, resistance might entail a combined Microenvironmental and spatial heterogeneity are of key
impact of multiple mechanisms operating both within single cells importance for immune therapies as well. For example, attrac-
(Hong et al., 2019; Shaffer et al., 2017; Vander Velde et al., 2019) tion of immune cells is mediated by paracrine action of chemo-
and across different functionally distinct subpopulations (Cha- kines and cytokines, key suppressors of immune responses
bon et al., 2016; Piotrowska et al., 2015). Thus, the development CD73 and IDO1 act in a paracrine manner, and hypoxia (Noman
of clinical resistance is likely to involve not only different sources et al., 2015) and acidification (Pilon-Thomas et al., 2016) can
of pre-existing ITH, but also different inputs for additional diver- potently suppress cytotoxic immunity. The location of immune
sification. For example, resistance to EGFR inhibitors in a cell line cells within tumors is also highly heterogeneous, which is likely
model of EGFR mutant lung cancer can develop from both pre- influenced by stromal cells and is associated with probability
existing resistance-conferring mutations, but also from drug- of response to immune checkpoint inhibitors (Hirata and Sahai,
tolerant cells, acquiring diverse resistance mechanisms (Hata 2017). Based on the frequency and location of immune cells,
et al., 2016; Ramirez et al., 2016). Although the role of microen- tumors can be classified as immune cold (immune desert),
Review
Figure 4. Optimal Therapies for
Heterogeneous Tumors
The effective treatment of heterogeneous tumors
requires optimized therapy to minimize the evolu-
tion of therapeutic resistance. There are three
general approaches by which this can be ach-
ieved: (1) combination therapies targeting different
tumor subpopulations, different dependencies, or
both cell-autonomous and non-cell-autonomous
functions; (2) therapies combining tumor-targeting
agents with compounds that reduce ITH, such as
HDAC or BET inhibitors; and (3) adaptive therapy
when ITH is maintained and the competition of
drug-resistant and drug-sensitive cells is limited by
alternating therapy and no treatment.
Review
2019; Hinohara et al., 2018; Liau et al., 2017). Although current Anagnostou, V., Smith, K.N., Forde, P.M., Niknafs, N., Bhattacharya, R., White,
J., Zhang, T., Adleff, V., Phallen, J., Wali, N., et al. (2017). Evolution of neoan-
strategies primarily focus on short-term cytotoxic/cytostatic ef- tigen landscape during immune checkpoint blockade in non-small cell lung
fects of the drugs, consideration of longer-term effects, medi- cancer. Cancer Discov. 7, 264–276.
ated by reduction of ITH might be warranted, even when the
Andor, N., Graham, T.A., Jansen, M., Xia, L.C., Aktipis, C.A., Petritsch, C., Ji,
agents are not effective in the short term. Reducing microenvi- H.P., and Maley, C.C. (2016). Pan-cancer analysis of the extent and conse-
ronmental heterogeneity might also be beneficial, both in the quences of intratumor heterogeneity. Nat. Med. 22, 105–113.
short and long term. Although anti-angiogenic agents did not Angelo, M., Bendall, S.C., Finck, R., Hale, M.B., Hitzman, C., Borowsky, A.D.,
live up to the initial promise of eliminating tumors by blocking Levenson, R.M., Lowe, J.B., Liu, S.D., Zhao, S., et al. (2014). Multiplexed ion
their access to oxygen/nutrients, the reverse idea––normalizing beam imaging of human breast tumors. Nat. Med. 20, 436–442.
blood vasculature using lower doses of anti-angiogenic Angelova, M., Mlecnik, B., Vasaturo, A., Bindea, G., Fredriksen, T., Lafontaine,
drugs––might still prove to be clinically useful (Jain, 2014). In L., Buttard, B., Morgand, E., Bruni, D., Jouret-Mourin, A., et al. (2018). Evolu-
tion of metastases in space and time under immune selection. Cell 175, 751–
addition to improving drug delivery, which is currently the main 765.e716.
rationale behind the strategy, normalization of blood vasculature
Bakhoum, S.F., and Cantley, L.C. (2018). The multifaceted role of chromo-
could also reduce microenvironmental heterogeneity, while also
somal instability in cancer and its microenvironment. Cell 174, 1347–1360.
inhibiting selection for more invasive and aggressive subpopula-
tions (Robertson-Tessi et al., 2015). Beckman, R.A., and Loeb, L.A. (2006). Efficiency of carcinogenesis with and
without a mutator mutation. Proc. Natl. Acad. Sci. U S A 103, 14140–14145.
Concluding Remarks Bielski, C.M., Zehir, A., Penson, A.V., Donoghue, M.T.A., Chatila, W., Armenia,
J., Chang, M.T., Schram, A.M., Jonsson, P., Bandlamudi, C., et al. (2018).
Despite the recent explosion of interest toward cancer evolution Genome doubling shapes the evolution and prognosis of advanced cancers.
and ITH, our knowledge in this area remains rudimentary. Mov- Nat. Genet. 50, 1189–1195.
ing from observational studies toward deeper understanding of Bozic, I., Reiter, J.G., Allen, B., Antal, T., Chatterjee, K., Shah, P., Moon, Y.S.,
ITH and using this knowledge to improve clinical outcomes Yaqubie, A., Kelly, N., Le, D.T., et al. (2013). Evolutionary dynamics of cancer in
would require not only the development of novel methodologies response to targeted combination therapy. eLife 2, e00747.
to quantify different types of ITH, but also new frameworks to Brocks, D., Assenov, Y., Minner, S., Bogatyrova, O., Simon, R., Koop, C.,
conceptualize and integrate this knowledge. Achieving these Oakes, C., Zucknick, M., Lipka, D.B., Weischenfeldt, J., et al. (2014). Intratu-
mor DNA methylation heterogeneity reflects clonal evolution in aggressive
goals would necessitate moving from reductionistic approaches prostate cancer. Cell Rep. 8, 798–806.
of molecular oncology toward multidisciplinary studies, involving
meaningful integration of clinical observations, bioinformatical Bussard, K.M., and Smith, G.H. (2012). Human breast cancer cells are redir-
ected to mammary epithelial cells upon interaction with the regenerating mam-
analyses, and systems biology approaches with experimental mary gland microenvironment in-vivo. PLoS One 7, e49221.
and mathematical modeling.
Carmeliet, P., and Jain, R.K. (2000). Angiogenesis in cancer and other dis-
eases. Nature 407, 249–257.
ACKNOWLEDGMENTS
Chabon, J.J., Simmons, A.D., Lovejoy, A.F., Esfahani, M.S., Newman, A.M.,
We thank Dr. Daria Miroshnychenko for her help with designing Figure 2. This Haringsma, H.J., Kurtz, D.M., Stehr, H., Scherer, F., Karlovich, C.A., et al.
(2016). Circulating tumour DNA profiling reveals heterogeneity of EGFR inhib-
work was supported by the National Cancer Institute R35CA197623 (to K.P.),
itor resistance mechanisms in lung cancer patients. Nat. Commun. 7, 11815.
R00CA201606 (to M.J.), Susan G. Komen Breast Cancer Foundation
CCR17481976 (to A.M.), and the Breast Cancer Research Foundation (to K.P.). Chakrabarti, S., and Michor, F. (2017). Pharmacokinetics and drug interactions
determine optimum combination strategies in computational models of cancer
DECLARATION OF INTERESTS evolution. Cancer Res. 77, 3908–3921.
Colotta, F., Allavena, P., Sica, A., Garlanda, C., and Mantovani, A. (2009). Can-
K.P. is a member of the scientific advisory boards of Farcast Biosciences and cer-related inflammation, the seventh hallmark of cancer: links to genetic
Acrivon Therapeutics. instability. Carcinogenesis 30, 1073–1081.
REFERENCES Dawson, S.J., Tsui, D.W., Murtaza, M., Biggs, H., Rueda, O.M., Chin, S.F.,
Dunning, M.J., Gale, D., Forshew, T., Mahler-Araujo, B., et al. (2013). Analysis
of circulating tumor DNA to monitor metastatic breast cancer. N. Engl. J. Med.
Aceto, N., Bardia, A., Miyamoto, D.T., Donaldson, M.C., Wittner, B.S., 368, 1199–1209.
Spencer, J.A., Yu, M., Pely, A., Engstrom, A., Zhu, H., et al. (2014). Circulating
tumor cell clusters are oligoclonal precursors of breast cancer metastasis. Cell deCarvalho, A.C., Kim, H., Poisson, L.M., Winn, M.E., Mueller, C., Cherba, D.,
158, 1110–1122. Koeman, J., Seth, S., Protopopov, A., Felicella, M., et al. (2018). Discordant in-
heritance of chromosomal and extrachromosomal DNA elements contributes
Alexandrov, L.B., Nik-Zainal, S., Wedge, D.C., Aparicio, S.A., Behjati, S., Bian- to dynamic disease evolution in glioblastoma. Nat. Genet. 50, 708–717.
kin, A.V., Bignell, G.R., Bolli, N., Borg, A., Borresen-Dale, A.L., et al. (2013).
Signatures of mutational processes in human cancer. Nature 500, 415–421. Dewhurst, S.M., McGranahan, N., Burrell, R.A., Rowan, A.J., Grönroos, E., En-
desfelder, D., Joshi, T., Mouradov, D., Gibbs, P., Ward, R.L., et al. (2014).
Alfarouk, K.O., Ibrahim, M.E., Gatenby, R.A., and Brown, J.S. (2013). Riparian Tolerance of whole- genome doubling propagates chromosomal instability
ecosystems in human cancers. Evol. Appl. 6, 46–53. and accelerates cancer genome evolution. Cancer Discov. 4, 175–185.
Almendro, V., Cheng, Y.K., Randles, A., Itzkovitz, S., Marusyk, A., Ametller, E., Duelli, D.M., Padilla-Nash, H.M., Berman, D., Murphy, K.M., Ried, T., and Laz-
Gonzalez-Farre, X., Munoz, M., Russnes, H.G., Helland, A., et al. (2014). Infer- ebnik, Y. (2007). A virus causes cancer by inducing massive chromosomal
ence of tumor evolution during chemotherapy by computational modeling and instability through cell fusion. Curr. Biol. 17, 431–437.
in situ analysis of genetic and phenotypic cellular diversity. Cell Rep. 6,
514–527. Duncan, A.W., Taylor, M.H., Hickey, R.D., Hanlon Newell, A.E., Lenzi, M.L., Ol-
son, S.B., Finegold, M.J., and Grompe, M. (2010). The ploidy conveyor of
Anagnostou, V., Forde, P.M., White, J.R., Niknafs, N., Hruban, C., Naidoo, J., mature hepatocytes as a source of genetic variation. Nature 467, 707–710.
Marrone, K., Sivakumar, I.K.A., Bruhm, D.C., Rosner, S., et al. (2019). Dy-
namics of tumor and immune responses during immune checkpoint blockade Dunn, G.P., Old, L.J., and Schreiber, R.D. (2004). The three Es of cancer immu-
in non-small cell lung cancer. Cancer Res. 79, 1214–1225. noediting. Annu. Rev. Immunol. 22, 329–360.
Review
Estrella, V., Chen, T., Lloyd, M., Wojtkowiak, J., Cornnell, H.H., Ibrahim-Ha- distinct tumor immune microenvironments stratify triple-negative breast can-
shim, A., Bailey, K., Balagurunathan, Y., Rothberg, J.M., Sloane, B.F., et al. cers. J. Clin. Invest. 129, 1785–1800.
(2013). Acidity generated by the tumor microenvironment drives local invasion.
Cancer Res. 73, 1524–1535. Hangauer, M.J., Viswanathan, V.S., Ryan, M.J., Bole, D., Eaton, J.K., Matov,
A., Galeas, J., Dhruv, H.D., Berens, M.E., Schreiber, S.L., et al. (2017). Drug-
Fais, S., and Overholtzer, M. (2018). Cell-in-cell phenomena in cancer. Nat. tolerant persister cancer cells are vulnerable to GPX4 inhibition. Nature 551,
Rev. Cancer 18, 758–766. 247–250.
Fane, M., and Weeraratna, A.T. (2019). How the ageing microenvironment in- Hata, A.N., Niederst, M.J., Archibald, H.L., Gomez-Caraballo, M., Siddiqui,
fluences tumour progression. Nat. Rev. Cancer 20, 89–106. F.M., Mulvey, H.E., Maruvka, Y.E., Ji, F., Bhang, H.E., Krishnamurthy Radhak-
rishna, V., et al. (2016). Tumor cells can follow distinct evolutionary paths to
Feinberg, A.P. (2007). Phenotypic plasticity and the epigenetics of human dis- become resistant to epidermal growth factor receptor inhibition. Nat. Med.
ease. Nature 447, 433–440. 22, 262–269.
Filho, O.M., Viale, G., Trippa, L., Li, T., Yardley, D.A., Mayer, I.A., Abramson, Havel, J.J., Chowell, D., and Chan, T.A. (2019). The evolving landscape of bio-
V.G., Arteaga, C.L., Spring, L., Waks, A.G., et al. (2019). HER2 heterogeneity markers for checkpoint inhibitor immunotherapy. Nat. Rev. Cancer 19,
as a predictor of response to neoadjuvant T-DM1 plus pertuzumab: results 133–150.
from a prospective clinical trial. J. Clin. Oncol. 37, 502.
Hinohara, K., and Polyak, K. (2019). Intratumoral heterogeneity: more than just
Flavahan, W.A., Gaskell, E., and Bernstein, B.E. (2017). Epigenetic plasticity mutations. Trends Cell Biol. 29, 569–579.
and the hallmarks of cancer. Science 357, https://doi.org/10.1126/science.
aal2380. Hinohara, K., Wu, H.J., Vigneau, S., McDonald, T.O., Igarashi, K.J., Yama-
moto, K.N., Madsen, T., Fassl, A., Egri, S.B., Papanastasiou, M., et al.
Foo, J., and Michor, F. (2014). Evolution of acquired resistance to anti-cancer (2018). KDM5 histone demethylase activity links cellular transcriptomic hetero-
therapy. J. Theor. Biol. 355, 10–20. geneity to therapeutic resistance. Cancer Cell 34, 939–953.e9.
Fox, E.J., Prindle, M.J., and Loeb, L.A. (2013). Do mutator mutations fuel Hirata, E., Girotti, M.R., Viros, A., Hooper, S., Spencer-Dene, B., Matsuda, M.,
tumorigenesis? Cancer Metastasis Rev. 32, 353–361. Larkin, J., Marais, R., and Sahai, E. (2015). Intravital imaging reveals how BRAF
inhibition generates drug-tolerant microenvironments with high integrin beta1/
Gaiti, F., Chaligne, R., Gu, H., Brand, R.M., Kothen-Hill, S., Schulman, R.C., FAK signaling. Cancer Cell 27, 574–588.
Grigorev, K., Risso, D., Kim, K.T., Pastore, A., et al. (2019). Epigenetic evolution
and lineage histories of chronic lymphocytic leukaemia. Nature 569, 576–580. Hirata, E., and Sahai, E. (2017). Tumor microenvironment and differential re-
sponses to therapy. Cold Spring Harb Perspect. Med. 7, https://doi.org/10.
Gallaher, J.A., Enriquez-Navas, P.M., Luddy, K.A., Gatenby, R.A., and Ander- 1101/cshperspect.a026781.
son, A.R.A. (2018). Spatial heterogeneity and evolutionary dynamics modulate
time to recurrence in continuous and adaptive cancer therapies. Cancer Res. Hoefflin, R., Lahrmann, B., Warsow, G., Hubschmann, D., Spath, C., Walter,
78, 2127–2139. B., Chen, X., Hofer, L., Macher-Goeppinger, S., Tolstov, Y., et al. (2016).
Spatial niche formation but not malignant progression is a driving force for in-
Gast, C.E., Silk, A.D., Zarour, L., Riegler, L., Burkhart, J.G., Gustafson, K.T., tratumoural heterogeneity. Nat. Commun. 7, ncomms11845.
Parappilly, M.S., Roh-Johnson, M., Goodman, J.R., Olson, B., et al. (2018).
Cell fusion potentiates tumor heterogeneity and reveals circulating hybrid cells Holohan, C., Van Schaeybroeck, S., Longley, D.B., and Johnston, P.G. (2013).
that correlate with stage and survival. Sci. Adv. 4, eaat7828. Cancer drug resistance: an evolving paradigm. Nat. Rev. Cancer 13, 714–726.
Gatenby, R.A., and Gillies, R.J. (2008). A microenvironmental model of carci- Hong, S.P., Chan, T.E., Lombardo, Y., Corleone, G., Rotmensz, N., Bravaccini,
nogenesis. Nat. Rev. Cancer 8, 56–61. S., Rocca, A., Pruneri, G., McEwen, K.R., Coombes, R.C., et al. (2019). Single-
cell transcriptomics reveals multi-step adaptations to endocrine therapy. Nat.
Gejman, R.S., Chang, A.Y., Jones, H.F., DiKun, K., Hakimi, A.A., Schietinger, Commun. 10, 3840.
A., and Scheinberg, D.A. (2018). Rejection of immunogenic tumor clones is
limited by clonal fraction. eLife 7, https://doi.org/10.7554/eLife.41090. Jacobsen, B.M., Harrell, J.C., Jedlicka, P., Borges, V.F., Varella-Garcia, M.,
and Horwitz, K.B. (2006). Spontaneous fusion with, and transformation of
Gerlinger, M., Rowan, A.J., Horswell, S., Math, M., Larkin, J., Endesfelder, D., mouse stroma by, malignant human breast cancer epithelium. Cancer Res.
Gronroos, E., Martinez, P., Matthews, N., Stewart, A., et al. (2012). Intratumor 66, 8274–8279.
heterogeneity and branched evolution revealed by multiregion sequencing.
N. Engl. J. Med. 366, 883–892. Jain, R.K. (2014). Antiangiogenesis strategies revisited: from starving tumors
to alleviating hypoxia. Cancer Cell 26, 605–622.
Gillies, R.J., Brown, J.S., Anderson, A.R.A., and Gatenby, R.A. (2018). Eco-
evolutionary causes and consequences of temporal changes in intratumoural Jamal-Hanjani, M., Wilson, G.A., McGranahan, N., Birkbak, N.J., Watkins,
blood flow. Nat. Rev. Cancer 18, 576–585. T.B.K., Veeriah, S., Shafi, S., Johnson, D.H., Mitter, R., Rosenthal, R., et al.
(2017). Tracking the evolution of non-small-cell lung cancer. N. Engl. J. Med.
Gillies, R.J., Verduzco, D., and Gatenby, R.A. (2012). Evolutionary dynamics of 376, 2109–2121.
carcinogenesis and why targeted therapy does not work. Nat. Rev. Cancer 12,
487–493. Janiszewska, M., Liu, L., Almendro, V., Kuang, Y., Paweletz, C., Sakr, R.A.,
Weigelt, B., Hanker, A.B., Chandarlapaty, S., King, T.A., et al. (2015). In situ sin-
Godek, K.M., Venere, M., Wu, Q., Mills, K.D., Hickey, W.F., Rich, J.N., and gle-cell analysis identifies heterogeneity for PIK3CA mutation and HER2 ampli-
Compton, D.A. (2016). Chromosomal instability affects the tumorigenicity of fication in HER2-positive breast cancer. Nat. Genet. 47, 1212–1219.
glioblastoma tumor-initiating cells. Cancer Discov. 6, 532–545.
Jenkinson, G., Pujadas, E., Goutsias, J., and Feinberg, A.P. (2017). Potential
Goldman, A., Majumder, B., Dhawan, A., Ravi, S., Goldman, D., Kohandel, M., energy landscapes identify the information-theoretic nature of the epigenome.
Majumder, P.K., and Sengupta, S. (2015). Temporally sequenced anticancer Nat. Genet. 49, 719–729.
drugs overcome adaptive resistance by targeting a vulnerable chemo-
therapy-induced phenotypic transition. Nat. Commun. 6, 6139. Jimenez-Sanchez, A., Memon, D., Pourpe, S., Veeraraghavan, H., Li, Y., Var-
gas, H.A., Gill, M.B., Park, K.J., Zivanovic, O., Konner, J., et al. (2017). Hetero-
Granja, J.M., Klemm, S., McGinnis, L.M., Kathiria, A.S., Mezger, A., Corces, geneous tumor-immune microenvironments among differentially growing me-
M.R., Parks, B., Gars, E., Liedtke, M., Zheng, G.X.Y., et al. (2019). Single- tastases in an ovarian cancer patient. Cell 170, 927–938 e920.
cell multiomic analysis identifies regulatory programs in mixed-phenotype
acute leukemia. Nat. Biotechnol. 37, 1458–1465. Joshi, K., Robert de Massy, M., Ismail, M., Reading, J.L., Uddin, I., Woolston,
A., Hatipoglu, E., Oakes, T., Rosenthal, R., Peacock, T., et al. (2019). Spatial
Grivennikov, S.I., Greten, F.R., and Karin, M. (2010). Immunity, inflammation, heterogeneity of the T cell receptor repertoire reflects the mutational land-
and cancer. Cell 140, 883–899. scape in lung cancer. Nat. Med. 25, 1549–1559.
Gruosso, T., Gigoux, M., Manem, V.S.K., Bertos, N., Zuo, D., Perlitch, I., Saleh, Kalbasi, A., and Ribas, A. (2019). Tumour-intrinsic resistance to immune
S.M.I., Zhao, H., Souleimanova, M., Johnson, R.M., et al. (2019). Spatially checkpoint blockade. Nat. Rev. Immunol. 20, 25–39.
Review
Kandoth, C., McLellan, M.D., Vandin, F., Ye, K., Niu, B., Lu, C., Xie, M., Zhang, Lu, X., and Kang, Y. (2009). Efficient acquisition of dual metastasis organotrop-
Q., McMichael, J.F., Wyczalkowski, M.A., et al. (2013). Mutational landscape ism to bone and lung through stable spontaneous fusion between MDA-MB-
and significance across 12 major cancer types. Nature 502, 333–339. 231 variants. Proc. Natl. Acad. Sci. U S A 106, 9385–9390.
Karnoub, A.E., Dash, A.B., Vo, A.P., Sullivan, A., Brooks, M.W., Bell, G.W., Ri- Mann, K.M., Newberg, J.Y., Black, M.A., Jones, D.J., Amaya-Manzanares, F.,
chardson, A.L., Polyak, K., Tubo, R., and Weinberg, R.A. (2007). Mesenchymal Guzman-Rojas, L., Kodama, T., Ward, J.M., Rust, A.G., van der Weyden, L.,
stem cells within tumour stroma promote breast cancer metastasis. Nature et al. (2016). Analyzing tumor heterogeneity and driver genes in single myeloid
449, 557–563. leukemia cells with SBCapSeq. Nat. Biotechnol. 34, 962–972.
Keren, L., Bosse, M., Marquez, D., Angoshtari, R., Jain, S., Varma, S., Yang, Mariathasan, S., Turley, S.J., Nickles, D., Castiglioni, A., Yuen, K., Wang, Y.,
S.R., Kurian, A., Van Valen, D., West, R., et al. (2018). A structured tumor-im- Kadel, E.E., III, Koeppen, H., Astarita, J.L., Cubas, R., et al. (2018). TGFbeta
mune microenvironment in triple negative breast cancer revealed by multi- attenuates tumour response to PD-L1 blockade by contributing to exclusion
plexed ion beam imaging. Cell 174, 1373–1387.e1319. of T cells. Nature 554, 544–548.
Kersten, K., de Visser, K.E., van Miltenburg, M.H., and Jonkers, J. (2017). Marusyk, A., Almendro, V., and Polyak, K. (2012). Intra-tumour heterogeneity:
Genetically engineered mouse models in oncology research and cancer med- a looking glass for cancer? Nat. Rev. Cancer 12, 323–334.
icine. EMBO Mol. Med. 9, 137–153.
Marusyk, A., Tabassum, D.P., Janiszewska, M., Place, A.E., Trinh, A., Rozhok,
Khan, K.H., Cunningham, D., Werner, B., Vlachogiannis, G., Spiteri, I., Heide, A.I., Pyne, S., Guerriero, J.L., Shu, S., Ekram, M., et al. (2016). Spatial proximity
T., Mateos, J.F., Vatsiou, A., Lampis, A., Damavandi, M.D., et al. (2018). Lon- to fibroblasts impacts molecular features and therapeutic sensitivity of breast
gitudinal liquid biopsy and mathematical modeling of clonal evolution forecast cancer cells influencing clinical outcomes. Cancer Res. 76, 6495–6506.
time to treatment failure in the PROSPECT-C phase II colorectal cancer clinical
trial. Cancer Discov. 8, 1270–1285. Mazor, T., Pankov, A., Johnson, B.E., Hong, C., Hamilton, E.G., Bell, R.J.A.,
Smirnov, I.V., Reis, G.F., Phillips, J.J., Barnes, M.J., et al. (2015). DNA methyl-
Knoechel, B., Roderick, J.E., Williamson, K.E., Zhu, J., Lohr, J.G., Cotton, ation and somatic mutations converge on the cell cycle and define similar
M.J., Gillespie, S.M., Fernandez, D., Ku, M., Wang, H., et al. (2014). An epige- evolutionary histories in brain tumors. Cancer Cell 28, 307–317.
netic mechanism of resistance to targeted therapy in T cell acute lympho-
blastic leukemia. Nat. Genet. 46, 364–370. Mazor, T., Pankov, A., Song, J.S., and Costello, J.F. (2016). Intratumoral het-
erogeneity of the epigenome. Cancer Cell 29, 440–451.
Korenchan, D.E., and Flavell, R.R. (2019). Spatiotemporal pH heterogeneity as
a promoter of cancer progression and therapeutic resistance. Cancers (Basel) McGranahan, N., Furness, A.J., Rosenthal, R., Ramskov, S., Lyngaa, R., Saini,
11, https://doi.org/10.3390/cancers11071026. S.K., Jamal-Hanjani, M., Wilson, G.A., Birkbak, N.J., Hiley, C.T., et al. (2016).
Clonal neoantigens elicit T cell immunoreactivity and sensitivity to immune
Korkaya, H., Liu, S., and Wicha, M.S. (2011). Regulation of cancer stem cells checkpoint blockade. Science 351, 1463–1469.
by cytokine networks: attacking cancer’s inflammatory roots. Clin. Cancer
Res. 17, 6125–6129. McGranahan, N., and Swanton, C. (2017). Clonal heterogeneity and tumor
evolution: past, present, and the future. Cell 168, 613–628.
LaBerge, G.S., Duvall, E., Grasmick, Z., Haedicke, K., and Pawelek, J. (2017).
A melanoma lymph node metastasis with a donor-patient hybrid genome
Meads, M.B., Gatenby, R.A., and Dalton, W.S. (2009). Environment-mediated
following bone marrow transplantation: a second case of leucocyte-tumor
drug resistance: a major contributor to minimal residual disease. Nat. Rev.
cell hybridization in cancer metastasis. PLoS One 12, e0168581.
Cancer 9, 665–674.
Landau, D.A., Clement, K., Ziller, M.J., Boyle, P., Fan, J., Gu, H., Stevenson, K.,
Merlo, L.M., Pepper, J.W., Reid, B.J., and Maley, C.C. (2006). Cancer as an
Sougnez, C., Wang, L., Li, S., et al. (2014). Locally disordered methylation
evolutionary and ecological process. Nat. Rev. Cancer 6, 924–935.
forms the basis of intratumor methylome variation in chronic lymphocytic leu-
kemia. Cancer Cell 26, 813–825.
Merlo, L.M.F., Shah, N.A., Li, X., Blount, P.L., Vaughan, T.L., Reid, B.J., and
Lazova, R., Laberge, G.S., Duvall, E., Spoelstra, N., Klump, V., Sznol, M., Maley, C.C. (2010). A comprehensive survey of clonal diversity measures in
Cooper, D., Spritz, R.A., Chang, J.T., and Pawelek, J.M. (2013). A melanoma Barrett’s esophagus as biomarkers of progression to esophageal adenocarci-
brain metastasis with a donor-patient hybrid genome following bone marrow noma. Cancer Prev. Res. 3, 1388–1397.
transplantation: first evidence for fusion in human cancer. PLoS One 8,
e66731. Miller, A.M., Shah, R.H., Pentsova, E.I., Pourmaleki, M., Briggs, S., Distefano,
N., Zheng, Y., Skakodub, A., Mehta, S.A., Campos, C., et al. (2019). Tracking
Leder, K., Pitter, K., Laplant, Q., Hambardzumyan, D., Ross, B.D., Chan, T.A., tumour evolution in glioma through liquid biopsies of cerebrospinal fluid. Na-
Holland, E.C., and Michor, F. (2014). Mathematical modeling of PDGF-driven ture 565, 654–658.
glioblastoma reveals optimized radiation dosing schedules. Cell 156, 603–616.
Milo, I., Bedora-Faure, M., Garcia, Z., Thibaut, R., Perie, L., Shakhar, G., De-
Lengauer, C., Kinzler, K.W., and Vogelstein, B. (1997). Genetic instability in riano, L., and Bousso, P. (2018). The immune system profoundly restricts intra-
colorectal cancers. Nature 386, 623–627. tumor genetic heterogeneity. Sci. Immunol. 3, https://doi.org/10.1126/sciim-
munol.aat1435.
Li, L., McCormack, A.A., Nicholson, J.M., Fabarius, A., Hehlmann, R., Sachs,
R.K., and Duesberg, P.H. (2009). Cancer-causing karyotypes: chromosomal Miroshnychenko, D., Baratchart, E., Ferrall-Fairbanks, M.C., Vander Velde, R.,
equilibria between destabilizing aneuploidy and stabilizing selection for onco- Laurie, M.A., Bui, M.M., Altrock, P.M., Basanta, D., and Marusyk, A. (2020).
genic function. Cancer Genet. Cytogenet. 188, 1–25. Spontaneous cell fusions as a mechanism of parasexual recombination in tu-
mor cell populations. BioRxiv. https://doi.org/10.1101/2020.03.09.984419.
Liau, B.B., Sievers, C., Donohue, L.K., Gillespie, S.M., Flavahan, W.A., Miller,
T.E., Venteicher, A.S., Hebert, C.H., Carey, C.D., Rodig, S.J., et al. (2017). Misale, S., Yaeger, R., Hobor, S., Scala, E., Janakiraman, M., Liska, D., Val-
Adaptive chromatin remodeling drives glioblastoma stem cell plasticity and torta, E., Schiavo, R., Buscarino, M., Siravegna, G., et al. (2012). Emergence
drug tolerance. Cell Stem Cell 20, 233–246 e237. of KRAS mutations and acquired resistance to anti-EGFR therapy in colorectal
cancer. Nature 486, 532–536.
Lin, J.R., Fallahi-Sichani, M., and Sorger, P.K. (2015). Highly multiplexed imag-
ing of single cells using a high-throughput cyclic immunofluorescence method. Modi, S., Saura, C., Yamashita, T., Park, Y.H., Kim, S.B., Tamura, K., Andre, F.,
Nat. Commun. 6, 8390. Iwata, H., Ito, Y., Tsurutani, J., et al. (2019). Trastuzumab deruxtecan in previ-
ously treated HER2-positive breast cancer. N. Engl. J. Med. 382, 610–621.
Lloyd, M.C., Cunningham, J.J., Bui, M.M., Gillies, R.J., Brown, J.S., and Gate-
nby, R.A. (2016). Darwinian dynamics of intratumoral heterogeneity: not solely Mohlin, S., Wigerup, C., Jogi, A., and Pahlman, S. (2017). Hypoxia, pseudohy-
random mutations but also variable environmental selection forces. Cancer poxia and cellular differentiation. Exp. Cell Res. 356, 192–196.
Res. 76, 3136–3144.
Morris, L.G., Riaz, N., Desrichard, A., Senbabaoglu, Y., Hakimi, A.A., Makarov,
Lovly, C.M., and Shaw, A.T. (2014). Molecular pathways: resistance to kinase V., Reis-Filho, J.S., and Chan, T.A. (2016). Pan-cancer analysis of intratumor
inhibitors and implications for therapeutic strategies. Clin. Cancer Res. 20, heterogeneity as a prognostic determinant of survival. Oncotarget 7,
2249–2256. 10051–10063.
Review
Naba, A., Clauser, K.R., Hoersch, S., Liu, H., Carr, S.A., and Hynes, R.O. Rosenthal, R., Cadieux, E.L., Salgado, R., Bakir, M.A., Moore, D.A., Hiley, C.T.,
(2012). The matrisome: in silico definition and in vivo characterization by pro- Lund, T., Tanic, M., Reading, J.L., Joshi, K., et al. (2019). Neoantigen-directed
teomics of normal and tumor extracellular matrices. Mol. Cell Proteomics 11, immune escape in lung cancer evolution. Nature 567, 479–485.
M111 014647.
Rossi, G., and Ignatiadis, M. (2019). Promises and pitfalls of using liquid biopsy
Nathanson, D.A., Gini, B., Mottahedeh, J., Visnyei, K., Koga, T., Gomez, G., for precision medicine. Cancer Res. 79, 2798–2804.
Eskin, A., Hwang, K., Wang, J., Masui, K., et al. (2014). Targeted therapy resis-
tance mediated by dynamic regulation of extrachromosomal mutant EGFR Russo, M., Crisafulli, G., Sogari, A., Reilly, N.M., Arena, S., Lamba, S., Bartolini,
DNA. Science 343, 72–76. A., Amodio, V., Magri, A., Novara, L., et al. (2019). Adaptive mutability of colo-
rectal cancers in response to targeted therapies. Science 366, 1473–1480.
Neftel, C., Laffy, J., Filbin, M.G., Hara, T., Shore, M.E., Rahme, G.J., Richman,
A.R., Silverbush, D., Shaw, M.L., Hebert, C.M., et al. (2019). An integrative Rye, I.H., Trinh, A., Saetersdal, A.B., Nebdal, D., Lingjaerde, O.C., Almendro,
model of cellular states, plasticity, and genetics for glioblastoma. Cell 178, V., Polyak, K., Borresen-Dale, A.L., Helland, A., Markowetz, F., and Russnes,
835–849 e821. H.G. (2018). Intratumor heterogeneity defines treatment-resistant HER2+
breast tumors. Mol. Oncol. 12, 1838–1855.
Niknafs, N., Zhong, Y., Moral, J.A., Zhang, L., Shao, M.X., Lo, A., Makohon-
Moore, A., Iacobuzio-Donahue, C.A., and Karchin, R. (2019). Characterization Senovilla, L., Vitale, I., Martins, I., Tailler, M., Pailleret, C., Michaud, M., Gal-
of genetic subclonal evolution in pancreatic cancer mouse models. Nat. Com- luzzi, L., Adjemian, S., Kepp, O., Niso-Santano, M., et al. (2012). An immuno-
mun. 10, 5435. surveillance mechanism controls cancer cell ploidy. Science 337, 1678–1684.
Noman, M.Z., Hasmim, M., Messai, Y., Terry, S., Kieda, C., Janji, B., and Seth, S., Li, C.Y., Ho, I.L., Corti, D., Loponte, S., Sapio, L., Del Poggetto, E.,
Chouaib, S. (2015). Hypoxia: a key player in antitumor immune response. A re- Yen, E.Y., Robinson, F.S., Peoples, M., et al. (2019). Pre-existing functional
view in the theme: cellular responses to hypoxia. Am. J. Physiol. Cell Physiol. heterogeneity of tumorigenic compartment as the origin of chemoresistance
309, C569–C579. in pancreatic tumors. Cell Rep. 26, 1518–1532 e1519.
Patel, A.P., Tirosh, I., Trombetta, J.J., Shalek, A.K., Gillespie, S.M., Wakimoto, Shaffer, S.M., Dunagin, M.C., Torborg, S.R., Torre, E.A., Emert, B., Krepler, C.,
H., Cahill, D.P., Nahed, B.V., Curry, W.T., Martuza, R.L., et al. (2014). Single- Beqiri, M., Sproesser, K., Brafford, P.A., Xiao, M., et al. (2017). Rare cell vari-
cell RNA-seq highlights intratumoral heterogeneity in primary glioblastoma. ability and drug-induced reprogramming as a mode of cancer drug resistance.
Science 344, 1396–1401. Nature 546, 431–435.
Pilon-Thomas, S., Kodumudi, K.N., El-Kenawi, A.E., Russell, S., Weber, A.M., Sharma, S.V., Lee, D.Y., Li, B., Quinlan, M.P., Takahashi, F., Maheswaran, S.,
Luddy, K., Damaghi, M., Wojtkowiak, J.W., Mule, J.J., Ibrahim-Hashim, A., and McDermott, U., Azizian, N., Zou, L., Fischbach, M.A., et al. (2010). A chro-
Gillies, R.J. (2016). Neutralization of tumor acidity improves antitumor re- matin-mediated reversible drug-tolerant state in cancer cell subpopulations.
sponses to immunotherapy. Cancer Res. 76, 1381–1390. Cell 141, 69–80.
Piotrowska, Z., Niederst, M.J., Karlovich, C.A., Wakelee, H.A., Neal, J.W., Shipitsin, M., Campbell, L.L., Argani, P., Weremowicz, S., Bloushtain-Qimron,
Mino-Kenudson, M., Fulton, L., Hata, A.N., Lockerman, E.L., Kalsy, A., et al. N., Yao, J., Nikolskaya, T., Serebryiskaya, T., Beroukhim, R., Hu, M., et al.
(2015). Heterogeneity underlies the emergence of EGFRT790 wild-type clones (2007). Molecular definition of breast tumor heterogeneity. Cancer Cell 11,
following treatment of T790M-positive cancers with a third-generation EGFR 259–273.
inhibitor. Cancer Discov. 5, 713–722.
Shu, S., Lin, C.Y., He, H.H., Witwicki, R.M., Tabassum, D.P., Roberts, J.M.,
Pisco, A.O., and Huang, S. (2015). Non-genetic cancer cell plasticity and ther- Janiszewska, M., Huh, S.J., Liang, Y., Ryan, J., et al. (2016). Response and
apy-induced stemness in tumour relapse: ’what does not kill me strengthens resistance to BET bromodomain inhibitors in triple-negative breast cancer.
me’. Br. J. Cancer 112, 1725–1732. Nature 529, 413–417.
Ramirez, M., Rajaram, S., Steininger, R.J., Osipchuk, D., Roth, M.A., Morinishi, Singh, A., and Settleman, J. (2010). EMT, cancer stem cells and drug resis-
L.S., Evans, L., Ji, W., Hsu, C.H., Thurley, K., et al. (2016). Diverse drug-resis- tance: an emerging axis of evil in the war on cancer. Oncogene 29, 4741–4751.
tance mechanisms can emerge from drug-tolerant cancer persister cells. Nat.
Commun. 7, 10690. Siravegna, G., Mussolin, B., Buscarino, M., Corti, G., Cassingena, A., Crisafulli,
G., Ponzetti, A., Cremolini, C., Amatu, A., Lauricella, C., et al. (2015). Clonal
Reuben, A., Gittelman, R., Gao, J., Zhang, J., Yusko, E.C., Wu, C.J., Emerson, evolution and resistance to EGFR blockade in the blood of colorectal cancer
R., Zhang, J., Tipton, C., Li, J., et al. (2017). TCR repertoire intratumor hetero- patients. Nat. Med. 21, 827.
geneity in localized lung adenocarcinomas: an association with predicted neo-
antigen heterogeneity and postsurgical recurrence. Cancer Discov. 7, Spitzer, M.H., and Nolan, G.P. (2016). Mass cytometry: single cells, many fea-
1088–1097. tures. Cell 165, 780–791.
Ribas, A., and Wolchok, J.D. (2018). Cancer immunotherapy using checkpoint Stacker, S.A., Williams, S.P., Karnezis, T., Shayan, R., Fox, S.B., and Achen,
blockade. Science 359, 1350–1355. M.G. (2014). Lymphangiogenesis and lymphatic vessel remodelling in cancer.
Nat. Rev. Cancer 14, 159–172.
Risom, T., Langer, E.M., Chapman, M.P., Rantala, J., Fields, A.J., Boniface, C.,
Alvarez, M.J., Kendsersky, N.D., Pelz, C.R., Johnson-Camacho, K., et al. Storchova, Z., and Pellman, D. (2004). From polyploidy to aneuploidy, genome
(2018). Differentiation-state plasticity is a targetable resistance mechanism instability and cancer. Nat. Rev. Mol. Cell Biol. 5, 45–54.
in basal-like breast cancer. Nat. Commun. 9, 3815.
Tang, Y.C., and Amon, A. (2013). Gene copy-number alterations: a cost-
Robertson-Tessi, M., Gillies, R.J., Gatenby, R.A., and Anderson, A.R. (2015). benefit analysis. Cell 152, 394–405.
Impact of metabolic heterogeneity on tumor growth, invasion, and treatment
outcomes. Cancer Res. 75, 1567–1579. Thrane, K., Eriksson, H., Maaskola, J., Hansson, J., and Lundeberg, J. (2018).
Spatially resolved transcriptomics enables dissection of genetic heterogeneity
Rockne, R.C., and Scott, J.G. (2019). Introduction to mathematical oncology. in stage III cutaneous malignant melanoma. Cancer Res. 78, 5970–5979.
JCO Clin Cancer Inform 3, 1–4.
Tomasetti, C., and Vogelstein, B. (2015). Cancer etiology. Variation in cancer
Roe, J.S., Hwang, C.I., Somerville, T.D.D., Milazzo, J.P., Lee, E.J., Da Silva, B., risk among tissues can be explained by the number of stem cell divisions. Sci-
Maiorino, L., Tiriac, H., Young, C.M., Miyabayashi, K., et al. (2017). Enhancer ence 347, 78–81.
reprogramming promotes pancreatic cancer metastasis. Cell 170, 875–
888 e820. Tomlinson, I.P., Novelli, M.R., and Bodmer, W.F. (1996). The mutation rate and
cancer. Proc. Natl. Acad. Sci. U S A 93, 14800–14803.
Roesch, A., Vultur, A., Bogeski, I., Wang, H., Zimmermann, K.M., Speicher, D.,
Korbel, C., Laschke, M.W., Gimotty, P.A., Philipp, S.E., et al. (2013). Over- Turner, K.M., Deshpande, V., Beyter, D., Koga, T., Rusert, J., Lee, C., Li, B.,
coming intrinsic multidrug resistance in melanoma by blocking the mitochon- Arden, K., Ren, B., Nathanson, D.A., et al. (2017). Extrachromosomal onco-
drial respiratory chain of slow-cycling JARID1B(high) cells. Cancer Cell 23, gene amplification drives tumour evolution and genetic heterogeneity. Nature
811–825. 543, 122–125.
Review
Vander Velde, R., Yoon, N., Marusyk, V., Durmaz, A., Dhawan, A., Myroshny- Yuan, Y. (2016). Spatial heterogeneity in the tumor microenvironment. Cold
chenko, D., Lozano-Peral, D., Desai, B., Balynska, O., Poleszhuk, J., et al. Spring Harb Perspect. Med. 6, https://doi.org/10.1101/cshperspect.a026583.
(2019). Resistance to targeted therapies as a multifactorial, gradual adaptation
to inhibitor specific selective pressures. bioRxiv, 504837. Zhang, A.W., McPherson, A., Milne, K., Kroeger, D.R., Hamilton, P.T., Miranda,
A., Funnell, T., Little, N., de Souza, C.P.E., Laan, S., et al. (2018). Interfaces of
Verhaak, R.G.W., Bafna, V., and Mischel, P.S. (2019). Extrachromosomal malignant and immunologic clonal dynamics in ovarian cancer. Cell 173,
oncogene amplification in tumour pathogenesis and evolution. Nat. Rev. Can- 1755–1769 e1722.
cer 19, 283–288.
Vogelstein, B., and Kinzler, K.W. (2004). Cancer genes and the pathways they Zhang, J., Cunningham, J.J., Brown, J.S., and Gatenby, R.A. (2017). Inte-
control. Nat. Med. 10, 789–799. grating evolutionary dynamics into treatment of metastatic castrate-resistant
prostate cancer. Nat Commun. 8, 1816.
Wilson, T.R., Fridlyand, J., Yan, Y., Penuel, E., Burton, L., Chan, E., Peng, J.,
Lin, E., Wang, Y., Sosman, J., et al. (2012). Widespread potential for growth- Zhang, J., Fujimoto, J., Zhang, J., Wedge, D.C., Song, X., Zhang, J., Seth, S.,
factor-driven resistance to anticancer kinase inhibitors. Nature 487, 505–509. Chow, C.W., Cao, Y., Gumbs, C., et al. (2014). Intratumor heterogeneity in
localized lung adenocarcinomas delineated by multiregion sequencing. Sci-
Wodarz, D., and Komarova, N.L. (2005). Emergence and prevention of resis- ence 346, 256–259.
tance against small molecule inhibitors. Semin. Cancer Biol. 15, 506–514.
Zhang, J., Yang, P.L., and Gray, N.S. (2009). Targeting cancer with small mole-
Wolf, Y., Bartok, O., Patkar, S., Eli, G.B., Cohen, S., Litchfield, K., Levy, R., Ji-
cule kinase inhibitors. Nat. Rev. Cancer 9, 28–39.
menez-Sanchez, A., Trabish, S., Lee, J.S., et al. (2019). UVB-induced tumor
heterogeneity diminishes immune response in melanoma. Cell 179, 219–
235.e21. Zhang, L., and Vijg, J. (2018). Somatic mutagenesis in mammals and its impli-
cations for human disease and aging. Annu. Rev. Genet. 52, 397–419.
Yates, L.R., Gerstung, M., Knappskog, S., Desmedt, C., Gundem, G., Van Loo,
P., Aas, T., Alexandrov, L.B., Larsimont, D., Davies, H., et al. (2015). Subclonal Zhang, N., Magee, B.B., Magee, P.T., Holland, B.R., Rodrigues, E., Holmes,
diversification of primary breast cancer revealed by multiregion sequencing. A.R., Cannon, R.D., and Schmid, J. (2015). Selective advantages of a para-
Nat. Med. 21, 751–759. sexual cycle for the yeast Candida albicans. Genetics 200, 1117–1132.
Review
Engineering CAR-T Cells
for Next-Generation Cancer Therapy
Mihe Hong,1,4 Justin D. Clubb,1,4 and Yvonne Y. Chen1,2,3,*
1Department of Chemical and Biomolecular Engineering, University of California–Los Angeles, Los Angeles, CA 90095, USA
2Department of Microbiology, Immunology, and Molecular Genetics, University of California–Los Angeles, Los Angeles, CA 90095, USA
3Parker Institute for Cancer Immunotherapy Center at UCLA, Los Angeles, CA 90095, USA
4These authors contributed equally
*Correspondence: yvonne.chen@ucla.edu
https://doi.org/10.1016/j.ccell.2020.07.005
SUMMARY
T cells engineered to express chimeric antigen receptors (CARs) with tumor specificity have shown remark-
able success in treating patients with hematologic malignancies and revitalized the field of adoptive cell
therapy. However, realizing broader therapeutic applications of CAR-T cells necessitates engineering
approaches on multiple levels to enhance efficacy and safety. Particularly, solid tumors present unique chal-
lenges due to the biological complexity of the solid-tumor microenvironment (TME). In this review, we high-
light recent strategies to improve CAR-T cell therapy by engineering (1) the CAR protein, (2) T cells, and (3) the
interaction between T cells and other components in the TME.
INTRODUCTION growth factor receptor 2 (HER2) fused to the CD3z signaling domain
can elicit tumor-specific cytotoxicity (Eshhar et al., 1993; Moritz
Chimeric antigen receptors (CARs) are synthetic receptors that et al., 1994; Stancovski et al., 1993), but T cells expressing these
enable T cells to recognize tumor-associated antigens (TAAs) ‘‘first-generation’’ CARs that included only the CD3z chain for T-
in a major histocompatibility complex (MHC)-independent cell signaling generally failed to elicit potent antitumor effects.
manner. CAR-T cells targeting the pan–B-cell marker CD19 In the following years, second- and third-generation CARs
have shown unprecedented response rates in treating refractory emerged that included one or two costimulatory domains,
B cell malignancies (Maude et al., 2014; Neelapu et al., 2017) and respectively, drawing from the biological understanding that
became the first genetically modified cell-based therapy to the endogenous TCR requires association with other costimula-
receive US Food and Drug Administration approval (Bouchkouj tory or accessory molecules for robust signaling (Chen and Flies,
et al., 2019; O’Leary et al., 2019). However, the development 2013). Most commonly derived from CD28 or 4-1BB, these cos-
of effective CAR-T cell therapy for non–B-cell malignancies has timulatory domains conferred more potent antitumor cytotox-
required more sophisticated engineering approaches to over- icity, increased cytokine production, and improved proliferation
come tumor-defense mechanisms such as immunosuppression, and persistence of CAR-T cells (Haynes et al., 2002; Imai et al.,
antigen escape, and physical barriers to entry into solid tumors. 2004). The choice of costimulatory domain has an impact on a
In this review, we examine current and prospective strategies to wide range of properties, including metabolic pathways (Kawa-
engineer CARs, T cells that express CARs or tumor-specific lekar et al., 2016), T-cell memory development (Kalos et al.,
T cell receptors (TCRs), and the interaction between engineered 2011; Kawalekar et al., 2016), and antigen-independent tonic
T cells and the tumor microenvironment (TME), with particular signaling (Long et al., 2015), prompting further research into
focus on improving the efficacy and safety of adoptive T cell ther- other costimulatory domains. For example, a third-generation
apy for the treatment of solid tumors (Figure 1). CAR with OX40 and CD28 costimulatory domains repressed
CD28-induced secretion of interleukin (IL)-10, an anti-inflamma-
ENGINEERING THE CAR PROTEIN tory cytokine that compromises T-cell activity (Hombach et al.,
2012). In addition, the inducible T-cell costimulator (ICOS) costi-
Evolution of CAR Designs mulatory domain in combination with either CD28 or 4-1BB
Kuwana et al. (1987) reported the first proof of principle of costimulation increased in vivo persistence, and MyD88/CD40
combining antibody-type antigen specificity with T-cell signaling costimulation improved in vivo proliferation of CAR-T cells (Col-
by fusing the TCR constant region to the variable regions of a bac- linson-Pautz et al., 2019; Guedan et al., 2018).
terial antigen-recognizing antibody. Single-chain variable frag- More recently, fourth-generation CARs that incorporate addi-
ments (scFvs), composed of the variable heavy (VH) and light (VL) tional stimulatory domains, commonly referred to as ‘‘armored’’
chains of a monoclonal antibody (mAb) separated by a flexible CARs, have been reported. In one example, Chmielewski et al.
linker, are still commonly used as the extracellular antigen-sensing (2014) engineered armored CAR-T cells termed ‘‘T cells redir-
domain of CARs. The first reports of tumor-targeting CARs demon- ected for universal cytokine-mediated killing’’ (TRUCK) to
strated that an scFv recognizing antigens such as human epidermal secrete the proinflammatory cytokine IL-12 to stimulate innate
Cancer Cell 38, October 12, 2020 ª 2020 Elsevier Inc. 473
ll
Review
Figure 1. CAR-T Cell Engineering
Approaches
Strategies to engineer CAR-T cells for improved
function in solid tumors include a focus on CAR
engineering, T cell engineering, and TME interac-
tion optimization.
Adapter-Dependent CARs
Alternatively, T cells can be engineered to
express a receptor that must be comple-
mented by an additional protein compo-
nent before the resulting complex is
meaning their CARs are constitutively expressed and are always capable of translating antigen recognition to T-cell activation
capable of signaling upon antigen stimulation. However, this is (Figure 2C). Urbanska et al. (2012) reported a ‘‘universal’’ recep-
not always desirable when CAR-derived toxicity is an anticipated tor comprising a biotin-binding domain fused to an intracellular
risk. In addition to the off-target and ‘‘on-target, off-tumor’’ tox- T-cell signaling domain. T cells expressing this biotin-specific re-
icities discussed previously, various systemic toxicities have ceptor can, in principle, be directed against any target cell that
also been observed in patients treated with CAR-T cells. Com- has been labeled with a biotinylated antibody. Two advantages
mon examples include cytokine release syndrome (CRS) (Fitz- of this strategy include the abilities to (1) control the ON/OFF
gerald et al., 2017; Grupp et al., 2013; Schuster et al., 2017), state of the T cell through administering or withholding the bio-
neurotoxicity or ‘‘CAR-related encephalopathy syndrome’’ tinylated antibody and (2) use the same biotin-specific receptor
(CRES) (Grupp et al., 2013; Lee et al., 2019a; Schuster et al., to target a wide variety of TAAs by changing the specificity of
2017), tumor lysis syndrome, and anaphylaxis (Maus et al., the biotinylated antibody. The concept of using a universal
2013). It has been hypothesized that modulation of CAR-T cell CAR coupled with one or more antigen-targeting adaptor pro-
activity in space and time can prevent or significantly lessen teins to overcome tumor heterogeneity and mitigate toxicity
the severity of toxicities associated with CAR-T cell therapy while soon opened the floodgates for other adapter-dependent CAR
maintaining its antitumor efficacy. designs (Bachmann, 2019; Cartellieri et al., 2016; Herzig et al.,
One approach to modulate CAR-T cell activity is to regulate the 2019; Lee et al., 2019c; Qi et al., 2020; Raj et al., 2019; Rodgers
presence of functional CARs on the surface of engineered T cells et al., 2016), including a small-molecule adapter, which binds to
by adjusting either the stability or the conformation of the CAR both fluorescein isothiocyanate (FITC) and folate, that alleviated
protein itself (Figure 2B). As examples of the former strategy, CRS-like toxicity in an NSG mouse model (Lee et al., 2019b).
