Article
Cell-of-Origin Patterns Dominate the Molecular
Classification of 10,000 Tumors from 33 Types of
Cancer
Graphical Abstract
Highlights
d An integrative data clustering method is applied to reclassify
human tumors
d Cell-of-origin influences, but does not fully determine, tumor
classification
d Immune features and copy-number aberrations define the
most mixed tumor groups
d Multi-cancer groups reveal new features with potential
clinical utility
Authors
Katherine A. Hoadley, Christina Yau,
Toshinori Hinoue, ..., Joshua M. Stuart,
Christopher C. Benz, Peter W. Laird
Correspondence
hoadley@med.unc.edu (K.A.H.),
peter.laird@vai.org (P.W.L.)
In Brief
Comprehensive, integrated molecular
analysis identifies molecular
relationships across a large diverse set of
human cancers, suggesting future
directions for exploring clinical
actionability in cancer treatment.
Hoadley et al., 2018, Cell 173, 291–304
April 5, 2018 ª 2018 Elsevier Inc.
https://doi.org/10.1016/j.cell.2018.03.022
Article
Cell-of-Origin Patterns Dominate
the Molecular Classification of 10,000 Tumors
from 33 Types of Cancer
Katherine A. Hoadley,1,21,* Christina Yau,2,3,21 Toshinori Hinoue,4,21 Denise M. Wolf,5,21 Alexander J. Lazar,6,21
Esther Drill,7,21 Ronglai Shen,7,21 Alison M. Taylor,8,9,21 Andrew D. Cherniack,8,9,21 Ve´ steinn Thorsson,10,21
Rehan Akbani,6,21 Reanne Bowlby,11,21 Christopher K. Wong,12,21 Maciej Wiznerowicz,13,14,15 Francisco Sanchez-Vega,16
A. Gordon Robertson,11 Barbara G. Schneider,17 Michael S. Lawrence,8,18 Houtan Noushmehr,19,20 Tathiane M. Malta,19,20
The Cancer Genome Atlas Network, Joshua M. Stuart,12 Christopher C. Benz,2 and Peter W. Laird4,22,*
1Department of Genetics, Lineberger Comprehensive Cancer Center, the University of North Carolina at Chapel Hill, Chapel Hill, NC 27599, USA
2Buck Institute for Research on Aging, Novato, CA 94945, USA
3Department of Surgery, University of California, San Francisco, San Francisco, CA 94115, USA
4Van Andel Research Institute, Grand Rapids, MI 49503, USA
5Department of Laboratory Medicine, University of California, San Francisco, San Francisco, CA 94115, USA
6Department of Bioinformatics and Computational Biology, The University of Texas MD Anderson Cancer Center, Houston, TX 77030, USA
7Department of Epidemiology and Biostatistics, Memorial Sloan Kettering Cancer Center, New York, NY 10065, USA
8Broad Institute of Harvard and MIT, Cambridge, MA 02142, USA
9Department of Medical Oncology, Dana-Farber Cancer Institute, Boston, MA 02215, USA
10Institute for Systems Biology, Seattle, WA 98109, USA
11Canada’s Michael Smith Genome Sciences Centre, BC Cancer Agency, Vancouver, BC V5Z 1L3, Canada
12Department of Biomolecular Engineering, Center for Biomolecular Sciences and Engineering, University of California, Santa Cruz,
Santa Cruz, CA 95064, USA
13Poznan University of Medical Sciences, 61-701 Poznan, Poland
14Greater Poland Cancer Centre, 61-866 Poznan, Poland
15International Institute for Molecular Oncology, 60-203 Poznan, Poland
16Marie-Jose´ e and Henry R. Kravis Center for Molecular Oncology, Memorial Sloan Kettering Cancer Center, New York, NY 10065, USA
17Department of Medicine, Division of Gastroenterology, Vanderbilt University Medical Center, Nashville, TN 37232, USA
18Massachusetts General Hospital Cancer Center and Department of Pathology, Harvard Medical School, Charlestown, MA 02129, USA
19Department of Neurosurgery, Henry Ford Health System, Detroit, MI 48202, USA
20Department of Genetics, University of Sao Paulo, Ribeirao Preto, SP, 14049-900, Brazil
21These authors contributed equally
22Lead Contact
*Correspondence: hoadley@med.unc.edu (K.A.H.), peter.laird@vai.org (P.W.L.)
https://doi.org/10.1016/j.cell.2018.03.022
SUMMARY
We conducted comprehensive integrative molecular
analyses of the complete set of tumors in The Cancer
Genome Atlas (TCGA), consisting of approximately
10,000 specimens and representing 33 types of
cancer. We performed molecular clustering using
data on chromosome-arm-level aneuploidy, DNA hypermethylation,
mRNA, and miRNA expression levels
and reverse-phase protein arrays, of which all, except
for aneuploidy, revealed clustering primarily organized
by histology, tissue type, or anatomic origin.
The influence of cell type was evident in DNA-methylation-based
clustering, even after excluding sites
with known preexisting tissue-type-specific methylation.
Integrative clustering further emphasized the
dominant role of cell-of-origin patterns. Molecular
similarities among histologically or anatomically
related cancer types provide a basis for focused
pan-cancer analyses, such as pan-gastrointestinal,
pan-gynecological, pan-kidney, and pan-squamous
cancers, and those related by stemness features,
which in turn may inform strategies for future therapeutic
development.
INTRODUCTION
Genomic and other molecular analyses across many types of
cancer have revealed a striking diversity of genomic aberrations,
altered signaling pathways, and oncogenic processes. We
hypothesized that this diversity arises from endogenous
factors, such as developmental and differentiation programs
and epigenetic states of the originating cells, in conjunction
with exogenous factors, such as mutagenic exposures,
pathogens, and inflammation. Here, we performed an integrative
analysis of approximately 10,000 human samples representing
33 different cancers, to provide the first comprehensive view of
the molecular factors that distinguish different neoplasms in
The Cancer Genome Atlas (TCGA).
Cell 173, 291–304, April 5, 2018 ª 2018 Elsevier Inc. 291
This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/).
In 2014, TCGA Research Network reported an interim analysis
of 3,527 tumors from 12 different cancer types (Pan-Cancer-12),
integrating six genome-wide platforms that assayed tumor
DNA (exome sequencing, DNA methylation, and copy number),
RNA (mRNA and microRNA sequencing), and a cancer-relevant
set of proteins and phosphoproteins (Hoadley et al., 2014). The
analysis tested the hypothesis that molecular signatures might
provide a taxonomy that differed from the current organ- and
tissue-histology-based pathology classification (Hoadley et al.,
2014). This effort extended beyond cancer subtype classification
by individual molecular platforms by employing an integrated
clustering algorithm to identify higher-level structures and
relationships. These integrated subtypes shared mutations,
copy-number alterations, pathway commonalities, and microenvironment
characteristics that appeared influential in the
new molecular taxonomy, beyond any phenotypic contributions
from tumor stage or tissue of origin. We estimated that at least
one in ten cancer patients might be classified (and perhaps
treated) differently using such a molecular taxonomy, rather
than the current histopathology-based classification.
Given that the earlier analysis included only a third of the final
set of TCGA tumors, it seemed appropriate to analyze all 33 tumor
types (called the PanCancer Atlas) to address the intriguing
questions left unanswered: whether the inclusion of many more
tumors and tumor types enhances the number of cross-tissue
associations, produces additional convergent and/or divergent
integrated molecular subtypes, and significantly increases the
fraction of cancer patients whose classification or treatment
might be affected by this new taxonomic approach.
We present a new PanCancer Atlas integrative analysis using
iCluster (Shen et al., 2009, 2012) identifying 28 distinct molecular
subtypes arising from the 33 different tumor types analyzed
across at least four different TCGA platforms. We confirmed
significant taxonomic divergences from and convergences with
the routinely used clinical tumor classification system. We employed
a new 2D visualization approach, TumorMap (Newton
et al., 2017), to intepret the relationships between the samples
and iClusters. The PanCancer Atlas molecular classification also
provides a rationale for several TCGA analyses based on organ
systems or differentiation states, including pan-gastrointestinal
(GI) (Liu et al., 2018), pan-gynecological (gyn) (Berger et al.,
2018), pan-kidney (Ricketts et al., 2018), pan-squamous (Camp-
belletal.,2018),andcancerstemnessfeatures(Maltaetal.,2018).
RESULTS
Specimens and Tumor Types
This PanCancer study encompassed 11,286 tumor samples from
33 cancer types, for which molecular data were available from at
leastoneofthefiveassayplatforms.Ofthese,9,759hadcomplete
data for 4 platforms: aneuploidy, DNA methylation, mRNA and
miRNA. RPPA protein data were available for a subset of samples
(7,858). Hematologic and lymphatic malignancies included acute
myeloid leukemia (LAML), lymphoid neoplasm diffuse large B cell
lymphoma (DLBC), andthymoma (THYM). Solid tumor types were
from gynecologic (ovarian [OV], uterine corpus endometrial
carcinoma [UCEC], cervical squamous cell carcinoma and endocervical
adenocarcinoma [CESC], and breast invasive carcinoma
[BRCA]), urologic (bladder urothelial carcinoma [BLCA], prostate
adenocarcinoma [PRAD], testicular germ cell tumors [TGCT], kidney
renal clear cell carcinoma [KIRC], kidney chromophobe
[KICH], and kidney renal papillary cell carcinoma [KIRP]), endocrine
(thyroid carcinoma [THCA] and adrenocortical carcinoma
[ACC]), core gastrointestinal (esophageal carcinoma [ESCA],
stomach adenocarcinoma [STAD], colon adenocarcinoma
[COAD], and rectum adenocarcinoma [READ]), developmental
gastrointestinal (liver hepatocellular carcinoma [LIHC], pancreatic
adenocarcinoma[PAAD],andcholangiocarcinoma[CHOL]),head
and neck (head and neck squamous cell carcinoma [HNSC]), and
thoracic (lung adenocarcinoma [LUAD], lungsquamouscellcarcinoma
[LUSC], and mesothelioma [MESO]) organ systems. Cancers
of the central nervous system (glioblastoma multiforme
[GBM] and brain lower-grade glioma [LGG]) and soft tissue (sarcoma
[SARC] and uterine carcinosarcoma [UCS]) were represented,
as were cancers from neural-crest-derived tissues,
such as pheochromocytoma and paraganglioma (PCPG), and
melanocytic cancers of the skin (skin cutaneous melanoma
[SKCM]) and eye (uveal melanoma [UVM]). (For a complete list
of the TCGA cancer-type abbreviations, please see https://
gdc.cancer.gov/resources-tcga-users/tcga-code-tables/tcga-
study-abbreviations.)
Clustering by Individual Platforms
We explored the sample groupings from each individual
assay platform. Using aneuploidy (AN), CpG hypermethylation
(METH), mRNA (MRNA), miRNA (MIR), and protein (P), the resultant
number of groups ranged from 10 to 25 (Figure 1). While
cell-of-origin was a dominant feature of the classification, we
observed tumors from different cancer types grouping and
samples within a cancer type dispersing across groups.
Hierarchical clustering of 10,522 samples by chromosome
arm-level aneuploidy yielded ten groups (Figure 1A; Table S1).
Samples were split mainly by those with few alterations (AN7),
those with moderate alterations (AN6,8-10), and those with many
alterations (AN1-5). Over one-third of the samples displayed relatively
sparse aneuploidy in AN7; these were enriched for THCA,
LAML, PRAD, and THYM. We observed more distinct clustering
by cell-of-origin among higher-aneuploid tumors. For example,
AN2, characterized by chromosome (chr) 13 gain and chr18 loss,
was strongly enriched for gastrointestinal tumors (COAD, READ,
and STAD), and chromosomal instability (CIN) ESCA. Consistent
with previous results (Hoadley et al., 2014), squamous (lung,
head and neck, and esophageal) tumors clustered together by
aneuploidy patterns, particularly 3p loss and 3q gain (AN3).
Unsupervised clustering of 10,814 tumors using DNA methylation
data with 3,139 CpG sites that were hypermethylated in at
least one tumor type identified 25 groups. Despite the exclusion
of loci known to be involved in tissue-specific DNA methylation,
tumors originating from the same organ often aggregated by
cancer-type-specific hypermethylation (Figure 1B; Table S2).
