tumor-infiltrating lymphocytes in glioblastoma are ...drgutman.org/original_papers/38a.pdf · clin...

2013;19:4951-4960. Published OnlineFirst July 17, 2013.Clin Cancer Res W. Caleb Rutledge, Jun Kong, Jingjing Gao, et al. Classwith Specific Genomic Alterations and Related to Transcriptional Tumor-Infiltrating Lymphocytes in Glioblastoma Are Associated

Updated version

10.1158/1078-0432.CCR-13-0551doi:

Access the most recent version of this article at:

Cited Articles

http://clincancerres.aacrjournals.org/content/19/18/4951.full.html#ref-list-1

This article cites by 36 articles, 10 of which you can access for free at:

E-mail alerts related to this article or journal.Sign up to receive free email-alerts

Subscriptions

Reprints and

[email protected]

To order reprints of this article or to subscribe to the journal, contact the AACR Publications Department at

Permissions

[email protected]

To request permission to re-use all or part of this article, contact the AACR Publications Department at

on October 30, 2013. © 2013 American Association for Cancer Research. clincancerres.aacrjournals.org Downloaded from

Published OnlineFirst July 17, 2013; DOI: 10.1158/1078-0432.CCR-13-0551





















http://clincancerres.aacrjournals.org/lookup/doi/10.1158/1078-0432.CCR-13-0551

http://clincancerres.aacrjournals.org/content/19/18/4951.full.html#ref-list-1

http://clincancerres.aacrjournals.org/cgi/alerts

mailto:[email protected]

mailto:[email protected]

http://clincancerres.aacrjournals.org/











Human Cancer Biology

Tumor-Infiltrating Lymphocytes in Glioblastoma AreAssociated with Specific Genomic Alterations and Relatedto Transcriptional Class

W.CalebRutledge1, JunKong2,4, JingjingGao2,4, David A. Gutman2,4,5, Lee A.D. Cooper2,4,5, Christina Appin3,Yuna Park3, Lisa Scarpace7, Tom Mikkelsen7, Mark L. Cohen8, Kenneth D. Aldape9, Roger E. McLendon10,Norman L. Lehman11, C. Ryan Miller12, Matthew J. Schniederjan6, Cameron W. Brennan13, Joel H. Saltz2,3,4,5,Carlos S. Moreno2,3,4,5, and Daniel J. Brat2,3,4,5

AbstractPurpose: Tumor-infiltrating lymphocytes (TIL) have prognostic significance in many cancers, yet their

roles in glioblastoma have not been fully defined. We hypothesized that TILs in glioblastoma are associated

with molecular alterations, histologies, and survival.

Experimental Design: We used data from The Cancer Genome Atlas (TCGA) to investigate molecular,

histologic, and clinical correlates of TILs in glioblastomas. Lymphocytes were categorized as absent, present,

or abundant in histopathologic images from 171 TCGA glioblastomas. Associations were examined

between lymphocytes and histologic features, mutations, copy number alterations, CpG island methylator

phenotype, transcriptional class, and survival.Wevalidatedhistologic findingsusingCD3G gene expression.

Results:We found apositive correlation betweenTILs and glioblastomaswith gemistocytes, sarcomatous

cells, epithelioid cells, and giant cells. Lymphocytes were enriched in themesenchymal transcriptional class

and strongly associated with mutations inNF1 and RB1. These mutations are frequent in the mesenchymal

class and characteristic of gemistocytic, sarcomatous, epithelioid, and giant cell histologies. Conversely, TILs

were rare in glioblastomas with small cells and oligodendroglioma components. Lymphocytes were

depleted in the classical transcriptional class and in EGF receptor (EGFR)-amplified and homozygous

PTEN-deleted glioblastomas. These alterations are characteristic of glioblastomas with small cells and

glioblastomas of the classical transcriptional class. No association with survival was shown.

Conclusions: TILs were enriched in glioblastomas of the mesenchymal class, strongly associated with

mutations inNF1 andRB1 and typical of histologies characterized by thesemutations. Conversely, TILswere

depleted in the classical class, EGFR-amplified, and homozygous PTEN-deleted tumors and rare in

histologies characterized by these alterations. Clin Cancer Res; 19(18); 4951–60. �2013 AACR.

IntroductionGlioblastoma is the most common and highest grade

astrocytoma (World Health Organization, grade IV).

Despite advances in diagnosis and treatment, glioblastomaremains incurable with an average survival of 15 months(1). Although characterized by dramatic molecular andhistologic heterogeneity, current adjuvant treatments inhib-it cell divisionnonspecifically andare not tailored to specificsubsets (1, 2). Molecular classifications of glioblastomahave revealed distinct subclasses with clinical relevance,opening the way for therapies to be directed at class-specificmechanisms (3, 4). Immunotherapy, for example, has greatpotential as a tailored therapy. Already U.S. Food and DrugAdministration (FDA)-approved for prostate cancer andmetastatic melanoma and in clinical trials for other cancertypes, immunotherapy remains promising for patients withglioblastoma, offering high tumor specificity and long-termtumor surveillance (5–7).

Tumor-infiltrating lymphocytes (TIL) are present in sub-sets of glioblastoma, yet their biologic activities and clinicalrelevance have not been fully explored. The presence of TILssuggests an antitumor adaptive immune response andmight be expected to impact survival. Lymphocytes are

Authors' Affiliations: 1Emory University School ofMedicine, Departmentsof 2Biomedical Informatics and 3Pathology and LaboratoryMedicine,4Cen-ter for Comprehensive Informatics, 5Winship Cancer Institute, EmoryUniversity; 6Department of Pathology, Children's Healthcare of Atlanta,Atlanta, Georgia; 7Department of Neurosurgery, Henry Ford Hospital,Detroit, Michigan; 8Department of Pathology, Case Western ReserveUniversity, Cleveland, Ohio; 9Department of Pathology, University of TexasMDAndersonCancer Center, Houston, Texas; 10Department of Pathology,Duke University Medical Center, Durham, North Carolina; 11Department ofPathology, Ohio State University, Columbus, Ohio; 12Department ofPathology, University of North Carolina, Chapel Hill, North Carolina; and13Department of Neurosurgery, Memorial Sloan Kettering Cancer Center,New York, New York

Corresponding Author: Daniel J. Brat, Department of Pathology andLaboratoryMedicine, Emory University Hospital, G-167, 1364 Clifton RoadNE, Atlanta, GA 30322. Phone: 404-712-1266; Fax: 404-727-3133; E-mail:[email protected]

doi: 10.1158/1078-0432.CCR-13-0551

�2013 American Association for Cancer Research.