CARs have been fused to degradation tags that can be inactivated Expanding on the concept of constitutively expressing a uni-
either by small-molecule binding (Weber et al., 2020) or under hyp- versal receptor on the T-cell surface combined with an externally
oxic conditions (Juillerat et al., 2017). The default state in such sys- administered adaptor protein to dictate antigen specificity, Cho
tems is ‘‘off,’’ and the removal or inactivation of the degradation et al. (2018) reported a SUPRA CAR system that can program
tag is required to stabilize the CAR protein, thus enabling CAR- a variety of Boolean logic gates in engineered T cells by express-
mediated T-cell activation upon antigen stimulation. An alternative ing multiple base receptors, each with multiple potential adaptor
design in which the CAR protein is present by default but turned protein partners reconstituted by leucine-zipper dimerization.
off through the administration of a small-molecule drug has also Such a system supports the possibility of simultaneously
been demonstrated (Juillerat et al., 2019). increasing specificity through the use of AND or AND-NOT gates
Instead of modulating protein half-life, controlling the confor- while addressing tumor heterogeneity by targeting multiple anti-
mation of the CAR protein can also regulate the availability of gens with different adaptor proteins. However, the versatility of
functional CARs, such that the receptor is signaling-competent multi-component systems comes at the cost of an increased
only under specified conditions. For example, the antigen-bind- number of parameters that must be optimized, including the
ing domain of CARs can be ‘‘masked’’ by a built-in inhibitory half-life, biodistribution, and interaction dynamics among the
Safety Controls on CAR-T Cell Activity Regulated CAR Expression to Improve CAR-T Cell Safety
In addition to engineering the CAR protein itself, engineering ap- As an alternative to post-translational regulation of CAR stability
proaches applied at a cellular level to modulate T-cell activity can and function, transcriptional regulation can provide another
also significantly affect therapeutic efficacy and safety. For tunable handle to improve CAR-T cell safety (Figure 3B). For
example, transient CAR expression resulting from mRNA elec- example, in the Tet-ON system, the small molecule doxycycline
troporation instead of viral integration can mitigate toxicities (Dox) acts as an ‘‘ON switch,’’ whereby CAR expression is
induced by CARs that cross-recognize healthy tissue (Beatty induced by the reverse tetracycline transactivator (rtTA) protein
et al., 2014; Tchou et al., 2017; Wiesinger et al., 2019). Another only in the presence of Dox (Sakemura et al., 2016). Conversely,
safeguard against severe toxicity is the implementation of sui- in the Tet-OFF system, Dox acts as an ‘‘OFF switch’’ by abolish-
cide genes that enable the depletion of engineered T cells by ing the ability of tetracycline transactivator (tTA) to activate CAR
the administration of small-molecule drugs (Figure 3A) (Marin transcription; this system was used to reversibly inhibit delete-
et al., 2012). For example, CAR-T cells targeting the CD44v6 an- rious CAR signaling and T-cell fratricide in CD5 CAR-T cells (Ma-
tigen for the treatment of leukemia and myeloma have been en- monkin et al., 2018). In addition, response to a tumor-associated
gineered to express herpes simplex virus thymidine kinase (HSV- environmental cue could be achieved through inducible pro-
TK), which enabled effective elimination of the CAR-T cells upon moters responsive to hypoxia-inducible factor (HIF)-1a, acti-
exposure to ganciclovir in vitro (Casucci et al., 2018). However, vating CAR expression only in hypoxic environments (Ede
expression of viral HSV-TK raises concerns of immunogenicity et al., 2016). Finally, as previously described, the synNotch re-
and requires 3 days of ganciclovir exposure to achieve effective ceptor enables inducible CAR transcription upon binding to a
T-cell elimination (Marin et al., 2012). An alternative approach us- membrane-bound ligand (Roybal et al., 2016; Srivastava
ing the inducible caspase 9 (iCasp9) suicide gene has been et al., 2019).
shown to eliminate >90% of engineered cells within 30 min of
drug administration in human patients (Di Stasi et al., 2011). Site-Specific CAR Transgene Insertion and Allogeneic
iCasp9 contains the intracellular domain of the pro-apoptotic Compatibility Engineering
protein caspase 9 fused to FK506-binding protein (FKBP). The The examples of regulated gene expression provided above uti-
small molecule AP1930 facilitates the dimerization of FKBP lize synthetic inducible promoters, and the entire gene-expres-
and activation of the fused caspase 9, inducing apoptosis in cells sion cassette is typically integrated into the T-cell genome using
expressing the iCasp9 protein (Straathof et al., 2005). In addition, retroviral or lentiviral vectors, leading to variable integration sites
the expression of transgenes for the surface proteins CD20 or and copy numbers. An alternative method to achieve dynamic
truncated epidermal growth factor receptor (tEGFR) can also CAR expression profiles is to integrate the transgene into
facilitate suicide mechanisms. The US Food and Drug Adminis- specific genetic loci regulated by endogenous transcriptional
tration-approved mAbs rituximab and cetuximab bind to CD20 machinery. A variety of gene-editing technologies, including
and tEGFR, respectively, and the resulting antibody-dependent clustered regularly interspaced short palindromic repeats
cellular cytotoxicity can be used to eliminate T cells engineered (CRISPR) and CRISPR-associated protein 9 (Cas9), transcription
to express these antigens (Griffioen et al., 2009; Serafini et al., activator-like effector nucleases (TALENs), and zinc-finger nu-
2004; Wang et al., 2011). cleases (ZFNs), have made this a feasible approach in T-cell en-
While suicide genes can efficiently deplete CAR-T cells to gineering (Chen, 2015). For example, Eyquem et al. (2017)
counter toxicities, the activation of a suicide gene also results demonstrated that CRISPR-Cas9-mediated insertion of the
in the irreversible termination of the therapy. An alternative CD19 CAR transgene into the TCRa constant (TRAC) locus re-
reversible strategy utilizes dasatinib, a tyrosine kinase inhibitor sults in CAR-T cells with superior in vivo function compared
In the studies referenced above, endogenous gene disruption (RNP) complex (Liu et al., 2017) or by electroporation of mRNA
and CAR transgene integration were independently executed, encoding Cas9 and sgRNAs into CAR-T cells (Ren et al.,
yielding heterogeneous CAR-T cell populations. Georgiadis 2017a). Furthermore, triple-targeting of the TRAC, B2M, and
et al. (2018) coupled gene-editing to CAR integration by incorpo- FAS loci using a lentiviral vector encoding both the sgRNAs
rating a single-guide RNA (sgRNA) element into the U3 region of and a CD19-targeting CAR and electroporation of Cas9 mRNA
the 30 long terminal repeat sequence of the CAR-encoding lenti- resulted in reduced alloreactivity and prolonged T-cell survival
viral vector. After electroporation of Cas9 mRNA, magnetic by abrogating pro-apoptotic Fas/FasL signaling (Ren et al.,
bead-based selection for edited cells resulted in highly enriched 2017b). Finally, CRISPR-Cas9-mediated deletion has been
CAR+ TCR populations. In another strategy, Ren et al. (2017b) used to improve the safety of CAR-T cells by disrupting the
used a ‘‘one-shot’’ CRISPR system with multiple sgRNA expression of CRS-associated cytokines, such as granulocyte-
expression cassettes in one CAR-encoding lentiviral vector to macrophage colony-stimulating factor (GM-CSF) (Sterner
simultaneously knock out the endogenous TCR and Beta-2 mi- et al., 2019).
croglobulin (b2M), an essential subunit of the HLA-I molecule,
to generate CD19 CAR-T cells suitable for allogeneic therapy. Receptors that Rewire an Inhibitory Input to a
Stimulatory Output
Knockout of Negative Regulators While CAR-T cells have shown promising therapeutic efficacy
Gene-editing technologies can also be used to abrogate the against hematologic malignancies, their targeting of solid tumors
expression of negative regulators of T-cell activity (Table 1). Tu- has been stymied by a number of challenges. In addition to
mor cells frequently upregulate ligands to immune-checkpoint antigen heterogeneity, tumor cells produce tumorigenic and
receptors, such as cytotoxic T-lymphocyte-associated antigen immunosuppressive factors in the TME, and this inhibitory envi-
4 (CTLA-4) and PD-1 expressed on T cells, leading to inhibition ronment is further compounded by immunosuppressive cell
of T-cell activity in the TME (Buchbinder and Desai, 2016). Im- types such as MDSCs and Tregs (Labanieh et al., 2018). Among
mune-checkpoint blockade using antibodies targeting CTLA-4, the immunosuppressive soluble factors commonly found in the
PD-1, or PD-1 ligand (PD-L1) has revolutionized the field of im- TME, transforming growth factor b (TGF-b) is a particularly
muno-oncology in recent years (Wei et al., 2018). An alternative potent cytokine that drives T-cell differentiation into Tregs, as
approach to checkpoint blockade is the ablation of checkpoint- well as macrophage polarization to the immunosuppressive
receptor expression on engineered T cells. Menger et al. (2016) M2 phenotype (Tormoen et al., 2018). The engineering of a
demonstrated that PD-1 knockout through TALEN-mediated ed- TGF-b-responsive CAR demonstrated that CARs can be used
iting in melanoma-reactive CD8+ T cells and fibrosarcoma-reac- to (1) sense a soluble factor and (2) rewire T cells to convert an
tive polyclonal T cells enhanced the persistence of the modified inhibitory signal to a trigger of antitumor activity (Figure 4) (Chang
T cells, augmenting their antitumor activity against syngeneic tu- et al., 2018; Hou et al., 2018). TGF-b-responsive CAR-T cells
mor models and establishing long-term antitumor memory. In were demonstrated to proliferate and produce T helper type 1
addition, CRISPR-Cas9-mediated disruption of PD-1 in CD19 (Th1)-associated cytokines in the presence of soluble TGF-b,
CAR-T cells enhanced CAR-T cell cytotoxicity toward PD-L1+ tu- as well as protect nearby cells from the inhibitory effects of
mor xenografts in vivo (Rupp et al., 2017). The first CRISPR- TGF-b, likely through the combined effects of TGF-b sequestra-
Cas9-edited cell therapy trial conducted in the United States tion and paracrine Th1 cytokine signaling (Chang et al., 2018;
evaluated the adoptive transfer of autologous T cells genetically Hou et al., 2018).
modified to express a New York esophageal squamous cell car- Signal rewiring can also be achieved with ‘‘switch receptors,’’
cinoma 1 (NY-ESO-1)-targeting TCR while eliminating endoge- which are chimeras comprising an ectodomain that binds a sup-
nous TCRab and PD-1 expression; recent results from the trial pressive molecule fused to an endodomain that drives a stimula-
confirmed the safety and feasibility of CRISPR-edited cell ther- tory pathway (Figure 4). IL-4 is a cytokine with complex roles in
apy for cancer (Stadtmauer et al., 2020). the TME, such as inducing M2 polarization, promoting tumor
In addition to PD-1 knockout, elimination of the metabolic growth, and suppressing tumor-specific effector T cells (Tor-
regulator REGNASE-1 and CD3-signaling regulator diacylgly- moen et al., 2018; Wilkie et al., 2010). IL-4 switch receptors con-
cerol kinase (DGK) has also been shown to enhance T-cell func- sisting of the IL-4 receptor a (IL-4Ra) ectodomain fused to either
tion in vitro and in vivo (Jung et al., 2018; Riese et al., 2013; Wei the IL-7Ra endodomain or the bc receptor subunit common to IL-
et al., 2019). Similar to PD-1, lymphocyte-activation gene 3 2 and IL-15 signaling have been shown to paradoxically enhance
(LAG3) is known as a T-cell exhaustion marker, and it has been T-cell proliferation in the presence of IL-4 (Leen et al., 2014;
found to inhibit the effector function of T cells and enhance the Wilkie et al., 2010). Consequently, the coexpression of the IL-
suppressive function of Tregs (Zhang et al., 2017). However, 4Ra:bC switch receptor in CAR-T cells augmented their anti-
CRISPR-Cas9-mediated deletion of LAG3 in CD19 CAR-T cells tumor capacity (Wilkie et al., 2010). More recently, Roth et al.
did not result in a discernable phenotype, perhaps due to (2020) used a pooled knockin screen using CRISPR-Cas9
compensatory effects from PD-1 (Woo et al., 2012; Zhang coupled with single-cell RNA sequencing (scRNA-seq) to eval-
et al., 2017). uate a panel of transgenes integrated into the TRAC locus.
Several studies have reported the feasibility of multiple knock- Through this analysis, a novel TGFbR2:4-1BB switch receptor
outs using CRISPR-Cas9 in CAR-T cells to achieve the dual was identified as the lead candidate for improved T-cell fitness
goals of reducing alloreactivity and improving T-cell function. Tri- and solid-tumor clearance.
ple-targeting of the TRAC, B2M, and PD-1 loci has been demon- Switch receptors can also mitigate the suppressive effects
strated by electroporation of Cas9:sgRNA ribonucleoprotein of immune checkpoints. For example, CTLA-4:CD28 and
PD-1:CD28 switch receptors were shown to enhance the activity The aberrant expression of the colony-stimulating factor 1
of tumor-specific T cells (Liu et al., 2016; Shin et al., 2012). As (CSF-1) in the TME drives macrophages to the M2 phenotype
another checkpoint receptor, T-cell immunoreceptor with Ig and promotes tumor growth. Several clinical trials of small mol-
and ITIM domains (TIGIT) binds to CD155 or CD112 on tumor ecules and mAbs targeting the CSF-1/CSF-1R axis in combina-
cells, and these interactions hinder cytokine production and tion with other mAbs and/or chemotherapy are under way for the
effector function of T cells. To blunt these effects, Hoogi et al. treatment of solid tumors (NCT01525602, NCT02777710,
(2019) designed a TIGIT:CD28 switch receptor that enhanced NCT02323191, NCT02760797, NCT02923739). In the context
cytokine production and activation of T cells, as well as delayed of CAR-T cells, the coexpression of CSF-1R, which T cells do
tumor growth in vivo when combined with a melanoma-spe- not naturally express, was shown to confer CSF-1 responsive-
cific TCR. ness and activated the RAS/MEK/Erk kinase pathway to
Yet another set of engineering strategies focuses on address- enhance T-cell proliferation, cytokine production, and CSF-1-
ing tumorigenic factors that are not directly expressed on T cells driven chemotaxis (Lo et al., 2008).
or tumor cells (Figure 4). For example, vascular endothelial
growth factor-A (VEGF-A) is a tumor-derived soluble factor that Metabolic Reprogramming of T Cells
promotes tumor growth and metastasis by facilitating angiogen- T-cell metabolism, or the manner in which T cells utilize nutrient
esis. Chinnasamy et al. (2012) cotransduced T cells with a CAR sources, has consequential effects on their differentiation state
targeting VEGF receptor 2 (VEGFR2) and an inducible transgene and effector function. The architecture of the CAR protein can
encoding IL-12. These CAR-T cells demonstrated trafficking to have an impact on the metabolic profiles of CAR-T cells. For
the tumor vasculature, elimination of VEGFR+ MDSC subtypes example, T cells expressing CARs with 4-1BB costimulation
that participate in tumor angiogenesis, and IL-12-mediated favor the oxidative breakdown of fatty acids characteristic of
solid-tumor regression. In addition, T cells expressing a CAR the central-memory phenotype, accompanied by enhanced pro-
that targets fibroblast activation protein (FAP) eliminated FAP+ liferation and persistence, while T cells expressing CARs with
tumor stromal fibroblasts that support tumor growth, augment- CD28 costimulation favor aerobic glycolysis characteristic of
ing T-cell immunotherapy (Wang et al., 2014). However, strate- the effector-memory phenotype (Kawalekar et al., 2016).
gies targeting non-tumor tissues can cause significant toxicity. Due to the intimate relationship between T-cell metabolism
For example, Tran et al. (2013) reported that CAR-T cells and function, reprogramming the metabolic profile of CAR-T
cross-reactive to human and mouse FAP cause off-target bone cells can potentially increase their clinical efficacy. The TME of
marrow toxicities and cachexia. solid tumors has an overabundance of potassium (K+) released
by necrotic tumor cells, which increases the intracellular K+ con-
Transgene Expression to Promote T-Cell Function centration in infiltrating T cells, downregulating Protein kinase B
In addition to abolishing or rewiring inhibitory signals, the overex- (Akt)/mammalian target of rapamycin (mTOR) signaling and im-
pression of stimulatory signals can also improve the activity of pairing T-cell activation after TCR ligation. As a counterstrategy,
tumor-specific T cells. The constitutive expression of costimula- the overexpression of K+ channels was shown to increase Akt/
tory ligands CD80 and 4-1BBL in PSMA-targeting CAR-T cells mTOR activity and rescue T-cell effector functions by facilitating
showed superior activity against prostate tumor cells by K+ efflux and lowering intracellular K+ levels (Eil et al., 2016). In
engaging costimulatory receptors in cis (autocostimulation) addition, elevated K+ in the TME and the resulting perturbation
and in trans (transcostimulation of bystander cells) (Stephan of the transmembrane electrochemical gradient limit nutrient up-
et al., 2007). Furthermore, Zhao et al. (2015) coexpressed the take by T cells. Interestingly, ex vivo conditioning and activation
4-1BBL costimulatory ligand with a second-generation CD19- of tumor-specific CD8+ T cells in elevated K+ to mimic such func-
targeting CAR with CD28 costimulation, and this combined cos- tional starvation in the TME led to epigenetic and metabolic re-
timulation led to improved antitumor functions driven by the programming that maintained T-cell stemness—demonstrated
continuous activation of the interferon regulatory factor 7 by improved persistence, engraftment, self-renewal, and multi-
(IRF7)/interferon beta (IFNb) pathway. Yet another strategy har- potency—thereby enhancing their antitumor function in vivo
nesses the endogenous IL-7 signaling mechanism that confers (Vodnala et al., 2019). Further, Geiger et al. (2016) showed sup-
improved persistence in tumor-specific T cells. Shum et al. plementing L-arginine to balance increased arginine metabolism
(2017) engineered a constitutively active IL-7 receptor and coex- in activated T cells promotes the central-memory phenotype and
pressed it with a GD2-targeting CAR, which resulted in improves antitumor activity.
enhanced survival under repeated tumor challenges in neuro- Altering the expression levels of metabolic genes can also pro-
blastoma and glioblastoma xenograft models. mote an advantageous metabolic profile. Phosphoenolpyruvate
(2020) recently demonstrated the feasibility of adenoviral trans- express CCR2b, the dominant isoform of the chemokine recep-
duction of a CAR into primary macrophages. The resulting tor of CCL2, resulting in enhanced T-cell homing to CCL2-ex-
CAR-M cells exhibited tumor-specific phagocytosis, inflamma- pressing neuroblastoma and malignant pleural mesothelioma
tory cytokine production, polarization of bystander macro- xenografts, respectively (Craddock et al., 2010; Moon et al.,
phages to the immunostimulatory M1 phenotype, and cross-pre- 2011). Similarly, CAR-T cells that coexpress CCR4 showed
sentation of the TAA to bystander T cells. Although a comparison improved migration toward tumors expressing CCL17 and
between CAR-T and CAR-M cells was not evaluated, the estab- CCL22 in vivo, while those expressing CXCR1 or CXCR2 ex-
lished role of macrophages as professional antigen-presenting hibited enhanced homing toward tumor-derived IL-8 (Di Stasi
cells (APCs) warrants the potential of CAR-M cells to more effec- et al., 2009; Jin et al., 2019). Once CAR-T cells reach the tumor
tively stimulate an adaptive antitumor immune response. site, their infiltration is hindered by the high-density structural
CAR-natural killer (NK) cells can be generated from cord blood extracellular matrix (ECM) associated with solid-tumor nodules.
or iPSCs (Li et al., 2018; Liu et al., 2020), making them an attrac- Accordingly, CAR-T cells engineered to express heparinase, an
tive candidate for allogeneic, off-the-shelf products. Moreover, enzyme that degrades ECM, have been shown to improve tumor
CD19-targeting CAR-NK cells have achieved robust clinical effi- infiltration and overall survival in multiple xenograft models (Car-
cacy without inducing CRS, neurotoxicity, or GvHD in patients uana et al., 2015).
with B-cell lymphoid tumors, highlighting their relative safety CAR-T cells that successfully reach solid tumors are next
compared with their T-cell counterparts (Liu et al., 2020). CAR- faced with a multitude of suppressive and evasive features that
NK cells have been shown to exert potent and specific cytotox- induce CAR-T cell dysfunction (Newick et al., 2017). To improve
icity toward a variety of tumor models, including leukemia, mul- therapeutic efficacy in this immunosuppressive environment,
tiple myeloma, ovarian cancer, and glioblastoma (Chu et al., CAR-T cells have been engineered to produce proteins that (1)
2014; Genbler et al., 2016; Li et al., 2018; Quintarelli et al., improve CAR-T cell function in an autocrine fashion; (2) disrupt
2020), as well as toward immunosuppressive cell types such immunosuppressive elements; and/or (3) induce TME remodel-
as MDSCs and follicular helper T cells (TFH) (Parihar et al., ing to enhance the endogenous antitumor immune response
2019; Reighard et al., 2020). Lastly, natural killer T (NKT) cells (Figure 5). Each of these strategies is discussed in detail below.
possess antitumor and tumor-homing capabilities, and GD2-tar-
geting CAR-NKT cells that harness these inherent advantages Autocrine Stimulation of CAR-T Cells in the TME
exhibited effective localization to and lysis of neuroblastoma Cytokines are signaling proteins with the ability to drastically
cells without significant toxicity (Xu et al., 2019). Taken together, augment or abrogate CAR-T cell function. Coexpressing the
these developments highlight both the potential for cell-based CAR with immunostimulatory cytokines could significantly
immunotherapy to expand beyond T cells and the applicability enhance CAR-T cell proliferation, survival, and effector function
of the CAR technology across a variety of immune cell types. in the immunosuppressive TME. For example, T cells that consti-
tutively coexpress a CD19-targeting CAR plus IL-2, IL-7, IL-15,
ENGINEERING CAR-T CELL INTERACTIONS WITH THE or IL-21 have been shown to achieve greater in vivo tumor con-
TUMOR AND TME trol compared with T cells expressing the CAR alone (Markley
and Sadelain, 2010). Interestingly, although the receptor com-
The immune-evasive and immunosuppressive nature of the TME plexes of these four cytokines contain the common gamma
contributes to the poor therapeutic efficacy of CAR-T cells chain (gc), each cytokine differentially affected the proliferation,
observed in solid tumors. Hallmarks of the TME, which have subtype differentiation, and function of the engineered T cells,
been extensively reviewed elsewhere (Gajewski et al., 2013; underscoring the complexity of T-cell biology and variety of
Whiteside, 2008), include (1) physical barriers to tumor penetra- potential outcomes achievable through different engineering
tion by immune cells, (2) upregulated checkpoint ligands, (3) a strategies (Markley and Sadelain, 2010). T cells coexpressing
pro-tumor stromal niche, (4) abundant immunosuppressive and IL-12, IL-15, IL-18, and/or IL-21 plus a CAR targeting a variety
pro-metastatic soluble factors, and (5) modulated expression of antigens have also been described, resulting in improved effi-
of chemokines to preferentially recruit leukocytes with an immu- cacy, proliferation, and/or persistence in vivo (Batra et al., 2020;
nosuppressive phenotype. These factors have in turn driven the Hoyos et al., 2010; Hu et al., 2017; Koneru et al., 2015; Krenciute
design of CAR-T cells that respond to TME elements to enhance et al., 2017; Pegram et al., 2012). However, constitutive overex-
CAR-T cell efficacy (Figure 5). pression of immunostimulatory cytokines can also increase
toxicity (Zhang et al., 2011). Regulatory strategies previously dis-
Tumor Homing and Penetration cussed in this review, such as inducible promoters, can be im-
The efficacy of CAR-T cell therapy in solid tumors is significantly plemented to modulate cytokine production and associated
hindered by poor immune-cell infiltration (Newick et al., 2017). T- toxicity (Liu et al., 2019).
cell migration is regulated through chemokine axes. Tumor cells
can upregulate or downregulate chemokines, as well as modu- Disruption of Immunosuppressive Axes
late chemokine expression by tumor-associated cells, contrib- The expression of immune-checkpoint receptors and ligands
uting to the poor recruitment of CAR-T cells (Oelkrug and Ram- such as PD-1 and PD-L1 are prevalent in the TME, and they
age, 2014). Engineering CAR-T cells to overexpress receptors can potently inhibit CAR-T cell cytotoxicity and induce anergy
for chemokines that are overexpressed in the TME can turn a tu- (Drake et al., 2006). Thus, immune-checkpoint blockade has
mor’s defense mechanism against itself. For example, GD2- and strong synergistic potential with CAR-T cell therapy, and several
mesothelin-targeting CAR-T cells have been engineered to co- ongoing clinical trials are evaluating combination therapy with
CAR-T cells and exogenously administered checkpoint inhibi- in that the DNR does not transduce any signals that can stimulate
tors (Grosser et al., 2019). Furthermore, CAR-T cells have been the engineered T cell. It remains to be seen whether the stimula-
engineered to secrete immune-checkpoint inhibitors, including tory effects of the CAR and switch receptors will confer addi-
anti-PD-1 scFvs and antiPD-L1 antibodies, or to express PD-1 tional clinical benefits compared with the DNR.
dominant-negative receptors (DNRs) (Chen et al., 2017; Cher- In the TME, IL-6 is often overexpressed by tumor cells, tumor-
kassky et al., 2016; Li et al., 2017; Rafiq et al., 2018; Suarez associated macrophages (TAMs), and other resident cells (Ku-
et al., 2016). In addition to enhancing efficacy, this approach mari et al., 2016). IL-6 supports tumorigenesis through a number
may also avoid toxicities associated with systemic immune- of mechanisms and plays a central role in the induction of CRS
checkpoint blockade by restricting checkpoint inhibitor distribu- after CAR-T cell infusion (Lee et al., 2014). Systemic administra-
tion to the immediate environment of the producer T cells. For tion of tocilizumab, an mAb targeting IL-6 receptor alpha (IL-
example, it has been shown that anti-PD-1 scFvs secreted by 6Ra), has become standard treatment for CRS after CAR-T cell
intraperitoneally (IP) injected CAR-T cells remained localized at therapy (Kotch et al., 2019). More recently, CD19-targeting
the injection site. However, when an equal number of conven- CAR-T cells that coexpress a non-signaling, membrane-bound
tional CAR-T cells were administered IP with exogenous anti- IL-6 receptor (mbaIL6) were shown to sequester IL-6 while
PD-1 antibody, the antibody was detected systemically within retaining in vivo antitumor efficacy (Tan et al., 2020). However,
3 h (Rafiq et al., 2018). it remains to be seen if CAR-T cells engineered in this fashion
The solid-tumor milieu also houses a diverse collection of sol- can prevent CRS.
uble factors that promote tumorigenesis and inhibit CAR-T cell
function. For example, prostaglandin E2 (PGE2) is a bioactive
lipid often upregulated in tumors, where it contributes to tumor TME Remodeling to Promote the Endogenous Immune
survival through regulation of cell proliferation, migration, Response
apoptosis, and angiogenesis (Ricciotti and FitzGerald, 2011; Tumors are adept at selectively attracting or evading subsets of
Wang and Dubois, 2006). In the context of CAR-T cell therapy, leukocytes, including CAR-T cells, to promote immune regulation
PGE2, along with adenosine, inhibits T-cell signaling and activa- or suppression (Rabinovich et al., 2007). In addition, tumors are
tion though the activation of protein kinase A (PKA), thereby often capable of inducing an immunosuppressive or pro-metasta-
reducing T-cell proliferation and effector function (Newick tic phenotype on the local stroma, as well as an anti-inflammatory
et al., 2016). In two solid-tumor models that highly express or dysfunctional phenotype on resident leukocytes (Morgan and
PGE2, CAR-T cells engineered to express a peptide inhibitor of Schambach, 2018). Another approach to enhancing the efficacy
ezrin-mediated PKA translocation to the immune synapse ex- of CAR-T cell therapy is to reverse this immunosuppressive cell
hibited improved tumor infiltration and killing (Newick et al., niche through remodeling the tumor-cellular composition and
2016). Similarly, elevated concentrations of bioreactive chemi- phenotype. To realize this, CAR-T cells have been engineered to
cals such as reactive oxygen species (ROS) in the TME play an secrete cytokines or other soluble factors that induce TME remod-
important role in tumorigenesis (Weinberg et al., 2019). Catalase eling in a paracrine or endocrine fashion.
is an enzyme that facilitates the decomposition of hydrogen In germinal-center lymphomas, loss of herpesvirus entry
peroxide (H2O2), an ROS that impairs T-cell activity in the TME. mediator (HVEM) expression induces the secretion of non-
Increasing intracellular levels of catalase by coexpressing the redundant stroma-activating factors, resulting in acute
catalase gene in HER2-and carcinoembryonic antigen (CEA)- lymphoid-stroma activation. The hyperactivated stroma recruits
specific CAR-T cells has been shown to enable CAR-T cells to TFH cells, which support malignant B cells through CD40/CD40L
metabolize the suppressive H2O2, improving their tumor cytolytic interactions and cytokine stimulation. As a counterstrategy,
capacity (Ligtenberg et al., 2016). CD19 CAR-T cells engineered to secrete a soluble form of
The aberrant expression of cytokines in the TME plays a crit- HVEM were shown to enhance tumor control in vivo (Boice
ical role in tumor progression and resistance to CAR-T cell ther- et al., 2016).
apy. In particular, TGF-b plays a multiplexed role in cancer pro- CAR-T cells engineered to secrete IL-12 have been shown to
gression through interactions with tumor cells, stroma, and both remodel the TME by reprogramming TAMs to an M1 phenotype
innate and adaptive immune cells to induce (1) the secretion of and decreasing the presence of MDSCs and Tregs in syngeneic
immunosuppressive chemokines, cytokines, and growth fac- mouse models (Chinnasamy et al., 2012; Liu et al., 2019; Yeku
tors; (2) ECM remodeling and matrix deposition; (3) immunosup- et al., 2017). Similarly, CAR-T cells that constitutively secrete
pressive reprogramming of macrophages, neutrophils, and IL-18 can alter the TME makeup by increasing intratumoral M1
T cells; and (4) inhibited maturation or proliferation of T cells macrophage, activated dendritic cell (DC), and activated NK
and NK cells (Pickup et al., 2013). To ablate these powerful ef- cell numbers, while decreasing M2 macrophage and Treg levels.
fects, CAR-T cells have been engineered to express a TGF-b A direct comparison of CAR-T cells expressing IL-12- to IL-18
DNR that potently inhibits endogenous TGF-b signaling, result- indicated that IL-18 is more effective at remodeling the immuno-
ing in T cells with enhanced proliferation and antitumor efficacy suppressive TME in a syngeneic murine pancreatic cancer
in a prostate cancer xenograft model (Kloss et al., 2018). Based model (Chmielewski and Abken, 2017). Furthermore, CD19
on these results, a phase 1 clinical trial has been initiated to CAR-T cells expressing IL-18 induced the expansion of endoge-
assess T cells coexpressing a PSMA CAR and the DNR for the nous CD8+ T cells, NK cells, NKT cells, and DCs in the bone
treatment of relapsed and refractory metastatic prostate cancer marrow, potentially contributing to the control of tumors with
(NCT03089203). The DNR is distinct from the TGF-b-targeting heterogeneous CD19 expression in a syngeneic mouse model
CAR and TGF-b switch receptor discussed in a previous section (Avanzi et al., 2018).
Casucci, M., Falcone, L., Camisa, B., Norelli, M., Porcellini, S., Stornaiuolo, A., Drake, C.G., Jaffee, E., and Pardoll, D.M. (2006). Mechanisms of immune
Ciceri, F., Traversari, C., Bordignon, C., Bonini, C., and Bondanza, A. (2018). evasion by tumors. Adv. Immunol. 90, 51–81.
Extracellular NGFR spacers allow efficient tracking and enrichment of fully
functional CAR-T cells co-expressing a suicide gene. Front. Immunol. 9, 507. Ede, C., Chen, X., Lin, M.Y., and Chen, Y.Y. (2016). Quantitative analyses of
core promoters enable precise engineering of regulated gene expression in
Cella, M., Scheidegger, D., Palmer-Lehmann, K., Lane, P., Lanzavecchia, A., mammalian cells. ACS Synth. Biol. 5, 395–404.
and Alber, G. (1996). Ligation of CD40 on dendritic cells triggers production
of high levels of interleukin-12 and enhances T cell stimulatory capacity: T-T Eil, R., Vodnala, S.K., Clever, D., Klebanoff, C.A., Sukumar, M., Pan, J.H.,
help via APC activation. J. Exp. Med. 184, 747–752. Palmer, D.C., Gros, A., Yamamoto, T.N., Patel, S.J., et al. (2016). Ionic immune
suppression within the tumour microenvironment limits T cell effector function.
Chang, Z.L., Lorenzini, M.H., Chen, X., Tran, U., Bangayan, N.J., and Chen, Nature 537, 539–543.
Y.Y. (2018). Rewiring T-cell responses to soluble factors with chimeric antigen
receptors. Nat. Chem. Biol. 14, 317–324. Eliopoulos, A.G., Davies, C., Knox, P.G., Gallagher, N.J., Afford, S.C., Adams,
D.H., and Young, L.S. (2000). CD40 induces apoptosis in carcinoma cells
Chen, L., and Flies, D.B. (2013). Molecular mechanisms of T cell co-stimulation through activation of cytotoxic ligands of the tumor necrosis factor superfam-
and co-inhibition. Nat. Rev. Immunol. 13, 227–242. ily. Mol. Cell Biol. 20, 5503–5515.
Chen, N., Morello, A., Tano, Z., and Adusumilli, P.S. (2017). CAR T-cell intrinsic Eshhar, Z., Waks, T., Gross, G., and Schindler, D.G. (1993). Specific activation
PD-1 checkpoint blockade: a two-in-one approach for solid tumor immuno- and targeting of cytotoxic lymphocytes through chimeric single chains con-
therapy. Oncoimmunology 6, e1273302. sisting of antibody-binding domains and the gamma or zeta subunits of the
immunoglobulin and T-cell receptors. Proc. Natl. Acad. Sci. U S A 90, 720–724.
Chen, Y.Y. (2015). Efficient gene editing in primary human T cells. Trends Im-
munol. 36, 667–669. Eyquem, J., Mansilla-Soto, J., Giavridis, T., van der Stegen, S.J., Hamieh, M.,
Cunanan, K.M., Odak, A., Gonen, M., and Sadelain, M. (2017). Targeting a
Cherkassky, L., Morello, A., Villena-Vargas, J., Feng, Y., Dimitrov, D.S., Jones, CAR to the TRAC locus with CRISPR/Cas9 enhances tumour rejection. Nature
D.R., Sadelain, M., and Adusumilli, P.S. (2016). Human CAR T cells with cell- 543, 113–117.
intrinsic PD-1 checkpoint blockade resist tumor-mediated inhibition. J. Clin.
Invest. 126, 3130–3144. Fedorov, V.D., Themeli, M., and Sadelain, M. (2013). PD-1– and CTLA-4–
based inhibitory chimeric antigen receptors (iCARs) divert off-target immuno-
Chinnasamy, D., Yu, Z., Kerkar, S.P., Zhang, L., Morgan, R.A., Restifo, N.P., therapy responses. Sci. Transl. Med. 5, 2–15ra172.
and Rosenberg, S.A. (2012). Local delivery of interleukin-12 using T cells tar-
geting VEGF receptor-2 eradicates multiple vascularized tumors in mice. Fitzgerald, J.C., Weiss, S.L., Maude, S.L., Barrett, D.M., Lacey, S.F., Melen-
Clin. Cancer Res. 18, 1672–1683. horst, J.J., Shaw, P., Berg, R.A., June, C.H., Porter, D.L., et al. (2017). Cytokine
release syndrome after chimeric antigen receptor T cell therapy for acute
Chmielewski, M., and Abken, H. (2017). CAR T cells releasing IL-18 convert to lymphoblastic leukemia. Crit. Care Med. 45, e124–e131.
T-Bet(high) FoxO1(low) effectors that exhibit augmented activity against
advanced solid tumors. Cell Rep. 21, 3205–3219. Fraietta, J.A., Nobles, C.L., Sammons, M.A., Lundh, S., Carty, S.A., Reich, T.J.,
Cogdill, A.P., Morrissette, J.J.D., DeNizio, J.E., Reddy, S., et al. (2018). Disrup-
Chmielewski, M., Hombach, A.A., and Abken, H. (2014). Of CARs and tion of TET2 promotes the therapeutic efficacy of CD19-targeted T cells. Na-
TRUCKs: chimeric antigen receptor (CAR) T cells engineered with an inducible ture 558, 307–312.
cytokine to modulate the tumor stroma. Immunol. Rev. 257, 83–90.
Fry, T.J., Shah, N.N., Orentas, R.J., Stetler-Stevenson, M., Yuan, C.M., Ram-
Cho, J.H., Collins, J.J., and Wong, W.W. (2018). Universal chimeric antigen re- akrishna, S., Wolters, P., Martin, S., Delbrook, C., Yates, B., et al. (2018).
ceptors for multiplexed and logical control of T cell responses. Cell 173, 1426– CD22-targeted CAR T cells induce remission in B-ALL that is naive or resistant
1438.e11. to CD19-targeted CAR immunotherapy. Nat. Med. 24, 20–28.
Hou, A.J., Chang, Z.L., Lorenzini, M.H., Zah, E., and Chen, Y.Y. (2018). TGF-b– Kumari, N., Dwarakanath, B.S., Das, A., and Bhatt, A.N. (2016). Role of inter-
responsive CAR-T cells promote anti-tumor immune function. Bioeng. Transl. leukin-6 in cancer progression and therapeutic resistance. Tumour Biol. 37,
Med. 3, 75–86. 11553–11572.
Labanieh, L., Majzner, R.G., and Mackall, C.L. (2018). Programming CAR-T Marin, V., Cribioli, E., Philip, B., Tettamanti, S., Pizzitola, I., Biondi, A., Biagi, E.,
cells to kill cancer. Nat. Biomed. Eng. 2, 377–391. and Pule, M. (2012). Comparison of different suicide-gene strategies for the
safety improvement of genetically manipulated T cells. Hum. Gene Ther.
Lai, J., Mardiana, S., House, I.G., Sek, K., Henderson, M.A., Giuffrida, L., Chen, Methods 23, 376–386.
A.X.Y., Todd, K.L., Petley, E.V., Chan, J.D., et al. (2020). Adoptive cellular ther-
apy with T cells expressing the dendritic cell growth factor Flt3L drives epitope Markley, J.C., and Sadelain, M. (2010). IL-7 and IL-21 are superior to IL-2 and
spreading and antitumor immunity. Nat. Immunol. 21, 914–926. IL-15 in promoting human T cell-mediated rejection of systemic lymphoma in
immunodeficient mice. Blood 115, 3508–3519.
Lee, D.W., Gardner, R., Porter, D.L., Louis, C.U., Ahmed, N., Jensen, M.,
Grupp, S.A., and Mackall, C.L. (2014). Current concepts in the diagnosis and Maude, S.L., Frey, N., Shaw, P.A., Aplenc, R., Barrett, D.M., Bunin, N.J., Chew,
management of cytokine release syndrome. Blood 124, 188–195. A., Gonzalez, V.E., Zheng, Z., Lacey, S.F., et al. (2014). Chimeric antigen re-
ceptor T cells for sustained remissions in leukemia. N. Engl. J. Med. 371,
Lee, D.W., Santomasso, B.D., Locke, F.L., Ghobadi, A., Turtle, C.J., Brudno, 1507–1517.
J.N., Maus, M.V., Park, J.H., Mead, E., Pavletic, S., et al. (2019a). ASTCT
consensus grading for cytokine release syndrome and neurologic toxicity asso- Maus, M., Haas, A., Beatty, G., Albelda, S., Levine, B., Liu, X., Zhao, Y., Kalos,
ciated with immune effector cells. Biol. Blood Marrow Transpl. 25, 625–638. M., and June, C. (2013). T cells expressing chimeric antigen receptors can
cause anaphylaxis in humans. Cancer Immunol. Res. 1, 26–31.
Lee, Y.G., Chu, H., Lu, Y., Leamon, C.P., Srinivasarao, M., Putt, K.S., and Low,
P.S. (2019b). Regulation of CAR T cell-mediated cytokine release syndrome- Menger, L., Sledzinska, A., Bergerhoff, K., Vargas, F.A., Smith, J., Poirot, L.,
like toxicity using low molecular weight adapters. Nat. Commun. 10, 2681. Pule, M., Hererro, J., Peggs, K.S., and Quezada, S.A. (2016). TALEN-mediated
inactivation of PD-1 in tumor-reactive lymphocytes promotes intratumoral T-
Lee, Y.G., Marks, I., Srinivasarao, M., Kanduluru, A.K., Mahalingam, S.M., Liu, cell persistence and rejection of established tumors. Cancer Res. 76,
X., Chu, H., and Low, P.S. (2019c). Use of a single CAR T cell and several bis- 2087–2093.
pecific adapters facilitates eradication of multiple antigenically different solid
tumors. Cancer Res. 79, 387–396. Mestermann, K., Giavridis, T., Weber, J., Rydzek, J., Frenz, S., Nerreter, T.,
Mades, A., Sadelain, M., Einsele, H., and Hudecek, M. (2019). The tyrosine ki-
Leen, A.M., Sukumaran, S., Watanabe, N., Mohammed, S., Keirnan, J., Yana-
nase inhibitor dasatinib acts as a pharmacologic on/off switch for CAR T cells.
gisawa, R., Anurathapan, U., Rendon, D., Heslop, H.E., Rooney, C.M., et al.
Sci. Transl. Med. 11, eaau5907.
(2014). Reversal of tumor immune inhibition using a chimeric cytokine recep-
tor. Mol. Ther. 22, 1211–1220. Montel-Hagen, A., Seet, C.S., Li, S., Chick, B., Zhu, Y., Chang, P., Tsai, S.,
Sun, V., Lopez, S., Chen, H.C., et al. (2019). Organoid-induced differentiation
Li, S., Siriwon, N., Zhang, X., Yang, S., Jin, T., He, F., Kim, Y.J., Mac, J., Lu, Z.,
of conventional T cells from human pluripotent stem cells. Cell Stem Cell 24,
Wang, S., et al. (2017). Enhanced cancer immunotherapy by chimeric antigen
376–389.
receptor-modified T cells engineered to secrete checkpoint inhibitors. Clin.
Cancer Res. 23, 6982–6992.
Moon, E.K., Carpenito, C., Sun, J., Wang, L.C., Kapoor, V., Predina, J., Powell,
Li, Y., Hermanson, D.L., Moriarity, B.S., and Kaufman, D.S. (2018). Human D.J., Jr., Riley, J.L., June, C.H., and Albelda, S.M. (2011). Expression of a func-
iPSC-derived natural killer cells engineered with chimeric antigen receptors tional CCR2 receptor enhances tumor localization and tumor eradication by
enhance anti-tumor activity. Cell Stem Cell 23, 181–192. retargeted human T cells expressing a mesothelin-specific chimeric antibody
receptor. Clin. Cancer Res. 17, 4719–4730.
Ligtenberg, M.A., Mougiakakos, D., Mukhopadhyay, M., Witt, K., Lladser, A.,
Chmielewski, M., Riet, T., Abken, H., and Kiessling, R. (2016). Coexpressed Morgan, M.A., and Schambach, A. (2018). Engineering CAR-T cells for
catalase protects chimeric antigen receptor-redirected T cells as well as improved function against solid tumors. Front. Immunol. 9, 2493.
bystander cells from oxidative stress-induced loss of antitumor activity.
J. Immunol. 196, 759–766. Moritz, D., Wels, W., Mattern, J., and Groner, B. (1994). Cytotoxic T lympho-
cytes with a grafted recognition specificity for ERBB2-expressing tumor cells.
Liu, E., Marin, D., Banerjee, P., Macapinlac, H.A., Thompson, P., Basar, R., Proc. Natl. Acad. Sci. U S A 91, 4318–4322.
Nassif Kerbauy, L., Overman, B., Thall, P., Kaplan, M., et al. (2020). Use of
CAR-transduced natural killer cells in CD19-positive lymphoid tumors. Neelapu, S.S., Locke, F.L., Bartlett, N.L., Lekakis, L.J., Miklos, D.B., Jacob-
N. Engl. J. Med. 382, 545–553. son, C.A., Braunschweig, I., Oluwole, O.O., Siddiqi, T., Lin, Y., et al. (2017). Ax-
icabtagene ciloleucel CAR T-cell therapy in refractory large B-cell lymphoma.
Liu, X., Ranganathan, R., Jiang, S., Fang, C., Sun, J., Kim, S., Newick, K., Lo, N. Engl. J. Med. 377, 2531–2544.
A., June, C.H., Zhao, Y., and Moon, E.K. (2016). A chimeric switch-receptor
targeting PD1 augments the efficacy of second-generation CAR T cells in Newick, K., O’Brien, S., Moon, E., and Albelda, S.M. (2017). CAR T cell therapy
advanced solid tumors. Cancer Res. 76, 1578–1590. for solid tumors. Annu. Rev. Med. 68, 139–152.
Liu, X., Zhang, Y., Cheng, C., Cheng, A.W., Zhang, X., Li, N., Xia, C., Wei, X., Newick, K., O’Brien, S., Sun, J., Kapoor, V., Maceyko, S., Lo, A., Pure, E.,
Liu, X., and Wang, H. (2017). CRISPR-Cas9-mediated multiplex gene editing Moon, E., and Albelda, S.M. (2016). Augmentation of CAR T-cell trafficking
in CAR-T cells. Cell Res. 27, 154–157. and antitumor efficacy by blocking protein kinase A localization. Cancer Immu-
nol. Res. 4, 541–551.
Liu, Y., Di, S., Shi, B., Zhang, H., Wang, Y., Wu, X., Luo, H., Wang, H., Li, Z., and
Jiang, H. (2019). Armored inducible expression of IL-12 enhances antitumor O’Leary, M.C., Lu, X., Huang, Y., Lin, X., Mahmood, I., Przepiorka, D., Gavin,
activity of glypican-3-targeted chimeric antigen receptor-engineered T cells D., Lee, S., Liu, K., George, B., et al. (2019). FDA approval summary: tisagen-
in hepatocellular carcinoma. J. Immunol. 203, 198–207. lecleucel for treatment of patients with relapsed or refractory B-cell precursor
acute lymphoblastic leukemia. Clin. Cancer Res. 25, 1142–1146.
Lo, A.S., Taylor, J.R., Farzaneh, F., Kemeny, D.M., Dibb, N.J., and Maher, J.
(2008). Harnessing the tumour-derived cytokine, CSF-1, to co-stimulate T- Oelkrug, C., and Ramage, J.M. (2014). Enhancement of T cell recruitment and
cell growth and activation. Mol. Immunol. 45, 1276–1287. infiltration into tumours. Clin. Exp. Immunol. 178, 1–8.
Long, A.H., Haso, W.M., Shern, J.F., Wanhainen, K.M., Murgai, M., Ingaramo, Ohno, M., Ohkuri, T., Kosaka, A., Tanahashi, K., June, C.H., Natsume, A., and
M., Smith, J.P., Walker, A.J., Kohler, M.E., Venkateshwara, V.R., et al. (2015). Okada, H. (2013). Expression of miR-17-92 enhances anti-tumor activity of T-
4-1BB costimulation ameliorates T cell exhaustion induced by tonic signaling cells transduced with the anti-EGFRvIII chimeric antigen receptor in mice
of chimeric antigen receptors. Nat. Med. 21, 581–590. bearing human GBM xenografts. J. Immunother. Cancer 1, 21.
MacLeod, D.T., Antony, J., Martin, A.J., Moser, R.J., Hekele, A., Wetzel, K.J., Parihar, R., Rivas, C., Huynh, M., Omer, B., Lapteva, N., Metelitsa, L.S., Gott-
Brown, A.E., Triggiano, M.A., Hux, J.A., Pham, C.D., et al. (2017). Integration of schalk, S.M., and Rooney, C.M. (2019). NK cells expressing a chimeric acti-
a CD19 CAR into the TCR alpha chain locus streamlines production of alloge- vating receptor eliminate MDSCs and rescue impaired CAR-T cell activity
neic gene-edited CAR T cells. Mol. Ther. 25, 949–961. against solid tumors. Cancer Immunol. Res. 7, 363–375.
Pickup, M., Novitskiy, S., and Moses, H.L. (2013). The roles of TGFbeta in the Rupp, L.J., Schumann, K., Roybal, K.T., Gate, R.E., Ye, C.J., Lim, W.A., and
tumour microenvironment. Nat. Rev. Cancer 13, 788–799. Marson, A. (2017). CRISPR/Cas9-mediated PD-1 disruption enhances anti-tu-
mor efficacy of human chimeric antigen receptor T cells. Sci. Rep. 7, 737.
Qasim, W., Zhan, H., Samarasinghe, S., Adams, S., Amrolia, P., Stafford, S.,
Butler, K., Rivat, C., Wright, G., Somana, K., et al. (2017). Molecular remission Sadelain, M., Riviere, I., and Riddell, S. (2017). Therapeutic T cell engineering.
of infant B-ALL after infusion of universal TALEN gene-edited CAR T cells. Sci. Nature 545, 423–431.
Transl. Med. 9, eaaj2013.
Sakemura, R., Terakura, S., Watanabe, K., Julamanee, J., Takagi, E., Miyao,
Qi, J., Tsuji, K., Hymel, D., Burke, T.R., Hudecek, M., Rader, C., and Peng, H. K., Koyama, D., Goto, T., Hanajiri, R., Nishida, T., et al. (2016). A Tet-On induc-
(2020). Chemically programmable and switchable CAR-T therapy. Angew. ible system for controlling CD19-chimeric antigen receptor expression upon
Chem. Int. Ed. 59, 12178–12185. drug administration. Cancer Immunol. Res. 4, 658–668.
Qin, H., Ramakrishna, S., Nguyen, S., Fountaine, T.J., Ponduri, A., Stetler-Ste-
Sasaki, K., Kohanbash, G., Hoji, A., Ueda, R., McDonald, H.A., Reinhart, T.A.,
venson, M., Yuan, C.M., Haso, W., Shern, J.F., Shah, N.N., and Fry, T.J. (2018).
Martinson, J., Lotze, M.T., Marincola, F.M., Wang, E., et al. (2010). miR-17-92
Preclinical development of bivalent chimeric antigen receptors targeting both
expression in differentiated T cells - implications for cancer immunotherapy.
CD19 and CD22. Mol. Ther. Oncolytics 11, 127–137.
J. Transl Med. 8, 17.
Quintarelli, C., Sivori, S., Caruso, S., Carlomagno, S., Falco, M., Boffa, I., Or-
Scharping, N.E., Menk, A.V., Moreci, R.S., Whetstone, R.D., Dadey, R.E., Wat-
lando, D., Guercio, M., Abbaszadeh, Z., Sinibaldi, M., et al. (2020). Efficacy
kins, S.C., Ferris, R.L., and Delgoffe, G.M. (2016). The tumor microenvironment
of third-party chimeric antigen receptor modified peripheral blood natural killer
represses T cell mitochondrial biogenesis to drive intratumoral T cell metabolic
cells for adoptive cell therapy of B-cell precursor acute lymphoblastic leuke-
insufficiency and dysfunction. Immunity 45, 374–388.
mia. Leukemia 34, 1102–1115.
Rabinovich, G.A., Gabrilovich, D., and Sotomayor, E.M. (2007). Immunosup- €cker, J.C. (2002). Induction of T cell develop-
Schmitt, T.M., and Zúñiga-Pflu
pressive strategies that are mediated by tumor cells. Annu. Rev. Immunol. ment from hematopoietic progenitor cells by delta-like-1 in vitro. Immunity
25, 267–296. 17, 749–756.
Rafiq, S., Hackett, C.S., and Brentjens, R.J. (2020). Engineering strategies to Schuster, S.J., Svoboda, J., Chong, E.A., Nasta, S.D., Mato, A.R., Anak, O.,
overcome the current roadblocks in CAR T cell therapy. Nat. Rev. Clin. Oncol. Brogdon, J.L., Pruteanu-Malinici, I., Bhoj, V., Landsburg, D., et al. (2017).
17, 147–167. Chimeric antigen receptor T cells in refractory B-cell lymphomas. N. Engl. J.
Med. 377, 2545–2554.
Rafiq, S., Yeku, O.O., Jackson, H.J., Purdon, T.J., van Leeuwen, D.G., Drakes,
D.J., Song, M., Miele, M.M., Li, Z., et al. (2018). Targeted delivery of a PD-1- Serafini, M., Manganini, M., Borleri, G., Bonamino, M., Imberti, L., Biondi, A.,
blocking scFv by CAR-T cells enhances anti-tumor efficacy in vivo. Nat. Bio- Golay, J., Rambaldi, A., and Introna, M. (2004). Characterization of CD20-
technol. 36, 847–856. transduced T lymphocytes as an alternative suicide gene therapy approach
for the treatment of graft-versus-host disease. Hum. Gene Ther. 15, 63–76.
Raj, D., Yang, M.H., Rodgers, D., Hampton, E.N., Begum, J., Mustafa, A., Lor-
izio, D., Garces, I., Propper, D., Kench, J.G., et al. (2019). Switchable CAR-T Shah, N.N., Highfill, S.L., Shalabi, H., Yates, B., Jin, J., Wolters, P.L., Ombrello,
cells mediate remission in metastatic pancreatic ductal adenocarcinoma. A., Steinberg, S.M., Martin, S., Delbrook, C., et al. (2020). CD4/CD8 T-cell selec-
Gut 68, 1052–1064. tion affects chimeric antigen receptor (CAR) T-cell potency and toxicity: updated
results from a phase I anti-CD22 CAR T-cell trial. J. Clin. Oncol. 38, 1938–1950.
Reighard, S.D., Cranert, S.A., Rangel, K.M., Ali, A., Gyurova, I.E., de la Cruz-
Lynch, A.T., Tuazon, J.A., Khodoun, M.V., Kottyan, L.C., Smith, D.F., et al. Shalabi, H., Kraft, I.L., Wang, H.W., Yuan, C.M., Yates, B., Delbrook, C., Zimbel-
(2020). Therapeutic targeting of follicular T Cells with chimeric antigen recep- man, J.D., Giller, R., Stetler-Stevenson, M., Jaffe, E.S., et al. (2018). Sequential
tor-expressing natural killer cells. Cell Rep. Med. 1, 100003. loss of tumor surface antigens following chimeric antigen receptor T-cell thera-
pies in diffuse large B-cell lymphoma. Haematologica 103, e215–e218.
Ren, J., Liu, X., Fang, C., Jiang, S., June, C.H., and Zhao, Y. (2017a). Multiplex
genome editing to generate universal CAR T cells resistant to PD1 Inhibition. Shin, J.H., Park, H.B., Oh, Y.M., Lim, D.P., Lee, J.E., Seo, H.H., Lee, S.J., Eom,
Clin. Cancer Res. 23, 2255–2266. H.S., Kim, I.H., Lee, S.H., and Choi, K. (2012). Positive conversion of negative
signaling of CTLA4 potentiates antitumor efficacy of adoptive T-cell therapy in
Ren, J., Zhang, X., Liu, X., Fang, C., Jiang, S., June, C.H., and Zhao, Y. (2017b). murine tumor models. Blood 119, 5678–5687.
A versatile system for rapid multiplex genome-edited CAR T cell generation.
Oncotarget 8, 17002–17011. Shum, T., Omer, B., Tashiro, H., Kruse, R.L., Wagner, D.L., Parikh, K., Yi, Z.,
Sauer, T., Liu, D., Parihar, R., et al. (2017). Constitutive signaling from an engi-
Ricciotti, E., and FitzGerald, G.A. (2011). Prostaglandins and inflammation. Ar-
neered IL7 receptor promotes durable tumor elimination by tumor-redirected
terioscler Thromb. Vasc. Biol. 31, 986–1000.
T cells. Cancer Discov. 7, 1238–1247.
Ridge, J.P., Di Rosa, F., and Matzinger, P. (1998). A conditioned dendritic cell
Siska, P.J., van der Windt, G.J., Kishton, R.J., Cohen, S., Eisner, W., MacIver,
can be a temporal bridge between a CD4+ T-helper and a T-killer cell. Nature
N.J., Kater, A.P., Weinberg, J.B., and Rathmell, J.C. (2016). Suppression of
393, 474–478.
Glut1 and glucose metabolism by decreased Akt/mTORC1 signaling drives
Riese, M.J., Wang, L.C., Moon, E.K., Joshi, R.P., Ranganathan, A., June, C.H., T cell impairment in B cell leukemia. J. Immunol. 197, 2532–2540.
Koretzky, G.A., and Albelda, S.M. (2013). Enhanced effector responses in acti-
vated CD8+ T cells deficient in diacylglycerol kinases. Cancer Res. 73, Sommermeyer, D., Hudecek, M., Kosasih, P.L., Gogishvili, T., Maloney, D.G.,
3566–3577. Turtle, C.J., and Riddell, S.R. (2016). Chimeric antigen receptor-modified
T cells derived from defined CD8+ and CD4+ subsets confer superior antitumor
Rodgers, D.T., Mazagova, M., Hampton, E.N., Cao, Y., Ramadoss, N.S., reactivity in vivo. Leukemia 30, 492–500.
Hardy, I.R., Schulman, A., Du, J., Wang, F., Singer, O., et al. (2016). Switch-
mediated activation and retargeting of CAR-T cells for B-cell malignancies. Srivastava, S., Salter, A.I., Liggitt, D., Yechan-Gunja, S., Sarvothama, M.,
Proc. Natl. Acad. Sci. U S A 113, E459–E468. Cooper, K., Smythe, K.S., Dudakov, J.A., Pierce, R.H., Rader, C., and Riddell,
S.R. (2019). Logic-gated ROR1 chimeric antigen receptor expression rescues
Roth, T.L., Li, P.J., Blaeschke, F., Nies, J.F., Apathy, R., Mowery, C., Yu, R., T cell-mediated toxicity to normal tissues and enables selective tumor target-
Nguyen, M.L.T., Lee, Y., Truong, A., et al. (2020). Pooled knockin targeting ing. Cancer Cell 35, 489–503.
for genome engineering of cellular immunotherapies. Cell 181, 1–17.
Stadtmauer, E.A., Fraietta, J.A., Davis, M.M., Cohen, A.D., Weber, K.L., Lan-
Roybal, K.T., Rupp, L.J., Morsut, L., Walker, W.J., McNally, K.A., Park, J.S., caster, E., Mangan, P.A., Kulikovskaya, I., Gupta, M., Chen, F., et al. (2020).
and Lim, W.A. (2016). Precision tumor recognition by T cells with combinatorial CRISPR-engineered T cells in patients with refractory cancer. Science 367,
antigen-sensing circuits. Cell 164, 770–779. eaba7365.
Sterner, R.M., Sakemura, R., Cox, M.J., Yang, N., Khadka, R.H., Forsman, Wei, J., Long, L., Zheng, W., Dhungana, Y., Lim, S.A., Guy, C., Wang, Y.,
C.L., Hansen, M.J., Jin, F., Ayasoufi, K., Hefazi, M., et al. (2019). GM-CSF in- Wang, Y.D., Qian, C., Xu, B., et al. (2019). Targeting REGNASE-1 programs
hibition reduces cytokine release syndrome and neuroinflammation but en- long-lived effector T cells for cancer therapy. Nature 576, 471–476.
hances CAR-T cell function in xenografts. Blood 133, 697–709.
Wei, S.C., Duffy, C.R., and Allison, J.P. (2018). Fundamental mechanisms of
Straathof, K.C., Pule, M.A., Yotnda, P., Dotti, G., Vanin, E.F., Brenner, M.K., immune checkpoint blockade therapy. Cancer Discov. 8, 1069–1086.
Heslop, H.E., Spencer, D.M., and Rooney, C.M. (2005). An inducible caspase
9 safety switch for T-cell therapy. Blood 105, 4247–4254. Weinberg, F., Ramnath, N., and Nagrath, D. (2019). Reactive oxygen species in
the tumor microenvironment: an overview. Cancers (Basel) 11, 1191.
Suarez, E.R., Chang, D.K., Sun, J., Sui, J., Freeman, G.J., Signoretti, S., Zhu,
Q., and Marasco, W.A. (2016). Chimeric antigen receptor T cells secreting anti- Whiteside, T.L. (2008). The tumor microenvironment and its role in promoting
PD-L1 antibodies more effectively regress renal cell carcinoma in a humanized tumor growth. Oncogene 27, 5904–5912.
mouse model. Oncotarget 7, 34341–34355.
Wiesinger, M., Marz, J., Kummer, M., Schuler, G., Dorrie, J., Schuler-Thurner,
Tan, A.H.J., Vinanica, N., and Campana, D. (2020). Chimeric antigen receptor- B., and Schaft, N. (2019). Clinical-scale production of CAR-T cells for the treat-
T cells with cytokine neutralizing capacity. Blood Adv. 4, 1419–1431. ment of melanoma patients by mRNA transfection of a CSPG4-specific CAR
under full GMP compliance. Cancers (Basel) 11, 1198.
Tchou, J., Zhao, Y., Levine, B.L., Zhang, P.J., Davis, M.M., Melenhorst, J.J.,
Kulikovskaya, I., Brennan, A.L., Liu, X., Lacey, S.F., et al. (2017). Safety and ef- Wilkie, S., Burbridge, S.E., Chiapero-Stanke, L., Pereira, A.C., Cleary, S., van
ficacy of intratumoral injections of chimeric antigen receptor (CAR) T cells in der Stegen, S.J., Spicer, J.F., Davies, D.M., and Maher, J. (2010). Selective
metastatic breast cancer. Cancer Immunol. Res. 5, 1152–1161. expansion of chimeric antigen receptor-targeted T-cells with potent effector
function using interleukin-4. J. Biol. Chem. 285, 25538–25544.
