The new england
journal of medicine
n engl j med 372;26 nejm.org June 25, 2015
2481
established in 1812
June 25, 2015
vol. 372 no. 26
The authors’ full names and academic
degrees are listed in the Appendix. Address
reprint requests to Dr. Daniel J. Brat at
the Department of Pathology and Labora-
tory Medicine, Winship Cancer Institute,
Emory University Hospital, G-167, 1364
Clifton Rd. N.E., Atlanta, GA 30322, or at
dbrat@ emory . edu.
* The authors are members of the Cancer
Genome Atlas Research Network, and
their names, affiliations, and roles are
listed in Supplementary Appendix 1,
available at NEJM.org.
This article was published on June 10, 2015,
at NEJM.org.
N Engl J Med 2015;372:2481-98.
DOI: 10.1056/NEJMoa1402121
Copyright © 2015 Massachusetts Medical Society.
BACKGROUND
Diffuse low-grade and intermediate-grade gliomas (which together make up the
lower-grade gliomas, World Health Organization grades II and III) have highly
variable clinical behavior that is not adequately predicted on the basis of histo-
logic class. Some are indolent; others quickly progress to glioblastoma. The un-
certainty is compounded by interobserver variability in histologic diagnosis. Muta-
tions in IDH, TP53, and ATRX and codeletion of chromosome arms 1p and 19q
(1p/19q codeletion) have been implicated as clinically relevant markers of lower-
grade gliomas.
METHODS
We performed genomewide analyses of 293 lower-grade gliomas from adults, in-
corporating exome sequence, DNA copy number, DNA methylation, messenger
RNA expression, microRNA expression, and targeted protein expression. These
data were integrated and tested for correlation with clinical outcomes.
RESULTS
Unsupervised clustering of mutations and data from RNA, DNA-copy-number, and
DNA-methylation platforms uncovered concordant classification of three robust,
nonoverlapping, prognostically significant subtypes of lower-grade glioma that
were captured more accurately by IDH, 1p/19q, and TP53 status than by histologic
class. Patients who had lower-grade gliomas with an IDH mutation and 1p/19q
codeletion had the most favorable clinical outcomes. Their gliomas harbored mu-
tations in CIC, FUBP1, NOTCH1, and the TERT promoter. Nearly all lower-grade
gliomas with IDH mutations and no 1p/19q codeletion had mutations in TP53
(94%) and ATRX inactivation (86%). The large majority of lower-grade gliomas
without an IDH mutation had genomic aberrations and clinical behavior strikingly
similar to those found in primary glioblastoma.
CONCLUSIONS
The integration of genomewide data from multiple platforms delineated three
molecular classes of lower-grade gliomas that were more concordant with IDH,
1p/19q, and TP53 status than with histologic class. Lower-grade gliomas with an
IDH mutation either had 1p/19q codeletion or carried a TP53 mutation. Most lower-
grade gliomas without an IDH mutation were molecularly and clinically similar to
glioblastoma. (Funded by the National Institutes of Health.)
abstract
Comprehensive, Integrative Genomic Analysis of Diffuse
Lower-Grade Gliomas
The Cancer Genome Atlas Research Network*
The New England Journal of Medicine
Downloaded from nejm.org at UNIV OF NC/ACQ SRVCS on February 20, 2018. For personal use only. No other uses without permission.
Copyright © 2015 Massachusetts Medical Society. All rights reserved.
n engl j med 372;26 nejm.org June 25, 2015
2482
The
new england journal
of
medicine
D
iffuse low-grade and intermedi-
ate-grade gliomas (World Health Organi-
zation [WHO] grades II and III, hereafter
called lower-grade gliomas) (see the Glossary)
are infiltrative neoplasms that arise most often
in the cerebral hemispheres of adults and include
astrocytomas, oligodendrogliomas, and oligoas-
trocytomas.
1,2
Because of their highly invasive
nature, complete neurosurgical resection is im-
possible, and the presence of residual tumor re-
sults in recurrence and malignant progression,
albeit at highly variable intervals. A subset of
these gliomas will progress to glioblastoma
(WHO grade IV gliomas) within months, where-
as others remain stable for years. Similarly, sur-
vival ranges widely, from 1 to 15 years, and some
lower-grade gliomas have impressive therapeutic
sensitivity.