ClinicalCancer

Research

www.aacrjournals.org 4951

present within the stroma of other cancers including mel-anoma, colorectal, and ovarian carcinoma, where theirpresence is associated with improved patient outcomes(8–17). A recent study suggested that TILs have prognosticsignificance in glioblastoma, yet others have shown lym-phocytes may impart a less favorable prognosis (18–21).

A complex relationship exists between the developmentof cancer, the immunologic response to it, and the immu-nosuppression it causes. In glioblastoma, tumor-mediatedimmunosuppression is thought to explain functional def-icits in cytotoxic lymphocytes (22). Regulatory lymphocytesare elevated in patients with glioblastoma and suppressantitumor activity by inhibiting secretion of cytotoxic cyto-kines from effector lymphocytes (23, 24). Thus, whilelymphocytes and other immune cells infiltrate glioblasto-ma, an immunosuppressive tumor milieu likely preventssuccessful immune-mediated tumor eradication. Recentclinical trials have focused on augmenting the antitumorimmune response with tumor vaccines or reversing tumor-mediated immunosuppression.

Molecular alterations in glioblastoma and other cancersmay be immunogenic. For example, microsatellite instabil-ity andmethylation aberrations are independent predictorsof lymphocyte density in colorectal cancer (14). Given itsmolecular and histologic heterogeneity, subsets of glioblas-tomamay be more immunogenic and responsive to immu-notherapy, yet little is known about the relation betweenimmune cells and the tumor microenvironment or molec-ular classes in glioblastoma. Therefore, the purpose of thisstudy was to identify molecular and histologic correlates ofthe immune response in glioblastoma.We usedwhole-slidedigitized images of glioblastomas fromTheCancerGenomeAtlas (TCGA) linked to multiplatform molecular data.These may provide insight into the biologic significance ofTILs in glioblastoma and impact therapy.

Materials and MethodsWe conducted an integrated molecular and histologic

analysis using data from TCGA, which molecularly charac-terized glioblastomas across multiple platforms, includingsingle-nucleotide polymorphism (SNP) genotyping,mRNA

and microRNA profiling, DNA sequencing, and methyla-tion analysis (2). Clinical data, including treatment andsurvival, is also available.

Digitized images used for morphologic analysisPermanent section histologic slides from TCGA glioblas-

tomas were provided by contributing institutions. Slideswere scanned and digitized at 20� resolution by a high-throughput digital scanner (Aperio, Inc.) at the TCGABiospecimen Core Resource located at the InternationalGenomics Consortium (Intgen). The number of slidesavailable for review ranged from1 to 9 per case (median, 3).

Ratings of TILs and other histopathologic featuresTCGA consortiumneuropathologists annotated digitized

permanent section histologic slides from 122 TCGA casesfor 18 histopathologic features including inflammation,angiogenesis, necrosis, lymphocytes, commonmorpholog-ic subtypes, and others (Table 1). "Inflammation" as ageneral category was defined by the presence of inflamma-tory cells, including lymphocytes, macrophages, neutro-phils, or their combination. Lymphocytes were identifiedas small round cells with scant cytoplasm and darkly stain-ing nuclei. Lymphocytes and all other histopathologic fea-tures were categorized as absent (0), present (1þ), orabundant (2þ) by 2 neuropathologists (M.L. Cohen, K.D. Aldape, R.E. McLendon, N.L. Lehman, C.R. Miller, M.J.Schniederjan, D.J. Brat) and adjudicated by a third. Caseswith a complete absence of lymphocytes were labeled 0(absent). Caseswith lymphocytes in less than 50%of tumortissue were categorized 1þ (present), whereas cases withlymphocytes in �50% were categorized 2þ (abundant;Fig. 1). In addition to these 122 cases, an additional 49TCGA cases from Emory University Hospital and HenryFord Health System were annotated for the same 18 histo-pathologic features using the same criteria by 2 TCGAneuropathologists (D.J. Brat, M.J. Schniederjan). All digi-tized histologic sections and TCGA neuropathology ratingswere obtained from the TCGA portal (http://tcga-data.nci.nih.gov.proxy.library.emory.edu/tcga/tcgaHome2.jsp; lastaccessed November 28, 2012). Digitized images and corre-sponding neuropathology ratings were the primary sourceof data.

Mutations, copy number alterations, methylationstatus, and transcriptional class

We obtained TCGA mutation and copy number datafrom theMemorial SloanKetteringCancerGenomics Portal(http://www.cbioportal.org/public-portal; last accessedNovember 28, 2012; ref. 25). Mutation data from wholeexome sequencing (NextGen) was obtained for 99 cases.Putative copy number calls determined with GISTIC 2.0were obtained for 153 cases. CpG island methylator phe-notype (G-CIMP) status was available for 124 cases. Tran-scriptional class labels for the Verhaak classification wereobtained for 162 cases from the TCGA Advanced WorkingGroup. The updated Verhaak labeling extends the originallabeled set presented by Verhaak and colleagues by using

Translational RelevanceTumor-infiltrating lymphocytes (TIL) are present in a

subset of glioblastomas, suggesting that some are morecapable of eliciting an adaptive immune response. Wehypothesized that TILs are differentially distributedamong specific molecular classes of glioblastoma andused histologic data linked to multiplatform molecularanalysis from The Cancer Genome Atlas (TCGA) toidentify correlates. We have shown that TILs are associ-ated with specific genomic alterations and related totranscriptional class, whichmay provide insight into thebiologic significance of TILs in glioblastoma and predictresponse to immunotherapy.

Rutledge et al.

Clin Cancer Res; 19(18) September 15, 2013 Clinical Cancer Research4952

the originally labeled samples along with AffymetrixHT_HG-U133A data to label previously unclassified sam-ples (3).