Torikai, H., Reik, A., Liu, P.Q., Zhou, Y., Zhang, L., Maiti, S., Huls, H., Miller,
Wilkie, S., van Schalkwyk, M.C., Hobbs, S., Davies, D.M., van der Stegen, S.J.,
J.C., Kebriaei, P., Rabinovich, B., et al. (2012). A foundation for universal T-
Pereira, A.C., Burbridge, S.E., Box, C., Eccles, S.A., and Maher, J. (2012). Dual
cell based immunotherapy: T cells engineered to express a CD19-specific
targeting of ErbB2 and MUC1 in breast cancer using chimeric antigen recep-
chimeric-antigen-receptor and eliminate expression of endogenous TCR.
tors engineered to provide complementary signaling. J. Clin. Immunol. 32,
Blood 119, 5697–5705.
1059–1070.
Torikai, H., Reik, A., Soldner, F., Warren, E.H., Yuen, C., Zhou, Y., Crossland,
Woo, S.R., Turnis, M.E., Goldberg, M.V., Bankoti, J., Selby, M., Nirschl, C.J.,
D.L., Huls, H., Littman, N., Zhang, Z., et al. (2013). Toward eliminating HLA
Bettini, M.L., Gravano, D.M., Vogel, P., Liu, C.L., et al. (2012). Immune inhibi-
class I expression to generate universal cells from allogeneic donors. Blood
tory molecules LAG-3 and PD-1 synergistically regulate T-cell function to pro-
122, 1341–1349.
mote tumoral immune escape. Cancer Res. 72, 917–927.
Tormoen, G.W., Crittenden, M.R., and Gough, M.J. (2018). Role of the immu- Xu, X., Huang, W., Heczey, A., Liu, D., Guo, L., Wood, M., Jin, J., Courtney,
nosuppressive microenvironment in immunotherapy. Adv. Radiat. Oncol. 3, A.N., Liu, B., Di Pierro, E.J., et al. (2019). NKT cells coexpressing a GD2-spe-
520–526. cific chimeric antigen receptor and IL15 show enhanced in vivo persistence
and antitumor activity against neuroblastoma. Clin. Cancer Res. 25,
Tran, E., Chinnasamy, D., Yu, Z., Morgan, R.A., Lee, C.C., Restifo, N.P., and 7126–7138.
Rosenberg, S.A. (2013). Immune targeting of fibroblast activation protein trig-
gers recognition of multipotent bone marrow stromal cells and cachexia. Yang, W., Bai, Y., Xiong, Y., Zhang, J., Chen, S., Zheng, X., Meng, X., Li, L.,
J. Exp. Med. 210, 1125–1135. Wang, J., Xu, C., et al. (2016). Potentiating the antitumour response of CD8+
T cells by modulating cholesterol metabolism. Nature 531, 651–655.
Urbanska, K., Lanitis, E., Poussin, M., Lynn, R.C., Gavin, B.P., Kelderman, S.,
Yu, J., Scholler, N., and Powell, D.J., Jr. (2012). A universal strategy for adop- Yeku, O.O., Purdon, T.J., Koneru, M., Spriggs, D., and Brentjens, R.J. (2017).
tive immunotherapy of cancer through use of a novel T-cell antigen receptor. Armored CAR T cells enhance antitumor efficacy and overcome the tumor
Cancer Res. 72, 1844–1852. microenvironment. Sci. Rep. 7, 10541.
Valton, J., Guyot, V., Marechal, A., Filhol, J.M., Juillerat, A., Duclert, A., Duch- Zah, E., Lin, M.Y., Silva-Benedict, A., Jensen, M.C., and Chen, Y.Y. (2016). T
ateau, P., and Poirot, L. (2015). A multidrug-resistant engineered CAR T cell for cells expressing CD19/CD20 bispecific chimeric antigen receptors prevent an-
allogeneic combination immunotherapy. Mol. Ther. 23, 1507–1518. tigen escape by malignant B cells. Cancer Immunol. Res. 4, 498–508.
Vodnala, S.K., Eil, R., Kishton, R.J., Sukumar, M., Yamamoto, T.N., Ha, N.H., Zah, E., Nam, E., Bhuvan, V., Tran, U., Ji, B.Y., Gosliner, S.B., Wang, X.,
Lee, P.H., Shin, M., Patel, S.J., Yu, Z., et al. (2019). T cell stemness and Brown, C.E., and Chen, Y.Y. (2020). Systematically optimized BCMA/CS1 bis-
dysfunction in tumors are triggered by a common mechanism. Science 363, pecific CAR-T cells robustly control heterogeneous multiple myeloma. Nat.
eaau0135. Commun. 11, 2283.
Wang, D., and Dubois, R.N. (2006). Prostaglandins and cancer. Gut 55, Zhang, L., Kerkar, S.P., Yu, Z., Zheng, Z., Yang, S., Restifo, N.P., Rosenberg,
115–122. S.A., and Morgan, R.A. (2011). Improving adoptive T cell therapy by targeting
and controlling IL-12 expression to the tumor environment. Mol. Ther. 19,
Wang, L.C., Lo, A., Scholler, J., Sun, J., Majumdar, R.S., Kapoor, V., Antzis, M., 751–759.
Cotner, C.E., Johnson, L.A., Durham, A.C., et al. (2014). Targeting fibroblast
activation protein in tumor stroma with chimeric antigen receptor T cells can Zhang, Y., Zhang, X., Cheng, C., Mu, W., Liu, X., Li, N., Wei, X., Liu, X., Xia, C.,
inhibit tumor growth and augment host immunity without severe toxicity. Can- and Wang, H. (2017). CRISPR-Cas9 mediated LAG-3 disruption in CAR-T
cer Immunol. Res. 2, 154–166. cells. Front. Med. 11, 554–562.
Wang, X., Chang, W.C., Wong, C.W., Colcher, D., Sherman, M., Ostberg, J.R., Zhao, Z., Condomines, M., van der Stegen, S.J.C., Perna, F., Kloss, C.C., Gun-
Forman, S.J., Riddell, S.R., and Jensen, M.C. (2011). A transgene-encoded set, G., Plotkin, J., and Sadelain, M. (2015). Structural design of engineered
cell surface polypeptide for selection, in vivo tracking, and ablation of engi- costimulation determines tumor rejection kinetics and persistence of CAR
neered cells. Blood 118, 1255–1263. T cells. Cancer Cell 28, 415–428.
Article
Pediatric low-grade gliomas (pLGG) are frequently driven by genetic alterations in the RAS-mitogen-acti-
vated protein kinase (RAS/MAPK) pathway yet show unexplained variability in their clinical outcome. To
address this, we characterized a cohort of >1,000 clinically annotated pLGG. Eighty-four percent of cases
harbored a driver alteration, while those without an identified alteration also often exhibited upregulation
of the RAS/MAPK pathway. pLGG could be broadly classified based on their alteration type. Rearrange-
ment-driven tumors were diagnosed at a younger age, enriched for WHO grade I histology, infrequently pro-
gressed, and rarely resulted in death as compared with SNV-driven tumors. Further sub-classification of clin-
ical-molecular correlates stratified pLGG into risk categories. These data highlight the biological and clinical
differences between pLGG subtypes and opens avenues for future treatment refinement.
Significance
Pediatric low-grade gliomas (pLGG) are the most common brain tumors affecting children. This integrated clinicopathologic
and genomic analysis of >1,000 pLGG defines molecular subgroups with distinct biological drivers and clinical features.
RAS/MAPK pathway is activated near universally in pLGG, regardless of the presence of a clear genomic activator. Further,
although many alterations converge on the RAS/MAPK pathway, clinical presentation and outcome are highly variable de-
pending on the type of underlying alteration. This information helped define clinical risk groups of pLGG with different pro-
gression-free and overall survival which likely require different treatment strategies. As modernized treatment regimens uti-
lize alteration-specific agents, we provide the framework for molecular classification of pLGG reflecting unique biological
mechanisms driving the disease that likely promote different therapeutic susceptibilities.
Cancer Cell 37, 569–583, April 13, 2020 ª 2020 Elsevier Inc. 569
Peter Dirks,1,2,20 Michael D. Taylor,1,2,20 James T. Rutka,1,2,20 Suzanne Laughlin,21,22 Manohar Shroff,21,22 Mary Shago,2,6
Lili-Naz Hazrati,2,23 Colleen D’Arcy,2,23,24 Vijay Ramaswamy,1,3,25 Ute Bartels,3,25 Annie Huang,1,2,3 Eric Bouffet,3,25
Matthias A. Karajannis,15 Mariarita Santi,10,11 David W. Ellison,8 Uri Tabori,1,3,26,28 and Cynthia Hawkins1,2,23,29,*
16Department of Pathology, Memorial Sloan Kettering Cancer Center, New York, NY, USA
17Radiation Medicine Program, Princess Margaret Cancer Centre, University Health Network, Toronto, ON, Canada
18Department of Radiation Oncology, Faculty of Medicine, University of Toronto, Toronto, ON, Canada
19Department of Surgery, University of Toronto, Toronto, ON, Canada
20Department of Neurosurgery, The Hospital for Sick Children, Toronto ON, Canada
21Department of Radiology, The Hospital for Sick Children, Toronto ON, Canada
22Department of Medical Imaging, University of Toronto, Toronto, ON, Canada
23Department of Pathology, The Hospital for Sick Children, Toronto, ON, Canada
24Department of Anatomical Pathology, The Alfred Hospital, Prahran, VIC, Australia
25Department of Paediatrics, University of Toronto, Toronto, ON, Canada
26Department of Medical Biophysics, University of Toronto, Toronto, ON, Canada
27These authors contributed equally
28Co-senior authors
29Lead Contact
*Correspondence: cynthia.hawkins@sickkids.ca
https://doi.org/10.1016/j.ccell.2020.03.011
which primarily arise in the cerebral hemispheres and inevitably annotated resource which includes representation of the rarest
transform to higher-grade glioma, pLGG can occur throughout pLGG molecular entities and their clinical features.
the CNS and rarely transform (Broniscer et al., 2007; Mistry
et al., 2015). Nevertheless, outcome and response to therapy is RESULTS
highly variable. If complete surgical resection is successful,
10-year progression-free survival (PFS) exceeds 85%, but drops Patient Cohort
below 50% if there is radiologically visible residual tumor (Wisoff Our population-based cohort of pLGG consisted of 976 patients
et al., 2011). In deep-seated midline locations, gross total resec- (<19 years) followed and treated at the Hospital for Sick Children
tion (GTR) is not often achievable and adjuvant therapy may be (Toronto, ON, Canada) from 1986 to 2017 (Table S1). For each
required, often with unsatisfactory tumor control and/or long- patient we collected demographic, treatment and outcome data.
term morbidity (Krishnatry et al., 2016; Nageswara and Packer, At the population level, tumors were distributed equally between
2014). Which patients require these therapies and who will benefit the diencephalon (n = 313, 32%; n = 124, 13% were from neurofi-
from them is not yet well understood. bromatosis type 1 [NF1] patients), cerebral hemispheres (n = 265,
Over the last decade, molecular profiling studies have incre- 27%), and the cerebellum (n = 252, 26%), whereas pure brainstem
mentally identified key genetic events in pLGG that converge on (n = 92, 9.4%), spinal cord (n = 41, 4.2%), and extensively dissem-
the RAS-mitogen-activated protein kinase (RAS/MAPK) pathway. inated tumors (n = 13, 1.3%) were less frequent (Figure 1A). Upon
Most commonly, these are somatic events involving BRAF or pathologic review of non-NF1 cases (n = 843), pilocytic astrocy-
germline NF1 alterations (Collins et al., 2015; Jones et al., 2008; toma (n = 303, 36%) was the most common diagnosis (excluding
Schindler et al., 2011; Dougherty et al., 2010; Lassaletta et al., LGG, not otherwise specified [NOS]) across tumor locations (Fig-
2017; Listernick et al., 1999; Uusitalo et al., 2016; Seminog and ure 1B) with the exception of the cerebral hemispheres, which
Goldacre, 2013). In addition to these common pLGG alterations, was histologically diverse (Figure 1C). The median age of diagnosis
rarer alterations affecting RAS/MAPK signaling, including those was 7.6 years (range 0–18.7 years), with pLGG in the cerebral
involving FGFR1/2/3, NTRK2, RAF1, ALK, and ROS1 (Zhang hemispheres being diagnosed at a later age (median = 10.7 years)
et al., 2013; Jones et al., 2013; Qaddoumi et al., 2016; Guerreiro as compared with other tumor locations (p < 0.0001, ANOVA; all
Stucklin et al., 2019), as well as non-RAS/MAPK alterations, pairwise comparisons adjusted p < 0.0001, t test) (Figure 1D).
such as MYB, MYBL1, IDH1, and H3F3A (Qaddoumi et al., There was a significant association between tumor location,
2016; Tatevossian et al., 2010; Ramkissoon et al., 2013; Bando- PFS, and overall survival (OS) (p < 0.0001, log rank test) (Figures
padhayay et al., 2016; Ryall et al., 2016; Hartmann et al., 2009) 1E and S1), with 10-year PFS and OS best for patients with tumors
have been identified in small numbers of cases. However, several in the cerebellum (89% and 99%, respectively), and worst for those
key issues remain undefined: (1) Are all NF1, BRAF fused and with extensively disseminated disease (0% and 67%, respec-
BRAF mutant tumors the same? (2) What is the mechanism of tively). Importantly, only 7.5% of patients succumbed to their dis-
tumorigenesis in pLGG without an identifiable genetic alteration? ease (median time to death = 3.9 years, median OS follow-up =
(3) What are the clinical features of the rare alterations in pLGG 15.9 years) despite 33% experiencing tumor progression (median
and is their outcome unique? (4) Can molecular alterations help time to progression = 2.3 years, median PFS follow-up = 5.9 years)
provide biological insights for disease stratification? (Figure 1F).
To answer these questions and provide a population-based
resource for the pediatric neuro-oncology community, we Characteristics of NF1-Driven pLGG
molecularly characterized >1,000 pLGG with comprehensive Patients with the genetic pre-disposition disorder NF1 are diag-
clinical data. This enabled us to provide a statistically robust, nosed using a series of clinical observations and tests indicative
pLGG were significantly more likely to succumb to their dis- p = 0.0058, log rank test). When investigating BRAF in grade
ease (n = 24, 13%, p < 0.0001, Fisher’s exact test versus rear- I tumors alone, the pattern remained (5-year PFS of 72% and
rangement-driven) and/or progress (n = 80, 44%, p < 0.0001, 56% for KIAA1549-BRAF and BRAF p.V600E, respectively,
Fisher’s exact test versus fusion-driven) (Table 1; Figures 4B– p = 0.0176). Although the numbers are too small to allow statis-
4D). Furthermore, rearrangement-driven pLGG were diagnosed tical comparisons between SNVs and fusions for other genes,
at a significantly younger age (median 6.6 versus 10.9 years, the same trend is evident. For example, for FGFR1, patients
p < 0.0001, t test) and were enriched for WHO grade I histology with FGFR1-TACC1 or FGFR1 TKD tumors had similar
(p < 0.0001, Fisher’s exact test) (Figure 4B; Table 1). This outcome to those with KIAA1549-BRAF fusions (5-year PFS
pattern is evident in BRAF, where tumors with KIAA1549- of 69% for all). Patients with FGFR1 SNVs, on the other
BRAF have superior outcome to BRAF p.V600E (5-year PFS hand, were more similar to BRAF p.V600E (5-year PFS of
of 69% for KIAA1549-BRAF versus 52% for BRAF p.V600E, 53% and 52% for FGFR1 SNV and BRAF p.V600E,
respectively). Importantly, where FGFR1-TACC1 and FGFR1- riched in pilocytic astrocytoma (n = 73/83, 88%, p < 0.0001, Fish-
TKD tumors did not contain additional alterations, FGFR1 er’s exact test) and in cerebellar tumors (n = 60/83, 72%, p <
SNVs often did, sometimes resulting in late deaths. 0.0001, Fisher’s exact test) (Figures S4B and S4C). Interestingly,
the only KIAA1549-BRAF fusion seen in hemispheric tumors was
Characteristics of Fusion-Driven pLGG 15:09 (Figure S4B). 15:09 was also the primary fusion seen in tu-
BRAF Fusions mors with extensive dissemination (n = 5, 83%, p < 0.0001, Fish-
KIAA1549-BRAF was the most frequent alteration in pLGG er’s exact test). 15:09 fusions were associated with a worse PFS
(35%) and was almost exclusively a single-event driver (p = 0.0003, log rank test, Figure S4D), with a 5-year PFS of 59%
(n = 175/180, 97%), with four cases also having a CDKN2A dele- as compared with 77%–100% for other fusion subtypes. BRAF
tion and one present in a patient with NF1 (Figure 2A). These five fusions not involving KIAA1549 occurred exclusively in adoles-
rare cases are still alive (median follow-up = 7.5 years). cents with no progression events (median follow-up = 3.7 years),
KIAA1549-BRAF was significantly enriched in pilocytic astrocy- while 15:11 was only observed in two infants who rapidly expe-
toma (n = 150/180, 83%, p < 0.0001, Fisher’s exact test), and in rienced tumor progression and died (Figures S4D and S4E).
cerebellar tumors (n = 100/180, 56%, p = 0.0002, Fisher’s exact Identifying the specific fusion breakpoints of KIAA1549-BRAF
test) (Figure 5A). will be important to properly ascertain their propensity for certain
Because of the large number of tumors, we could sub-stratify clinical features and impact on outcome.
the BRAF fusions into subgroups based on their breakpoints FGFR1/2 Fusions and FGFR1 TKD
(Figure S4A). The most common KIAA1549-BRAF fusion FGFR fusions/TKD were observed in 6.1% of the cohort (Fig-
involved exon 16 in KIAA1549 and exon 9 in BRAF (16:09). ure 2B). FGFR1-TACC1 pLGG were often cystic lesions, most
Like all KIAA1549-BRAF fusions, 16:09 was significantly en- commonly pilocytic astrocytoma (n = 7/14, 50%) and occurred
of 55%) BRAF p.V600E tumors had inferior PFS to KIAA1549- 19%). Of those not lost to follow-up, 43% (n = 12) progressed
BRAF (5-year PFS of 69%) fused tumors (p = 0.0139 and rapidly (median of 2.2 years, 5-year PFS of 53%). Despite this,
p = 0.0356, respectively, log rank test, Figures S7A and S7B), only two cases had late deaths after 13.7 and 15.5 years, respec-
despite their OS not being significantly different (p = 0.1169 tively, both of whom had additional alterations.
and 0.1888, respectively, log rank test, Figures S7C and S7D). IDH1 p.R132H
FGFR1 Mutations IDH1 mutations are common in adult lower-grade gliomas,
FGFR1 point mutations were observed in 1.5% of pLGG and pri- arising in approximately 70% of grade II and III tumors (Parsons
marily consisted of p.N546K and p.K656E. Histologically, these et al., 2008; Yan et al., 2009, Balss et al., 2008). In pLGG, IDH1
tumors were most frequently dysembryoplastic neuroepithelial tu- p.R132H mutations were extremely rare, accounting for only
mors (n = 13, 41%) or pilocytic astrocytoma (n = 9, 28%), diag- 0.8% of cases (Figure 2B). Most IDH1 p.R132H patients pre-
nosed in older children (p = 0.032, Fisher’s exact test), and often sented with a prolonged history of seizures, sometimes years
(n = 16, 50%) co-occurred with multiple genetic alterations, before the biopsy was performed. All tumors were restricted to
including NF1 (n = 7, 22%) or additional RAS/MAPK pathway mu- the cerebral hemispheres and were either oligodendroglioma
tations (n = 11, 34%) (Figures 2A and 6B). Interestingly, in some or diffuse astrocytoma (Figure 6C). Patients harboring IDH1
cases, multiple point mutations in FGFR1 were seen (n = 6, p.R132H were diagnosed in late childhood (median: 15.7 years),
with the youngest patient diagnosed at 8.9 years (Figure 6C). SNVs, often co-occurred with other alterations (25%), most often
Fifty percent of IDH1 p.R132H pLGG progressed within a median with BRAF p.V600E (Figures 2A and 6D). Although morphologically
of 5.1 years (5-year PFS of 56%) despite only one succumbing to and clinically different than midline HGG, H3.3 p.K27M patients
their disease (Figure 6C). progressed early (median time to progression = 0.8 years), with
H3.3 p.K27M all patients eventually succumbing to their disease (Figure 6D).
H3F3A mutations are common in childhood high-grade gliomas These data support the role of H3.3 p.K27M as a marker of aggres-
and DIPG and confer a dismal outcome (Khuong-Quang et al., sive behavior regardless of the initial morphology and
2012; Buczkowicz et al., 2014). In our pLGG series, all H3F3A mu- presentation.
tations were p.K27M, with no p.G34R/V alterations identified
(Table S1). These cases were restricted to the midline (dienceph- Molecular-Based Risk Stratification for pLGG
alon [n = 8] and brainstem [n = 4]), enriched for diffuse astrocy- Based on the above data, we define a risk stratification for children
tomas (n = 8, 67%, p = 0.0011, Fisher’s exact test), and like other with pLGG (Figure 7A). pLGG harboring gene fusions or germline
nosed at a later age, consistent with these alterations being ac- d METHOD DETAILS
quired later in development. Furthermore, SNV-driven pLGG co- B Nucleic Acid Extraction
occurred with additional SNVs, but never co-harbored fusion B Droplet Digital PCR
events; displaying a pattern of mutual exclusivity as seen across B NanoString nCounter Analysis
many cancer types (Gao et al., 2018). Finally, SNV-driven pLGG B NanoString nCounter Vantage 3D for Protein Analysis
were associated with poorer outcome. Future work will need to B Fluorescent In Situ Hybridization
pLGG to elucidate why the observed clinical differences occur. B SNP Array
system for pLGG (Figure 7A). Genetic rearrangements, including B Whole Transcriptome Sequencing
all fusions and duplications, as well as germline NF1 inactivation, B Gene Set Enrichment Analysis
are considered low risk. As these tumors are rarely fatal, we pro- B Genetic Analysis at Collaborating Institutions
pose they should be managed expectantly, with careful consid- d QUANTIFICATION AND STATISTICAL ANALYSIS
eration of additional treatment after surgery, and radiation d DATA AND CODE AVAILABILITY
excluded from all post-operative treatment. For example, for
SUPPLEMENTAL INFORMATION
asymptomatic NF1-driven pLGG, surveillance is justified. How-
ever, if patients display progressive symptoms, most often as Supplemental Information can be found online at https://doi.org/10.1016/j.
vision loss, treatment with chemotherapy (Mahoney et al., ccell.2020.03.011.
2000; Packer et al., 1997) or targeted inhibitors (Banerjee et al.,
2017; Fangusaro et al., 2019) can be beneficial. Beyond 10 years ACKNOWLEDGMENTS
these tumors are much less likely to recur and the frequency of
follow-up may potentially be reduced. Intermediate-risk pLGG This work was supported by endowed funds from the Government of Canada
through Genome Canada and the Ontario Genomics Institute (OGI-121); A
are SNV-driven tumors, including those with BRAF or FGFR1
Kid’s Brain Tumor Cure; Brain Tumor Research Assistance and Information
SNVs. These frequently harbor more than one alteration and Network; The Pediatric Low-Grade Astrocytoma Foundation; Meagan’s
have a higher risk of recurrence which extends beyond 10 years. Walk; B.r.a.i.n.child Canada; Canadian Cancer Society (grant no. 702296); Ca-
These patients may require multiple treatments and longer nadian Institutes of Health Research (grant no. 159805); Restracomp
S.R., M.Z., U.T., and C.H. conceived the project and led the study. Molecular D’Angelo, F., Ceccarelli, M., Tala, Garofano, L., Zhang, J., Frattini, V., Caruso,
characterization of samples from HSC were completed by S.R., M.Z., K.F., F.P., Lewis, G., Alfaro, K.D., Bauchet, L., et al. (2019). The molecular landscape
L.N., A.G.-S., J.B., A.L., M.J., and J.S. under the supervision of A.A., P.E.K., of glioma in patients with neurofibromatosis 1. Nat. Med. 25, 176–187.
S.J., M.S., U.T., and C.H. Molecular characterization of samples from partner- Dobin, A., Davis, C.A., Schlesinger, F., Drenkow, J., Zaleski, C., Jha, S., Batut,
ing institutions were completed by R.G.T., W.O., I.Q., L.F.S., M.I.L., A.J.W., P., Chaisson, M., and Gingeras, T.R. (2013). STAR: ultrafast universal RNA-seq
S.G., M.R., T.B., D.S.T., N.L., and C.D’A. under supervision from M.A.K., M. aligner. Bioinformatics 29, 15–21.
Santi., and D.W.E. R.G.T., W.O., I.Q., L.F.S., M.I.L., A.J.W., S.G., M.R., T.B., Dougherty, M.J., Santi, M., Brose, M.S., Ma, C., Resnick, A.C., Sievert, A.J.,
D.S.T., N.L., C.D’A., A.K., G.M.I., J.D., P.D., M.D.T., J.T.R., S.L., M. Shroff., Storm, P.B., and Biegel, J.A. (2010). Activating mutations in BRAF characterize
M. Shago., L.-N.H., V.R., E.B., U.B., A.H., M.A.K., M. Santi., and D.W.E. pro- a spectrum of pediatric low-grade gliomas. Neuro Oncol. 12, 621–630.
vided patient-related materials and/or clinical data used in this study. Histo-
Fangusaro, J., Onar-Thomas, A., Young Poussaint, T., Wu, S., Ligon, A.H.,
pathological analysis was completed by A.A., M. Shago., C.D’A., and C.H.
Lindeman, N., Banerjee, A., Packer, R.J., Kilburn, L.B., Goldman, S., et al.
Array and NGS data were processed and analyzed by S.R., R.S., and C.L. un-
(2019). Selumetinib in paediatric patients with BRAF-aberrant or neurofibro-
der supervision from U.T. and C.H. Statistical analysis was done by S.R., M.Z.,
matosis type 1-associated recurrent, refractory, or progressive low-grade gli-
R.S., and C.L. U.T. and C.H provided overall supervision for the project and
oma: a multicentre, phase 2 trial. Lancet Oncol. 20, 1011–1022.
wrote the manuscript with S.R. and M.Z. and with input from M.A.K., M. Santi.,
and D.W.E. All authors reviewed the manuscript and figures before Fina, F., Barets, D., Colin, C., Bouvier, C., Padovani, L., Nanni-Metellus, I.,
submission. Ouafik, L., Scavarda, D., Korshunov, A., Jones, D.T., et al. (2017). Droplet dig-
ital PCR is a powerful technique to demonstrate frequent FGFR1 duplication in
dysembryoplastic neuroepithelial tumors. Oncotarget 8, 2104–2113.
DECLARATION OF INTERESTS Gao, Q., Liang, W.W., Foltz, S.M., Mutharasu, G., Jayasinghe, R.G., Cao, S.,
Liao, W.W., Reynolds, S.M., Wyczalkowski, M.A., Yao, L., et al. (2018).
The authors declare no competing interests.
Driver fusions and their implications in the development and treatment of hu-
man cancers. Cell Rep. 23, 227–238.
Received: January 3, 2020
Revised: January 27, 2020 Ge, H., Liu, K., Juan, T., Fang, F., Newman, M., and Hoeck, W. (2011).
Accepted: March 12, 2020 FusionMap: detecting fusion genes from next-generation sequencing data at
Published: April 13, 2020 base-pair resolution. Bioinformatics 27, 1922–1928.
Greenberg, M.L., Barr, R.D., DiMonte, B., McLaughlin, E., and Greenberg, C.
(2003). Childhood cancer registries in Ontario, Canada: lessons learned from
REFERENCES a comparison of two registries. Int. J. Cancer 105, 88–91.
Guerreiro Stucklin, A.S., Ryall, S., Fukuoka, K., Zapotocky, M., Lassaletta, A.,
Aghajan, Y., Levy, M.L., Malicki, D.M., and Crawford, J.R. (2016). Novel
Li, C., Bridge, T., Kim, B.5, Arnoldo, A., Kowalski, P.E., et al. (2019). Alterations
PPP1CB-ALK fusion protein in a high-grade glioma of infancy. BMJ Case
in ALK/ROS1/NTRK/MET drive a group of infantile hemispheric gliomas. Nat.
Rep. 16, 2016.
Commun. 10, 4343.
Anders, S., Pyl, P.T., and Huber, W. (2015). HTSeq-A Python framework to Gutmann, D.H., Ferner, R.E., Listernick, R.H., Korf, B.R., Wolters, P.L., and
work with high-throughput sequencing data. Bioinformatics 31, 166–169. Johnson, K.J. (2017). Neurofibromatosis type 1. Nat. Rev. Dis. Primers
Balss, J., Meyer, J., Mueller, W., Korshunov, A., Hartmann, C., and von 3, 17004.
Deimling, A. (2008). Analysis of the IDH1 codon 132 mutation in brain tumors. Hartmann, C., Meyer, J., Balss, J., Capper, D., Mueller, W., Christians, A.,
Acta Neuropathol. 116, 597–602. Felsberg, J., Wolter, M., Mawrin, C., Wick, W., et al. (2009). Type and fre-
Bandopadhayay, P., Ramkissoon, L.A., Jain, P., Bergthold, G., Wala, J., Zeid, quency of IDH1 and IDH2 mutations are related to astrocytic and oligoden-
R., Schumacher, S.E., Urbanski, L., O’Rourke, R., Gibson, W.J., et al. (2016). droglial differentiation and age: a study of 1,010 diffuse gliomas. Acta
MYB-QKI rearrangements in angiocentric glioma drive tumorigenicity through Neuropathol. 118, 469–474.
a tripartite mechanism. Nat. Genet. 48, 273–282. Jacob, K., Quang-Khuong, D.A., Jones, D.T., Witt, H., Lambert, S., Albrecht,
Banerjee, A., Jakacki, R.I., Onar-Thomas, A., Wu, S., Nicolaides, T., Young S., Witt, O., Vezina, C., Shirinian, M., Faury, D., et al. (2011). Genetic aberra-
Poussaint, T., Fangusaro, J., Phillips, J., Perry, A., Turner, D., et al. (2017). A tions leading to MAPK pathway activation mediate oncogene-induced senes-
phase I trial of the MEK inhibitor selumetinib (AZD6244) in pediatric patients cence in sporadic pilocytic astrocytomas. Clin. Cancer Res. 17, 4650–4660.
with recurrent or refractory low-grade glioma: a Pediatric Brain Tumor Jones, D.T., Kocialkowski, S., Liu, L., Pearson, D.M., Ba €cklund, L.M.,
Consortium (PBTC) study. Neuro Oncol. 19, 1135–1144. Ichimura, K., and Collins, V.P. (2008). Tandem duplication producing a novel
Article
MYC Drives Temporal Evolution of Small
Cell Lung Cancer Subtypes
by Reprogramming Neuroendocrine Fate
Abbie S. Ireland,1 Alexi M. Micinski,1 David W. Kastner,1 Bingqian Guo,1 Sarah J. Wait,1 Kyle B. Spainhower,1
Christopher C. Conley,2 Opal S. Chen,3 Matthew R. Guthrie,1 Danny Soltero,1 Yi Qiao,4 Xiaomeng Huang,4
Szabolcs Tarapcsák,4 Siddhartha Devarakonda,5 Milind D. Chalishazar,1 Jason Gertz,1 Justin C. Moser,6 Gabor Marth,4
Sonam Puri,7 Benjamin L. Witt,8,9 Benjamin T. Spike,1 and Trudy G. Oliver1,10,*
1Department of Oncological Sciences, Huntsman Cancer Institute, University of Utah, Salt Lake City, UT 84112, USA
2Huntsman Cancer Institute Bioinformatic Analysis Shared Resource, Huntsman Cancer Institute, University of Utah, Salt Lake City, UT
84112, USA
3Huntsman Cancer Institute High-Throughput Genomics Shared Resource, Huntsman Cancer Institute, University of Utah, Salt Lake City, UT
84112, USA
4Utah Center for Genetic Discovery, Eccles Institute of Human Genetics, University of Utah, Salt Lake City, UT 84112, USA
5Division of Medical Oncology, Department of Medicine, Washington University School of Medicine, St. Louis, MO 63110, USA
6HonorHealth Research Institute, Scottsdale, AZ 85254, USA
7Department of Internal Medicine, University of Utah, Salt Lake City, UT 84112, USA
8Department of Pathology, University of Utah, Salt Lake City, UT 84112, USA
9ARUP Laboratories at University of Utah, Salt Lake City, UT 84108, USA
10Lead Contact
*Correspondence: trudy.oliver@hci.utah.edu
https://doi.org/10.1016/j.ccell.2020.05.001
SUMMARY
Small cell lung cancer (SCLC) is a neuroendocrine tumor treated clinically as a single disease with poor out-
comes. Distinct SCLC molecular subtypes have been defined based on expression of ASCL1, NEUROD1,
POU2F3, or YAP1. Here, we use mouse and human models with a time-series single-cell transcriptome anal-
ysis to reveal that MYC drives dynamic evolution of SCLC subtypes. In neuroendocrine cells, MYC activates
Notch to dedifferentiate tumor cells, promoting a temporal shift in SCLC from ASCL1+ to NEUROD1+ to YAP1+
states. MYC alternatively promotes POU2F3+ tumors from a distinct cell type. Human SCLC exhibits intratu-
moral subtype heterogeneity, suggesting that this dynamic evolution occurs in patient tumors. These
findings suggest that genetics, cell of origin, and tumor cell plasticity determine SCLC subtype.
Significance
SCLC has historically been treated as a single disease, but is now recognized to comprise multiple molecular subtypes. We
find that MYC directly activates NOTCH signaling to reprogram neuroendocrine SCLC from ASCL1+ to NEUROD1+ to YAP1+
non-neuroendocrine states. Therefore, SCLC molecular subtypes are not distinct, but rather represent dynamic stages of
MYC-driven tumor evolution. We find that individual human SCLC tumors are composed of multiple molecular subtypes.
Given the reported unique therapeutic vulnerabilities of each subtype, we postulate that SCLC tumors represent a ‘‘moving
therapeutic target’’ that may require more general, combinatorial, or plasticity-directed therapeutic approaches to combat
this transcriptional flexibility. We speculate that molecular subsets of other cancer types may also represent dynamic stages
of tumor evolution.
60 Cancer Cell 38, 60–78, July 13, 2020 ª 2020 Elsevier Inc.
ll
Article
A B
C D
F G H
on the initiating cell of origin (Park et al., 2016; Wang et al., 2013). in SCLC with 25% of tumors harboring loss-of-function alter-
The childhood cerebellar tumor medulloblastoma comprises ge- ations in Notch receptors (George et al., 2015). During lung injury,
netic subtypes (i.e., WNT, SHH) and subtypes that appear to pro- Notch2 marks an NE-stem cell population (NEstem) that undergoes
ceed along a developmental trajectory or continuum (i.e., group self-renewal in the absence of Notch activation (Ouadah et al.,
C and D) (Hovestadt et al., 2019). For most cancers, the relative 2019). Loss of Notch function in the context of Rb1 and Trp53
contribution of oncogenic pathway alterations, cell of origin, and loss is postulated to lock cells in a self-renewing NEstem-like state
biological plasticity to the overall tumor phenotype (i.e., subtype) and thereby contribute to SCLC. However, Notch is active in a
is unknown. subset of human SCLC, and Notch activation can promote non-
Small cell lung cancer (SCLC) has historically been treated as a NE SCLC fate (Lim et al., 2017). Moreover, an unknown signal
single disease without patient stratification. SCLC exhibits ge- activates Notch signaling during lung injury to promote transit
netic loss of both tumor suppressors RB1 and TP53, along amplification and deprogramming of NEstem cells to non-NE fates
with mutually exclusive expression of MYCL, MYC, or MYCN (Ouadah et al., 2019). While MYC and NOTCH have both been
(Bragelmann et al., 2017; Dammert et al., 2019; George et al., implicated in non-NE cell fate in SCLC, a functional relationship
2015; Peifer et al., 2012; Poirier et al., 2015; Rudin et al., 2012). between MYC and NOTCH in this setting has not been defined.
Large-scale gene expression analyses of human tumors and Here, we investigate the function of MYC in the origins and re-
cell lines suggest that SCLC comprises four distinct molecular lationships among SCLC molecular subtypes.
subtypes based on expression of lineage-defining transcription
factors: ASCL1 (SCLC-A), NEUROD1 (SCLC-N), POU2F3 RESULTS
(SCLC-P), or YAP1 (SCLC-Y) (Rudin et al., 2019). Studies sug-
gest that SCLC subtypes have unique therapeutic vulnerabilities MYC Drives Multiple SCLC Molecular Subtypes In Vivo
(Cardnell et al., 2017; Chalishazar et al., 2019; Dammert et al., To determine which SCLC molecular subtypes are promoted by
2019; Huang et al., 2018a, 2018b; Mollaoglu et al., 2017; Owoni- MYC, we analyzed tumors in Myc-driven (Rb1fl/fl;Trp53fl/fl;
koko et al., 2019), emphasizing the clinical importance of under- Lox-Stop-Lox [LSL]-MycT58A, RPM) and Mycl-driven
standing these subtypes. The origins and relationships among (Rb1fl/fl;Trp53fl/fl;Rbl2fl/fl, RPR2) SCLC GEMMs (Figure S1A) at
SCLC molecular subtypes are currently unknown (Poirier et al., time points with similar tumor burden (Kim et al., 2016; Schaffer
2020; Rudin et al., 2019). et al., 2010). We used adenoviral-Cgrp promoter-Cre viruses
The SCLC-ASCL1 subtype comprises 70% of human SCLC. (Ad-Cgrp-Cre) to specifically transform PNECs (Sutherland
ASCL1 is a master regulator of neuroendocrine (NE) fate that is et al., 2011). Early in situ lesions in RPM and RPR2 models exhibit
required for pulmonary NE cell (PNEC) development, and labels high expression of NE markers, including ASCL1, SYP, CGRP,
adult PNECs (Ito et al., 2001). The adult PNEC is a cell of origin UCHL1, and DLL3 (Figures 1A, 1B, and S1B). Invasive RPR2
for ASCL1+ tumors in multiple Rb1/Trp53 (RP)-null genetically en- tumors retained an NE-high identity, whereas invasive RPM tu-
gineered mouse models (GEMMs) of SCLC (Meuwissen et al., mors displayed significantly reduced NE marker expression (Fig-
2003; Park et al., 2011; Sutherland et al., 2011). In these GEMMs, ures 1A, 1B, and S1B). Bulk gene expression data from human
ASCL1 is essential for tumor development (Borromeo et al., 2016; SCLC cell lines has shown that Myc, Notch/Rest, Hippo/Yap1,
Kim et al., 2016). MYCL is amplified or highly expressed in the and epithelial-mesenchymal transition (EMT) pathways correlate
SCLC-A subtype and is necessary for SCLC-A development (Bor- with non-NE fate in SCLC (Zhang et al., 2018). We analyzed
romeo et al., 2016; Dooley et al., 2011; George et al., 2015; Huijb- expression of SCLC subtype markers and a subset of the non-
ers et al., 2014; Kim et al., 2016; McFadden et al., 2014). In NE fate markers in RPM and RPR2 tumors. Invasive RPM tumors
contrast, the other three SCLC subtypes representing 30% of gained expression of non-NE markers, including NEUROD1,
tumors (SCLC-N, SCLC-P, SCLC-Y) tend to exhibit amplification YAP1, HES1, ZEB1, and CD44 (Figures 1C, 1D, and S1B).
or overexpression of MYC, and exhibit a low or non-NE cell fate MYC expression was high in both in situ and invasive RPM tu-
(George et al., 2015; Rudin et al., 2019). We showed that Myc mors, and not detectable in RPR2 tumors (Figure S1B). Interest-
expression drives a non-NE SCLC phenotype in RP GEMMs ingly, POU2F3 was rarely detected in RPM tumors derived from
and tumors express NEUROD1 (Mollaoglu et al., 2017). However, PNECs (Figure 1E). However, tumors initiated in RPM mice with a
the relationship between SCLC-A and SCLC-N, and whether MYC general promoter (Ad-CMV-Cre), but not a club or alveolar type II
drives SCLC-P or SCLC-Y subtypes is unknown. (AT2) promoter, were enriched for POU2F3 expression (Figures
Notch signaling regulates PNEC fate and SCLC tumorigenesis 1E, 1F, and S1C). POU2F3 was not detected in Mycl-associated
with dichotomous functions. Notch acts as a tumor suppressor RPR2 tumors initiated from a general or NE promoter (Figure 1E).
(C) Representative IHC for non-NE markers in in situ or invasive tumors in indicated GEMMs infected with Ad-Cgrp-Cre.
(D) IHC quantification from (C).
(E) Representative IHC for POU2F3 in in situ or invasive tumors in indicated GEMMs infected with cell-type-specific Cre viruses. Positive control (+) is adult mouse
skin. IHC quantification in the right panel.
(F) Percentage of RPM tumors per animal that are POU2F3+ by IHC after cell-type-specific Cre viral infection. Indicated p values relative to Cmv.
(G) Left panel: percent of POU2F3+ (P) tumors expressing subtype markers analyzed by IHC from serial sections (n = 27 total tumors). Right panel: average
percentage of positive cells for each marker within individual POU2F3+ tumors.
(H) Representative IHC from serial sections of POU2F3+ tumors analyzed for SCLC subtype markers. Left color label indicates classification as in (G).
Data from n = 6–9 mice per genotype, except for Ccsp- and Spc-Cre mice, n = 5. Number of tumors scored by manual H-Score method is indicated within bar
graphs. Scale bars, 25 mm. Mean ± SEM, Student’s two-tailed unpaired t test, *p < 0.045, **p < 0.009, ***p = 0.002, ****p < 0.0001; n.s., not significant. See also
Figure S1.
A C
H&E
Classic
A
RPM
(Rb1, Trp53, MycT58A)
Variant
In situ
- -
ASCL1+ NEUROD1 YAP1
4 11 21
Days
Single-cell suspension
B
Day 4 7 11 14 17 21
RPM
Day 1 5 11 13 22 35
RPR2
D E
Day 5 7 10 12 14 17 19 21 24
ASCL1
MYC+
Vector MycT58A
MYC+
Day 22 25 30 22 25 30 Vector MycT58A
EPCAM
Day 28 32 35 37 28 32 35 37
INSM1 MYC
REST INSM1
ASCL1
NOTCH2
YAP1
N2ICD
N2ICD
NOTCH2
N2ICD ZEB1 YAP1
HES1 YAP1
NEUROD1
MYC NEUROD1
HSP90
NKX2-1 HSP90
H1963
HSP90 H889
F
Circularity: 0.59 ± 0.06 Circularity: 0.64 ± 0.07
Roundness: 0.67 ± 0.03 Roundness: 0.65 ± 0.05
Vector
Vector
Cluster size: 79.4 microns2 ± 35.3 Cluster size: 137.3 microns2 ± 29.5
MycT58A
H889 H1963
Analysis of serial sections of POU2F3+ tumors revealed that sessed markers of SCLC subtypes. Cells at day 5 displayed an
44% lacked expression of other SCLC subtype markers, NE-high identity, marked by ASCL1 and NE marker expression
whereas the majority of the remaining tumors expressed (Figure 2D). NEUROD1 was transiently expressed from days 5
POU2F3 and NEUROD1 (Figures 1G and 1H). The fraction of to 12, followed by expression of the non-NE markers REST,
cells expressing other subtype markers within POU2F3+ tumors YAP1, NOTCH2, and NOTCH2’s active cleaved intracellular
was relatively minor (<16% of cells) (Figures 1G and 1H). These domain (N2ICD), and HES1 at days 10–24. In contrast, control
data suggest that MYC-driven SCLC-P tumors predominantly RPR2 tumor cells expressed ASCL1, and did not induce non-
arise from an unknown cell of origin that is not a PNEC, club, NE markers (Figure S2C). Ectopic MycT58A-Ires-Gfp expression
or AT2 cell, and that MYC (as opposed to MYCL) promotes in RPR2 cells triggered a variant morphology of looser clusters
SCLC-P development. Since POU2F3 is a master driver of the or chains, and induced non-NE markers (Figures S2C and
tuft cell lineage (Huang et al., 2018b), future studies will be S2D). This finding indicates that the NE-high to non-NE fate tran-
required to determine whether MYC-driven SCLC-P tumors arise sition is not simply a consequence of prolonged culture.
from tuft cells. To test if MYC is sufficient to alter tumor cell fate in established
Together, these data suggest that in the context of Rb1 and human SCLC cells, we expressed MycT58A-Ires-Gfp in NE-high,
Trp53 loss, MYC can promote SCLC-N and SCLC-Y molecular MYCL-associated classic lines. Ectopic MYCT58A expression
subtypes from an ASCL1+ PNEC cell of origin. Furthermore, altered cell morphology from classic to variant-like, and induced
MYC drives SCLC progression in PNECs from an NE-high to non-NE marker expression (Figures 2E and 2F). The levels of
non-NE phenotype. These findings are consistent with the ectopic MYC overexpression were consistently less than that
enrichment of MYC expression in human SCLC-N, -Y, and -P of MYC-high SCLC cell lines and RPM tumors (Figure S2E), sug-
subtypes (Rudin et al., 2019), and demonstrate that MYC is a gesting this phenotype is not due to supraphysiological MYC
driver of these SCLC molecular subtypes in vivo. levels. NEUROD1 expression was not detected in these experi-
ments (Figures 2E and S2C), suggesting that NEUROD1 expres-
MYC Drives SCLC Subtype Evolution In Vitro sion is either early and transient (as in Figure 2D) or that YAP1
Static observations of tumor histology provide limited insight into expression can be induced by MYC without NEUROD1 expres-
the potential temporal connections between SCLC subtypes. To sion. Despite the gain of non-NE marker expression and variant
address this point, we macro-dissected central portions of the morphology induced by MYC, mouse and human MYC-express-
lung of Ad-Cgrp-Cre-infected RPM mice before detection of ing cells retained expression of some NE markers, including
macroscopic lesions at a time point with predominantly ASCL1 (Figures 2E and S2C), suggesting either a
ASCL1+ in situ tumors (Figure 2A). We cultured dissociated cells heterogeneous population, and/or the presence of hybrid NE/
and within 3–4 days, large clusters of tumor cells grew in suspen- non-NE cells. Overall, these data demonstrate MYC’s ability to
sion (Figure 2B). The tumor cells initially form tight, round, spher- promote non-NE tumor cell fate and subtype evolution from
ical aggregates that resemble classic, NE-high human SCLC cell SCLC-A to SCLC-Y in tumor cells in vivo and in human
lines (Figures 2B and 2C), but eventually form amorphous clus- SCLC cells.
ters with a ‘‘chain-link’’ morphology that resembles variant,
non-NE human SCLC cells (Figures 2B, 2C, and S2A). This strik- Human SCLC Subtypes Correspond with MYC-Driven
ing transition occurred reproducibly over a period of 20 days, Evolution
consistent with the time frame from early-to-invasive SCLC pro- Next, we sought to investigate the transcriptional states of MYC-
gression in the RPM GEMM (Mollaoglu et al., 2017). There was a driven tumor evolution in human SCLC tumors. We performed
minor level of cell death during culture that was relatively consis- bulk RNA sequencing (RNA-seq) of RPM tumor cells from Ad-
tent throughout the time course (Figure S2B). Importantly, early- Cgrp-Cre-infected mice at eight time points during the transition
stage RPR2 tumor cells maintained a classic morphology in spanning 21 days in culture. Analysis of NE and non-NE pathway
culture (Figures 2B and S2A). genes confirmed a temporal loss of NE identity, and a subse-
To characterize the RPM tumor cell transition, we harvested quent gain of non-NE signaling pathways, including Notch/
RPM tumor cells at multiple time points during culture and as- Rest, Hippo/Yap1, and EMT (Figures 3A and S3A) (Zhang
A B
NE markers EMT NE genes Non-NE genes
10 Neurod1 6 Twist2
Enrichment score
Chga Vim 0.8
0.0
8 Ascl1 Tgfb2 0.6
-0.2
0.4
Syp 4 Snai1 -0.4
6 Twist1
0.2
Uchl1 0.0 -0.6
4 Dll3 Snai2
Insm1 2 Tgfbr2
log2(RPKM fold change+1)
2 Nkx2-1 Zeb1 RPM early RPM late RPM early RPM late
Epcam
0 0 NES 5.15 p<0.0001 NES -3.75 p<0.0001
0 5 10 15 20 25 0 5 10 15 20 25
log2(RPKM+1)
Hes1 Wwtr1 6
4 Rest 4 Yap1 Yap1
Notch2 Tead3 4 Pou2f3
2 Jag2 2 Ctgf
Rbpj 2
0 0
0 5 10 15 20 25 0 5 10 15 20 25
0
Days in culture 0 5 10 15 20 25
Days in culture
D
Day 3-5 Day 7-10 Day 14-21
Enrichment score
Enrichment score
Enrichment score
0.20 0.05
0.25
0.20 0.15 0.00
0.15 0.10 -0.05
hSCLC-A
Day 3-5 Day >7 Day 7-10 Day <7, >10 Day 14-21 Day <14
Enrichment score
Enrichment score
0.15 0.05
0.25 0.00
0.20 0.10 -0.05
0.15 -0.10
hSCLC-N
Day 3-5 Day >7 Day 7-10 Day <7, >10 Day 14-21 Day <14
Enrichment score
Enrichment score
-0.10 -0.10
0.05
-0.15 0.00
-0.15
-0.20 -0.05
-0.20
Day 3-5 Day >7 Day 7-10 Day <7, >10 Day 14-21 Day <14
et al., 2018). Gene set enrichment analyses (GSEA) of an estab- scription factor genes Ascl1, Neurod1, and Yap1 during the tran-
lished 50 gene signature comprising 25 NE and 25 non-NE- sition (Figure 3C). Consistent with a lack of POU2F3 in RPM tu-
related genes from human SCLC cell lines (Zhang et al., 2018) mors derived from PNECs (Figure 1E), we observed extremely
demonstrated that MYC promotes a shift from NE to non-NE low counts of Pou2f3 mRNA during the transition (Figure 3C).
transcriptional states (Figures 3B and S3A). We observed To determine whether the expression of the subtype-defining
dynamic and sequential expression of the subtype-defining tran- transcription factors correlate with gene expression patterns in
A B
Days Day 4 11 17
25
4 7 11 14 17 21 7 14 21
tSNE 2
0
-25
(n = 15,434)
-50
-25 0 25
tSNE 1
Expression
2 2
5
6 2
Insm1 Epcam Hes1 Cd44
Day 4 (8174) 14 (5093) 1 Tumor cells 5 B/T cells Max
7 (4281) 17 (5086) 2 Stromal/myeloid 6 NK cells
11 (4371) 21 (4514) 3 AT2/club cells 7 Low quality cells
4 Neutrophils
D F
Pseudotime
Early Late
Epcam
log2 expression
3
Ascl1
Calca
0 Tcf4
Dll3
Pseudotime Day 4 7 11 14 17 21
Epha5
-3
Pou3f4
Early Late Syt11
Hdac9
Mycl
Sox11
Hes6
4% 26% 39% 30% Fez1
Myt1l
Sox1
Ncam
Uchl1
Bex2
RPM1 RPM2 Bex4
Chga
RPM3 RPM4 Myt1
4 11 17 Chgb
Day
7 14 21 Syp
Bex1
Insm1
RPM1 RPM2 RPM3 RPM4
(n = 2107) (n = 809) (n = 248) (n = 772) Runx1
Vim
Wwtr1
Anxa1
E Tgfb1
Tgfbr2
Neat1
Lgals1
Tead2
Twist2
Ascl1 Neurod1 Yap1 Pou2f3 Bmp4
Tuba1c
Cd44
Rest
Hes1
Tgfb2
Min
Expression
Mycl Myc
Max
human tumors, we clustered 81 human SCLC tumors and 51 Next, we performed scRNA-seq on four invasive RPM-Cas9
human SCLC cell lines according to molecular subtype using tumors, one predominantly ASCL1-high (RPM1) and three that
bulk RNA-seq data (Figure S3B). We then created human had varying levels of NEUROD1 and YAP1 expression (RPM2-
SCLC-subtype-specific gene signatures using the most highly 4) (Figures S4D and S4E). Combining the RPM transition time
expressed genes per subtype (Table S1), and applied GSEA to points and tumor samples, we constructed pseudotime trajec-
determine whether each signature was enriched or depleted dur- tories to identify predicted transcriptional relationships among
ing the RPM transition (Figure 3D). Consistent with the patterns the tumor cells. Unsupervised pseudotime ordering of combined
of transcription factor gene expression (Figure 3C), days 3–5 tumor cells predicted a single lineage trajectory that corre-
and days 7–10 cells were enriched for human SCLC-A and sponds with early to late transition time points (Figures 4D–4F).
SCLC-N signatures, and depleted for the SCLC-Y signature. In Faceted pseudotime trajectories reveal that cells of the RPM tu-
contrast, day 14–21 cells were enriched for the SCLC-Y mors progress along the same trajectory as transitioning cells
signature and depleted for SCLC-A and SCLC-N signatures, in vitro, with the bulk of tumor cells in the earliest stages of pseu-
demonstrating that MYC-driven tumor cell evolution in the dotime and fewer cells progressing to later stages (Figure 4D).
mouse corresponds with subtype-defining transcriptional signa- RPM2-4 tumors had more cells in the later time points than
tures in human tumors. We confirmed that Rb1 and Trp53 were RPM1, consistent with their lower ASCL1 levels and higher
recombined in RPM tumor cells (Figures S3C and S3D). Whole- YAP1 protein expression (Figures 4D and S4D).
genome sequencing of early and late time point RPM tumor cells Expression of Ascl1, Neurod1, and Yap1 were consistent with
also confirmed complete loss of expected regions of Rb1 and MYC-driven temporal evolution, whereas Pou2f3 was rarely de-
Trp53 with no detectable copy number variations (Figures S3E tected (Figure 4E). Mycl was associated with Ascl1 expression in
and S3F). Together, MYC-driven tumor cell evolution in vitro re- early time points and absent during tumor cell progression (Fig-
flects the temporal phenotypic changes observed in the RPM ure 4E), consistent with studies showing that ASCL1 and MYCL
GEMM, and places three of four human SCLC subtypes along are coordinately expressed in chemo-naive SCLC, and reduced
a defined MYC-driven trajectory from SCLC-A to SCLC-N to in chemotherapy-relapsed SCLC (Wagner et al., 2018). Expres-
SCLC-Y. sion of the top 500 differentially expressed genes across pseu-
dotime demonstrate broad loss of NE genes followed by gain
of non-NE genes (Figure 4F and Table S2).
MYC-Driven SCLC Subtypes Progress along a Single We detected a small number of Ascl1-high cells at day 4 (n = 8
Evolutionary Trajectory cells) that distributed to the late end of pseudotime (Figures 4D
To better understand the transcriptional changes during MYC- and S4F); only one of these cells clustered with late-stage tumor
driven SCLC progression, we performed single-cell RNA-seq cells in tSNE space (Figures S4F and S4G), suggesting that these
(scRNA-seq). We isolated unsorted early-stage tumor cells cells are transcriptionally dissimilar from both early- and late-
from RPM-Rosa26-LSL-Cas9-Ires-Gfp (RPM-Cas9) mice as stage tumor cells. As an orthogonal approach to predicting
they were transitioning from days 4 to 21 in culture at six distinct cellular trajectories, we performed diffusion mapping (Angerer
time points (Figure 4A). Approximately 4–8,000 cells were et al., 2016; Coifman et al., 2005; Giraddi et al., 2018). Diffusion
captured per time point (n = 31,519 total cells). Days 4 and 7 cells mapping revealed the same cellular trajectory from early, NE-
comprised both tumor and non-tumor cell populations and the high to late, non-NE cell states in transitioning cells in culture
non-tumor populations were depleted in culture by day 11 (Fig- and in tumors (Figure S4H). Diffusion components (DCs) that
ures 4A and S4A). For downstream analyses, non-tumor and best distribute time points in culture were compounded to define
low-quality cells were filtered out of the time course (Figures a principal curve predicting diffusion pseudotime (i.e., cellular
4B and S4A). Minor variation in gene expression due to cell-cycle trajectory) (Figure S4I). Diffusion pseudotime placed a small
genes was regressed out for downstream analyses (Figure S4B). number of day 4 cells late in pseudotime (Figure S4J), but these
Unsupervised tSNE clustering of tumor cells based on top highly cells were transcriptionally distinct from other day 4 cells and the
expressed genes revealed at least three distinct clusters corre- bulk of late cells from other time points and tumors (Figure S4K).
sponding with the vast majority of days 4–7, 11, and 14–21 cells, Diffusion analysis also revealed some features distinguishing pri-
respectively (Figure 4B). Cells in the days 4–7 cluster expressed mary tumor cells and transitioning cultured cells (e.g., DC6, DC4)
high levels of Ascl1 and NE markers, which were largely absent in (Figure S4L). Monocle and diffusion map pseudotime coordi-
days 11–21 cells that instead expressed high levels of non-NE nates were significantly, positively correlated (Figure S4M).
markers (Figure 4C). Although we captured only a few Neu- Together, pseudotime analyses are consistent with our data sug-
rod1-expressing cells in the time course, these cells clustered gesting that MYC drives the temporal evolution of SCLC fate
with NE-high cells (Figure S4C). from NE to non-NE states in culture and in tumors in vivo.
(B) Following removal of non-tumor and low-quality cells, RPM tumor cells labeled by day in tSNE space using Monocle 2.
(C) Expression of individual NE and non-NE marker genes in tSNE space from cells in (B).
(D) Pseudotime trajectory by Monocle 2 from early to late time points of primary RPM transition and four RPM-Cas9 tumors from Ad-Cgrp-Cre-infected mice.
