ARTICLE

Genomic Analysis of Immune Cell Infiltrates

Across 11 Tumor Types

Michael D Iglesia Joel S Parker Katherine A Hoadley Jonathan S SerodyCharles M Perou Benjamin G VincentAffiliations of authors Lineberger Comprehensive Cancer Center (MDI JSP KAH JSS CMP BGV) Curriculum in Genetics and Molecular Biology (MDI JSP) Departmentof Medicine (JSS BGV) Department of Genetics (KAH CMP) Department of Microbiology and Immunology (JSS) and Department of Pathology amp Laboratory Medicine(CMP) University of North Carolina Chapel Hill Chapel Hill NC

Authors contributed equally to this workCorrespondence to Benjamin Vincent MD Lineberger Comprehensive Cancer Center University of North Carolina 125 Mason Farm Road 27599 Chapel Hill NC(e-mail benjamin_vincentmeduncedu)

Abstract

Background Immune infiltration of the tumor microenvironment has been associated with improved survival for somepatients with solid tumors The precise makeup and prognostic relevance of immune infiltrates across a broad spectrum oftumors remain unclearMethods Using mRNA sequencing data from The Cancer Genome Atlas (TCGA) from 11 tumor types representing 3485tumors we evaluated lymphocyte and macrophage gene expression by tissue type and by genomic subtypes defined withinand across tumor tissue of origin (Cox proportional hazards Pearson correlation) We investigated clonal diversity of B-cellinfiltrates through calculating B-cell receptor (BCR) repertoire sequence diversity All statistical tests were two-sidedResults High expression of T-cell and B-cell signatures predicted improved overall survival across many tumor typesincluding breast lung and melanoma (breast CD8_T_Cells hazard ratio [HR] frac14 036 95 confidence interval [CI] frac14 016 to 081P frac14 01 lung adenocarcinoma B_Cell_60gene HRfrac14071 95 CIfrac14058 to 087 P frac14 780E-04 melanoma LCK HRfrac14086 95CIfrac14079 to 094 P frac14 675E-04) Macrophage signatures predicted worse survival in GBM as did B-cell signatures in renaltumors (Glioblastoma Multiforme [GBM] macrophages HRfrac14162 95 CIfrac14117 to 226 P frac14 004 renal B_Cell_60geneHRfrac14117 95 CIfrac14104 to 132 P frac14 009) BCR diversity was associated with survival beyond gene segment expression inmelanoma (HRfrac14267 95 CIfrac14132 to 540 P frac14 02) and renal cell carcinoma (HRfrac14036 95 CIfrac14015 to 087 P frac14 006)Conclusions These data support existing studies suggesting that in diverse tissue types heterogeneous immune infiltratesare present and typically portend an improved prognosis In some tumor types BCR diversity was also associated withsurvival Quantitative genomic signatures of immune cells warrant further testing as prognostic markers and potentialbiomarkers of response to cancer immunotherapy

The complex interplay between solid tumors and host immu-nity has been widely studied but remains incompletely under-stood In multiple tumor types tumor-infiltrating lymphocytes(TILs) have been associated with clinical outcomes (1ndash8) For ex-ample CD8thorn TILs have been shown to be favorably prognosticin melanoma colorectal breast ovarian and nonndashsmall celllung cancer In selected tumors it has been demonstrated thatthese CD8thorn TILs are able to specifically kill tumor cells (910)

Several schemas have been developed to leverage immune infil-tration as a prognostic factor (1112) Solid tumors are thoughtto cultivate an immunosuppressive microenvironment thatpromotes exhaustion of TILs and induction of a protumor in-flammatory wound-healing response (13) This is supported bydata that regulatory T-cells (Treg) tumor-associated macro-phages (TAMs) andor myeloid-derived suppressor cells(MDSCs) predict worse outcomes in melanoma renal cell

AR

TIC

LE

Received December 21 2015 Revised March 7 2016 Accepted April 26 2016

copy The Author 2016 Published by Oxford University Press All rights reserved For Permissions please e-mail journalspermissionsoupcom

1 of 11

JNCI J Natl Cancer Inst (2016) 108(11) djw144

doi 101093jncidjw144

ArticleFirst published online June 22 2016

at University of N

orth Carolina at C

hapel Hill on June 22 2016

httpjncioxfordjournalsorgD

ownloaded from

carcinoma breast ovarian bladder prostate and nonndashsmallcell lung cancer (14ndash16) The wide impact and clinical relevanceof the multifaceted tumor-associated immune response makesit critical to develop a more thorough understanding of thisphenomenon

Next-generation sequencing and large-scale genomics havebecome critical to our understanding of human cancersThrough efforts such as The Cancer Genome Atlas (TCGA)Project genomic data on a wide variety of tumors have becomeavailable By combining diverse datasets groupings both withinand across tumor types have emerged (17) This has deepenedour understanding of distinctions within tumor types and high-lighted similarities between previously distinct tumor typessuch as identification of the ldquosquamousrdquo genomic subtypewhich combines lung squamous head and neck and somebladder cancers into a single group (1718) While many tumortypes are thought to harbor prognostic TILs the interplay be-tween genomic subtype and the antitumor immune responsehas not been adequately explored