Validation datasetCD3G is the gene that encodes the T-cell surface

marker CD3. We used CD3G expression data from TCGAas a marker of lymphocytes to validate our histologicfindings in the same set of tumors. CD3G expression datawere obtained from the TCGA portal (http://tcga-data.nci.nih.gov.proxy.library.emory.edu/tcga/tcgaHome2.jsp; last accessed February 1, 2012).

Statistical analysesAssociations between lymphocytes and other histopath-

ologic features were assessed using the Mantel–Haenzel c2

and exact test and Spearman correlation for ordinal com-parisons (0, 1þ, 2þ vs. 0, 1þ, 2þ). The Mantel–Haenzelexact test was used when the cell count was less than 5 in25% or more of cells. The c2 and Fisher’s exact tests wereused for all dichotomous comparisons (0 vs. 1þ, 2þ com-bined or 0, 1þ combined vs. 2þ). We combined categoriesto identify associations driven by cases with either complete

absence or abundance of lymphocytes. The association oflymphocytes with mutations, copy number alterations(CNA), and G-CIMP status were also assessed using c2 andFisher’s exact tests. Fisher’s exact test was used when the cell

Figure 1. Representative examples of permanent section histologic slidesfrom TCGA cases. Lymphocytes were identified as small round cells withscant cytoplasm and darkly staining nuclei. Cases with a completeabsence of lymphocytes were labeled 0 (absent; top). Cases withlymphocytes present in less than 50% of tumor tissue were categorized1þ (present; middle). Cases with lymphocytes present in �50% werecategorized 2þ (abundant; bottom).

Table 1. Eighteen histopathologic featureswere categorized as absent (0), present (1þ), orabundant (2þ) in 171 TCGA glioblastomas

Histopathologic feature

Mantel–Haenzelc2 P

Spearmancorrelation

Inflammation <0.001 0.93Pseudopalisading necrosis 0.01 �0.19Zonal necrosis 0.04 0.16Neutrophils 0.09 0.13Macrophages <0.001 0.32Microvascular hyperplasia 0.59 �0.04Endothelial hyperplasia 0.05 �0.15Epithelial metaplasia 0.04 0.16Gemistocytes 0.01 0.22Giant cells 0.02 0.18Oligodendroglial cells 0.01 �0.19Sarcomatous metaplasia <0.01 0.22Small cells <0.01 �0.24Mineralization 0.68 0.03Satellitosis 0.20 �0.10White matter invasion 0.09 �0.13Cortex invasion 0.71 �0.03

NOTE: TILs are associated with specific histopathologicfeatures in GBMs. Mantel–Haenzel c2 test and Spearmancorrelation were used to examine associations betweenlymphocytes (0, 1þ, 2þ) and other histopathologic features(0, 1þ, 2þ). Mantel–Haenzel exact c2 test was used when25% of the cells had less than 5 observations. Effectivesample size was 171. Bolded P values are <0.05.

Tumor-Infiltrating Lymphocytes in Glioblastoma

www.aacrjournals.org Clin Cancer Res; 19(18) September 15, 2013 4953

count was less than 5 in 25% or more of cells. Associationsbetween lymphocytes and transcriptional class were exam-ined using c2 and Fisher’s exact tests. The Spearman corre-lation was used to measure the correlation between lym-phocytes and CD3G expression. Wilcoxon 2-sample orKruskal–Wallis testswere used todetect differences inCD3Gexpression according to lymphocytes, mutation, or copynumber status and transcriptional class.

Survival analysisClinical data were obtained from the TCGA data portal.

Survival was taken as "days to death" for uncensoredpatients and "days to last follow-up" for right-censoredpatients. The association between TILs and survival wasexamined using the log-rank test.

All P values reported are 2-sided and regarded as statis-tically significant if P < 0.05. The software used for statisticalanalysis was SAS Version 9.3 (SAS Institute Inc.).

ResultsTILs are differentially distributed in glioblastoma

Within the 171 glioblastomas reviewed for this study,TILs were absent (0) in 93 cases (54%), present (1þ) in 59cases (35%), and abundant (2þ) in 19 cases (11%). Therewas no significant association between the number of slidesanalyzed for each case and the level of lymphocytesannotated.

TILs are associated with specific histopathologicfeatures in glioblastoma

Glioblastomas show a tremendous degree of histologicvariability and numerous morphologic subtypes have beenrecognized including fibrillary, gemistocytic, epithelioid,small cell, giant cell, gliosarcoma, and glioblastoma witholigodendroglioma component. Todeterminewhether TILscorrelatedwith specific glioblastomamorphologies or otherhistopathologic features, we examined their associationsusing theMantel–Haenzel c2 test and Spearman correlation(Table 1). All variables, including lymphocytes and otherhistopathologic features, were categorized as 0, 1þ, or 2þ.We detected a strong positive correlation between TILs andspecific tumor cell morphologies that included gemisto-cytes, sarcomatous cells, epithelioid cells, and giant cells (allP < 0.05; Fig. 2). Conversely, TILs were depleted in glio-blastomas characterized by small cells and oligodendroglialcells (both P < 0.05). Among other features analyzed, TILswere most tightly correlated with the findings of inflam-mation (as a general category) and macrophages (both P <0.05). TILs were also statistically associated with both formsof necrosis annotated (pseudopalisading and zonal; bothP < 0.05).

TILs are associated with specific mutations inglioblastoma

In the initial TCGA analysis of glioblastoma, 8 genes wereidentified as significantly mutated, including TP53, PTEN,NF1, EGFR, ERBB2, RB1, PIK3R1, and PIK3CA (genesattaining a false discovery rate < 0.1; ref. 2). A subsequent,

unbiased genomic analysis identified recurrent mutationsin isocitrate dehydrogenase 1 (IDH1; ref. 26). We examinedassociations between lymphocytes (0, 1þ, 2þ) and muta-tions (mutant vs. wild-type) for these genes using c2 andFisher’s exact test (Table 2). Our data set contained no caseswith ERBB2 mutations.

Figure 2. Representative examples of glioblastoma morphologicsubtypes associated with the presence of lymphocytes in TCGApermanent section histologic slides. TILs are enriched in glioblastomaswith sarcomatous cells (top), gemistocytes (middle), and giant cells(bottom).