Faceted pseudotime plots indicated on the top right and bottom right panels. Dashed insets in RPM tumors highlight percentage of cells in late stages of
progression; total post-QC number of tumor cells indicated below RPM1-4 labels.
(E) Expression of indicated genes projected onto pseudotime space as in (D).
(F) Heatmap of top 500 differentially expressed genes over pseudotime from the primary RPM tumor cell transition and 4 RPM tumors.
See also Figure S4 and Table S2.
A B
ASCL1 NEUROD1 YAP1
% positive
83%
3.4% ASCL1 + YAP1
50 NEUROD1 + YAP1
86.8% 50.8% 87% ASCL1 + NEUROD1 + YAP1
25 1 subtype
3.4% >1 subtypes
13.6% 17%
0
In situ Invasive In situ Invasive
C
100 4.1%
Day 3-5
Invasive tumors
20.3%
80 Day 7-10
% expression
16.1% Day 12-14
60 Day 19-21
40 59.5%
20
0
RPM tumors (n = 10) Avg
D E
100 100
Ascl1 only Ascl1 only
35%
75 Neurod1 only 75 Neurod1 only
Yap1 only Yap1 only
% cells
% cells
0
0
Day 4 7 11 14 17 21
1
g
PM
PM
PM
PM
Av
R
High
Expression
Ascl1
Low
88% 40% 59% 32%
High
Expression
Neurod1
Low
33% 16% 10% 9%
High
Expression
Yap1
Low
2% 44% 35% 62%
MYC-Driven Murine Tumors Exhibit Intratumoral tumors and in mouse RPM versus RPR2 tumors were used to
Molecular Subtype Heterogeneity define a conserved MYC-ChIP score (Table S3). Application of
Pseudotime analyses suggest that individual RPM tumors are the MYC-ChIP score to the time-series data revealed a small,
composed of cells representing multiple SCLC molecular sub- but significant increase in MYC activity from day 11 onward (Fig-
types (Figures 4D and 4E). To test this more comprehensively, ure 6E), validating the ENRICHR predictions (Figure 6B). These
we analyzed serial sections of individual RPM tumors at the in analyses suggest that days 7–11 represent a key transition state,
situ or invasive stage for subtype markers (Figures 5A and 5B). characterized by loss of NE identity and high MYC activity.
In situ tumors were predominantly ASCL1+ and harbored a single Because we observed Notch/Rest pathway induction during
molecular subtype marker (87%). In contrast, invasive tumors MYC-driven SCLC progression, we analyzed the expression of
predominantly harbored two or more subtypes, either ASCL1 Notch-related machinery and target genes during the transition.
and NEUROD1, or all three molecular subtypes (ASCL1, NEU- Expression of Notch-inhibitory factors were increased at early
ROD1, and YAP1). CIBERSORT analyses from an additional time points (Figures S5B and S5C), including the inhibitory Notch
ten RPM tumors also predicted intratumoral heterogeneity with ligand gene Dll3, Hes6 (HES6 is a repressor of HES1), Fbxw7
cells throughout the stages of MYC-driven progression (Fig- (FBXW7 promotes NOTCH degradation), and Ncor2 (NCOR2
ure 5C). Analyses of the RPM tumors by scRNA-seq suggest functions in a Notch-corepressor complex) (Gratton et al.,
that there is minimal co-expression of these markers within indi- 2003; Matsumoto et al., 2011; Saunders et al., 2015). In contrast,
vidual cells, with the exception of Ascl1 and Neurod1 (Figure 5D), pro-Notch signaling genes increased in expression over time,
consistent with their close connection in pseudotime. The including the Notch target genes Hes1 and Rest, and the Notch
average abundance of Ascl1, Neurod1, and Yap1 in these tu- receptor gene, Notch2 (Figures S5B and S5C). GSEA of Notch
mors is also relatively consistent with bulk RNA-seq and protein signaling and REST-transcriptional targets by bulk RNA-seq
analyses (Figures 5B–5D). scRNA-seq analyses of the RPM reveals increased Notch pathway activity and repressive REST
transition in culture also shows little overlap between Ascl1, Neu- activity in late compared with early transition time points (Fig-
rod1, and Yap1 in individual cells (Figure 5E), with day 11–21 ure S5D). Importantly, multiple Notch signaling components
cells resembling later stages of tumor progression. Thus, inde- were identified as MYC target genes, including Notch2, Hes1,
pendent methods suggest that MYC drives the evolution of mul- Hes6, and Jag2 (Figure 6F). Consistently, NOTCH2 and HES1
tiple SCLC subtypes in vivo. protein levels were induced by MYC in multiple cell types (Fig-
ures 2D, 2E, and S2C). Pro-Notch signaling genes NOTCH2,
MYC Activates Notch Signaling during NE HES1, and REST are also preferentially enriched in MYC-high
Dedifferentiation human SCLC (Figure 6G), while Notch-inhibitory HES6 is
To uncover transcriptional patterns associated with MYC-driven reduced, suggesting broad positive regulation of Notch signaling
tumor cell evolution, we performed differential gene expression by MYC in SCLC. Together, these data suggest that MYC in-
analysis using the scRNA-seq data (Figure 6A). ENRICHR anal- creases NOTCH/REST activity to destabilize NE identity during
ysis of top differentially expressed genes between the major MYC-driven SCLC evolution.
clusters of cells (days 4–7 versus day 11 versus days 14–21)
identified transcription factors predicted to be important for Notch Activation Is Required for MYC-Driven Tumor
SCLC fate (Figure 6B) (Wooten et al., 2019), including REST, Evolution
SUZ12, TCF3, NEUROD1, NHLH1, MYC, and SOX2. As Notch/ To determine whether Notch signaling is required for MYC-
Rest signaling can promote non-NE fate in SCLC (Lim et al., driven SCLC evolution, early RPM tumor cells were cultured in
2017), and REST is a top predicted regulator of the earliest vehicle control or the gamma-secretase inhibitor (GSI-IX) (e.g.,
changes promoted by MYC, we focused on the NE phenotypic DAPT) to block Notch signaling. In contrast to control cells that
switch. Using the human-derived NE score vectors from (Zhang began to convert to variant morphology on day 10, DAPT-treated
et al., 2018) (Table S3), we assigned an NE score to every cell in RPM cells did not switch to variant morphology until approxi-
the RPM time-series and bulk tumors (Figure 6C). The NE score mately day 20 (Figure 7A). Notch blockade increased and pro-
accurately predicted the decrease in NE identity over time with a longed ASCL1, INSM1, EPCAM, and NEUROD1 expression
dramatic reduction in NE score between days 7 and 11 compared with control cells (Figures 7A, S6A, and S6B). In
(Figure 6D). contrast, DAPT treatment delayed or blocked expression of
To determine how MYC promotes this transition at a mecha- non-NE markers (Figures 7B, S6A, and S6B). Interestingly,
nistic level, we performed chromatin-immunoprecipitation when DAPT is added to Notch-active fully progressed RPM tu-
sequencing (ChIP-seq) for MYC in invasive RPM tumors (n = 4) mor cells rather than cells in the classic/early state, we observe
representing a spectrum of tumor cell states (Figure S5A). The equivalent expression of non-NE markers and no reversion to NE
top 50 upregulated genes bound by MYC in RPM tumors whose marker expression, despite efficient Notch signaling blockade
expression was enriched in MYC-high versus -low human SCLC (Figure S6C). Therefore, it is possible that MYC-driven tumor
(B) IHC quantification from serial sections in (A) where individual tumors have detection of one or more than one (>1) subtypes.
(C) CIBERSORT analyses of bulk RNA-seq data from RPM tumors with average (Avg) percentage similarity to gene expression signatures of RPM transition cells
in culture.
(D) Top panel: percentage of cells per tumor expressing subtype-defining genes with average (Avg) across tumors. Bottom panel: individual RPM 1–4 trajectories
with localization of positive cells in pseudotime. Percentage of total tumor cells expressing indicated genes shown below trajectories.
(E) Percentage of cells expressing subtype-defining genes in the RPM transition experiment from Figure 4D.
A B
47 11 14 17 21 Day ENCODE and ChEA consensus TFs TF-Gene Coocurrence
REST ENCODE ISL1
Ascl1 REST CHEA INSM1
log2 expression
2 SOX2 CHEA MYT1
Epcam 1 TCF3 CHEA SOX11
SUZ12 CHEA NEUROD1
Insm1 0 SUZ12 ENCODE TOX3
Bex1 -1 FOSL2 ENCODE ST18
-2 RUNX1 CHEA NEUROD1
Uchl1 TCF3 ENCODE NEUROD4
GATA2 ENCODE NHLH1
Hes6
ENCODE and ChEA consensus TFs ENCODE and ChEA consensus TFs
MAX ENCODE MYC ENCODE
Hes1 MYC ENCODE SIN3A ENCODE
MYC CHEA TAF1 ENCODE
Odc1 TAF1 ENCODE MAX ENCODE
BRCA1 ENCODE BRCA1 ENCODE
Lgals1 SIN3A CHEA MYC CHEA
ATF2 ENCODE E2F1 CHEA
Lgals3 E2F6 CHEA PML ENCODE
ZMIZ1 CHEA NEF2L2 CHEA
Jun TCF3 ENCODE E2F6 ENCODE
Eno1
ENCODE and ChEA consensus TFs ChEA consensus TFs 2016
Vim TAF7 ENCODE E2F1 MESCs
NELFE ENCODE XRN2 HELA
Gstm1 TAF1 CHEA RUNX2 MC3T3
MYC ENCODE MYC MESCs
TCF3 ENCODE EKLF Erythrocyte
PML CHEA MYC MESCs
Fos BRCA1 ENCODE OCT4 MEFs
Fosb GATA1 CHEA SOX2 MEFs
E2F1 CHEA MYC MESCs
Egr1 SOX2 ENCODE SOX2 MESCs
Eef1a1
Up in days 4-7 v day 11 Up in day 11 v days 14-21
Up in day 11 v days 4-7 Up in days 14-21 v day 11
C D tumor
E
4 11 17 RPM1 RPM3
tumor
4 11 17 RPM1 RPM3
Day
Day
Expression 7 14 21 RPM2 RPM4 7 14 21 RPM2 RPM4
NE score Early ns
Min Max ***
pseudotime **** **** **** ns
***
0.8 **** 1.0 ****
0.4 0.5
tSNE 2
Min Max
0
0.0 0.0
-25
Late -0.4 -0.5
-50 pseudotime
4 7 11 14 17 21
RPM1
RPM2
RPM3
RPM4
4 7 11 14 17 21
RPM1
RPM2
RPM3
RPM4
-25 0 25
tSNE 1 Day Day
F G
MYC H3k27Ac E-box motif
[0-3.32] [0-5.64]
RPM-1
15 **** 40 * MYC-high
NOTCH2 FPKM
RPM-2
MYC-low
HES1 FPKM
RPM-3
RPM-4
30
[0-1.96] [0-2.47] 10
RPM-2
RPM-3
20
RPM-4 5
10
Notch2-> Hes1-> 0 0
4 *** 400 ns
HES6 FPKM
REST FPKM
[0-7.13] [0-8.53]
3 300
2 200
[0-2.22] [0-4.41] 1 100
0 0
Jag2-> Hes6->
evolution may be irreversible, but it remains to be determined Human SCLC Exhibits Intratumoral Molecular Subtype
whether blockade of other non-NE-related pathways could Heterogeneity
revert SCLC to an NE-high state. Next, we treated RPM mice We sought to determine the abundance of RPM time point gene
with early-stage tumors with DAPT to block Notch signaling signatures in bulk RNA-seq data from human SCLC tumors and
in vivo and monitored tumor growth by microCT imaging for cell lines (George et al., 2015; Newman et al., 2015). CIBERSORT
10 days, since control animals live an average of 12 days after analyses revealed that most human tumors are predicted to have
tumor detection by microCT imaging. DAPT treatment led to a cells resembling multiple stages of MYC-driven progression,
significant decrease in tumor burden in DAPT-treated versus regardless of their classified SCLC subtype (Figure 8A), reminis-
control mice (Figure 7C). Quantification of H&E-stained tissue re- cent of a similar approach predicting varying proportions of
vealed that tumor burden was significantly reduced by DAPT SCLC phenotypes in human tumors (Wooten et al., 2019). The
treatment (Figure 7D). DAPT-treated RPM tumors exhibited an majority of tumors harbor cells that resemble the NE-high
increase in classic morphology indicated by smaller cells with ASCL1+ early time point RPM cells (Figures 8A and 8B). The hu-
high nuclear:cytoplasmic ratios compared with control tumors man SCLC-N subtype was enriched for the day 7–10 time point
and a significant increase in NE identity, measured by ASCL1 signature and the human SCLC-Y samples exhibit a significantly
and DLL3 levels (Figures 7E and 7F). Moreover, DAPT-treated tu- higher percentage of non-NE late time point signatures (Figures
mors exhibited a decrease in tumor progression compared with 8A and 8B). This suggests that individual human tumors have
control tumors evident by reduced expression of NEUROD1 and cells in multiple stages of tumor progression, with MYC-high tu-
YAP1. We did not detect differences in HES1 expression, likely mors more likely to have cells at the latest stages of progression.
because HES1 levels were already low in control tumors at these Interestingly, some SCLC-A tumors were predicted to have a
time points (Figures 7E and 7F). MYC levels were high in both higher percentage of late-stage cells; these tumors had moder-
treatment groups with >90% of cells positive for MYC, suggest- ately higher MYC levels and significantly higher levels of
ing that the impact of Notch inhibition on cell state is not due to NOTCH2, HES1, and REST, consistent with a reduction in NE
reducing MYC levels. Treatment of tumor-bearing RPM mice identity (Figures 8A and 8C).
with a second Notch inhibitor (dibenzazepine) also significantly To verify these predictions in human tissue, we obtained 21
reduced tumor burden compared with control animals, but was chemo-naive human SCLC biopsies (n = 6 limited and 15 exten-
highly toxic limiting analysis to day 7 (Figures S6D and S6E). sive stage), since surgical specimens in SCLC are rare and diffi-
Together, these findings suggest that Notch inhibition signifi- cult to obtain due to metastatic disease. We performed immuno-
cantly inhibits MYC-driven tumor progression. histochemistry for ASCL1, NEUROD1, YAP1, and MYC on serial
Loss-of-function NOTCH mutations occur in 25% of human sections. Many tumors harbored cells with more than one sub-
SCLC, with mutually exclusive alterations in NOTCH1, -2, -3, and type, with the majority of tumors having some frequency of
-4 (George et al., 2015). We hypothesized that MYC requires ASCL1 and/or NEUROD1 protein (Figures 8D and 8E), consistent
intact NOTCH to drive SCLC subtype evolution. Using published with human gene expression data (Figure 8A). Approximately
functional classifications of NOTCH mutations, we grouped hu- 14% of samples harbored detectable MYC protein and, of these,
man SCLC tumors and cell lines as either NOTCH wild-type none of them were in the ASCL1-only group. Finally, we obtained
(WT) or silent, non-damaging, or damaging. Consistent with one fresh human SCLC biopsy from a patient who briefly re-
our hypothesis, MYC expression is significantly increased in sponded to carboplatin and etoposide and then progressed on
NOTCH WT or silent mutant compared with NOTCH-damaging carboplatin-irinotecan for scRNA-seq analyses. The biopsy
mutant samples (Figure 7G). All MYC-high SCLC samples (n = harbored distinct ASCL1+ and NEUROD1+ populations, and
30) are predicted to have intact NOTCH (Figure 7G, left panel). high MYC expression overlapped specifically with the NEU-
Many MYC-expressing tumors with intact NOTCH exhibit low ROD1-high population (Figures 8F, 8G, S7A, and S7B). While
NE scores, whereas all of the NOTCH-damaging mutant tumors these data warrant a more comprehensive analysis in human tis-
are NE-high (Figure 7G, right panel). We mined a recent cell line sue, these findings are consistent with bioinformatic predictions
genomics database (SCLC_CellMiner) (Tlemsani et al., 2020), suggesting that tumors frequently harbor cells representing mul-
which allowed a similar analysis with 50 additional human tiple SCLC subtypes.
SCLC cell lines (Figure 7H). There was a significant difference
in the quantity and proportion of MYC-high NOTCH WT versus DISCUSSION
MYC-high NOTCH mutant samples that were NE-low, consistent
with our model that MYC promotes non-NE progression by acti- While SCLC has historically been treated as a single disease,
vating NOTCH signaling. GSEA suggests that Notch signaling recent studies have converged on the concept that SCLC is
and MYC activity are enriched in NOTCH WT human SCLC composed of at least four molecular subsets with unique thera-
compared with samples with damaging mutations (Figure 7I; Ta- peutic vulnerabilities. Recent bioinformatic approaches suggest
ble S4). Together, these results suggest that MYC depends on that SCLC subtypes may represent dynamic states of transition
NOTCH to promote NE dedifferentiation, and that Notch (Wooten et al., 2019). Our functional data here support that hy-
blockade can inhibit MYC-driven tumor progression. pothesis and suggest that MYC has the capacity to shift SCLC
(G) Gene expression from George et al. (2015) analyzed according to MYC status (n = 59 MYC-low and n = 11 MYC-high tumors). Mean ± SEM, Student’s two-
tailed unpaired t test, ****p < 0.0001, ***p < 0.0007, *p < 0.03; n.s., not significant.
For boxplots, the median and interquartile range are shown (lower bar is the 25th percentile, upper bar is the 75th percentile, and endpoints indicate minimum and
maximum values). See also Figure S5 and Table S3.
A B
DMSO DAPT
Day 4 10 13 15 20 Day 4 7 10 13 15 4 7 10 13 15
**
ASCL1
DMSO
**
INSM1
*
ZEB1
YAP1
DAPT
N2ICD
REST
DMSO
DAPT HSP90
C Day 1 Day 10 D
DAPT Control DAPT **
Control 40
30 Control
* 21.0%
**
% tumor volume
30 DAPT
20
6.7%
ns 20
10
10
DAPT
0
Day 1 7 10 0
E F G
MYC-high ns
Control DAPT ASCL1 DLL3 15 ns
* 1.0
** **** ns
MYC log2(TPM+1)
% pos tumor cells
H&E
NE Score
non-damaging
50 50 0.0
damaging
5
-0.5
0 0
ASCL1
0 -1.0
NEUROD1 YAP1
****
H
% pos tumor cells
DLL3
NE-high 28 14
0 0 0.0
NE-low 25 1
-0.5
p = 0.0055
YAP1
HES1 MYC
**** -1.0
ns NOTCH WT Mutant
100
% pos tumor cells
100
HES1
50 50 I
Notch signaling MYC ChIP score
NOTCH WT vs mutant
Enrichment score
Enrichment score
0.20 0.20
0 0 0.15 0.15
MYC
0.10 0.10
molecular subtypes. We find, in the context of Rb1 and Trp53 loss, cological approaches that activate Notch in SCLC (Augert et al.,
MYC can promote three of the four molecular subsets from a 2019; Oser et al., 2019) to determine whether they can promote
PNEC cell of origin that proceed in a temporal evolution from tumor progression, particularly in collaboration with MYC.
SCLC-A to SCLC-N to SCLC-Y in vivo. Studies in a limited number The RPM primary cell culture model should be a valuable tool
of human SCLC cell lines and patient-derived xenograft models to better understand the role of additional signaling pathways in
also suggest that MYC can convert ASCL1+ SCLC to a variant MYC-driven tumor evolution. Lineage-tracing approaches in this
morphology with NEUROD1 expression (Johnson et al., 1986; Pa- model in vitro and in vivo would be a powerful complement to the
tel et al., 2019; Simpson et al., 2020). Interestingly, we observed studies here. As some of our assays suggest that SCLC can
SCLC-P in RPM mice when an unknown cell type was targeted progress from SCLC-A to SCLC-Y without evidence of the
that we speculate could be the tuft cell (Huang et al., 2018b; Rudin SCLC-N subtype, further studies are warranted to determine
et al., 2019). These data demonstrate that cell of origin, genetics, whether NEUROD1 is required or dispensable for SCLC progres-
and tumor cell plasticity can determine SCLC subtype. sion. The functions of Hippo/Yap1 and EMT in SCLC (also
Our data suggest that MYC requires Notch pathway activity to induced by MYC) are less well understood than Notch signaling
promote tumor progression. Notch activity can promote non-NE (Horie et al., 2016; Jia et al., 2018; McColl et al., 2017). Previous
fate in Mycl-associated SCLC models in the absence of MYC studies suggest that SCLC-Y human tumors are often RB1 WT
expression (George et al., 2015; Lim et al., 2017). In the RPR2 (McColl et al., 2017). We observe MYC-high YAP1+ tumor cells
model, tumors with MYC-negative Notch-active cells do not that appear to lack Rb1, suggesting there may be more than
develop variant morphology or express NEUROD1 or YAP1, in one mechanism to induce YAP1 in SCLC.
contrast to MYC-expressing Notch-active cells in our study. Our work builds on an emerging concept that MYC and MYCL
Together, these findings suggest that Notch activity alone is are not functionally redundant in SCLC. MYC and MYCL corre-
not sufficient to drive SCLC-N and SCLC-Y subtypes, and that late with distinct gene expression and methylation profiles, and
MYC and NOTCH likely cooperate with one another to drive localize to distinct super enhancers (Borromeo et al., 2016;
SCLC progression. Christensen et al., 2014; Poirier et al., 2015). Functional studies
Recently identified NEstem cells are enriched for Notch2 and modulating MYC reveal MYC’s capacity to change cell fate,
Hes1 expression (Ouadah et al., 2019). Hes1 is particularly en- morphology, drug sensitivity, and molecular subtype (Dammert
riched in RPM time points where MYC-driven tumor cells are et al., 2019; Mollaoglu et al., 2017; Patel et al., 2019). In other tu-
transitioning to a non-NE fate. HES1 is also enriched in mor types, such as medulloblastoma, MYC and MYCN also
ASCL1+ human SCLC tumors that exhibit a more non-NE fate appear to have distinct roles (Vo et al., 2016), illustrating diver-
in the absence of high MYC, consistent with a model whereby gent functions for MYC family members in cancer.
Notch can promote non-NE fate, but may be limited in its ability We find that many SCLC tumors are composed of cells repre-
to promote progression without MYC. We identify Notch2 and senting multiple molecular subtypes at different frequencies,
Hes1 as MYC target genes, suggesting that MYC could be the suggesting that bulk analysis methods may only identify the
unidentified signal that induces Notch signaling to deprogram most abundant state in a given tumor. Studies are increasingly
NEstem cells to other cell fates during lung injury (Ouadah et al., identifying unique therapeutic vulnerabilities for SCLC subtypes.
2019). We propose a model whereby NEstem-like tumors with For example, MYC-high tumors are preferentially sensitive to in-
defective Notch signaling are locked in an NE-high state, hibition of AURKA/B, CHK1, IMPDH1/2, and arginine deprivation
whereas tumors with intact Notch can be reprogrammed by (Cardnell et al., 2017; Chalishazar et al., 2019; Dammert et al.,
MYC to non-NE fates—explaining the dichotomous nature of 2019; Huang et al., 2018a; Mollaoglu et al., 2017; Sen et al.,
Notch signaling in SCLC (Figure 8H). While our pharmacological 2017). POU2F3+ SCLC cells are sensitive to IGF1R inhibitors
studies suggest that Notch activity is critical for MYC-driven tu- (Huang et al., 2018b), while ASCL1+ tumors express more
mor progression, genetic Notch knockout studies are warranted DLL3 and are sensitive to DLL3-targeting drugs (Cardnell et al.,
to fully test this model. These data warrant evaluation of pharma- 2017; Saunders et al., 2015). Considering data that SCLC
(C) Representative microCT (mCT) imaging from vehicle control- (corn oil) (n = 6) and 20 mg/kg DAPT-treated (n = 8) RPM mice. Lung tumors pseudocolored in
yellow and heart outlined in red. Quantification of microCT imaging data for total tumor burden (percentage of lung tumor volume/total lung volume) in the graph
on the right at the indicated days. Mean ± SEM, Student’s two-tailed unpaired t test, **p < 0.004, *p < 0.03; n.s., not significant.
(D) Representative H&E from vehicle control- (n = 7) or DAPT-treated (n = 6) RPM mice at day 10 following treatment. Scale bar, 2,000 mm. Quantification of
average tumor burden (% tumor area/total lung area) in right panel with boxplots where each dot represents one animal, Student’s two-tailed unpaired t test,
**p < 0.009.
(E) Representative H&E and IHC in vehicle control- (n = 7) or DAPT-treated (n = 6) RPM mice at day 10 after treatment. Scale bars, 20 mm.
(F) Digital IHC quantification from lung tumor tissue (percentage of positive tumor cells) in (E) where each dot represents a tumor. Student’s two-tailed unpaired t
test, ****p < 0.0001, **p < 0.03; n.s., not significant.
(G) Left panel: MYC expression in normalized transcripts per million (TPM) + 1 grouped by NOTCH status with MYC-high samples in blue; RNA-seq data from n =
70 human tumors (George et al., 2015) and n = 48 human cell lines (CCLE), excluding POU2F3+ samples. Right panel: NE score applied to the same samples.
‘‘Non-damaging’’ indicates missense or in-frame deletions, and ‘‘damaging’’ indicates frameshift deletions, nonsense, or splice variant mutations. Mean ± SEM,
Student’s two-tailed unpaired t test with Welch’s correction, *p < 0.01; n.s., not significant.
(H) NE score applied to RNA-seq data from n = 70 human tumors (George et al., 2015) and n = 98 human cell lines (SCLC_CellMiner), excluding POU2F3+
samples; Student’s two-tailed unpaired t test with Welch’s correction, ***p = 0.0009. Contingency table analyzed by Fisher’s exact test.
(I) GSEA comparing hSCLCs with NOTCH WT or silent (‘‘WT’’) versus ‘‘damaging’’ mutations from samples in (G). Normalized enrichment scores (NES) indicated.
For boxplots, the median and interquartile range are shown (the lower bar is the 25th percentile, the upper bar is the 75th percentile, and endpoints indicate
minimum and maximum values). See also Figure S6 and Table S4.
G
% expression log2(TPM+1)
HES1 MYC
YAP1 NEUROD1 ASCL1 MYC
0
2
4
6
8
10
0
2
4
6
8
0
5
100
20
40
60
80
ll
0
10
0
S02243
S01524
NCIH1092
NCIH69
S01578
NCIH2196
CORL47
NCIH1436
20
NCIH510
20
SHP77
S02382
S00356
S02290
COLO668
NCIH1618
HCI-4
NCIH209
NCIH146
40
NCIH1930
40
CORL95
DMS153
DMS79
NCIH1836
**
ns
DMS454
S02241
ASCL1
S02284
60
NCIH2081
S02297
60
NEUROD1
S02234
CORL88
p = 0.0014
NCIH1963
p = 0.0652
NCIH1105
S01366
S01297
S02360
80
S02248
80
S02065
S01861
S00050
NCIH1184
Predicted NOTCH status:
NCIH2029
S00472
S02342
NOTCH2 REST S00022
Expression Expression S02397
NCIH889
0
2
4
6
8
0
1
2
3
4
5
S02354
NCIH2066
HCI-12
DMS53
0
0
S00832
S02242
ASCL1
S01512
S02294
NCIH1876
S01248
% late in SCLC-A
S02350
Low
Low
S00838
High
High
S02352
20
S01453
20
S02295
S02120
S02293
S00836
40
S02353
S02351
40
S01728
wild-type or silent
S02285
S02378
****
S00827
****
S00035
S02163
60
S00825
S02287
S02322
60
p < 0.0001
HCI-3
S02244
S00830
p < 0.0001
S01792
S01242
S02328
80
S00831
S02249
S02194
80
S01698
S01864
S02347
MYC
S01563
MYCL
S00213
S02093
S02299
S02291
CORL24
non-damaging
S02139
HCC33
D
SCLC21H
SCLC22H
CORL279
NCIH2171
NCIH2227
NCIH1694
LU135
NCIH524
NCIH82
Expression Expression S01494
DMS273
NCIH446
HCI-20
S01873
NCIH1341
NCIH196
SW1271
HCI-27
damaging
NCIH841
ASCL1+
DMS114
HCI-30*
SBC5
HCI-11
S02298
Low
Low
NCIH1339
High
High
NCIH2286
NEUROD1 YAP1
S02246
HCI-4*
HCI-21
HCI-3*
HCI-39
MYC
HES1
REST
MYCL
NOTCH2
Day 3-5
Day 7-10
black = tumor
HCI-37
HCI-23
YAP1+
HCI-10
HCI-13* HCI-16 HCI-31
HCI-5* HCI-20
B
H
HCI-14
HCI-6
0
20
40
60
80
HCI-12
tSNE 2 AS
HCI-36
N C
EU L1
-20
20
R
HCI-34*
0
HCI-26
O
D
1
****
**
YA
NEUROD1+
P1
-20
% Day 7-10 expression
n = 1021
0
10
20
30
AS
3/21
N C
1 subtype
EU L1
*
3 subtypes
2 subtypes
R
MYC+
O
D
1
0
YA
tSNE 1
P1
0%
14%
AS
N C
20
EU L1
*
18/21
4
1 O
33% 1 subtype
D
**
3 subtypes
67% 2 subtypes
1
Cluster
MYC-
ns
5 YA
2
P1
0%
39%
61%
86%
6
3
AS
N C
EU L1
R
O
D
1
****
****
YA
P1
subtypes have distinct therapeutic vulnerabilities, this suggests d EXPERIMENTAL MODEL AND SUBJECT DETAILS
that dynamically evolving tumors represent a ‘‘moving therapeu- B Human SCLC Samples
tic target,’’ adopting unique therapeutic vulnerabilities as they B Clinical Details
progress. Given many failed clinical trials with targeted therapies B Cell Lines
in SCLC, we speculate that the reason why chemotherapy (a B Mice
‘‘blunt instrument’’) has remained the most effective therapeutic d METHOD DETAILS
option is likely due to its non-specific cytotoxicity in the multiple B Mouse Lung Tumor Initiation
subtypes of SCLC that evolve during progression. B In Vivo DAPT and DBZ
As clinical trials begin to assess biomarkers of SCLC subtype B MicroCT Analysis and Quantification
and potentially enroll patients based on these subtypes (Owoni- B Immunohistochemistry
koko et al., 2019; Poirier et al., 2020), it will be critical to assess B Primary Tumor Cell Isolation
subtype heterogeneity and anticipate tumor evolution to other B Quantification of Tumor Cell Morphology
subtypes. Multiple studies have implicated an increase in MYC B Live/Dead Assays
and a decrease in MYCL and ASCL1 in chemotherapy-resistant B Immunoblotting
mouse and human SCLC (Chalishazar et al., 2019; Farago et al., B PCR for Recombination Efficiency
2019; Huang et al., 2018a; Mollaoglu et al., 2017; Stewart et al., B MYC Overexpression and Virus Production
2020; Wagner et al., 2018) and have correlated high MYC with B DAPT Treatments In Vitro
shorter patient survival and more aggressive, drug-resistant B ChIP-seq in RPM Tumors
phenotypes (Carney et al., 1985; Gazdar et al., 1985; Johnson B Whole Genome Sequencing (WGS)
et al., 1986). Recent studies also demonstrate an increase in in- B Mouse Tumor and Timepoint Bulk RNA-seq
tratumoral transcriptional heterogeneity in chemo-resistant pa- B Bulk RNA-seq Data Analysis
tient samples (Stewart et al., 2020). Together, these findings sug- B Human Genomics Data Analysis
gest the provocative notion that therapy selects for and/or B CIBERSORT
promotes the latest stages of tumor cell progression identified B Single Cell RNA-seq Sample and Library Prep
here. We speculate that cancers in other tissues may also harbor B 10X Single Cell RNA-seq Data Processing
cells in transcriptionally dynamic states of progression, and d QUANTIFICATION AND STATISTICAL ANALYSIS
would potentially benefit from either more blunt, combinatorial,
or plasticity-directed therapies. SUPPLEMENTAL INFORMATION
Detailed methods are provided in the online version of this paper ACKNOWLEDGMENTS
and include the following:
We thank members of the Oliver Lab for technical assistance and mouse col-
d KEY RESOURCES TABLE ony management, R. Olsen, R. Dahlgren, and L. Houston for administrative
support, A. Andersen for editorial guidance, and B. Dalley, K. Rondem, and
d RESOURCE AVAILABILITY
C. Stubben for bioinformatics support. We are grateful to our late colleague
B Lead Contact Dr. Adi Gazdar for insightful discussions and inspiration for this work. We
B Materials Availability acknowledge support from the National Cancer Institute (NCI) of the NIH under
B Data and Code Availability award P30CA042014 to Huntsman Cancer Institute for the use of core facilities
Article
MD, USA
2Lymphoid Malignancies Branch, National Cancer Institute, National Institutes of Health, Bethesda, MD 20892, USA
3Department of Molecular Biology and Biochemistry, Simon Fraser University, Burnaby, BC V5A 1S6, Canada
4BostonGene, Waltham, MA 02453, USA
5Office of Intramural Research, Center for Information Technology, National Institutes of Health, Bethesda, MD 20892, USA
6British Columbia Cancer, Vancouver, BC V5Z 4E6, Canada
7Lead Contact
*Correspondence: lstaudt@mail.nih.gov
https://doi.org/10.1016/j.ccell.2020.03.015
SUMMARY
The development of precision medicine approaches for diffuse large B cell lymphoma (DLBCL) is confounded
by its pronounced genetic, phenotypic, and clinical heterogeneity. Recent multiplatform genomic studies re-
vealed the existence of genetic subtypes of DLBCL using clustering methodologies. Here, we describe an
algorithm that determines the probability that a patient’s lymphoma belongs to one of seven genetic
subtypes based on its genetic features. This classification reveals genetic similarities between these DLBCL
subtypes and various indolent and extranodal lymphoma types, suggesting a shared pathogenesis. These
genetic subtypes also have distinct gene expression profiles, immune microenvironments, and outcomes
following immunochemotherapy. Functional analysis of genetic subtype models highlights distinct vulnera-
bilities to targeted therapy, supporting the use of this classification in precision medicine trials.
INTRODUCTION Rosenwald et al., 2002). This COO methodology has proved use-
ful in understanding the varied responses of patients with diffuse
Initial progress toward a molecular diagnosis of DLBCL subtypes large B cell lymphoma (DLBCL) to targeted therapies such as
came with the advent of gene expression profiling, which was ibrutinib, an inhibitor of B cell receptor (BCR) signaling (Wilson
used to define two prominent ‘‘cell-of-origin’’ (COO) subtypes et al., 2015b). Nonetheless, the COO distinction does not fully
comprising 80%–85% of cases, termed germinal center B cell- account for the heterogeneous responses and outcomes
like (GCB) and activated B cell-like (ABC), with the remaining following either R-CHOP therapy or targeted therapy. This is
cases declared ‘‘unclassified’’ (Alizadeh et al., 2000; Rosenwald likely because gene expression profiling provides a phenotypic
et al., 2002). This classification accounted for some of the het- description of cancers rather than a genetic description that en-
erogeneity in the clinical outcome following R-CHOP (rituximab compasses tumor pathogenesis more directly.
plus cyclophosphamide, doxorubicin, vincristine, and predni- While recurrent genetic aberrations in individual genes have
sone) chemotherapy (Alizadeh et al., 2000; Lenz et al., 2008a; elucidated oncogenic mechanisms in DLBCL, progress toward
Significance
We describe a taxonomy of diffuse large B cell lymphoma (DLBCL) consisting of seven genetic subtypes and provide a prob-
abilistic method to classify a patient’s tumor within this taxonomy. Several DLBCL subtypes are genetically related to
distinct indolent lymphoma types, suggesting that these subtypes may arise from clinically occult precursors. We infer
oncogenic pathway activity and therapeutic vulnerabilities of the DLBCL subtypes using their genetic and gene expression
profiles supplemented by functional identification of essential genes using loss-of-function CRISPR/Cas9 screens of cell-
line models of the subtypes. To foster precision medicine studies of DLBCL, we have implemented our probabilistic algo-
rithm on a publicly accessible server.
Cancer Cell 37, 551–568, April 13, 2020 Published by Elsevier Inc. 551
a genetic classification of DLBCL tumors required the integration with these TP53 features, which we term ‘‘A53’’ (aneuploid with
of genomic data from multiple analytic platforms to identify TP53 inactivation). We also observed that mutations in TET2,
genes that were recurrently altered by mutations, translocations, P2RY8, and SGK1 were recurrently mutated among the geneti-
and/or copy-number alterations (Chapuy et al., 2018; Schmitz cally unassigned cases (10.1%, 6.9%, and 6.9% of cases,
et al., 2018). Mathematically distinct clustering methods were respectively), with the majority (54%) of SGK1 mutations trun-
used to assort DLBCL tumors into genetic subtypes that are cating the protein. TET2 mutations were significantly associated
characterized by genomic aberrations in subtype-specific hall- with truncating SGK1 mutations (p = 0.001) and with P2RY8 mu-
mark genes. The potential clinical utility of this genetic classifica- tations (p = 0.033), leading us to create a second new seed class
tion was evident by the association of the subtypes with based on these features, which we term ‘‘ST2’’ (SGK1 and TET2
outcome following R-CHOP therapy. mutated). Using the six seed classes defined above, the
Many clustering methods are limited by the necessity to place GenClass algorithm assigned 54% of the cases.
a tumor sample into no more than one subtype and by the fact LymphGen next develops a separate Bayesian predictor
that the subtype assignment of a particular tumor can vary model for each GenClass subtype, which determines the prob-
when different tumors are included during the clustering pro- ability that a tumor belongs to the subtype based on its genetic
cess. Such methods are therefore not appropriate in clinical features (Figure 1A). The algorithm defines subtype predictor
settings where molecular diagnoses are required in real time features that distinguish the subtype from all other cases
for individual tumors. We have therefore developed an algorithm (p % 0.001, Fisher’s exact test, prevalence >0.2) and uses
to classify an individual patient’s tumor based on the probability the prevalence of the feature in the subtype and its prevalence
of belonging to a particular genetic subtype, allowing for the pos- in other cases to estimate the likelihood that a tumor with that
sibility that the tumor may have acquired more than one genetic feature belongs to the subtype. These likelihood estimates are
program during its evolution. then used in Bayes formula to calculate the probability that an
individual tumor belongs to a subtype based on its constella-
RESULTS tion of genetic features. Thus, for each DLBCL tumor, Lymph-
Gen calculates six probabilities, one for each GenClass-defined
Development of the LymphGen Genetic Subtype subtype. We defined tumors with subtype probabilities
Classifier of >90% or 50%–90% as ‘‘core’’ or ‘‘extended’’ subtype mem-
We created the LymphGen algorithm to provide a probabilistic bers, respectively. Tumors that were core members of more
classification of a tumor from an individual patient into a genetic than one subtype were termed ‘‘genetically composite’’ (Fig-
subtype. We define a genetic subtype as a group of tumors that ure S1). In the NCI cohort, the LymphGen algorithm identified
is enriched for genetic aberrations in a set of subtype predictor 47.6% core cases, 9.8% extended cases, and 5.7% genetically
genes. These subtype predictor genes are identified by consid- composite cases (Figure 1B). Altogether 329 (63.1%) of the 574
ering each possible combination of genetic aberrations (i.e., mu- cases in the NCI cohort were classified, which is substantially
tations, copy-number alterations, or fusions) as a separate greater than the 46.6% classified previously (Schmitz et al.,
genetic ‘‘feature’’ and scoring a tumor as positive for a feature 2018) (Figures 1B and 1C). The inability of LymphGen to clas-
if one or more of its genetic aberrations is observed. LymphGen sify the remaining cases stemmed from three issues: some tu-
uses the presence or absence of each subtype predictor feature mors had a few features characteristic of one or more subtype
to provide a probability that a tumor belongs to the subtype. but not enough to be classified, some had unique features that
To implement LymphGen in DLBCL, we first needed to define were not recurrent in DLBCL, and others had very few genetic
the genetic subtypes to which a tumor could be assigned. For features at all.
this subtype discovery effort, we used the GenClass algorithm In the resulting genetic taxonomy, each of the DLBCL COO
(Schmitz et al., 2018), which begins with several ‘‘seed’’ tumor gene expression subgroups included multiple genetic subtypes,
subsets and iteratively sorts tumors into and out of these seeds, with ABC tumors enriched for MCD, GCB tumors enriched for
converging on a classification that maximizes the genetic EZB and ST2, and unclassified tumors enriched for BN2 (Fig-
distinctiveness of the resulting subtypes (Figure 1A). First, we ure 1D). Conversely, some genetic subtypes were largely
chose seeds representing the four previously identified genetic composed of tumors belonging to the same gene expression
subtypes: MCD (including MYD88L265P and CD79B mutations), subgroup (MCD, N1, EZB), while others comprised different
BN2 (including BCL6 translocations and NOTCH2 mutations), gene expression subgroups, with BN2, A53, and ST2 being the
N1 (including NOTCH1 mutations), and EZB (including EZH2 mu- most phenotypically diverse.
tations and BCL2 translocations). Among the remaining cases in
our cohort (hereafter ‘‘NCI cohort’’ Schmitz et al., 2018), TP53 Genetic Attributes of DLBCL Subtypes
was the most frequently mutated gene (25.2%) that was not To display the genetic composition of the subtypes, we selected
also significantly enriched in one of the previous subtypes. a set of genetic features that were significantly associated with a
TP53 inactivation has been previously associated with aneu- subtype (p % 0.01) and were present in >10% of the subtype
ploidy in DLBCL (Bea et al., 2004; Chapuy et al., 2018; Monti (Figures 2 and S2A). Many subtype-defining mutations are likely
et al., 2012). In the NCI cohort, tumors with a homozygous due to activation-induced deaminase-dependent somatic hy-
TP53 deletion (5.9%) or the combination of a heterozygous permutation (Schmitz et al., 2018), which in many cases
TP53 deletion and a TP53 mutation (8.7%) had the most aneu- produced truncating mutations in subtype-specific tumor-sup-
ploidy, as assessed by the number of gains and losses of chro- pressor genes (e.g., PRDM1, ETV6, TOX, HLA-A, HLA-B,
mosomal segments. We therefore formed a seed class of cases HLA-C in MCD, TNFRSF14 in EZB, and NFKBIA in ST2).
B C D
MYD88L265P and CD79B mutations, the genetic hallmarks of 72% of which also have a BCL6 translocation. Mutations target-
MCD, cooperatively activate nuclear factor kB (NF-kB) via the ing components of the BCR-dependent NF-kB pathway
My-T-BCR supercomplex involving MYD88, TLR9, and the (PRKCB, BCL10, TNFAIP3, TNIP1) are also prominent in BN2,
BCR (Phelan et al., 2018). MCD tumors also frequently delete suggesting that these tumors rely on this pathway for survival. In-
the CDKN2A tumor-suppressor locus, encoding the cell-cycle teractions with T and natural killer (NK) cells are potentially
inhibitors p16INK4A and p15INK4B as well as p19ARF, which stabi- compromised in BN2 by CD70 deletions. CCND3 mutations
lizes p53. The viability of MCD cells is likely sustained by BCL2, likely foster vigorous proliferation in BN2, as in Burkitt’s lym-
which is upregulated epigenetically and by copy number gain/ phoma (BL) (Schmitz et al., 2012).
amplification (Figure S2B). Another prominent theme in MCD tu- Epigenetic dysregulation is a defining attribute of EZB due to
mors is immune evasion (see below). inactivation of several epigenetic regulators (KMT2D, CREBBP,
BN2 is characterized by mutations that activate NOTCH2 or EP300, ARID1A, IRF8, MEF2B, EBF1) and mutational activation
inactivate the NOTCH antagonist SPEN in 50% of tumors, of EZH2 (Mlynarczyk et al., 2019; Pasqualucci and Dalla-Favera,
2018). Inactivation of the S1PR2/GNA13 pathway in EZB alters (Mintz et al., 2019). STAT6 mutations may modulate the ability of
GC B cell migration and signaling (Muppidi et al., 2014). Phos- TFH-derived interleukin-4 (IL-4) to promote plasma cell differenti-
phatidylinositol 3-kinase (PI3K) signaling in EZB is promoted by ation (Weinstein et al., 2016).
inactivating mutations and deletions of PTEN and MIR17HG ST2 is named for its recurrent SGK1 and TET2 mutations. ST2
amplification (Pfeifer et al., 2013). Recurrent REL amplification tumors acquire TET2 truncating mutations suggesting a tumor-
may deregulate EZB metabolism and growth, as in normal GC suppressor function, as in mouse GC B cell lymphomagenesis
B cells (Heise et al., 2014). (Dominguez et al., 2018). The majority of SGK1 mutations are
Several genetic lesions in EZB potential perturb their interac- truncating, suggesting a tumor-suppressor function that may
tion with T follicular helper (TFH) cells. Major histocompatibility involve PI3K signaling, since SGK1 is an AKT-family kinase
complex (MHC) class II expression and function in EZB may be (Di Cristofano, 2017). JAK/STAT signaling is likely promoted in
compromised by EZH2 activation (Ennishi et al., 2019b) and ST2 by inactivation of SOCS1 (Linossi and Nicholson, 2015),
inactivation of CIITA (Steidl et al., 2011) and HLA-DMB (Denzin inactivation of DUSP2 (Lu et al., 2015), and by known STAT3-
and Cresswell, 1995), potentially modulating interactions be- activating mutations (Y640F, D661Y) (Crescenzo et al., 2015). In-
tween lymphoma cells and TFH cells. The survival of EZB lym- activating mutations in ST2 targeting P2RY8 and its signaling
phoma cells that inactivate herpesvirus entry mediator mediator GNA13 prevent responses to S-geranylgeranyl-L-
(TNFRSF14) may be enhanced by TFH-mediated CD40 signaling glutathione, which spatially confines normal GC B cells and
B E F G
inhibits their AKT activity (Lu et al., 2019). Finally, some ST2 tu- Relationship of DLBCL Genetic Subtypes to Other
mors apparently activate NF-kB by inactivating IkBa (NFKBIA) Lymphoid Malignancies
(Baeuerle and Baltimore, 1988). While DLBCLs typically present clinically in lymph nodes and other
A53 is characterized by TP53 mutations and deletions. A53 immune tissues, primary extranodal lymphomas present as tu-
tumors also acquire homozygous deletions and mutations tar- mors involving various non-lymphoid organs. Primary extranodal
geting 53BP1 (TP53BP1), a DNA-damage sensor that pre- lymphomas frequently acquire MYD88L265P and/or CD79B muta-
vents aneuploidy (Celeste et al., 2002), consistent with the tions as well as other MCD-defining mutations (Figures 3A, 3B,
recurrent gains and losses of chromosome arms in A53. and S3B). Notably, in the NCI cohort, MCD DLBCL tumors spread
Some A53 abnormalities have been previously associated secondarily to extranodal sites in 30% of cases, and 46% of these
with ABC DLBCL, including: deletion of 6q, harboring the occurred at sites that can give rise to primary extranodal lym-
tumor suppressors TNFAIP3 and PRDM1; gain/amplification phomas—the central nervous system (CNS), vitreo-retina, testis,
of 3q (Lenz et al., 2008b); focal amplification of NFKBIZ (Nogai and breast—whereas other DLBCL subtypes spread to these sites
et al., 2013); amplification of CNPY3 (Phelan et al., 2018); and significantly less often (Figure 3C). Primary extranodal lymphomas
BCL2 amplification. Additional focal deletions target the often arise in the CNS, ocular vitreo-retina, and testis, all consid-
tumor suppressors p73 and ING1 (Tallen and Riabowol, ered ‘‘immune-privileged’’ sites because they tolerate allografts
2014). Finally, A53 tumors frequently delete or mutationally and permit only selective access by immune cells (Shechter
inactivate b2-microglobulin (B2M), providing a mechanism et al., 2013). In this regard it is notable that 72.5% of MCD tumors
of escape from immune surveillance (Challa-Malladi acquire homozygous deletions, truncating mutations, or translo-
et al., 2011). cations that could allow them to evade immune surveillance by
N1 is characterized by gain-of-function NOTCH1 mutations, several mechanisms including (1) reduced antigen presentation
similar to those in chronic lymphocytic leukemia and mantle due to MHC class I or TAP1 inactivation, (2) decreased T cell acti-
cell lymphoma. These tumors additionally acquire mutations tar- vation due to gene fusions that elevate expression of CD274 and
geting B cell differentiation regulators (ID3, BCOR) and IkB ki- PDCD1LG2, encoding PD-L1 and PD-L2, respectively, and (3)
nase b (IKBKB), including the V203I isoform that constitutively diminished NK activation due to CD58 inactivation (Challa-Malladi
activates NF-kB (Cardinez et al., 2018). et al., 2011) (Figures 3D and S3A).
D E
and GCB-4, defined as the subset of genes characteristically ex- their strong association with DHIT by gene set enrichment anal-
pressed in normal GC B cells that are expressed more highly in ysis (Figure 5C). A linear model combining these signatures was
BL than in DLBCL (Dave et al., 2006) (Table S4). Signatures of in- significantly correlated with DHIT (Figure 5D) and accounted for
termediate- and dark-zone GC B cells were also correlated 60.2% of DHIT variance among EZB cases within GCB.
(GCB-9 and GCB-10, respectively), suggesting that DHIT re- GCB patients with DHIT– tumors had a better survival than
flects dynamic changes in GC B cell differentiation. DHIT was those with DHIT+ tumors (Figure S6A), in part reflecting the
also correlated with signatures of MYC activity, notably enrichment among DHIT– cases of the prognostically favorable
MYCUp-4, consisting of genes that are induced by MYC and ST2, BN2, and A53 subtypes. Within the EZB subset of GCB,
are direct MYC binding targets (Zeller et al., 2006). DHIT also the survival of DHIT+ was significantly worse than DHIT–, which
correlated with a signature of adverse survival in DLBCL was not true in the non-EZB subset (Figure 5E), leading us to
(Prolif-6) that includes MYC and its target genes GNL3 and confine our investigation of DHIT to EZB.
NPM3 (Rosenwald et al., 2002). We next explored the association of genetic features with the
We therefore hypothesized that DHIT is a composite signature DHIT+ and DHIT– subsets of EZB GCB cases (Figure 5F). The
that reflects both GC B cell differentiation and MYC activity. We majority of EZB-defining genetic features were observed
used GCB-4 and MycUp-4 to represent these phenotypes, given comparably in these two subsets with the exception of GNA13
E F
mutations, which were more prevalent among DHIT+ than DHIT– p value = 0.004, see STAR Methods). Given the association of
cases (p = 0.025). In keeping with a role for MYC in DHIT+ cases, MYC abnormalities with DHIT+ EZB cases, we hereafter refer
MYC rearrangements, amplifications, and mutations were signif- to DHIT+ EZB as ‘‘EZB-MYC+’’ and DHIT– EZB as ‘‘EZB-MYC–’’.
icantly enriched in these tumors (p = 0.0079). Mutations or homo- Genes preferentially mutated in EZB-MYC+ were also
zygous deletions of TP53 were more than twice as prevalent in frequently mutated in BL whereas those preferentially mutated
DHIT+ than DHIT– cases, while the tumor suppressor DDX3X in EZB-MYC– were not (Figure 5G). However, some genetic ab-
was mutated in one-third of DHIT+ tumors but never in DHIT– tu- errations that define BL, such as ID3 and TCF3 mutations
mors. FOXO1, a transcription factor (TF) that is inactivated by (Schmitz et al., 2012), were not observed in EZB-MYC+. Thus,
PI3K signaling, was targeted by mutations more than 3 times the genetic program adopted by EZB-MYC+ is shared by BL,
as often in DHIT+ cases. Conversely, DHIT– cases were signifi- but these lymphomas are genetically distinct.
cantly enriched in mutations targeting the NF-kB regulators
A20 (TNFAIP3) and CARD11, as well as deletions of the tumor Phenotypic Distinctions between Genetic Subtypes
suppressor TP73. We were able to use the genetic data to create Gene expression signatures offer glimpses into tumor pheno-
a probabilistic model of EZB-MYC+ versus EZB-MYC– that could types that are differentially manifested in DLBCL genetic sub-
distinguish these subtypes effectively (Figure S6B; permutation types (Schmitz et al., 2018). We identified signatures whose
0.8 0.8
0.6 0.6
DLBCL
Normalized Signature Average
-0.8 -0.8
-1.0 -1.0
-2.0 -2.0
-3.0
Signature NFkB-10 p53Up-1 NotchUp-6 PI3KUp-1 JAK2Up-2 JAKUp-1 GCThUp-1 CD4T-2 CD8T-7 Treg-1 NK-3 MPhage-1 MPhage-2 DC-9 Stromal-1
Phenotype NF-κB p53 Notch PI3 kinase JAK2 kinase JAK1 kinase GC TFH CD4 CD8 Regulatory Natural killer Type 1 Type 2 Activated Fibrosis
activated activated activated actvated activated activated cells T cells T cells T cells cells macrophages macrophages dendritic cells (good prognosis)
average expression was significantly associated with one or 10 4) and their respective target genes. Although both IRF4
more genetic subtypes (Figure 6 and Table S5). The subtypes and OCT2 promote GC B cell differentiation to plasma cells
differed with respect to various malignant attributes, with MCD (Hodson et al., 2016; Sciammas et al., 2006), MCD had low
and EZB-MYC+ highly expressing signatures of proliferation expression of a plasma cell signature, likely due to inactivation
and MYC activity, while N1 instead expressed a signature of of Blimp-1 (PRDM1), which is required for plasma cell
quiescence. MYC induces ribosome biogenesis (Dang, 2013) differentiation.
and, accordingly, EZB-MYC+ tumors highly expressed a ribo- NF-kB target genes were highly expressed in MCD but not
somal protein signature. Metabolic distinctions between the EZB, as expected (Davis et al., 2001), but were also high in
subtypes included high expression of glycolytic pathway en- BN2 and ST2. As expected, p53 target genes were lowest in
zymes in ST2, consistent with a Warburg effect, and high expres- A53 tumors, and NOTCH target genes were significantly upregu-
sion of lipid synthetic enzymes in EZB-MYC+. lated in N1 and BN2. ST2 expressed a signature of PI3K signaling
The subtypes appeared to derive from distinct B cell differen- highly, potentially due to SGK1 inactivation, as well as a signa-
tiation stages, with a signature of GC B cells characterizing EZB ture of JAK2 signaling, consistent with SOCS1 and DUSP2
and ST2. EZB-MYC– tumors resembled GC light-zone cells inactivation.
whereas EZB-MYC+ tumors resembled intermediate-zone cells The tumor microenvironments of the subtypes were strikingly
(Milpied et al., 2018), which are likely generated from light-zone discordant: N1 expressed signatures of multiple immune cell
cells by MYC expression (Dominguez-Sola et al., 2012). Genes types while A53 and EZB-MYC+ tumors were relatively low for
repressed by BCL6 in the GC were lowest in EZB-MYC+ tumors, all immune signatures. T cell signatures were selectively low in
and genes transactivated by another GC TF, TCF3, were highest MCD, potentially stemming from defective antigen presentation
in EZB and ST2 tumors. due to genetic abnormalities or to their high IL10 expression rela-
MCD and N1 lacked GC signature expression, with N1 instead tive to other DLBCLs (p = 1.7 3 10 11) (Mittal and Roche, 2015).
expressing a memory B cell signature (Suan et al., 2017), sug- A GC TFH signature was highest in EZB-MYC– tumors, consistent
gesting a post-GC origin. MCD tumors had high expression of with their similarity to GC light-zone cells, but was lower in EZB-
the TFs IRF4 (p = 5.1 3 10 7) and OCT2 (POU2F2) (p = 1.3 3 MYC+ and ST2, despite their GC derivation. The stromal-1
F G
Our investigation provides insight into the mechanisms by that endow the B cell precursor with the hallmarks of cancer.
which tumors in a DLBCL genetic subtype acquire a shared ge- Support for this model comes from the observation that the ge-
netic program (Figure 8B). In one model, the epigenetic nature of netic subtypes differ in the expression of B cell differentiation sig-
a subtype’s cell of origin necessitates certain oncogenic events natures. Alternatively, a precursor B cell may randomly acquire a
B C D
Figure 8. Implications of the DLBCL Genetic Subtypes for Pathogenesis and Therapy
(A) Summary of the relationship between DLBCL COO subgroups and the genetic subtypes (left). The genetic themes, phenotypic attributes, clinical correlates,
and treatment implications of each subtype are shown at right. Prevalences were estimated using the NCI cohort, adjusting for a population-based distribution of
COO subgroups (see STAR Methods). dep., dependent; FDC, follicular dendritic cell; LZ, light zone; IZ, intermediate zone.
(B) Models of selection for shared genetic features in DLBCL subtypes.
(C) Models accounting for genetic attributes shared by DLBCL genetic subtypes and indolent NHLs.
(D) Model of EZB-MYC+ and EZB-MYC– evolution.
