Biology of Human Tumors

Stromal CD8þ T-cell Density—A PromisingSupplement to TNM Staging in Non–Small CellLung CancerTomDonnem1,2, SigurdM.Hald2, Erna-ElisePaulsen1,2, ElinRichardsen3,4, SamerAl-Saad3,4,Thomas K. Kilvaer1, Odd Terje Brustugun5,6, Aslaug Helland5,6, Marius Lund-Iversen6,MettePoehl7, 8, 9, KarenEgeOlsen9,10, Henrik J.Ditzel8,11,OlfredHansen8,9, KhalidAl-Shibli12,Yury Kiselev4,13, Torkjel M. Sandanger14, Sigve Andersen1, Francesco Pezzella15,Roy M. Bremnes1,2, and Lill-Tove Busund3,4

Abstract

Purpose: Immunoscore is a prognostic tool defined to quantifyin situ immune cell infiltrates, which appears to be superior to thetumor–node–metastasis (TNM) classification in colorectal can-cer. In non–small cell lung cancer (NSCLC), no immunoscore hasbeen established, but in situ tumor immunology is recognized ashighly important. We have previously evaluated the prognosticimpact of several immunological markers in NSCLC, yielding thedensity of stromal CD8þ tumor-infiltrating lymphocytes (TIL) asthe most promising candidate. Hence, we validate the impact ofstromal CD8þ TIL density as an immunoscore in NSCLC.

Experimental Design: The prognostic impact of stromal CD8þ

TILs was evaluated in four different cohorts from Norway andDenmark consisting of 797 stage I–IIIA NSCLC patients. TheTromso cohort (n¼ 155) was used as training set, and the resultswere further validated in the cohorts from Bodo (n ¼ 169), Oslo

(n ¼ 295), and Denmark (n ¼ 178). Tissue microarrays andclinical routine CD8 staining were used for all cohorts.

Results: Stromal CD8þ TIL density was an independent prog-nostic factor in the total material (n ¼ 797) regardless of theendpoint: disease-free survival (P < 0.001), disease-specific sur-vival (P < 0.001), or overall survival (P < 0.001). Subgroupanalyses revealed significant prognostic impact of stromalCD8þ TIL density within each pathologic stage (pStage). Inmultivariate analysis, stromal CD8þ TIL density and pStage wereindependent prognostic variables.

Conclusions: Stromal CD8þ TIL density has independentprognostic impact in resected NSCLC, adds prognostic impactwithin each pStage, and is a good candidate marker forestablishing a TNM-Immunoscore. Clin Cancer Res; 21(11);2635–43. �2015 AACR.

IntroductionIn non–small cell lung cancer (NSCLC), staging according to

the tumor–node–metastasis (TNM) system and histologic sub-type has been the most relevant clinicopathologic variables forroutine prognostication and treatment (1). TNM summarizesdata on the extent of tumor burden (T), spread to regional lymphnodes (N) and evidence of metastasis (M). The American JointCommittee of Cancer and the Union Internationale Contre leCancer (UICC) define and periodically update the TNM classifi-cation system (2). In lung cancer, the current 7th edition of thelung cancer staging system is based on an initiative undertaken bythe International Association for the Study of Lung Cancer(IASLC; ref. 1). The TNM staging system has shown its value, butprovides incomplete prognostic information as clinical outcomesvary among patients in the same TNM stage.

Hence, new strategies to improve outcome prediction andtreatment selection are warranted. Extensive literature has dem-onstrated highly prognostic impact by in situ immune cell infil-trates in tumors. In colorectal cancer, a combination of the densityof tumor-infiltrating lymphocytes (TIL; CD3þ T cells, pan-lym-phocyte marker; CD45ROþ, memory T cells; CD8þ T cells, cyto-toxic T cells, CTL) and location (central tumor vs. invasive mar-gins) has defined the Immunoscore as a new component for theclassification of colorectal cancer, designated TNM-Immune

1Department of Oncology, University Hospital of North Norway,Tromso, Norway. 2Institute of Clinical Medicine, The Arctic Universityof Norway, Tromso, Norway. 3Department of Clinical Pathology, Uni-versity Hospital of North Norway, Tromso, Norway. 4Institute of Med-ical Biology, The Arctic University of Norway, Tromso, Norway.5Department of Oncology, Oslo University Hospital, The NorwegianRadium Hospital, Oslo, Norway. 6Institute of Clinical Medicine, Univer-sity of Oslo, Oslo, Norway. 7Department of Oncology, Rigshospitalet,Copenhagen, Denmark. 8Department of Oncology,Odense UniversityHospital, Odense, Denmark. 9Institute of Clinical Research, Universityof Southern Denmark, Odense, Denmark. 10Department of Pathology,Odense University Hospital, Odense, Denmark. 11Institute of MolecularMedicine, University of Southern Denmark, Odense, Denmark.12Department of Pathology, Nordland Hospital, Bodo, Norway.13Department of Pharmacy, The Arctic University of Tromso, Tromso,Norway. 14DepartmentofCommunityMedicine,TheArticUniversityofTromso,Tromso,Norway. 15NuffieldDepartmentofClinical LaboratorySciences,UniversityofOxford, JohnRadcliffeHospital,Oxford,UnitedKingdom.

Note: Supplementary data for this article are available at Clinical CancerResearch Online (http://clincancerres.aacrjournals.org/).

R.M. Bremnes and L.-T. Busund contributed equally to this article.

Corresponding Author: Tom Donnem, Department of Oncology/Institute ofClinical Medicine, University Hospital of North Norway/The Arctic University ofNorway, 9038 Tromso, Norway. Phone: 47-91145107; Fax: 47-77626779; E-mail:tom.donnem@uit.no

doi: 10.1158/1078-0432.CCR-14-1905

�2015 American Association for Cancer Research.

ClinicalCancerResearch

www.aacrjournals.org 2635

on July 7, 2018. © 2015 American Association for Cancer Research. clincancerres.aacrjournals.org Downloaded from

Published OnlineFirst February 13, 2015; DOI: 10.1158/1078-0432.CCR-14-1905

(TNM-I). An international consortium has been initiated tovalidate and promote the Immunoscore in the routine clinicalcolorectal cancer setting (3–6).