Rutledge et al.


Among cases with mutational data, lymphocytes wereabsent (0) in 52, present (1þ) in 38, and abundant (2þ) in9. We found lymphocytes were strongly associated withmutations in neurofibromatosis 1 (NF1) and retinoblasto-ma 1 (RB1) (both P < 0.05). Nine cases with absent (0)lymphocytes were NF1-mutant (9 of 52, 17%), whereas44% of cases with abundant (2þ) lymphocytes harboredmutations in NF1 (4 of 9). Two cases with absent (0)lymphocytes harbored mutations in RB1 (2 of 52, 4%). Ofcases with lymphocytes present (1þ), 6 were RB1 mutants(6 of 32, 16%). Two cases with abundant (2þ) lymphocytesharbored mutations in RB1 (2 of 9, 22%).There was a trend toward significance for TP53mutations

(P ¼ 0.12), particularly when cases with present (1þ) andabundant (2þ) lymphocytes were combined (P ¼ 0.06).Twenty-five percent of cases with absent (0) lymphocyteswere TP53-mutant (13 of 52), whereas 43% of cases withpresent (1þ) or abundant (2þ) lymphocytes harboredmutations in TP53 (20 of 47).As mutations in TP53 are characteristic of both the

proneural andmesenchymal transcriptional class, we inves-tigated whether mutations were overrepresented in 1 ofthese 2 transcriptional classes (3). Twenty cases harboredmutations in TP53 and had present (1þ) or abundant (2þ)lymphocytes. Nine (45%) belonged to the mesenchymaltranscriptional class and 5 (25%)belonged to the proneuralclass (P > 0.05).

TILs are associated with specific CNAs in glioblastomaSignificant CNAs were also identified in the initial TCGA

global analysis publication, including amplifications ofEGFR, CDK4, PDGFRA, MDM2, MDM4, MET, CDK6,MYCN, CCND2, PIK3CA, and AKT3 and deletions ofCDKN2A/B, PTEN, CDKN2C, RB1, PARK2, and NF1 (2).We examined the association between TILs and these recur-rent amplifications and deletions using c2 and Fisher’s exacttests (Table 3). Lymphocytes were depleted in EGFR-ampli-fied tumors (P<0.05). AmplificationofEGFRwaspresent in65% of cases with absent (0) lymphocytes (53 of 82)

comparedwith 48%of cases with present (1þ) or abundant(2þ) lymphocytes. Only 35% of cases with abundant (2þ)lymphocytes were EGFR-amplified.

TILs were also depleted in tumors with homozygousdeletions of PTEN (P < 0.05). Seventy-eight percent ofcases with homozygous PTEN deletion (14 of 18) hadabsent (0) lymphocytes. Seventeen percent of PTEN-deleted cases had present (1þ) lymphocytes (3 of 18),whereas only 5% of cases had abundant (2þ) lympho-cytes (1 of 18). There was a trend toward significance forNF1 deletions (P ¼ 0.07). No other associations betweenTILs and CNAs were noted.

TILs are not associated with the CIMPG-CIMP–positive tumors represent less than 10% of

glioblastomas but have significantly improved survivalcomparedwithG-CIMP–negative tumors (27). IDH1muta-tions have been shown to establish the G-CIMP phenotype(28). We examined the association between TILs and theCIMP using c2 and Fisher’s exact tests (data not shown).TILs were not associated with G-CIMP status (P > 0.05).

TILs are related to transcriptional classVerhaak and colleagues identified 4 transcriptional clas-

ses of glioblastoma using TCGA gene expression data:proneural, neural, classical, andmesenchymal (3). To deter-mine whether there was an association between transcrip-tional class and TILs, we examined each class for its distri-bution of lymphocytes using c2 and Fisher’s exact tests(Table 4). We found that TILs were strongly associated withtranscriptional class (P < 0.05).

TILs were enriched in the mesenchymal class comparedwith all other classes combined (P < 0.05). Forty-twopercent of cases (30 of 72) with a lymphocytic infiltrate(1þ or 2þ) belonged to the mesenchymal class. However,when classical tumors were excluded, there was no statis-tically significant enrichment of TILs in the mesenchymalclass compared with the proneural and neural classes (P >0.05). Cases with abundant lymphocytes were heavily

Table 2. TILs are associated with specific mutations in glioblastoma

Gene nc2 P(0, 1þ, 2þ)

c2 P (0 vs. 1þ,2þ combined)

c2 P value (0, 1þcombined vs. 2þ)

TP53 33 0.13 0.06 0.15PTEN 29 0.15 0.06 0.44EGFR 26 0.82 0.54 1.00NF1 16 0.03 0.74 0.04RB1 10 0.04 0.04 0.22PIK3R1 9 0.67 0.49 1.00PIK3CA 7 0.72 1.00 0.50IDH1 4 0.53 1.00 0.32ERBB2 0 — — —

NOTE: c2 test was used to examine associations between lymphocytes (0, 1þ, 2þ) and mutations (0, 1). Fisher's exact test was usedwhen 25% of the cells had less than 5 observations. Effective sample size was 99. Categories were combined to identify associationsdriven by cases with either complete absence or abundance of lymphocytes. Bolded P values are <0.05.



enriched inmesenchymal transcriptional class. Seventy-onepercent of cases (12 of 17) with abundant lymphocytes(2þ) belonged to the mesenchymal class, whereas 12%were proneural, 12% neural, and 6% were classical, repre-senting statistically significant enrichment (P < 0.05).Whenclassical cases were excluded, there was a still a strong trendtoward statistically significant enrichment of abundant(2þ) TILs in the mesenchymal class compared with theproneural and neural classes (P ¼ 0.07).

Conversely, TILs were depleted in the classical tran-scriptional case compared with all other classes combined(P < 0.05). Seventy-four percent of classical cases (31 of42) were characterized by absent (0) lymphocytes, repre-senting statistically significant depletion (P < 0.05). Only

one case with abundant (2þ) lymphocytes belonged tothe classical transcriptional class (1 of 17). When mes-enchymal tumors were excluded, there was still a trendtoward statistically significant depletion of TILs (1þ and2þ) in the classical class compared with the proneuraland neural classes (P ¼ 0.053).