This result suggests that cancer-associated DNA hypermethylation
in human cancers is influenced by pre-existing
cell-type-specific chromatin marks or transcriptional programs,
and not just by cell-type-specific DNA methylation patterns.
Tumors within an organ system tended to co-cluster. Consistent
with the aneuploidy analysis, squamous cell carcinomas
292 Cell 173, 291–304, April 5, 2018
(HNSC, ESCA, LUSC, and CESC) associated closely in METH2
and METH3. Gastrointestinal adenocarcinomas (ESCA, STAD,
COAD and READ) were represented in a branch containing
METH10 through METH13.
Unsupervised consensus clustering of 10,165 tumors by
mRNA expression profiles identified 25 groups that contained
at least 40 samples (Figure 1C; Table S3). While tumor type
was a driving feature for many groups, several groups were
comprised of tumors from different organ types. Samples with
squamous morphology components (BLCA, CESC, ESCA,
HNSC, and LUSC) grouped together. Similarly, tumors with
tissue or organ similarities or proximity also grouped
together. These included neuroendocrine and glioma tumors
(GBM, LGG and PCPG), melanomas of the skin and eye
(SKCM and UVM), clear cell and papillary renal carcinomas
(KIRC and KIRP), adrenal cortical and chromophobe renal
(ACC and KICH), hepatocellular and cholangiocarcinomas
(LIHC and CHOL), a gastrointestinal group (COAD, READ,
non-squamous ESCA, READ, and STAD), a digestive system
group (PAAD, STAD, and a few ESCA), hematologic and
lymphatic cancers (LAML, DLBC, and THYM), and two mixed
lung cancer groups (LUAD and LUSC).
Figure 1. Platform-Specific Classification of 10,000 TCGA Cancer Tumor Samples across 33 Cancer Types
(A) Aneuploidy (AN). Unsupervised consensus clustering of 10,522 tumors and chromosomal arm-level amplifications or deletions.
(B) DNA hypermethylation (METH). Clustering of cancer-associated DNA methylation profiles in 10,814 tumors at 1,035 CpG sites lacking DNA methylation in
normal tissues (left) and leukocytes (right). DNA methylation b-values are represented as a color gradient from low (blue) to high (red).
(C) mRNA (MRNA). Unsupervised consensus clustering of 10,165 tumors and variably expressed genes.
(D) microRNA (MIR). Unsupervised hierarchical clustering of 743 expressed mature strands in 10,170 tumors.
(E) Protein (P). Unsupervised hierarchical clustering of 7,858 tumor samples from 32 cancer types across 216 cancer-relevant proteins and phosphoproteins.
Tumor types are color-coded as shown in the lower-right corner.
See also Tables S1–S5.
Cell 173, 291–304, April 5, 2018 293
Unsupervised hierarchical clustering of miRNA expression
profiles from 10,170 tumors yielded 15 groups (Figure 1D;
Table S4). While six groups contained only a single cancer type,
the remaining nine groups each represented a mix of cancer
types. These included a squamous-enriched group (MIR2), a
pan-kidney group (MIR11), and a pan-GI-enriched group (MIR6).
Hierarchical clustering of protein expression data from
7,858 samples across 32 tumor types (LAML did not have
protein data) revealed ten distinct protein (P) groups (Figure 1E;
Table S5). P1 (GBM, LGG) and P2 (DLBC, SARC, PCPG, UCS,
THYM, and metastatic SKCM) were distinguished from the
remaining 8 groups, largely corresponding to mesenchymallike
tumor types with high EMT signatures. Similar to the other
individual data platforms, samples from related organ systems
grouped together: luminal breast and gynecologic cancers
(BRCA-Luminal, UCEC, and OV), plus some liver samples
(LIHC) with high levels of ER-alpha, AR and IGFBP2 comprised
the majority of the P3 and P4 groups. In addition, a pan-kidney
(P6) and a pan-GI (P8) group were identified.
Integrative Clustering across Data Types
We used clustering of cluster assignments (COCA) algorithm
(Hoadley et al., 2014) to assess the overlap of platform-specific
memberships from each of the five molecular platforms (aneuploidy,
mRNA, miRNA, DNA methylation, and RPPA) (Figure 2A).
Many samples similarly grouped together by multiple platformspecific
cluster memberships, both in groups that were defined
by a single tumor type and in tumor types that co-clustered,
such as KIRC and KIRP (pan-kidney). Gastrointestinal
tumors (COAD, READ, STAD, and ESCA adenocarcinomas)
co-clustered in the mRNA, miRNA, and RPPA platforms but
were represented by several distinct DNA methylation clusters.
Squamous histology cancers (LUSC, HNSC, CESC, ESCA, and
BLCA) were similarly classified by the miRNA, mRNA and
RPPA data but were further divided by the aneuploidy and
DNA methylation data. Within pan-gyn cancers (BRCA, OV,
UCEC, and UCS), RPPA data suggested that ovarian serous
cystadenocarcinoma (OV) and UCEC (and ER+ LIHC) shared
similarities at the protein level, whereas miRNA, mRNA, and
DNA methylation data were grouped by their organ sites. Also
of note, 13% of BRCA formed a subtype distinct from the majority
of other BRCA, influenced by the mRNA and DNA methylation
platforms.
While COCA showed high consistency across most data platforms,
we found less concordance for aneuploidy, where more
than a third of the samples were defined by few to no aneuploidy
events. This group, AN7, included almost all the THCA and LAML
samples, while not well defined by aneuploidy had strong
concordance among the other data platforms. COCA is less
powerful when the molecular patterns are not strong enough to
specify a distinct group on multiple individual platforms. To complement
this analysis, we explored joint clustering across all platforms
simultaneously.
We performed integrative molecular subtyping with iCluster
using the four most complete data types (copy number, DNA
methylation, mRNA, and miRNA) across 9,759 tumor samples,
identifying 28 iClusters (Figure 2B; Table S6). The relative contribution
of each platform to the overall clustering was quantified
by summing the different platform feature weights on the iCluster
latent variables. Copy-number alterations contributed 47% to
the overall integrated clustering results, followed by the transcriptome
(mRNA and miRNA) at 42%, and DNA methylation
at 11%.
For 16 of the tumor types, over 80% of samples grouped
together in the same iCluster. Eight iClusters were dominated by
a single tumor type (C24:LAML, C11:LGG [IDH1 mut], C6:OV,
C8:UCEC, C12:THCA, C16:PRAD, C26:LIHC, C14:LUAD). Others
contained tumors from similar or related cells or tissues: C28:pankidney
(KIRC, KIRP), C15:SKCM/UVM-melanoma of the skin
(SKCM) and eye (UVM), C23:GBM/LGG (IDH1wt), and C5:CNS/
endocrine. Six tumor types had more diverse iCluster membership,
with less than 50% of tumors represented in a given iCluster
(BLCA, UCS, HNSC, ESCA, STAD, and CHOL).
The pan-GI cohort separated into three iClusters (C1, C4, and
C18), primarily driven by differences in DNA methylation
profiles. C1:STAD (Epstein-Barr virus [EBV]-CIMP) consisted
of hypermethylated EBV-associated tumors, and C18:pan-GI
(MSI) consisted mostly of microsatellite instability (MSI) tumors
of STAD and COAD. C4:pan-GI (CRC) was predominantly
COAD and READ with chromosomal instability (CIN) and a
distinct aneuploidy profile (Figure 2B). The pan-squamous
cohort formed three iClusters (C10, C25, and C27). The
majority of LUSC fell into C10:pan-SCC, and nearly all CESC
fell into C27:pan-SCC (human papillomavirus [HPV]). Even
though all squamous iClusters were characterized by chromosome
3q amplification, unique features defined C10:pan-SCC
(9p deletion) and C25:pan-SCC (Chr11 amp) (Figure 2B).
Among mixed tumor type iClusters, three were defined
by copy-number alterations. C7:mixed was characterized by
chr9 deletion, C2:BRCA (HER2 amp) mainly consisted of
ERBB2-amplified tumors (BRCA, BLCA, and STAD), and
C13:mixed (Chr8 del) contained highly aneuploid tumors,
including a mixture of BRCA-Basal, UCEC (CN-high subtype),
UCS, and BLCA. C3 and C20 were defined by their non-tumorcell
components including immune and stromal features.
We explored the non-tumor components of the iClusters in
more detail. We estimated the stromal fraction as 1 minus tumor
purity and the leukocyte fraction based on DNA methylation (Figure
3). C20 had the highest median stromal fraction followed by
C14:LUAD, C10:pan-SCC, and C3 (Figure 3A). Each of these
iClusters also displayed elevated leukocyte fractions (Figure 3B).
To estimate how much of the stromal fraction was due to immune
cell infiltration, we plotted the stromal fraction versus the
leukocyte fraction (Figure 3C). In C3, more of the stromal fraction
was defined by leukocytes than in C20. C3 contained predominately
mesenchymal cancers, which we labeled C3:mesenchymal
(immune). C20 tumors were predominately mixed epithelial
cancers, which we labeled C20:mixed (stromal/immune).
To characterize composition and relative homogeneity of each
iCluster, we computed the dominant-cancer-type proportion
within each iCluster and plotted it against the mean iCluster
silhouette width, a measure of within-group homogeneity (Figure
2C). The silhouette widths ranged from À0.05 to 0.59,
with the highest silhouette widths belonging to single-cancertype-dominant
iClusters (C11:LGG [IDH1 mut], C12:THCA,
C16:PRAD, and C24:LAML). Interestingly, 6 of the 7 pan-organ
294 Cell 173, 291–304, April 5, 2018
system iClusters (pan-GI: C1, C4, C18; pan-SCC: C25, C27, and
pan-kidney: C28) had similar ranges of silhouette widths to those
of single cancer-type dominant iClusters, suggesting that these
were as robust as the cancer-type-dominant iClusters. iClusters
driven by a shared specific chromosomal alteration (e.g.,
C13:mixed [chr8 del]) tended to compose multiple tumor types
and appeared to have among the lowest silhouette widths,
suggesting substantial molecular heterogeneity.
We used a Sankey diagram to further visualize the relationship
between the iCluster classification, cancer types, and organ systems
(Figure 2D). Pan-kidney mapped almost entirely to C28,
except for KICH, which grouped with ACC in C9, characterized
Figure 2. Cross-Platform Classification Revealed Genomic, Epigenomic, and Transcriptomic Similarities and Differences across
Cancer Types
(A) COCA clusters. Membership for individual clusters for each of the five molecular platforms—aneuploidy (AN), methylation (Meth), miRNA expression (miR),
mRNA, and RPPA—is displayed as a separate binary membership variable in a distinct row. For the mRNA platform, only clusters containing >40 samples were
considered. Samples are labeled for membership of each platform-specific cluster (red, member; white, non-member; gray, not evaluated on the platform). Order
of samples and platform-specific clusters were determined by hierarchical clustering using a binary distance matrix and average linkage. Column annotation
shows cancer type and tissue organ systems of each sample; row annotations reflect the platform for each classification (bright pink, AN; purple, Meth; light
turquoise, miR; dark turquoise, mRNA; orange, RPPA).
(B) iCluster. Data used for integrated analysis of iClusters. RPPA data are also included in the heatmap to visualize proteomic patterns across the integrated
clusters.
(C) iCluster robustness versus composition. Pie charts show the cancer-type composition within each iCluster and the size is proportional to the membership
size. The cancer type accounting for the highest proportion of members within the iCluster was considered the dominant cancer type. The y coordinate of each pie
center reflects this dominant cancer-type proportion; the x coordinate was determined by the iCluster silhouette width.
(D) Relationship of TCGA tumor type, iCluster, and Pan-Organ system. The Sankey diagram demonstrates the tumor-type composition of each iCluster. The pancancer
designations are shown on the right.
See also Tables S6 and S7.
Cell 173, 291–304, April 5, 2018 295
(legend on next page)
296 Cell 173, 291–304, April 5, 2018
by a high frequency of hypodiploid samples (Davis et al., 2014;
Zheng et al., 2016). However, pan-GI, pan-gyn, and pan-squamous
were distributed among multiple iClusters. C20:mixed
(stromal/immune) was fairly heterogeneous, including pan-GI,
pan-gyn, and pan-squamous. Pan-gyn and pan-squamous
overlapped, as cervical cancer is primarily a squamous cell
carcinoma. This analysis demonstrated that the iClusters were
strongly influenced by the cell type of origin for the individual
cancers, though this relationship was not absolute.