The CD8 marker can be found on natural killer and dendriticcells, but is predominantly expressed on the surface on cytotoxic Tcells. CD8þ T cells are a crucial component of the cellular immunesystem and are pivotal in cell-mediated antitumor immuneresponses (7, 8). CD8þ T cells undergo a period of massiveexpansion, activation, and differentiation to cytotoxic T cells witheffector functions. Once the pathogenic process is resolved, mosteffector CD8þ cells undergo apoptosis, leaving a subset of long-lived memory cells (9, 10). The association of immune cellinfiltrates and prognosis has been reviewed in different malig-nancies, and in a majority of the studies, memory T cells andcytotoxic T cells are predictive of a favorable clinical outcome (8).In NSCLC, we and others have found CD8þ T cells to havesignificant prognostic effect alone or in combination with othermarkers (11–16).

Although our group has studied more than 100 prognosticmarkers related to in situ tumor immunology (11, 13, 14,17, 18),angiogenesis (19–21), and epithelial–mesenchymal transition(22) in NSCLC, stromal CD8þ score (CD8þ TIL density in thetumor-related stroma) has been one of the markers with thestrongest prognostic impact independent of other clinicopath-ologic variables including TNM classification. Furthermore,cytotoxic CD8þ T cells are of increasing interest in lung canceras novel targeted therapies are aiming to stimulate the immunedefense by blocking mechanisms that may inhibit cytotoxic Tlymphocytes, e.g., programmed death-1 (PD1), PD-ligand 1(PD-L1), and cytotoxic T-lymphocyte–associated antigen 4(CTLA-4; ref. 23).

In the interest of establishing an Immunoscore in resectedNSCLC patients, we evaluated our most promising candidatemarker, stromal CD8þ TIL density, in four different NSCLCcohorts (three Norwegian and one Danish). In the Danish vali-dation set, the impact of tissuemicroarray (TMA) core localization(central vs. peripheral tumor areas) has been explored (24),allowing us to evaluate the impact of stromal CD8þ TILslocalization.

Materials and MethodsStudy design and patients

Patients surgically resected for stage I–IIIA NSCLC at the Uni-versity Hospital of North Norway (UNN, Tromso) and NordlandHospital (NH, Bodo) from 1990 through 2004 were included inthis study (19). In the original cohort, 335 patients were includedfrom these two hospitals and results on immunological markersreported in 2008 (11). However, patients have been restagedaccording to the revised 7th edition of UICC TNM classificationand reclassified according to the new pathologic classification oflung cancer, excluding 11 previous bronchioalveolar carcinomas(25, 26). In addition, both cohorts now include follow-up data asof January 2011, hence leaving 155patients fromTromso and169fromBodo. In this study, the Tromso cohort was used as a trainingcohort to define the optimal prognostic cutoffs for stromal CD8density. In contrast with the exploratory approach in the Tromsocohort, the cutoffs from the Tromso cohort were set as a pre-defined scoring system in the validation cohorts (Bodo, Oslo, andDenmark).

Two additional validation cohorts of consecutive NSCLC stageI–IIIA patients from Oslo University Hospital, the NorwegianRadium Hospital (OUS; 2006–2011, n ¼ 295), and OdenseUniversity Hospital, Denmark (1992–1999, n¼ 178), were used.

Exclusion criteria for patients in either the training or validationcohorts were as follows: (i) inadequate paraffin-embedded fixedtissue blocks; (ii) other malignancy within 5 years before NSCLCdiagnosis; (iii) radiotherapy or chemotherapy before surgery; and(iv) or less than two cores with tumor stroma evaluable. All studycohorts were approved by the Regional Committee for Medicaland Health Research Ethics (Tromso and Bodo: protocol ID:2011/2503; Oslo: protocol ID 2009/1904; Denmark: protocolID: 20080018). The reporting of clinicopathologic variables,survival data, and biomarker expression was conducted in accor-dance with the REMARK guidelines (27).

TMA constructionLung cancer specimens were histologically reviewed by expe-

rienced pathologists. In all cohorts, representative areas of vitaltumor tissue were carefully selected and marked on the hema-toxylin and eosin stain (H&E) slide and sampled for the TMAblocks. Regardless of cohort, the large majority of the coresshowed a mixture of epithelial tumor tissue and tumor-relatedstroma (Fig. 1). In the Danish cohort, designated cores from thetumor center and from the invasive margin were made (Fig. 1B).In the Norwegian cohorts, the localization of the cores (centraltumor vs. invasivemargin) was not noted. TMAs weremade usinga tissue arrayer instrument (Beecher Instruments). The detailedmethodology has been previously reported (19, 24). Tromso andBodo used 0.6-mm cores, whereas Oslo and Denmark used 1.0-mm cores. In all cohorts, multiple 4-mm sections were cut with aMicron microtome before antibody staining for immunohisto-chemical analysis (Table 1).

ImmunohistochemistryCD8þ staining is well validated and in routine clinical use at all

included hospitals.In the Norwegian cohorts, Ventana Benchmark, XT automated

slide stainer (Ventana Medical System), was used for immuno-histochemistry (IHC). For the Tromso/Bodo cohorts, predilutedprimary CD8 antibody from Ventana Medical Systems/Roche

Translational Relevance

The tumor–node–metastasis (TNM) staging system pro-vides the most reliable guidelines for the routine prognosti-cation and treatment of non–small cell lung cancer (NSCLC).However, several in situ immunological studies show tumor-infiltrating lymphocytes to be of vital importance in suppres-singNSCLC development. In colorectal cancer, the presence ofCD8þ cytotoxic T cells seems to be a useful prognostic marker,and large ongoing studies aimat implementing a combinationof Immunoscore with TNM classification, TNM-I, in a routineclinical setting. This study validates one of themost promisingin situ immunological markers in NSCLC, stromal CD8þ

density, in tumor tissues of four different cohorts of Norwe-gian and Danish NSCLC patients. We conclude that stromalCD8þ density adds significant prognostic impact to the estab-lished TNM staging in stage I–IIIA NSCLC, hence is a prom-ising candidate in establishing an NSCLC TNM-I.

Donnem et al.