Cancer immunotherapy has been pursued as an alternativeor complement to cytotoxic chemotherapy and radiotherapyInhibition of the immune checkpoint proteins CTLA-4 PD-1 andPD-L1 reduces the ability of the tumor microenvironment to sup-press host antitumor immunity (19ndash21) and immune checkpointinhibitors have shown clinical responses in diverse cancers (20ndash24) As these and other immune-targeted therapies gain wide-spread clinical usage a key question is identification of tumorcharacteristics that predict response Evidence in melanoma andbladder cancer suggests that responders may harbor clonallyrestricted antitumor TILs that are able to respond after immuno-suppression has been lifted (2325) In this work we use mRNA-seq data for a large number of diverse tumors to analyze theprevalence and prognostic relevance of tumor-immune infiltratesand evaluate the clonal diversity of tumor-infiltrating B-cells

Methods

Datasets

The dataset used comprised mRNA-seq data from 3485 TCGAtumors (see TCGA Data Portal at httpstcga-datancinihgovtcga CGHub at httpscghubucscedu) which was originallydescribed in Hoadley 2014 with the following modifications(17) AML samples were removed Data for 329 melanoma sam-ples were added (26) All samples were assayed by mRNA-seqas described by the TCGA Research Network (27) Gene expres-sion values were represented as RSEM data normalized withineach sample to the upper quartile of total reads (37) For genesignature and hierarchical clustering analyses gene expressionvalues were median-centered across all 3485 tumors Genomicsubtypes within tissue types and TCGA Pan-Cancer Cluster ofClusters Assignments (COCA) subtypes were obtained from pre-vious publications of those TCGA tissue types (171827ndash33) NoCOCA subtype was assigned to melanoma samples and thusmelanoma was assigned to its own group in COCA subtypeanalyses (34)

Gene Signatures and BCR Expression and Diversity

Several immune gene signatures were previously developed byunsupervised hierarchical clustering of mRNA-seq expressiondata for 728 breast tumor samples (35) In addition to theseother tumor-infiltrating lymphocyte and macrophage gene

signatures were obtained from published studies Multiple sig-natures were used for each cell type to reduce bias from individ-ual gene lists and included IGG_Cluster (36) B_Cell (37) B_Cells(12) B_Cell_cluster (35) B_Cell_60gene (38) and TNBC_B_Cell(39) are Bplasma cell signatures T_Cell (37) T_Cells (12)T_Cell_cluster (35) CD8 (37) CD8_cluster (35) LCK (40) andTNBC_T-Cell (39) are T-cell signatures with the CD8 andCD8_cluster signatures specifically representing CD8thorn T-cellsMacTh1_cluster (35) CD68_cluster (35) Mac_CSF1 (41) andMacrophages (12) are all macrophagemonocyte signatures Foreach signature the mean expression of all genes in the signa-ture was used as the signature score for a given tumorpatientExpression levels for all BCR gene segments overall BCR genesegment expression and BCR diversity values were calculatedaccording to previously described methods (35)

Statistical Analysis

Gene signature expression levels were compared using one-wayanalysis of variance (ANOVA) Correlations were evaluated us-ing Pearson coefficient calculation Univariate survival analyseswere performed by Cox proportional hazards modeling witheach signature tested as a continuous variable MultivariableCox proportional hazards models were used to evaluate theprognostic value of individual gene expression signatures whencombined with tumor typesubtype information and clinicalvariables All multivariable survival analyses included one geneexpression signature as well as patient age and pathologic tu-mor size (T) node (N) and metastasis (M) stages Pathologic TN and M stages were binned into T1 vs Tgt1 N0 vs Ngt 0 andM0 vs Mgt0 groups to reduce the number of variables in eachmodel and to standardize across tumor types Multiple testingcorrection was done by calculating q-values for each P valueand statistically significant q-values were confirmed using theKaplan-Meier estimator and log-rank testing of high-expressionvs low-expression groups All survival analyses were performedin R version 2131 using the ldquosurvivalrdquo package (4243) To as-sess the statistical significance of association between BCR di-versity and survival multivariable Cox proportional hazardmodels were constructed for each tumor tissue type where vari-ables included BCR gene segment expression diversity and theexpressiondiversity interaction term Separate models were fitusing the expression term alone and statistical significancewas assessed by performing analysis of variance comparing thefull and reduced models using the likelihood ratio test imple-mented in R (43) A P value of less than 05 was considered sta-tistically significant and all statistical tests were two-sided