CD3G gene expression is positively correlated withlymphocytes, mutations in TP53, RB1 and themesenchymal transcriptional class and negativelycorrelatedwithEGFR amplification,PTENdeletion, andthe classical transcriptional class

We used CD3G expression data to validate findingsuncovered by our morphologic analysis. There was a

Table 4. TILs are enriched in the mesenchymal class and depleted in the classical class

Transcriptional class

Lymphocytes Classical Mesenchymal Neural Proneural Total

0 31 27 13 19 901þ 10 18 9 18 552þ 1 12 2 2 17Total 42 57 24 39 162

NOTE: c2 test was used to examine associations between lymphocytes (0, 1þ, 2þ) and transcriptional class. Fisher's exact test wasused when 25% of the cells had less than 5 observations. Effective sample size was 162.

Table 3. TILs are depleted in EGFR-amplified and homozygous PTEN-deleted tumors

CNA %c2 P value(0, 1þ, 2þ)

c2 P value (0 vs. 1þ,2þ combined)

c2 P value (0, 1þcombined vs. 2þ)

AmplificationsEGFR 49.3 0.05 0.04 0.06CDK4 14.5 0.37 0.22 0.47PDGFRA 14.1 0.61 0.73 0.47MDM4 9.9 0.23 0.24 0.64MDM2 9.3 0.81 0.82 1.00MET 8.8 0.57 0.79 0.48CDK6 7.0 0.75 0.65 0.69CCND2 4.0 1.00 1.00 1.00AKT3 3.2 1.00 1.00 1.00MYCN 2.6 1.00 0.62 1.00PIK3CA 2.4 0.69 0.60 1.00

DeletionsCDKN2A 62.0 0.23 0.09 0.48CDKN2B 61.0 0.23 0.09 0.48PTEN 10.3 0.09 0.03 0.69RB1 3.6 0.87 1.00 0.57CDKN2C 3.6 0.71 1.00 0.51PARK2 2.0 1.00 1.00 1.00NF1 1.2 0.07 0.10 0.30

NOTE: c2 test was used to examine associations between lymphocytes (0, 1þ, 2þ) and CNAs. Fisher's exact test was usedwhen 25%of the cells had less than 5 observations. Effective sample size was 153. Categories were combined to identify associations driven bycases with either complete absence or abundance of lymphocytes.

Rutledge et al.


positive correlation (0.30) between the histologic catego-rization of lymphocytes and CD3G expression (P < 0.001).Glioblastomas with higher levels of TILs had significantlyhigherCD3G expression (P < 0.001). Those with absent (0),present (1þ), and abundant (2þ) lymphocytes had CD3Gexpression of 15.34 (SD � 7.81), 17.55 (SD � 10.99), and20.23 (SD � 13.51), respectively.We examined levels of CD3G expression in cases harbor-

ing mutations in RB1, TP53, and NF1. Cases with RB1mutations tended to have higher levels of CD3G expressionthan wild-type cases (19.1 vs. 16.2). Cases with TP53mutations also showed higher levels of CD3G expressionthan wild-type (17.4 vs. 16.0). Cases with NF1 mutationshad lower levels of CD3G expression compared with wild-type (15.6 vs. 16.5; all P > 0.05).We also investigatedCD3G expression in EGFR-amplified

and PTEN-deleted cases. EGFR-amplified cases had lowerlevels of CD3G expression than wild-type (15.6 vs. 16.7).Likewise, PTEN-deleted cases also had lower levels ofCD3Gexpression (15.5 vs. 16.3; both P > 0.05).Finally, we analyzedCD3G expression levels by transcrip-

tional class. There was a statistically significant difference inexpression according to transcriptional class (P < 0.001).Cases belonging to the mesenchymal transcriptional classhad the highest CD3G expression (19.6), which was signif-icantly higher than that of other classes combined (15.1; P <0.001). Conversely, CD3G expression was significantlylower in the classical transcriptional class compared withall other classes (14.8 vs. 17.1; P < 0.001).

TILs are not associated with prolonged survivalTILs were not associated with prolonged survival in

univariate analyses. We compared the survival of cases withabsent (0) lymphocytes, present (1þ), and abundant (2þ)lymphocytes using a log-rank test (P > 0.05; Fig. 3). Wefound patient age was a highly significant predictor ofsurvival (P < 0.001); however, lymphocytes did not vary

according to age. The mean age of patients with absent (0),present (1þ), and abundant (2þ) lymphocytes was 56.9(SD� 14.6), 57.3 (SD� 12.1), and 54.8 years (SD� 13.1),respectively (P > 0.05). TILs were not a significant predictorof survival in an age-adjusted Cox proportional hazardsmodel (data not shown; P > 0.05). If tumors belonging tothe classical transcriptional class were excluded, there wereno survival differences according to TILs among proneural,neural, and mesenchymal tumors (P > 0.05). Similarly, ifmesenchymal tumors were excluded, there were no survivaldifferences according to TILs among proneural, neural, andclassical tumors (P > 0.05).

DiscussionLymphocytes are present in the stroma of many human

cancers. The histologic finding of TILs is often associatedwith prolonged survival. Tumors with TILs may have dis-tinctive clinicopathologic features or underlying geneticalterations, which could be relevant for future tailoredtherapies.

Although prognostically significant in other cancers, theclinical relevance and genetic associations of TILs in glio-blastoma remain unclear. The TCGA data set offers anopportunity to study the relationship betweenmorphologicfeatures,molecular alterations, and survival in glioblastoma(3, 4). Verhaak and colleagues used TGCA gene expressionprofiles to identify 4 transcriptional classes, whereas inde-pendent studies of genomemethylation uncovered a hyper-methylated, G-CIMPþ subset characterized by IDHmutations and overproduction of the oncometabolite 2-hydroxyglutarate (2-HG; refs. 3, 26, 29). Recent evidencesuggests that glioblastomas within the mesenchymal tran-scriptional class have a better response to immunotherapy,suggesting this class may be more immunogenic (30). Wehypothesized that TILs are differentially distributed withinspecificmorphologic andmolecular classesof glioblastoma.