‘‘founder’’ genetic lesion, the nature of which dictates the subse- resistance of such genetically composite lymphomas is dictated
quent selection of secondary genetic lesions. For example, MYC largely by one of its genetic programs or is influenced by each
overexpression kills normal cells unless they also have lesions program.
that prevent cell death, such as the BCL2 translocations that Several of the DLBCL genetic subtypes have intriguing similar-
occur in the EZB-MYC+ subtype (Evan et al., 1992). Our proba- ities to more indolent lymphoma types: BN2 resembles MZLs,
bilistic approach raises a third, hybrid possibility. A substantial EZB resembles FL, and ST2 resembles both NLPHL and
subset of DLBCL tumors (5.7%) had a high probability of THRLBCL. Three models could account for these genetic rela-
belonging to more than one genetic subtype. This suggests a tionships (Figure 8C). A ‘‘direct evolution’’ model suggests that
model in which one genetic program is adopted by a tumor some DLBCL patients have a concurrent but undiagnosed low-
initially and a second is subsequently acquired because it con- grade malignancy that acquires additional genetic lesions, trans-
fers an additional selective advantage (Figure 8B). Future work forming it into DLBCL. Consistent with this model, pathologists
will be needed to understand whether therapeutic sensitivity or recognize histologically ‘‘composite lymphomas’’ that have, at
B Combination Features
Received: October 19, 2019
B LymphGen Methodology
Revised: January 3, 2020
B Measures of Feature Significance Accepted: March 16, 2020
B Gene List Selection Published: April 13, 2020
B Feature Selection within a Gene
B Hierarchical Feature Selection within a Gene REFERENCES
B Example 1: ETV6 in MCD
Agarwal, R., Chan, Y.C., Tam, C.S., Hunter, T., Vassiliadis, D., Teh, C.E.,
B Example 2: IRF4 in MCD
Thijssen, R., Yeh, P., Wong, S.Q., Ftouni, S., et al. (2019). Dynamic molecular
B Single-Class Sample Prediction monitoring reveals that SWI-SNF mutations mediate resistance to ibrutinib
B Example 1b: ETV6 in MCD plus venetoclax in mantle cell lymphoma. Nat. Med. 25, 119–129.
d EXAMPLE 2B: IRF4 IN MCD Alizadeh, A.A., Eisen, M.B., Davis, R.E., Ma, C., Lossos, I.S., Rosenwald, A.,
B Combining Models to Generate Final Sample Call Boldrick, J.C., Sabet, H., Tran, T., Yu, X., et al. (2000). Distinct types of diffuse
B Application of LymphGen to Imperfect Data large B cell lymphoma identified by gene expression profiling. Nature 403,
B Evaluation of Model Performance on Subclasses of 503–511.
Features Amin, N.A., Seymour, E., Saiya-Cork, K., Parkin, B., Shedden, K., and Malek,
B Prediction on Validation Sets S.N. (2016). A quantitative analysis of subclonal and clonal gene mutations
before and after therapy in chronic lymphocytic leukemia. Clin. Cancer Res.
B Model Verification via Gene Cross-Validation
+ – 22, 4525–4535.
B Genetic Prediction of EZB-MYC and EZB-MYC
Baeuerle, P.A., and Baltimore, D. (1988). I kappa B: a specific inhibitor of the
B Phosphoprotein Analysis of BCR Signaling
NF-kappa B transcription factor. Science 242, 540–546.
B CRISPR Screens
Bea, S., Colomo, L., Lopez-Guillermo, A., Salaverria, I., Puig, X., Pinyol, M.,
B Analysis of Tumor Suppressor Genes
Rives, S., Montserrat, E., and Campo, E. (2004). Clinicopathologic significance
B Mouse Xenograft Experiments and prognostic value of chromosomal imbalances in diffuse large B cell lym-
B Publicly Available Data Used in Study phomas. J. Clin. Oncol. 22, 3498–3506.
d QUANTIFICATION AND STATISTICAL ANALYSIS Bea, S., Valdes-Mas, R., Navarro, A., Salaverria, I., Martin-Garcia, D., Jares,
B Estimation of DLBCL Genetic Subtype Prevalence P., Gine, E., Pinyol, M., Royo, C., Nadeu, F., et al. (2013). Landscape of somatic
Monti, S., Chapuy, B., Takeyama, K., Rodig, S.J., Hao, Y., Yeda, K.T., Richter, J., Schlesner, M., Hoffmann, S., Kreuz, M., Leich, E., Burkhardt, B.,
Inguilizian, H., Mermel, C., Currie, T., Dogan, A., et al. (2012). Integrative anal- Rosolowski, M., Ammerpohl, O., Wagener, R., Bernhart, S.H., et al. (2012).
ysis reveals an outcome-associated and targetable pattern of p53 and cell cy- Recurrent mutation of the ID3 gene in Burkitt lymphoma identified by inte-
cle deregulation in diffuse large B cell lymphoma. Cancer Cell 22, 359–372. grated genome, exome and transcriptome sequencing. Nat. Genet. 44,
1316–1320.
Muppidi, J.R., Schmitz, R., Green, J.A., Xiao, W., Larsen, A.B., Braun, S.E., An,
J., Xu, Y., Rosenwald, A., Ott, G., et al. (2014). Loss of signalling via Galpha13 Rosenwald, A., Wright, G., Chan, W.C., Connors, J.M., Campo, E., Fisher, R.I.,
in germinal centre B cell-derived lymphoma. Nature 516, 254–258. Gascoyne, R.D., Muller-Hermelink, H.K., Smeland, E.B., Giltnane, J.M., et al.
(2002). The use of molecular profiling to predict survival after chemotherapy
Nakamura, T., Tateishi, K., Niwa, T., Matsushita, Y., Tamura, K., Kinoshita, M.,
for diffuse large-B cell lymphoma. N. Engl. J. Med. 346, 1937–1947.
Tanaka, K., Fukushima, S., Takami, H., Arita, H., et al. (2016). Recurrent muta-
tions of CD79B and MYD88 are the hallmark of primary central nervous system Rossi, D., Trifonov, V., Fangazio, M., Bruscaggin, A., Rasi, S., Spina, V., Monti,
lymphomas. Neuropathol. Appl. Neurobiol. 42, 279–290. S., Vaisitti, T., Arruga, F., Fama, R., et al. (2012). The coding genome of splenic
marginal zone lymphoma: activation of NOTCH2 and other pathways regu-
Ngo, V.N., Young, R.M., Schmitz, R., Jhavar, S., Xiao, W., Lim, K.H., lating marginal zone development. J. Exp. Med. 209, 1537–1551.
Kohlhammer, H., Xu, W., Yang, Y., Zhao, H., et al. (2011). Oncogenically active
Rui, L., Drennan, A.C., Ceribelli, M., Zhu, F., Wright, G.W., Huang, D.W., Xiao,
MYD88 mutations in human lymphoma. Nature 470, 115–119.
W., Li, Y., Grindle, K.M., Lu, L., et al. (2016). Epigenetic gene regulation by
Nogai, H., Wenzel, S.S., Hailfinger, S., Grau, M., Kaergel, E., Seitz, V., Wollert- Janus kinase 1 in diffuse large B cell lymphoma. Proc. Natl. Acad. Sci. U S A
Wulf, B., Pfeifer, M., Wolf, A., Frick, M., et al. (2013). IkappaB-zeta controls the 113, E7260–E7267.
constitutive NF-kappaB target gene network and survival of ABC DLBCL.
Saito, T., Chiba, S., Ichikawa, M., Kunisato, A., Asai, T., Shimizu, K.,
Blood 122, 2242–2250.
Yamaguchi, T., Yamamoto, G., Seo, S., Kumano, K., et al. (2003). Notch2 is
Okosun, J., Bodor, C., Wang, J., Araf, S., Yang, C.Y., Pan, C., Boller, S., preferentially expressed in mature B cells and indispensable for marginal
Cittaro, D., Bozek, M., Iqbal, S., et al. (2014). Integrated genomic analysis iden- zone B lineage development. Immunity 18, 675–685.
tifies recurrent mutations and evolution patterns driving the initiation and pro-
Scherer, F., Kurtz, D.M., Newman, A.M., Stehr, H., Craig, A.F., Esfahani, M.S.,
gression of follicular lymphoma. Nat. Genet. 46, 176–181.
Lovejoy, A.F., Chabon, J.J., Klass, D.M., Liu, C.L., et al. (2016). Distinct biolog-
Okosun, J., Wolfson, R.L., Wang, J., Araf, S., Wilkins, L., Castellano, B.M., ical subtypes and patterns of genome evolution in lymphoma revealed by
Escudero-Ibarz, L., Al Seraihi, A.F., Richter, J., Bernhart, S.H., et al. (2016). circulating tumor DNA. Sci. Transl. Med. 8, 364ra155.
Recurrent mTORC1-activating RRAGC mutations in follicular lymphoma.
Schmitz, R., Wright, G.W., Huang, D.W., Johnson, C.A., Phelan, J.D., Wang,
Nat. Genet. 48, 183–188.
J.Q., Roulland, S., Kasbekar, M., Young, R.M., Shaffer, A.L., et al. (2018).
Parry, M., Rose-Zerilli, M.J., Gibson, J., Ennis, S., Walewska, R., Forster, J., Genetics and pathogenesis of diffuse large B cell lymphoma. N. Engl. J.
Parker, H., Davis, Z., Gardiner, A., Collins, A., et al. (2013). Whole exome Med. 378, 1396–1407.
sequencing identifies novel recurrently mutated genes in patients with splenic Schmitz, R., Young, R.M., Ceribelli, M., Jhavar, S., Xiao, W., Zhang, M., Wright,
marginal zone lymphoma. PLoS One 8, e83244. G., Shaffer, A.L., Hodson, D.J., Buras, E., et al. (2012). Burkitt lymphoma path-
Parry, M., Rose-Zerilli, M.J., Ljungstrom, V., Gibson, J., Wang, J., Walewska, ogenesis and therapeutic targets from structural and functional genomics.
R., Parker, H., Parker, A., Davis, Z., Gardiner, A., et al. (2015). Genetics and Nature 490, 116–120.
Singh, M., Jackson, K.J.L., Wang, J.J., Schofield, P., Field, M.A., Koppstein, Yonese, I., Takase, H., Yoshimori, M., Onozawa, E., Tsuzura, A., Miki, T.,
D., Peters, T.J., Burnett, D.L., Rizzetto, S., Nevoltris, D., et al. (2020). Mochizuki, M., Miura, O., and Arai, A. (2019). CD79B mutations in primary vit-
Lymphoma driver mutations in the pathogenic evolution of an iconic human reoretinal lymphoma: diagnostic and prognostic potential. Eur. J. Haematol.
autoantibody. Cell 180, 878–894.e19. 102, 191–196.
Spina, V., Khiabanian, H., Messina, M., Monti, S., Cascione, L., Bruscaggin, A., Young, R.M., Phelan, J.D., Wilson, W.H., and Staudt, L.M. (2019). Pathogenic
Spaccarotella, E., Holmes, A.B., Arcaini, L., Lucioni, M., et al. (2016). The ge- B cell receptor signaling in lymphoid malignancies: new insights to improve
netics of nodal marginal zone lymphoma. Blood 128, 1362–1373. treatment. Immunological Rev. 291, 190–213.
Young, R.M., Wu, T., Schmitz, R., Dawood, M., Xiao, W., Phelan, J.D., Xu, W.,
Steidl, C., Shah, S.P., Woolcock, B.W., Rui, L., Kawahara, M., Farinha, P.,
Menard, L., Meffre, E., Chan, W.C., et al. (2015). Survival of human lymphoma
Johnson, N.A., Zhao, Y., Telenius, A., Neriah, S.B., et al. (2011). MHC class
cells requires B cell receptor engagement by self-antigens. Proc. Natl. Acad.
II transactivator CIITA is a recurrent gene fusion partner in lymphoid cancers.
Sci. U S A 112, 13447–13454.
Nature 471, 377–381.
Zamo, A., Pischimarov, J., Schlesner, M., Rosenstiel, P., Bomben, R., Horn,
Suan, D., Krautler, N.J., Maag, J.L.V., Butt, D., Bourne, K., Hermes, J.R.,
H., Grieb, T., Nedeva, T., Lopez, C., Haake, A., et al. (2018). Differences be-
Avery, D.T., Young, C., Statham, A., Elliott, M., et al. (2017). CCR6 defines
tween BCL2-break positive and negative follicular lymphoma unraveled by
memory B cell precursors in mouse and human germinal centers, revealing
whole-exome sequencing. Leukemia 32, 685–693.
light-zone location and predominant low antigen affinity. Immunity 47, 1142–
1153.e4. Zeller, K.I., Zhao, X., Lee, C.W., Chiu, K.P., Yao, F., Yustein, J.T., Ooi, H.S.,
Orlov, Y.L., Shahab, A., Yong, H.C., et al. (2006). Global mapping of c-Myc
Suehara, Y., Sakata-Yanagimoto, M., Hattori, K., Nanmoku, T., Itoh, T., Kaji,
binding sites and target gene networks in human B cells. Proc. Natl. Acad.
D., Yamamoto, G., Abe, Y., Narita, K., Takeuchi, M., et al. (2018). Liquid biopsy
Sci. U S A 103, 17834–17839.
for the identification of intravascular large B cell lymphoma. Haematologica
103, e241–e244. Zhang, J., Jima, D., Moffitt, A.B., Liu, Q., Czader, M., Hsi, E.D., Fedoriw, Y.,
Dunphy, C.H., Richards, K.L., Gill, J.I., et al. (2014). The genomic landscape
Tallen, G., and Riabowol, K. (2014). Keep-ING balance: tumor suppression by
of mantle cell lymphoma is related to the epigenetically determined chromatin
epigenetic regulation. FEBS Lett. 588, 2728–2742.
state of normal B cells. Blood 123, 2988–2996.
Timens, W., Visser, L., and Poppema, S. (1986). Nodular lymphocyte predom- Zhou, X.A., Louissaint, A., Jr., Wenzel, A., Yang, J., Martinez-Escala, M.E.,
inance type of Hodgkin’s disease is a germinal center lymphoma. Lab. Invest. Moy, A.P., Morgan, E.A., Paxton, C.N., Hong, B., Andersen, E.F., et al.
54, 457–461. (2018a). Genomic analyses identify recurrent alterations in immune evasion
Tsukamoto, T., Nakano, M., Sato, R., Adachi, H., Kiyota, M., Kawata, E., genes in diffuse large B cell lymphoma, leg type. J. Invest. Dermatol. 138,
Uoshima, N., Yasukawa, S., Chinen, Y., Mizutani, S., et al. (2017). High-risk 2365–2376.
follicular lymphomas harbour more somatic mutations including those in the Zhou, Y., Liu, W., Xu, Z., Zhu, H., Xiao, D., Su, W., Zeng, R., Feng, Y., Duan, Y.,
AID-motif. Sci. Rep. 7, 14039. Zhou, J., and Zhong, M. (2018b). Analysis of genomic alteration in primary cen-
Vater, I., Montesinos-Rongen, M., Schlesner, M., Haake, A., Purschke, F., tral nervous system lymphoma and the expression of some related genes.
Sprute, R., Mettenmeyer, N., Nazzal, I., Nagel, I., Gutwein, J., et al. (2015). Neoplasia 20, 1059–1069.
Article
*Correspondence: ddenardo@wustl.edu
https://doi.org/10.1016/j.ccell.2020.02.008
SUMMARY
Here, we utilized spontaneous models of pancreatic and lung cancer to examine how neoantigenicity shapes
tumor immunity and progression. As expected, neoantigen expression during lung adenocarcinoma devel-
opment leads to T cell-mediated immunity and disease restraint. By contrast, neoantigen expression in
pancreatic ductal adenocarcinoma (PDAC) results in exacerbation of a fibro-inflammatory microenvironment
that drives disease progression and metastasis. Pathogenic TH17 responses are responsible for this neoan-
tigen-induced tumor progression in PDAC. Underlying these divergent T cell responses in pancreas and lung
cancer are differences in infiltrating conventional dendritic cells (cDCs). Overcoming cDC deficiency in early-
stage PDAC leads to disease restraint, while restoration of cDC function in advanced PDAC restores tumor-
restraining immunity and enhances responsiveness to radiation therapy.
INTRODUCTION have revealed that many PDAC patients indeed harbor intratu-
moral T cells and potentially actionable neoantigens that can
Pancreatic ductal adenocarcinoma (PDAC) is notoriously resis- elicit T cell responses (Bailey et al., 2016a, 2016b; Balachandran
tant to immunotherapy, including cytokine therapy, adoptive et al., 2017; Cristescu et al., 2018; Poschke et al., 2016). To
T cell therapy and checkpoint blockade strategies (Brahmer develop therapies to revive T cell responses to neoantigens in
et al., 2012; Kunk et al., 2016; Royal et al., 2010). Failure of these PDAC, it is critical to understand how this endogenous T cell
therapies has been attributed to CD8+ T cell scarcity and pro- response becomes ineffectual.
found immunosuppression in the PDAC microenvironment The magnitude and persistence of a T cell response against a
(Beatty et al., 2015; Stromnes et al., 2014; Zhang et al., 2016). tumor is dependent on initial priming by antigen-presenting cells.
However, recent studies have challenged this paradigm and Conventional dendritic cells (cDCs) have been recognized as
Significance
T cell-directed immunotherapies have not been effective for the majority of pancreatic cancer patients, in part due to our
limited understanding of how T cell immunity is subverted in this disease. We sought to identify mechanisms for this failure
using spontaneous mouse models. We report that endogenous antigen-specific responses in pancreatic ductal adenocar-
cinoma are aberrant due to a scarcity of dendritic cells, which favors the expansion of tumor-promoting TH17 immunity.
Restoring conventional dendritic cells (cDCs) in pancreatic cancer can enhance CD8+ T cell and TH1 activity to ultimately
help control disease. These findings expand our understanding of T cell ineffectiveness in pancreatic cancer, and propose
combinatorial strategies to modulate cDCs in conjunction with existing therapies for pancreatic cancer and similar solid
malignancies.
Cancer Cell 37, 289–307, March 16, 2020 ª 2020 Elsevier Inc. 289
critical mediators of antigen-priming and T cell activity, with recognition, and central tolerance. Using KPC-OG mice and
Batf3/Irf8-dependent CD103+ CD24+ cDC1s being responsible cell lines derived from tumors (KP-OG cells), we found that
for CD8+ cytotoxic T lymphocyte (CTL) cross-priming and Irf4- PDAC cells express OVA and GFP, which can be repressed
dependent CD11b+ CD172a+ cDC2s being implicated in helper by administration of doxycycline (Figure 1A). Notably, similar
CD4+ T cell (TH) priming (Gardner and Ruffell, 2016). In addition to other lineage-tracing studies in this model (Rhim et al.,
to initial T cell priming, cDCs have been implicated in T cell- 2012), GFP and OVA were expressed concomitant with early
dependent tumor killing and response to immunotherapies (Bin- transformation and induction of metaplasia. As such, >95%
newies et al., 2019; de Mingo Pulido et al., 2018; Roberts et al., of cytokeratin 19+ (CK19) ductal cells co-expressed GFP (Fig-
2016; Salmon et al., 2016; Spranger et al., 2017). Studies on ure 1B). Tumor lines derived from KPC-OG mice express major
antigen-presenting cells (APCs) in PDAC models have focused histocompatibility complex 1 (MHCI) at levels equivalent to
on tolerogenic subsets (Barilla et al., 2019; Bellone et al., 2014; those of traditional KPC mice (Figure S1A). To assess functional
Jang et al., 2017; Ochi et al., 2012). Presently, more granular antigen presentation by KPC-OG tumor cells, we performed
studies of DC subsets in the PDAC context are needed to distin- T cell killing assays ex vivo using OVA-specific OT-I CD8+
guish the differential impact of cDCs from monocytic and other T cells and found that T cells both recognized and killed
inflammatory APC subsets. KPC-OG-derived cells (Figures 1C and S1B). To verify that
Studies in cancers of different etiologies have shown that neo- endogenous antigen-specific T cells generated in OG+ mice
antigen-directed immunity can be subverted by diverse mecha- were not subjected to central tolerance prior to tumorigenesis,
nisms (DuPage et al., 2011, 2012; Gubin et al., 2014; Schietinger we vaccinated tumor-free p48-Cre;Trp53fl/fl;OG (PC-OG) mice
et al., 2016); mechanisms of immune evasion in PDAC are thus with OVA and observed a clear enrichment of dextramer+
worth elucidating. However, existing genetic models for PDAC OVA-specific CD8+ T cells in peripheral blood and draining
have not been amenable to study the heterogeneous interac- lymph nodes (Figure S1C). Furthermore, we implanted KPC-
tions between developing tumors and the host adaptive immu- OG-derived tumor cells into PC-OG or PC (control) littermates
nity due to a dearth of tumor-specific neoepitopes (Evans in the presence or absence of doxycycline. We observed that
et al., 2016). Transplanted PDAC models are constrained by grafted antigen-positive KP-OG cells grew equally slowly in
their lack of stroma and a very distinct inflammatory/immunized PC-OG and PC mice and that doxycycline repression of OVA
milieu upon tumor grafting, which can mask de novo immune expression led to tumor progression (Figure S1D). Together,
responses (Spear et al., 2019). In this study, we sought to deter- these data suggest that p48-Cre-driven OVA neoantigen in
mine how antigen-specific anti-tumor immunity becomes dysre- developing pancreatic tumors is presented and recognized by
gulated during progression of autochthonous PDAC. T cells not subject to thymic deletion. Notably, we found that
untreated KPC-OG mice had OVA-specific T cell density equiv-
RESULTS alent to that of mice with doxycycline withdrawal at birth, so we
did not treat with doxycycline for the remainder of the studies.
Neoantigen Expression during Pancreas Cancer To determine the impact of neoantigen expression during
Development Elicits Antigen-Specific Responses tumor initiation, we analyzed immune infiltrates in pre-
The ‘‘KPC’’ genetic mouse model of pancreas cancer has been cancerous lesions of KPC-OG or KPC mice. At the early stage
widely used because of its fidelity to human PDAC, notably acti- of tumorigenesis (6 weeks), we observed increased infiltration
vating mutations in Kras(G12D) and loss of Trp53, associated of CD8+ and CD4+ T cells and B cells in KPC-OG mice
desmoplasia, and inflammation (Hingorani et al., 2003; Morton compared with KPC littermates (Figures 1D and 1E). Assess-
et al., 2010). The model also mirrors human disease in its resis- ing antigen-specific responses, we observed increased
tance to both cytotoxic and immunotherapies (Beatty et al., numbers of OVA-dextramer+ CD8+ T cells in pre-malignant
2011; Gopinathan et al., 2015). However, KPC mice seldom pancreas, draining lymph nodes (dLNs), and spleens of
develop additional genetic alterations that drive prominent neo- early-stage KPC-OG compared with non-tumor-bearing PC-
antigens for studying immune surveillance and evasion (Evans OG mice (Figure 1F). Interestingly, compared with control an-
et al., 2016; Li et al., 2018). Studies that have assessed the imals, the dextramer+ CD8+ T cells in KPC-OG pancreas had
impact of antigenicity have utilized heterotopic or orthotopic higher Ki67+ frequency, but >30% of these cells were PD1hi/
tumor grafts that do not recapitulate de novo pancreas cancer TIM3hi, suggesting an early exhausted/dysfunctional pheno-
progression and thus may have very divergent immune contex- type (Figures S1E and S1F). This recapitulates observations
ture (Spear et al., 2019). To study antigen-specific responses in liver cancer models that found a reversibly dysfunctional
in the context of de novo pancreas cancer development, we phenotype (Schietinger et al., 2016). To ascertain whether
engineered a mouse designed to express a model neoantigen there was a systemic response toward tumor neoantigens,
chicken ovalbumin (OVA) bicistronically with green fluorescent we measured OVA immunoglobulin G (IgG) levels in serum.
protein (GFP) under the control of both Cre activation and tetra- We found total IgG1 to be similar between KPC-OG and
cycline repression (R26tm1(LSL-OG) or OG, Figure 1A). The KPC littermates, but OVA-specific IgG1 titers in KPC-OG
presence of neoepitopes for CD8+ T cell, CD4+ T cells, and B serum were markedly higher than in age-matched controls
cells allows us to study OVA-specific cellular and humoral immu- (Figure S1G). Together, these data suggest that there is an an-
nity raised during the course of tumor progression. These ‘‘OG’’ tigen-directed immune response in KPC-OG pancreas during
mice were crossed to KPC mice to create ‘‘KPC-OG’’ mice. initial stages of tumorigenesis. These observations emphasize
We first sought to test several key parameters of this model, that early pancreatic lesions do not grow in an immune-privi-
including antigen expression and presentation, CD8+ T cell leged environment.
p48/Cre 8
OFF DOX
STOP Kras G12D
KPC-OG
**
– + 4 **
Doxycyline
ON DOX
0
GFP (27 kDa) CTL : tumor
1
:1
:1
0
5:
10
20
ratio
β-actin (42 kDa) Ag non-specific T cells
OVA-specific OT-I
GFP CK19 DAPI
0.9
0.4 1.5
0.6
1.0
0.2
0.3 0.5
CD8a PAN-K DAPI CD4 PAN-K DAPI 0 B220 DAPI 0
0
* * ns * * * *
1.0 1.40 60 0.8 4 1.5 2.00
Treg (% of CD4+ T cells)
0.8
cells per g tissue (x 106)
0.6 1.90
0.5 1.05 0.6 3
per dLN (x 103)
1.40
per g (x 104)
per g (x 105)
40 1.0
0.4 1.05
0.70 0.4 2
0.3
0.70
0.2 20 0.5
0.35 0.2 1 0.35
0.1
0 0 0 0 0 0 0
pancreas panc-dLN spleen
Figure 1. Neoantigen Expression during Pancreas Cancer Development Elicits Antigen-Specific Responses
(A) Genetic loci for KPC-OG model and immunoblot for OVA and GFP expression in KPC-OG-derived cell line 72 h after doxycycline withdrawal. Representative of
three independent cell lines.
(B) Gross images (left) of pancreatic tissue at 6 weeks in KPC-OG mice on or off doxycycline, and immunofluorescence images (right) of pancreatic tumors at
36 weeks in KPC-OG mice on or off doxycycline. Scale bar, 100 mm.
(C) KPC-OG tumor-derived cell line depicting GFP fluorescence after 24-h co-culture with antigen-specific (OT-I TCR) or non-specific (C57Bl/6) activated CD8+
T cells (CTL) consistent across three independent cell lines. n = 3/group.
(D) Representative images and quantification of CD8+ T cells, CD4+ T cells, and B220+ B cells in 6-week-old KPC-OG and KPC mice. n = 10 mice/group. Scale
bar, 100 mm.
(E) Density of CD8+ T cells, CD4+ TH, CD4+ TREG, and CD19+CD22+ B cells measured by flow cytometry in early-stage KPC-OG and KPC mice. n = 5–8
mice/group.
(F) Density of OVA-specific CD8+ T cells in pancreas, pancreas dLNs, and spleen of early-stage KPC-OG and PC-OG mice. n = 3–8 mice/group.
Data were consistent across two independent experiments. Data are presented as mean ± SEM. For comparisons between two groups, Student’s two-tailed
t test was used. ns, not significant; *p < 0.05. See also Figure S1.
Neoantigen Expression Accelerates PDAC Progression We utilized the KPC-OG mouse (p53fl/+) and validated our find-
but Restrains Lung Adenocarcinomas ings in the KPPC-OG model (p53fl/fl), which exhibits faster
To evaluate the impact of neoantigen expression on pancreatic progression. Surprisingly, in both models we found that OG
cancer progression, we employed three distinct PDAC models. expression accelerated tumor progression at every stage of
50 * *
KPC
KPC
100 80
KPC
12
α-SMA+ (% of tissue)
lesions (% of tissue)
percent of lesions
80 40 9
60 5
60 ns 30 4
* 40
40 * 20 3
KPC-OG
KPC-OG
KPC-OG
20 2
20 10
1
0 0 0 0
KPC
KPC-OG
N3
oc
N2
N1
KPC
KPC-OG
en
nI
nI
nI
Pa
Pa
Pa
ad
H&E KPC (n=12) KPC (n=12) Sirius Red α-SMA
KPC-OG (n=12) KPC-OG (n=12)
G H
E F *
15 * 1.5 * 1.5 ns
2.5 * 100
14
KPC
KPC
high-grade (% of total)
100
per g tissue (x 106)
2.0 80
overall survival (%)
80
p48-Cre model
10 1.0 1.0
1.5 60 25% met
60
incidence
1.0 40
5 0.5 0.5 40
KPC-OG
KPC-OG
0.5 20
* 20
0 0 0 0 0 0 14
TAMs Granul- Mono- Eosino- Time (d) 40 60 80 100
KPC
KPC-OG
ocytes cytes phils
KPPC littermates (n=14)
62.5% met
*
KPC (n=5) KPC-OG (n=6) KPPC-OG (n=20) H&E H&E incidence
I J K
8 * 14.0 * 7.2 * 100
cells per g tissue (x 106)
percent of lesions
80
Pdx-CreER model
80 6 T 10.5 5.4
40% met
60 *
60 4 7.0 3.6
incidence 40
40 *
2 3.5 1.8 20 ns
20
KPL-OG
0 8 0 0 0 0
Time (d post-tam) 150 200 250
gr 1
gr 2
3
e
e
KPL (n=5) KPL (n=6)
ad
ad
ad
iKPC (n=10) **
gr
100% met KPL-OG (n=5) KPL-OG (n=6)
iKPC-OG (n=8)
incidence
H&E
L M N
GFP CK19 DAPI
100
GFP+ tumor cells
80 80 80
KPPC-OG
60
60 * 60
40
40 20 40
T*
KPL-OG
OT-I Rx
0
GFP DAPI
0 15 30 45 60 0
Time (d since Rx )
T
KPC-OG (n=8) untreated (n=18) untreated (n=6)
KPL-OG (n=6) OT-I + IL-2 Rx (n=10) GFP CK19 DAPI OT-I Rx (n=6)
Figure 2. Neoantigen Expression Accelerates PDAC Progression but Restrains Lung Adenocarcinomas
(A) Representative hematoxylin-eosin (H&E) images with quantification of lesions in early-stage KPC-OG and KPC mice. n = 12 mice/group.
(B) Lesion grades for early-stage KPC-OG and KPC pancreata. n = 12 mice/group.
(C) Sirius Red staining with quantification in KPC-OG and KPC mice. n = 12 mice/group.
(D) a-SMA staining with quantification in KPC-OG and KPC mice. n = 12 mice/group.
(E) Flow-cytometric quantification of various myeloid infiltrates in KPC-OG and KPC mice. n = 5–6 mice/group.
(F) Kaplan-Meier survival curve for KPPC-OG mice compared with KPPC littermates. n = 14–20 mice/group.
(G) Representative histology of late-stage KPC-OG and KPC tumors with quantification of high-grade tumors. n = 14–16 mice/group.
(H) Representative H&E images of late-stage KPC-OG and KPC livers with quantification of metastases. n = 14 mice/group.
(I) Kaplan-Meier survival curve for Pdx1-Cre-ER-driven iKPC-OG mice compared with iKPC littermates, with quantification of liver metastases. n = 8–10
mice/group.
(J) Density of CD8+ T cells in early-stage KPL-OG and KPL lung lesions. n = 5 mice/group.
(K) Representative H&E images of early-stage KPL-OG and KPL lung with quantification of lesion area and grade. Lesions demarcated by yellow line. n = 5
mice/group.
(L) Representative immunofluorescence images and quantification of GFP (green) expression in CK19+ (red) tumors of late-stage KPC-OG or KPL-OG tumors.
GFP-negative lesions in KPL-OG tumors are demarcated by yellow arrowhead. n = 6–8 mice/group.
(M) Overall survival since start of treatment for KPPC-OG mice undergoing OT-I adoptive transfer therapy compared with untreated controls. n = 10–18
mice/group.
(N) Representative immunofluorescence images and quantification of GFP (green) expression in CK19+ (red) tumors of KPPC-OG mice subjected to OT-I
adoptive transfer compared with untreated controls. n = 6 mice/group.
Data were consistent across two independent experiments. Data are presented as mean ± SEM. For comparisons between two groups, Student’s two-tailed
t test was used. For survival analyses, log-rank (Mantel-Cox) test was used. ns, not significant; *p < 0.05, **p < 0.01. Scale bars, 500 mm (A, C, D, G, H, and K) and
100 mm (L and N). See also Figure S2; Tables S1 and S2.
lesions (% of tissue)
* ns
60 42 24
control
40
control
control
18 *
α-SMA+ (% of tissue)
16 12
8 28
0 21
14 6
80 *
percent of lesions
60 ns 7
α-CD4
α-CD4
α-CD4
* * * 0 0
40 * * KPC KPC-OG KPC KPC-OG
control
α-CD19/α-B220
α-CD4
control
α-CD19/α-B220
α-CD4
20
D CD4+ TH cells
E CD4+ TH cells IL17A+ TH TNFα+ TH
RORγt+ GATA3+
KPC KPC-OG 2.8 * 2.20 * KPC KPC-OG 0.5
* 0.3 *
2.4% 12.6% 10% <1% 12% 4.5%
cells per g tissue (x 105)
2.1 0.4
1.65
PE-A :: GATA3
F 60 G 25 H 5
* * *
sirius red+ (% of tissue)
lesions (% of tissue)
α-SMA+ (% of tissue)
20 4
α-IL17A,F
α-IL17A,F
α-IL17A,F
40
15 3
10 2
20
5 1
H&E Sirius Red α-SMA
KPC 0 KPC 0 KPC 0
KPC-OG KPC-OG KPC-OG KPC-OG KPC-OG KPC-OG KPC-OG KPC-OG KPC-OG
+ α-CD4 + α-IL17A/α-IL17F + α-CD4 + α-IL17A/α-IL17F + α-CD4 + α-IL17A/α-IL17F
I
10 * 14.0 * 24 *
ns
pERK+ cells (% of total)
8
control
10.5 18
6
7.0 12
4
3.5 6
2
0 0 0
α-CD4
α-CD4
α-CD4
18 * 18 * 56 *
pERK+ cells (% of lesion)
ns
48
40
12 12
32
α-IL17A,F
α-IL17A,F
α-IL17A,F
24
6 6
16
8
0 0 0
p-ERK1/2 p-STAT3 (Y705) p-EGFR (Y1068)
KPC (n=10) KPC-OG (n=10) KPC-OG + α-CD4 (n=10) KPC-OG + α-IL17A/α-IL17F (n=8)
Figure 3. Pro-inflammatory CD4+ T Cell Responses Drive PDAC Acceleration in Response to Neoantigen
(A) Representative H&E images with quantification of pancreatic lesion area (top) and grade (bottom) in early-stage KPC-OG mice subjected to depletion of CD4+
T cells or CD19+B220+ B cells. n = 8–12 mice/group.
(legend continued on next page)
294 Cancer Cell 37, 289–307, March 16, 2020
Pathogenic CD4+ T Cell Responses Drive PDAC cDCs Are Fewer and Less Functional in PDAC Compared
Acceleration in Response to Neoantigen with Lung Cancer
Our data suggest that neoantigen expression leads to adap- We sought to determine the cellular origins for differences in
tive immune responses that surprisingly drive tumor progres- response to neoantigenicity in the lung and pancreas. We per-
sion. Previous studies have shown that CD4+ T cells or acti- formed immune profiling of major innate immune cell subsets
vated B cells can drive pathogenic inflammation and in both early- and late-stage lung and pancreas tumor tissues
accelerate PDAC progression (Barilla et al., 2019; Gunderson and found that cDCs were among the most divergent (Figures
et al., 2016; McAllister et al., 2014; Pylayeva-Gupta et al., 4A, S4A, and S4B). In early pre-malignant stages, we observed
2016; Zhang et al., 2014). Thus, we evaluated whether CD4+ that CD103+ CD24hi cDC1s were 10-fold fewer and CD172a+
T or B cells were critical for the early-stage disease progres- CD11b+ cDC2s were 4-fold fewer in the pancreas when
sion observed in OG+ mice. We found that while B cell deple- compared with lung. The disparity in cDC1s was magnified at
tion did not alter tumor progression, CD4+ T cell depletion led later stages, with cDC1s in PDAC being 79-fold less than in
to a decrease in pre-malignant disease burden and PANIN counterpart lung adenocarcinomas (Figure 4B). These observa-
grade, reduced collagen density, and decreased a-SMA+ tions were not affected by OG expression. We also observed
fibroblast accumulation (Figures 3A–3C, S3A, and S3B). These fewer migratory CD103+ cDC1s and CD11b+ cDC2s in pancreas
data suggest CD4+ T cells accelerate pancreatic neoplasia in dLNs of KPC-OG mice when compared with lung-dLNs of KPL-
response to neoantigen expression. We next evaluated the OG mice, but no major difference in resident DC populations
polarization of CD4+ T cell responses in OG+ mice and (Figures 4C and S4C). To determine possible differences in
observed higher numbers of pancreas-infiltrating ROR-gt+ cDC localization between the pancreas and lung TME, we trans-
and GATA3+ CD4+ TH cells, and more TH cells producing IL- planted irradiated KPC and KPL mice with Zbtb46GFP and
17A, TNF-a, IL-4, and IL-10, consistent with dominant TH17 Snx22GFP bone marrow. The Zbtb46GFP reporter model marks
and TH2 responses (Figures 3D, 3E, and S3C–S3E). By all cDCs (Satpathy et al., 2012) while the Snx22GFP model labels
contrast, we did not see increased frequency of Tbet+ or inter- Batf3-dependent cDC1s (Bra €hler et al., 2018) (Figure S4D). Using
feron-g (IFN-g)-producing TH cells in KPC-OG (Figure S3F). immunohistochemistry to quantify GFP+ cells, we observed pat-
Surprisingly, partial depletion of FOXP3+ TREGS did not affect terns similar to our flow-cytometric results with markedly more
disease progression in KPC-OG tumors (Figures S3G and Zbtb46-GFP+ cDCs and Snx22-GFP+ cDC1s in lung when
S3H). To determine the function of this enhanced TH17 signa- compared with pancreas, both in pre-malignant tissues and
ture observed in early lesions, we depleted cytokines neces- late-stage cancer (Figure 4D). We next analyzed cDC localization
sary for their activity. Upon neutralizing pro-inflammatory IL- and found that both Zbtb46-GFP+ cDCs and Snx22-GFP+
17 signaling, we observed lower disease burden and patho- cDC1s localized close to lung tumor cells (>60% were within
logical fibrosis (Figures 3F–3H). Correspondingly, we found 5 mm or less), while in PDAC cDCs were more distant from tumor
the increased expression of phosphorylated (p)-ERK, p- nests (Figure 4E). Analysis of lung and pancreatic tumoral cDCs
STAT3, and p-EGFR signaling observed in KPC-OG lesions found that cDC1s and cDC2s had lower co-stimulatory and
was attenuated upon CD4+ T cell depletion or IL-17 neutraliza- maturation markers in PDAC (Figure S4E), and pancreatic
tion (Figure 3I). We next compared CD4+ T cell polarization in cDC1s were less functional at antigen presentation in ex vivo as-
early-stage pancreatic and lung lesions and found higher fre- says (Figure S4F). The higher cDC1 density found in lung cancer
quency of GATA3+ and ROR-gt+ TH cells in KPC-OG pancreas also paralleled higher OVA-specific CD8+ T cell density, suggest-
compared with KPL-OG lung tumors. By contrast, KPL-OG ing a critical role in antigen-specific T cell immunity (Figure 4F).
lung-infiltrating TH cells were more TH1-skewed, with This was further supported by the observation that depleting
increased frequency of Tbet+ and IFN-g-producing cells (Fig- cDC1s prior to KPL-OG lung tumor initiation via Batf3 / bone
ures S3I and S3J). Overall, these data indicate that, in contrast marrow transplant results in drastically reduced CD8+ T cell infil-
to lung, pancreatic neoantigen expression results in enhanced tration (Figure S4G).
pathogenic TH17 responses that can facilitate progression To determine whether these observations in mouse models
(Alam et al., 2015; McAllister et al., 2014; Zhang et al., held true in human pancreatic tumors, we analyzed cDC density
2016, 2018). in human PDAC tissue by mass cytometry and publicly available
(B) Sirius red staining with quantification in early-stage KPC-OG mice subjected to indicated depletions. n = 8–12 mice/group.
(C) a-SMA staining with quantification in early-stage KPC-OG mice subjected to indicated depletions. n = 8–12 mice/group.
(D) Representative flow-cytometry plots of RORgt and GATA3 bias in TH cells of KPC-OG and KPC tumors, with cellular density of RORgt+ TH17 and GATA3+ TH2
cells quantified. n = 3–6 mice/group.
(E) Representative flow-cytometry plots of IL-17A and TNF-a expression in TH cells of KPC-OG and KPC tumors, with cellular density quantified. n = 3–6
mice/group.
(F) Representative H&E image with quantification of pancreatic lesion area of early-stage KPC-OG mice subjected to IL-17A and IL-17F neutralization. n = 8–10
mice/group.
(G) Representative Sirius red staining of KPC-OG mice subjected to IL-17A and IL-17F neutralization, with quantification. n = 8–10 mice/group.
(H) Representative a-SMA staining of KPC-OG mice subjected to IL-17A and IL-17F neutralization, with quantification. n = 8–10 mice/group.
(I) Representative p-ERK1/2, p-STAT3, and p-EGFR immunohistochemistry staining in KPC-OG mice subjected to indicated depletions, with quantification over
overall tissue area and lesion area. n = 8–10 mice/group.
Data were consistent across two independent experiments and pooled. Data are presented as mean ± SEM. For comparisons between two groups, Student’s
two-tailed t test was used. ns, not significant; *p < 0.05, **p < 0.01. Scale bars, 500 mm. See also Figure S3.
KPL-OG
** * 0.70 ** 0.6 * 3 * 8 *
1.5 ns 3 ns
0.53 6
1.0 2 0.4 2
TAMs
eosinophils 0.35 4
granulocytes ns ns
0.5 ns 1 0.2 1
monocytes * 0.17 2
CD103 cDC1
CD11b cDC2 0 0 0 0 0 0
DN cDC early stage late stage late stage
mo-DC KPC early KPC KPL
KPL early
mean (log10) cells/g KPC-OG early KPL-OG early KPC-OG KPL-OG KPC-OG KPL-OG
3.5 7.0
D E
2.0 60 80
4 ** **
KPC
Snx22-GFP+ cells
60
Zbtb46-GFP+ cells
Snx22-GFP+ cells
3 1.5 40
2 1.0 ** 40 **
* ** 20
K-S: 0.648 20 K-S: 0.735
1 0.5
KPL
0 0
0 0 10 30 50 70 90 10 30 50 70 90
distance from lesion (μm) distance from lesion (μm)
F G *
H
6
** **
early KPL-OG 4.2 * 1.40 * 2 2
OVA-specific CD8+ T cells
late KPL-OG
t-SNE dim 2
5 1.05 1 1
2.8
Z-score
0.70 0
0
4 early KPC-OG 1.4
late KPC-OG 0.35 -1
macrophages
neutrophils
-1
3 0 CD45 immune cells; n=11
+
-2
3 4 5 6 7
cDC1
cDC2
t-SNE dim 1
PAAD
LUAD
cDC1 cells per g tissue (Log10)
(177)
(230)
KPC-OG KPL-OG
early
in late stage tumors
5.25
per dLN (x 103)
1.8 KPC-Z 24 60
early
21.5 KPL-Z *
16 * 40 3.50
32.3 KPC-Z
8 20 1.75
57.1 KPL-Z
0.2 uninv LN
control
0 0 0
migr migr migr migr
0 ZsGreen CD103+ CD11b+ CD103+ CD11b+
Figure 4. cDCs Are Fewer and Less Functional in PDAC Compared with Lung Cancer
(A) Heatmap depicting mean density (log scale) of major myeloid cell infiltrates in advanced KPC-OG pancreatic and KPL-OG lung tumors. n = 5–10 mice/group.
(B) CD103+ cDC1 and CD11b+ cDC2 density in pancreas and lung tumors, at (left) early stage and (right) late stage. n = 5–10 mice/group.
(C) Migratory cDC1 and cDC2 density in respective draining lymph nodes of late-stage KPC-OG pancreatic tumors and KPL-OG lung tumors. n = 7 mice/group.
(D) Immunohistochemistry for tumor cytokeratin (CK7/19) expression, and Zbtb46-GFP+ (pink) cDCs in late-stage KPC or KPL bone marrow chimeras. Right:
Zbtb46-GFP+ cDC density in non-tumor (WT) tissue and late-stage tumors. Far right: Snx22-GFP+ cDC1 density in WT and late-stage tumors. n = 4–6 mice/group.
Scale bar, 100 mm.
(E) Frequency distribution of Zbtb46-GFP+ cDC and Snx22-GFP+ cDC1 proximity to nearest CK7/19+ tumor cell. n = 3–5 mice/group.
(F) Tumoral cDC1 density (log scale) plotted against OVA-specific CD8+ T cell density (log scale) across tissue and stage. Right: OVA-specific CD8+ T cell density
in early- and late-stage tumors. n = 5–8 mice/group.
(G) Phenograph of CD45+ immune infiltrates from human PDAC patient CyTOF samples (pooled), with quantification of individual cellular fractions (log scale).
n = 11.
(H) Z–normalized cDC1 infiltration score between pancreatic (PAAD, n = 177) and lung (LUAD, n = 230) adenocarcinoma based on conserved cDC1 gene
signature.
(I) Representative histogram indicating ZsGreen in migratory cDC1s and cDC2s from respective draining lymph nodes of KPC-Z or KPL-Z tumors at denoted time
points. Right: percentage of migratory cDC subsets that have ZsGreen antigen in respective lymph nodes at denoted time points. n = 3–4 mice/group.
(J) Density of OVA-specific CD8+ T cells in draining lymph nodes of early-stage tumors. n = 5–8 mice/group.
Data were consistent across two independent experiments, except in (A), (B), (F), (G), and (H), where they were pooled across multiple experiments. Data are
presented as mean ± SEM, except in (H) where box plot denotes 10th to 90th percentile, middle line indicates median, range lines indicate maximal values, and
data points beyond indicate outliers (>1.53 range). For comparisons between any two groups, Student’s two-tailed t test was used. Frequency distributions were
compared using non-parametric Kolmogorov-Smirnov test. ns, not significant; *p < 0.05, **p < 0.01. See also Figure S4.
lesions (% of tissue)
Flt3L i.p. 1.0 40
1.0
0 9 12
0.5 20
0.5
H&E
0 0 0
KPC (n=5) KPC-OG (n=6) KPC-OG (+Flt3L) (n=6) KPC KPC-OG KPC-OG (+Flt3L)
C control + Flt3L 28
* *
D control + Flt3L 5
* *
α-SMA+ (% of tissue)
21
3
14
2
Sirius Red
α-SMA
1
0 0
KPC KPC-OG KPC-OG (+Flt3L) KPC KPC-OG KPC-OG (+Flt3L)
3.0 * * 0.4 * *
2.20 control +Flt3L 1.6
cells per g tissue (x 105)
2.0
cells per g tissue (x 105)
1.2
1.10 0.2 0.8
KPC (n=3) KPC-OG (n=6) KPC-OG (+Flt3L) (n=6) KPC (n=3) KPC-OG (n=6) KPC-OG (+Flt3L) (n=6)
G H I
control + Flt3L 1.6
ns 50 0.60 *
IFNγ+ TNFα+ CD8+ T cell
CD8+ T cells within 30 μm
0.55
relative frequency (%)
40
oer g tissue (x 104)
/ cm2 lesion ( x 103)
1.2 0.35
CD8+ T cell
30 0.26
0.8 **
CD8α CK19
0 0 0
10 30 50 70 90
control
+Flt3L
control
+Flt3L
J
+Flt3L +Flt3L + α-CD8α +Flt3L + α-IFNγ 90 * *
lesions (% of tissue)
60
all KPC-OG
control (n=12)
30
+Flt3L (n=9)
H&E
Figure 5. Mobilizing cDCs into Early Pancreatic Lesions Can Reverse Fibro-Inflammatory Responses
(A) Schematic of Flt3L administration in KPC-OG mice (starting at P30), with quantification of cDC1 and cDC2 density in pancreata of KPC-OG mice either treated
or not with Flt3L and control KPC mice. n = 5–6 mice/group.
(legend continued on next page)
298 Cancer Cell 37, 289–307, March 16, 2020
We speculated that the lack of impact of Flt3L on CTL and TH gen-specific CD8+ T cells in the dLNs of treated KPPC-OG
cell infiltration could be due to ineffective licensing of incoming tumors. There was a 6.7-fold increase in OVA-specific CD8+
cDCs, allowing for the immature DCs to become tolerogenic in T cells in dLNs of the anti-CD40 plus Flt3L treatment cohort,
the TME. To overcome this barrier, we either intratumorally compared with a 3.5-fold increase in the anti-CD40-only cohort
injected an STING agonist (RR-S2-CDA) to influence IFN-depen- (Figure 6J). This suggested that combination treatment was effi-
dent DC maturation (Corrales et al., 2015; Sivick et al., 2018) or cacious in improving T cell priming in PDAC dLNs.
administered CD40-agonist IgGs to improve licensing and Combining Flt3L administration with either CD40 or STING ag-
enhance APC function and survival (Beatty et al., 2011; Byrne onism resulted in improved disease control beginning a week
and Vonderheide, 2016). Unlike other tumor models (Kinkead into either treatment (Figure 6K). Combining CD40 agonism
et al., 2018; Ma et al., 2019), we found that neither CD40 nor and Flt3L was also effective at slowing tumor progression in
STING agonist alone enhanced cDC abundance in the PDAC the non-antigenic KPPC model (Figures S6F and S6G) and was
TME. However, combination of either agent with Flt3L worked dependent on T cells for efficacy (Figures S6H and S6I). Histo-
synergistically to drive massive influx of cDC1s and cDC2s, logically, tumors receiving Flt3L and anti-CD40 had no major
with CD40 agonist showing clear superiority and a >64-fold changes in total PDPN+ fibroblast presence, but substantially
increase in cDC1s (Figure 6D). Phenotypically, cDC1s and lower collagen deposition and a reduction in desmoplastic
cDC2s had higher MHCI and MHCII expression and modestly a-SMA+ fibroblasts (Figures 6L, 6M, and S6J). These data sug-
higher CD80/CD86 expression upon CD40 agonist and Flt3L gested that Flt3L plus anti-CD40 treatment might either result
treatment (Figure S6A). Notably, combination treatment with in collagen degradation and/or indirectly switch fibroblast
CD40 agonist and Flt3L did not mobilize more pre-cDCs than phenotype. In agreement with potential myeloid-dependent
Flt3L treatment alone, suggesting that the observed synergism matrix remodeling, we observed higher matrix metallopepti-
was in the PDAC TME (Figures S6B and S6C). dase-9 production in tumors treated with Flt3L plus anti-CD40
While CD40 or STING agonist alone modestly enhanced CD8+ (Figure S6K) (Long et al., 2016).
T cell infiltration, combination with Flt3L triggered markedly
enhanced intratumoral CD8+ CTL and CD4+ TH cell infiltration cDC-Directed Therapy Renders PDAC Responsive to
without TREG induction (Figure 6E). Additionally, CD40 agonist Radiation Therapy
plus Flt3L combination treatment increased the abundance of While we found that CD40 agonist plus Flt3L reshaped the im-
intratumoral OVA-specific CD8+ T cells (Figure 6F). Histological- mune response, we did not observe tumor regression. A possible
ly, we found more CD8+ CTLs were in close proximity and explanation is that immune priming and ‘‘antigen spill’’ by tumor
frequently in contact with PDAC cells of mice treated with the cell death is very limited in this model (Vonderheide, 2018). One
CD40 agonist plus Flt3L combination (Figure 6G). Notably, the effective modality to induce immunogenic cell death and boost
CD40 agonist plus Flt3L treatment elicited integrated anti-tumor CTL priming by APCs is radiation therapy (RT) (Ngwa et al.,
responses involving marked increases in infiltrating NK, NKT, 2018; Twyman-Saint Victor et al., 2015). RT in PDAC patients
and gdT cells (Figures 6H and S6D) and resulted in significant has limited benefit and is mostly palliative (Balaban et al.,
loss of GFP in OG tumors, suggesting immune evasion under 2016), possibly because the TME does not support induction
T cell pressure (Figure S6E). Due to the substantial changes of tumor immunity. To test whether cDCs could provide the
observed in cDC numbers upon anti-CD40 plus Flt3L combina- necessary induction and synergize with RT, we combined anti-
tion treatment, we next tested priming capacity in PDAC dLNs CD40 agonist and Flt3L treatment with radiotherapy. We utilized
using KPPC-ZsG mice. We found a considerable increase in computed tomography-guided RT to provide three fractionated
ZsGreen+ migratory cDC1s trafficking tumor antigen to dLNs 8-Gy doses directed to KPPC-OG pancreata after cDCs were
of CD40 agonist plus Flt3L-treated mice (Figure 6I). To determine mobilized. While RT alone only modestly affected tumor progres-
whether this enhanced antigen trafficking results in better sion, the triple therapy resulted in tumor regression in the major-
peripheral cross-priming, we analyzed the enrichment of anti- ity of KPPC-OG animals (Figures 7A–7C). Similar results were
(B) Representative H&E images of early-stage KPC-OG mice either treated or not with Flt3L, with quantification of lesion area. n = 8–12 mice/group; mice from
Figure 2 are included in this and following analyses.
(C) Sirius Red staining with quantification in KPC-OG mice either treated or not with Flt3L. n = 8–12 mice/group.
(D) a-SMA staining with quantification in KPC-OG mice either treated or not with Flt3L. n = 8–12 mice/group.
(E) Density of RORgt+ TH17 and GATA3+ TH2 cells in early-stage KPC-OG mice treated as indicated. n = 3–6 mice/group.
(F) Representative flow-cytometry plots of IL-17A and TNF-a expression in TH cells of KPC-OG mice either treated or not with Flt3L, with IL-17A+ and TNF-a+
IL-17A+ TH cellular density quantified. n = 3–6 mice/group.
(G) Representative immunohistochemistry of CD8+ T cells (brown) and CK19+ tumor lesions (pink) in early-stage KPC-OG mice either treated or not with Flt3L.
n = 6 mice/group.
(H) Cumulative CD8+ T cell density within 30 mm of CK19+ lesions, and distribution of CD8+ T cell proximity to nearest tumor cell in KPC-OG mice treated as
indicated. n = 6 mice/group.
(I) Density of IFN-g+ TNF-a+ cytotoxic CD8+ T cells in KPC-OG mice treated as indicated. n = 6 mice/group.
(J) Representative H&E images of early-stage KPC-OG mice treated with Flt3L and anti-CD8 or anti-IFN-g depletion antibodies, with quantification of lesion area.
n = 7–12 mice/group.
Data were consistent across two independent experiments, except in (B), (C), (D), (H), (I), and (J), where they were pooled across multiple experiments. Data are
presented as mean ± SEM. For comparisons between any two groups, Student’s two-tailed t test was used. Frequency distributions were compared using non-
parametric Kolmogorov-Smirnov test. ns, not significant; *p < 0.05, **p < 0.01. Scale bars, 500 mm (B, C, D, J) and 100 mm (G). See also Figure S5.
% of CD4+ T cells
3 6 9 2.4 1.5
4 28
2 4 6 1.6 1.0 21
2 14
1 2 3 0.8 0.5
7
9
0 0 0 0 0
14 untreated Flt3L untreated Flt3L
OVA-specific CD8+ T
35
% of CD4+ T cells
2.62 2.4 ns
3.15
2 28 4
0 0 0 0 0 0
untreated Flt3L α-CD40 untreated Flt3L α-CD40 untreated Flt3L
STINGag + Flt3L α-CD40 + Flt3L STINGag + Flt3L α-CD40 + Flt3L α-CD40 α-CD40
+ Flt3L
1.0
0 0 0 0
0.3
volume measure (cm3)
volume measure (cm3)
cells per dLN (x 103)
normalized to mode
0.2
3 5.4 * 0.2 0.1
L M
untreated α-CD40 + Flt3L untreated α-CD40 + Flt3L
12 * 10 *
8
% of tumor tissue
% of tumor tissue
9
Trichrome (collagen)
6
6
4
3
α-SMA
0 0
untreated Flt3L untreated Flt3L
α-CD40 α-CD40 + Flt3L α-CD40 α-CD40 + Flt3L
Figure 6. Enhancing cDC Infiltration and Activation in Established PDAC Leads to Disease Stabilization
(A) Schematic of Flt3L administration in ultrasound-diagnosed KPPC-OG mice. n = 5–8 mice/group.
(B) Density of (left) CD103+ cDC1s and CD11b+ cDC2s in tumors, and (right) migratory cDC1 and cDC2 populations in respective dLNs of KPPC-OG mice treated
with Flt3L. n = 5–8 mice/group.
(C) Density of CD8+ T cells and CD4+ TH cells, and frequency of CD4+ TREG in tumors of KPPC-OG mice treated with Flt3L. n = 5–8 mice/group.
(D) Density of cDC1s and cDC2s in tumors of KPPC-OG mice treated as described. n = 7–8 mice/group.
(E) Density of CD8+ T cells and CD4+ TH cells, and frequency of CD4+ TREGS in tumors of KPPC-OG mice treated as described. n = 7–8 mice/group.
(F) Density of OVA-specific CD8+ T cells in tumors of treated KPPC-OG mice. n = 5–8 mice/group.
(G) Representative immunohistochemistry of CD8+ T cells (brown) and CK19+ tumor lesions (pink) in KPPC-OG mice treated as indicated. Right: cumulative CD8+
T cell density within 50 mm of CK19+ lesions. n = 5–8 mice/group.
(H) Density of tumor-infiltrating NK cells, NKT cells, and gd-T cells in treated KPPC-OG mice. n = 5–8 mice/group.
(I) Representative flow histogram indicating ZsGreen in migratory cDC1s from draining nodes of KPPC-Z treated as indicated. Right: absolute number of
migratory cDC subsets that have ZsGreen. n = 3–4 mice/group.
(J) Density of OVA-specific CD8+ T cells in draining lymph nodes of treated KPPC-OG mice. n = 5–8 mice/group.
(K) Tumor growth quantified by ultrasound measurements over 2 weeks of treatment. Right: individual traces of untreated and anti-CD40 plus Flt3L combination
cohorts. n = 5–8 mice/group.
(L) Representative Masson’s trichrome staining with quantification in KPPC-OG mice treated as denoted. n = 5–8 mice/group.
(M) Representative a-SMA staining with quantification in KPPC-OG mice treated as denoted. n = 5–8 mice/group.