3-5
Current treatment varies with the
extent of resection, histologic class, grade, and
the results of ancillary testing and includes clini-
cal monitoring, chemotherapy, and radiation
therapy, with salvage options available in the
event of treatment failure.
6-8
Although the histopathological classification
of lower-grade gliomas is time-honored, it suf-
fers from high intraobserver and interobserver
variability and does not adequately predict clini-
cal outcomes.
9,10
Consequently, clinicians increas-
ingly rely on genetic classification to guide clini-
cal decision making.
11-14
Mutations in IDH1 and
IDH2 (two very similar genes, hereafter referred
to collectively as IDH) characterize the majority
of lower-grade gliomas in adults and define a
subtype that is associated with a favorable prog-
nosis.
15-17
Lower-grade gliomas with both an IDH
mutation (i.e., a mutation in either IDH1 or IDH2)
and deletion of chromosome arms 1p and 19q
(1p/19q codeletion), which occurs most often in
oligodendrogliomas, have better responses to ra-
diochemotherapy and are associated with longer
survival than diffuse gliomas without these al-
terations.
5,18
TP53 and ATRX mutations are more
frequent in astrocytomas and are also important
markers of clinical behavior.
19
To gain additional
insight, we performed a comprehensive, integra-
tive analysis of 293 lower-grade gliomas from
adults, using multiple advanced molecular plat-
forms. We performed an unsupervised analysis
of integrated whole-genome molecular data to
determine whether we could identify biologic
classes of disease with clinically distinct behav-
ior and to determine whether these classes were
captured more accurately by molecular-marker
status than by histologic class.
Methods
Patients
The tumor samples we analyzed were from 293
adults with previously untreated lower-grade glio-
mas (WHO grades II and III), including 100 astro-
cytomas, 77 oligoastrocytomas, and 116 oligo-
dendrogliomas. Pediatric lower-grade gliomas
were excluded; their molecular pathogenesis is
distinct from that of lower-grade gliomas in
adults.
20,21
Diagnoses were established at the con-
tributing institutions; neuropathologists in our
consortium reviewed the diagnoses and ensured
the quality of the diagnoses and of the tissue for
molecular profiling (see Supplementary Appen-
dix 1, available with the full text of this article
at NEJM.org, for sample inclusion criteria). Pa-
tient characteristics are described in Table 1, and
in Table S1 (Supplementary Appendix 2) and
Table S2 in Supplementary Appendix 1. We ob-
tained appropriate consent from relevant institu-
tional review boards, which coordinated the con-
sent process at each tissue-source site; written
informed consent was obtained from all partici-
pants. The patients’ ages, tumor locations, clini-
cal histories and outcomes, tumor histologic
classifications, and tumor grades were typical of
adults with a diagnosis of diffuse glioma.
1,2
Analytic Platforms
We performed exome sequencing (289 samples),
DNA copy-number profiling (285), messenger
RNA (mRNA) sequencing (277), microRNA se-
quencing (293), DNA methylation profiling (289),
TERT promoter sequencing (287), and reverse-
phase protein lysate array (RPPA) profiling (255).
22
Complete data for all platforms were available
for 254 samples. Whole-genome sequencing and
low-pass whole-genome sequencing were per-
formed on 21 and 52 samples, respectively. Mo-
lecular data were frozen on January 31, 2014, and
clinical data were frozen on August 25, 2014. We
also performed an unsupervised analysis (i.e., an
analysis in which the categories are not known
before computation) that integrated results from
multiple platforms, including cluster of clusters
(CoC) and OncoSign.
23
In brief, CoC is a second-
level clustering of class assignments derived from
each individual molecular platform. OncoSign
A video summary
is available at
NEJM.org
The New England Journal of Medicine
Downloaded from nejm.org at UNIV OF NC/ACQ SRVCS on February 20, 2018. For personal use only. No other uses without permission.
Copyright © 2015 Massachusetts Medical Society. All rights reserved.
n engl j med 372;26 nejm.org June 25, 2015
2483
Genomic Analysis of Lower-Grade Gliomas
classifies tumors on the basis of similarities in
recurrent mutations and copy-number variations.