Clin Cancer Res; 21(11) June 1, 2015 Clinical Cancer Research2636

on July 7, 2018. © 2015 American Association for Cancer Research. clincancerres.aacrjournals.org Downloaded from

Published OnlineFirst February 13, 2015; DOI: 10.1158/1078-0432.CCR-14-1905

(clone C8/144, 760-4250) was used, and for the Oslo cohort,prediluted primary CD8 antibody from Roche (clone SP57, 790-4460) was used. The slides were baked for 60�C overnight,deparaffinized in EZ Prep, and submitted to heat-induced epitoperetrieval. Cell conditioning-1 protocol (CC1Mild) for 30minutesat 95�C epitope retrieval was used for Tromso/Bodo cohortsand CC1 Standard for 60 minutes at 95�C for the Oslo cohort.The endogenous peroxidase activity was quenched using 3%hydrogen peroxide. The primary antibody was applied and theslides were incubated for 32 minutes at 36�C, followed bywashing in buffer and visualization. Visualization method wascarried out with the Iview DAB Detection Kit for Tromso/Bodo cohorts and the ultraview DAB Detection Kit for the Oslocohort. 3,30-diaminobenzidine was used as a chromogen for allcohorts.

In the Danish cohort, Ventana Benchmark, Ultra automatedslide stainer (Ventana Medical System), was used. The primaryantibody CD8 was from Dako M7103 (clone C8/144B). Theslides were baked at 75�C for 4 minutes, deparaffinized in EZPrep, and submitted to heat-induced epitope retrieval. CC1 Mildfor 36 minutes at pH 8.5 and 99�C was used. The endogenousperoxidase activity was quenched using 3% hydrogen peroxide.The primary antibody dilution was applied and the slides wereincubated for 32 minutes at 36�C, followed by washing in bufferand visualization using theOptiViewDABDetectionKitwith 3,30-diaminobenzidine as a chromogen.

Negative and positive staining controls were placed simulta-neously with all TMAs. In all cohorts, slides were counterstainedwith hematoxylin to visualize the nuclei. For each cohort, stainingwas done in one single experiment.

To validate different IHC procedures, new TMA slides fromDenmark were stained and reevaluated with the same IHC pro-cedure as for the Oslo cohort.

Scoring of IHCIn all cohorts, samples were anonymized and independently

scored by two experienced pathologists in the Norwegiancohorts (K. Al-Shibli and S. Al-Saad in the Tromso and Bodocohorts; S. Al-Saad and E. Richardsen in the Oslo cohort) andone experienced pathologist (K.E. Olsen) and one oncologist(M. Poehl) in the Danish cohort. The second scoring of theDanish cohort (validating the staining procedure/scoring) wasdone by S. Al-Saad and E. Richardsen. In case of disagreement,the slides were reexamined and the observers reached a con-sensus. By light microscopy, the tissue sections were scored forthe degree of infiltration of CD8þ lymphocytes. The percen-tages of CD8þ lymphocytes compared with the total amount ofnucleated cells in the stromal compartments were assessed.Based on experiences from one of our previous studies (11),scoring cutoff points at 5%, 25%, or 50% for each core accordingto the degree of cell densitieswere used in the training set (Tromsocohort): 0% to 5% ¼ 0, 5% to 25% ¼ 1, 25% to 50% ¼ 2, and

Figure 1.A, low, intermediate, and high stromalCD8þ density in the Tromso (6-mmcores), Bodo (6-mm cores), Oslo(1.0-mm cores), and the Danish cohort(1.0-mm cores). B, in the Danishcohort, designated cores from thetumor center (blue) and from theinvasive margin (yellow) were made.In the Norwegian cohorts, thelocalization of the cores (central tumorvs. invasive margin) was not noted. C,in all cohorts, the large majority of thecores showed a mixture of epithelialtumor tissue (red) and tumor-relatedstroma (green). Within each core, thepercentages of CD8þ lymphocytescompared with the total amount ofnucleated cells in the stromalcompartments (green areas) wereassessed and defined as stromalCD8þ density. Low density: �25%;intermediate density: >25% to � 50%;high density: >50%.

CD8þ T Cells in NSCLC

www.aacrjournals.org Clin Cancer Res; 21(11) June 1, 2015 2637

on July 7, 2018. © 2015 American Association for Cancer Research. clincancerres.aacrjournals.org Downloaded from

Published OnlineFirst February 13, 2015; DOI: 10.1158/1078-0432.CCR-14-1905

Table

1.Clinicopatho

logic

characteristicsin

each

coho

rtan

din

thetotalmaterial

Training

coho

rtTromso

,Norw

ay(U

NN)

Validationco

hort

Bodo,

Norw

ay(N

H)

Validationco

hort

Oslo,

Norw

ay(O

US)

Validationco

hort

Oden

se,Den

mark

Total

coho

rt

Num

ber

ofpatients

155

169

295

178

797

Tim

eofinclusion

1990–2005

1990–2005

2006–2011

1992–

1999

1990–2

011

Med

ianag

ein

years

66.6

(ran

ge,

38.8–8

4.7)

67.3(ran

ge,

27.5–8

2.1)

66.5

(ran

ge,

39.1–8

4.1)

64.1(ran

ge,

39.4–8

2.4)

65.2(ran

ge,

27.5–8

4.7)

Last

follo

w-up

Janu

ary20

11Janu

ary20

11March

2014

Janu

ary20

10Janu

ary20

10–M

arch

2014

Med

ianfollo

w-uptimeofthe

survivors

inmonths

110.7

(ran

ge,

76.3–222

.2)

102.8(ran

ge,

72.9–2

34.0)

52.4

(ran

ge,

34.9–9

9.4)

161.7

(ran

ge,

120.6–210.8)

65.5(ran

ge,

34.9–3

4.0)

Sco

ring

system

Exp

loratory

Predefi

ned

Predefi

ned

Predefi

ned

Predefi

ned

Different

cutoffsystem

tested

,and

both

averag

ean

dmaxim

umscore

evalua

ted.

Optimal

progno

stic

impactcutoffdefi

ned

andused

inallcoho

rts.

Maxim

umscore

Maxim

umscore

Maxim

umscore

Maxim

umscore

Low

den

sity:�

25%

Low

den

sity:�2

5%Lo

wden

sity:�2

5%Lo

wden

sity:�2

5%Interm

ediate

den

sity:

>25%

–�50

%Interm

ediate

den

sity:

>25%

–�50

%Interm

ediate

den

sity:

>25%

–�50

%Interm

ediate

den

sity:

>25%

–�50

%Highden

sity:>5

0%

Highden

sity:>5

0%

Highden

sity:>

50%

Highden

sity:>

50%

End

points

DSS,D

FS,O

SDSS,D

FS,O

SDSS,D

FS,O

SDSS,D

FS,O

SDSS,D

FS,O

STMAco

resize

0.6

mm

0.6

mm

1.0mm

1.0mm

0.6–1.0

mm

CD8þIHCstaining

Clinical

routine

Clinical

routine

Clinical

routine

Clinical

routine

Clinical

routine

Num

ber

ofco

resfrom

each

patients

Four

Four

Variable,m

inim

umofthree

Four,twofrom

centralan

dtw

ofrom

invasive

margins

Variable,b

utat

leastpotentially

threeev

alua

ble

from

each

patient.