Tumor Maps of Organ Systems
We visualized the samples by calculating Euclidean distances
between the iCluster latent variables for all sample pairs and projecting
the distances onto a 2D layout with TumorMap (Figure 4A;
Table S7) (Newton et al., 2017). We overlaid the tumor-type colors
to reveal that tumors systematically assembled along the major
organ systems (Figure 4B), lending further support for the organ-systemgroups
explored in accompanying papers (Figure 4C)
(Berger et al., 2018; Campbell et al., 2018; Liu et al., 2018; Malta
Figure 3. Cellularity of the Tumor Microenvironment among iCluster Samples
(A) Stromal fraction of tumor samples. The stromal fraction, defined by subtracting tumor purity (estimated by ABSOLUTE) from one, is shown for 9,057 TCGA
tumor samples, segregated by iCluster membership.
(B) Leukocyte fraction. Leukocyte fraction, estimated from DNA methylation arrays, for 9,417 tumor samples, for each iCluster, with the exception of C24:LAML
and C21:DLBC.
(C) Leukocyte fraction versus stromal fraction. Points near the diagonal correspond to tumor samples in which non-tumor stromal cells are nearly all immune cells,
and points away from the diagonal correspond to a more mixed or a non-immune stromal tumor microenvironment. Points in the upper-left triangle of each plot
are estimation artifacts.
A B C
D
E F
Figure 4. The iCluster TumorMap
(A–F) The map layout was computed from sample Euclidean similarity in the iCluster latent space, and similar samples are positioned in close proximity to each
other. Each spot represents a single sample and is colored to represent attributes as described for each panel including (A) iCluster, (B) disease type, and (C)
organ system. Organ systems highlighted include pan-kidney, red; pan-gyn, orange; pan-GI, blue; pan-squamous, purple; and those that overlap pan-gyn and
pan-squamous, light purple.
(D) Subtypes from the pan-kidney analysis (Ricketts et al., 2018). Clear cell renal cell carcinoma (ccRCC), green; papillary renal cell carcinoma type 1 (PRCC T1),
blue; papillary renal cell carcinoma type 2 (PRCC T2), yellow; unclassified papillary renal cell carcinoma (PRCC Unc.), dark gray; CpG island methylator phenotype
renal cell carcinoma (RCC-CIMP), red; and chromophobe renal cell carcinoma (ChRCC), purple.
(E) Subtypes from the pan-gyn group (Berger et al., 2018). Not hypermutated, with low copy-number changes (non-HM CNV low), red; hypermutated, with low
copy-number changes (HM), blue; high levels of leukocyte infiltration (immune), green; low AR or PR expression (AR/PR low), orange; and high androgen receptor
(AR) or progesterone receptor (PR) expression (AR/PR high), dark gray.
(F) Subtypes from the pan-GI group (Liu et al., 2018). High Epstein-Barr virus (EBV) burden, red; microsatellite instability (MSI), blue; hypermutated without MSI
(HM-SNV), gold; chromosomal instability tumors (CIN), purple; and genome stable (GS) with low aneuploidy, green. The gray dots represent non-highlighted
diseases.
Cell 173, 291–304, April 5, 2018 297
et al., 2018; Ricketts et al., 2018). More subtle differences within
individual iClusters were apparent, potentially signifying important
distinctions from the dominant cell-of-origin-associated
signals. Kidney tumors separated into KICH, KIRC, and KIRP
(Ricketts et al., 2018), and CIMP kidney tumors were positioned
near the Pan-GI CIMP tumors, suggesting similarities driven
by DNA hypermethylation data (Figure 4D). Pan-gyn subtypes
displayed partial overlap (Berger et al., 2018) (Figure 4E). Pangyn
samples were broadly distributed, accounting for at least
5% of samples in 11 of the 28 iClusters. However, the majority
of cervical cancers fell into the squamous C27:pan-SCC (HPV)
with HPV-positive HNSC and BLCA, whereas other samples fell
primarily within C6:OV, C19:BRCA (luminal) and C8:UCEC,
reflecting their cell-of-origin and hormonal dependency (Berger
et al., 2018). The pan-GI tumors separated into distinct molecular
subtypes represented by MSI tumors, hypermutated-SNV tumors,
genome-stable tumors, CIN tumors, and EBV-associated
gastric cancers (Liu et al., 2018) (Figure 4F).
The TumorMap landscape showed that tumors with similar
pathologic classification tended to assemble together, even
though histopathologic information was not used in the map
generation (Figure 5A). This result underscores the influence
BLCA
ACC
BLCA
BRCA
CESC
CHOL
GBM GBM/LGG
HNSC
KICH
KIRC
KIRP
LAML
LIHC
LUAD
LUSC
KIRP
PAAD
PCPG
PRAD
COAD/
READ
MESO
SARC
SKCM
STAD
STAD
ESCA/STAD
LGG
TGCT
THCA
UCECUCS
OV
THYM
UVM
DLBC ESCA
STAD/COAD
CESC
BRCA
HighLow
Stemness index
DNA methylation-based stemness index
HighLow
Stemness index
BLCA
ACC
BLCA
BRCA
CESC
CHOL
GBM GBM/LGG
HNSC
KICH
KIRC
KIRP
LAML
LIHC
LUAD
LUSC
KIRP
PAAD
PCPG
PRAD
COAD/
READ
MESO
SARC
SKCM
STAD
STAD
ESCA/STAD
LGG
TGCT
THCA
UCECUCS
OV
THYM
UVM
DLBC ESCA
STAD/COAD
CESC
BRCA
mRNA expression-based stemness index
Wound Healing (C1)
TGF-beta Dominant (C6)
Lymphocyte Depleted (C4)Inflammatory (C3)
Immunologically Quiet (C5)
IFN-gamma Dominant (C2)
BLCA
ACC
BLCA
BRCA
CESC
CHOL
GBM GBM/LGG
HNSC
KICH
KIRC
KIRP
LAML
LIHC
LUAD
LUSC
KIRP
PAAD
PCPG
PRAD
COAD/
READ
MESO
SARC
SKCM
STAD
STAD
ESCA/STAD
LGG
TGCT
THCA
UCECUCS
OV
THYM
UVM
DLBC ESCA
STAD/COAD
CESC
BRCA
Immune subtype
Squamous cell carcinoma
SarcomaOther carcinoma
OtherLymphomaLeukemia
Adenocarcinoma
BLCA
ACC
BLCA
BRCA
CESC
CHOL
GBM GBM/LGG
HNSC
KICH
KIRC
KIRP
LAML
LIHC
LUAD
LUSC
KIRP
PAAD
PCPG
PRAD
COAD/
READ
MESO
SARC
SKCM
STAD
STAD
ESCA/STAD
LGG
TGCT
THCA
UCECUCS
OV
THYM
UVM
DLBC ESCA
STAD/COAD
CESC
BRCA
HistopathologyA B
C D
Figure 5. Sample Characteristics in the
Context of the iCluster TumorMap
(A–D) The TumorMap layout is as described for
Figure 4.
(A) Histopathology. Colors indicate major
histopathology types. Adenocarcinoma, yellow;
squamous cell carcinoma, purple; other carcinomas,
green; sarcomas, light blue; leukemias,
dark blue; lymphomas, magenta; and other, red.
(B) Immune subtypes. Wound-healing group, red;
IFN-gamma, yellow; inflammatory group, green;
lymphocyte-depleted, light blue; immunologically
quiescent, dark blue; and transforming growth
factor (TGF)-beta activity, magenta.
(C and D) Stemness signatures for (C) mRNA
and (D) DNA methylation from Malta et al. (2018)
are displayed. Increasing red colors indicate
increasing stemness index.
of the cell of origin on the molecular patterns
observed in cancer and provides
further support for the pan-squamous
sub-analysis (Campbell et al., 2018).
Immune-signaling subtypes identified in
Thorsson et al. (2018) also co-localized
on the TumorMap, indicating relationships
between the iClusters, histopathology,
and the types of immune infiltration
(Figure 5B). Pan-squamous tumors
shared predominant wound healing
and interferon (IFN)-gamma-dominant
immune signatures.
Cancer stemness has been proposed
as a possible mechanism for treatment
resistance and as a driver of the ability
of subpopulations to repopulate new
metastatic niches (Jin et al., 2017). Two stemness indices (Malta
et al., 2018), based on mRNA expression and on DNA methylation
data, revealed aggregation of high stemness tumors
across distinct regions of the TumorMap (Figures 5C and 5D).
TGCT showed strong enrichment of both signatures while
others, such as LAML, showed strong enrichment only for the
mRNA-based signature.
Mutational Assessment of iClusters
We did not use tumor mutation data in generating iClusters due
to sparsity of mutations; however, we did use mutational burden
and signatures for characterization. Overall somatic mutation
burden varied among iClusters. Melanomas and lung adenocarcinomas
have been shown to have relatively high mutation rates,
and we observed similar results with C15:SKCM/UVM and
C14:LUAD (Lawrence et al., 2013). Pan-GI and pan-squamous
were also associated with overall higher somatic mutational
burdens (Figure 6A). Mutation frequencies varied widely within
the two iClusters with the most diverse tumor compositions:
C3:mesenchymal (immune) and C20:mixed (stromal/immune).
Mutational signatures (Covington et al., 2016) also varied
among iClusters. Expected signatures were apparent, such as
298 Cell 173, 291–304, April 5, 2018
enrichment for UVB signatures in C15:SKCM/UVM, smoking in
C14:LUAD, and POLE mutation in hypermutated samples of
C8:UCEC and C4:pan-GI (CRC) (Figure 6B). We also found
enhanced signatures in a few of our pan-organ groups such as
C18:pan-GI (MSI), which showed enrichment of known (CpG,
toxins) and unknown mutational signatures, some of which are
likely related to the high proportion of mismatch-repair deficient
tumors in this group (Figure 6B).
Pathway Characteristics of the PanCancer iCluster
Subtypes
We compared the PARADIGM-inferred activation of $19,000
pathway features (Vaske et al., 2010), as well as expressionbased
scores of 22 gene programs defined previously
(Hoadley et al., 2014), and 18 canonical targetable pathways,
to identify differential pathway characteristics across the 28
iClusters (Figure 7; Table S8). C28:pan-kidney was characterized
by high hypoxia signaling, retinoid metabolism, low proliferation,
PPAR-RXR pathway and immune-related signaling, including
C9:ACC/KICH
C23:GBM/LGG
(IDH1wt)
C24:LAML
C11:LGG
(IDH1m
ut)
C16:PRAD
C28:Pan−Kidney
C12:THCA
C22:TGCT
C17:BRCA
(Chr8q
am
p)
C2:BRCA
(HER2am
p)
C19:BRCA
(Lum
inal)
C21:DLBC
C6:OVC26:LIHC
C4:Pan−GI(CRC)
C5:CNS/Endocrine
C20:Mixed
(Strom
al/Im
m
une)
C3:Mesenchym
al(Im
m
une)
C7:Mixed
(Chr9del)
C14:LUAD
C1:STAD
(EBV−CIMP)
C27:Pan−SCC
(HPV)
C10:Pan−SCC
C13:Mixed
(Chr8del)
C25:Pan−SCC
(Chr11am
p)
C15:SKCM/UVM
C8:UCEC
C18:Pan−GI(MSI)
0.2 0.4 0.6 0.8 1
MutSig 5 (Smoking)
MutSig 4 (APOBEC)
MutSig 9 (APOBEC)
MutSig 7 (Arsenic?)
MutSig 11
MutSig 12
MutSig 15 (TMZ)
MutSig 21 (ABOBEC3G)
MutSig 3
MutSig 1 (AC>AN;AT>AN)
MutSig 20
MutSig 8 (C>G)
MutSig 18 (T>A)
MutSig 10 (Toxin)
MutSig 13 (Toxin/Liver)
MutSig 17
MutSig 2 (Smoking?)