Data were pooled across multiple independent experiments for all treatments. Data are presented as mean ± SEM. For comparisons between any two groups,
Student’s two-tailed t test was used. ns, not significant; *p < 0.05, **p < 0.01. Scale bars, 100 mm (G) and 500 mm (L, M). See also Figure S6.
untreated
% change in volume
Flt3L 0.20 150
(D7 to D14)
0.15 100
0.10 50
α-CD40 + Flt3L
RT
(8Gy x 3)
+ RT
0.05 0
*
9 -50
0 7 14
14 days since treatment start
D KI E F G
tumor implant 600
0.9 100
day –8 *
400
% change in volume
RT 80
Flt3L
volume measure (cm3)
H KPPC I J
100
0.6
** 80
0 Flt3L 0.5
volume measure (cm3)
α-CD40
RT 0.4 60
(8Gy x 3) 0.3
RT 40 **
0.2
0.1 20
RT
9 0
0 7 14 21 28 35 20 40 60
14 days since treatment start days since treatment start
of RANK and non-canonical NF-kB signaling upon CD40 ligation B Cytotoxicity and Degranulation Assays
B Bone Marrow Chimerism
is known to enhance DC survival (Hou and Van Parijs, 2004; Miga
et al., 2001; Ouaaz et al., 2002). Flt3L and CD40 agonism have B Vaccination Strategy
been shown to independently enhance IFN-g and IL-12 produc- B Immunohistochemical Staining
B Immunofluorescence Staining
tion (Borges et al., 1999; Chaudhry et al., 2006; Garris et al.,
B Multiparametric Flow Cytometry
2018), and their combined activity could impose TH1 immunity
in the otherwise suppressive TME. Even though CD40 agonism B Mass Cytometry
B ELISA
has been shown to rely on Batf3-dependent DCs in PDAC (Byrne
B Ex Vivo cDC Cross-Presentation Assay
and Vonderheide, 2016), it is also known to be mediated by
TAMs (Hoves et al., 2018; Stromnes et al., 2019). Multiple B Western Immunoblot
B RNA Isolation
myeloid cell types could be mediating CD40-dependent immu-
B RNA Sequencing and Analysis
nity in the PDAC TME; this redundancy could be of benefit to
patients with extremely low numbers of cDCs. B TCGA Patient Analysis
The described cDC-targeted strategy may have the added d QUANTIFICATION AND STATISTICAL ANALYSIS
B Image Analysis
benefit of altering the TME for fully integrated immune killing.
The increase in intratumoral NK and NKT cells via Flt3L can B Statistical Analysis
have a profound effect on sustaining DC-T cell interaction d DATA AND SOFTWARE AVAILABILITY
and TH1 help (Barry et al., 2018; Böttcher et al., 2018; Nair
SUPPLEMENTAL INFORMATION
and Dhodapkar, 2017). Additionally, stromal remodeling and
resolution of fibrosis can facilitate drug delivery and enhance
Supplemental Information can be found online at https://doi.org/10.1016/j.
CTL activity in the TME (Feig et al., 2013; Jiang et al., 2016; ccell.2020.02.008.
Stromnes et al., 2015). Notably, CD40 agonism has been
shown to reduce pathogenic fibrosis by enhancing the matrix ACKNOWLEDGMENTS
remodeling properties of tumor-infiltrative monocytes (Long
et al., 2016). In addition to facilitating CTL activity and access, S.H. was supported by NCI predoctoral-to-postdoctoral fellowship
F99CA223043. J.K.S. was supported by NCI R01CA181745, and the Small An-
stromal remodeling has a positive effect on cDC migration, an-
imal Radiation Research Platform was purchased with S10OD020136. D.G.D.
tigen sampling, and activity (Boissonnas et al., 2013; Hope was supported by NCI P50CA196510, R01CA177670, R01CA203890,
et al., 2017); this can be advantageous for sustaining DCs in P30CA09184215, and BJC Cancer Frontier Fund. C.J.D. was supported by
similarly ‘‘insulated’’ solid tumors. 1DP5OD026427. We thank Brian Ruffell for his feedback while drafting the
Going forward, it becomes important to experimentally deter- manuscript. We thank the Genome Technology Access Center for help with
mine the best treatment window in which to administer inductive genomic analyses, supported by NCI P30CA91842 and by ICTS/CTSA
RT to maximize priming capacity of mobilized cDCs and main- UL1RR024992 from the NCRR and the NIH. We also thank the Digestive Dis-
ease Research Core Center for histological help, supported by NIDDK
tain their T cell-assistive function in the TME. It might also be
P30DK052574. Imaging experiments were performed through the use of
germane to combine this treatment with DC agonistic pathways Washington University Center for Cellular Imaging supported by CDI-CORE-
such as OX40 or 4-1BB to further benefit T cell co-stimulation. 2015-505 and the Foundation for Barnes-Jewish Hospital.
Meaningful interventions in PDAC will likely necessitate such
combinations (Baird et al., 2016; Hammerich et al., 2019; Rech AUTHOR CONTRIBUTIONS
et al., 2018). In conclusion, increasing the numbers and/or activ-
Conceptualization, S.H. and D.G.D.; Methodology, S.H., C.J.D., J.K.S., and
ity of scarce cDCs in PDAC is a distinct strategy that could D.G.D.; Investigation, S.H., V.E.K., B.H.H., C.Z., M.A.B., B.L.K., J.P.T.,
improve cytotoxic or immune-based therapies that by them- G.D.H., J.M.B., C.M., K.B.L., and K.A.A.; Writing – Original Draft, S.H.; Writing –
selves are poorly effective in the treatment of this disease. Review & Editing, S.H., R.C.F., and D.G.D.; Funding, S.H. and D.G.D.;
Beatty, G.L., Winograd, R., Evans, R.A., Long, K.B., Luque, S.L., Lee, J.W., Cristescu, R., Mogg, R., Ayers, M., Albright, A., Murphy, E., Yearley, J., Sher,
Clendenin, C., Gladney, W.L., Knoblock, D.M., Guirnalda, P.D., and X., Liu, X.Q., Lu, H., Nebozhyn, M., et al. (2018). Pan-tumor genomic
Vonderheide, R.H. (2015). Exclusion of T cells from pancreatic carcinomas in biomarkers for PD-1 checkpoint blockade-based immunotherapy. Science
mice is regulated by Ly6Clow F4/80+ extratumoral macrophages. 362, eaar3593.
Gastroenterology 149, 201–210. Dallal, R.M., Christakos, P., Lee, K., Egawa, S., Son, Y.I., and Lotze, M.T.
Bedrosian, A.S., Nguyen, A.H., Hackman, M., Connolly, M.K., Malhotra, A., (2002). Paucity of dendritic cells in pancreatic cancer. Surgery 131, 135–138.
Ibrahim, J., Cieza-Rubio, N.E., Henning, J.R., Barilla, R., Rehman, A., et al. de Mingo Pulido, Á., Gardner, A., Hiebler, S., Soliman, H., Rugo, H.S.,
(2011). Dendritic cells promote pancreatic viability in mice with acute pancre- Krummel, M.F., Coussens, L.M., and Ruffell, B. (2018). TIM-3 regulates
atitis. Gastroenterology 141, 1915–1926.e4. CD103+ dendritic cell function and response to chemotherapy in breast
Bellone, G., Carbone, A., Smirne, C., Scirelli, T., Buffolino, A., Novarino, A., cancer. Cancer Cell 33, 60–74.e66.
Stacchini, A., Bertetto, O., Palestro, G., Sorio, C., et al. (2014). Cooperative in- De Monte, L., Reni, M., Tassi, E., Clavenna, D., Papa, I., Recalde, H., Braga,
duction of a tolerogenic dendritic cell phenotype by cytokines secreted by M., Di Carlo, V., Doglioni, C., and Protti, M.P. (2011). Intratumor T helper
pancreatic carcinoma cells. J. Immunol. 177, 3448–3460. type 2 cell infiltrate correlates with cancer-associated fibroblast thymic
Article
1Department of Medical Oncology, Dana-Farber Cancer Institute and Harvard Medical School, Boston, MA 02215, USA
2Department of Cancer Biology, Dana-Farber Cancer Institute, Harvard Medical School, Boston, MA 02215, USA
3Department of Biological Chemistry and Molecular Pharmacology, Harvard Medical School, Boston, MA 02215, USA
4Department of Informatics and Analytics, Dana-Farber Cancer Institute, Boston, MA 02215, USA
5Center for Functional Cancer Epigenetics, Dana-Farber Cancer Institute, Boston, MA 02215, USA
6Belfer Center for Applied Cancer Science, Dana-Farber Cancer Institute, Boston, MA 02215, USA
7Blais Proteomics Center, Dana-Farber Cancer Institute, Harvard Medical School, Boston, MA 02215, USA
8Department of Oncologic Pathology, Dana-Farber Cancer Institute, Boston, MA 02215, USA
9Division of Hematology & Medical Oncology, Laura and Isaac Perlmutter Cancer Center, New York University Langone Medical Center, New
SUMMARY
Eradicating tumor dormancy that develops following epidermal growth factor receptor (EGFR) tyrosine
kinase inhibitor (TKI) treatment of EGFR-mutant non-small cell lung cancer, is an attractive therapeutic strat-
egy but the mechanisms governing this process are poorly understood. Blockade of ERK1/2 reactivation
following EGFR TKI treatment by combined EGFR/MEK inhibition uncovers cells that survive by entering a
senescence-like dormant state characterized by high YAP/TEAD activity. YAP/TEAD engage the epithelial-
to-mesenchymal transition transcription factor SLUG to directly repress pro-apoptotic BMF, limiting drug-
induced apoptosis. Pharmacological co-inhibition of YAP and TEAD, or genetic deletion of YAP1, all deplete
dormant cells by enhancing EGFR/MEK inhibition-induced apoptosis. Enhancing the initial efficacy of
targeted therapies could ultimately lead to prolonged treatment responses in cancer patients.
INTRODUCTION 2009; Rosell et al., 2012; Soria et al., 2018). However, acquired
resistance inevitably develops, limiting clinical efficacy (Cortot
Epidermal growth factor receptor (EGFR) tyrosine kinase inhibi- and Ja€nne, 2014). In most cases, resistance arises after a dra-
tors (TKIs) are the standard of care for patients with advanced matic initial response followed by a stable minimal residual dis-
EGFR-mutant non-small cell lung cancer (NSCLC) (Mok et al., ease (MRD), or dormant, state, with subsequent gradual growth
Significance
Genotype-directed therapy rarely, if ever, leads to complete tumor eradication. The residual tumors, following drug treat-
ment, can serve as reservoirs for the development of acquired drug resistance. Using EGFR-mutant lung cancer as a model
system, we identify YAP activation as a major counter regulatory mechanism promoting cell survival and dormancy
following sustained inhibition of EGFR and its downstream signaling, through transcriptional repression of the pro-
apoptotic protein BMF. Targeting YAP, indirectly using a tankyrase inhibitor or directly using a TEAD inhibitor, enhances
the apoptotic effects of EGFR/MEK inhibition. Our findings suggest that targeting YAP/TEAD following genotype-directed
therapy may be one strategy to enhance initial drug-induced apoptosis, reduce residual disease, and as such lead to
improved treatment outcomes.
104 Cancer Cell 37, 104–122, January 13, 2020 ª 2019 Elsevier Inc.
Arrien A. Bertram,1 Benjamin L. Ebert,1,14,16 Rameen Beroukhim,1,14,17 Pratiti Bandopadhayay,11,14,18 Mark M. Awad,1
Prafulla C. Gokhale,6,15 Paul T. Kirschmeier,1,6 Jarrod A. Marto,2,7,8,21 Fernando D. Camargo,13,22 Rizwan Haq,17,19
Cloud P. Paweletz,1,6 Kwok-Kin Wong,9 David A. Barbie,1 Henry W. Long,1,5 Nathanael S. Gray,2,3 and
€nne1,6,20,24,*
Pasi A. Ja
12Department of Pediatrics, Harvard Medical School, Boston, MA 02215, USA
13Stem Cell Program, Boston Children’s Hospital, Boston, MA 02215, USA
14Broad Institute of MIT and Harvard, Cambridge, MA 02142, USA
15Experimental Therapeutics Core, Dana-Farber Cancer Institute, Boston, MA 02210, USA
16Howard Hughes Medical Institute, Dana-Farber Cancer Institute, Boston, MA 02215, USA
17Department of Medicine, Harvard Medical School, Boston, MA 02115, USA
18Dana-Farber/Boston Children’s Cancer and Blood Disorders Center, Boston, MA 02115, USA
19Department of Molecular and Cellular Oncology, Dana-Farber Cancer Institute, Boston, MA 02215, USA
20Lowe Center for Thoracic Oncology, Dana-Farber Cancer Institute, 450 Brookline Avenue, LC4114, Boston, MA 02215, USA
21Department of Pathology, Brigham and Women’s Hospital and Harvard Medical School, Boston, MA 02215, USA
22Department of Stem Cell and Regenerative Biology, Harvard University, Cambridge, MA 02138, USA
23These authors contributed equally
24Lead Contact
*Correspondence: pasi_janne@dfci.harvard.edu
https://doi.org/10.1016/j.ccell.2019.12.006
of a drug-resistant tumor. Previous preclinical studies suggest to re-colonization of wells within 8 weeks (Figure 1A). The
that, following EGFR TKI treatment, EGFR-mutant tumor cells combination of O and the MEK inhibitor trametinib (T) prevents
can enter a drug-tolerant state, reminiscent of dormancy in pa- any measurable regrowth (Figure 1A). However, few viable cells
tients, allowing cells to evade apoptosis and survive (Hata can still be detected after 15 weeks of treatment (Figure 1B). We
et al., 2016; Sharma et al., 2010). Over time, the drug-tolerant used live-cell imaging and observed that the OT-treated cells
cells can acquire resistance through mutational or non-muta- surviving the initial apoptosis remained in a largely non-prolifer-
tional mechanisms (Hata et al., 2016). While the establishment ative, or dormant, state throughout the treatment period. How-
of this state seems to be largely stochastic and dictated mostly ever, within days following drug withdrawal the cells began to
by epigenetic mechanisms (Guler et al., 2017; Sharma et al., proliferate and re-colonize the wells (Figure 1C, Video S1). This
2010), the mechanistic bases of how cancer cells evade the phenomenon was consistent across EGFR-mutant NSCLC
initial apoptosis in response to drug treatment––the absolute cell lines (Figures 1C and S1B). These observations suggest
requirement to enter the drug-tolerant state––or maintain toler- that while combined EGFR/MEK inhibition eliminates cells in
ance in the presence of drug, are poorly understood. which reactivation of ERK signaling occurs following single-
Our previous work demonstrated that despite sustained agent EGFR inhibition, a separate population enters a dormant
EGFR inhibition following EGFR TKI treatment, reactivation state, surviving combined EGFR/MEK inhibition. There was no
of ERK1/2 occurs within just a few days (Ercan et al., 2012; evidence of reactivation of EGFR and/or ERK signaling in the
Tricker et al., 2015). Concomitant inhibition of MEK effectively dormant cells during treatment (Figure 1D) and EGFR signaling
prevents reactivation of ERK1/2, results in a greater initial was restored in cells that grew following drug washout (Fig-
apoptotic response, and leads to a more durable tumor con- ure 1D). These cells were still sensitive to OT, and morphologi-
trol in vitro and in vivo than single-agent EGFR inhibition cally indistinguishable from the untreated cells (Figure S1C),
(Ercan et al., 2012; Tricker et al., 2015). The EGFR (osimertinib) suggesting that we did not select out a subclone with a pre-ex-
and MEK (selumetinib) inhibitor combination has been studied isting resistance mutation (Hata et al., 2016). To formally address
in patients resistant to previous EGFR TKIs and is also under whether the establishment of dormancy following OT treat-
evaluation as initial therapy for advanced EGFR-mutant ment is pre-determined or a stochastic process, we barcoded
NSCLC in a phase II clinical trial (NCT03392246; Ramalingam PC-9 cells using the EvoSeq library (Feldman et al., 2019),
et al., 2019). However, even with this combination, acquired treated the cells with DMSO, gefitinib (G), O, or OT for 3 weeks,
resistance can still develop (Tricker et al., 2015). In this study sequenced DNA from the remaining cells and analyzed the
we set out to elucidate the mechanisms that allow cancer findings as described (Bhang et al., 2015) with some modifica-
cells to evade apoptosis and survive despite combined tions. We observed a large fraction of shared barcodes within
EGFR/MEK inhibition. the G-treated (data not shown) and O-treated (Figures 1E, 1F,
and S2A) cells, strongly suggesting selection of pre-existing
RESULTS clones, consistent (for G) with previous studies (Hata et al.,
2016). In contrast, the vast majority of barcodes in the OT-
Combined EGFR and MEK Inhibition Results in a Stable treated cells were unique (Figures 1E, 1F, and S2A). Comparison
but Reversible Dormant State of the shared barcodes between O and OT cells demonstrated
Combined EGFR/MEK inhibition prevents the reactivation of that 89% of the barcodes identified in the O group are not
ERK1/2 following EGFR inhibition and delays the onset of drug present in the OT group (Figure S2B). These findings suggest
resistance in vitro and in vivo (Tricker et al., 2015) (Figure S1A). that, while resistance to O likely occurs through a selection
In PC-9 cells, treatment with single-agent osimertinib (O) leads process of a pre-existing clone, the ability of cells to enter
(G) GSEA of Hallmark gene sets comparing dormant cells versus DMSO-treated control cells. Normalized enrichment scores (NES) for gene sets with false
discovery rate (FDR) < 0.1 in at least two cell lines are shown.
(H) Senescence-associated b-galactosidase (SA-b-Gal) staining of cells treated as indicated for 10 days. Scale bars, 100 mm.
(I) Quantification of (H).
(J) GSEA of senescence signature comparing dormant, OT-treated PC-9 cells versus control cells.
(K) Immunofluorescence (IF) staining for H3K9Me3 in control cells or dormant cells treated with OT for 10 days. Scale bars, 20 mm.
Mean ± SEM are shown in all plots except (I) where mean ± SD are shown. ANOVA (I) or t test (K) were used for statistical analyses. ***p < 0.001, **p < 0.01. See
also Figures S1–S3.
(G) Regrowth of EGFR-mutant NSCLC cells after washout following a 3-week treatment with the indicated drug combinations.
(H) Western blot analysis of YAP protein levels in YAP1 knockout (KO) and control (CTRL) cells.
(I) Proliferation of cells in (H) treated as indicated for 21 days, followed by drug washout.
(J) Mice bearing CTRL or YAP1 KO cell xenograft tumors were treated with vehicle or OT followed by treatment cessation and follow-up. Data with at least 6/8 live
mice per group are plotted. Right: tumor volumes at time of regrowth, indicated by an arrow.
Mean ± SEM are shown in all plots except (F), where mean ± SD are shown. ANOVA was used for statistical analyses in all but (J), where t test was used.
***p < 0.001, *p < 0.05. See also Figure S4.
(H) Left: western blot analysis of YAP protein levels in YAP1 KO cells transduced with wild-type YAP1. Right: cells were treated with OT and analyzed as in (G).
Only data from drug-treated cells is shown.
(I) Proportions of YAPhigh cells in PC-9 YAP reporter cell populations treated as indicated. Scale bars, 150 mm.
(J) Different means for EGFR-mutant NSCLC cells to avoid apoptosis following EGFR inhibition.
Mean ± SEM are shown in all plots except (I), where SD is shown. ANOVA was used for statistical analyses in all but (D), where Fisher’s exact test was used.
***p < 0.001. See also Figure S5.
(F) Western blot analysis of BMF, BIM, and YAP expression in PC-9 HA-BMF cells transfected with non-targeting (NT) or YAP siRNA and treated as indicated
for 24 h.
(G) qPCR analysis of BMF expression in CTRL or YAP1 KO cells transduced as indicated, and following treatment with either DMSO or OT for 24 h.
(H) Peak apoptosis over 72-h treatment in PC-9 and HCC4006 cells transfected with NT or BMF siRNA.
(I) The mechanism of YAP/TEAD-mediated suppression of apoptosis in EGFR-mutant NSCLC cells following EGFR/MEK inhibition.
Mean ± SD are shown in all plots except (H), where mean ± SEM is shown. ANOVA was used for statistical analyses. ***p < 0.001, **p < 0.01; n.s., not significant
(p > 0.05). See also Figure S7.
(G) qPCR analysis of BMF expression in cells in (E) following 24 h treatment as indicated.
(H) Apoptosis in PC-9 and HCC4006 cells treated as indicated.
(I) Regrowth of PC-9 and HCC4006 cells after drug washout following a 2-week treatment as indicated.
Mean ± SEM are shown in all plots except (E), where mean ± SD is shown. ANOVA was used for statistical analyses. ***p < 0.001, **p < 0.01. See also Figure S8.
Bhang, H.C., Ruddy, D.A., Krishnamurthy Radhakrishna, V., Caushi, J.X., Costa, D.B., Halmos, B., Kumar, A., Schumer, S.T., Huberman, M.S., Boggon,
Zhao, R., Hims, M.M., Singh, A.P., Kao, I., Rakiec, D., Shaw, P., et al. T.J., Tenen, D.G., and Kobayashi, S. (2007). BIM mediates EGFR tyrosine ki-
(2015). Studying clonal dynamics in response to cancer therapy using high- nase inhibitor-induced apoptosis in lung cancers with oncogenic EGFR muta-
complexity barcoding. Nat. Med. 21, 440–448. tions. PLoS Med. 4, 1669–1680.
Bhola, P.D., and Letai, A. (2016). Mitochondria-judges and executioners of cell Cragg, M.S., Kuroda, J., Puthalakath, H., Huang, D.C.S., and Strasser, A.
death sentences. Mol. Cell 61, 695–704. (2007). Gefitinib-induced killing of NSCLC cell lines expressing mutant EGFR
Blanke, C.D., Rankin, C., Demetri, G.D., Ryan, C.W., Von Mehren, M., requires BIM and can be enhanced by BH3 mimetics. PLoS Med. 4,
Benjamin, R.S., Raymond, A.K., Bramwell, V.H.C., Baker, L.H., Maki, R.G., 1681–1690.
et al. (2008). Phase III randomized, intergroup trial assessing imatinib mesylate Debacq-Chainiaux, F., Erusalimsky, J.D., Campisi, J., and Toussaint, O.
at two dose levels in patients with unresectable or metastatic gastrointestinal (2009). Protocols to detect senescence-associated beta-galactosidase (SA-
stromal tumors expressing the kit receptor tyrosine kinase: S0033. J. Clin. bgal) activity, a biomarker of senescent cells in culture and in vivo. Nat.
Oncol. 26, 626–632. Protoc. 4, 1798–1806.
Med, a new journal from Cell Press, publishes transformative, evidence-based science across
the clinical and translational research continuum – from large-scale clinical trials to translational
studies with demonstrable functional impact, offering novel insights in disease understanding.
We aim to elevate the global standard of medical research by accelerating translation of bench
research to the clinic, serving as a hub for engagement between all stakeholders, improving
reproducibility, and changing medical practice.
Elevate your research. Submit your paper today.
cell.com/med
ll
Article
Predicting Drug Response and Synergy
Using a Deep Learning Model of Human Cancer Cells
Brent M. Kuenzi,1,5 Jisoo Park,1,5 Samson H. Fong,1,2 Kyle S. Sanchez,1 John Lee,1 Jason F. Kreisberg,1 Jianzhu Ma,4
and Trey Ideker1,2,3,6,*
1Division of Genetics, Department of Medicine, University of California San Diego, La Jolla, CA 92093, USA
2Department of Bioengineering, University of California San Diego, La Jolla, CA 92093, USA
3Department of Computer Science and Engineering, University of California San Diego, La Jolla, CA 92093, USA
4Department of Computer Science, Purdue University, West Lafayette, IN 47907, USA
5These authors contributed equally
6Lead Contact
*Correspondence: tideker@ucsd.edu
https://doi.org/10.1016/j.ccell.2020.09.014
SUMMARY
Most drugs entering clinical trials fail, often related to an incomplete understanding of the mechanisms gov-
erning drug response. Machine learning techniques hold immense promise for better drug response predic-
tions, but most have not reached clinical practice due to their lack of interpretability and their focus on mono-
therapies. We address these challenges by developing DrugCell, an interpretable deep learning model of
human cancer cells trained on the responses of 1,235 tumor cell lines to 684 drugs. Tumor genotypes induce
states in cellular subsystems that are integrated with drug structure to predict response to therapy and,
simultaneously, learn biological mechanisms underlying the drug response. DrugCell predictions are accu-
rate in cell lines and also stratify clinical outcomes. Analysis of DrugCell mechanisms leads directly to the
design of synergistic drug combinations, which we validate systematically by combinatorial CRISPR,
drug-drug screening in vitro, and patient-derived xenografts. DrugCell provides a blueprint for constructing
interpretable models for predictive medicine.
672 Cancer Cell 38, 672–684, November 9, 2020 ª 2020 Elsevier Inc.
ll
Article
Morgan fingerprint
model interpretation is now a rapidly growing subfield within ma- the impact of genetic mutations on cellular growth response
chine learning, with a growing arsenal of approaches for achieving and, simultaneously, identify the most relevant molecular path-
models with not only high predictive accuracy, but also high ways driving those predictions. Building from this paradigm,
descriptive accuracy (Murdoch et al., 2019). One major strategy we now describe DrugCell, a VNN that simulates the response
has been to use prior knowledge or data to add structure to the of human cancer cells to therapeutic chemical compounds.
model, which can then be interpreted. Applied to genomics, such DrugCell couples the inner workings of the model to the hierar-
a strategy has been used to recast the thousands of measured mo- chical structure of human cell biology, allowing for response pre-
lecular features of a tumor as states on a much smaller number of dictions for any drug in any cancer and intelligent design of effec-
functional modules (Cortés-Ciriano et al., 2016; Yang et al., 2019). tive combination therapies.
For example, a recent study mapped raw molecular measurements
to a set of pre-defined metabolic pathways drawn from prior knowl- RESULTS
edge bases; the states of these pathways predict antibiotic resis-
tance in Escherichia coli, with particular pathway features emerging Design and Training of an Interpretable Neural Network
as candidate mechanisms of resistance (Yang et al., 2019). Organi- of Drug Response
zation of molecular features into predictive modules can also be The cellular drug response is a complex phenomenon that de-
accomplished using prior data as opposed to literature-curated pends on both biological and chemical factors (Turner et al.,
knowledge. Such an approach was recently exemplified by Deep- 2015). Current black-box models of drug response that use
Profile, which analyzed a large collection of leukemia expression both these factors have begun to reach the limits of predictive
profiles to extract a low-dimensional representation of these data performance (Table S1). We therefore aimed to design a model
as a set of functional gene modules; these modules are then that maintains this high level of predictive capability while gaining
used as interpretable features for drug response prediction (Dincer mechanistic interpretability of the model predictions. To capture
et al., 2018). Apart from model-based approaches, a second major both determinants of drug response in an interpretable model,
strategy to increase model interpretability has been to perform post we devised DrugCell as a neural network with two branches (Fig-
hoc analysis of model features or feature weights to interpret the ure 1A, STAR Methods). The first branch was a VNN modeling
underlying drug response mechanisms (Chiu et al., 2019; Iorio the hierarchical organization of molecular subsystems in a hu-
et al., 2016; Murdoch et al., 2019). For example, the weights as- man cell, drawn from 2,086 biological processes documented
signed to each input gene by a black-box neural network model in the Gene Ontology (GO) database (Ashburner et al., 2000) (Fig-
are subjected to gene set enrichment analysis (Subramanian ure S1A). Each of these subsystems, from those involving small
et al., 2005) to identify pathways regulating the predicted drug protein complexes (e.g., b-catenin destruction complex) to
response (Sakellaropoulos et al., 2019). These pathways, however, larger signaling pathways (e.g., MAPK signaling pathway) to
were not used during modeling or validated experimentally. overarching cellular functions (e.g., glycolysis), was assigned a
To more explicitly link the structure of a machine learning bank of artificial neurons to represent the state of that subsystem
model to cellular functions, we recently developed a visible neu- (Figure 1B). Connectivity of neurons was set to mirror the biolog-
ral network (VNN) simulating a simple eukaryotic cell, Saccharo- ical hierarchy, so that neurons accept inputs only from child sub-
myces cerevisiae (Ma et al., 2018; Yu et al., 2018). This model, systems and send outputs only to parent systems, with connec-
called DCell, was made mechanistically interpretable, or tion weights determined during training. The use of multiple
‘‘visible,’’ by directly mapping the neurons of a deep neural neurons per subsystem (here six, see STAR Methods) allowed
network into a large hierarchy of known and putative molecular cellular subsystems to be multifunctional, with distinct states
components and pathways. DCell is able to accurately predict able to adopt a range of values along multiple dimensions
A B C
rho = 0.8
1.0 1.0
0.6
0.0 0.0
0.4
-0.5 -0.5
0.2
p < 0.0001 p > 0.05
-1.0 -0.5 0.5 1.0 -1.0 -0.5 0.0 0.5 1.0
0.0
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1.0
0.0
Inf
Spearman rho
0.8
D E 0.6
1.0 0.4
Spearman correlation (DrugCell)
G ML tine
46 030
rb et 4
da el
le
11 lit 1
Ti 02 l
Te an 8
po ib
de
6 e
Pa Doc 136
C c 9
v 5
en ax
36 x
KX zo
ni tin
N Pa 2-3
si
a
ris
SK -
nc
Performance
Vi
0.0
ZI
0.4
-0.5
(Copley, 2012). The input layer of the hierarchy mapped to the recording the mutational status (1 = mutated, 0 = non-mutated)
mutation status of genes. The six neurons at the VNN output, of the top 15% most frequently mutated genes in cancer (n =
corresponding to the root of the hierarchy, represented the 3,008; median mutated genes per cell line = 73; Figure S1E).
embedded state of the whole cell based on its genotype (Fig- Each drug’s chemical structure was represented by an average
ure 1B). In total, the VNN used 12,516 neurons distributed hier- of 81 activated bits in the Morgan fingerprint vector, with each bit
archically across six distinct layers (STAR Methods, Figure S1B). typically representing fewer than 10 molecular fragments (Fig-
The second branch of DrugCell was a conventional artificial neu- ures S1F and S1G). DrugCell was trained to associate each ge-
ral network (ANN) embedding the Morgan fingerprint of a drug, a notype-drug pair with its corresponding drug response,
canonical vector representation of chemical structure (Figure 1C, measured by the area under the dose-response curve (AUC,
STAR Methods) (Rogers and Hahn, 2010). Outputs from the two STAR Methods). The DrugCell model and its codebase are avail-
branches of the model, the VNN embedding cell genotype and able for public download on GitHub (https://github.com/
the ANN embedding drug structure, were combined in a single idekerlab/DrugCell).
layer of neurons, which were then integrated to generate
the response of a given genotype to a particular treatment Interpretable Modeling of Drug Response Has No
(Figure 1A). Performance Loss
To train the model, we harmonized data from two large cancer We first sought to assess the prediction accuracy of DrugCell us-
drug screening resources: the Cancer Therapeutics Response ing the Spearman correlation (rho) between predicted and
Portal (CTRP) v2 and the Genomics of Drug Sensitivity in Cancer observed AUC values in 5-fold cross validation (STAR Methods).
(GDSC) database (Seashore-Ludlow et al., 2015; Yang et al., The total accuracy over all cell line-drug pairs was rho = 0.80
2013). The combined dataset consisted of 509,294 cell line- (Figure 2A). Further insight was achieved by computing the pre-
drug pairs, covering 684 drugs and 1,235 cell lines (Figure S1C, diction accuracy for each drug individually, revealing a subpop-
STAR Methods). All major tissue types were represented, with ulation of drugs with very high prediction accuracy (30% of
hematopoietic and lung lineages the most prevalent (Figure S1D). drugs with rho > 0.5) amid a much wider general distribution
Each cell-line genotype was represented by a binary vector (range 0.29 to +0.83, median 0.37). These accuracies were
significantly higher than those achieved for elastic net (median individual subsystems within the VNN and found that many were
rho = 0.35), a state-of-the-art regression technique used in in agreement with subsystem activities measured experimentally
many previous approaches to drug response prediction (Eskio- by an independent analysis of protein abundances and phos-
cak et al., 2017; Iorio et al., 2016; Kuenzi et al., 2019; Potts phorylation states using reverse-phase protein arrays (RPPAs;
et al., 2015) (Figure 2B). DrugCell’s drug-by-drug predictive per- Figure S3A; STAR Methods). For example, DrugCell accurately
formance was not significantly different from that of a conven- captured MAPK pathway activity within the subsystem embed-
tional black-box ANN with matching numbers of neurons, layers, ding of Regulation of MAPK cascade (Figure S3B), which signif-
and connections (Figure 2C). It was also comparable to previous icantly correlated with ERK1/2 phosphorylation (Figure S3C).
efforts to incorporate chemical features of drugs into the Overall, the majority of DrugCell subsystems were well corre-
response prediction (e.g., structure and physiochemical proper- lated with the RPPA measurements of those subsystems (note
ties such as solubility, lipophilicity, and molecular weight), and it bimodal distribution of correlation in Figure S3A). Other accu-
outperformed models that predict response using biological fea- rately captured subsystems included Proteolysis (Figure S3D),
tures alone (e.g., expression of biomarkers, point mutation, copy Regulation of PI3K signaling (Figure S3E), and Cell-cycle arrest
number variation, and microsatellites; Table S1). Finally, since (Figure S3F).
knowledge of tissue type can be predictive of drug response Inspection of the drug embedding from the ANN revealed a
even in the absence of other information (Iorio et al., 2016), we stratification of drugs based on their mechanisms of action
considered that some of the performance of these models might within major drug target classes (Figure 3E). The distance be-
be due to their ability to recognize the tissue type of a cell line tween each pair of drugs in the chemical structure embedding
from its input data (i.e., its mutational profile). Accordingly, we did not correlate with their overall chemical similarity (Fig-
compared DrugCell with an equivalent neural network model ure S4A), consistent with previous studies of drug activity and
trained on drug structure and a tissue label only (STAR Methods). chemical structure (Breinig et al., 2015). Since the training data
DrugCell vastly outperformed this tissue-only model (median consisted solely of drugs and drug-like molecules, the chemical
rho = 0.18; Figure 2D), indicating that the model had learned in- structural embedding did not stratify drugs on chemical features
formation from somatic mutations beyond the tissue of origin. such as membrane permeability (Figure S4B), solubility (Fig-
Compounds for which DrugCell predictions were most accu- ure S4C), or pharmacodynamic properties (Lipinski; Figure S4D).
rate came from diverse target classes, including chemothera- Together these results suggest that DrugCell is able to learn key
peutics (e.g., vincristine, teniposide) and targeted therapies features of the genotype that govern drug sensitivity and resis-
(e.g., GSK461364 targeting PLK1, KX2-391 targeting Src; Fig- tance, as well as features of chemical structure that govern
ure 2E). DrugCell maintained the specificity of the training data drug biological activity.
in that its predictions were specific to individual classes of drugs Since DrugCell’s VNN is structured according to the hierarchy
(e.g., MEK inhibitor predictions were highly specific) and did not of biological subsystems comprising a human cell, its output (ge-
simply reflect general drug toxicity (Figure S2A). Predictive per- notype embedding) is the result of state changes in particular
formance for a drug did not strongly correlate with the number subsystems within that hierarchy. To identify the most important
of cell line-drug pairs used for training, nor with the structural of these subsystems, we scored subsystems by the degree to
complexity of a compound (number of activated bits; Figures which their states were significantly more predictive of a drug
S2B and S2C). We did find that compounds eliciting a larger response than the states of their child subsystems using the rela-
range of cell-line responses tended to be more predictable (Fig- tive local improvement in predictive power metric (RLIPP, STAR
ure S2D). Similarly, individual cell lines (Figure S2E) and tissue Methods) (Ma et al., 2018). As an initial proof of concept, we used
types (Figure S2F), which elicit a large range of responses, RLIPP scoring to identify subsystems important for the cellular
were in general highly predictable. response to taxol (paclitaxel), an agent that stabilizes microtu-
bules (Figures 2E and 3F, Table S2). Among the top scores for
DrugCell Learns Mechanisms that Mediate Specific paclitaxel, many subsystems were metabolic processes (hyper-
Drug Responses geometric p < 0.05; Figures 3G and 3H), including Response to
Having evaluated predictive ability, we next turned to mecha- cAMP (top score) along with Insulin secretion in response to
nistic interpretation. This task was aided by the two model glucose and Response to glucose. We confirmed by inspection
branches, which dissect the effects of genotype on the configu- that the states of these subsystems had the ability to stratify
ration of cell systems (genotype embedding) from the effects of paclitaxel sensitive versus resistant cell lines (e.g., Response to
chemical structure on drug activity within the cell (drug embed- cAMP subsystem, Figure 3I). Given these underlying metabolic
ding, Figure 1A). We visually inspected these embeddings by pathways, we hypothesized that paclitaxel efficacy might be
plotting the top two principal components (Figures 3A–3E). The modulated by metabolic perturbation. We therefore exposed
genotype embedding from the VNN revealed a separation of ge- A427 cells to three different treatments – paclitaxel, the glycol-
notypes according to mutations known to confer specific drug ysis inhibitor 2-deoxyglucose (2-DG), or a combination of the
sensitivities, such as activating mutations in BRAF (Figure 3A) two – and found that the combination was substantially more
that promote sensitivity to the MEK inhibitor selumetinib (Fig- effective than either individual compound (Figure 3J).
ure 3B). The genotype embedding also distinguished mutations A similar analysis was performed for the next (second-most)
leading to drug resistance, such as mutations in EGFR (Yin et al., important subsystem, Regulation of ubiquitin-protein trans-
2019), LKB1 (Shimamura et al., 2013), or BRAF (Ma et al., 2017) ferase activity (Figure 3G, Table S2). We combined paclitaxel
(Figure 3C) that confer resistance to the BET-family inhibitor JQ- with perturbation of ubiquitin-dependent protein degradation
1 (Figure 3D). We similarly inspected the DrugCell embeddings of via the proteasome inhibitor bortezomib (Figure S5A). We found
A B C D
Sensitive Resistant Sensitive Resistant
BRAF mutations EGFR, BRAF,
0.0 0.4 0.8 1.2 or LKB1 mutations 0.0 0.4 0.8 1.2
Selumetinib AUC JQ-1 AUC
E H
Chemical structure embedding (PC2)
Insulin
secretion in
response to
glucose
Response to
cAMP
Response to
glucose
Target class
EGFR
MEK
CDK
BRAF
PARP
BRD4
Chemical structure
embedding (PC1)
Sensitive Resistant
F Sensitive Resistant G I 0.0 0.4 0.8 1.2 J
Metabolic pathway
Paclitaxel AUC
Response to cAMP embedding (PC3)
to cAMP ***
Importance for paclitaxel response
100
Regulation
in response 80
to glucose
70
Response
to glucose 60
10
n = 239 n = 239 50
Top 5% of subsystems Paclitaxel (8nM)
Genotype embedding (PC1) Response to cAMP
2-DG (800µM)
embedding (PC1)
that these treatments were antagonistic, consistent with recent pathways were not identified by earlier analyses of genetic
findings showing that glycolysis is subject to negative physical mutations (Table S4) and were distinct from those identified
regulation by ubiquitin ligases at the cytoskeleton (Park et al., by differential mRNA expression analysis of paclitaxel sensitive
2020). Ubiquitin and subsystems were also identified for doce- versus resistant lines (Figures S5B and S5C). Unlike the
taxel, a sister compound (Table S3). Notably, these DrugCell glycolytic perturbations emerging from DrugCell analysis,
A B C D E
P1
300
AP
R
-5 0 5
PA
TP
M
120
40
200
Area under 80
fitness curve 20
40 100
of double KO
Select 0 0 0
of lic en res ER
on via an se
bi eo n
im e
us
accurately
sy sfe hyl t
on
D e lo it y
t o lic t io
ic om
pl tion
lo R
al thy t
du n
n
n
pa en
en
in
rg on
ul
ns tio
io
su tra m me
N
iv
st ras ati
se sp iza
ag
A pm
s
ct
de d N
pm
de act
p
ou za
tra la
predicted
N n fi mu pled
9 op
st
ck
e
l
em e
et
qu H eve
ve
im u
gn me
e
compounds
ot
A ed -co
re ing t o
n
sc t or
d
t
iti 3K
er
ia r ip
g. on tex
er ran en
AS N
n
g. sp m
. o Re lay
,
R fD
ed sc
ri
T m
la
si
co
bi of
-m an
o
e
f il
t
in
f u g.
Ig Tr
sp
al
Po mR cti
th
t in
re ati
e
in
Ac
re A
eb
of
Vi
ul
r
by
La
R
eg
s.
s.
176 cancer genes
R
F G H NS
*
* BLM
3 POLK
TDP2
2.0
Identify subsystems 1
mediating drug 2
response
in combination with
in combination with
in combination with
PARP1 KO
MAP2K1 KO
TP53 KO
-1 1.0
CDK7
0
-2 0.5 KDM1A
TOP2A
-1
-3
PLK1 0.0 PRMT5
LIG1
-4 MPHOSPH8
-2
CDK7
CDK1 -0.5
Dual gene KOs to validate top subsystem
Gene KOs in
top nutlin-3
subsystems
Gene KOs in
top olaparib
subsystems
Gene KOs in
random
subsystems
Gene KOs in
random
subsystems
Gene KOs in
top trametinib
subsystems
Gene KOs in
random
subsystems
combination treatments suggested by differentially expressed which had broad representation of cancer signaling pathways
pathways were not successful at enhancing paclitaxel efficacy (MCF7 cells; Figure 4B). The top five important subsystems in
(Figure S5D). the response to each drug were identified (RLIPP analysis; Fig-
Moving beyond paclitaxel to examine the important subsys- ures 4C–4E), along with the genes in those subsystems covered
tems identified for other drugs, we found that some of these sub- by the CRISPR library. Combinatorial disruption of MAPK1 with
systems corresponded to previously identified mechanisms of genes in trametinib subsystems (Figure 4C) resulted in signifi-
drug sensitivity, while many others were novel pathways war- cantly more cell killing than observed for genes from random un-
ranting further investigation. In particular, we examined 60 drugs important subsystems (Figure 4F). A similar cell killing effect (Fig-
for which pan-cancer diagnostic gene mutations had been re- ure 4G) was observed for combinatorial disruption of PARP1 with
ported by an earlier analysis of the GDSC dataset using type II genes in olaparib subsystems (Figure 4D). In contrast, combina-
error ANOVA modeling (Iorio et al., 2016). For a number of drugs, torial disruption of TP53 with genes in nutlin-3 subsystems (Fig-
DrugCell recovered the previously reported diagnostic gene(s) ure 4E) had effects on cell growth that were not significantly
within the top subsystem (4 drugs) or top 10 subsystems (14 different from random (Figure 4H). This result was expected, as
drugs, upper 0.4th percentile of subsystems). For the vast major- TP53 knockout has the opposite effect compared with nutlin-3,
ity, however (56 drugs), DrugCell achieved better predictive per- which leads to p53 activation. These results, together with the
formance by consulting additional, or different, markers than had preliminary results from paclitaxel, provide systematic support
been previously reported (Table S4). for the importance of top response pathways identified by
Given the extent of novel drug response pathways, we sought DrugCell.
to systematically investigate the indicated mechanisms (Fig-
ure 4A; STAR Methods), focusing on trametinib, a MEK1 inhibi- Identified Subsystems Represent Synergistic Drug
tor; olaparib, a PARP1 inhibitor; and nutlin-3, an MDM2 antago- Combination Opportunities
nist that stabilizes and activates p53. CRISPR knockouts of each The parallel pathway inhibition theory of drug synergy (Yeh et al.,
of the three drug targets (MEK1, PARP1, TP53) were combined 2009) holds that two drugs will be synergistic if they inhibit sepa-
with knockouts of each gene in a custom CRISPR/Cas9 library, rate pathways that regulate a common essential function
A B C D 40 *** E
***
Pathway 1 Pathway 2 Pathway(s) MoA
Negative
A S X A S regulation of Regulation of
yn yn
...
...
...
top 5 bottom 5
VNN ANN
Essential cellular Drug response Synergistic Non-synergistic
function D2 partner drugs partner drugs
D’2 -20
Cell response
Druggable target Active signaling
to etoposide
= y
| = rgy
= ed
Inactive signaling
)| rg
Gene in pathway
75
70
)
(n ict
76
D2 yne
D’ yne
d
re
Assess synergy
-s
2)
tp
|(D ted
|(D on
( D1, D2 ), ( D1, D’2 )
No
n
ic
1,
1,
ed
ed
ct
Pr
i
ed
Pr
0 = wild-type
1 = mutant Odds ratio
F G H FLT1 J 0 2 4 6 8 10 12 14
(Figure 5A). The branched architecture of the DrugCell model been tested across a panel of 39 cell lines (Figure 5C). We then
(Figure 1A) mirrors this parallel pathway structure, in that the bio- analyzed the top 5 and bottom 5 DrugCell subsystems for each
logical activity of a drug is learned by the drug embedding of these compounds to nominate synergistic and non-synergistic
branch, and the parallel pathways are learned by the genotype drug combinations. We found that drug combinations nominated
embedding branch (Figure 5B). Subsystems important for pre- by DrugCell were strongly and significantly enriched for synergis-
dicting a drug response may therefore represent synergistic tic cell killing outcomes, in contrast to combinations predicted to
drug combination opportunities. Exactly such parallelism was be non-synergistic or random combinations (Figure 5D).
used to nominate the combination treatments in the above anal- One such example was etoposide, a topoisomerase inhibitor
ysis (i.e., 2-DG as synergistic with paclitaxel). that leads to DNA damage (Table S5). Among the top etoposide
To further explore this concept, we used RLIPP scores to rank subsystems were the major kinase signaling pathways PI3K-
subsystems regulating sensitivity to 25 drugs in the DeepSynergy AKT (Regulation of PI3K activity, PI3K; Figure 5E) and RAF-
database (Preuer et al., 2018), in which all pairs of 25 drugs had MEK-ERK (Negative regulation of ERK1/ERK2 cascade, ERK;
Figure 5E). Indeed, etoposide synergized strongly with AKT and ating characteristic curve (ROC; Figure 6A, STAR Methods).
MEK inhibition across the majority of cell lines tested in DeepSy- We found that DrugCell was able to accurately identify subsys-
nergy (Figure 5F). We further validated the observed synergy by tems that correspond to effective drug combinations in PDX tu-
deleting the target of etoposide, TOP2, using CRISPR/Cas9 mors (auROC = 0.75; Figure 6B) with relatively few false positives
gene editing in A549 cells, either alone or in combination with and negatives (Figure 6C).
core genes in PI3K-AKT signaling (PIK3CA) or RAF-MEK-ERK For example, DrugCell analysis of BKM-120, a PI3K inhibitor,
signaling (MAP2K1). We observed that deletion of TOP2 with identified Negative regulation of ERK1 + ERK2 cascade as an
either PIK3CA or MAP2K1 demonstrated significant loss of cell important subsystem for BKM-120 response, suggesting a com-
viability compared with single-gene knockout (Figure 5G). bination of PI3K + MAPK pathway inhibitors (BKM-120 + encor-
APC, whose subsystem (b-catenin destruction complex) was afenib). This combination significantly increased PFS across the
not identified by RLIPP (Table S5), did not show this same PDX panel compared with monotherapy (Figure 6D). Similarly,
pattern (Figure 5G). Similarly, etoposide did not synergize with DrugCell identified DNA damage response, signal transduction
the proteasome inhibitor bortezomib (Figure 5F), consistent by p53 class mediator resulting in cell-cycle arrest as an impor-
with the proteasome subsystem not being identified by DrugCell tant subsystem for abraxane response, suggesting combination
(Figure 5E). chemotherapy with an agent inducing DNA damage and cell-cy-
Further inspection suggested that the relationship between cle arrest (abraxane + gemcitabine). This combination similarly
PI3K signaling, ERK signaling, and etoposide sensitivity significantly improved PFS (Figure 6D). For the combinations
captured by DrugCell could be roughly approximated by a logic that were not prioritized by DrugCell (not in top 20% of subsys-
function integrating the mutational status of six genes (Figures tems by RLIPP), these combinations indeed failed to significantly
5H and 5I; STAR Methods). Among these, FLT1 (Das et al., improve PFS (Figures 6C and 6E). These results suggested that
2005) and PIN1 (Mathur et al., 2011) had previously been shown DrugCell has utility in guiding design of combination therapies in
to regulate etoposide response, whereas DUSP1, PIK3R4, SRC, patient tumors.
and RPS6KA6 had not. Considered individually, any one of these
genes was mutated rarely in cancer cell lines, with limited power DrugCell Predicts the Response of Estrogen Receptor-
to predict etoposide sensitivity versus resistance (mutation fre- Positive Metastatic Breast Cancer Patients to mTOR and
quencies 0.9%–8.9%; odds ratios <2; Figure 5J). Considered CDK4/6 Inhibitors
as an integrated circuit, however, these gene mutations Last, we sought to evaluate whether DrugCell could be used
converge on PI3K or ERK subsystems to create a powerful clinically to stratify cancer patients into responsive and non-
network-based biomarker of drug response (odds ratio 7.8; Fig- responsive patient populations. We obtained and analyzed
ure 5J). We also noted that these two pathways represent only a aggregated clinical trial data (Smyth et al., 2020) from 221 es-
portion of the full DrugCell model, which predicts etoposide trogen receptor (ER)-positive metastatic breast cancer patients
sensitivity with an odds ratio of 14.3. who had undergone multiple rounds of therapy, including an ER
antagonist (fulvestrant) in addition to treatment with an mTOR
DrugCell Improves Progression-Free Survival of inhibitor (everolimus) or CDK4/6 inhibitor (ribociclib). For this
Patient-Derived Xenograft Models analysis (STAR Methods), we predicted patient response to
We next wished to move beyond cell lines to predict and inter- either mTOR or CDK4/6 inhibition using our pre-trained Drug-
pret drug responses in the in vivo setting of patient-derived xeno- Cell model. We considered a patient to be DrugCell (+) if they
graft models (PDX; Figure 6A, STAR Methods). To do so, we ac- were predicted to be sensitive to either therapy and DrugCell
cessed the PDX Encyclopedia (Gao et al., 2015), in which 399 ( ) if they were predicted to be insensitive to both therapies.
PDX tumors of varying tissue types had been screened against DrugCell (+) patients had significantly longer overall survival
a total of 40 different monotherapies and 27 combination thera- than DrugCell ( ) patients (48.2 versus 33.6 months, p =
pies. The genotypes of each PDX had also been established 0.018; Figure 7A).
(Gao et al., 2015), which were provided to DrugCell to make We next interrogated the mechanisms underlying the differ-
response predictions to each monotherapy. We considered a ential sensitivity between DrugCell (+) and DrugCell ( ) patients
PDX tumor to be sensitive to a therapy (DrugCell (+)) if its pre- by performing an RLIPP analysis for both the mTOR and the
dicted AUC was beneath the median predicted for all PDX- CDK4/6 inhibitors. Notably, we found that both drug responses
drug pairs; otherwise this tumor was labeled as insensitive were modulated by ER-related subsystems (Figures 7B and
(DrugCell ( )). DrugCell (+) tumors demonstrated significantly 7C), consistent with their use in ER-positive breast cancer
longer progression-free survival (PFS) than DrugCell ( ) tumors (Hare and Harvey, 2017; Pernas et al., 2018). We also found
(2.19 versus 1.58 months, p = 9.4 3 10 10, log rank test). How- that the major mechanisms of action of both drugs were among
ever, given the overall insensitivity of these PDX tumors to mono- the top pathways, with PI3K signaling being especially impor-
therapy, corresponding to the short observed PFS observed for tant for response to mTOR inhibitors and CDK activity being
both DrugCell classes, we wished to evaluate how well DrugCell important for CDK4/6 inhibitor activity (Figures 7B and 7C).
is able to suggest effective drug combinations. We used RLIPP Interestingly, TOR signaling was also identified for CDK4/6 in-
scoring to rank subsystems by importance in mediating drug re- hibitors (Figure 7B), and CDK activity was identified for mTOR
sponses to six primary drugs, filtering this list to those that con- inhibitors (Figure 7C), suggesting that these drugs could be
tained secondary drug targets. The observed PFS of each of an effective combination therapy, a finding supported by recent
these (primary, secondary) combinations was used to estimate preclinical studies (Michaloglou et al., 2018; Occhipinti
the prediction sensitivity and specificity along a receiver oper- et al., 2020).
6 small molecules
RLIPP
Evaluate ROC
combination analysis
efficacy
Identify subsystems mediating response
399 PDX
and design drug combinations
tumor samples
B D E
1.0 Predicted effective by DrugCell Predicted ineffective by DrugCell
0.8
True positive rate
100 100
Untreated
Encorafenib Paclitaxel
80
80 NS Paclitaxel + LCL161
BKM120
0.4 *** BKM120 +
60 60
Encorafenib
0.2 40 40
auROC = 0.75 20 20
0.0
0.0 0.2 0.4 0.6 0.8 1.0 0 0
0 5 10 15 20 25 30 35 0 2 4 6 8 10
False negative rate Progression free survival (months) Progression free survival (months)
C
100 100
Untreated Untreated
5 2
Yes
Drug combination
80 80
Abraxane Trastuzumab
60
*** Abraxane +
60 NS
INC280
Gemcitabine INC280 +
Trastuzumab
40 40
1 5 20 20
No
0 0
0 2 4 6 8 10 0 5 10 15 20 25
Drug combination
effective in PDX tumor samples
signaling pathway
p = 0.018
20
Positive
regulation 15.3 ... Patient
response
viates parent subsystems at subsequent layers of
Median OS Median OS of CDK to CDK4/6i
33.6 months 48.2 months activity the hierarchy. RLIPP scores are displayed inside
n = 117 n = 104
0 each node.
Negative
regulation of 5.1 ...
0 20 40 60 80 100 120 TOR signaling (D) Scatterplot of the absolute (x axis) and per-
Overall survival (months) centage (y axis) difference in mutation frequencies
of genes between DrugCell (+) and DrugCell ( )
D Mutation E patients. Red points represent genes mutated more
Frequency frequently in DrugCell (+) patients. Blue points
0
Percent difference in mutation frequency
80 80
ESR1
(E) Survival curves for AKT1-mutant and wild-type
AKT1
60 ATM 60 patients treated with CDK4/6 or mTOR inhibitors in
NTRK1
RUNX1
TP53 any line of therapy. The p value indicates signifi-
40 CBFB 40 cance by log rank test.
CREBBP
7.3 months
p = 0.877
20 PIK3CA 20
Median OS Median OS
DrugCell (+) 35.8 months 43.1 months
DrugCell (–) n = 68 n = 153
0 0
0 5 10 15 20 25 35 0 20 40 60 80 100 120
Absolute difference in mutation frequency Overall survival (months)
Such engagement of human reasoning and follow-up experi- combinatorial number of pairwise and higher order drug combi-
mentation helps greatly to increase accountability and trust in nations necessary for training.
the predictions of a machine learning model. In contrast, conven- If the favorable performance observed in PDX samples (Fig-
tional black-box predictive modeling yields only a model ure 6) and ER-positive breast cancer patients (Figure 7) con-
output—the drug response—without further information by tinues in further clinical studies, DrugCell and its successors
which to build trust in the process. have the potential to substantially expand the set of clinically
DrugCell is a flexible model that is amenable to both auto- meaningful mutations. DrugCell translates the mutational status
mated and semi-automated combinatorial drug design. First, of approximately 3,000 genes into treatment recommendations.
the importance of each cellular subsystem is scored by DrugCell Although we have not fully studied which of these genes are
during a response to monotherapy. These important subsystems absolutely required for DrugCell prediction accuracy, RLIPP
are then annotated with second points of intervention, such as analysis suggests that many of them are—1,467 of the 2,086
the PI3K or ERK pathway in the response to etoposide (Figures subsystems are assigned relatively high importance (RLIPP
5E–5G). To follow up on this analysis, drug combinations can >10) for at least one drug, collectively covering 2,855 genes.
be selected automatically based on the druggable targets pre- This breadth of information contrasts with the fewer genes
sent in top DrugCell subsystems. Alternatively, if DrugCell is be- included in current cancer mutation panels such as MSK-
ing used in a clinical context, its recommendations can be pro- IMPACT or FoundationOne CDx (468 and 324 genes, respec-
vided to physician-scientists (e.g., a molecular tumor board) tively), which were designed to be queried manually by a physi-
who consider the recommended combinations in light of other cian (Cheng et al., 2015; Harris, 2017). Moreover, since we
biological knowledge not explicitly used in modeling, such as po- currently do not understand the clinical implications of the major-
tential toxicities and specific information about the case. After ity of cancer mutations, there is little consensus on what genes
careful consideration of all relevant information, the ultimate should be included in these pan-cancer mutation panels
treatment decision remains in the hands of the physician and (Nguyen and Gocke, 2017) or on how physicians should act on
the patient. Such need for human accountability is not unique the results. An increase in the number of clinically meaningful
to drug response prediction but is a central tenet of high-stakes cancer mutations, facilitated by interpretable machine learning
applications of machine learning (Rudin, 2019). models such as DrugCell, could further motivate the case for
Notably, previous models trained on monotherapy responses complete genomic sequencing of cancer patients (Katsanis
(Ammad-ud-din et al., 2017; Cortés-Ciriano et al., 2016; Iorio and Katsanis, 2013; Kuenzi and Ideker, 2020).
et al., 2016; Zhang et al., 2015) have not attempted to suggest Future work may also elect to integrate mutations with addi-
combination therapies. Rather, drug combinations have been tional levels of molecular information such as epigenetic states,
predicted using models of synergy trained directly on data gene expression, or microenvironmental influences. This inte-
from pairwise drug treatments (Preuer et al., 2018). This brute gration could be accomplished by pre-processing multiple
force approach faces the challenge of scalability, given the layers of information to derive a profile of gene scores for each
cell line or tumor, which would then be input to DrugCell. Extra B Assessing the Correspondence of Learned Subsystem
levels of information could also be integrated by adding new Embeddings to Measured Subsystem Activities
visible or conventional neural network branches alongside exist- B Differential Expression Analysis
ing ones. Alternatively, the effects of specific mutations on gene B Synergy Determinations
functions could be incorporated by a metric such as the Com- B Translation of Continuous Cell Response (AUC) to Bi-
bined Annotation-Dependent Depletion score (Rentzsch et al., nary Cell Response
2019) or by including gene structural domains as an additional B Identification of Boolean Logic Combinations
layer of the hierarchy. B PDX Tumor Analysis
Another opportunity is to structure the DrugCell system hierar- B Breast Cancer Patient Analysis
chy from ’omics data rather than literature curation (GO), as has
previously been done in budding yeast (Kramer et al., 2014; Ma SUPPLEMENTAL INFORMATION
et al., 2018). A data-driven, rather than literature-curated, hierar-
chy has the potential to incorporate new gene-subsystem asso- Supplemental Information can be found online at https://doi.org/10.1016/j.
ccell.2020.09.014.
ciations as well as entirely new subsystems into the model. It also
has the potential to revise and tailor subsystem definitions in GO,
ACKNOWLEDGMENTS
which are generic, to their particular contexts relevant to cancer.