TMAthickn

ess

4mm

4mm

4mm

4mm

4mm

Distributionstromal

CD8þ

den

sity

score

(%)1:low,

2:interm

ediate,3

:high

1:35

(23)

1:47(28)

1:77

(27)

1:50

(28)

1:20

9(27)

2:80(52)

2:90(53)

2:156(53)

2:77

(43)

2:403(52)

3:39

(25)

3:29

(17)

3:56

(19)

3:47(26)

3:171(22)

Missing

:1Missing

:3Missing

:6Missing

:4Missing

:14

Donnem et al.

Clin Cancer Res; 21(11) June 1, 2015 Clinical Cancer Research2638

on July 7, 2018. © 2015 American Association for Cancer Research. clincancerres.aacrjournals.org Downloaded from

Published OnlineFirst February 13, 2015; DOI: 10.1158/1078-0432.CCR-14-1905

>50% ¼ 3. In this training cohort, only 2 patients had an averageor maximum score evaluated as 0; hence, scores 0 and 1 weremerged and the following cutoffs were used in the validation sets:low density: �25%; intermediate density: >25% to �50%; highdensity: >50%. These percentages are also easy to follow andreproduce in daily practice. Both the average and the maximumscore from each patient were assessed in the training set, resultingin an optimal significant prognostic impact (P¼ 0.004, endpointDSS), when using a maximum score. Hence, the maximum scorewas used in the validation sets.

Statistical analysisThe Kaplan–Meier method was used to analyze the association

between marker expression and survival endpoints. Disease-spe-cific survival (DSS)was determined from the date of surgery to thetime of lung cancer death, overall survival (OS) from the date ofsurgery to death, and disease-free survival (DFS) as the time fromthe date of surgery to time of first relapse. The statistical signif-icance of differences between survival curveswas assessedwith thelog-rank test. The c2 test was used to examine the associationbetween molecular marker expression and various clinicopatho-logic variables. In the second IHC score of theDanish cohort, eachobserver was compared for interobserver reliability by use of atwo-way random effect model with absolute agreement defini-tion. The intraclass correlation coefficient (reliability coefficient,ICC) was obtained from these results. Variables with a P value of<0.25 from univariate analyses entered the multivariate analysis,using the Cox proportional hazards model (backward stepwise,probability for stepwise entry and removal was set at 0.05 and0.10). P values < 0.05 were considered statistically significant.

ResultsPatient characteristics

Demographic, clinical, and histopathologic variables for eachcohort and in the total material are listed in Table 1 and Supple-mentary Tables S1–S3. Median age in the total material was 65.2(range, 27.5–84.7) years (Table 1). The majority of patients weremen (64%), but there was a higher rate of women in the cohortwith the most recent inclusion period (Oslo, 49% women). The

Oslo cohort also had the highest frequency of adenocarcinomas(63%) compared with 48% in the total population.

Marker expression and correlationsThe stromal CD8þ TIL density score distribution in each cohort

is shown in Table 1. There was no significant difference in thestromal CD8þ density score (low vs. intermediate vs. high)between the different cohorts (P ¼ 0.18). In the total material,there was no significant correlation between stromal CD8þ den-sity and the clinicopathologic variables.

Univariate analysesAs shown in Supplementary Table S1, tumor differentiation

(P < 0.001), pStage (P < 0.001), T stage (P < 0.001), and N stage(P < 0.001) were significant prognostic indicators for DSS in thetotal material (Supplementary Tables S2 and S3 show theresults with DFS and OS as endpoints). Figure 2 presents theprognostic impact of stromal CD8þ density in the total materialwith three different endpoints: DFS (P < 0.001), DSS (P <0.001), and OS (P < 0.001). Five-year survival rates for stromalCD8þ high, intermediate, and low score in the total materialwere as follows: DFS: 68%, 59%, and 43%; DSS: 74%, 63%, and49%; OS: 61%, 50%, and 41%, respectively. The prognosticimpact in each cohort with DFS and DSS as endpoint is shownin Fig. 3. Using OS as endpoint, the prognostic impact ofstromal CD8 density in each cohort was as follows: Tromso(P ¼ 0.004), Bodo (P ¼ 0.28), Oslo (P ¼ 0.22), and Odense(P ¼ 0.11), all showing the same tendency as with DFS and DSSas endpoints.

The frequencies of recurrences, lung cancer–related deaths, andoverall deaths in each cohort were as follows: Tromso: 51%, 45%,and 79%; Bodo: 42%, 41%, and 82%;Oslo: 37%, 28%, and 41%;Denmark: 52%, 51%, and 82%, respectively.

When stratified by histology and pStage, CD8þ density hadsignificant prognostic impact in all subgroups with DSS as end-point (Table 2). Using OS as endpoint, all subgroups showedsignificant prognostic impact, except only showing a tendency inpStage II (P ¼ 0.09), adenocarcinomas (P ¼ 0.09), and large-cellcarcinoma (LCC; P ¼ 0.06). Using DFS as endpoint, stromal

Figure 2.DFS (A), DSS (B), and OS (C) curves according to low, intermediate, and high stromal CD8þ T-cell density in the total material. The maximum score for each patientwas used.

CD8þ T Cells in NSCLC

www.aacrjournals.org Clin Cancer Res; 21(11) June 1, 2015 2639

on July 7, 2018. © 2015 American Association for Cancer Research. clincancerres.aacrjournals.org Downloaded from

Published OnlineFirst February 13, 2015; DOI: 10.1158/1078-0432.CCR-14-1905

CD8þ density showed significant prognostic impact in all sub-groups, except within pStage I (P ¼ 0.07) and LCC (P ¼ 0.29).

In the Danish population, we were able to stratify the stromalCD8 density according to the central parts of the tumor versus thetumor periphery/invasive margins. As shown in SupplementaryFig. S1, the prognostic impact of stromal CD8þ density with DSSas endpoint was highly significant at the tumor margins (P ¼

0.008) in contrast with the central tumor (P ¼ 0.67). CD8þ

density in the tumor periphery/invasive margins showed 5-yearsurvival rates of 80% (high density), 61% (intermediate density),and 42% (low density). The prognostic impact of stromal CD8density in the invasive margin was also significant for DFS(P ¼ 0.021) and tended toward significance for OS (P ¼ 0.09).