MutSig 16 (8−oxoG)
MutSig 14 (UVB)
MutSig 6 (CpG)
MutSig 19 (POLE)
C24:LAML
C11:LGG
(IDH1m
ut)
C12:THCA
C5:CNS/Endocrine
C22:TGCT
C9:ACC/KICH
C3:Mesenchym
al(Im
m
une)
C16:PRAD
C2:BRCA
(HER2am
p)
C19:BRCA
(Lum
inal)
C17:BRCA
(Chr8q
am
p)
C23:GBM/LGG
(IDH1wt)
C28:Pan−Kidney
C20:Mixed
(Strom
al/Im
m
une)
C8:UCEC
C27:Pan−SCC
(HPV)
C13:Mixed
(Chr8del)
C6:OVC26:LIHCC1:STAD
(EBV−CIMP)
C4:Pan−GI(CRC)
C25:Pan−SCC
(Chr11am
p)
C15:SKCM/UVM
C21:DLBC
C7:Mixed
(Chr9del)
C14:LUAD
C10:Pan−SCC
C18:Pan−GI(MSI)
0.01
0.1
1
10
100
1000
Somaticmutationfrequency(/Mb)
n=119 496 292 208 131 474 428 477 548 181 140 617 748 142 451 202 189 285 49 375 460 272 32 200 454 607 256 373
A
B
Figure 6. Mutation Patterns of iClusters
(A) Somatic mutation frequency (log10) per iCluster
sorted by median mutations per megabase.
Somatic mutation frequencies were calculated
using a filtered MC3 mutation annotation file to
determine the total number of mutations per
sample, normalized by whole-exome sequencing
coverage as described in Knijnenburg et al. (2018).
Bars represent median mutation frequency for
each iCluster.
(B) Mutational signatures (Covington et al., 2016)
enriched in iClusters. Mutational signature scores
were scaled per sample by the overall mutation
rate. The means of scaled signature scores
were calculated for each iCluster and log10transformed.
Hierarchical clustered data are
displayed in the heatmap (blue, low; red, high).
immune checkpoints PD-1 and CTLA4.
However, KICH co-clustered with ACC
in C9:ACC/KICH, lacking hypoxic and
immune signals and showing low activity
in nearly all pathways. Both these tumor
types have previously been characterized
as hypodiploid (Davis et al., 2014; Zheng
et al., 2016).
Despite having very different cancer
type compositions, the pan-squamous
iClusters C10:pan-SCC, C25:pan-SCC
(chr11 amp), and C27:pan-SCC (HPV)
shared many pathway characteristics. All
had high levels of squamous-cell-related
signaling (dNp63 and TAp63 complexes
and GP6), proliferation-related pathways,
relatively high hypoxia, immune-related
signaling, and high basal signaling.
Although the Pan-GI iClusters
C1:STAD (EBV-CIMP), C4:pan-GI (CRC),
and C18:pan-GI (MSI) shared some
common characteristics such as relatively high proliferation
signaling, these iClusters diverged in some respects. Immunerelated
signaling was high in C1:STAD (EBV-CIMP) and
C18:pan-GI (MSI), but not in C4:pan-GI (CRC). In addition,
C20:mixed (stromal/immune) contained 32% Pan-GI samples
and also displayed strong immune-related signaling. Beta-catenin/cell-cell
adhesion signaling appeared high in C4:pan-GI
(CRC), C18:pan-GI (MSI), and C20:mixed (stromal/immune),
but not in the smaller C1:STAD (EBV-CIMP).
Most UCS co-clustered with a subset of Basal BRCA, UCEC
and BLCA in C13:mixed (chr8 del), with high basal signaling
and proliferation in the absence of immune activation. Interestingly,
another subset of Basal breast cancers co-clustered with
squamous cancers in the C20:mixed (stromal/immune), which
also had high basal signaling and proliferation, but activated
immune signaling. OV and UCEC shared a number of pathway
similarities with cervical cancers and a subset of Basal breast
cancers despite falling into different iClusters. These similarities
included high proliferation and DNA repair pathways and basal
Cell 173, 291–304, April 5, 2018 299
Figure 7. Pathway Features Characterizing
the PanCancer-33 iCluster Subtypes
(A) PARADIGM pathway heatmap. Regulatory nodes
with differential PARADIGM-inferred pathway levels
(IPL) with at least 15 downstream regulatory targets
with differential inferred activities between iClusters
are shown for one versus rest comparisons. Samples
are arranged by iCluster order; regulatory nodes are
hierarchically clustered using 1-Pearson correlation
as distance and average linkage. Red-blue intensities
represent median-centered IPLs from low (blue) to
high (red).
(B) Gene programs and canonical pathway values.
The 22 Gene Programs (Hoadley et al., 2014) and 20
pathway signatures reflecting drug targets and
canonical pathways (found in Table S4 of Hoadley
et al. [2014]) were hierarchically clustered using
1-Pearson distance and complete linkage and are
shown with samples arranged by iCluster subtypes in
numerical order. Red-blue intensities represent
signature scores from low (blue) to high (red).
See also Tables S8 and S9.
300 Cell 173, 291–304, April 5, 2018
signaling. Although the estrogen-signaling gene program (GP7)
was very high in the breast cancer iClusters C2:BRCA (HER2
amp) and C19:BRCA (luminal), that program did not appear to
be high in the other gynecological cancers.
DISCUSSION
With nearly three times more tumors and tumor types profiled in
this PanCancer Atlas analysis, we were able to detect more
integrated molecular subtypes than we had reported in the
original Pan-Cancer-12 analysis (Hoadley et al., 2014). We first
performed unsupervised consensus clustering of tumor
profiles from each of the 5 platforms, revealing from 10 to 25
platform-specific molecular subsets within $10,000 tumors,
each showing significant compositional heterogeneity based
on classical tumor taxonomy (Figure 1). Aneuploidy classifications
were weakly consistent with other classifications, in
part due to low numbers of arm-level copy-number events in
one-third of the tumors. We explored cross-platform cluster
relationships using COCA and employed iCluster to integrate
the multiplatform molecular data simultaneously into a final
28-cluster solution.
While a third of iClusters were mostly homogeneous for a single
tumor type, the other two-thirds showed varying degrees of
heterogeneity. The most diverse group, C20:mixed (stromal/
immune), contained a remarkable 25 tumor types (Figures 2C
and 2D). Most of the heterogeneous iClusters, including
C20:mixed (stromal/immune), contained tumor types that fell
within four major cell-of-origin, or organ system, patterns (Figure
2D): pan-GI, pan-gyn, pan-squamous, and pan-kidney.
Individual cluster assignments, COCA, and iCluster-determined
molecular subsets were concordant, and confirmed the multiplatform
co-clustering of different kidney malignancies (pankidney),
various gastrointestinal malignancies (pan-GI), diverse
squamous cell malignancies (pan-squamous) and most gynecological
malignancies (pan-gyn) into molecular subgroups, each
with subordinate platform-specific subsets (Figure 2A). Consequently,
these four major cell-of-origin patterns are the subject
of separate in-depth reports detailing their distinguishing
genomic and molecular features (Berger et al., 2018; Campbell
et al., 2018; Liu et al., 2018; Malta et al., 2018; Ricketts et al.,
2018). These iCluster assignments have potential clinical utility,
and their multi-platform basis suggests that this new subclassification
system might further improve the management of the
1%–3% of all cancer patients newly diagnosed with cancer of unknown
primary (CUP). Using either RNA (Hainsworth et al., 2013)
or DNA methylation (Moran et al., 2016) profiling has recently led
to improved patient outcomes by better defining the tissues of
origin for this diverse group of life-threatening malignancies.
While separate spatial co-localization of the four major cell-oforigin
patterns was generally evident in the TumorMap visualization
(Figure 4), heterogeneity was also apparent between subsets
within these individual iClusters, even those with generally
similar tumor type, organ system, and histopathology. This indicates
that while iCluster groupings were strongly influenced by
organ and cell-of-origin patterns, this influence did not fully
determine their molecular groupings such as seen in our largest
and most heterogeneous iCluster, C20:mixed (stromal/immune),
which contained 25 of our 33 tumor types. The spatial relationships
of C20:mixed (stromal/immune) tumors to C10:pan-SCC
and C13:mixed (chr8 del) tumors may be determined in part by
their different mRNA and DNA methylation-based stemness signatures
(Figures 5C and 5D).
Interrogation of individual iClusters for their differentiating
PARADIGM pathway features, canonical pathways, and gene
programs amenable to drug targeting identified strong immunerelated
signaling features for both C3:mesenchymal (immune)
and C20:mixed (stromal/immune) tumors, suggesting that
they may share potential susceptibility to immunotherapy. We
noted that C20:mixed (stromal/immune) and C3:mesenchymal
(immune) tumors were commonly enriched for gene programs
representing PD1, CTLA4, and GP2-T cell/B cell activation (Figure
7B), indicating that new therapies targeting these specific
immune pathways might be appropriate. Another potentially clinically
relevant similarity was upregulation of different druggable
growth factor signaling pathways (Figure 7B). In particular, our
PARADIGM analysis showed that C3:mesenchymal (immune)
and C20:mixed (stromal/immune) tumors shared upregulated
JAK2/STAT1,3,6 signaling with C14:LUAD tumors and
C10:pan-SCC, pointing to the possibility of treating these diverse
iCluster tumors with JAK-STAT agents currently approved to
treat rheumatoid arthritis, myelofibrosis, polycythemia vera, and
other non-malignant diseases (Banerjee et al., 2017).
Compared to the seemingly discohesive groupings of
the 17 heterogeneous iClusters, the 11 most homogeneous
iClusters (C6:OV, C8:UCEC, C11:LGG [IDH1 mut], C12:THCA,
C14:LUAD, C15:SKCM/UVM, C16:PRAD, C19:BRCA [luminal],
C21:DLBC, C24:LAML, C26:LIHC) had higher silhouette widths,
uniform tumor types, and histopathologies, but showed surprising
degrees of spatial discohesion in the TumorMap. These
anatomically homogeneous iClusters also showed mixed types
of immune infiltration and variable degrees of stemness, attesting
to their underlying molecular heterogeneity, as previously
reported (Cancer Genome Atlas Network, 2015; Cancer Genome
Atlas Research Network, 2011, 2012, 2014a, 2014b, 2015a,
2015b, 2017; Cancer Genome Atlas Research Network et al.,
2013a, 2013b; Robertson et al., 2017).
While malignancies arising from the same anatomical site
have traditionally been treated clinically as a single entity,
histologic and molecular sub-classifications are now routinely
used to determine treatments for subtypes of lung, breast,
gastrointestinal, skin and bone marrow derived malignancies.
As drugs become increasingly clinically available to target
such cancer-driving pathway targets as ALK, EGFR, ERBB2,
ERa, KIT, BRAF, and ABL1, the traditional system of anatomic
cancer classification should be supplemented by a classification
system based on molecular alterations shared by tumors
across different tissue types (Hoadley et al., 2014; Saunders
et al., 2012). This concept has led to the development of socalled
basket or umbrella trials, such as the NCI-MATCH study,
to investigate the feasibility and validity of this new clinical
approach (Ramos et al., 2015). However, exceptions that challenge
this concept have also become apparent from such
notable examples as the unpredictable clinical responses to a
potent BRAF inhibitor across diverse malignancies all
expressing the same BRAF mutation (Saunders et al., 2012).
Cell 173, 291–304, April 5, 2018 301
Integrated molecular tumor profiling such as described here,
and in our previous Pan-Cancer-12 analysis, may improve
basket-trial design by considering both mutations and
oncogenic signaling pathways along with consideration of
each tumor’s tissue-specific or cell-of-origin context (Hoadley
et al., 2014).
STAR+METHODS
Detailed methods are provided in the online version of this paper
and include the following:
d KEY RESOURCES TABLE
d CONTACT FOR REAGENT AND RESOURCE SHARING
d EXPERIMENTAL MODEL AND SUBJECT DETAILS
B Human Subjects
d METHOD DETAILS
B Sample Processing
B Pathology Review
B Somatic Copy-Number Alterations
B DNA methylation
B RNA Data Batch Correction
B mRNA
B miRNA
B Protein
B Integrative clustering with iCluster
B Cancer Immune Subtypes
B Leukocyte and Stromal Fraction Estimates
B TumorMap
B PARADIGM
B Gene Programs/Canonical pathways
d QUANTIFICATION AND STATISTICAL ANALYSES
d DATA AND SOFTWARE AVAILABILITY
SUPPLEMENTAL INFORMATION
Supplemental Information includes nine tables and can be found with this
article online at https://doi.org/10.1016/j.cell.2018.03.022.