The complete data sets are provided in Table
S1 (Supplementary Appendix 2). The primary
sequence files are deposited in CGHub (https:/ /
cghub . ucsc . edu); all other data, including muta-
tion annotation files, are deposited at the Cancer
Genome Atlas Data Coordinating Center (http://
cancergenome . nih . gov). Sample lists, data ma-
trixes, and supporting data are available at the
Cancer Genome Atlas lower-grade glioma publi-
cation page (https:/ / tcga-data . nci . nih . gov/ docs/
publications/ lgg_2015).
Statistical Analysis
The statistical analysis included Fisher’s exact
test for associations of categorical variables,
one-way analysis of variance for association with
continuous outcomes, Kaplan–Meier estimates
of survival with log-rank tests among strata, and
Cox proportional-hazards regression for multi-
ple-predictor models of survival. A complete
description of the methods is provided in Sup-
plementary Appendix 1.
Results
Histologic and Molecular Subtypes
To compare the results from molecular plat-
forms with both histologic classification and
classification based on markers frequently used
in clinical practice (IDH mutation and 1p/19q
codeletion), we classified lower-grade gliomas
into three categories: gliomas with an IDH muta-
tion and 1p/19q codeletion, gliomas with an IDH
mutation and no 1p/19q codeletion, and gliomas
with wild-type IDH. We found a strong correla-
tion between the presence of an IDH mutation
and 1p/19q codeletion and the oligodendroglio-
ma histologic class (69 of 84 samples) (Table 1,
Adjusted Rand index: A measure of the similarity between two data clusterings, adjusted for chance grouping of the ele-
ments.
Cluster of clusters (CoC) analysis: A method of obtaining clusters (e.g., of patient samples) that represent a consensus
among the individual data types (in this study, we incorporated DNA methylation, DNA copy number, mRNA ex-
pression, and microRNA expression into the analysis).
Double-minute chromosome–breakpoint-enriched region (DM-BER): As detected by whole-exome and whole-genome
sequencing, highly amplified gene regions that are connected by DNA rearrangement breakpoints and allow cancer
cells to maintain high levels of oncogene amplification.
Exon: The portion of a gene that encodes amino acids to form a protein.
Fusion transcript: A transcript composed of parts of two separate genes joined together by a chromosomal rearrange-
ment, in some cases with functional consequences for oncogenesis, therapy, or both.
Glioblastoma: The highest-grade (World Health Organization grade IV) and most frequently occurring form of diffusely
infiltrative astrocytoma. It arises most often in the cerebral hemispheres of adults and is distinguished histopatho-
logically from diffuse lower-grade astrocytomas (grades II and III) by the presence of necrosis or microvascular pro-
liferation.
Lower-grade glioma: A diffusely infiltrative low-grade or intermediate-grade glioma (World Health Organization grade II
or III) that arises most often in the cerebral hemispheres of adults and includes astrocytomas, oligodendrogliomas,
and oligoastrocytomas.
Methylation: The attachment of methyl groups to DNA at cytosine bases. Methylation is correlated with reduced tran-
scription of the gene immediately downstream of the methylated site.
microRNA: A short regulatory form of RNA that binds to a target RNA and generally suppresses its translation by ribo-
somes.
Molecular subtype: Subgroup of a tumor type based on molecular characteristics (rather than, e.g., histologic or clinical
features); in this study, a molecular subtype is one of three classes based on IDH mutation and 1p/19q codeletion
status.
Mutation frequency: The number of mutations detected per megabase of DNA.
Significantly mutated gene: A gene with a greater number of mutations than expected on the basis of the background
mutation rate, which suggests a role in oncogenesis.
Whole-exome sequencing: Sequencing of the coding regions, or exons, of an entire genome.
Whole-genome sequencing: Sequencing of the entire genome.
Glossary
The New England Journal of Medicine
Downloaded from nejm.org at UNIV OF NC/ACQ SRVCS on February 20, 2018. For personal use only. No other uses without permission.