Multivariate analysisResults from the multivariate analysis are presented in Table 3.

All clinicopathologic variables with P < 0.25 and stromal CD8þ

density from the univariate analyses were entered into the mul-tivariate analysis. However, as pStage is based on a combinationof T stage and N stage, multivariate analysis was done includingpStage and not T stage or N stage (Table 3). Pathologic stage,differentiation, and stromal CD8þ density were identified asindependent prognostic factors regardless of endpoints (DSS,DFS, or OS).

Validating the Danish cohortTMAs from the Danish cohort were reevaluated with the same

IHCprocedure and scoredby the samepathologists as for theOslocohort. The interobserver reliability coefficient between thescorers was good (ICC ¼ 0.89, P > 0.001). At a patient level(maximum score for each patient), the correlation coefficient, r,was 0.68, kappa value was 0.57, and P < 0.001 when comparingthe results from the first and the second staining/scoring. Theprognostic impact of stromal CD8þ density in the validation ofthe Danish cohort, using the same predefined cutoffs, was similarto the original results: DFS, P¼ 0.009 versus P¼ 0.024; DSS, P¼0.005 versus P ¼ 0.009; OS, P ¼ 0.066 versus P ¼ 0.11.

DiscussionIn this large study of 797 NSCLC patients, stromal CD8þ TIL

density was a strong independent prognostic factor for DFS, DSS,and OS in the total material. There is a significant prognosticimpact or at least the same trend in all four cohorts. Subgroupanalyses revealed a significant prognostic impact within eachpStage. In the Danish cohort, marker localization was furtheraddressed and the prognostic impact of stromal CD8þ TILs wasrelated to an invasive margin/tumor periphery location.

The strength of this study lies in the number of cohorts andvalidated NSCLC patients. The cutoff was predefined for valida-tion sets. Cutoff points of 25% and 50% are easy to follow andreproduce in daily practice, supported by the fact that the distri-bution of low, intermediate, and high score was similar in allcohorts and also independent of the core size. Finally, CD8 IHCstaining has limited technical challenges, is well validated, andestablished as a routine marker at most pathologic departments.

Though CD8 is a routine clinical marker, we reevaluated theDanish population to show that the results were independent ofthe staining and scoring procedure. Some variation in the score isexpected due to the fact that new slides provided are from adifferent depth of the tumor block. Even though the interobservercorrelation coefficient was good, this may also to a limited degreeaffect the score. However, reassuringly, there was a highly signif-icant correlation between the first and the second Danish stainingprocedure/scoring as well as a similar prognostic impact. How-ever, as the prognostic impact of the intermediate group variesmost between the different cohorts, one may consider a two-category approach in a clinical setting.

Figure 3.DFS and DSS curves according to low, intermediate, and high stromal CD8þ

T-cell density in the four different cohorts; A and B, University Hospitalof North Norway (Tromso, n ¼ 155); C and D, Central Nordland Hospital(Bodo, n ¼ 169); E and F, Oslo University Hospital (Oslo, n ¼ 295); G and H,Danish cohort (Odense University Hospital, n ¼ 178). The maximumscore for each patient was used.

Donnem et al.

Clin Cancer Res; 21(11) June 1, 2015 Clinical Cancer Research2640

on July 7, 2018. © 2015 American Association for Cancer Research. clincancerres.aacrjournals.org Downloaded from

Published OnlineFirst February 13, 2015; DOI: 10.1158/1078-0432.CCR-14-1905

The size of the TMA coresmay also be an issue, but showing thesame trends regardless of core size is reassuring as we are evalu-ating the percentages of lymphocytes compared with the totalamount of nucleated cells in the stromal compartment.

In the colorectal TNM-I, several immune cell combinationshave been tried and the most promising combination so farhas included CD8 and CD3 with scores from both the centralparts of the tumor and the invasive margins (3–6). We have

Table 2. The prognostic impact of stromal CD8þ density in the total material stratified by histology and pathologic stage (DSS, univariate analyses, log-rank test);the maximum score for each patient was used

Total (n ¼ 797, 1990–2011)Number (%) 5-year DSS (%) Median (Months) P

Stromal CD8þ density, total material <0.001Low CD8þ 209 (27) 49 57.6Intermediate CD8þ 403 (51) 63 127.4High CD8þ 171 (22) 74 191.1

HistologySquamous cell carcinoma 0.006Low CD8þ 99 (28) 51 64.1Intermediate CD8þ 172 (49) 71 NRHigh CD8þ 78 (23) 77 NR

Adenocarcinoma 0.041Low CD8þ 96 (26) 48 57.1Intermediate CD8þ 205 (54) 57 75.4High CD8þ 74 (20) 68 189.6

Large cell carcinoma 0.025Low CD8þ 14 (24) 32 54.0Intermediate CD8þ 26 (44) 61 98.1High CD8þ 19 (32) 83 NR

Pathologic stageStage I 0.028Low CD8þ 99 (25) 65 175.2Intermediate CD8þ 211 (53) 74 174.9High CD8þ 88 (22) 85 191.1

Stage II 0.023Low CD8þ 80 (30) 40 41.5Intermediate CD8þ 128 (47) 58 98.1High CD8þ 63 (23) 61 142.1

Stage IIIA 0.012Low CD8þ 30 (26) 18 22.4Intermediate CD8þ 64 (56) 40 43.1High CD8þ 20 (18) 68 NR

NOTE: Bold numbers are significant results.Abbreviation: NR, not reached.