ACKNOWLEDGMENTS
Weare gratefultothe patients and familieswho contributed tothis study.Wealso
thank the NCI TCGA Program Office and NHGRI coupterpart for organizational
and logistical support. This work was supported by NIH grants (U54
HG003273, U54 HG003067, U54 HG003079, U24 CA143799, U24 CA143835,
U24 CA143840, U24 CA143843, U24 CA143845, U24 CA143848, U24
CA143858, U24 CA143866, U24 CA143867, U24 CA143882, U24 CA143883,
U24 CA144025, and P30 CA016672).
AUTHOR CONTRIBUTIONS
Conceptualization: K.A.H., J.M.S., C.C.B., and P.W.L. Data Curation: K.A.H.,
A.D.C., V.T., R.A., R.B., and T.H. Formal Analysis: K.A.H., C.Y., T.H.,
D.M.W., E.D., R.S., A.M.T., A.D.C., V.T., R.A., R.B., C.K.W., F.S.-V., A.G.R.,
M.S.L., and T.M.M. Composition of Figures and Graphical Abstract: T.H.,
A.G.R., D.M.W., C.Y., and P.W.L. Writing – Original Draft: K.A.H., C.Y., T.H.,
D.M.W., A.J.L., A.M.T., V.T., R.A., M.W., A.G.R., B.G.S., C.C.B., and P.W.L.
Writing – Review & Editing: K.A.H., C.Y., T.H., D.M.W., A.J.L., E.D., R.S.,
A.M.T., A.D.C., V.T., R.A., R.B., C.K.W., M.W., F.S.-V., A.G.R., B.G.S.,
M.S.L., H.N., T.M.M., J.M.S., C.C.B., and P.W.L. Supervision: K.A.H.,
and P.W.L.
DECLARATION OF INTERESTS
Michael Seiler, Peter G. Smith, Ping Zhu, Silvia Buonamici, and Lihua Yu are
employees of H3 Biomedicine, Inc. Parts of this work are the subject of a
patent application: WO2017040526 titled ‘‘Splice variants associated with
neomorphic sf3b1 mutants.’’ Shouyoung Peng, Anant A. Agrawal, James
Palacino, and Teng Teng are employees of H3 Biomedicine, Inc. Andrew D.
Cherniack, Ashton C. Berger, and Galen F. Gao receive research support
from Bayer Pharmaceuticals. Gordon B. Mills serves on the External Scientific
Review Board of Astrazeneca. Anil Sood is on the Scientific Advisory Board for
Kiyatec and is a shareholder in BioPath. Jonathan S. Serody receives funding
from Merck, Inc. Kyle R. Covington is an employee of Castle Biosciences, Inc.
Preethi H. Gunaratne is founder, CSO, and shareholder of NextmiRNA Therapeutics.
Christina Yau is a part-time employee/consultant at NantOmics. Franz
X. Schaub is an employee and shareholder of SEngine Precision Medicine, Inc.
Carla Grandori is an employee, founder, and shareholder of SEngine Precision
Medicine, Inc. Robert N. Eisenman is a member of the Scientific Advisory
Boards and shareholder of Shenogen Pharma and Kronos Bio. Daniel J. Weisenberger
is a consultant for Zymo Research Corporation. Joshua M. Stuart is
the founder of Five3 Genomics and shareholder of NantOmics. Marc T.
Goodman receives research support from Merck, Inc. Andrew J. Gentles is
a consultant for Cibermed. Charles M. Perou is an equity stock holder, consultant,
and Board of Directors member of BioClassifier and GeneCentric Diagnostics
and is also listed as an inventor on patent applications on the Breast
PAM50 and Lung Cancer Subtyping assays. Matthew Meyerson receives
research support from Bayer Pharmaceuticals; is an equity holder in, consultant
for, and Scientific Advisory Board chair for OrigiMed; and is an inventor of
a patent for EGFR mutation diagnosis in lung cancer, licensed to LabCorp.
Eduard Porta-Pardo is an inventor of a patent for domainXplorer. Han Liang
is a shareholder and scientific advisor of Precision Scientific and Eagle Nebula.
Da Yang is an inventor on a pending patent application describing the use of
antisense oligonucleotides against specific lncRNA sequence as diagnostic
and therapeutic tools. Yonghong Xiao was an employee and shareholder of
TESARO, Inc. Bin Feng is an employee and shareholder of TESARO, Inc.
Carter Van Waes received research funding for the study of IAP inhibitor
ASTX660 through a Cooperative Agreement between NIDCD, NIH, and Astex
Pharmaceuticals. Raunaq Malhotra is an employee and shareholder of
Seven Bridges, Inc. Peter W. Laird serves on the Scientific Advisory Board
for AnchorDx. Joel Tepper is a consultant at EMD Serono. Kenneth Wang
serves on the Advisory Board for Boston Scientific, Microtech, and Olympus.
Andrea Califano is a founder, shareholder, and advisory board member of
DarwinHealth, Inc. and a shareholder and advisory board member of Tempus,
Inc. Toni K. Choueiri serves as needed on advisory boards for Bristol-Myers
Squibb, Merck, and Roche. Lawrence Kwong receives research support
from Array BioPharma. Sharon E. Plon is a member of the Scientific Advisory
Board for Baylor Genetics Laboratory. Beth Y. Karlan serves on the Advisory
Board of Invitae.
Received: November 19, 2017
Revised: February 12, 2018
Accepted: March 8, 2018
Published: April 5, 2018
REFERENCES
Alencar, A., and Polley, T. (2011). DrL (VxOrd). http://wiki.cns.iu.edu/pages/
viewpage.action?pageId=1704113.
Banerjee, S., Biehl, A., Gadina, M., Hasni, S., and Schwartz, D.M. (2017). JAKSTAT
signaling as a target for inflammatory and autoimmune diseases: current
and future prospects. Drugs 77, 521–546.
Beck, A.H., Espinosa, I., Edris, B., Li, R., Montgomery, K., Zhu, S., Varma, S.,
Marinelli, R.J., van de Rijn, M., and West, R.B. (2009). The macrophage colonystimulating
factor 1 response signature in breast carcinoma. Clin. Cancer Res.
15, 778–787.
Berger, A.C., Korkut, A., Kanchi, R.S., Hegde, A.M., Lenoir, W., Liu, W., Liu, Y.,
Fan, H., Shen, H., Ravikumar, V., et al. (2018). A comprehensive Pan-Cancer
302 Cell 173, 291–304, April 5, 2018
molecular study of gynecologic and breast cancers. Cancer Cell 33. https://
doi.org/10.1016/j.ccell.2018.03.014.
Calabro` , A., Beissbarth, T., Kuner, R., Stojanov, M., Benner, A., Asslaber, M.,
Ploner, F., Zatloukal, K., Samonigg, H., Poustka, A., and Su¨ ltmann, H. (2009).
Effects of infiltrating lymphocytes and estrogen receptor on gene expression
and prognosis in breast cancer. Breast Cancer Res. Treat. 116, 69–77.
Campbell, J.D., Yau, C., Bowlby, R., Liu, Y., Brennan, K., Fan, H., Taylor, A.M.,
Wang, C., Walter, V., Akbani, E., et al. (2018). Genomic, pathway network, and
immunologic features distinguishing squamous carcinomas. Cell Rep. 23
https://doi.org/10.1016/j.celrep.2018.03.063.
Cancer Genome Atlas Network (2012). Comprehensive molecular portraits of
human breast tumours. Nature 490, 61–70.
Cancer Genome Atlas Network (2015). Genomic classification of cutaneous
melanoma. Cell 161, 1681–1696.
Cancer Genome Atlas Research Network (2011). Integrated genomic analyses
of ovarian carcinoma. Nature 474, 609–615.
Cancer Genome Atlas Research Network, Kandoth, C., Schultz, N., Cherniack,
A.D., Akbani, R., Liu, Y., Shen, H., Robertson, A.G., Pashtan, I., Shen, R., Benz,
C.C., et al. (2013a). Integrated genomic characterization of endometrial
carcinoma. Nature 497, 67–73.
Cancer Genome Atlas Research Network, Ley, T.J., Miller, C., Ding, L.,
Raphael, B.J., Mungall, A.J., Robertson, A., Hoadley, K., Triche, T.J., Jr., Laird,
P.W., Baty, J.D., et al. (2013b). Genomic and epigenomic landscapes of adult
de novo acute myeloid leukemia. N. Engl. J. Med. 368, 2059–2074.
Cancer Genome Atlas Research Network (2014a). Comprehensive molecular
profiling of lung adenocarcinoma. Nature 511, 543–550.
Cancer Genome Atlas Research Network (2014b). Integrated genomic characterization
of papillary thyroid carcinoma. Cell 159, 676–690.
Cancer Genome Atlas Research Network (2015a). The molecular taxonomy of
primary prostate cancer. Cell 163, 1011–1025.
Cancer Genome Atlas Research Network, Brat, D.J., Verhaak, R.G., Aldape,
K.D., Yung, W.K., Salama, S.R., Cooper, L.A., Rheinbay, E., Miller, C.R.,
Vitucci, M., Morozova, O., et al. (2015b). Comprehensive, integrative genomic
analysis of diffuse lower-grade gliomas. N. Engl. J. Med. 372, 2481–2498.
Cancer Genome Atlas Research Network (2017). Comprehensive and integrative
genomic characterization of hepatocellular carcinoma. Cell 169, 1327–
1341.e23.
Carter, S.L., Cibulskis, K., Helman, E., McKenna, A., Shen, H., Zack, T., Laird,
P.W., Onofrio, R.C., Winckler, W., Weir, B.A., et al. (2012). Absolute quantification
of somatic DNA alterations in human cancer. Nat. Biotechnol. 30,
413–421.
Chang, H.Y., Sneddon, J.B., Alizadeh, A.A., Sood, R., West, R.B., Montgomery,
K., Chi, J.T., van de Rijn, M., Botstein, D., and Brown, P.O. (2004). Gene
expression signature of fibroblast serum response predicts human cancer
progression: similarities between tumors and wounds. PLoS Biol. 2, E7.
Cherniack, A.D., Shen, H., Walter, V., Stewart, C., Murray, B.A., Bowlby, R.,
Hu, X., Ling, S., Soslow, R.A., Broaddus, R.R., et al.; Cancer Genome Atlas
Research Network (2017). Integrated molecular characterization of uterine
carcinosarcoma. Cancer Cell 31, 411–423.
Chu, J., Sadeghi, S., Raymond, A., Jackman, S.D., Nip, K.M., Mar, R.,
Mohamadi, H., Butterfield, Y.S., Robertson, A.G., and Birol, I. (2014).
BioBloom tools: fast, accurate and memory-efficient host species sequence
screening using bloom filters. Bioinformatics 30, 3402–3404.
Covington, K., Shinbrot, E., and Wheeler, D.A. (2016). Mutation signatures
reveal biological processes in human cancer. bioRxiv, https://doi.org/10.
1101/036541.
Davidson, G.S., Wylie, B.N., and Boyack, K.W. (2001). Cluster stability and the
use of noise in interpretation of clustering. In IEEE Information Visualization
2001, INFOVIS 2001. (IEEE).
Davis, C.F., Ricketts, C.J., Wang, M., Yang, L., Cherniack, A.D., Shen, H.,
Buhay, C., Kang, H., Kim, S.C., Fahey, C.C., et al.; The Cancer Genome Atlas
Research Network (2014). The somatic genomic landscape of chromophobe
renal cell carcinoma. Cancer Cell 26, 319–330.
Hainsworth, J.D., Rubin, M.S., Spigel, D.R., Boccia, R.V., Raby, S., Quinn, R.,
and Greco, F.A. (2013). Molecular gene expression profiling to predict the
tissue of origin and direct site-specific therapy in patients with carcinoma
of unknown primary site: a prospective trial of the Sarah Cannon research
institute. J. Clin. Oncol. 31, 217–223.
Hoadley, K.A., Yau, C., Wolf, D.M., Cherniack, A.D., Tamborero, D., Ng, S.,
Leiserson, M.D.M., Niu, B., McLellan, M.D., Uzunangelov, V., et al.; Cancer
Genome Atlas Research Network (2014). Multiplatform analysis of 12 cancer
types reveals molecular classification within and across tissues of origin.