For instance, we found that in its current form DrugCell contains We thank Dr. Gaudenz Danuser for his valuable insights into mechanisms of
a number of subsystems that have misleading labels based on taxol sensitivity. We gratefully acknowledge the support for this work provided
GO naming conventions. For example, Labyrinthine develop- by grants from the National Institutes of Health to T.I. (GM103504, CA209891,
ment was among the top subsystems for trametinib, which ES014811), J.F.K. (CA243885), and B.M.K. (CA212456).
was initially puzzling, but upon further inspection corresponds
to MAPK cascade genes with well-known involvement in cancer AUTHOR CONTRIBUTIONS
proliferation (e.g., MAP2K1, MAPK1, GRB2, FGFR2). Incorpo- B.M.K., J.P., S.H.F., J.M., J.F.K., and T.I. designed the study and developed
rating data-driven hierarchies into DrugCell provides a route to the conceptual ideas. B.M.K. collected all the input sources and additional
relabel such subsystems and revise their specific gene contents. data. J.P., B.M.K., and J.M. implemented the main algorithm. B.M.K. and
Finally, given that DrugCell inputs a full drug structure, it can J.P. implemented all other computational methods and analyses. B.M.K.,
potentially be used to design compounds de novo. Leveraging S.H.F., J.L., and K.S. performed all experiments and B.M.K. analyzed the re-
advancements in reinforcement learning for drug design (Zha- sults. B.M.K., J.P., S.H.F., J.F.K., and T.I. wrote the manuscript with sugges-
tions from other authors.
voronkov et al., 2019), it may then be possible to design com-
pounds for maximal efficacy against any given genomic
DECLARATION OF INTERESTS
background.
T.I. is a co-founder of Data4Cure, Inc., and has an equity interest. T.I. has an
equity interest in Ideaya BioSciences, Inc. The terms of this arrangement
STAR+METHODS
have been reviewed and approved by the University of California San Diego
in accordance with its conflict of interest policies.
Detailed methods are provided in the online version of this paper
and include the following: Received: April 8, 2020
Revised: August 7, 2020
d KEY RESOURCES TABLE Accepted: September 22, 2020
d RESOURCE AVAILABILITY Published: October 22, 2020
B Lead Contact
B Data and Code Availability REFERENCES
B Materials Availability
Ammad-ud-din, M., Khan, S.A., Wennerberg, K., and Aittokallio, T. (2017).
d EXPERIMENTAL MODEL AND SUBJECT DETAILS
Systematic identification of feature combinations for predicting drug response
B Cell Culture and Reagents with Bayesian multi-view multi-task linear regression. Bioinformatics 33,
d METHOD DETAILS i359–i368.
B Defining a Hierarchy of Genes and Cellular Sub- Ashburner, M., Ball, C.A., Blake, J.A., Botstein, D., Butler, H., Cherry, J.M.,
systems Davis, A.P., Dolinski, K., Dwight, S.S., Eppig, J.T., et al. (2000). Gene ontology:
B Pharmacogenomics Data Processing and Morgan tool for the unification of biology. The Gene Ontology Consortium. Nat. Genet.
Fingerprint Encoding 25, 25–29.
B Neural Network Configuration, Training and Evaluation Bani, M.R., Nicoletti, M.I., Alkharouf, N.W., Ghilardi, C., Petersen, D., Erba, E.,
B Implementation of Alternative Models for Comparison Sausville, E.A., Liu, E.T., and Giavazzi, R. (2004). Gene expression correlating
with response to paclitaxel in ovarian carcinoma xenografts. Mol. Cancer Ther.
of Predictive Performance
3, 111–121.
B Ranking Important Subsystems in DrugCell
Baptista, D., Ferreira, P.G., and Rocha, M. (2020). Deep learning for drug
B Comparing Important DrugCell Subsystems to Predic-
response prediction in cancer. Brief. Bioinform bbz171, https://doi.org/10.
tive Biomarkers Reported by Alternative Models 1093/bib/bbz171.
B Viability Assays
Barretina, J., Caponigro, G., Stransky, N., Venkatesan, K., Margolin, A.A., Kim,
B Combinatorial CRISPR-Cas9 Gene Knockouts and S., Wilson, C.J., Lehár, J., Kryukov, G.V., Sonkin, D., et al. (2012). The Cancer
Systematic Evaluation Cell Line Encyclopedia enables predictive modelling of anticancer drug sensi-
d QUANTIFICATION AND STATISTICAL ANALYSIS tivity. Nature 483, 603–607.
Costa, C., Wang, Y., Ly, A., Hosono, Y., Murchie, E., Walmsley, C.S., Huynh, Kuenzi, B.M., Rix, L.L.R., Kinose, F., Kroeger, J.L., Lancet, J.E., Padron, E.,
T., Healy, C., Peterson, R., Yanase, S., et al. (2020). PTEN loss mediates clin- and Rix, U. (2019). Off-target based drug repurposing opportunities for tivan-
ical cross-resistance to CDK4/6 and PI3Ka inhibitors in breast cancer. Cancer tinib in acute myeloid leukemia. Sci. Rep. 9, 606.
Discov. 10, 72–85. Li, J., Zhao, W., Akbani, R., Liu, W., Ju, Z., Ling, S., Vellano, C.P., Roebuck, P.,
Costello, J.C., Heiser, L.M., Georgii, E., Gönen, M., Menden, M.P., Wang, N.J., Yu, Q., Eterovic, A.K., et al. (2017). Characterization of human cancer cell lines
Bansal, M., Ammad-ud-din, M., Hintsanen, P., Khan, S.A., et al. (2014). A com- by reverse-phase protein arrays. Cancer Cell 31, 225–239.
munity effort to assess and improve drug sensitivity prediction algorithms. Nat. Liu, T., Sun, H., Zhu, D., Dong, X., Liu, F., Liang, X., Chen, C., Shao, B., Wang,
Biotechnol. 32, 1202–1212. M., Wang, Y., et al. (2017). TRA2A promoted paclitaxel resistance and tumor
Cover, T., and Hart, P. (1967). Nearest neighbor pattern classification. IEEE progression in triple-negative breast cancers via regulating alternative
Trans. Inf. Theor. 13, 21–27. splicing. Mol. Cancer Ther. 16, 1377–1388.
Das, B., Yeger, H., Tsuchida, R., Torkin, R., Gee, M.F.W., Thorner, P.S., Loewe, S. (1953). The problem of synergism and antagonism of combined
Shibuya, M., Malkin, D., and Baruchel, S. (2005). A hypoxia-driven vascular drugs. Arzneimittelforschung 3, 285–290.
endothelial growth factor/flt1 autocrine loop interacts with hypoxia-inducible Ma, J., Yu, M.K., Fong, S., Ono, K., Sage, E., Demchak, B., Sharan, R., and
factor-1a through mitogen-activated protein kinase/extracellular signal-regu- Ideker, T. (2018). Using deep learning to model the hierarchical structure
lated kinase 1/2 pathway in neuroblastoma. Cancer Res. 65, 7267–7275. and function of a cell. Nat. Methods. 15, 290–298.
Dincer, A.B., Celik, S., Hiranuma, N., and Lee, S.-I. (2018). DeepProfile: deep Ma, Y., Wang, L., Neitzel, L.R., Loganathan, S.N., Tang, N., Qin, L., Crispi, E.E.,
learning of cancer molecular profiles for precision medicine. BioRxiv. https:// Guo, Y., Knapp, S., Beauchamp, R.D., et al. (2017). The MAPK pathway regu-
doi.org/10.1101/278739. lates intrinsic resistance to BET inhibitors in colorectal cancer. Clin. Cancer
Eskiocak, B., McMillan, E.A., Mendiratta, S., Kollipara, R.K., Zhang, H., Res. 23, 2027–2037.
Humphries, C.G., Wang, C., Garcia-Rodriguez, J., Ding, M., Zaman, A., et al. Mathur, R., Chandna, S., N Kapoor, P., and S Dwarakanath, B. (2011). Peptidyl
(2017). Biomarker accessible and chemically addressable mechanistic sub- prolyl isomerase, Pin1 is a potential target for enhancing the therapeutic effi-
types of BRAF melanoma. Cancer Discov. 7, 832–851. cacy of etoposide. Curr. Cancer Drug Targets 11, 380–392.
Esteva, A., Robicquet, A., Ramsundar, B., Kuleshov, V., DePristo, M., Chou, Menden, M.P., Iorio, F., Garnett, M., McDermott, U., Benes, C.H., Ballester,
K., Cui, C., Corrado, G., Thrun, S., and Dean, J. (2019). A guide to deep P.J., and Saez-Rodriguez, J. (2013). Machine learning prediction of cancer
learning in healthcare. Nat. Med. 25, 24. cell sensitivity to drugs based on genomic and chemical properties. PLoS
Gao, H., Korn, J.M., Ferretti, S., Monahan, J.E., Wang, Y., Singh, M., Zhang, One 8, https://doi.org/10.1371/journal.pone.0061318.
C., Schnell, C., Yang, G., Zhang, Y., et al. (2015). High-throughput screening Michaloglou, C., Crafter, C., Siersbaek, R., Delpuech, O., Curwen, J.O.,
using patient-derived tumor xenografts to predict clinical trial drug response. Carnevalli, L.S., Staniszewska, A.D., Polanska, U.M., Cheraghchi-Bashi, A.,
Nat. Med. 21, 1318–1325. Lawson, M., et al. (2018). Combined inhibition of mTOR and CDK4/6 is
Garnett, M.J., Edelman, E.J., Heidorn, S.J., Greenman, C.D., Dastur, A., Lau, required for optimal blockade of E2F function and long-term growth inhibition
K.W., Greninger, P., Thompson, I.R., Luo, X., Soares, J., et al. (2012). in estrogen receptor–positive breast cancer. Mol. Cancer Ther. 17, 908–920.
Systematic identification of genomic markers of drug sensitivity in cancer cells. Moos, P.J., and Fitzpatrick, F.A. (1998). Taxane-mediated gene induction is in-
Nature 483, 570–575. dependent of microtubule stabilization: induction of transcription regulators
Hare, S.H., and Harvey, A.J. (2017). mTOR function and therapeutic targeting and enzymes that modulate inflammation and apoptosis. Proc. Natl. Acad.
in breast cancer. Am. J. Cancer Res. 7, 383–404. Sci. U S A 95, 3896–3901.
Rentzsch, P., Witten, D., Cooper, G.M., Shendure, J., and Kircher, M. (2019). Yang, W., Soares, J., Greninger, P., Edelman, E.J., Lightfoot, H., Forbes, S.,
CADD: predicting the deleteriousness of variants throughout the human Bindal, N., Beare, D., Smith, J.A., Thompson, I.R., et al. (2013). Genomics of
genome. Nucleic Acids Res. 47, D886–D894. Drug Sensitivity in Cancer (GDSC): a resource for therapeutic biomarker dis-
covery in cancer cells. Nucleic Acids Res. 41, D955–D961.
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 Yeh, P.J., Hegreness, M.J., Aiden, A.P., and Kishony, R. (2009). Drug interac-
and microarray studies. Nucleic Acids Res. 43, e47. tions and the evolution of antibiotic resistance. Nat. Rev. Microbiol. 7,
460–466.
Robinson, M.D., and Oshlack, A. (2010). A scaling normalization method for
differential expression analysis of RNA-seq data. Genome Biol. 11, R25. Yin, Y., Sun, M., Zhan, X., Wu, C., Geng, P., Sun, X., Wu, Y., Zhang, S., Qin, J.,
Rogers, D., and Hahn, M. (2010). Extended-connectivity fingerprints. J. Chem. Zhuang, Z., et al. (2019). EGFR signaling confers resistance to BET inhibition in
Inf. Model. 50, 742–754. hepatocellular carcinoma through stabilizing oncogenic MYC. J. Exp. Clin.
Cancer Res. 38, 83.
Rudin, C. (2019). Stop explaining black box machine learning models for high
stakes decisions and use interpretable models instead. Nat. Mach. Intell. 1, Yu, M.K., Ma, J., Fisher, J., Kreisberg, J.F., Raphael, B.J., and Ideker, T. (2018).
206–215. Visible machine learning for biomedicine. Cell 173, 1562–1565.
Sakellaropoulos, T., Vougas, K., Narang, S., Koinis, F., Kotsinas, A., Polyzos, Zeng, X., Zhu, S., Liu, X., Zhou, Y., Nussinov, R., and Cheng, F. (2019).
A., Moss, T.J., Piha-Paul, S., Zhou, H., Kardala, E., et al. (2019). A deep deepDR: a network-based deep learning approach to in silico drug reposition-
learning framework for predicting response to therapy in cancer. Cell Rep. ing. Bioinformatics 35, 5191–5198.
29, 3367–3373.e4. Zhang, N., Wang, H., Fang, Y., Wang, J., Zheng, X., and Liu, X.S. (2015).
Sanjana, N.E., Shalem, O., and Zhang, F. (2014). Improved vectors and Predicting anticancer drug responses using a dual-layer integrated cell line-
genome-wide libraries for CRISPR screening. Nat. Methods 11, 783–784. drug network model. PLoS Comput. Biol. 11, e1004498.
Seashore-Ludlow, B., Rees, M.G., Cheah, J.H., Cokol, M., Price, E.V., Coletti, Zhavoronkov, A., Ivanenkov, Y.A., Aliper, A., Veselov, M.S., Aladinskiy, V.A.,
M.E., Jones, V., Bodycombe, N.E., Soule, C.K., Gould, J., et al. (2015). Aladinskaya, A.V., Terentiev, V.A., Polykovskiy, D.A., Kuznetsov, M.D.,
Harnessing connectivity in a large-scale small-molecule sensitivity dataset. Asadulaev, A., et al. (2019). Deep learning enables rapid identification of potent
Cancer Discov. 5, 1210–1223. DDR1 kinase inhibitors. Nat. Biotechnol. 37, 1038–1040.
We give authors the same level of editorial excellence and support that they expect from flagship
journals such as Cell and foster deeper engagement between scientists, clinicians, and industry
professionals, increasing the reach and visibility of research that can positively impact human health.
Article
Conserved Interferon-g Signaling
Drives Clinical Response to Immune
Checkpoint Blockade Therapy in Melanoma
Catherine S. Grasso,1,2,* Jennifer Tsoi,1 Mykola Onyshchenko,1,17 Gabriel Abril-Rodriguez,1 Petra Ross-Macdonald,3
Megan Wind-Rotolo,3 Ameya Champhekar,1 Egmidio Medina,1 Davis Y. Torrejon,1 Daniel Sanghoon Shin,1 Phuong Tran,1
Yeon Joo Kim,1 Cristina Puig-Saus,1,5 Katie Campbell,1 Agustin Vega-Crespo,1 Michael Quist,1 Christophe Martignier,4
Jason J. Luke,6,18 Jedd D. Wolchok,5,7 Douglas B. Johnson,8 Bartosz Chmielowski,1 F. Stephen Hodi,5,9
Shailender Bhatia,10 William Sharfman,11 Walter J. Urba,12 Craig L. Slingluff, Jr.,13 Adi Diab,14 John B.A.G. Haanen,15
Salvador Martin Algarra,16 Drew M. Pardoll,11 Valsamo Anagnostou,11 Suzanne L. Topalian,11 Victor E. Velculescu,11
Daniel E. Speiser,4 Anusha Kalbasi,1 and Antoni Ribas1,5,19,*
1Jonsson Comprehensive Cancer Center at the University of California, Los Angeles (UCLA), Los Angeles, CA, USA
2Cedars-Sinai Medical Center, Los Angeles, CA, USA
3Translational Bioinformatics, Bristol-Myers Squibb, Hopewell, NJ, USA
4University of Lausanne, Lausanne, Switzerland
5Parker Institute for Cancer Immunotherapy, San Francisco, CA, USA
6University of Chicago, Chicago, IL, USA
7Memorial Sloan Kettering Cancer Center, New York, NY, USA
8Vanderbilt-Ingram Cancer Center, Nashville, TN, USA
9Dana Farber Cancer Institute, Boston, MA, USA
10University of Washington, Seattle, WA, USA
11Bloomberg-Kimmel Institute for Cancer Immunotherapy and Sidney Kimmel Comprehensive Cancer Center, Johns Hopkins University
SUMMARY
We analyze the transcriptome of baseline and on-therapy tumor biopsies from 101 patients with advanced
melanoma treated with nivolumab (anti-PD-1) alone or combined with ipilimumab (anti-CTLA-4). We find
that T cell infiltration and interferon-g (IFN-g) signaling signatures correspond most highly with clinical
response to therapy, with a reciprocal decrease in cell-cycle and WNT signaling pathways in responding bi-
opsies. We model the interaction in 58 human cell lines, where IFN-g in vitro exposure leads to a conserved
transcriptome response unless cells have IFN-g receptor alterations. This conserved IFN-g transcriptome
response in melanoma cells serves to amplify the antitumor immune response. Therefore, the magnitude
of the antitumor T cell response and the corresponding downstream IFN-g signaling are the main drivers
of clinical response or resistance to immune checkpoint blockade therapy.
INTRODUCTION with melanoma and other cancers (Ribas and Wolchok, 2018;
Sharma and Allison, 2015b). Pathological studies performed
Immune checkpoint blockade (ICB) therapy with antibodies tar- in tumor biopsies from treated patients support that clinical re-
geting the cytotoxic T lymphocyte-associated protein 4 (CTLA- sponses induced by the use of ICB are mediated by tumor-infil-
4) or the programmed cell death-1 receptor (PD-1) blocks two trating T cells that have been re-activated by inhibiting these
main negative regulators of antitumor immune responses immune checkpoints (Sharma et al., 2019; Tumeh et al.,
(Curran et al., 2010; Sharma and Allison, 2015a; Wei et al., 2014). Transcriptomic and genomic sequencing of baseline
2017) and induces durable responses in a subset of patients and on-therapy biopsies from patients treated with ICB allow
500 Cancer Cell 38, 500–515, October 12, 2020 ª 2020 Elsevier Inc.
ll
Article
A RNAseq analysis B
7 7
Nivo
(Ipi- 10 13
99 experi CRPR
101 enced)
On-
Baseline 13 22
treatment
13
Nivo 10
Reasons for Sample Reasons for Sample 8
(Ipi- SD
Removal Removal naïve) 8
• 1 Biopsied at 24h • 1 Unscheduled 3
• 1 Unscheduled 16
• 0 Insufficient coverage
• 0 Insufficient coverage • 4 QC failure (Low
• 5 QC failure (No Pearson Correlation/ 13
Melanin expression/ Male with XIST
Low Pearson expression) 22 PD
Correlation/ Male with Nivo/
• 2 NE by RECIST Ipi 16
XIST expression) • 7 Ocular melanoma
• 2 NE by RECIST 3
• 8 Ocular melanoma 9 9
outcomes (Figure S2A). When analyzing OS and PFS according change from baseline to on-therapy was greater in biopsies of
to tumor response assessment, patients with CRPR unsurpris- patients with a response to therapy, and there were increases
ingly had prolonged survival compared with those with stable in T cell infiltrates in biopsies of patients whether or not they
disease (SD) or progressive disease (PD) (Figure S2B). Based had an objective clinical response to therapy, and regardless
on these observations, for the RNA-seq analyses we compared of receiving combination therapy or nivolumab alone (Figures
patients receiving combination therapy to those receiving nivolu- 2A and 2B).
mab monotherapy (merging the groups that had or had not pre- With the goal to define the main drivers explaining the overall
viously received ipilimumab), and clinical response was defined biopsy data, we then analyzed the whole-transcriptome RNA-
as CRPR compared with SD or PD. seq dataset using principal-component analysis. Analysis of
the first two principal components showed that the signature
The Dominant Signal Associated with Response or for T cell score derived from MCP-counter RNA-seq data decon-
Resistance to ICB Therapy in Patient Biopsies Is volution was the main factor organizing the data, and it segre-
Mediated by T Cells gated with response to therapy (with the exception of 19 outlier
Consistently, it has been shown that high levels of CD8 T cell samples, 12 of which were matched from the same patients
infiltration of tumors at baseline and on-treatment are predictive with baseline and on-therapy biopsies, Figure 2C). Of note, in
of clinical response to ICB in patients with advanced melanoma previous work we had validated the MCP-counter T cell score
(Chen et al., 2016; Gide et al., 2019; Tumeh et al., 2014). We with pathological analysis of immune infiltrates in corresponding
used the method denominated microenvironment cell popula- patient biopsies (Grasso et al., 2018). In the CheckMate 038 bi-
tions-counter (MCP-counter) (Becht et al., 2016) for RNA-seq opsy dataset, the best organization of the samples was accord-
data deconvolution to define immune cell types. We applied ing to the degree of T cell infiltration regardless of whether this
an optimal pooled t test, since only a subset of the samples was observed in baseline or on-therapy biopsies, with or without
were paired. Using these approaches, we documented that bi- prior treatment with ipilimumab, single or combination ICB ther-
opsies of patients with a response to therapy had higher base- apy, as well as clinical response or no response to therapy, with
line levels of T cells than those with PD (Figure 2A). They also the paired CRPR samples showing the most consistent
had significantly higher baseline levels of B lineage cells, alignment along the T cell score vector (Figures 2C and 2D). Fig-
myeloid dendritic cells, and natural killer cells (Figure 2A). The ure S3 shows the degree to which the T cell score dominates the
Table 1. Baseline Patient Characteristics data, with the large number of genes correlated (n = 2,047) and
anti-correlated (n = 1,087) with this score based on a Pearson
n (%)
N = 101 correlation cutoff of ±0.3. This supplemental figure also includes
information about the effect of the melanoma subtype (cuta-
Age (years)
neous, mucosal, or acral) on the tumor transcriptome. The coor-
Median 56
dinated alignment of the correlated and anti-correlated genes
Range 22–89 on-treatment is reflected in the significant upregulation of the
Sex, n (%) correlated genes and the significant downregulation from base-
Male 58 (57.4) line to on-treatment in CRPR biopsies (optimal pooled t test p <
Female 43 (42.6) 1 3 10 4 and p < 1 3 10 4, respectively, Figure 2C). The corre-
ECOG performance status, n (%) lated genes accounted not only for systematic changes in the tu-
0 70 (69.3) mor cell gene expression, but also in the tumor microenviron-
ment. Except for neutrophils, the MCP-counter immune cell
1 29 (28.7)
signatures were highly correlated with the T cell signature (R2
Not reported 2 (2.0)
R0.6, Figures S4A and S4B). Use of t-SNE embedding to cluster
Stage at study entry, n (%) genes with their closest MCP-counter immune cell signature re-
III 8 (7.9) sulted in only two main clusters: genes that correlated with the
IV 93 (92.1) T cell score and genes that were anti-correlated (Figure S4C).
Prior anti-CTLA-4 therapy, n (%) These data indicate a truly coordinated immune response,
Yes 30 (29.7) including large numbers of genes, which was dominated by
No 71 (70.3) changes in T cells. We acknowledge that this conclusion is
limited by being derived from bulk RNA-seq analyses as
BRAF mutation status, n (%)
opposed to single-cell RNA-seq. Based on these analyses, we
Positive 33 (32.7)
conclude that ICB therapy has the potential to increase T cell in-
Negative 61 (60.4) filtrates regardless of whether there are clinical responses to
Not reported 7 (6.9) therapy, and that a greater degree of T cell infiltrate increases
PD-L1 status at baseline, n (%) from baseline to on-therapy biopsies is associated with
PD-L1 negative 30 (29.7) improved response to therapy.
(TPS baseline = 0)
PD-L1 positive 48 (47.5) Interferon-g Response Genes Play an Integral Role in
(TPS baseline >0) Response and Resistance to Therapy as Mediators of
NA (TPS baseline = NA) 23 (22.8) the T Cell Program
Metastatic staging, n (%) We examined the expression of effector cytokines and toxic
M0 2 (2.0)
granules induced by TCR engagement with cognate antigen,
including the cytotoxic molecules perforin and granzyme B, tu-
M1A 17 (16.8)
mor necrosis factor alpha (TNF-a) family members FAS, TRAIL
M1B 15 (14.9)
(TNFSF10), and TNF-a, and interferon-g, applying an optimal
M1C 53 (52.5) pooled t test. The expression of perforin, granzyme B, TRAIL,
Not reported (stage III)/unknown 14 (13.9) and TNF-a followed the pattern of expression of interferon-g,
Brain metastases, n (%) and was higher in biopsies of patients with a clinical response
Yes (*) 13 (12.9) to therapy (Figure S5). As exposure to most of these molecules
No 80 (79.2) results in cytotoxic death of cancer cells, we reasoned that the
Not reported (stage III) 8 (7.9)
cancer cell expression of interferon response genes may be rele-
vant to attracting other immune cells into the tumor microenvi-
Lactate dehydrogenase, n (%)
ronment and amplifying the antitumor immune response once
Normal (%ULN) 72 (71.3)
it is initiated.
Elevated (>ULN, %2*ULN) 24 (23.8)
Highly elevated (>2*ULN) 4 (4.0) Melanoma Cells Have a Uniform Gene Set Response to
Not reported 1 (1.0) Interferon-g Provided That They Can Signal through the
Melanoma primary site, n (%) Interferon-g Receptor
Acral 4 (4.0) With the goal of addressing the question if differential responses
Cutaneous 71 (70.3) to ICB may be due to a different inherent ability of melanoma
cells from different patients to respond to interferon-g, we set
Mucosal 11 (10.9)
Other 15 (14.9)
ECOG, Eastern Cooperative Oncology Group; M1a, metastases to tration of lactate dehydrogenase; PD-L1, programmed death
skin, subcutaneous tissues, or distant lymph nodes; M1b, metasta- ligand 1.
ses to lung; M1c, metastases to all other visceral sites or distant *In addition to all patients enrolled in part 4, a number of patients in parts
metastases at any site combined with an increased serum concen- 1–3 also had reported brain metastases.
B C D
Figure 2. Immune Cell Infiltration in Patient Samples from Clinical Study CheckMate 038
(A) Boxplot the MCP-counter T cell score according to response to therapy combining all treatment groups using an optimal pooled t test since data are paired
and unpaired.
(B) Boxplots showing the average expression of the genes correlated with the T cells broken down by response and pre- and on-treatment using an optimal
pooled t test since data are paired and unpaired.
out to examine the transcriptome changes in a panel of ex-vivo- To identify a set of genes up- and downregulated in response
cultured melanoma cells exposed to interferon-g. For this study, to interferon-g in typical melanoma cell lines, we used all the
we used 57 previously established and described cell lines from samples that were not experimentally modified (excluding cases
cultures of human melanoma biopsies or surgical resections with JAK1, JAK2, and B2M loss generated by CRISPR-Cas9, as
(Atefi et al., 2014; Neubert et al., 2017; Shin et al., 2017; Tsoi well as the AR samples but including naturally occurring JAK and
et al., 2018) and the primary melanocyte line, HeMa, as a non- B2M loss cases). We performed a paired t test between cell lines
malignant pigmented cell control. This cell line panel included before and after treatment with interferon-g and observed a sig-
three variants of corresponding parental melanoma cell lines nificant decrease in the expression of 1,176 genes and increase
that had been rendered resistant to BRAF inhibitors by contin- in the expression of 549 genes after applying a false discovery
uous in vitro drug exposure, denoted as AR for acquired resis- rate (FDR) cutoff of 0.01. These changes clustered with the
tance (Atefi et al., 2014; Nazarian et al., 2010; Tsoi et al., 2018); changes in the primary melanocyte line HeMa, included as a
three cell lines that had developed Janus kinase 1 (JAK1) or normal control, indicating that normal interferon-g signaling
JAK2 loss-of-function mutations in patients; as well as variants was intact in all of these cases (Figure S8B). This transcriptome
of parental cell lines that had JAK1 (N = 5), JAK2 (N = 5), or signature includes previously reported interferon-g response
beta-2 microglobulin (B2M) (N = 2) loss of function generated genes, melanoma-specific responses, as well as a deeper look
in vitro through CRISPR-Cas9 gene editing, as these are genetic into the genes whose expression decreases in response to inter-
events known to be associated with resistance to anti-PD-1 ther- feron-g (Figure S9).
apy (Garcia-Diaz et al., 2017; Gettinger et al., 2017; Sade-Feld-
man et al., 2017; Shin et al., 2017; Sucker et al., 2017; Torrejon Interferon-g Triggers Melanoma Cells to Increase
et al., 2020; Zaretsky et al., 2016). These cell lines were exposed Expression of Interferon Pathway Genes, Antigen
to 5 ng/mL of interferon-g for 6 h to assess immediate whole- Presenting Machinery, and T Cell-Attracting
transcriptome changes by RNA-seq analysis. Chemokines
We found that a cell line’s gene expression profile was most We next focused the analysis on changes in known genes that
correlated with itself at baseline and upon exposure to inter- are related to interferon-g exposure, including interference with
feron-g, indicating that cell line identity is a major contributor cell proliferation (the first functional effect that led to the descrip-
to gene expression (Figure S6). In addition, we observed that tion of interferons), upregulation of antigen presentation machin-
the baseline samples were not correlated with each other, nor ery, enhancement of immune cell-attracting chemokines, as well
were the interferon-g-exposed samples correlated with each as genes involved in positive and negative downstream inter-
other (Figure S6). However, for the 46 cell lines that responded feron signaling pathways (Bach et al., 1997). Figure 3A shows
to interferon-g, the changes between on-treatment and baseline the response as fold changes, while Figure S10 shows it in terms
for interferon-g were highly correlated as a result of consistent of baseline and on-treatment gene expression. The overall
large changes in interferon-g response genes (Figure S7). Fig- change in transcripts involved in cell proliferation was low after
ure S8A shows the high level of concordance among expression 6 h of in vitro interferon-g exposure (Figure 3A). There was evi-
changes of known interferon-g response genes, making it dence of increased expression of multiple HLA class I and II
possible to identify outlier gene expression changes, such as genes, with the most consistent increase being in HLA-E. The
the two samples whose JAK2 expression decreased signifi- two major clusters in the data were defined by whether or not
cantly compared with the rest of the samples. Analysis of all HLA class II genes are expressed at baseline. There was a
increased and decreased gene expression showed that there very strong and consistent increase in the expression of addi-
were two broad groups of cell lines: the great majority (n = 46) tional genes in the antigen presentation pathway, in particular
that could signal through the interferon-g receptor and had large NLRC5 (also known as HLA class I transactivator) and CIITA
changes in the expression of the same set of interferon-g (HLA class II transactivator), TAP transporters and proteasome
response genes; and a distinct set of 12 cell lines that had subunits. Another set of genes with large increases were those
JAK1 or JAK2 natural or induced loss-of-function mutations involved in the interferon signaling pathway itself, representing
and were largely unable to signal through the interferon-g recep- an amplification of the interferon-g signal, including JAK2 (but
tor (Figures 3 and S8B). Of note, there were no major differences not JAK1), several signal transductors and activators of tran-
in gene expression profiles induced by interferon-g exposure in scription and interferon response factors (IRFs), in particular
cell lines with acquired resistance to BRAF inhibitors or in B2M the transcription factor IRF1, which serves as a well-defined pos-
knockout cell lines. Therefore, this large panel of melanoma itive control gene of interferon-g response (Garcia-Diaz et al.,
cell lines, each of which had a unique gene expression signature, 2017), going up in all the samples with JAKs intact (Figure 3B).
responded consistently to interferon-g exposure. With this gene group, there was also an increase in negative
(C) Principal-component analysis showing all RNA-seq samples, paired and unpaired, plotted on the first two components. The size of the data point is pro-
portional to the T cell score. Open circles correspond to pre-treatment samples, while closed circles correspond to on-treatment samples. The data points are
colored by response (red, PD; green, SD; blue, CRPR). The vector for the T cell score is plotted as well. Outlier samples are labeled by case. The 12 paired outlier
samples were either due to quality control issues that had not been detected, or may be due to an actual different biology. For example, cases 48, 20038, and
30022 all had interferon-g signaling either going down or not changing on-treatment, while cases 30004 and 20001 have G2M cell-cycle genes that either increase
or did not respond.
(D) Boxplots of MCP-counter immune cell deconvolution types according to response to therapy.
Mixed t test for paired and unpaired samples using an optimal pooled t test since data are paired and unpaired (*p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001).
See also Figures S3–S5.
Cancer Cell 38, 500–515, October 12, 2020 505
ll
Article
regulators of the pathway including SOCS genes, as well as PD- FDR cutoff of 0.05, Figure S13). The SD and PD samples did
L1 (CD274), which are also well characterized interferon-g imme- not change consistently from pre-treatment to on-treatment,
diate response genes (Garcia-Diaz et al., 2017; Shin et al., 2017) except for four genes, ADI1, ARHGEF9, POU6F2, and RGPD4,
(Figure 3C). Chemokines, in particular CXCL9, CXCL10, and which were significantly different on-treatment relative to pre-
CXCL11, were increased more consistently than immune treatment SD cases (after applying an optimal pooled t test
effector molecules and physiologically serve to attract more with an FDR cutoff of 0.05). Only one gene in baseline samples
T cells in response to interferon-g exposure. These increases was associated with response, as expression of GPR31 was
in gene expression were very inconsistent or non-existent in significantly different between biopsies of patients with CRPR
the 12 cell lines that had JAK1 or JAK2 natural or induced or PD (p = 8.4 3 10 7, after applying an optimal pooled t test
loss-of-function mutations. Overall, melanoma cell lines have a with an FDR cutoff of 0.05).
near uniform transcriptome response to interferon-g exposure, To understand the major pathways driving response or lack of
with increases in transcripts for antigen presentation, interferon response, we applied gene set enrichment analysis (GSEA) using
pathway, and chemokine genes, that is lost if there are loss-of- the hallmark gene signatures to identify enrichment of gene sets
function mutations in interferon-g receptor pathway signaling changing on-treatment in CRPR (Figure 5A). We identified a set
at the level of JAK1 or JAK2. of gene signatures that were also present when we applied
GSEA to genes significantly altered in cell lines after interferon-
Analysis of Interferon-g Response Genes in Patient g treatment (Figure 5B). When analyzed individually for the two
Biopsies Reveals Increased HLA Expression as the top-scoring pathways according to response to therapy, most
Major Difference between Response and Lack of biopsies with CRPR had increases in interferon-g genes and de-
Response to ICB Therapy creases in G2M checkpoint genes, while biopsies from patients
We analyzed the CheckMate 038 patient biopsies for interferon- with PD had increases or decreases of these gene sets with no
g response genes before and during ICB therapy. For this anal- apparent directionality (Figure S14). This implies that inter-
ysis, we acknowledge that cells other than melanoma cells may feron-g response, presumably resulting from higher tumor anti-
be responsible for the expression of interferon-g response genes gen-specific T cell infiltration, is the major driver of clinical
when analyzing bulk RNA-seq from biopsies as opposed to the response in the CRPR samples. Despite the caveat of the mixed
experiment using human cell lines or single-cell RNA-seq from cell content of the tumor microenvironment contributing to the
fresh tumor biopsies. Biopsies of patients with PD and SD bulk RNA-seq transcriptome analysis, the CRPR and SD cases
included samples with and without expression of interferon had high increases in interferon-g response gene sets that
response genes at baseline (Figure 4). There was evidence of were consistent with the interferon-g 6 h in vitro human cell line
an increase in the frequency and intensity of gene expression data, while there were PD cases for which all the interferon-g
in on-therapy biopsies, in particular for chemokines and immune genes were downregulated (Figures S14 and S15). Similarly,
effector molecules (Figure 4). Biopsies of patients with CRPR for the HALLMARK_G2M_CHECKPOINT gene set, most of the
had higher baseline expression of interferon response genes, CRPR and SD cases were decreased, consistent with the inter-
which increased substantially in the on-therapy biopsies (Fig- feron-g 6-h in vitro exposure data (Figures S14 and S15). Howev-
ure 4). The largest difference between the PD and SD groups er, there were biopsies from multiple patients with PD and one
compared with the CRPR group was in the marked increase in with SD for which all the cell-cycle genes increased. Other indi-
antigen presentation pathway genes, in particular in HLA class vidual Hallmark gene sets similarly revealed fold changes
I and class II genes, B2M, TAP1, TAP2, NLRC5, and CIITA. After consistent with interferon-g treatment for CRPR cases, while
normalizing the data by leukocyte common antigen ([LCA], CD45 changes for the PD cases were the opposite of what would be
gene) expression to account for changes in immune cell infil- expected based on interferon-g biology. This is true both for
trates, the HLA class I gene expression did not increase. This additional immune gene sets (Figures S16 and S17) and for addi-
does not mean that it did not increase in the tumor, but rather tional cell-cycle gene sets for E2F targets and mitotic spindle
that the predicted increase in HLA class I expression due to inter- (Figure S18). The main observation was that the gene expression
feron-g signaling was not large enough to be detected against profile of biopsies from patients with clinical response is well
the background of other changes in the tumor microenvironment defined and consistent with exposure to interferon-g, while there
(Figures S11 and S12). Therefore, the main difference between was a large diversity of gene expression changes in SD and
interferon-g genes in biopsies of patients with response and PD cases.
resistance to ICB was an increase in antigen presentation genes
in biopsies obtained during therapy, a conclusion that can be Changes in WNT, MYC, and T Cell Exclusion Programs as
confounded by being derived from bulk RNA-seq analyses. Tumor-Intrinsic Responses to Interferon-g
Several publications have recently shown that high WNT
CRPR Cases Were Associated with Consistent Changes signaling contributes to immune exclusion (Grasso et al., 2018;
in Hallmark Pathways, While PD Cases Involve Different Luke et al., 2019; Nsengimana et al., 2018; Spranger et al.,
Pathways for Each Sample 2015; Spranger and Gajewski, 2018). We used a nine-gene
We found a large set of genes that were significantly different be- RNA-seq-based WNT score to assess the level of WNT signaling
tween on-treatment CRPR and on-treatment PD, likely driven in in each biopsy (Nsengimana et al., 2018). The WNT score is the
part by the large number of consistent changes in the on-treat- geometric average of APC, APC2, CTNNB1, MYC, SOX11,
ment CRPR relative to the pre-treatment CRPR (1,935 up and SOX2, TCF12, TCF7, and VEGFA. Analysis of WNT gene score
3,503 down after applying an optimal pooled t test with an revealed that biopsies of patients with CRPR exhibited a
Figure 4. Interferon-g-Induced Changes in Gene Expression in Specimens in Clinical Study CheckMate 038
Heatmap of key interferon-g response genes in biopsies obtained at baseline and on-treatment with immune blockade therapy, separated by response to
therapy. Samples are ordered by T cell score and annotated by ipilimumab naive status, monotherapy versus combination therapy and melanoma subtype.
Genes are organized by class: anti-proliferative (black), antigen presentation (white), chemoattractants (cyan), cytotoxic effectors (orange), feedback/signaling
(purple).
See also Figures S11 and S12.
A B
Significantly Up and Down Genes in T-test for CRPR Pre to Post Significantly Up and Down Genes in T-test for IFNG 6h Treatment
HALLMARK_INTERFERON_GAMMA_RESPONSE HALLMARK_INTERFERON_GAMMA_RESPONSE
HALLMARK_ALLOGRAFT_REJECTION HALLMARK_INTERFERON_ALPHA_RESPONSE
HALLMARK_INFLAMMATORY_RESPONSE HALLMARK_TNFA_SIGNALING_VIA_NFKB
HALLMARK_INTERFERON_ALPHA_RESPONSE HALLMARK_ALLOGRAFT_REJECTION
HALLMARK_COMPLEMENT HALLMARK_COMPLEMENT
HALLMARK_TNFA_SIGNALING_VIA_NFKB HALLMARK_INFLAMMATORY_RESPONSE
HALLMARK_IL2_STAT5_SIGNALING HALLMARK_IL6_JAK_STAT3_SIGNALING
HALLMARK_IL6_JAK_STAT3_SIGNALING HALLMARK_IL2_STAT5_SIGNALING
HALLMARK_MYC_TARGETS_V1 HALLMARK_G2M_CHECKPOINT
HALLMARK_E2F_TARGETS HALLMARK_MYC_TARGETS_V1
HALLMARK_G2M_CHECKPOINT HALLMARK_E2F_TARGETS
HALLMARK_MYC_TARGETS_V2 HALLMARK_MITOTIC_SPINDLE
HALLMARK_MTORC1_SIGNALING HALLMARK_UV_RESPONSE_DN
HALLMARK_OXIDATIVE_PHOSPHORYLATION HALLMARK_ESTROGEN_RESPONSE_EARLY
HALLMARK_DNA_REPAIR HALLMARK_MTORC1_SIGNALING
0 20 40 60 80 100 120 140 160 180 200 0 20 40 60 80 100 120 140 160 180 200
# Genes in Overlap # Remaining Genes in Gene Set # Genes in Overlap # Remaining Genes in Gene Set
Figure 5. Global Changes in Gene Expression in Clinical Study CheckMate 038 Consistent with Interferon-g-Induced Changes in Biopsies
from Patients with Response but Not without Response to Therapy
Gene set enrichment analysis (GSEA) using hallmark gene sets for significantly up- or downregulated genes for CRPR post-treatment relative to pre-treatment for
clinical dataset CheckMate 038 (A) and significantly up- or downregulated genes for interferon-g 6-h treatment for the unmodified human cell lines (B). At most
2,000 of most significant genes by q value submitted to GSEA for each class (maximum set by GSEA).
See also Figures S13–S18.
consistent significant decrease in WNT score in the on-therapy mor. Figure S21 shows a heatmap of these two sets of genes for
biopsies, while there was no change in the score in biopsies of the CheckMate 038 dataset. CRPR cases showed a significant
patients with SD or PD (after applying an optimal pooled t test) decrease in the immune excluded-high genes (p < 1.4 3 10 4,
(Figure 6A). To explore if this effect was correlated with the de- after applying an optimal pooled t test, Figures 6F and 6G). These
gree of change in T cell infiltration, we plotted the WNT score genes did not change significantly in response to interferon-g in
with T cell infiltration by RNA-seq deconvolution with baseline our in vitro testing in melanoma cell lines (the immune excluded-
and on-therapy biopsies tracked with arrows. Again, we down genes do go up significantly, p < 0.006, after applying a
acknowledge that this analysis is limited by the bulk RNA-seq paired t test). We similarly observed significant downregulation
that includes cellular changes in the tumor microenvironment in CDK4 restricted to CRPR cases (after applying an optimal
in responding biopsies, which may make it harder to interpret pooled t test, Figure 6H), which was reported as the driver of
the results. Despite this caveat, we noted that biopsies of pa- the immune exclusion signature (Jerby-Arnon et al., 2018).
tients with PD or SD had a rather random distribution in this Based on our melanoma cell line cohort, interferon-g does not
WNT/T cell space, while the majority of biopsies from patients appear to significantly alter CDK4 expression. On the other
with CRPR had a trend of going from a high WNT/low T cell to hand, MYC genes followed the same pattern (after applying an
a low WNT/high T cell when comparing the baseline and on-ther- optimal pooled t test, Figure 6I), consistent with the WNT gene
apy biopsies (Figure 6B). A key advantage of the data derived score (Figure 6A), while also going down significantly in response
from the cell lines before and after exposure to interferon-g to interferon-g. We previously reported, using a separate set of
in vitro was that the results were not confounded by tumor purity. biopsies, that PAK4 was overexpressed in biopsies of patients
We observed the same significant downward trend in the WNT with low T cell infiltration and lack of response to PD-1 blockade
gene score (p < 7.8 3 10 6 after applying a paired t test) in the therapy, and that in experimental mouse models the T cell exclu-
interferon-g responsive melanoma cell lines (Figure 6C), as six sion and anti-PD-1 resistance could be reversed by PAK4 inhibi-
of the genes comprising the nine gene WNT score were signifi- tion (Abril-Rodrı́guez et al., 2020). In the CheckMate 038 biopsy
cantly decreased in response to interferon-g exposure. MYC, a dataset, PAK4 expression did not differ significantly between
WNT target gene used to monitor changes in WNT signaling, baseline and on-treatment for patients with PD or SD, but
was similarly downregulated in response to interferon-g expo- expression was significantly (p < 2.7 3 10 4) downregulated in
sure across most samples (Figures 6D and S19). Upstream on-treatment biopsies of patients with CRPR (after applying an
WNT signaling genes also changed consistently in response to optimal pooled t test, Figure 6J). This change was not explained
interferon-g exposure, including DKK1 and the frizzled gene fam- by interferon-g signaling, since in our melanoma cell line cohort,
ily (Figure S20A). In fact, upregulation of FZD5 was one of the PAK4 did not change significantly in response to interferon-g.
highest increases across all samples (Figure S20B). Interestingly, all of these biomarkers of immune exclusion were
highly correlated, except for the T cell exclusion-up signature,
Expression of Genes that Drive Immune Exclusion Are which, as expected, was anti-correlated (Pearson correlation,
Selectively Decreased in Biopsies during Response to Figure 6K). Together, these data indicate that ICB therapy results
Therapy in decreased expression of immune exclusion gene programs in
Recently, Jerby-Arnon et al. (2018) developed an immune exclu- biopsies of patients who experience clinical response to therapy.
sion signature by single-cell RNA-seq of melanoma cells from bi- While interferon-g from T cells directly decreases WNT signaling
opsies with low levels of immune infiltration. This work identified and MYC, decreases in the Jerby-Arnon immune exclusion
a set of transcripts originating from tumor cells that were posi- down signature, CDK4, or PAK4 expression are likely down-
tively or negatively associated with immune exclusion by that tu- stream effects in clinically responding tumors.
A B
F I
G J
H K
E
Figure 6. Immune Exclusion Signatures and WNT Signaling in Biopsies of Patients Receiving ICB Therapy
(A) Boxplot of the nine gene WNT score (the geometric average of APC, APC2, CTNNB1, MYC, SOX11, SOX2, TCF12, TCF7, and VEGFA), by RNA-seq according
to response to therapy in patient biopsies using an optimal pooled t test since data are paired and unpaired (*p < 0.05, **p < 0.01, ***p < 0.001).
(B) Plot of WNT score versus T cell score with arrows connecting pre-treatment and on-treatment and one figure for each response PD (red), SD (green), and
CRPR (blue).
(C–E) Pre- and post-treatment gene expression levels of the WNT score (C), MYC (D), and WNT5A (E), the WNT ligand initiating the downstream WNT signaling
programming (red, up; blue, down).
(F–J) Boxplots of immune exclusion genes/signatures by RNA-seq according to response to therapy: Jerby-Arnon et al. associated immune exclusion gene set 1
up in immune excluded tissues (F) and set 2 down in immune excluded samples (G), CDK4 (H), MYC (I), and PAK4 (J) using an optimal pooled t test since data are
paired and unpaired (*p < 0.05, **p < 0.01, ***p < 0.001).
(K) Heatmap showing the level of correlation between the immune exclusion genes and signatures.
See also Figures S19–S21.
therapy in our series, in particular the increase in MHC genes in d KEY RESOURCES TABLE
on-therapy biopsies. However, a significant component of the in- d RESOURCE AVAILABILITY
crease in expression of MHC genes had to be from the infiltration B Lead Contact
with hematopoietic lineage cells into responding tumors as as- B Materials Availability
sessed by LCA correction. Therefore, future analyses would B Data and Code Availability
need to use techniques allowing determination of cancer cell- d EXPERIMENTAL MODEL AND SUBJECT DETAILS
intrinsic antigen processing and presentation pathway. B Clinical Trial and Biopsy Collections
The baseline expression of several T cell immune exclusion sig- B Human Melanoma Cell Lines
natures has been associated with lack of response to ICB in d METHODS DETAILS
different series, most of which also have overlapping gene sets B RNA Sequencing
(Auslander et al., 2018; Jerby-Arnon et al., 2018; Jiang et al., B RNA Immune Deconvolution Using MCP-Counter
2018; Liu et al., 2019). In our series, baseline and on-therapy bi- B Interferon-Gamma In Vitro Exposure and Transcrip-
opsies of patients without a response to ICB therapy had a higher tome Analysis
expression of the Jerby-Arnon et al. (2018) signatures of T cell d QUANTIFICATION AND STATISTICAL ANALYSIS
exclusion, with the relevant feature that responding biopsies dis- B Statistics and Survival Analysis
played decreases in expression of WNT and MYC genes on-ther-
apy, while non-responding biopsies continued to express the im- SUPPLEMENTAL INFORMATION
mune exclusion genes. Overall, our data together with
transcriptome analyses of other biopsy series (Auslander et al., Supplemental Information can be found online at https://doi.org/10.1016/j.
2018; Ayers et al., 2017; Cabrita et al., 2020; Chen et al., 2016; Cris- ccell.2020.08.005.
tescu et al., 2018; Fehrenbacher et al., 2016; Gide et al., 2019; Hel-
mink et al., 2020; Hugo et al., 2016; Jerby-Arnon et al., 2018; Jiang ACKNOWLEDGMENTS
et al., 2018; Liu et al., 2019; Petitprez et al., 2020; Riaz et al., 2017;
C.S.G. was funded by the Parker Institute for Cancer Immunotherapy. J.T. was
Rodig et al., 2018; Roh et al., 2017; Sade-Feldman et al., 2018), im- supported by the NIH Ruth L. Kirschstein Institutional National Research Ser-
plies the presence of feedback loops between the tumor and CD8 vice Award no. T32-CA009120. G.A-R. was supported by the Isabel & Harvey
T cells mediated by interferon-g that modulates immune exclusion Kibel Fellowship award and the Alan Ghitis Fellowship Award for Melanoma
secondary to WNT signaling (Luke et al., 2019). Research. D.Y.T. was supported by a Young Investigator Award from ASCO,
In conclusion, T cell infiltration and expression of interferon-g- a grant from the Spanish Society of Medical Oncology for Translational
regulated genes increase in biopsies of patients receiving ICB Research in Reference Centers and the V Foundation-Gil Nickel Family En-
dowed Fellowship in Melanoma Research. K.C. was supported by the NIH
therapy regardless of clinical response, whereas the degree of
Ruth L. Kirschstein Institutional National Research Service Award no. T32-
HLA upregulation on-therapy and the decrease in expression CA009120 and a Cancer Research Institute Irvington Postdoctoral Fellowship.
of genes associated with immune exclusion are features of clin- J.D.W. was funded by the Parker Institute for Cancer Immunotherapy and NIH
ically responding biopsies. The difference between responsive grant P30 CA008748. V.A. was funded by US NIH grants CA121113, the V
and non-responsive melanomas does not seem to be due to Foundation, Swim Across America, the Allegheny Health Network–Johns Hop-
an inherent ability of some cancer cells to respond differently kins Research Fund, and the LUNGevity Foundation. D.M.P. and S.L.T. were
funded by NCI R01 CA142779, the Bloomberg-Kimmel Institute for Cancer
to interferon-g, as the quality of the interferon-g response pro-
Immunotherapy, and a Cancer Immunology Translational Cancer Research
gram was similar in the great majority of melanoma cell lines Grant (SU2C-AACR-DT1012) from Cancer Research Institute–Stand Up 2
tested. As we did not have single cells for RNA-seq analyses, Cancer (to A.R., D.M.P., and S.L.T.). Stand Up 2 Cancer is a program of the
we were not able to directly determine if the same was true in Entertainment Industry Foundation administered by the American Association
the tumor biopsies. In most cases, ICB was able to change the for Cancer Research. V.E.V. was funded in part by US NIH grants CA121113,
transcriptome of melanoma biopsies, but it did so to different de- CA006973, and CA233259, the Commonwealth Foundation, the Bloomberg-
Kimmel Institute for Cancer Immunotherapy, the Dr. Miriam and Sheldon G.
grees that correlate with favorable changes in the immune cell
Adelson Medical Research Foundation, the V Foundation, the Allegheny
infiltrate. The ability to provide a more powerful anti-melanoma Health Network–Johns Hopkins Research Fund, and the Mark Foundation
immune response with combination ICB explains a higher rate for Cancer Research. A.K. was supported by UCLA CTSI KL2 Award, Sarcoma
of responses. This information also provides hope that additional Alliance for Research through Collaboration Career Enhancement Program,
combinatorial strategies may dial up the antitumor immune Tower Cancer Research Foundation Young Investigator Award, and Radiolog-
response to become clinically meaningful in more patients, in ical Society for North America Research Scholar Grant. A.R. was funded by the
particular when combining ICB therapy with treatments that in- Parker Institute for Cancer Immunotherapy, NIH grants R35 CA197633, P01
CA244118, and P30 CA016042, the Ressler Family Fund, Ken and Donna
crease interferon signaling inside tumors to jump-start an anti-
Schultz Fund, and Cancer Immunology Translational Cancer Research Grant
tumor immune response when it is not already pre-existing, (SU2C-AACR-DT1012) from Cancer Research Institute–Stand Up 2 Cancer
such as the intratumoral administration of oncolytic viruses or with a supplemental grant from the Melanoma Research Alliance.
nucleotide sequences that trigger pattern recognition receptors
to produce interferons (Ribas et al., 2018a, 2018b; Torrejon et al., AUTHOR CONTRIBUTIONS
2020; Vanpouille-Box et al., 2019).