Results

Genomic Signatures of Immune Infiltrationin Human Cancer

To clarify the role of tumor-infiltrating immune cells in humancancer gene signatures associated with immune cell types wereevaluated in a diverse set of 3485 tumors representing 11 tumortypes Signatures chosen were previously published as corre-sponding to tumor-infiltrating immune cells in at least one solidtumor type A heat map of signature expression across all 3485samples ordered by tissue type and genomic subtype showedhigh expression across particular cancer types and subtypes(Figure 1) Immune signature expression in general was highestin the basal-like bladder basal-like and human epidermal

AR

TIC

LE

2 of 11 | JNCI J Natl Cancer Inst 2016 Vol 108 No 11

at University of N

orth Carolina at C

hapel Hill on June 22 2016

httpjncioxfordjournalsorgD

ownloaded from

growth factor 2 (HER2)ndashenriched breast immunoreactive ovar-ian and immune melanoma subtypes as well as in the headand neck clear cell renal cell lung adenocarcinoma and lungsquamous tissue types

While only the ovarian immunoreactive and melanoma im-mune subtypes were defined with immune infiltration as a keyfeature immune infiltration often segregated by tumor genomicsubtype Similarly assignment by Cluster of Cluster Analysisclassification yielded differences in immune gene signature ex-pression (Figure 2 Supplementary Figure 1 available online) (17)Within all tumor types immune gene signatures tended to bepositively correlated with the average Pearson correlation above63 for all tumor tissue or COCA types (Figure 3) Signatures repre-senting the same immune cell type in most cases were highlycorrelated regardless of tissue of origin or COCA type The lowestintersignature correlation was among glioblastoma sampleswhich showed low expression of lymphocyte-related gene signa-tures and high expression of signatures related to macrophagesand other myeloid-derived cell types

Prognostic Significance of ImmuneInfiltration Signatures

Univariate survival analyses for immune signature expressionwithin tissue and COCA types are summarized in Table 1 andSupplementary Table 1 (available online) with hazard ratios pro-vided in Supplementary Tables 2 and 3 (available online) Thenumber of events and median follow-up times varied substan-tially between tumor types Almost all immune signatures testedwere prognostic in the complete set of 3485 samples The breasthead and neck lung adenocarcinoma melanoma and endome-trial groups showed improved survival among patients with tu-mors scoring high for a variety of immune signatures (eg breastCD8_T_Cells hazard ratio [HR] frac14 036 95 confidence interval [CI]frac14 016 to 081 P frac14 01 head and neck B_Cell_Cluster HRfrac14 08295 CIfrac14 074 to 091 P frac14 222E-04 lung adenocarcinoma B_Cell_

60gene HRfrac14 071 95 CIfrac14 058 to 087 P frac14 780E-04 melanomaLCK HRfrac14 086 95 CIfrac14 079 to 094 P frac14 675E-04 endometrialCD8_cluster HRfrac14 071 95 CIfrac14 055 to 092 P frac14 001) When sam-ples were divided by COCA subtype the LUAD-enriched squa-mous UCEC and melanoma subtypes showed improved survivalwith high immune signature expression for at least four signa-tures each (LUAD-enriched HRfrac14 073069075077 95 CIfrac14 063to 086058 to 082061 to 092064 to 092 P frac14 132E-04214E-05006004 for IGG_ClusterB_Cell_60geneT_CellsCD8_clustersquamous HRfrac14 089086085087 95 CIfrac14 083 to 096078 to 095076 to 095079 to 096 P frac14 003003006004 for B_Cell_clusterT_Cell_clusterT_CellsCD8_cluster UCEC HRfrac14 071070069071 95 CIfrac14 054 to 093052 to 093050 to 096055 to 093P frac14 01020301 for TNBC_T-CellLCKT_CellsCD8_cluster mel-anoma HRfrac14 061086086080 95 CIfrac14 046 to 079080 to 092079 to 094071 to 091 P frac14 209E-04177E-05675E-04604E-04for B_CellB_Cell_60geneLCKMac_CSF1) Signatures representinglymphocytes more consistently predicted overall survival thandid other signatures the lung adenocarcinoma head and neckbreast and endometrial tumor types showed consistent prognos-tic ability for lymphocyte signatures and a decreased associationbetween survival and macrophage signatures Interestingly inclear cell renal cell carcinoma there was an association of B-cellsignature expression with worse survival (eg IGG_Cluster HRfrac14 122 95 CIfrac14 108 to 139 P frac14 002) Macrophage-related signatureswere less commonly prognostic though they were associatedwith worse survival overall and within GBM (eg MacrophagesHRfrac14 162 95 CIfrac14 117 to 226 P frac14 004)

Multivariable survival analyses of immune signatures wereperformed for each tumor tissue type individually incorporat-ing patient age pathologic stage and signature expression(Supplementary Tables 4-11 available online) Multivariablemodeling results largely supported the results seen in the uni-variate survival models with head and neck lung adenocarci-noma and melanoma showing consistent prognostic benefit forpatients with high immune gene expression (eg B_Cell_clusterHRfrac14 082077092 95 CIfrac14 073 to 091065 to 093086 to 098