Cell 158, 929–944.
Jin, X., Jin, X., and Kim, H. (2017). Cancer stem cells and differentiation
therapy. Tumour Biol. 39, 1010428317729933.
Knijnenburg, T., Wang, L., Zimmermann, M., Chambwe, N., Gao, G.,
Cherniack, A., Fan, H., Shen, H., Way, G., Greene, C., et al. (2018). Genomic
and Molecular Landscape of DNA Damage Repair Deficiency Across The
Cancer Genome Atlas. Cell Rep. 23 https://doi.org/10.1016/j.celrep.2018.
03.076.
Korn, J.M., Kuruvilla, F.G., McCarroll, S.A., Wysoker, A., Nemesh, J., Cawley,
S., Hubbell, E., Veitch, J., Collins, P.J., Darvishi, K., et al. (2008). Integrated
genotype calling and association analysis of SNPs, common copy number
polymorphisms and rare CNVs. Nat. Genet. 40, 1253–1260.
Langfelder, P., and Horvath, S. (2008). WGCNA: an R package for weighted
correlation network analysis. BMC Bioinformatics 9, 559.
Lawrence, M.S., Stojanov, P., Polak, P., Kryukov, G.V., Cibulskis, K.,
Sivachenko, A., Carter, S.L., Stewart, C., Mermel, C.H., Roberts, S.A., et al.
(2013). Mutational heterogeneity in cancer and the search for new cancerassociated
genes. Nature 499, 214–218.
Liu, Y., Sethi, N.S., Hinoue, T., Schneider, B.G., Cherniack, A.D., SanchezVega,
F., Seoane, J.A., Farshidfar, F., Bowlby, R., Islam, M., et al. (2018).
Comparative molecular analysis of gastrointestinal adenocarcinomas. Cancer
Cell 33. https://doi.org/10.1016/j.ccell.2018.03.010.
Malta, T.M., Sokolov, A., Gentles, A.J., Burzykowski, T., Poisson, L.,
Weinstein, J.N., Kaminska, B., Huelsken, J., Omberg, L., Gevaert, O., et al.
(2018). Comprehensive analysis of cancer stemness. Cell 173.
McCarroll, S.A., Kuruvilla, F.G., Korn, J.M., Cawley, S., Nemesh, J., Wysoker,
A., Shapero, M.H., de Bakker, P.I., Maller, J.B., Kirby, A., et al. (2008).
Integrated detection and population-genetic analysis of SNPs and copy
number variation. Nat. Genet. 40, 1166–1174.
Mermel, C.H., Schumacher, S.E., Hill, B., Meyerson, M.L., Beroukhim, R., and
Getz, G. (2011). GISTIC2.0 facilitates sensitive and confident localization of the
targets of focal somatic copy-number alteration in human cancers. Genome
Biol. 12, R41.
Mo, Q., Wang, S., Seshan, V.E., Olshen, A.B., Schultz, N., Sander, C., Powers,
R.S., Ladanyi, M., and Shen, R. (2013). Pattern discovery and cancer gene
identification in integrated cancer genomic data. Proc. Natl. Acad. Sci. USA
110, 4245–4250.
Moran, S., Martı´nez-Cardu´ s, A., Sayols, S., Musule´ n, E., Balan˜ a´ , C., EstivalGonzalez,
A., Moutinho, C., Heyn, H., Diaz-Lagares, A., de Moura, M.C.,
et al. (2016). Epigenetic profiling to classify cancer of unknown primary:
a multicentre, retrospective analysis. Lancet Oncol. 17, 1386–1395.
Newton, Y., Novak, A.M., Swatloski, T., McColl, D.C., Chopra, S., Graim, K.,
Weinstein, A.S., Baertsch, R., Salama, S.R., Ellrott, K., et al. (2017).
TumorMap: exploring the molecular similarities of cancer samples in an
interactive portal. Cancer Res. 77, e111–e114.
Olshen, A.B., Venkatraman, E.S., Lucito, R., and Wigler, M. (2004). Circular
binary segmentation for the analysis of array-based DNA copy number data.
Biostatistics 5, 557–572.
Ramos, A.H., Lichtenstein, L., Gupta, M., Lawrence, M.S., Pugh, T.J.,
Saksena, G., Meyerson, M., and Getz, G. (2015). Oncotator: cancer variant
annotation tool. Hum. Mutat. 36, E2423–E2429.
Ricketts, C.J., De Cubas, A.A., Fan, H., Smith, C.C., Lang, M., Reznik, E.,
Bowlby, R., Gibb, E.A., Akbani, R., Beroukhim, R., et al. (2018). The Cancer
Cell 173, 291–304, April 5, 2018 303
Genome Atlas Comprehensive Molecular Characterization of Renal Cell
Carcinoma. Cell Rep. 23 https://doi.org/10.1016/j.celrep.2018.03.075.
Robertson, A.G., Shih, J., Yau, C., Gibb, E.A., Oba, J., Mungall, K.L., Hess,
J.M., Uzunangelov, V., Walter, V., Danilova, L., et al. (2017). Integrative analysis
identifies four molecular and clinical subsets in uveal melanoma. Cancer Cell
32, 204–220.e15.
Saunders, C.T., Wong, W.S., Swamy, S., Becq, J., Murray, L.J., and
Cheetham, R.K. (2012). Strelka: accurate somatic small-variant calling from
sequenced tumor-normal sample pairs. Bioinformatics 28, 1811–1817.
Scrucca, L., Fop, M., Murphy, T.B., and Raftery, A.E. (2016). mclust 5:
clustering, classification and density estimation using gaussian finite mixture
models. R J. 8, 289–317.
Shen, R., Olshen, A.B., and Ladanyi, M. (2009). Integrative clustering of multiple
genomic data types using a joint latent variable model with application to
breast and lung cancer subtype analysis. Bioinformatics 25, 2906–2912.
Shen, R., Mo, Q., Schultz, N., Seshan, V.E., Olshen, A.B., Huse, J., Ladanyi,
M., and Sander, C. (2012). Integrative subtype discovery in glioblastoma using
iCluster. PLoS ONE 7, e35236.
Taylor, A.M., Shih, J., Ha, G., Gao, G.F., Zhang, X., Berger, A.C., Schumacher,
S.E., Wang, C., Hu, H., Liu, J., et al. (2018). Genomic and functional approachesto
understanding cancer aneuploidy. Cancer Cell 33. https://doi.org/10.
1016/j.ccell.2018.03.007.
Teschendorff, A.E., Gomez, S., Arenas, A., El-Ashry, D., Schmidt, M.,
Gehrmann, M., and Caldas, C. (2010). Improved prognostic classification of
breast cancer defined by antagonistic activation patterns of immune response
pathway modules. BMC Cancer 10, 604.
Thorsson, V., Gibbs, D.L., Brown, S.D., Wolf, D., Bortone, D.S., Yang, T.-H.O.,
Porta-Pardo, E., Gao, G., Plaisier, C.L., Eddy, J.A., et al. (2018). The immune
landscape of cancer. Immunity 48. https://doi.org/10.1016/j.immuni.2018.
03.023.
Vaske, C.J., Benz, S.C., Sanborn, J.Z., Earl, D., Szeto, C., Zhu, J., Haussler, D.,
and Stuart, J.M. (2010). Inference of patient-specific pathway activities from
multi-dimensional cancer genomics data using PARADIGM. Bioinformatics
26, i237–i245.
Wilkerson, M.D., and Hayes, D.N. (2010). ConsensusClusterPlus: a class
discovery tool with confidence assessments and item tracking. Bioinformatics
26, 1572–1573.
Wolf, D.M., Lenburg, M.E., Yau, C., Boudreau, A., and van ’t Veer, L.J. (2014).
Gene co-expression modules as clinically relevant hallmarks of breast cancer
diversity. PLoS ONE 9, e88309.
Zheng, S., Cherniack, A.D., Dewal, N., Moffitt, R.A., Danilova, L., Murray, B.A.,
Lerario, A.M., Else, T., Knijnenburg, T.A., Ciriello, G., et al.; Cancer Genome
Atlas Research Network (2016). Comprehensive pan-genomic characterization
of adrenocortical carcinoma. Cancer Cell 29, 723–736.
304 Cell 173, 291–304, April 5, 2018
STAR+METHODS
KEY RESOURCES TABLE
CONTACT FOR REAGENT AND RESOURCE SHARING
Further information and requests for resources and reagents should be directed to and will be fulfilled by the Lead Contact,
Peter W. Laird (Peter.Laird@vai.org). Sequence data hosted at the GDC is under controlled access. Details for gaining access can
be found at (https://gdc.cancer.gov/access-data/data-access-processes-and-tools).
REAGENT or RESOURCE SOURCE IDENTIFIER
Antibodies
RPPA antibodies RPPA Core Facility,
MD Anderson Cancer Center
https://www.mdanderson.org/research/research-resources/
core-facilities/functional-proteomics-rppa-core.html
Biological Samples
Tumor and normal tissue
and blood samples
TCGA Network https://portal.gdc.cancer.gov/legacy-archive/
Critical Commercial Assays
DNA/RNA AllPrep kit QIAGEN Cat# 80204
mirVana miRNA Isolation kit Ambion Cat# AM1560
QiaAmp blood midi kit QIAGEN Cat# 51185
AmpFISTR Identifiler kit Applied Biosystems Cat# A30737
RNA6000 nano Assay Agilent Cat# 5067-1511
Genome-Wide Human SNP Array 6.0 Affymetrix Cat# 901150
HumanMethylation450 Infinium Cat# WG-314-1002
HumanMethylation27 Infinium Cat# WG-311-2201
mRNA TruSeq kit Illumina Cat# RS-122-2001
Deposited Data
Raw genomic and clinical data NCI Genomic Data Commons https://portal.gdc.cancer.gov/legacy-archive/
MC3 mutation annotation file NCI Genomic Data Commons https://gdc.cancer.gov/about-data/publications/mc3-2017
Processed data files NCI Genomic Data Commons https://gdc.cancer.gov/about-data/publications/pancanatlas
Software and Algorithms
Copy number estimation Broad Institute http://archive.broadinstitute.org/cancer/
cga/copynumber_pipeline
Significant focal copy number
change – GISTIC 2.0
Mermel et al., 2011 http://software.broadinstitute.org/software/cprg/?q=node/31
Purity, ploidy, genome
doubling - ABSOLUTE
Carter et al., 2012 http://archive.broadinstitute.org/cancer/cga/absolute
Cluster analysis - ConsensusClusterPlus Wilkerson and Hayes, 2010 http://bioconductor.org/packages/release/bioc/
html/ConsensusClusterPlus.html
Integrative clustering of multiple
genomic data types (iCluster)
Shen et al., 2009 https://www.mskcc.org/sites/www.mskcc.org/files/
node/4281/documents/icluster-1.2.0.tar.gz
PARADIGM Vaske et al., 2010 http://sbenz.github.io/Paradigm/
TumorMap Newton et al., 2017 https://tumormap.ucsc.edu/
Mclust R package Scrucca et al., 2016 https://cran.r-project.org/web/packages/mclust/index.html
pheatmap v1.0.2 N/A https://www.rdocumentation.org/packages/
pheatmap/versions/1.0.2
Mbatch (EB++) MD Anderson Cancer Center http://bioinformatics.mdanderson.org/main/
TCGABatchEffects:Overview
DrL Alencar and Polley, 2011 http://wiki.cns.iu.edu/pages/viewpage.action?pageId=1704113
WGCNA Langfelder and Horvath, 2008 https://labs.genetics.ucla.edu/horvath/htdocs/
CoexpressionNetwork/Rpackages/WGCNA/
Cell 173, 291–304.e1–e6, April 5, 2018 e1
EXPERIMENTAL MODEL AND SUBJECT DETAILS
Human Subjects
Tumor tissue, adjacent normal tissue, and normal whole blood samples were obtained from patients at contributing centers with
informed consent according to their local Institutional Review Boards (IRBs, see below). Biospecimens were centrally processed
and DNA, RNA, and protein were distributed to TCGA analysis centers.