Copyright © 2015 Massachusetts Medical Society. All rights reserved.
n engl j med 372;26 nejm.org June 25, 2015
2484
The
new england journal
of
medicine
Characteristic
Total
(N = 278)†
IDH Mutation and
1p/19q Codeletion
(N = 84)
IDH Mutation and No
1p/19q Codeletion
(N = 139)
IDH Wild Type
(N = 55)
Histologic type‡ and grade‡ — no. (%)
Oligodendroglioma
Grade II 65 (23) 38 (45) 21 (15) 6 (11)
Grade III 44 (16) 31 (37) 6 (4) 7 (13)
Oligoastrocytoma
Grade II 41 (15) 9 (11) 30 (22) 2 (4)
Grade III 33 (12) 4 (5) 20 (14) 9 (16)
Astrocytoma
Grade II 30 (11) 1 (1) 24 (17) 5 (9)
Grade III 65 (23) 1 (1) 38 (27) 26 (47)
Age at diagnosis — yr‡
Mean 42.6±13.5 45.4±13.2 38.1±10.9 49.9±15.3
Range 14–75 17–75 14–70 21–74
Male sex — no. (%) 155 (56) 45 (54) 84 (60) 26 (47)
White race — no./total no. (%)§ 261/274 (95) 79/81 (98) 131/138 (95) 51/55 (93)
Year of diagnosis — no. (%)
Before 2005 38 (14) 10 (12) 18 (13) 10 (18)
2005–2009 88 (32) 30 (36) 44 (32) 14 (25)
2010–2013 152 (55) 44 (52) 77 (55) 31 (56)
Family history of cancer — no./total no. (%)¶
None 108/190 (57) 30/58 (52) 64/98 (65) 13/34 (38)
Primary brain cancer 11/190 (6) 2/58 (3) 7/98 (7) 2/34 (6)
Other cancers 72/190 (38) 26/58 (45) 27/98 (28) 19/34 (56)
Extent of resection — no./total no. (%)
Open biopsy 6/268 (2) 1/81 (1) 4/132 (3) 1/55 (2)
Subtotal resection 98/268 (37) 31/81 (38) 45/132 (34) 22/55 (40)
Gross total resection 164/268 (61) 49/81 (60) 83/132 (63) 32/55 (58)
Tumor location — no. (%)‡
Frontal lobe 172 (62) 68 (81) 84 (60) 20 (36)
Parietal lobe 23 (8) 5 (6) 13 (9) 5 (9)
Temporal lobe 74 (27) 9 (11) 40 (29) 25 (45)
Other‖ 9 (3) 2 (2) 2 (1) 5 (9)
Laterality — no./total no. (%)
Left 133/276 (48) 37/84 (44) 69/137 (50) 27/55 (49)
Midline 5/276 (2) 2/84 (2) 2/137 (1) 1/55 (2)
Right 138/276 (50) 45/84 (54) 66/137 (48) 27/55 (49)
White matter — no./total no. (%) 74/144 (51) 26/48 (54) 37/72 (51) 11/24 (46)
First presenting symptom — no./total no. (%)
Headache 64/252 (25) 15/72 (21) 39/129 (30) 10/51 (20)
Mental status change 22/252 (9) 7/72 (10) 10/129 (8) 5/51 (10)
Motor or movement change 18/252 (7) 6/72 (8) 7/129 (5) 5/51 (10)
Table 1. Clinical Characteristics of the Sample Set According to IDH Mutation and 1p/19q Codeletion Status.*
The New England Journal of Medicine
Downloaded from nejm.org at UNIV OF NC/ACQ SRVCS on February 20, 2018. For personal use only. No other uses without permission.
Copyright © 2015 Massachusetts Medical Society. All rights reserved.
n engl j med 372;26 nejm.org June 25, 2015
2485
Genomic Analysis of Lower-Grade Gliomas
and Table S2 in Supplementary Appendix 1), a
finding consistent with that in previous stud-
ies.
24,25
Glioma samples with an IDH mutation
and no 1p/19q codeletion (139 samples, 50% of
the cohort) represented a mixture of histologic
classes but were enriched for astrocytomas and
oligoastrocytomas. IDH wild-type samples were
mostly astrocytomas (31 of 55 samples) and
grade III gliomas (42 of 55 samples), but this
group included other histologic classes and
grades. Overall, classification based on IDH–
1p/19q status correlated strongly with the oligo-
dendroglioma histologic class but only modestly
with astrocytoma and oligoastrocytoma.