Figure 1 Concordant expression of genes from different immune cell types in specific solid tumor types and subtypes Tumors are ordered by column according to tis-

sue type and subtype Within signatures genes are ordered by row by unsupervised hierarchical clustering across all patients Log 2 gene expression is median-cen-

tered across all samples IGG_Cluster B_Cell B_Cells B_Cell_cluster B_Cell_60gene and TNBC_B_Cell are Bplasma cell signatures T_Cell T_Cells T_Cell_cluster CD8

CD8_cluster LCK and TNBC_T-Cell are T-cell signatures with the CD8 and CD8_cluster signatures specifically representing CD8thorn T-cells MacTh1_cluster

CD68_cluster Mac_CSF1 and Macrophages are all macrophagemonocyte signatures

AR

TIC

LE

M D Iglesia et al | 3 of 11

at University of N

orth Carolina at C

hapel Hill on June 22 2016

httpjncioxfordjournalsorgD

ownloaded from

Figu

re2

Imm

un

ege

ne

sign

atu

reex

pre

ssio

nby

tum

or

tiss

ue

typ

ean

dC

ance

rC

lust

ero

fC

lust

ers

Ass

ign

men

ts(C

OC

A)s

ubt

ypes

Box

plo

tso

fex

pre

ssio

no

fIG

G_C

lust

erC

D8

and

CD

68_c

lust

erge

ne

exp

ress

ion

sign

atu

res

asw

ella

s

gen

esPD

-1P

DL-

1an

dC

TLA

-4b

ytu

mo

rti

ssu

ety

pe

(AC

EG

IK

)or

CO

CA

subt

ype

(BD

FH

JL

)are

sho

wn

Sta

tist

ical

sign

ifica

nce

test

ing

was

do

ne

usi

ng

on

e-w

ayan

alys

iso

fva

rian

ce(A

NO

VA

)Ex

pre

ssio

no

fim

mu

ne

gen

es

was

con

sist

entl

yh

igh

inSK

CM

HN

SCL

UA

DL

USC

an

dK

IRC

tum

ors

AN

OV

AP

valu

esfo

rge

ne

or

sign

atu

reex

pre

ssio

nby

tiss

ue

typ

eC

OC

Agr

ou

pC

D8

Pfrac14

512

E-15

56

26E-

169

IGG

_Clu

ster

Pfrac14

101

E-23

86

42E-

224

CD

68_c

lust

er

Pfrac14

109

-208

37

7E-1

99P

D-1

Pfrac14

646

E-97

87

1E-1

03P

D-L

1Pfrac14

190

E-27

11

11E-

279

CT

LA-4

Pfrac14

466

E-11

74

85E-

127

All

stat

isti

calt

ests

wer

etw

o-s

ided

AR

TIC

LE

4 of 11 | JNCI J Natl Cancer Inst 2016 Vol 108 No 11

at University of N

orth Carolina at C

hapel Hill on June 22 2016

httpjncioxfordjournalsorgD

ownloaded from

P frac14 005330E-0401 for head and necklung adenocarcinomamelanoma) GBM ovarian and endometrial tumors were ex-cluded from multivariable analyses within tissue types as path-ologic staging data were not available

Prognostic BCR Gene Segment Expression

Previous investigations of TILs have suggested that manytumors contain clonally restricted T- andor B-cell infiltrates

(44ndash46) consistent with an antigen-driven response As previ-ously described mRNA-seq data can be used to estimate therelative clonal diversity of B-cell populations and discover as-sociations between specific BCR gene segment expression andimproved survival (35) Given our prior results that B-cell genesignature expression was more statistically significantly asso-ciated with improved OS than either T-cell gene signature ex-pression or the classical clinical prognostic variables of ageand lymph node involvement in breast cancer and that spe-cific BCR segments where expression was associated with OS

Figure 3 Correlation of lymphocyte infiltration signatures in tumors In (A) all samples and (B-L) within samples in each tissue type Pearson correlations are shown be-

tween all immune gene signatures

AR

TIC

LE

M D Iglesia et al | 5 of 11

at University of N

orth Carolina at C

hapel Hill on June 22 2016

httpjncioxfordjournalsorgD

ownloaded from

Tab

le1

Un

ivar

iate

Co

xp

rop

ort

ion

alP

valu

esfo

rim

mu

ne

gen

esi

gnat

ure

exp

ress

ion

vso

vera

llsu

rviv

alby

tiss

ue

typ

e

Var

iabl

eSt

ud

yA

llB

LCA

BR

CA

CR

CG

BM

HN

SCK

IRC

LUA

DLU

SCO

VSK

CM

UC

EC

No

34

8511

882

125

012

030

146

126

723

724

032

934

1

OS

med

ian

(ran

ge)

mo

240

128

242

199

106

205

372

193

210

286

433

272

(00

ndash369

9)