TCGA Project Management has collected necessary human subjects documentation to ensure the project complies with 45-CFR-46
(the ‘‘Common Rule’’). The program has obtained documentation from every contributing clinical site to verify that IRB approval has
been obtained to participate in TCGA. Such documented approval may include one or more of the following:
d An IRB-approved protocol with Informed Consent specific to TCGA or a substantially similar program. In the latter case, if the
protocol was not TCGA-specific, the clinical site PI provided a further finding from the IRB that the already-approved protocol is
sufficient to participate in TCGA.
d A TCGA-specific IRB waiver has been granted.
d A TCGA-specific letter that the IRB considers one of the exemptions in 45-CFR-46 applicable. The two most common
exemptions cited were that the research falls under 46.102(f)(2) or 46.101(b)(4). Both exempt requirements for informed
consent, because the received data and material do not contain directly identifiable private information.
d A TCGA-specific letter that the IRB does not consider the use of these data and materials to be human subjects research. This
was most common for collections in which the donors were deceased.
A total of 11,188 patients were analyzed in TCGA with at least one molecular-profiling platform. This study contained both males
and females with inclusions of genders dependent on tumor types. There were 5,769 females, 5,282 males and 137 missing information
about gender. TCGA’s goal was to characterize adult human tumors; therefore, the vast majority are over the age of 18. However,
there are 20 samples that are under the age of 18 that had tissue submitted prior to clinical data. Age was missing for 188 patients.
The range of ages was 10 – 90 (maxed 90 for protection of human subjects) with a median age of diagnosis of 60 years of age.
METHOD DETAILS
Sample Processing
RNA and DNA were extracted from tumor and adjacent normal tissue specimens using a modification of the DNA/RNA AllPrep kit
(QIAGEN). The flow-through from the QIAGEN DNA column was processed using a mirVana miRNA Isolation Kit (Ambion). This latter
step generated RNA preparations that included RNA < 200 nt suitable for miRNA analysis. DNA was extracted from blood using the
QiaAmp Blood Midi Kit (QIAGEN). Each specimen was quantified by measuring Abs260 with a UV spectrophotometer or by
PicoGreen assay. DNA specimens were resolved by 1% agarose gel electrophoresis to confirm high molecular weight fragments.
A custom Sequenom SNP panel or the AmpFISTR Identifiler (Applied Biosystems) was utilized to verify that tumor DNA and germline
DNA were derived from the same patient. Five hundred nanograms of each tumor and normal DNA were sent to QIAGEN for REPLI-g
whole genome amplification using a 100 mg reaction scale. Only specimens yielding a minimum of 6.9 mg of tumor DNA, 5.15 mg RNA,
and 4.9 mg of germline DNA were included in this study. RNA was analyzed via the RNA6000 Nano assay (Agilent) for determination of
an RNA Integrity Number (RIN), and only the cases with RIN > 7.0 were included in this study.
Pathology Review
Samples were systematically evaluated by pathologists to confirm the histopathologic diagnosis and any variant histology, using the
criteria of the most recent edition of the WHO / IARC Classification of Tumors relevant to each cancer type. All tumor samples were
assessed for tumor content (percent tumor nuclei). Any non-concordant diagnoses among the pathologists were re-reviewed and
resolution achieved after discussion.
Somatic Copy-Number Alterations
Somatic copy-number data were generated on Affymetrix SNP 6.0 arrays using standard protocols from the Genome Analysis
Platform of the Broad Institute (McCarroll et al., 2008). Briefly, preliminary copy number at each probe locus was inferred by Birdseed
analysis of raw .CEL files (Korn et al., 2008). Tangent normalization was then used to further refine genome-wide copy-number
estimates (https://www.broadinstitute.org/cancer/cga/copynumber_pipeline). Segmented copy-number data were generated using
Circular Binary Segmentation (Olshen et al., 2004). Regions corresponding to germline copy-number alterations were removed by
applying filters generated from normal samples. Gene-level copy number was generated by GISTIC 2.0 analysis (Mermel et al.,
2011). Purity and ploidy estimates were calculated using ABSOLUTE (Carter et al., 2012).
Chromosome arm-level copy-number calls were determined by clustering breakpoint locations and fraction of arm altered (further
detailed in Taylor et al., 2018). Hierarchical clustering was performed using a metric of Manhattan distance and Ward2 methods for
10,522 samples; this analysis identified 10 groups (Figure 1A). Aneuploidy scores reflect the overall aneuploidy burden, and the range
varied across tumor types. Most AN groups represented a mix of tumor types; however, tumor types with specific aneuploidy
e2 Cell 173, 291–304.e1–e6, April 5, 2018
patterns defined unique groups like AN9 enriched with GBM, characterized by chr7 gain and chr10 loss, and AN10 enriched for
TGCT, which all displayed chromosome ploidies greater than 2.
Cervical squamous tumors clustered in high aneuploidy clusters AN1 and AN5. These clusters were also enriched for other
Pan-gyn tumors, including ovarian, high-copy number endometrial, and uterine carcinosarcoma (Cherniack et al., 2017).
Gynecologic tumors with fewer copy-number alterations including Luminal breast cancers and other endometrial tumors grouped
separately in low aneuploidy clusters AN7 and AN8, respectively.
DNA methylation
Illumina Infinium DNA methylation arrays were used to obtain DNA methylation profiles of 10,814 tumors from 33 tumor types and
1,064 histologically normal tumor-adjacent tissue specimens representing 24 different tissue types. Data from two generations of
Infinium arrays, HumanMethylation27 (HM27) and HumanMethylation450 (HM450), were merged to generate a dataset for 22,601
probes shared between two platforms. To minimize systematic platform-specific effects, we normalized the HM27 data against
the HM450 data using a probe-by-probe proportional rescaling method. During data generation, a single technical replicate of the
same cell line control sample from either of two different DNA extractions (TCGA-07-0227/TCGA-AV-A03D) was included on each
plate as a control, and measured 44/198 times and 12/169 times on HM27 and HM450, respectively. These repeated-measurements
were therefore used for rescaling of the HM27 data to be comparable to HM450. For each probe within each platform, we computed
the median b-value across all technical replicates of each of the two TCGA IDs. We then combined the two extractions by taking the
mean of the two medians obtained for each of the two replicate TCGA IDs, and obtained a single summarized DNA methylation
readout (b-value) for the corresponding probe i for each platform, noted as Betahm27,i, and Betahm450,i, respectively. We then applied
a constrained (within the range of 0 to 1 for b-values) linear rescaling of the HM27 data for each probe and for each patient’s sample
using Betahm27,i and Betahm450,i. When the HM27 b-value of a patient’s sample j for probe i was smaller than the mean of median
replicate samples on the HM27 for that probe, we linearly rescaled the HM27 b-value Betahm27,i,j in the (0, Betahm27,i,j) space; and
when Betahm27,i,j was greater, we linearly rescaled the HM27 beta value Betahm27,i,j in the (Betahm27,i,j, 1) space; This translates
into the following mathematical computation: Beta hm450,i,j = Betahm27,i,j*(Betahm450,i/Betahm27,i), if Betahm27,i,j < Betahm27,i ; and
Beta hm450,i,j = 1-(1- Betahm27,i,j)*((1- Betahm450,i)/(1- Betahm27,i)), if Betahm27,i,j > Betahm27,i. After the between-platform normalization,
we further excluded 779 probes that still showed a consistent platform difference (mean b-value difference greater than or equal
to 0.1) in six or more tumor types.
Unsupervised clustering was performed based on promoter CpG sites that did not exhibit tissue-specific DNA methylation, but that
acquired hypermethylation in cancer. We used DNA methylation data from the histologically normal tissues and leukocytes to identify
11,275 sites that lacked tissue-specific DNA methylation (mean b-value < 0.2 in any tissue type and b-value > 0.3 in no more than five
samples across the entire set). To minimize the influence of variable tumor purity levels on a clustering result, we dichotomized the
data using a b-value of R 0.3 to define positive DNA methylation and < 0.3 to specify lack of methylation. The dichotomization not
only ameliorated the effect of tumor sample purity on the clustering, but also removed a great portion of residual batch/platform
effects that are mostly reflected in small variations near the two ends of the range of b-values. For clustering analysis of tumors,
we selected 3,139 CpG sites that were methylated at a b-value of R 0.3 in more than 10% of tumors within any of the 33 cancer types.
We performed unsupervised clustering of 10,814 tumors using hierarchical clustering with Ward’s method to cluster the distance
matrix computed with the Jaccard index. The dendrogram was cut at different levels, and resulting clusters were evaluated for
associations with tumor types and subtypes. The heatmap was generated using the original b-values for the top one-third
(n = 1,035) of the most variability methylated CpGs across tumors (Figure 1B). We chose 25 clusters for the subsequent crossplatform
analyses. We noted that a fraction of ESCA and STAD was found in METH9 with LUAD and PAAD, a result that may be
related to the low tumor cellularity of the cancers in this cluster. Three types of renal cell carcinomas, including KIRC, (KIRP and
KICH, aligned together in METH19, which interestingly also included THYM and THCA. Pan-GYN tumors separated into three major
groups, which appeared to reflect molecular subtypes within each tumor type. Luminal and HER2 breast (BRCA-Luminal) and
subtypes of UCEC lacking CIN organized into METH 4, 5 and 6. OV and UCEC with CIN-high grouped together in METH 22 and
23. Finally, Basal-like BRCA was found in METH 24 and 25.
RNA Data Batch Correction
The expression data for mRNA and miRNA were batch-corrected to adjust for platform differences between the GAII and HiSeq
Illumina sequencers. For mRNA, additional adjustments were made for different sequencing centers (The University of North
Carolina [UNC] and British Columbia Cancer Agency [BCCA]) and a plate effect observed in PRAD. For the mRNA data, first batch
312 and 320 PRAD were adjusted to remove batch effects. UNC GA samples (UCEC, COAD, READ) were adjusted to the UNC
HiSeq data. Genes with mostly zero reads or with residual batch effects ($10% of genes) were removed from the adjusted
samples and replaced with NAs. A similar adjustment was made for BCCA GAII-sequenced samples (LAML, STAD, ESCA) to
HiSeq. Genes were adjusted using a novel algorithm called EB++; a variant of the Empirical Bayes / ComBat algorithm with
training and testing features added.
The miRNA data were batch-corrected for GAII and HiSeq, as well as for two library construction protocols (MultiMACS and Direct).
Weakly expressed miRNAs were filtered by requiring miRNA mature strands to be expressed with an RPM of at least 10 in 10% of
primary tumors in each TCGA project resulting in 743 miRNAs across all 32 projects (miRNA sequencing was not performed on GBM).
Cell 173, 291–304.e1–e6, April 5, 2018 e3
The EB++ method was used to correct the Direct protocol to the MultiMACs protocol and the GAII to the HiSeq protocol similar to
what was done for mRNA.
mRNA
Upper quartile normalized RSEM data for batch-corrected mRNA gene expression were used for analysis. The matrix was filtered for
genes expressed in 60% or more of the samples. Unsupervised consensus clustering using Consensus Cluster Plus (Wilkerson and
Hayes, 2010) was performed on 10,165 tumors with 15,363 genes. At K = 43, we identified 25 major groups with at least 40 samples
per group (Figure 1C). Many of the sample groups contained > 90% of a single tumor type or subtype. These included OV, PRAD,
THCA, BRCA-Luminal, BRCA-Basal, LUAD, BLCA, CESC, UCEC, MESO, and TGCT. As observed in our previous publication
(Hoadley et al., 2014), Basal-like breast cancer split out as a separate group from the estrogen receptor (ER)-positive and HER2positive
breast cancers.
miRNA
We analyzed batch-corrected, normalized abundance (i.e., reads per million, RPM) data for 743 expressed mature strands (of 1212
miRBase v16 strands). The data matrix contained abundance profiles for 10,170 tumor samples. We hierarchically clustered the
data matrix with the pheatmap R package, using row-scaling, Pearson correlation coefficients for a distance metric, and ward.D2
clustering.
Unsupervised hierarchical clustering of batch-corrected miRNA mature-strand expression profiles from 10,170 tumors yielded a
15-group solution (Figure 1D). We observed six tumor-type-specific clusters. MIR5 contained OV, MIR8 BRCA, MIR12 LGG, MIR13
LIHC, MIR14 THCA, and MIR15 PRAD. Two clusters contained samples from two diseases. MIR7 contained two blood cancers:
DLBC and LAML, while MIR10 contained two types of melanomas: SKCM and UVM. MIR11 contained only the three kidney tumors:
KICH, KIRC and KIRP.