C.S.G. and A.R. designed the experiments and wrote the paper. C.S.G., P.R.-
M., M.W.-R., D.M.P., V.A., S.L.T., V.E.V., D.E.S., A.K., and A.R. provided over-
STAR+METHODS
all scientific oversight on the analyses. C.S.G., J.T., M.Q., G.A.-R., and E.M.
performed bioinformatics analyses. M.O. and P.T. performed experimental
Detailed methods are provided in the online version of this paper studies. A.C., D.Y.T., D.S.S., Y.J.K., C.P.-S., K.C., A.V.-C., C.M., D.E.S.,
and include the following: A.K., and A.R. provided specific expertise for bioinformatics data
Cell Mentor—an online resource from Cell Press and Cell Signaling Technology—empowers
early-career researchers with career insights, publishing advice, and techniques on experimental
processes and procedures. Now it’s even easier to tap into the knowledge and experience of
experts who’ve walked in your shoes.
cellmentor.com
ll
Article
The PD-1/PD-L1-Checkpoint Restrains T cell
Immunity in Tumor-Draining Lymph Nodes
Floris Dammeijer,1,2,11,* Mandy van Gulijk,1,2,11 Evalyn E. Mulder,3 Melanie Lukkes,1 Larissa Klaase,1
Thierry van den Bosch,4 Menno van Nimwegen,1 Sai Ping Lau,1,3 Kitty Latupeirissa,1 Sjoerd Schetters,5
Yvette van Kooyk,5 Louis Boon,6 Antien Moyaart,4 Yvonne M. Mueller,7 Peter D. Katsikis,7 Alexander M. Eggermont,8
Heleen Vroman,1,2 Ralph Stadhouders,1,9 Rudi W. Hendriks,1 Jan von der Thu € sen,4 Dirk J. Gru
€ nhagen,2,3
Cornelis Verhoef,2,3 Thorbald van Hall,10,12,* and Joachim G. Aerts1,2,12,13,*
1Department of Pulmonary Medicine, Erasmus Medical Center, Rotterdam, the Netherlands
2Erasmus MC Cancer Institute, Erasmus Medical Center, Rotterdam, the Netherlands
3Department of Surgical Oncology, Erasmus Medical Center, Rotterdam, the Netherlands
4Department of Pathology, Erasmus Medical Center, Rotterdam, the Netherlands
5Department of Molecular Cell Biology and Immunology, Amsterdam University Medical Center, Amsterdam, the Netherlands
6Polpharma Biologics, Utrecht, the Netherlands
7Department of Immunology, Erasmus Medical Center, Rotterdam, the Netherlands
8Princess Máxima Center, Utrecht, the Netherlands
9Department of Cell Biology, Erasmus Medical Center, Rotterdam, the Netherlands
10Department of Medical Oncology, Oncode Institute, Leiden University Medical Center, Leiden, the Netherlands
11These authors contributed equally
12These authors contributed equally
13Lead Contact
SUMMARY
INTRODUCTION cules following ICB therapy (Koyama et al., 2016; Nishino et al.,
2017). However, the predictive value of these proposed bio-
Drugs targeting the PD-1/PD-L1 pathway have revolutionized markers remains poor in the majority of tumors, while the rele-
the treatment of multiple cancer types including non-small cell vance of PD-L1-expression at other sites remains unknown.
lung cancer, renal cancer, and melanoma with a subset of pa- Furthermore, results from recent trials evaluating combination
tients experiencing durable responses. However, still a majority therapy with anti-PD-(L)1 and other co-inhibitory pathways in
of patients and cancer types do not, or only temporarily respond the TME have been disappointing (Havel et al., 2019; Hellmann
to these immune checkpoint blocking (ICB) drugs (Dammeijer et al., 2019; Long et al., 2019; Ready et al., 2019). Therefore, a
et al., 2017a; Ribas and Wolchok, 2018). PD-1/PD-L1 blocking more comprehensive interrogation of the molecular and spatial
antibodies are believed to act primarily in the tumor microenvi- mechanisms of anti-PD-1/PD-L1 therapy is needed to further
ronment (TME), by re-invigorating exhausted T cells and thereby boost immunotherapy efficacy.
reviving pre-existing anti-tumor immunity (Ribas and Wolchok, Several recent insights exploring the biology of the PD-1 re-
2018). Based on this hypothesis, several theories have been pro- ceptor and its corresponding ligand PD-L1 offer clues into
posed to explain the lack of ICB efficacy in patients, such as a what drives ICB efficacy. Besides tumor cells, myeloid cells ex-
lack of PD-L1 expression or T cell infiltration in the TME and up- pressing PD-L1 have been revealed to be essential for ICB effi-
regulation of other co-inhibitory receptors or suppressive mole- cacy as anti-PD-L1 antibodies remained effective in
Cancer Cell 38, 685–700, November 9, 2020 ª 2020 Elsevier Inc. 685
ll
Article
transplanted tumor cells lacking PD-L1 in a variety of models S1A). Tumor-specific CD8+ T cells in the TDLN were highly pro-
(Kleinovink et al., 2017; Lau et al., 2017; Lin et al., 2018; Tang liferative, expressed PD-1 with higher frequencies and to higher
et al., 2018). Furthermore, these seminal discoveries offer an levels than those in non-TDLNs (Figure 1B). In the TDLN, the pro-
explanation for the rather unexpected finding that PD-1 primarily portions of PD-1-expressing CD8+ T cells strongly correlated
acts by inhibiting signaling downstream of the CD28 costimula- with the frequency of tumor-specific CD8+ T cells and could
tory receptor following B7-ligation, proposedly by myeloid cells therefore serve as a marker for tumor-specificity (Figures 1C
(Hui et al., 2017; Kamphorst et al., 2017). Where this interaction and S1A). Therefore, we enumerated PD-1+ CD4+ and CD8+
and therapeutic disruption in case of anti-PD-1/PD-L1 anti- T cells in several solid tumor models and consistently found
bodies takes place, however, is still unknown, as current genetic higher frequencies of PD-1+ T cells in TDLNs compared with
and pharmacologic interventions (e.g., with the S1PR-blocking non-TDLNs, irrespective of cancer type, mouse genetic back-
agent FTY720) limit proper spatiotemporal analysis of ICB- ground, tumor localization or T cell subset, except for the
induced anti-tumor immune responses (Chamoto et al., 2017; KPC3 pancreatic cancer model (Figure 1D). This difference in
Lau et al., 2017; Lin et al., 2018). Additional insights into these dy- PD-1 expression did not appear to result from tumor metastasis
namics may improve ICB-response prediction and future immu- to the TDLN, as the frequency of CD45 cells was equally low in
notherapy development. both TDLNs and non-TDLNs (Figure S1B). The poorly immuno-
Recent investigations analyzing T cell receptor clonotypes in genic KPC3 pancreatic cancer cell line did not induce PD-1+
mouse and patient tumors before and after anti-PD-1 therapy T cells in the TDLN, suggesting that PD-1 expression is related
suggest the appearance of novel, previously non-existing clono- to immunogenicity of the tumor. In line with this hypothesis, intro-
types in ICB-treated tumors, and a limited expansion capacity of duction of the immunogenic OVA-antigen in KPC3 recapitulated
tumor-resident T cells following treatment (Sade-Feldman et al., the results of the other tumor models (Figure 1D). In addition to
2018; Yost et al., 2019). In contrast to terminally differentiated tu- high PD-1 display, tumor antigen-specific CD8+ T cells in TDLNs
mor-resident T cells, T cell factor 1 (TCF-1)-expressing progen- expressed other co-inhibitory receptors like TIM-3 and TIGIT,
itor-exhausted T cells have recently been described to generate resembling T cells from the tumor site (Figure 1E). In contrast
effector T cell progeny, however their exact origins remain un- to PD-1 expression, other parameters of lymphocyte activation
known (Jansen et al., 2019). These findings, together with an such as proliferation were not consistently increased across
abundance of B7-expressing antigen-presenting cells being different models (Figure S1C), but absolute T cell numbers
exposed to draining tumor-antigens prompted us to investigate were enhanced (Figure S1D).
the role of tumor-draining lymph nodes (TDLNs) in efficacy of To more comprehensively characterize both CD4+ and CD8+
anti-PD-L1-therapy in multiple pre-clinical tumor mouse models. PD-1+ tumor-specific T cells in TDLNs and their dynamics, we
We find that TDLNs harbor significant proportions of tumor-spe- infused naive CD45.1+ OT-I and OT-II cells in AE17-OVA-bearing
cific PD-1+ T cells co-localizing with PD-L1 expressing myeloid CD45.2+ mice and assessed their frequency and phenotype
cells including conventional dendritic cells (cDCs). Selective tar- across tissues over time (Figure 2A). As expected, CD45.1+ cells
geting of PD-L1 only in the TDLN, reveals that TDLN-localized initially accumulated in secondary lymphoid organs including the
T cells are capable of mounting effective anti-tumor immune re- TDLN, followed by egress and migration in blood to finally infil-
sponses thereby demonstrating that TDLN-resident T cells are trate tumor tissue 6 days post-injection (Figure 2B). Upon hom-
able to generate ICB-mediated immunity. Finally, we show the ing to TDLNs, CD45.1+ T cells upregulated PD-1 and prolifer-
role of this PD-1/PD-L1 interaction in the TDLN of stage II mela- ated, suggesting PD-1 to be associated with activation rather
noma patients, independent of known prognostic parameters. than T cell exhaustion (Figure 2C). Globally, PD-1+ T cells further
TDLNs in patients with early disease recurrence are enriched preferentially accumulated in TDLNs over time, where a sub-
for PD-1/PD-L1 interactions which seem to primarily occur be- group of cells expressed effector molecules including interleukin
tween T cells and CD11c+ DCs. In patients without disease (IL)-2, interferon (IFN)-g, and granzyme-B (Figure 2D). In addition
recurrence there are fewer PD-1/PD-L1 interactions in the to increased expression of effector molecules, PD-1+ CD4+ and
TDLN. These results offer unexplored insights in the role of CD8+ T cells expressed more CD28, CD44, CD69, and SLAMF6
TDLNs in generating effective anti-tumor immunity and chal- (Ly108) compared with their PD-1 counterparts (Figure S2)
lenge the current dogma that PD-1/PD-L1-blockade occurs pri- (Kurtulus et al., 2019). These results show initial activation of
marily at the tumor site. early-effector T cells in TDLNs, but not in non-TDLNs, prior to
PD-1 expression and migration to the tumor.
RESULTS
TDLNs Contain Abundant PD-L1high- Expressing Myeloid
+
TDLNs Are Enriched for PD-1 Tumor-Specific T Cells Cells including Migratory cDC2s
To gain insight into the activity of the PD-1/PD-L1-axis in LNs, we Next, we set out to quantify and characterize the ligands of PD-1
analyzed the frequencies and phenotype of tumor antigen-spe- on LN myeloid cells. LNs harbor a complex architecture of
cific CD8+ T cells in LNs of ovalbumin (OVA)-expressing AE17 myeloid cells including CD11b+ dendritic cells (DCs) and macro-
mesothelioma tumors. AE17-OVA tumor cells were injected phages lining the subcapsular (CD169+ subcapsular sinus mac-
intraperitoneally and at late-stage disease mediastinal LNs that rophages [SSMs]) and medullary sinuses (F4/80+ medullary si-
drain tumors in the peritoneal cavity (TDLNs) and inguinal control nus macrophages [MSMs]), type 1 and 2 conventional DCs
LNs (non-TDLNs) were analyzed. We found higher frequencies of (cDC1 and cDC2), and granulocytes interspersed between
OVA257-264-specific CD8+ T cells in the TDLN, compared with a T cells in the paracortex (Eisenbarth, 2019; Glodde et al., 2017;
near absence of these cells in a non-TDLN (Figures 1A and Gray and Cyster, 2012). TDLN-residing myeloid cells including
A 4
*
C
non-TDLN TDLN
% OVA(257-264)-reactive
3 60
0.13% 1.54%
**
% OVA257-264-reactive
2 51.2%
CD3
40
CD3
1
20
0
LN -
OVA257-264 Tetramer+
TD on
LN
N
TD
OVA257-264 Tetramer+ 0
CD3
+
-1
PD-1+ OVA-specific
-1
PD
p<0.0001
PD
B OVA257-264-specific CD8+ T cells CD8+ T cells 8 r2= 0.923
0.39%
* * 6
*** 1400
%PD1+ CD8+
CD3
100 100 PD-1
1200
4
% Proliferation (Ki67)
80 80
1000
PD-1 MFI
60 60 800 2
%PD-1+
0 0 0
LN -
or
T D on
LN
LN -
or
TD on
LN -
LN
m
or
T D on
LN
N
m
TD
Tu
m
TD
N
Tu
TD
Tu
D
15 CD8+ T cells 20 CD4+ Foxp3- T cells
* **
***
*** non-TDLN
TDLN 15 **
10
%PD1+
%PD1+
10
* *
*** *
5
p=0.09 ** *
p=0.07
5
p=0.43
0
AC29 AE17-OVA B16F10 MC38 s.c. MC38 i.p. KPC3 KPC3-OVA AC29 AE17-OVA B16F10 MC38 s.c. MC38 i.p. KPC3 KPC3-OVA
E
OVA257-264-reactive
CD8+ T cells CD4+ Foxp3- cells
CD8+ T cells
50 50 50
0 0 0
LN -
LN
or
TD on
LN -
LN -
LN
or
LN
or
TD on
TD on
m
N
m
TD
m
N
N
TD
TD
Tu
Tu
Tu
TP SP
DP Negative
A B CD8+ CD45.1+
1.25
1.00
3.0x105 Tumor
AE17-OVA i.p.
CD45.1+ OT-I/II-transfer i.v.
0.75 TDLN
% of Alive
CD45.2 +
ns
Blood
0 11 12 14 17 0.50 * * non-TDLN
***
**
Analysis
0.25
0.00
+24h +72hr +144hr
80 80 80 80
% of CD4+ CD45.1
% of CD4+ CD45.1
% of CD8+ CD45.1
% of CD8+ CD45.1
60 60 60 60
40 40 40 40
20 20 20 20
0 0 0 0
h
4h
2h
4h
h
4h
2h
4h
2h
h
4h
2h
44
44
44
+2
+7
4
+2
+7
+2
+7
+2
+7
+1
+1
+1
+1
D PD-1+ IL-2+
IFN-γ+ Granzyme-B+
3 4 10 10 5 25
** *
8 8 *
* *** 6 6 4 20
3 4 4
% of CD4+
2 4 4 *
+
% of CD8+
% of CD8+
% of CD4+
% of CD8
15
% of CD8+
*** 3
2
**
* 3 3
2 10 *
1 2 * * 2
1 5
1 1
1
0 0 0 0 0 0
t=12 t=14 t=17 t=12 t=14 t=17 t=12 t=14 t=17 t=12 t=14 t=17 t=12 t=14 t=17 t=12 t=14 t=17
non-TDLN
TDLN
Figure 2. T Cells Are Activated in TDLNs Prior to Migration and Activation in the Tumor
(A) Experimental design (n = 5–6 mice per group per time point).
(B) Frequencies of CD45.1+ cells were determined in tumor, TDLN, blood, and non-TDLN.
(C) PD-1 positivity as well as the percentage of cells undergoing proliferation of CD8+CD45.1+ cells in the TDLN.
(D) Percentages of PD-1+, IL-2+, IFN-g+, and granzyme-B+ were compared for CD8+ T cells and CD4+ T cells in TDLN (circles) and non-TDLN (squares).
Means and SEMs are shown, paired t tests were used to calculate statistical significance. *p < 0.05, **p < 0.01, ***p < 0.001. TDLN = tumor-draining lymph node,
i.v. = intravenous, IL-2 = interleukin-2, IFN-g = interferon-gamma.
SSMs and cDCs but not granulocytes expressed high levels of types of macrophages (SSM and MSM) in all tested tumor
PD-L1, approaching levels found on their intratumoral counter- models (Figure 3D). These findings were corroborated by multi-
parts (Figure 3A). Interestingly, TDLN myeloid cells lacked or dis- color confocal microscopy of TDLN tissue, where F4/80 + MSMs
played low surface expression of PD-L2, whereas PD-L2 was in the medulla and CD11c + DCs in the LN cortex expressed the
strongly present on all investigated cells in the TME. PD-L1 highest levels of PD-L1, whereas expression levels were negli-
expression and myeloid cell numbers were consistently higher gible in granulocytes (Figure 3E).
in TDLNs compared with non-TDLNs in the investigated solid tu- DCs can be subdivided into migratory and resident subsets
mor models, paralleling PD-1 positivity on T cells (Figures 3B and based on differential expression of CD11c and MHC-II (Fig-
3C). To evaluate which myeloid cells particularly expressed PD- ure S3A) (Ohl et al., 2004). Based on this distinction, we detected
L1 and therefore could be involved in suppressing PD-1-ex- a particularly strong increase in migratory cDC2s, especially in
pressing T cells, we quantified PD-L1 on the aforementioned TDLNs (Figure S3B). These cells expressed high levels of PD-
cell types and found especially high levels on cDC2s and both L1 in addition to CD80 compared with cDC1s (Figures S3C
A B
Ratio PD-L1 expression SSM
SSM MSM cDC2 cDC1 Granulocytes
p=0.06
300 *
Tumor
**
4849 (TAM) 4849 (TAM) 7578 2837 4164
**
PD-L1
TDLN
3689 5267 6361 3044 867 250
%increase (MFI)
non-
TDLN 2646
3 4
4171
3 4
5258
3
2248
3 4
1196
3 4
p=0.07
** **
4
200
Tumor
PD-L2
C non-TDLN
SSM TDLN
cDC2 cDC1 Granulocytes
8000 1500 600 1000
** * ***
* * * 800
6000 p=0.06
1000 400
600
Counts
4000
400
500 200
2000 200
0 0 0 0
B16F10 MC38 s.c. KPC3-OVA B16F10 MC38 s.c. KPC3-OVA B16F10 MC38 s.c. KPC3-OVA B16F10 MC38 s.c. KPC3-OVA
D E
9000 PD-L1+ CD11c+ CD169+
cDC2 SSM
7000
Cortex
6000
PD-L1 MFI
cDC2
5000
SSM
MSM
4000
cDC1 3000
Granu- 2000
locyte PD-L1+ F4/80+
1000
Medullary Sinus
PD-L1+ Ly6G+
Medulla
Figure 3. TDLN Myeloid Cells Express High Levels of PD-L1 Compared With Non-TDLNs
(A) Expression of PD-L1 and PD-L2 (KPC3-OVA) on myeloid cells in the tumor, TDLN and non-TDLN at late-stage disease.
(B) PD-L1 expression on subcapsular sinus macrophages (SSMs) was compared between TDLNs and non-TDLNs, with PD-L1 expression (MFI) on TDLN
macrophages divided over non-TDLN MFI multiplied 3100 to indicate percentage increase of expression.
(legend continued on next page)
Cancer Cell 38, 685–700, November 9, 2020 689
ll
Article
and S3D), which is in line with recent findings identifying migra- body, albeit not completely, whereas the antibody did not reach
tory PD-L1+ DCs to predominantly present tumor-derived anti- non-TDLN nor tumor cells, TAMs or circulating monocytes and
gens in the TDLN (Maier et al., 2020). These data indicate DCs (Figures S4B and S4C).
increased frequencies of PD-L1+ cells in TDLNs compared to
non-TDLNs, especially for macrophages and cDC2s. TDLN-Specific PD-L1 Antibody Elicits Anti-tumor T Cell
Immunity and Tumor Control
Low-Dose Intrapleural PD-L1 Antibody Administration Using these established doses for local (2.5 mg i.pl.) TDLN target-
Selectively Targets TDLNs ing and for systemic (200 mg i.pl.) overall targeting of PD-L1, we
In order to study the effect of selectively targeting the PD-1/PD-L1 examined the therapeutic effect of this ICB in 2 syngeneic tumor
axis in the TDLN on tumor progression we set up a system by models, AC29 mesothelioma and MC38 colon carcinoma (Figures
which we selectively target TDLNs with therapeutic PD-L1 anti- 5A and S5A). Systemic targeting of PD-L1 as well as local TDLN
bodies prohibited antibody availability in the tumor environment it- targeting resulted in decreased tumor burden and increased sur-
self. We examined the option to administer ICB antibody via the vival (Figures 5B and S5B). These therapeutic effects were quite
intrapleural route. This location drains directly to mediastinal strong, considering that not all PD-L1 molecules were blocked
LNs, which are the TDLNs of intraperitoneal tumors from where in TDLN at these low doses (Figure S4B). CD8+ tumor-infiltrating
excess antibody continues to enter the blood via the thoracic T cells (TILs) simultaneously expressing multiple co-inhibitory re-
duct (Kool et al., 2008). Therapeutic anti-PD-L1 blocking anti- ceptors were increased in MC38 tumor tissue, indicating a more
bodies were administered via intravenous (i.v.) or intrapleural exhausted phenotype, which is in line with recent findings in the
(i.pl.) routes and dispersed tissues were counterstained with a flu- same tumor model (Oh et al., 2020). Systemically administered
orescently labeled anti-rat IgG2a antibody, thereby obviating the anti-PD-L1 increased T cell proliferation in blood, whereas both
need to alter the therapeutic antibody itself potentially influencing TDLN-targeted and systemic treatment caused elevated fre-
drug pharmacokinetics (Figure 4A). Irrespective of the route of quencies of CD69+ cells (Figure 5C). Whereas increased tumor
administration, we could readily detect in vivo binding of the ther- infiltration of T cells was markedly induced by systemic anti-PD-
apeutic PD-L1 antibody on tumor cells and TAMs by flow cytom- L1 treatment, TDLN-targeted anti-PD-L1 specifically induced
etry within 24 hr post-injection (Figure 4B). The mediastinal TDLN KLRG-1+ effector T cell infiltration in the AC29 tumor model (Fig-
SSMs, MSMs and cDC2s were similarly reached by i.v. as well as ure 5C). In addition, while nearly all TILs were TOX-positive,
i.pl. administration and staining patterns paralleled those of direct TOX-MFI was significantly decreased in both treatment groups.
ex vivo staining with PD-L1 antibodies (Figure 4C). As expected Recent research has shown that tumors may harbor TCF1+
from the previous findings showing decreased PD-L1 expression CD8+ T cells with a stem-like phenotype, giving rise to distinct
in non-TDLNs, binding of the PD-L1-antibody to myeloid cells was populations, including terminally differentiated exhausted T cells
decreased in the absence of tumor (Figure 4D). Interestingly, i.pl. (Jansen et al., 2019). Similarly, we could identify a subset of TILs
injection appeared to especially reach TDLN subcapsular cells with a stem-like phenotype termed TEXprog, characterized by the
(SSMs & CD11b + cDC2s) more efficiently than the i.v. route in tu- surface marker SLAMF6+ (Ly108), being a surrogate marker for
mor-free animals in contrast to tumor-bearing mice in which anti- TCF-1, displaying low Ki-67, TIM-3, and CD39 (Figures 5D, 5E,
body binding on these cells was largely similar between injection and S5D–S5F) (Beltra et al., 2020). Intratumoral CD8+ and CD4+
routes (Figure 4D). This was not due to differences in PD-L1 TEXprog cells were increased by TDLN-targeted and systemic
expression on cells between treatment arms as PD-L1 expression PD-L1 blockade with decreased levels of proliferation indicating
(assessed in isotype-treated animals, Figure S4A) mirrored PD-L1 preserved stemness following treatment (Figure 5F).
antibody binding in all the leukocyte subsets interrogated irre-
spective of injection route (Figure 4E). These data suggest that FTY720 Abrogates Systemic and TDLN-Specific Anti-
antibody deposition in TDLNs is enhanced in the presence of tu- PD-L1 Immunotherapy Efficacy
mor, possibly via secondary drainage of antibody from permeable In order to confirm the role of the T cells activated in TDLN
tumor vasculature to afferent lymphatics, in addition to direct following PD-L1 treatment, we administered the S1P receptor
TDLN targeting (Figure 4F). As we could now successfully target agonist FTY720, which abrogates T cell egress from lymphoid or-
TDLNs via i.pl. injections we next examined the extent to which gans (Yanagawa et al., 1998) during anti-PD-L1 treatment. Reten-
the efficacy of ICB depends on TDLNs and therefore titrated the tion of T cells in lymphoid organs was confirmed by decreased fre-
i.pl.-administered anti-PD-L1-antibody dose to a level that al- quencies of T cells but not natural killer cells in peripheral blood
lowed selective blockade in the mediastinal TDLN, with no anti- (Figure S6A). FTY720 administration abrogated anti-PD-L1 treat-
body drainage to the intraperitoneal tumor or the circulation (Fig- ment efficacy and prevented influx of CD4+Th- and CD8+ TILs
ures S4B and S4C). At a near 100-fold lower anti-PD-L1-dosage but retained T cells exhibiting increased PD-1 expression (Figures
of 2.5 mg, macrophages and cDCs in the TDLN still bound the anti- 5G and 5H). In addition, FTY720 administration neutralized both
(C) Absolute myeloid cell numbers in subcutaneous tumor models. Means and SEMs are shown and paired t tests were performed.
(D) PD-L1 expression on CD169+ SSMs, F4/80+ medullary sinus macrophages (MSM), type 1 (cDC1) and 2 conventional dendritic cells (cDC2), and Ly6G+
granulocytes from TDLNs.
(E) Multicolor confocal microscopy of a representative TDLN section from an untreated mouse bearing AC29 tumor. Slides were stained for PD-L1, CD11c (DCs),
CD169 (SSM and MSM), Ly6G (granulocytes) and F4/80 (MSM and medullary cord macrophages; MCM). A 3-dimensional composite image was created after
spectral unmixing using single stains (lower left image).
*p < 0.05, **p < 0.01, ***p < 0.001. TDLN = tumor-draining lymph node. Paired t tests were performed to determine statistical significance
A B
αPD-L1-binding αPD-L1-binding αPD-L1-bound
2.5-100 μg αPD-L1 TAM
in 200 μl PBS
Tumor Cells TAM
isotype 100 ns 100 ns
200 μg αPD-L1 i.pl. Isotype
200 μg αPD-L1 i.v. 80 80
%Binding
%Binding
Mediastinal
LN (TDLN) 7
i.p. 10 AC29 60 60
Inguinal LN Tumor cells αPD-L1 i.pl
(non-TDLN)
40 40
αPD-L1 i.v
0 5 1011 20
20
C + Tumor D - Tumor
80 80 80
** 80
ns
%positive
60
%positive
60 60 60
40 40 40 40
20 20 20 20
0 0 0 0
iso i.pl. i.v. iso i.pl. i.v. iso i.pl. i.v. iso i.pl. i.v.
60 60 60 60
%positive
40 40 40 40
20 20 20 20
0 0 0 0
iso i.pl. i.v. iso i.pl. i.v. iso i.pl. i.v. iso i.pl. i.v.
E αPD-L1 i.v.
F
αPD-L1 i.pl.
+ Tumor - Tumor
cDC2 100 p=0.0067 SSM
100 p=0.0067 SSM PD-L1
Rho=0.93 cDC2 αPD-L1
Rho=0.93
afferent
lymphatics
Mean PD-L1
MSM
Expression
Subcapsular
Mean PD-L1
MSM
Expression
Sinus
Granulocyte
50 Granulocyte
50
cDC1
cDC1
B cell B cell
T cell T cell
0 HEV
0
0 20 40 60 80 100
0 20 40 60 80 100
Figure 4. Systemically Administered Anti-PD-L1 Antibodies Bind to PD-L1 Expressing Cells in the TDLN
(A) Experimental setup (n = 6–7 mice/group) and mode of antibody administration. Tumor-bearing mice (filled symbols) were treated with 200 mg isotype (gray)
intrapleurally (i.pl.) or alternatively with 200 mg anti-PD-L1 antibody i.pl (blue) or i.v. (turquoise) to target the TDLN directly or indirectly, respectively.
(B and C) (B) In vivo antibody expression was quantified on intratumoral cell types, and (C) on cells in the TDLN. Histograms showing PD-L1 expression patterns
are displayed.
(D) Quantification of antibody binding on cells in mediastinal LNs in the absence of tumor (open squares and circles).
(E) Mean PD-L1 expression on TDLN cells subsets derived from isotype-treated animals (stained with the BV711-antibody) was correlated to anti-PD-L1 antibody
binding (detected using the secondary anti-rat FITC-labeled antibody) for both i.pl. (left) and i.v. (right) injection routes. A Pearson correlation coefficient (Rho) was
determined.
(F) A graphical depiction of the proposed model showing that anti-PD-L1 antibodies reach TDLNs via intravascular and afferent lymphatics in the presence
of tumor.
Means and SEMs are shown and Mann-Whitney tests were performed indicating statistical significance. ns = not significant (p R 0.05), *p < 0.05, **p < 0.01. i.p. =
intraperitoneal, i.v. = intravenous, TDLN = tumor-draining lymph node, Ab = antibody, SSM = subcapsular sinus macrophages, MSM = medullary sinus mac-
rophages, cDC = conventional dendritic cell.
A B
100
10 ***
Isotype i.pl.
*** ** ***
80 2.5 μg i.pl.
***
200 μg i.pl. *** 9
Centimeters
10⁷ AC29 i.p. 200 μg Isotype
2.5 μg/200 μg αPD-L1 60
% Alive
8
Survival 40
0 10 13 17
7
20
0 6
0 20 40 60 80
l.
l.
-L g
-L g
l.
yp 0μg
i.p
i.p
D 0μ
i.p
D 5μ
Days
1
aP 20
aP 2.
20
e
ot
is
C Blood Isotype i.pl.
2.5 μg aPD-L1 i.pl. Tumor
200 μg aPD-L1 i.pl.
% of CD8+
6000
% of CD8+
% of CD45+
% of CD8+
60 60
% of CD8+
% of CD8+
10 10 5 4000
40 40
10 2000
5 5 20
20 0
0 0 0 0 0 0
% of CD8+ TEXprog
2000
+ Isotype ** 10
300 % of CD8+ 60
** 1500
MFI
MFI
200 1000 40
9.0% 5
+ aPD-L1
2.5 μg 100 500 20
0 0 0
0
TIM-3
G H
CD4+ Th cells CD8+ T cells PD-1+
ns
1500 10 30 100 p=0.05
***
*** *** ** *
8
Tumor weight (mg)
1000 20
% of CD45+
% of CD8+
% of CD45+
6 Isotype i.pl.
50 Isotype i.pl. + FTY720 (>day 9)
200 μg aPD-L1 i.pl.
4 200 μg aPD-L1 i.pl. + FTY720 (>day 9)
500 10
2
0 0 0 0
% of CD8+
TIM-3
20 5
SLAMF6
200 μg aPD-L1
0 0 + FTY720
Figure 5. Specific Targeting of PD-L1 in the TDLN by Low-Dose Intrapleural Injection of Anti-PD-L1 Enhances Clinical Responses in the AC29
Tumor Model and Is Abrogated by FTY720 Treatment
(A) Experimental setup (n = 8–15/group) with mice being treated with either LN-local (2.5 mg) or systemic anti-PD-L1 antibodies (200 mg) i.pl. and survival was
monitored.
(B) Kaplan-Meier curve of the experiment in A showing tumor survival. Log rank tests were used to determine statistical significance. In addition, abdominal
circumferences (being a measure of ascites volume) were measured on t = 21 post tumor-cell injection and displayed as violin-plots.
spontaneous and additional anti-PD-L1-mediated induction of tu- lated liposomes (CELs), which specifically deplete LN macro-
mor CD4+ and CD8+ TEXprog cells following local- and systemically phages upon phagocytosis (van Rooijen and Hendrikx, 2010).
administered anti-PD-L1 treatment (Figures 5I and S6B). These SSMs and MSMs, but not cDC2, were effectively depleted
results indicate that PD-L1 blockade does not solely rely on re- from the TDLN within 48 h at a dose of 5% CEL, whereas mac-
invigoration of TME-localized T cells but in fact amplifies priming rophages in non-TDLNs and in the intraperitoneal tumor re-
and activation of T cells, including TEXprog, from the TDLN. mained essentially unaltered (Figures 7A and 7B). Importantly,
repeated TDLN-localized CEL administration failed to abrogate
PD-L1 Antibody Blockade Amplifies TEXprog Induction anti-PD-L1 efficacy (Figures 7C and 7D), demonstrating a negli-
gible contribution of macrophages to anti-PD-L1 therapy in this
in TDLNs
To investigate the functional consequences of blocking the PD-1/ model, which is in line with recently published data in genetically
PD-L1 axis in TDLNs or TILs, we repeated the aforementioned modified models (Oh et al., 2020). We then examined TDLNs us-
ing multicolor confocal microscopy to visualize co-localization of
experiment in the aggressive AE17 mesothelioma expressing the
OVA model antigen allowing interrogation of tumor-specific anti-PD-L1-expressed cDC2s with CD8+ T cells. CD8+ T cells did
T cells (Figure S7A). CD45.1+ T cell proliferation was enhanced in not co-localize with SSMs and MSMs in TDLNs, but we
frequently found clusters of PD-L1 positive DCs and CD8+
TDLNs compared with non-TDLN following anti-PD-L1 treatment
(Figure S7B). CD45.1+ T cells were proliferating more and were T cells (Figure 7E). Together our data indicate an active role for
more frequently present in TDLN compared with tumor early TDLN in re-invigoration of tumor-specific T cells by ICB, most
likely via PD-L1-expressing cDC(2)s, but not macrophages.
following systemic anti-PD-L1 while no significant differences be-
tween isotype-treated animals were found (Figure S7C). Anti-PD-
L1 treatment markedly increased endogenous CD45.1 TEXprog PD-1/PD-L1 Interactions in TDLN but Not in Tumor
cells, starting in the TDLN after 24 h, followed by blood and tumor
Correlate with Prognosis in Melanoma Patients
on 72 and 144 h, respectively (Figure S7D). In an efficacy experi- As all the aforementioned data were derived from pre-clinical
ment, repeated dosing of anti-PD-L1 locally in the TDLN and sys- solid tumor models, it remained unclear whether PD-1/PD-L1
temically resulted in decreased AE17-OVA-tumor burden (Figures
interaction takes place in patient TDLNs. Melanoma is a disease
6A and 6B), without overt increase in (tumor-specific) TILs (Fig- with a highly variable prognosis depending on the presence of
ure 6C). To assess the replicative potential and phenotype of TILs distant and LN metastasis (stage III-IV), and in case of absence
following anti-PD-L1 therapy, we developed an ex vivo T cell stim-
of distant metastasis (stage I-II); tumor characteristics including
ulation assay in which CD8+ TILs were stimulated with cognate an- thickness (Breslow’s-depth), tumor histology, and presence of
tigen in the presence of their original TME with or without blocking ulceration (Schadendorf et al., 2015; Verver et al., 2018). The
PD-L1 antibodies (Figure 6D). Stimulation with SIINFEKL-peptide
melanoma setting is particularly suited for TDLN characteriza-
led to increased CD8+ T cell activation in this assay, with profound
tion, as these tissues are generally extracted for staging pur-
upregulation of PD-1 following the addition of anti-PD-L1 in vitro poses with the aim of identifying patients who may benefit
(Figure 6D). This induced activation system selectively triggered from adjuvant systemic (immuno-)therapy (Eggermont et al.,
CD8+ TILs and not CD4+ TILs due to the lack of MHC-II binding
2018). To identify whether melanoma TDLNs feature PD-1/PD-
OVA peptide. PD-L1 blockade in tumor cell cultures of mice treated L1-axis activity in the absence of LN metastasis, we stained for
with LN-targeted therapeutic ICB-induced TIL activity to much PD-1/PD-L1-interactions in TDLNs of systemic treatment-naive
higher extent than non-treated mice, as measured by increased
stage II melanoma using a proximity ligation assay (PLA, Fig-
cell frequencies, proliferation, and decrease in PD-1 levels (Fig- ure 8A). We studied PD-1/PD-L1-positivity in TDLNs of stage II
ure 6E). These findings demonstrate that PD-L1 blockade in the patients remaining disease free following surgery (n = 19) and pa-
TDLN alleviates the suppressive impact on intranodal tumor-spe-
tients with early distant disease recurrence (n = 15) (Table S1). As
cific T cells, resulting in trafficking to the tumor site where they expected, known prognostic factors from the primary tumor
display a much improved responsiveness to OVA tumor antigen. such as Breslow’s-depth, presence of ulceration, and nodular
histology were overrepresented in the short-recurrence-free sur-
Response to PD-L1-Blockade Occurs Independent from vival (RFS) cohort compared with patients remaining free of dis-
TDLN Macrophages ease long-term (Schadendorf et al., 2015; Verver et al., 2019). We
In order to investigate which PD-L1-expressing cells in the TDLN detected a significantly higher PD-1/PD-L1-interaction density in
were most likely responsible for inhibiting T cell responses, we patients with a short RFS (<48 months) compared with patients
made use of low-dose i.pl.-administered clodronate-encapsu- remaining disease free for more than 96 months (Figure 8B),
(C–E) (C) Blood was isolated +4 days after first injection with anti-PD-L1 treatment and characterized by multicolor flow cytometry. Tumor-infiltrating lymphocytes
(TILs) were characterized by multicolor flow cytometry for frequencies as well as for positivity for PD-1, KLRG1 and TOX (D) CD8+ TEXprog cells in tumor tissue were
characterized by positivity for SLAMF6 and further identified in (E) for expression of TIM3, Ki67, PD-1, TOX, CD39 and CD69.
(F) Frequencies of tumor-infiltrating CD8+ TEXprog cells and their level of proliferation (Ki67+).
(G) S1P receptor agonist FTY720 administration via drinking water and oral gavage at day 9 to 20 and isotype or anti-PD-L1 treatment (200 mg) i.p. at day 10, 13,
and 17 (9–10 mice per group).
(H) Frequencies of CD4+ Th cells and CD8+ T cells of CD45+ T cells and PD-1 positivity on CD8+ T cells were determined of TILs.
(I) Frequency of CD4+ and CD8+ TEXprog cells was determined for total CD4+ and CD8+ TILs, respectively.
Means and SEMs are shown and Mann-Whitney tests were performed indicating statistical significance. ns = not significant (p R 0.05), *p < 0.05, **p < 0.01. i.p. =
intraperitoneal, i.pl. = intrapleural, TEXprog = progenitor-exhausted T cells.
% of CD45+
200μg Isotype 4
% of CD8+
Grams
2.5μg/200μg αPD-L1
10
0 10 13 17 20 2 20
5
0 0 0
l.
l.
-L g
-L g
l.
l.
yp μg
-L g
l.
l.
l.
-L g
-L g
yp μg
i.p
i.p
-L g
D 0μ
l.
yp μ g
l.
D 5μ
i.p
i.p
D 0μ
i.p
i.p
i.p
D 0μ
D 5μ
D 5μ
i.p
i.p
ot 0
ot 0
aP 20
ot 0
1
1
aP 2.
e
1
is 20
aP 20
aP 20
aP 2.
e
is 20
aP 2.
e
is 20
OVA(257-264)-specific
CD8+ T cells
D E
+ Peptide & αPD-L1
Enzymatic
disgestion 7000
+Celltracker dye
+3d
6000
Analyze + Peptide
Untreated 5000
PD-1 MFI
2.5u g aPDL1 4000 No adds
200 ug aPDL1
3000
Tumor +/- OVA(257-264)-peptide
+/- 10μg/ml αPD-L1 antibody 2000
1000
0
80 ns
p=0.05
100 **
80 140
* ns
CD8+ T cells
%CD25+
%CD25+
ns
ns
60 200 130
60
*
ns
* * 120
40 40 50 ** *
150
110
20 20
100
100
0 0 0
CD4 T cells
+
150
T cells
*
6000 6000 * * 100
100 * **
5000 5000 0
PD-1 MFI
PD-1 MFI
4000 4000 *
* * 150
*
3000 3000 200 * 100 *
* 140 * ns
ns *
2000 2000
Total CD4+
ns
130
T cells
1000 1000
150 120
50
0 0
110
Peptide - + + Peptide - + +
αPD-L1 - - + αPD-L1 - - + 100
100
Pre-culture + - - - + - - - Peptide - + +
Post-culture - + + + - + + + αPD-L1 - - +
Peptide - - + + - - + +
αPD-L1 - - - + - - - +
Figure 6. Specific Targeting of PD-L1 in the TDLN Elicits Durable Anti-tumor Immune Responses Capable of Re-invigoration In Vitro
(A) Experimental design (n = 7–9 mice per group).
(B) Tumor weights of the different treatment groups.
(C) Frequencies of total CD8+ T cells and OVA(257–264)-specific CD8+ T cells in tumors as percentages of total alive CD45+ leukocytes or CD8+ T cells,
respectively.
(D) Design and validation of the in vitro culture system mimicking the tumor microenvironment. Means and SEMs are depicted with paired t tests used for
statistical analysis.
(E) Tumor single cell suspensions from isotype (gray), anti-PD-L1 LN-specifically (pink), and anti-PD-L1 systemically (purple) treated mice were cultured as
described in D.
Means and SEMs are shown, Mann-Whitney tests were used to evaluate statistical significance For E, treatment arms were normalized to isotype-treated animals
and baseline and post-3 day culture values for the several conditions are shown.
4000 ns
15
%alive
Count
500 ns
1000 200
2500
2000 ns
10
ns ns
0 0 1500
0
1000 5
500
Red Pulp Macrophage Marginal Metalophillic Ly6C-low Non-classical 0 0
1.0 Macrophage 6 Monocyte (blood) SSM MSM cDC1 cDC2 SSM Peritoneal RP TAM
0.4
(iLN) Mac Mac
0.8
0.3
4
0.6
%CD45+
0.2
0.4
2
0.1 0.2
0
0.0 0.0
l l l l l .
D
l l l l
BS .p .p .p .p .p p
l . l l l l
BS .p .p .p .p .p p
l .
P % i % i % i % i % i i.
BS .p .p .p .p .p p
P % i % i % i % i % i i.
100 Isotype *
P % i % i % i % i % i i. 5 10 2 0 5 0 00 0 % 5 1 0 2 0 5 0 00 0 %
5 10 20 50 00 0%
1 10 1 10 1 10 5% CEL
80 local aPDL1
5% CEL + local aPDL1
Percent survival
60
C
10⁷ AC29 i.p. 2.5 μg αPD-L1 40
5% CEL + 5% CEL
Survival
20
0 8 10 13 17
0
0 20 25 30 35
Days
E
αPD-L1
200 μg αPD-L1
αPD-L1+
reflecting possible ineffective anti-tumor immune surveillance. and CD68. Using this method, we observed high PD-1/PD-L1-
Similar results were obtained when contacts were numerically signaling in germinal centers largely devoid of CD8+ T cells,
quantified or alternatively calculated as percentage of total which is in line with previous data (containing primarily B cells
cell-surface area using automated software analysis (Methods and follicular T helper cells, excluded from analysis in 8A-C (Shi
S1). Interestingly, short RFS was associated with increased et al., 2018)). Also numerous interactions elsewhere in the LN
PD-1/PD-L1-contact density irrespective of the aforementioned cortex were present (Figure 8D). Similar to the murine data, these
prognostic factors, indicating that this process of PD-L1-medi- contacts were particularly established by CD8+ (and to a lesser
ated immune suppression arises independently of these primary degree CD8 ) T cells and CD11c+ DCs whereas CD68+ macro-
tumor characteristics (Figure 8C). To gain insight into which cells phages barely associated with T cells (Figure 8E). To investigate
were involved in PD-1/PD-L1-interaction, we combined the PLA whether PD-1/PD-L1-interactions occur in melanoma tumors to
with multicolor immunofluorescence staining for CD8, CD11c a similar extend as in TDLNs, we performed PLA on matched
A B ** ***
40X p=0.0064 p=0.0007
200 5
4
150
%cell-surface+
#contacts/hpf
PD-1 3
100
PD-L1 2
50 1
0 0
RFS RFS RFS RFS
<48m >96m <48m >96m
C ns ns ns
5 5 5
%cell-surface+
4 4 4
%cell-surface+
%cell-surface+
3 3 3
2 2 2
1 1 1
0 0 0
Ulcer- Ulcer- unknown SupSM NM other/
Brewlow Breslow unknown
ation+ ation- unknown
<4mm ≥4mm
D E
50
1. *
* *
%of cell-cell contacts
1.25X 40
30
*
1
20
10X
2 40X 10
2.
0
PD-1 CD8+ + + + + - - - -
CD11c+ CD8+
CD68+
CD11c+ - + - + - + - +
PD-1+ PD-L1+
PD-L1
+ - + + + +
CD68 - - -
2
1
0.50
ns
0.25
0.00
RFS RFS
<48m >96m
Figure 8. Stage II Melanoma TDLNs Harbor Frequent PD-1/PD-L1 Interactions, Which Associates With Early Distant Recurrence Following
Surgery, and Not in Primary Tumor Tissue
(A) Exemplary image magnified 340 of a stage II melanoma TDLN (sentinel node) displaying several PD-1/PD-L1-contacts stained using PLA.
primary tumor tissue of a subset of patients from whom material anti-PD-1 therapy, in which an extratumoral source of immuno-
was available. Intriguingly, PD-1/PD-L1 interactions in tumor tis- therapy-elicited T cells was suggested (Yost et al., 2019).
sue of melanoma patients (n = 9) were scarce compared with the TDLNs could therefore be pivotal in generating effective anti-
TDLN with the extent of the remainder of interactions not corre- tumor T cell responses following liberation by anti-PD-1/PD-L1
lating with RFS (Figure 8F). PD-L1 alone, however, was ex- antibodies, as has been previously shown to be the case for
pressed in tumor tissue showing that the lack of PLA-positivity other cancer immunotherapies (Spitzer et al., 2017). Surgical
did not result from absence of PD-L1-expression (Figure 8G). removal of the TDLN in mouse tumor models revealed a contri-
These findings support our initial findings in mice and highlight bution of TDLNs to ICB efficacy, but as LNs are known to amplify
a previously unidentified role for PD-1/PD-L1 interactions in hu- direct anti-tumor effects of immunotherapy, formal assessment
man TDLNs, possibly identifying high-risk patient groups for of their role in ICB therapy is lacking (Chamoto et al., 2017; Fran-
adjuvant immunotherapy. sen et al., 2018; Spitzer et al., 2017). The role of TDLNs as main
hubs in providing anti-tumor T cell immunity following ICB
DISCUSSION furthermore fits with recent insights into PD-1/PD-L1 biology
showing that PD-1 inhibits CD28-B7-mediated T cell co-stimula-
Our data formally establish a role for TDLNs in generating pri- tion, a process that takes place in a B7-rich environment such as
mary anti-tumor immune responses following anti-PD-L1 ICBs, LNs, and is less likely in the immune-suppressive tumor microen-
a checkpoint that is generally regarded to act primarily at the vironment (Hui et al., 2017; Kamphorst et al., 2017). Furthermore,
TME (Ribas and Wolchok, 2018). These data could offer an anti-tumor T cells in TDLNs are less exhausted compared with
explanation for the apparent clinical incongruences in which tu- TIL and thus may have a proliferative advantage with ICB
mor PD-L1-negative tumors may still respond to anti-PD-1 (Hope et al., 2019).
blockade, which has resulted in (chemo-)immunotherapy being Our tumor models evaluate the role of TDLN in a relatively early
first-line treatment in several metastatic cancers irrespective of stage, at a time that T cell priming following tumor cell inoculation
PD-L1 positivity (Eggermont et al., 2018; Gandhi et al., 2018; might still be occurring. The aggressiveness of the model pro-
Paz-Ares et al., 2018). Although multiple mechanisms will likely hibits treatment in later phases. TILs were not significantly
define response to ICB therapy, alleviating immune suppression decreased following FTY720 treatment on day 9, which suggests
in the TDLN could propel systemic anti-tumor T cell immunity to that initial priming has largely occurred in the preceding time win-
effectively control distant tumor sites. Our data reveal a critical dow (Figure 5H). Importantly, we do not exclude a role for PD-L1
role of PD-L1 on cDCs in the TDLN, without negating the involve- blockade in the TME, but have technically not been able to selec-
ment of this inhibitory ligand in the TME. tively administer antibody to the i.pl. or subcutaneously (s.c.)-
Our data show that PD-1 expression on T cells in the TDLN located tumors in order to directly compare the importance of
seems to correlate with antigenicity of the tumor, which might TDLN and TME to PD-L1 treatment efficacy. The exact contribu-
explain the conflicting results from others showing a minor de- tion of TDLN versus TME during PD-1/PD-L1 checkpoint
pendency on T cells from the TDLN (Siddiqui et al., 2019). Tu- blockade therapy remains to be elucidated; however, our data
mor-mutational burden has been identified as a predictive clearly unraveled the involvement of lymph nodes.
marker for ICB efficacy in patients, and our data strengthen the We identify PD-L1+ cDCs, most likely cDC2s, as main targets
role of TDLNs as possible mediators of ICB-responses as these of anti-PD-L1 antibodies in the TDLN, and not macrophages.
potentially immunogenic tumor-antigens are likely to drain or be Recently, Oh et al. showed that DC-specific genetic PD-L1-abla-
transported to TDLNs for further T cell induction (Samstein et al., tion phenocopied the effects of complete PD-L1-knockout mice
2019). Others have previously hinted toward a role for LNs in in bearing MC38 tumors, whereas macrophage-direct PD-L1-
generating anti-tumor immunity following PD-1/PD-L1 ICB in pa- elimination did not (Oh et al., 2020). However, as PD-L1 was ab-
tients by describing several dysfunctional T cell subsets arising lated systemically via the CD11c promotor, analyses into the site
following treatment (Li et al., 2019; Sade-Feldman et al., 2018). and preferred mechanism of PD-L1-PD-1-blockade mediated
Our data formally establish these notions and directly comple- anti-tumor rejection remained elusive. Classically, a division of
ment recent findings on TIL-clonality in patients treated with labor has been proposed with cDC2s predominantly inducing
(B) Quantification of the average number of contacts, in patients with an early (<48 months) recurrence of disease following surgery (n = 15) and no recurrence
after 96 months (n = 19) either via manual (left) and automated (right) quantification analysis. Statistical significance was determined using Mann-Whitney tests
and means plus SEMs are depicted.
(C) Using the outcome measure project in right panel B, patient tumors were divided according to Breslow’s-depth (tumor-thickness), presence of tumor ul-
ceration and histological subtype.
(D) TDLN-overview and zoom-in of germinal center naturally rich in PD-1/PD-L1-interactions (1) and paracortical area containing PD-1+ CD8+ T cells and PD-L1+
myeloid cells (2).
(E) Multiplexed images from 6 patient TDLNs high in PD-1/PD-L1 contacts (5 short, 1 long RFS) were constructed combining the PLA with CD8 (green), CD11c (yellow),
and CD68 (red), followed by counterstaining with hematoxylin (blue). Images magnified 340 were acquired from cortical LN regions, excluding germinal centers
(F) PD-1/PD-L1 interactions in primary tissue of stage II melanoma patients were stained using PLA (n = 9). The average number of contacts was enumerated from
patients with an early recurrence (n = 5) and no recurrence after 96 months (n = 4).
(G) A 320 magnified exemplary image of primary tumor tissue of stage II melanoma patients displaying PD-L1 expression (clone SP263; left) and PD-1/PD-L1
interactions using PLA (right).
Means and SEMs are shown, Mann-Whitney tests were used to calculate statistical significance. TDLN = tumor-draining lymph node, RFS = recurrence-free
survival, PLA = Proximity Ligation Assay, SupSM = Superficial spreading melanoma, NM = nodular melanoma.
CD4+ T cell activation and cDC1s priming CD8+ T cells (Wculek following blockade of PD-L1 in the TDLN. These data challenge
et al., 2020). We show, however, that cDC2s are significantly the current dogma that PD-1/PD-L1-checkpoint act primarily at
more PD-L1-positive and frequent in TDLNs compared with the effector (tumor) site and offers additional avenues for
cDC1s and co-localize with CD8+ T cells in the TDLN cortex, biomarker and combination-immunotherapy discovery.
challenging the current dogma. In agreement with our findings
are recent publications by different groups showing cDC2s to STAR+METHODS
be efficient CD8+ T cell stimulators in the context of solid cancer
and viral infection, in addition to their well-described role in acti- Detailed methods are provided in the online version of this paper
vating CD4+ T cells (Bosteels et al., 2020; Ruhland et al., 2020). and include the following:
Moreover, comprehensive analysis of tumor- and TDLN-DCs
by Maier et al. shows that both cDC1s and cDC2s adopt a reg- d KEY RESOURCES TABLE
ulatory phenotype upon apoptotic tumor-cell ingestion and up- d RESOURCE AVAILABILITY
regulate PD-L1 thereby preventing proper T cell induction (Maier B Lead Contact
et al., 2020). These combined findings provide a rationale for B Material Availability
manipulating cDC2s for the benefit of cancer immunotherapy. B Data and Code Availability
Our analysis of PD-1/PD-L1-axis activity in nonmetastatic TDLNs d EXPERIMENTAL MODEL AND SUBJECT DETAILS
of melanoma patients complements our findings in murine solid tu- B Mouse Models
mor models by identifying a subset of patients with high PD-1/PD- B Mouse Tumor Cell Lines
L1 interaction density that was associated with early disease B Melanoma Patient Cohort
relapse following surgery. Although these early stage (II) patients d METHOD DETAILS
generally have a favorable prognosis following primary resection, B In Vivo Experiments in Mouse Tumor Models
a minor subset eventually presents with distant recurrences bearing B Preparation of Single Cell Suspensions
a poor prognosis (Verver et al., 2019; von Schuckmann et al., 2019). B Flow Cytometry
It is tentative to speculate that PD-L1-mediated suppression of anti- B Multicolor Confocal Microscopy
tumor T cell responses in the TDLNs of these patients is involved in B TIL Re-stimulation Culture
the development or progression of distant metastasis, and future B PLA and Multiplex Staining (cmIHC)
research will likely shed more light on the validity of this hypothesis. B Quantification of PLA and Multiplex Stainings
An ancillary observation supporting this hypothesis is the fact that d QUANTIFICATION AND STATISTICAL ANALYSIS
the 2 patients who received primary anti-PD-1 antibodies at disease B Statistical Analysis
recurrence resulting in durable complete responses were the sec-
ond- and third-highest expressers of PD-1/PD-L1 in the TDLN (Fig- SUPPLEMENTAL INFORMATION
ure S8). Although the TDLNs bearing these high-density contacts
were excised at primary disease presentation, it is likely that distant Supplemental Information can be found online at https://doi.org/10.1016/j.
micrometastases bearing a similar genetic and TME-makeup ccell.2020.09.001.
Ruhland, M.K., Roberts, E.W., Cai, E., Mujal, A.M., Marchuk, K., Beppler, C., Verver, D., van Klaveren, D., van Akkooi, A.C.J., Rutkowski, P., Powell, B.,
Nam, D., Serwas, N.K., Binnewies, M., and Krummel, M.F. (2020). Robert, C., Testori, A., van Leeuwen, B.L., van der Veldt, A.A.M., Keilholz,
Visualizing synaptic transfer of tumor antigens among dendritic cells. Cancer U., et al. (2018). Risk stratification of sentinel node-positive melanoma patients
Cell 37, 786–799.e5. defines surgical management and adjuvant therapy treatment considerations.
Sade-Feldman, M., Yizhak, K., Bjorgaard, S.L., Ray, J.P., de Boer, C.G., Eur. J. Cancer 96, 25–33.
Jenkins, R.W., Lieb, D.J., Chen, J.H., Frederick, D.T., Barzily-Rokni, M., von Schuckmann, L.A., Hughes, M.C.B., Ghiasvand, R., Malt, M., van der Pols,
et al. (2018). Defining T cell states associated with response to checkpoint J.C., Beesley, V.L., Khosrotehrani, K., Smithers, B.M., and Green, A.C. (2019).
immunotherapy in melanoma. Cell 175, 998–1013.e20. Risk of melanoma recurrence after diagnosis of a high-risk primary tumor.
Samstein, R.M., Lee, C.-H., Shoushtari, A.N., Hellmann, M.D., Shen, R., JAMA Dermatol. 155, 688–693.
Janjigian, Y.Y., Barron, D.A., Zehir, A., Jordan, E.J., Omuro, A., et al. (2019).
Wculek, S.K., Cueto, F.J., Mujal, A.M., Melero, I., Krummel, M.F., and Sancho,
Tumor mutational load predicts survival after immunotherapy across multiple
D. (2020). Dendritic cells in cancer immunology and immunotherapy. Nat. Rev.
cancer types. Nat. Genet. 51, 202–206.
Immunol. 20, 7–24.
Schadendorf, D., Fisher, D.E., Garbe, C., Gershenwald, J.E., Grob, J.J.,
Halpern, A., Herlyn, M., Marchetti, M.A., McArthur, G., Ribas, A., et al. Yanagawa, Y., Sugahara, K., Kataoka, H., Kawaguchi, T., Masubuchi, Y., and
(2015). Melanoma. Nat. Rev. Dis. Primers 1, 15003. Chiba, K. (1998). FTY720, a novel immunosuppressant, induces sequestration
of circulating mature lymphocytes by acceleration of lymphocyte homing in
Shi, J., Hou, S., Fang, Q., Liu, X., Liu, X., and Qi, H. (2018). PD-1 controls follic-
rats. II. FTY720 prolongs skin allograft survival by decreasing T cell infiltration
ular T helper cell positioning and function. Immunity 49, 264–274.e4.
into grafts but not cytokine production in vivo. J. Immunol. 160, 5493–5499.
Siddiqui, I., Schaeuble, K., Chennupati, V., Fuertes Marraco, S.A., Calderon-
Copete, S., Pais Ferreira, D., Carmona, S.J., Scarpellino, L., Gfeller, D., Yost, K.E., Satpathy, A.T., Wells, D.K., Qi, Y., Wang, C., Kageyama, R.,
Pradervand, S., et al. (2019). Intratumoral Tcf1(+)PD-1(+)CD8(+) T cells with McNamara, K.L., Granja, J.M., Sarin, K.Y., Brown, R.A., et al. (2019). Clonal
stem-like properties promote tumor control in response to vaccination and replacement of tumor-specific T cells following PD-1 blockade. Nat. Med.
checkpoint blockade immunotherapy. Immunity 50, 195–211.e10. 25, 1251–1259.
Spitzer, M.H., Carmi, Y., Reticker-Flynn, N.E., Kwek, S.S., Madhireddy, D., Zhao, M., Kiernan, C.H., Stairiker, C.J., Hope, J.L., Leon, L.G., van Meurs, M.,
Martins, M.M., Gherardini, P.F., Prestwood, T.R., Chabon, J., Bendall, S.C., Brouwers-Haspels, I., Boers, R., Boers, J., Gribnau, J., et al. (2020). Rapid
et al. (2017). Systemic immunity is required for effective cancer immuno- in vitro generation of bona fide exhausted CD8+ T cells is accompanied by
therapy. Cell 168, 487–502.e15. Tcf7 promotor methylation. PLoS Pathog. 16, e1008555.
iScience is an interdisciplinary
open access journal from Cell Press
iScienceW\ISPZOLZIHZPJHUKHWWSPLKYLZLHYJO[OH[HK]HUJLZHZWLJPÄJÄLSKHJYVZZSPMLWO`ZPJHS
HUKLHY[OZJPLUJLZ0[»ZHUVWLUHJJLZZQV\YUHS^P[OJVU[PU\V\ZW\ISPJH[PVUZVYLZLHYJOPZ
PTTLKPH[LS`HJJLZZPISL6\YUVUVUZLUZLHWWYVHJO[VZ\ITPZZPVUZPZZPTWSLMHZ[HUKMHPYHUK
V\YJVTTP[TLU[[VPU[LNYP[`TLHUZ^LW\ISPZO[YHUZWHYLU[TL[OVKZ^P[OOPNOLKP[VYPHSZ[HUKHYKZ
cell.com/iscience
What is your superpower?
Gene functional analysis is ours.
Who we are
Cellecta is a leading provider of genomic products and services. Our
functional genomics portfolio includes gene knockout and knockdown
screens, custom and genome-wide CRISPR and RNAi libraries, construct
services, cell engineering, NGS kits and targeted expression profiling
products and services.
Switch to BD Horizon™ Red 718 Reagents to help achieve your resolution goals.
Class I Laser Products. For Research Use Only. Not for use in diagnostic or therapeutic procedures.
23-22635-00
BD, the BD Logo, Horizon, OptiBuild and Pharmingen are trademarks of Becton, Dickinson and Company or its affiliates.
All other trademarks are the property of their respective owners. © 2020 BD. All rights reserved.
I S R U P T IN
DE THERAP RUG DIGSC
EUTIC D
N
OV
U
IMM
ERY
OX I I E S
J A X NGL E
OT T U D
CI T Y
A SI
CY T
KIN
ON N
S
O
ST E R
UD AT I
Y TO ELEASE EVALU IMMU
TEST EFFICACY AND
USE A SINGLE HUMAN PBMC DONOR USE MULTIPLE HUMAN PBMC DONORS
to perform dose ranging studies and to recapitulate donor-to-donor variability
to allow lead selection. you’d expect to find in the clinic.