(00

ndash140

8)

(00

ndash234

3)

(00

ndash117

1)

(02

ndash50

5)(0

0ndash2

110

)(0

0ndash1

206

)(0

0ndash2

340

)(0

0ndash1

741

)(0

3ndash1

802

)(0

0ndash3

699

)(0

4ndash1

858

)

Even

tss

amp

leP

210

390

130

176

717

445

347

333

418

567

468

123

IGG

_CLU

STER

PFa

net

al2

011

(36)

310

E-06

4

40

28

80

04dagger

760

E-04

0

02dagger

003

3

51

33

76E-

04

05

TN

BC

_B-C

ell

PR

od

yet

al2

011

(39)

527

E-04

2

92

96

73

06

71E-

04

05

11

50

77

05

31

B_C

ell

PFa

net

al2

011

(36)

002

6

31

16

21

40

1dagger

25

02

13

99

209

E-04

6

3

B_C

ell_

clu

ster

PIg

lesi

aet

al2

014

(35)

525

E-10

5

10

28

30

01dagger

222

E-04

0

1dagger3

67E-

04

27

25

002

2

3

B_C

ell_

60ge

ne

PSc

hm

idt

etal

200

8(3

8)1

45E-

07

30

07

73

001

dagger3

28E-

04

009

dagger7

80E-

04

29

12

177

E-05

0

5

B_C

ells

PB

ind

eaet

al2

013

(12)

07

70

14

81

54

003

0

50

02

42

90

001

0

5

T_C

ell

PFa

net

al2

011

(36)

007

8

90

15

60

03dagger

03

28

01

28

54

703

E-04

4

7

TN

BC

_T-C

ell

PR

od

yet

al2

011

(39)

104

E-05

4

70

68

90

60

03

33

01

89

52

001

0

1

LCK

PFa

net

al2

011

(36)

962

E-06

4

70

79

30

80

05

45

02

67

49

675

E-04

0

01

T_C

ell_

clu

ster

PIg

lesi

aet

al2

014

(35)

230

E-10

2

80

49

70

1dagger

454

E-04

1

30

17

92

50

01

02

T_C

ells

PB

ind

eaet

al2

013

(12)

351

E-12

2

90

59

40

40

02

93

01

97

28

001

0

2

CD

8P

Fan

etal

201

1(3

6)0

11

60

19

70

02dagger

004

0

60

1Dagger9

69

50

02

04

CD

8_cl

ust

erP

Igle

sia

etal

201

4(3

5)1

18E-

11

26

03

92

01

dagger6

57E-

04

20

01

82

26

993

E-04

0

01

CD

8_T

_Cel

lsP

Bin

dea

etal

201

3(1

2)8

23E-

07

09

01

51

09

01

87

05

79

66

003

0

9

Mac

Th

1_cl

ust

erP

Igle

sia

etal

201

4(3

5)3

06

94

99

22

20

46

41

14

48

74

80E-

04

07

CD

68_c

lust

erP

Igle

sia

etal

201

4(3

5)0

4dagger

89

81

85

05

28

47

56

19

94

008

2

7

Mac

_CSF

1P

Bec

ket

al2

009

(41)

95

84

46

97

10

14

65

14

47

93

604

E-04

1

4

Mac

rop

hag

esP

Bin

dea

etal

201

3(1

2)1

65E-

05dagger

73

99

83

004

dagger4

71

13

72

92

40

04

59

In

dic

ates

that

surv

ival

was

was

stat

isti

call

ysi

gnifi

can

tly

asso

ciat

edw

ith

gen

esi

gnat

ure

exp

ress

ion

wit

hfa

lse-

dis

cove

ryra

te-a

dju

sted

Pva

luelt

05

and

con

firm

atio

nby

Kap

lan

-Mei

eres

tim

atio

nan

dlo

g-ra

nk

test

ing

daggerPva

lues

mar

ked

ind

icat

eh

igh

sign

atu

reex

pre

ssio

nis

stat

isti

call

ysi

gnifi

can

tly

(Plt

05)

asso

ciat

edw

ith

po

or

surv

ival

All

stat

isti

calt

ests

are

two

-sid

ed

DaggerIn

dic

ates

that

surv

ival

was

stat

isti

call

ysi

gnifi

can

tly

asso

ciat

edw

ith

gen

esi

gnat

ure

exp

ress

ion

wit

hfa

lse-

dis

cove

ryra

te-a

dju

sted

Pva

luelt

05

All

stat

isti

calt

ests

are

two

-sid

ed

AR

TIC

LE

6 of 11 | JNCI J Natl Cancer Inst 2016 Vol 108 No 11

at University of N

orth Carolina at C

hapel Hill on June 22 2016

httpjncioxfordjournalsorgD

ownloaded from

varied by tumor subtype we were interested in exploring BCRgene segment expression associations with survival in thebroader TCGA datasets Figure 4A illustrates BCR gene seg-ment expression across all tissue types highlighting the sametissue types identified by gene signature analysis While BCRexpression was able to distinguish certain samples and tumortypes as high expressers expression was high across multipleBCR gene segments in such samples To more specifically dis-sect the importance of oligoclonal B-cell populations we de-termined if individual BCR gene segment expression wasassociated with overall survival (Figure 4B) For melanoma andhead and neck squamous cell carcinoma expression of themajority of BCR gene segments was associated with prolongedsurvival whereas in renal cell carcinoma and glioblastomamultiforme most segments were associated with reduced