Table 3. Results of Cox regression analysis summarizing significant independent prognostic factors in the total material (n¼ 797) using three different endpoints;clinicopathologic variables with P < 0.25 and stromal CD8þ density from univariate analyses were included, and the maximum score for each patient was used

DSS DFS OSPrognostic factor HR (95% CI) P HR (95% CI) P HR (95% CI) P

Pathologic stage <0.001a <0.001a <0.001a

I 1.00 1.00 1.00II 1.99 (1.52–2.61) <0.001 1.89 (1.47–2.44) <0.001 1.49 (1.22–1.82) <0.001IIIA 3.19 (2.27–4.49) <0.001 2.90 (2.09–4.01) <0.001 2.39 (1.83– 3.13) <0.001

Differentiation 0.001a 0.001a 0.029a

Well 1.00 1.00 1.00Moderate 1.13 (0.73–1.74) 0.58 1.39 (0.91–2.13) 0.13 1.03 (0.76–1.38) 0.84Poor 1.90 (1.23– 2.93) 0.004 2.10 (1.37–3.24) 0.001 1.32 (0.97–1.78) 0.14

Histology 0.005a 0.001a NISCC 1.00 1.00AC 1.63 (1.021–2.60) 0.041 1.35 (0.89–2.04) 0.16LCC 1.10 (0.69–1.76) 0.69 0.84 (0.55–1.27) 0.39

Age, years NI NS�65 1.00>65 1.48 (1.22–1.77) <0.001

Stromal CD8þ density <0.001a <0.001a <0.001a

High 1.00 1.00 1.00Intermediate 1.48 (1.05–2.09) 0.026 1.37 (0.99–1.88) 0.053 1.26 (0.99–1.60) 0.065Low 2.31 (1.61–3.31) <0.001 2.27 (1.63–3.18) <0.001 1.74 (1.34–2.26) <0.001

NOTE: Bold numbers are significant results.Abbreviations: AC, adenocarcinoma; CI, confidence interval; NI, not included; NS, not significant; SCC, squamous cell carcinoma.aOverall significance as a prognostic factor. Age and gender were only included in OS analysis. Gender did not become significant.

CD8þ T Cells in NSCLC

www.aacrjournals.org Clin Cancer Res; 21(11) June 1, 2015 2641

on July 7, 2018. © 2015 American Association for Cancer Research. clincancerres.aacrjournals.org Downloaded from

Published OnlineFirst February 13, 2015; DOI: 10.1158/1078-0432.CCR-14-1905

previously addressed the issue of localization by examining theprognostic impact of CD8þ TILs within the cancer cell nests(epithelial lymphocytes) and in the tumor-related stroma (stro-mal lymphocytes; ref. 11). We observed stromal CD8þ densityto have the most promising prognostic impact. We did not,however, distinguish between stroma centrally in the tumorversus at the tumor periphery/invasive margins. Herein, wewere not able to address this issue in the three Norwegiancohorts as the stromal CD8þ density localization was notregistered. However, this could be examined in the Danishvalidation set because the localizations were recorded as centraltumor versus tumor periphery/invasive margins (24). In ourprevious study, we found the prognostic impact of stromalCD8þ density to be superior to CD8þ density in the cancernests. In this study, the Danish cohort shows that the greatestprognostic impact of stromal CD8þ TIL density was observed inthe tumor periphery/invasive margins. An obvious question is:would the prognostic impact of the total material would furtherimprove if the stromal CD8þ density had been scored only atthe invasive margins? In our previous study, we used an averagevalue for the cores of each patient (11). In our test set, weobserved that a maximum score resulted in a slightly improvedprognostic impact compared with the average score. The coresin the Norwegian cohorts are a mix from the central and theperiphery of the tumors, and in most cases, we expect one corefrom each patient to originate from the tumor periphery. Thismay explain why the maximum score appears to be the bestapproach. Nevertheless, CD8þ TIL localization in the tumorenvironment is an important matter that needs to be fullyaddressed in a prospective study.

Besides the localization of TILs, a pivotal question is whichcombination of immune cells should be included in an NSCLCimmunoscore. The prognostic impact of other immune cells,such as B cells, natural killer cells, myeloid-derived suppressorcells, macrophages, regulatory T cells, and subsets of T-helperpopulations (Th2, Th17), has been shown to differ dependinglargely on cancer type and stage (8). In contrast, the presence ofT cells, cytotoxic T cells, Th1 cells, and memory T cells has beenshown to be strongly associated with beneficial clinical out-come for most cancer types (8). In our previous studies, wehave explored most of these immune cells and found CD8þ

cells as one of the most promising candidates (11, 13,14,17,18). In this study, we have focused on this marker to explore itsrobustness with the intention to include it in a subsequentprospective clinical study on NSCLC TNM-I. Based on thecolorectal cancer TNM-I and previous NSCLC tumor in situimmunology studies, there are other promising immune mar-kers that should also be taken into account with respectto future prospective NSCLC TNM-I studies. These are CD4(T-helper cell), CD45ROþ (memory T cells), and CD3 (panT-cell marker; refs. 3–6, 8, 11, 17, 18, 28–30).

To establish an NSCLC TNM-I classification, the immunemarkers included need to add prognostic or predictive impact ina clinical setting. In this study, pStage showed 5-year survival ratesof 74% (stage 1), 53% (stage II), and 39% (stage IIIA), whereasstromal CD8þ density demonstrates a comparable significantprognostic impact with 5-year DSS rates of 74% (high density),63% (intermediate density), and 49% (low density). In theDanish validation set, the stromal CD8þ density in the tumorperiphery/invasive margins showed 5-year survival rates of 80%(high density), 61% (intermediate density), and 42% (low den-

sity). More importantly, CD8þ density appears to be a significantprognostic marker within each pStage (Table 2). For instance,among stage IIIA patients in the total material, the 5-year DSS isonly 18% in lowCD8þ density group versus 68% in patients withhigh stromal CD8þ density, demonstrating a large variationamong patients within the same pStage. Together with pStage,histology has been themost important clinicopathologic variablein clinical NSCLC decision making. The prognostic impact ofstromal CD8þ density is also significant across all histologicsubgroups using DSS as endpoint. The same tendency is seenwith DFS and OS as endpoint.

The predictive value of CD8þ has not been evaluated in thisstudy, but as cytotoxic CD8þT cells are essential in novel therapiestargeting the immune system, e.g., by blocking CD8þ T cell–related ligands (PD-L1 and PD-L2) and receptors (PD1 andCTLA-4), thiswouldbeof great interest for further investigation (23, 31).As indicated by our previous study (14), the potential predictiveimpact of stromal CD8þ density in an adjuvant NSCLC radio-therapy setting should also be examined. Further, in a recentreview, it was indicated that the presence of immune cells largelyreflected the underlying biology (3). Hence, immune biomarkerswill be both prognostic and predictive more often than non-immuno markers as many treatment modalities affect theimmune system.