We found that TILs were not uniformly distributedamong glioblastomas, suggesting some are more capableof eliciting an immune response. Indeed, lymphocytes wereabsent inmore thanhalf, as determinedmorphologically bya panel of TCGA neuropathologists. TILs were stronglyenriched in the mesenchymal transcriptional class of glio-blastoma, which suggests tumors belonging to this subtypeare more immunogenic. Seventy-one percent of tumorswith abundant (2þ) lymphocytes belonged to the mesen-chymal class. No other transcriptional class had more than12% with abundant (2þ) TILs, indicating a fundamentaldifference in transcriptional class with regard to TILs.

Variation of TILs could potentially be accounted forby the extent of necrosis, as themesenchymal class signatureis heavily influenced by the degree of necrosis and thepresence of lymphocytes was associated with necrosis inthis study (31). However, mesenchymal glioblastomaswith abundant (2þ) TILs did not have substantially differ-ent levels of necrosis than those with no TILs. Neither zonalor pseudopalisading necrosis were associated with tran-scriptional class. Moreover, molecular alterations associat-ed with the mesenchymal class are more likely to be

Product-limit survival estimates

1.0

0.8

0.6

0.4

0.2

0.0

+ Censored

0 20 40 60Survival (mo)

Sur

viva

l pro

babi

lity

Lymphocytes 0 1 2

80 100

Figure 3. Kaplan–Meier estimates of survival according to absent (0),present (1þ), or abundant (2þ) lymphocytes. TILs are not associatedwithimproved survival (log-rank P > 0.05).



responsible for the associationwith TILs. Lymphocyteswerestrongly associated with mutations in NF1 and RB1 and atrend was noted for TP53mutations. Mutations ofNF1 andRB1 are characteristic of the mesenchymal transcriptionalclass, and TP53 mutations are common in both the mes-enchymal and proneural subtypes. Interestingly, we foundthat TP53 mutant tumors with abundant (2þ) TILs weremore common in the mesenchymal than the proneuralclass, but this did not reach statistical significance.

We also found that TILs were more common in certainmorphologic subsets of glioblastomas, including thosewithsarcomatous, giant cell, epithelioid, and gemistocytic com-ponents. This was of interest since prior studies have shownthat gliosarcoma, giant cell glioblastoma, and gemistocyticastrocytomas are all characterized by a high frequency ofTP53mutations (32–34).Our own studies using TCGAdataindicated that sarcomatous components in glioblastomawere associated with NF1, RB1, and TP53 mutations; giantcells were associated with TP53 and RB1 mutations; andepithelioid cells were associated with NF1 and RB1 muta-tions (data not shown). We did not find an associationbetween TP53 mutations and gemistocytic cells, likelybecause of inclusion of tumors with low levels of thishistologic finding as compared with prior studies, whichincluded morphologically pure gemistocytic astrocytomasof lower grade. Nonetheless, we found that the presence ofTILs was strongly associated with the mesenchymal tran-scriptional class as well as the mutations and tumormorphologies associated with this gene signature.

We also found that TILs were rare in tumors of theclassical transcriptional class, suggesting that these mayexclude lymphocytes and prevent immune-mediated tumoreradication. Only one case with abundant (2þ) lympho-cytes belonged to the classical transcriptional class (1 of 17).We also noted that TILs were depleted in EGFR-amplifiedglioblastomas, which are frequent in the classical transcrip-tional class of glioblastomas but less common in otherclasses, includingmesenchymal. As small cell glioblastomashave been shown to have a high frequency EGFR amplifi-cation, wewere also encouraged that the histologic presenceof small cells within glioblastomas from the TCGA data setwas associated with TIL depletion (35). Recent evidencesuggests that EGFR activation may repress the adaptiveimmune response, potentially through attenuation ofMHCI and MHCII expression, and therefore explain therelation between EGFR amplification and lymphocytedepletion (36). We also found that TILs were depleted inPTEN-deleted tumors (homozygous). Loss of PTEN hasbeen shown to increase expression of the immunosuppres-sive protein B7 homolog 1 (B7-H1) and therefore mayaccount for the association between homozygous PTENdeletion and lymphocyte depletion (22).

We used CD3G expression data from TCGA to validateour morphologic findings. The CD3G gene encodes a pro-tein that forms the T-cell receptor–CD3 complex and ishighly specific to T lymphocytes and is present in all subsets.The correlationbetween lymphocytes andCD3G expressionwas highly statistically significant (P < 0.001), however, the

strength of the correlation was moderate (0.3). Althoughthe tissue used formolecular analysis by TCGAwas from thesame neoplasm as the tissue used to create slide images, thedistribution of TILs in tumors can be irregular and maycontribute to a weak correlation. In addition, the detectionand evaluation of TILs by molecular and pathologic meth-ods differ considerably and may also lead to a weakenedcorrelation. However, we were encouraged that mutationsassociated with TILs (RB1 and TP53) tended to have higherlevels of CD3G expression, whereas CNAs associated withdepletion of lymphocytes (EGFR amplification and homo-zygous PTEN deletion) had lower CD3G expression.Tumors of the mesenchymal transcriptional class had sig-nificantly higher CD3G expression than all other subtypes,whereas those of the classical subtype had significantlylower expression, which corroborates the associationsuncovered in our morphologic analysis.

Glioblastomas annotated as having zonal necrosis byTCGA neuropathologists did not have significantly higherlevels ofCD3G expression and thosewith pseudopalisadingnecrosis had lower levels of CD3G expression, suggestingthat TILs may indeed represent an antitumor adaptiveimmune response rather than a response to necrosis.