survival There was a high degree of overlap of prognostic BCRsegments across tumor types in which many BCRs were prog-nostic (melanoma head and neck glioblastoma renal celllung adenocarcinoma and breast) To understand the statisti-cal significance of individual BCR gene segment expressionwe assessed for each tumor type whether the number of prog-nostic BCR segments exceeded the number of prognostic genesexpected by chance To do this we used a bootstrap resam-pling approach by randomly choosing a subset of genes of thesame size as the number of total BCR gene segments (nfrac14 1000resamplings) and considering as the null distribution the 95confidence interval of the number of prognostic genes in theresampling dataset For seven tumor types the number ofprognostic BCR segments exceeded the upper confidence in-terval of the null distribution (Figure 4C)

Figure 4 B-cell receptor (BCR) gene segment expression and pattern of BCR gene segments where expression was associated with overall survival by tumor type A)

Unsupervised hierarchical clustering of average BCR gene segment expression by tissue type B) Grid of BCR gene segments where increased expression was statisti-

cally significantly associated with prolonged overall survival (OS red) and gene segments associated with diminished OS (blue) C) Number of gene segments with sta-

tistically significant association with OS by tumor type Cox proportional hazard models were built using Log10 gene expression for each gene segment Null

distributions were estimated by bootstrap resampling (nfrac14 1000) of 353 random genes and calculating the number where expression was associated with OS Error bars

show the 95 bootstrap confidence interval

AR

TIC

LE

M D Iglesia et al | 7 of 11

at University of N

orth Carolina at C

hapel Hill on June 22 2016

httpjncioxfordjournalsorgD

ownloaded from

Figu

re5

B-c

ellr

ecep

tor

(BC

R)v

aria

ble

gen

ese

gmen

tex

pre

ssio

nan

dd

iver

sity

bytu

mo

rty

pe

BC

RV

segm

ent

seq

uen

ced

iver

sity

(by

log1

0(ef

fect

ive

nu

mbe

ro

fsp

ecie

s))v

sB

CR

Vse

gmen

tex

pre

ssio

n(b

ylo

g10(

read

cou

nts

))fo

ral

ltu

-

mo

rty

pes

In

each

plo

ttu

mo

rso

fa

sin

gle

typ

ear

eco

lore

din

red

wit

hal

loth

ertu

mo

rsco

lore

din

gray

AR

TIC

LE

8 of 11 | JNCI J Natl Cancer Inst 2016 Vol 108 No 11

at University of N

orth Carolina at C

hapel Hill on June 22 2016

httpjncioxfordjournalsorgD

ownloaded from

Figure 5 shows BCR gene segment expression and sequencediversity for all samples and demonstrates the existence of ahigh-expression low-diversity group composed mainly of a sub-set of melanoma breast head and neck renal cell and ovariantumors To test the association of restricted BCR diversity withsurvival Cox proportional hazards models were built using BCR-variable gene segment expression diversity and the interactionterm between expression and diversity For each tissue type thismodel was compared with a reduced model containing BCR genesegment expression as the sole variable These models werecompared using analysis of variance to assess whether includingthe diversity term in the full model provided a statistically signifi-cantly more robust explanation of variance than the expressionterm alone This approach was used because B-cell gene segmentexpression was highly correlated with B-cell gene signature ex-pression which was statistically significantly associated withoverall survival in multiple tumor types (Table 1) The results ofthis analysis are shown in Figure 6 The full model was found tobe statistically significantly better at predicting survival in blad-der renal cell and melanoma datasets Decreased diversity wasassociated with improved survival in melanoma (HRfrac14 267 95CIfrac14 132 to 540 P frac14 02) In contrast decreased diversity was as-sociated with diminished overall survival in renal cell carcinoma(HRfrac14 036 95 CIfrac14 015 to 087 P frac14 006)

Discussion

Here we present several key aspects of a prognostically relevantimmune response determined from mRNA-seq data across adiverse set of human tumors Patients with apparently benefi-cial tumor-immune infiltrates had high levels of immune signa-ture expression for several cell types including cytotoxic andhelper T-cells Bplasma cells and macrophagesmonocytesAmong these highly infiltrated tumors a subset contained ahighly expressed population of BCRs with lower sequence diver-sity Squamous and basal-like COCA subtypes showed in-creased immune gene signature expression and prognosticsignificance The fact that these subtypes were defined compre-hensively by global gene expression protein levels DNA copynumber miRNA and DNA methylation suggests that subtype-defining features may be responsible for the shared immuno-genic phenotype