Each of the remaining 6 clusters contained at least four cancer types. MIR1 was largely UCEC, with substantial BRCA and BLCA,
plus smaller numbers of 6 other cancers. MIR2 contained predominantly squamous carcinomas including HNSC, LUSC, CESC and
BLCA, with smaller numbers of ESCA, LUAD, and minor BRCA and SARC. MIR3 contained largely PCPG, with SARC and ACC, and
smaller numbers of 8 other cancer types. MIR6, the Pan-GI group, was largely COAD and STAD, but also had substantial PAAD,
READ and ESCA, with smaller numbers of CHOL and LIHC. MIR4 was largely TGCT, with THYM and BLCA, with smaller numbers
of LIHC and SKCM. MIR9 was largely LUAD and SARC, with smaller numbers of MESO and LUSC.
Protein
Protein expression data were available for 7,858 samples from 32 of the 33 tumor types (LAML data were never generated) across
216 proteins and phosphoproteins. The data were generated using the reverse phase protein array (RPPA) platform. We used batch
effects-corrected RPPA data and median-centered them in both directions. We then clustered them using hierarchical clustering
from the R function hclust() with 1 – Pearson’s correlation coefficient as the distance metric and Ward as the linkage function. The
10 clusters were obtained by cutting the dendrogram using the cutree() function in R.
Hierarchical clustering of protein expression data revealed 10 distinct Protein (P) groups (Figure 1E). The dendrogram first
separated P1 and P2 from the remaining 8 clusters, which largely corresponded with the separation between mesenchymal-like
tumor types with high EMT signatures versus tumor types with low EMT signatures, respectively. Cluster 1 consisted of the brain
cancers (GBM, LGG), whereas cluster 2 contained DLBC, SARC, PCPG, UCS, THYM and metastatic SKCM. Those 2 clusters
were characterized by low levels of E-cadherin, EPPK1, RAB25 and Claudin 7. The brain cancers had high levels of PKC-alpha,
phosphoPKC-alpha, PKC-delta, ERK2, PEA15 and acetyl-A tubulin.
P3 and P4 consisted mainly of the Luminal breast and gynecologic cancers (BRCA-Lum8, UCEC7, OV), plus some liver samples
(LIHC). The clusters had high levels of ER-alpha, AR and IGFBP2. Interestingly, the LIHC samples in P4 had high levels of ER-alpha as
well, whereas those LIHC samples not in P4 had low ER-alpha levels. P6 was a Pan-kidney cluster with KIRC, KIRP and ACC and was
characterized by high levels of EMT based on low expression of the negative EMT markers E-cadherin, RAB25 and Claudin 7, as well
as low IGFBP2, FASN and Cyclin B1, and high GAPDH, CD26, and phosphoNDRG1. P8 was a Pan-GI cluster consisting of most of
the colorectal (COAD/READ) and gastric cancer (STAD) samples. In contrast to the Pan-kidney group, the Pan-GI group had a very
low EMT signature with high expression of RAB25, EPPK1 and Claudin 7. Other distinguishing features of the cluster included high
levels of cleaved CASPASE 7, TFRC, MYH11, TIGAR, and beta catenin. P9 and P10 were the most diverse and included some
samples from most of the tumor types. P10, in particular, had an enrichment of the squamous cancers with large proportions of
HNSC, LUSC, CESC, CHOL, and BLCA. This cluster had high levels of PAI1, cleaved CASPASE 7, ANNEXIN1, TFRC, P16INK4A,
ASNS, Cyclin B1, Cyclin E1, FASN and FOXM1.
Integrative clustering with iCluster
The iCluster clustering algorithm formulates the problem of subgroup discovery as a joint multivariate regression of multiple data
types with reference to a set of common latent variables, which represent the underlying 28 tumor subtypes (Mo et al., 2013;
Shen et al., 2009, 2012). Four molecular platforms - SCNA, DNA methylation, mRNA expression, and miRNA expression were
used as input. Data were pre-processed using the following procedures: For mRNA, and mature-strand miRNA sequence
e4 Cell 173, 291–304.e1–e6, April 5, 2018
data, poorly expressed genes were excluded based on median-normalized counts, and variance filtering led to a list of reduced
features for clustering. mRNA and miRNA expression features were log2 transformed, normalized and scaled before using them
as an input to iCluster. Pre-processing led to 3,217 mRNA and 382 miRNA features. Pre-processed DNA methylation data were
obtained from the methylation merged HM27 and HM450 platform datasets and included 3,139 hypermethylation features.
Circular Binary Segmented (CBS) SCNA data were further reduced to a set of 3,105 non-redundant regions as described
(Mo et al., 2013).
Cancer Immune Subtypes
To characterize the commonality and diversity of intratumoral immune states, we scored 160 published immune expression
signatures on all available TCGA PanCancerAtlas tumor samples, and performed cluster analysis to identify similarity modules
of multiple immune signature sets. The 160 immune expression signatures were selected based on extensive literature search,
utilizing diverse resources considered to be reliable and comprehensive, based on expert opinions of immuno-oncologists
(Thorsson et al., 2018). Eighty-three signatures were derived in the context of immune response studies in cancer, and the
remaining 77 are of general validity for immunity. TCGA RNA-seq values from the PanCancer Atlas normalized gene expression
matrix were scored for each of the 160 identified gene expression signatures using single-sample gene set enrichment (ssGSEA)
analysis, using the R package GSVA. Clusters of similar signature scores were identified by weighted gene correlation network
analysis (WGCNA) (Langfelder and Horvath, 2008). Based on the WGCNA analysis, five immuno-oncology-related immune
expression signatures: activation of macrophages/monocytes (Beck et al., 2009), overall lymphocyte infiltration (dominated by
T and B cells) (Calabro` et al., 2009), TGF-b response (Teschendorff et al., 2010), IFN-g response (Wolf et al., 2014), and wound
healing (Chang et al., 2004), robustly reproduced co-clustering of the immune signature sets, and were selected to perform cluster
analysis of all cancer types, with the exception of hematologic neoplasias (acute myeloid leukemia, LAML; diffuse large B cell
lymphoma, DLBC; and thymoma, THYM). Clustering of tumor samples scored on these five signatures was performed using
model-based clustering, using the mclust R package (Scrucca et al., 2016), with the number of clusters, K, determined by maximization
of Bayesian Information Criterion (BIC). Maximal BIC was found with a six-cluster solution, and the six resulting clusters
C1-C6 (with 2416, 2591, 2397, 1157, 385 and 180 cases, respectively) were characterized by a distinct distribution of scores over
the five representative signatures, and effectively categorized each TCGA sample as belonging to one of six cancer ‘‘immune
subtypes,’’ namely Wound Healing (C1), IFN-g Dominant (C2), Inflammatory (C3), Lymphocyte Depleted (C4), Immunologically
Quiet (C5), or TGF-b Dominant (C6). Additional details are found in Thorsson et al. (2018). The designations C1-C6 of immune
subtypes were made independently from iCluster designations in the current work.
Leukocyte and Stromal Fraction Estimates
Overall leukocyte content in 10,814 TCGA tumor aliquots was assessed by identifying DNA methylation probes with the greatest
differences between pure leukocyte cells and normal tissue, then estimating leukocyte content using a mixture model. From Illumina
Infinium DNA methylation platform arrays HumanMethylation450, 2000 loci were identified (200 for HumanMethylation27) that
were the most differentially methylated between leukocyte and normal tissues, 1000 in each direction. For each locus i, assuming
two populations (j), for each sample we have the following equation:
bi =
X2
j = 1
bijpj:
Using the tumor with the least evidence of leukocyte methylation as a surrogate for the beta value (b) for each locus in the pure tumor,
2000 estimates were made, solving for p. We took the mode of 200 estimates to avoid loci that violate the assumptions. Using the
estimated p and the measured b for tumor and leukocyte, with the same linear model, we solved for b (deconvoluted value) extracting
the leukocyte fraction (LF).
Stromal fraction (SF) was defined as the total non-tumor cellular component, obtained by subtracting tumor purity from unity.
Tumor purity was generated using ABSOLUTE (Carter et al., 2012) as detailed in Taylor et al., 2018.
TumorMap
We used the latent iCluster space (Table S7) to calculate Euclidean similarity between every pair of samples, where Euclidean
similarity = (1 / (1 + Euclidean_distance)) (https://tumormap.ucsc.edu/). The distances were used as input to generate a 2D layout
of the samples using the physics-based Distributed Recursive (Graph) Layout method (Alencar and Polley, 2011), previously known
as VxOrd (Davidson et al., 2001). DrL layout engine was used with each sample’s 28 most similar neighbors. DrL’s default settings
were used for ‘‘edge cutting’’ and ‘‘intermediate output interval’’ parameters, 0.8 and 0, respectively. Sample lists for attributes
(GI, gyn, kidney, stemness, squamous) were obtained from other working groups.
PARADIGM
The PARADIGM algorithm with the interaction-learning update (Chu et al., 2014; Vaske et al., 2010) was used to infer protein
activities in the context of gene regulatory pathways, based on gene expression and copy-number data. The method uses a
Cell 173, 291–304.e1–e6, April 5, 2018 e5
set of interactions from several sources (NCI-PID, Reactome, and KEGG) and superimposes them into a single network
(SuperPathway). The SuperPathway contained 7,369 proteins, 9,354 multi-protein complexes, 2,092 families, and 592 cellular
processes connected by 45,315 interactions. The PARADIGM algorithm was applied to 9,829 tumors with platformcorrected
expression data and gene-level copy-number alteration data from 33 cancer types to infer the integrated pathway levels
(IPLs) of the 19,504 SuperPathway features.
Pathway features characterizing each iCluster were identified by comparing each iCluster versus all others using the t test and
Wilcoxon Rank sum test with Benjamini-Hochberg (BH) false discovery rate (FDR) correction. An initial minimum variation filter
(at least 1 sample with absolute activity > 0.05) was applied; and the 15,502 features passing the minimum variation feature were
considered in this analysis. Features deemed significant (FDR corrected p < 0.05) by both tests and showing an absolute difference
in group means > 0.05 were selected. The selected pathway features were assessed for interconnectivity; regulatory nodes with
differential inferred IPLs that also had at least 15 differential downstream regulatory targets were identified.
Gene Programs/Canonical pathways
Twenty-two Gene Programs and 20 additional pathways were used to characterize the molecular, signaling, and pathway level
characteristics of the iCluster-based subtypes. The Gene Programs were identified in a previous PanCancer analysis of 12 tumor
types, by 1) assembling 6,898 gene signatures documented to contain gene sets that are coexpressed, coamplified, or function
together; 2) applying a bimodality filter to select only those signatures with bimodal (ON/OFF) expression; and 3) performing weighted
gene correlation network-based clustering (WGCNA) to identify a non-redundant set of expression modules/programs (see Hoadley
et al. [2014] and associated SI, Section 5, for details). These Gene Programs were evaluated in the PanCancer-33 dataset by
averaging the top most-correlated signatures from each module (Table S9). The 20 additional pathways represent known drug
targets or/and canonical cancer pathways (Table S4 of Hoadley et al. [2014]) and were evaluated as the mean expression level of
pathway genes.
QUANTIFICATION AND STATISTICAL ANALYSES
Quantitative and statistical methods are noted above according to their respective technologies and analytic approaches.
DATA AND SOFTWARE AVAILABILITY
The raw data, processed data and clinical data can be found at the legacy archive of the GDC (https://portal.gdc.cancer.gov/
legacy-archive/search/f) and the PancanAtlas publication page (https://gdc.cancer.gov/about-data/publications/pancanatlas).
The mutation data can be found here: (https://gdc.cancer.gov/about-data/publications/mc3-2017). TCGA data can also be explored
through the Broad Institute FireBrowse portal (http://gdac.broadinstitute.org) and the Memorial Sloan Kettering Cancer Center
cBioPortal (http://www.cbioportal.org). Details for software availability are in the Key Resource Table.
e6 Cell 173, 291–304.e1–e6, April 5, 2018