In summary, our analysis of digitized, whole slide hema-toxylin and eosin (H&E)-stained images fromTCGAhad theadvantage of a large number of cases and high-quality,multiplatform molecular analysis. It illustrates the powerof networks such as the TCGA to link multiplatformmolec-ular analysis to histopathologic findings in cancer withimplications for future therapy. The categorical classifica-tion of lymphocytes as 0, 1þ, and 2þ was not optimal asstatistical associations withmolecular and clinical variableswere not as strong. We have not adjusted for multiplecomparisons, as the purpose of the analysis was to identifycommon molecular alterations in glioblastoma related tothe immune response for future tissue-based analyses.However, our analysis was limited to alterations known assignificant and recurrent events in glioblastoma pathogen-esis. Furthermore, we are encouraged that each statisticallysignificant association has a biologic rationale. Finally, thisclassification does not account for the functional activity oflymphocytes or specific lymphocyte subsets. While effectorlymphocytes are thought to mediate the antitumorresponse, regulatory lymphocytes may suppress the cyto-toxic activity of effector lymphocytes. Thus, effector andregulatory lymphocytes may have distinct molecular andhistologic associations. Future studies that examine themolecular correlates of lymphocyte subsets will add sub-stantially to the field of immunotherapy.

Disclosure of Potential Conflicts of InterestJ.H. Saltz is a consultant/advisory board member for Appistry. No

potential conflicts of interest were disclosed by the other authors.

Authors' ContributionsConception and design:W.C. Rutledge, D.A. Gutman, L. Cooper, D.J. BratDevelopment ofmethodology: J. Kong, J. Gao,D.A. Gutman, L. Cooper, C.W. Brennan, D.J. BratAcquisitionofdata (provided animals, acquired andmanagedpatients,provided facilities, etc.): J. Kong, D.A. Gutman, L. Cooper, Y. Park, L.

Rutledge et al.


Scarpace, T. Mikkelsen, M.L. Cohen, R.E. McLendon, N.L. Lehman, C.R.Miller, M.J. Schniederjan, D.J. BratAnalysis and interpretation of data (e.g., statistical analysis, biosta-tistics, computational analysis):W.C. Rutledge, J. Kong, J. Gao, L. Cooper,Y. Park, K.D. Aldape, N.L. Lehman, C.W. Brennan, C.S. Moreno, D.J. BratWriting, review, and/or revision of the manuscript: W.C. Rutledge, J.Kong, J. Gao, D.A. Gutman, L. Cooper, T. Mikkelsen, M.L. Cohen, K.D.Aldape, R.E. McLendon, N.L. Lehman, C.R. Miller, M.J. Schniederjan, C.W.Brennan, C.S. Moreno, D.J. BratAdministrative, technical, or material support (i.e., reporting or orga-nizing data, constructing databases): C. Appin, L. Scarpace, D.J. BratStudy supervision: L. Scarpace, C.S. Moreno, J.H. Saltz, D.J. Brat

AcknowledgmentsThe authors thank the Research Pathology Lab of the Cancer Tissue and

Pathology Shared Resource within the Winship Cancer Institute for theirassistance with whole slide scanning and digital pathology, especially Can-

dace Chisolm. They also thank the leadership and support staff of the TCGAWorking Group and Expert Pathology Committee for their assistance in thegeneration of publicly available data.

Grant SupportThis work was supported by U.S. Public Health Service grants from the

Clinical and Translational Science Award program, NIH, NCRR ULRR025008, TL1 RR025010; the In Silico Center for Brain Tumor ResearchContract ST12-1100 (NCI-SAIC Frederick); and the Winship Cancer CenterSupport Grant (P30 CA138292).

The costs of publication of this article were defrayed in part by thepayment of page charges. This article must therefore be hereby markedadvertisement in accordance with 18 U.S.C. Section 1734 solely to indicatethis fact.

Received February 26, 2013; revised June12, 2013; accepted June 29, 2013;published OnlineFirst July 17, 2013.

References1. Stupp R, Mason WP, van den Bent MJ, Weller M, Fisher B,

Taphoorn MJ, et al. Radiotherapy plus concomitant and adjuvanttemozolomide for glioblastoma. N Engl J Med 2005;352:987–96.

2. Cancer Genome Atlas Research Network. Comprehensive genomiccharacterization defines human glioblastoma genes and core path-ways. Nature 2008;455:1061–8.

3. VerhaakRG,HoadleyKA,PurdomE,WangV,QiY,WilkersonMD, et al.Integrated genomic analysis identifies clinically relevant subtypes ofglioblastoma characterized by abnormalities in PDGFRA, IDH1, EGFR,and NF1. Cancer Cell 2010;17:98–110.

4. Phillips HS, Kharbanda S, Chen R, Forrest WF, Soriano RH, Wu TD,et al. Molecular subclasses of high-grade glioma predict prognosis,delineate a pattern of disease progression, and resemble stages inneurogenesis. Cancer Cell 2006;9:157–73.

5. Hodi FS, O'Day SJ, McDermott DF, Weber RW, Sosman JA, HaanenJB, et al. Improved survival with ipilimumab in patients with metastaticmelanoma. N Engl J Med 2010;363:711–23.

6. Kantoff PW, Higano CS, Shore ND, Berger ER, Small EJ, Penson DF,et al. Sipuleucel-T immunotherapy for castration-resistant prostatecancer. N Engl J Med 2010;363:411–22.

7. Heimberger AB, Sampson JH. Immunotherapy coming of age: whatwill it take to make it standard of care for glioblastoma? Neuro Oncol2011;13:3–13.

8. Clark WH Jr, Elder DE, Guerry Dt, Braitman LE, Trock BJ, Schultz D,et al. Model predicting survival in stage I melanoma based on tumorprogression. J Natl Cancer Inst 1989;81:1893–904.

9. Clemente CG, Mihm MC Jr, Bufalino R, Zurrida S, Collini P, CascinelliN. Prognostic value of tumor infiltrating lymphocytes in the verticalgrowth phase of primary cutaneous melanoma. Cancer 1996;77:1303–10.

10. DunnGP,Dunn IF, CurryWT. Focus on TILs: prognostic significanceoftumor infiltrating lymphocytes in human glioma. Cancer Immun2007;7:12.

11. Laghi L, Bianchi P, Grizzi F, Malesci A. How dense, how intense? Roleof tumour-infiltrating lymphocytes across colorectal cancer stages.Re: Nosho et al. Tumour-infiltrating T-cell subsets, molecular changesin colorectal cancer, and prognosis: cohort study and literature review.J Pathol 2010;222:350–66.