Most investigations into the role of TILs in human tumorshave focused on T-cells Multiple studies have shown T-cellTILs to be associated with response to immune checkpoint inhi-bition and survival (2347ndash49) This work adds to the growingbody of literature identifying T-cell TILs as a positive prognosticfeature Less work has been done evaluating the role of B-cellsin solid tumors though CD20 expression has been associatedwith improved survival in breast and ovarian cancer (1050)Here we demonstrate that in all tumor types analyzed B-cellgene expression including expression of BCR gene segmentswas highly correlated with expression of T-cell genes Thestrong correlation of immune signature expression across dis-tinct immune cell types illustrates that the tumor-immune in-filtrate is diverse but somewhat predicable and consistent witha supportive role for B-cell TILs in the antitumor immune re-sponse The addition of B-cell clonal diversity analysis repre-sents a novel aspect of tumor-immune research and highlightsthe importance of B-cell TILs in diverse solid tumor types

Our analysis of adaptive immune receptor repertoire diver-sity is limited by evaluation of BCR repertoires alone We fo-cused on the BCR for one major technical reason The diversity

estimate used here depends on the degree of sequence variationwithin the immunoglobulin-variable loci generated by somatichypermutation Because the TCR does not undergo somatichypermutation we are not able to measure TCR diversity usingthis approach Our study is also limited by use of gene expres-sion data to infer characteristics of the tumor-immune micro-environment without companion cellular assays which werenot obtainable from TCGA samples Although the immune genesignatures reported here have been published and well-validated to correspond to specific cellular populations (1235)we were unable to evaluate abundances of more complex im-mune cell phenotypes that may be important in tumor immu-nobiology and accessable via multiparameter IHC or flowcytometry Nonetheless our results are important in illustratingthe power of immune gene signature and BCR repertoire analy-sis applied to a large and diverse RNA-seq dataset It is unclearwhether immunohistochemistry-based methods of interrogat-ing the tumor-immune microenvironment such as theImmunoscore (11) will outperform genomics methods for pro-ductive biomarker development

For most of the tumor types tested immune infiltrates wereassociated with improved survival however there were threetumor types that were exceptions colorectal adenocarcinomaclear cell renal cell carcinoma and GBM There are many stud-ies indicating a beneficial role for TILs in human colorectal can-cer (51ndash53) The failure to detect these associations in TCGAdata may have to do with the lack of adequate follow-up lowevent rate and low number of patients with mismatch repairgene mutations (54) In renal cell carcinoma high expression ofB-cell signatures and decreased BCR diversity (consistent withan antigen-driven response) predicted poor survival Previouswork has suggested that this tumor type may contain abundant

Figure 6 Multivariable Cox proportional hazard modeling results for B-cell re-

ceptor (BCR) gene segment diversity vs overall survival Each column indicates

one tumor tissue type CoxPH survival models were fit with BCR variable gene

segment expression alone as an explanatory variable (Expression model) and in-

cluding both expression and diversity as explanatory variables (Expression amp

Diversity model) Dark gray bars show change in LR v2 statistic between the

Expression model and the null model and light gray bars show the change in LR

v2 statistic between the Expression model and the Expression amp Diversity model

(ie a measure of the extent of increased information included in the latter

model) Tumor types in which the Expression amp Diversity model provided a sta-

tistically significantly better fit than the Expression model alone (P lt 05 by LRT)

are indicated by All statistical tests were two-sided

AR

TIC

LE

M D Iglesia et al | 9 of 11

at University of N

orth Carolina at C

hapel Hill on June 22 2016

httpjncioxfordjournalsorgD

ownloaded from

regulatory B-cells (Breg) which may explain this findingImmune infiltration appeared to predict worse outcomes inGBM as well As in our data infiltrating macrophages andormicroglia often represent a large proportion of immune infil-trates (5556) The dominance of TAMs within the immune infil-trate of GBM is consistent with a negative clinical impact

In summary for most solid tumor types expression of im-mune gene signatures predicted a more favorable outcomeAnalysis of potential clonal restriction suggests that in some tu-mors with heavy immune infiltration B-cell clonal restrictionadds prognostic information This may be because of tumor an-tigenndashdriven expansion of specific B-cell clones Not all tumortypes show these common features with counterexamples be-ing GBM and clear cell renal cell carcinoma where high immuneinfiltration predicted a worse outcome Thus measures of im-mune cell phenotypes BCR (and potentially TCR) diversity andtissue type or subtype are needed to correctly interpret andlikely act upon these characteristics of the tumor microenvi-ronment High levels of pretreatment CD8thorn TILs have correlatedwith response to immune checkpoint inhibition in melanomaand bladder cancer (2347) It will be important to test whetherquantification of immune infiltrates using mRNA-seq predictsclinical response to these and other immunotherapeutic strate-gies going forward