Even though the stromal CD8þ density seems to be reliable,reproducible, clinically relevant, and biologically meaningful,some obstacles must be considered before a prospective clinicalstudy is initiated. TMAs as used in this study are mainlydesigned to reveal prognostic impact in large populations,whereas the sensitivity and specificity for each patient arecrucial in a clinical setting. To better reflect the tumor locali-zation on an individual level, there are some advantages withwhole slides as for instance the localization of the cytotoxic Tlymphocytes in the central tumor versus invasive margins. Thescoring system should be optimized, and the utilization ofdigital pathology and image analysis software may be advan-tageous. Further, we have been studying stage I–IIIA NSCLCpatients in which resected material was largely accessible. Forthe majority of NSCLC patients (stage IIIB/IV), only biopsyspecimens will be available. In the latter group, the prognosticimpact of stromal CD8þ density is to our knowledge notclarified and further studies need to be done.

In colorectal cancer, a worldwide task force has started tovalidate TNM-I in a routine clinical setting, with a goal to imple-ment Immunoscore in colorectal cancer staging. We propose thata similar concept may be useful for NSCLC prognostication andtreatment. Stromal CD8þ cytotoxic T-cell density seems to be agood candidate marker and should be included in prospectiveclinical trials in the effort of establishing an Immunoscore forNSCLC.

Disclosure of Potential Conflicts of InterestH.J. Ditzel reports receiving a commercial research grant from Genmab. No

potential conflicts of interest were disclosed by the other authors.

Authors' ContributionsConception and design: T. Donnem, S. Al-Saad, S. Andersen, R.M. Bremnes,L.-T. BusundDevelopment of methodology: T. Donnem, S.M. Hald, S. Al-Saad,R.M. Bremnes, L.-T. Busund

Clin Cancer Res; 21(11) June 1, 2015 Clinical Cancer Research2642

Donnem et al.

on July 7, 2018. © 2015 American Association for Cancer Research. clincancerres.aacrjournals.org Downloaded from

Published OnlineFirst February 13, 2015; DOI: 10.1158/1078-0432.CCR-14-1905

Acquisition of data (provided animals, acquired and managed patients,provided facilities, etc.): T. Donnem, S.M. Hald, E.-E. Paulsen, S. Al-Saad,O.T. Brustugun, A. Helland, M. Poehl, K.E. Olsen, H.J. Ditzel, O. Hansen,S. Andersen, R.M. Bremnes, L.-T. BusundAnalysis and interpretation of data (e.g., statistical analysis, biostatistics,computational analysis): T. Donnem, S.M. Hald, E.-E. Paulsen, E. Richardsen,S. Al-Saad, T. Kilvaer, O.T. Brustugun, M. Lund-Iversen, M. Poehl, K.E. Olsen,S. Andersen, R.M. BremnesWriting, review, and/or revision of the manuscript: T. Donnem, S.M. Hald,E.-E. Paulsen, E. Richardsen, S. Al-Saad, T. Kilvaer, O.T. Brustugun, A. Helland,M. Lund-Iversen, M. Poehl, H.J. Ditzel, O. Hansen, Y. Kiselev, T.M. Sandanger,S. Andersen, F. Pezzella, R.M. Bremnes, L.-T. BusundAdministrative, technical, or material support (i.e., reporting or organizingdata, constructingdatabases): T.Donnem, S.M.Hald, E. Richardsen, S. Al-Saad,

T. Kilvaer, O.T. Brustugun, A. Helland, K.E. Olsen, H.J. Ditzel, Y. Kiselev,S. Andersen, R.M. Bremnes, L.-T. BusundStudy supervision: T. Donnem, R.M. Bremnes, L.-T. Busund

Grant SupportThis study was financially supported by the Norwegian Cancer Society and

Northern Norway Health Region Authority.The costs of publication of this articlewere defrayed inpart by the payment of

page charges. This article must therefore be hereby marked advertisement inaccordance with 18 U.S.C. Section 1734 solely to indicate this fact.

Received July 25, 2014; revised January 15, 2015; accepted February 1, 2015;published OnlineFirst February 13, 2015.

References1. Detterbeck FC, Postmus PE, Tanoue LT. The stage classification of lung

cancer: diagnosis and management of lung cancer, 3rd ed: AmericanCollege Of Chest Physicians evidence-based clinical practice guidelines.Chest 2013;143:e191S–e210S.

2. Sobin LH, Compton CC. TNM seventh edition: what's new, what's chan-ged: communication from the international union against cancer and theAmerican Joint Committee On Cancer. Cancer 2010;116:5336–9.

3. Angell H, Galon J. From the immune contexture to the immunoscore: therole of prognostic and predictive immune markers in cancer. Curr OpinImmunol 2013;25:261–7.

4. Galon J, Pages F, Marincola FM, Angell HK, ThurinM, Lugli A, et al. Cancerclassification using the immunoscore: a worldwide task force. J Transl Med2012;10:205.

5. Galon J, Angell HK, Bedognetti D,Marincola FM. The continuum of cancerimmunosurveillance: prognostic, predictive, and mechanistic signatures.Immunity 2013;39:11–26.

6. Galon J,Mlecnik B, BindeaG, Angell HK, Berger A, Lagorce C, et al. Towardsthe introduction of the `Immunoscore' in the classification of malignanttumours. J Pathol 2014;232:199–209.

7. BarryM,BleackleyRC.Cytotoxic T lymphocytes: all roads lead todeath.NatRev Immunol 2002;2:401–9.

8. FridmanWH, Pages F, Sautes-Fridman C, Galon J. The immune contexturein human tumours: impact on clinical outcome. Nat Rev Cancer 2012;12:298–306.

9. Klebanoff CA, Gattinoni L, Restifo NP. CD8þ T-cell memory in tumorimmunology and immunotherapy. Immunol Rev 2006;211:214–24.

10. Restifo NP, Dudley ME, Rosenberg SA. Adoptive immunotherapy forcancer: harnessing the T cell response. Nat Rev Immunol 2012;12:269–81.

11. Al-Shibli KI, Donnem T, Al-Saad S, Persson M, Bremnes RM, Busund LT.Prognostic effect of epithelial and stromal lymphocyte infiltration in non-small cell lung cancer. Clin Cancer Res 2008;14:5220–7.

12. Dieu-Nosjean MC, Antoine M, Danel C, Heudes D, Wislez M, Poulot V,et al. Long-term survival for patients with non-small-cell lung cancer withintratumoral lymphoid structures. J Clin Oncol 2008;26:4410–7.

13. Donnem T, Al-Shibli K, Andersen S, Al-Saad S, Busund LT, Bremnes RM.Combination of low vascular endothelial growth factor A (VEGF-A)/VEGFreceptor 2 expression and high lymphocyte infiltration is a strong andindependent favorable prognostic factor in patients with nonsmall celllung cancer. Cancer 2010;116:4318–25.