12. Menon AG, Janssen-van Rhijn CM,Morreau H, Putter H, Tollenaar RA,van de Velde CJ, et al. Immune system and prognosis in colorectalcancer: a detailed immunohistochemical analysis. Lab Invest 2004;84:493–501.

13. Ropponen KM, Eskelinen MJ, Lipponen PK, Alhava E, Kosma VM.Prognostic value of tumour-infiltrating lymphocytes (TILs) in colorectalcancer. J Pathol 1997;182:318–24.

14. NoshoK, BabaY, TanakaN, Shima K,HayashiM,Meyerhardt JA, et al.Tumour-infiltrating T-cell subsets, molecular changes in colorectalcancer, and prognosis: cohort study and literature review. J Pathol2010;222:350–66.

15. Galon J, Costes A, Sanchez-Cabo F, Kirilovsky A,Mlecnik B, Lagorce-Pages C, et al. Type, density, and location of immune cells withinhuman colorectal tumors predict clinical outcome. Science 2006;313:1960–4.

16. Zhang L, Conejo-Garcia JR, Katsaros D, Gimotty PA, Massobrio M,Regnani G, et al. Intratumoral T cells, recurrence, and survival inepithelial ovarian cancer. N Engl J Med 2003;348:203–13.

17. Curiel TJ, Coukos G, Zou L, Alvarez X, Cheng P, Mottram P, et al.Specific recruitment of regulatory T cells in ovarian carcinoma fostersimmune privilege and predicts reduced survival. Nat Med 2004;10:942–9.

18. Yang I, Tihan T, Han SJ, Wrensch MR, Wiencke J, Sughrue ME, et al.CD8þ T-cell infiltrate in newly diagnosed glioblastoma is associatedwith long-term survival. J Clin Neurosci 2010;17:1381–5.

19. Safdari H, Hochberg FH, Richardson EP Jr. Prognostic value of roundcell (lymphocyte) infiltration in malignant gliomas. Surg Neurol 1985;23:221–6.

20. Rossi ML, Jones NR, Candy E, Nicoll JA, Compton JS, Hughes JT,et al. The mononuclear cell infiltrate compared with survival in high-grade astrocytomas. Acta Neuropathol 1989;78:189–93.

21. Heimberger AB, Abou-Ghazal M, Reina-Ortiz C, Yang DS, Sun W,Qiao W, et al. Incidence and prognostic impact of FoxP3þregulatory T cells in human gliomas. Clin Cancer Res 2008;14:5166–72.

22. Parsa AT, Waldron JS, Panner A, Crane CA, Parney IF, Barry JJ,et al. Loss of tumor suppressor PTEN function increases B7-H1expression and immunoresistance in glioma. Nat Med 2007;13:84–8.

23. Fecci PE, Ochiai H, Mitchell DA, Grossi PM, Sweeney AE, Archer GE,et al. SystemicCTLA-4blockadeameliorates glioma-induced changesto the CD4þ T cell compartment without affecting regulatory T-cellfunction. Clin Cancer Res 2007;13:2158–67.

24. Fecci PE, Mitchell DA, Whitesides JF, Xie W, Friedman AH, Archer GE,et al. Increased regulatory T-cell fraction amidst a diminished CD4compartment explains cellular immune defects in patients with malig-nant glioma. Cancer Res 2006;66:3294–302.

25. Cerami E, Gao J, Dogrusoz U, Gross BE, Sumer SO, Aksoy BA, et al.The cBio cancer genomics portal: an open platform for exploringmultidimensional cancer genomics data. Cancer Discov 2012;2:401–4.

26. Parsons DW, Jones S, Zhang X, Lin JC, Leary RJ, Angenendt P, et al.An integrated genomic analysis of human glioblastoma multiforme.Science 2008;321:1807–12.

27. Noushmehr H, Weisenberger DJ, Diefes K, Phillips HS, Pujara K,Berman BP, et al. Identification of a CpG island methylator phenotypethat defines a distinct subgroup of glioma. Cancer Cell 2010;17:510–22.

28. Turcan S, Rohle D, Goenka A, Walsh LA, Fang F, Yilmaz E, et al. IDH1mutation is sufficient to establish the glioma hypermethylator pheno-type. Nature 2012;483:479–83.



29. Yan H, Parsons DW, Jin G, McLendon R, Rasheed BA, Yuan W,et al. IDH1 and IDH2 mutations in gliomas. N Engl J Med 2009;360:765–73.

30. Prins RM, Soto H, Konkankit V, Odesa SK, Eskin A, Yong WH, et al.Gene expression profile correlates with T-cell infiltration and relativesurvival in glioblastoma patients vaccinated with dendritic cell immu-notherapy. Clin Cancer Res 2011;17:1603–15.

31. Cooper LA,GutmanDA,ChisolmC,AppinC, Kong J, RongY, et al. Thetumor microenvironment strongly impacts master transcriptional reg-ulators and gene expression class of glioblastoma. Am J Pathol2012;180:2108–19.

32. Reis RM, Konu-Lebleblicioglu D, Lopes JM, Kleihues P, Ohgaki H.Genetic profile of gliosarcomas. Am J Pathol 2000;156:425–32.

33. Meyer-Puttlitz B, Hayashi Y, Waha A, Rollbrocker B, Bostrom J,Wiestler OD, et al. Molecular genetic analysis of giant cell glioblasto-mas. Am J Pathol 1997;151:853–7.

34. Watanabe K, Peraud A, Gratas C, Wakai S, Kleihues P, Ohgaki H. p53and PTEN gene mutations in gemistocytic astrocytomas. Acta Neu-ropathol 1998;95:559–64.

35. Burger PC, Pearl DK, Aldape K, Yates AJ, Scheithauer BW, Passe SM,et al. Small cell architecture–a histological equivalent of EGFR ampli-fication in glioblastoma multiforme? J Neuropathol Exp Neurol2001;60:1099–104.

36. Pollack BP, Sapkota B, Cartee TV. Epidermal growth factor receptorinhibition augments the expression of MHC class I and II genes. ClinCancer Res 2011;17:4400–13.

Rutledge et al.


tumor-infiltrating lymphocytes in glioblastoma are ...drgutman.org/original_papers/38a.pdf · clin...

Documents