Multiplatform Integrative Analysis
To determine whether advanced molecular pro-
filing could subdivide lower-grade gliomas into
discrete sets that are associated with biologic
characteristics of disease, we performed unsuper-
vised clustering of molecular data derived from
four independent platforms and found well-
defined clusters based on DNA methylation (five
clusters) (Fig. S1 through S5 in Supplementary
Appendix 1), gene expression (four clusters)
(Fig. S6 and S7 in Supplementary Appendix 1 and
Table S7 [Supplementary Appendix 6]), DNA copy
number (three clusters) (Fig. S8 in Supplementary
Appendix 1), and microRNA expression (four
clusters) (Fig. S9 and S10 in Supplementary Ap-
pendix 1 and Table S8 [Supplementary Appendix 7]
and Table S9 [Supplementary Appendix 8]).
22,26,27
To integrate data and compare the resulting
biologic classes with histologic classes and sub-
types based on IDH–1p/19q status, cluster group
assignments from the four individual platforms
(DNA methylation, mRNA, DNA copy number,
and microRNA) were used for a second-level CoC
analysis, resulting in three CoC clusters with dis-
tinctive biologic themes (Fig. 1). We found a strong
correlation between CoC cluster assignment and
molecular subtypes defined on the basis of IDH–
1p/19q codeletion status: most lower-grade glio-
mas with wild-type IDH were in the CoC cluster
that included mRNA cluster R2, microRNA clus-
ter Mi3, DNA methylation cluster M4, and DNA
copy number cluster C2. Another CoC cluster con-
tained almost all gliomas with an IDH mutation
and 1p/19q codeletion and included primarily
clusters R3, M2 and M3, and C3. The third CoC
cluster was highly enriched for gliomas with an
IDH mutation and no 1p/19q co deletion and in-
cluded clusters R1, M5, C1, and Mi1.
To determine the relative strength of clinical
schemes for the classification of lower-grade
gliomas in capturing the biologic subsets revealed
by CoC analysis, we compared the correlation
between IDH–1p/19q subtype and CoC cluster
assignment with the correlation between histo-
logic class and CoC cluster assignment. Whereas
90% of samples with a specific IDH–1p/19q desig-
nation mapped one-to-one with a predominant
CoC cluster, only 63% of samples within a specific
histologic class showed this predominant map-
ping. Moreover, the concordance between IDH–
1p/19q status and CoC cluster assignment was
much greater than that between histologic sub-
type and CoC cluster assignment (adjusted Rand
index, 0.79 vs. 0.19) (Table S2E in Supplementary
Appendix 1), which indicates that IDH–1p/19q
Characteristic
Total
(N = 278)†
IDH Mutation and
1p/19q Codeletion
(N = 84)
IDH Mutation and No
1p/19q Codeletion
(N = 139)
IDH Wild Type
(N = 55)
Seizure 135/252 (54) 38/72 (53) 70/129 (54) 27/51 (53)
Sensory or visual change 13/252 (5) 6/72 (8) 3/129 (2) 4/51 (8)
* Plus–minus values are means ±SD. Categorical distributions were compared with the use of Fisher’s exact test. Analysis of variance was used
to compare age between groups.
† IDH–1p/19q status was not determined for 11 cases with clinical information.
‡ P<0.01 for the difference among the molecular subtypes.
§ Race was self-reported. Of the 261 patients who reported their ethnic background, 5% identified themselves as Hispanic or Latino.
¶ Included are patients for whom responses to questions regarding a family history of any cancer (192 patients) and a family history of primary
brain cancer (197 patients) were available. P<0.05 for the difference among the molecular subtypes.
‖ One case (with wild-type IDH) was in the cerebellum, three cases were in the occipital lobe (two with IDH mutation and 1p/19q codeletion
and one with an IDH mutation and no codeletion), and five cases were listed as “supratentorial, not otherwise specified” (one with an IDH
mutation and no codeletion and four with wild-type IDH).
Table 1. (Continued.)
The New England Journal of Medicine
Downloaded from nejm.org at UNIV OF NC/ACQ SRVCS on February 20, 2018. For personal use only. No other uses without permission.
Copyright © 2015 Massachusetts Medical Society. All rights reserved.