14. Hald SM, Bremnes RM,Al-Shibli K, Al-Saad S, Andersen S, StenvoldH, et al.CD4/CD8 co-expression shows independent prognostic impact in resectednon-small cell lung cancer patients treated with adjuvant radiotherapy.Lung Cancer 2013;80:209–15.

15. Hiraoka K, Miyamoto M, Cho Y, Suzuoki M, Oshikiri T, Nakakubo Y, et al.Concurrent infiltration by CD8þ T cells and CD4þ T cells is a favourableprognostic factor in non-small-cell lung carcinoma. Br J Cancer 2006;94:275–80.

16. Kawai O, Ishii G, Kubota K, Murata Y, Naito Y, Mizuno T, et al. Predom-inant infiltration of macrophages and CD8(þ) T Cells in cancer nests is asignificant predictor of survival in stage IV nonsmall cell lung cancer.Cancer 2008;113:1387–95.

17. Al-Shibli K, Al-Saad S, Donnem T, Persson M, Bremnes RM, Busund LT.The prognostic value of intraepithelial and stromal innate immunesystem cells in non-small cell lung carcinoma. Histopathology 2009;55:301–12.

18. Al-Shibli K, Al-Saad S, Andersen S, Donnem T, Bremnes RM, Busund LT.The prognostic value of intraepithelial and stromal CD3-, CD117- andCD138-positive cells in non-small cell lung carcinoma. APMIS 2010;118:371–82.

19. Donnem T, Al-Saad S, Al-Shibli K, Delghandi MP, Persson M, Nilsen MN,et al. Inverse prognostic impact of angiogenic marker expression in tumorcells versus stromal cells in non small cell lung cancer. Clin Cancer Res2007;13:6649–57.

20. Donnem T, Al-Saad S, Al-Shibli K, Andersen S, Busund LT, Bremnes RM.Prognostic impact of platelet-derived growth factors in non-small cell lungcancer tumor and stromal cells. J Thorac Oncol 2008;3:963–70.

21. Donnem T, Al-Saad S, Al-Shibli K, Busund LT, Bremnes RM. Co-expression of PDGF-B and VEGFR-3 strongly correlates with lymphnode metastasis and poor survival in non-small-cell lung cancer. AnnOncol 2010;21:223–31.

22. Al-Saad S, Al-Shibli K, DonnemT, PerssonM, Bremnes RM, Busund LT. Theprognostic impact of NF-kappaB p105, vimentin, E-cadherin and Par6expression in epithelial and stromal compartment in non-small-cell lungcancer. Br J Cancer 2008;99:1476–83.

23. Pardoll DM. The blockade of immune checkpoints in cancer immuno-therapy. Nat Rev Cancer 2012;12:252–64.

24. PohlM,OlsenKE,Holst R,DitzelHJ,HansenO. Tissuemicroarrays in non-small-cell lung cancer: reliability of immunohistochemically-determinedbiomarkers. Clin Lung Cancer 2014;15:222–30.

25. Rami-Porta R, Crowley JJ, Goldstraw P. The revised TNM staging system forlung cancer. Ann Thorac Cardiovasc Surg 2009;15:4–9.

26. Travis WD, Brambilla E, Riely GJ. New pathologic classification of lungcancer: relevance for clinical practice and clinical trials. J Clin Oncol 2013;31:992–1001.

27. McShane LM, Altman DG, Sauerbrei W, Taube SE, Gion M, Clark GM.Reporting recommendations for tumor marker prognostic studies(REMARK). J Natl Cancer Inst 2005;97:1180–4.

28. Bremnes RM, Al-Shibli K, Donnem T, Sirera R, Al-Saad S, Andersen S,et al. The role of tumor-infiltrating immune cells and chronic inflam-mation at the tumor site on cancer development, progression, andprognosis: emphasis on non-small cell lung cancer. J Thorac Oncol2011;6:824–33.

29. Fridman WH, Dieu-Nosjean MC, Pages F, Cremer I, Damotte D, Sautes-Fridman C, et al. The immune microenvironment of human tumors:general significance and clinical impact. Cancer Microenviron 2013;6:117–22.

30. Prado-Garcia H, Aguilar-Cazares D, Flores-Vergara H, Mandoki JJ, Lopez-Gonzalez JS. Effector, memory and naive CD8+ T cells in peripheral bloodand pleural effusion from lung adenocarcinoma patients. Lung Cancer2005;47:361–71.

31. Sznol M, Chen L. Antagonist antibodies to PD-1 and B7-H1 (PD-L1) in thetreatment of advanced human cancer. Clin Cancer Res 2013;19:1021–34.

www.aacrjournals.org Clin Cancer Res; 21(11) June 1, 2015 2643

CD8þ T Cells in NSCLC

on July 7, 2018. © 2015 American Association for Cancer Research. clincancerres.aacrjournals.org Downloaded from

Published OnlineFirst February 13, 2015; DOI: 10.1158/1078-0432.CCR-14-1905

2015;21:2635-2643. Published OnlineFirst February 13, 2015.Clin Cancer Res   Tom Donnem, Sigurd M. Hald, Erna-Elise Paulsen, et al.  

Small Cell Lung Cancer−Staging in Non A Promising Supplement to TNM−− T-cell Density+Stromal CD8

  Updated version

  10.1158/1078-0432.CCR-14-1905doi:

Access the most recent version of this article at:

  Material

Supplementary

  http://clincancerres.aacrjournals.org/content/suppl/2015/02/17/1078-0432.CCR-14-1905.DC1

Access the most recent supplemental material at:

   

   

  Cited articles

  http://clincancerres.aacrjournals.org/content/21/11/2635.full#ref-list-1

This article cites 31 articles, 5 of which you can access for free at:

  Citing articles

  http://clincancerres.aacrjournals.org/content/21/11/2635.full#related-urls

This article has been cited by 4 HighWire-hosted articles. Access the articles at:

   

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

  Subscriptions

Reprints and

  .pubs@aacr.org

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

  Permissions

  Rightslink site. Click on "Request Permissions" which will take you to the Copyright Clearance Center's (CCC)

.http://clincancerres.aacrjournals.org/content/21/11/2635To request permission to re-use all or part of this article, use this link

on July 7, 2018. © 2015 American Association for Cancer Research. clincancerres.aacrjournals.org Downloaded from

Published OnlineFirst February 13, 2015; DOI: 10.1158/1078-0432.CCR-14-1905