Funding

This study was supported by funds from the following sourcesU24-CA143848-06 (The Cancer Genome Atlas) the NationalCancer Institute Breast Specialized Programs of ResearchExcellence program (P50-CA58223-09A1) RO1-CA195740-01 theBreast Cancer Research Foundation UNC University CancerResearch Fund UNC Oncology Clinical Translational ResearchTraining Program (5K12CA120780) and William Guy ForbeckResearch Foundation

Notes

The funders had no role in study design the data collection oranalysis the decision to publish or the preparation of the man-uscript We would like to thank the Perou and Vincent labs forhelpful suggestions

References1 Balch CM Riley LB Bae YJ et al Patterns of human tumor-infiltrating lym-

phocytes in 120 human cancers Arch Surg 1990125(2)200ndash2052 Lee S Margolin K Tumor-infiltrating lymphocytes in melanoma Curr Oncol

Rep 201214(5)468ndash4743 Kandalaft LE Motz GT Duraiswamy J et al Tumor immune surveillance and

ovarian cancer lessons on immune mediated tumor rejection or toleranceCancer Metastasis Rev 201130(1)141ndash151

4 Anagnostou VK Brahmer JR Cancer Immunotherapy A Future ParadigmShift in the Treatment of Non-Small Cell Lung Cancer Clin Cancer Res 201521(5)976ndash984

5 Finke JH Rayman PA Ko JS et al Modification of the tumor microenviron-ment as a novel target of renal cell carcinoma therapeutics Cancer J 201319(4)353ndash364

6 Liakou CI Narayanan S Ng Tang D et al Focus on TILs Prognostic signifi-cance of tumor infiltrating lymphocytes in human bladder cancer CancerImmun 2007710

7 Schoenfeld JD Immunity in head and neck cancer Cancer Immunol Res 20153(1)12ndash17

8 Lee AH Gillett CE Ryder K et al Different patterns of inflammation andprognosis in invasive carcinoma of the breast Histopathology 200648(6)692ndash701

9 Yee C Thompson JA Byrd D et al Adoptive T cell therapy using antigen-specific CD8thorn T cell clones for the treatment of patients with metastatic

melanoma in vivo persistence migration and antitumor effect of trans-ferred T cells Proc Natl Acad Sci U S A 200299(25)16168ndash16173

10 Milne K Kobel M Kalloger SE et al Systematic analysis of immune infiltratesin high-grade serous ovarian cancer reveals CD20 FoxP3 and TIA-1 as posi-tive prognostic factors PLoS One 20094(7)e6412

11 Galon J Pages F Marincola FM et al Cancer classification using theImmunoscore a worldwide task force J Transl Med 201210205

12 Bindea G Mlecnik B Tosolini M et al Spatiotemporal dynamics of intratu-moral immune cells reveal the immune landscape in human cancerImmunity 201339(4)782ndash795

13 Schreiber RD Old LJ Smyth MJ Cancer immunoediting integrating immu-nityrsquos roles in cancer suppression and promotion Science 2011331(6024)1565ndash1570

14 Gobert M Treilleux I Bendriss-Vermare N et al Regulatory T cells recruitedthrough CCL22CCR4 are selectively activated in lymphoid infiltrates sur-rounding primary breast tumors and lead to an adverse clinical outcomeCancer Res 200969(5)2000ndash2009

15 Biswas SK Allavena P Mantovani A Tumor-associated macrophages func-tional diversity clinical significance and open questions SeminImmunopathol 201335(5)585ndash600

16 Ostrand-Rosenberg S Sinha P Myeloid-derived suppressor cells linking in-flammation and cancer J Immunol 2009182(8)4499ndash4506

17 Hoadley KA Yau C Wolf DM et al Multiplatform analysis of 12 cancer typesreveals molecular classification within and across tissues of origin Cell 2014158(4)929ndash944

18 Brennan CW Verhaak RG McKenna A et al The somatic genomic landscapeof glioblastoma Cell 2013155(2)462ndash477

19 Hodi FS Butler M Oble DA et al Immunologic and clinical effects of antibodyblockade of cytotoxic T lymphocyte-associated antigen 4 in previously vacci-nated cancer patients Proc Natl Acad Sci U S A 2008105(8)3005ndash3010

20 Herbst RS Soria JC Kowanetz M et al Predictive correlates of response to theanti-PD-L1 antibody MPDL3280A in cancer patients Nature 2014515(7528)563ndash567

21 Topalian SL Hodi FS Brahmer JR et al Safety activity and immune corre-lates of anti-PD-1 antibody in cancer N Engl J Med 2012366(26)2443ndash2454

22 Phan GQ Yang JC Sherry RM et al Cancer regression and autoimmunity in-duced by cytotoxic T lymphocyte-associated antigen 4 blockade in patientswith metastatic melanoma Proc Natl Acad Sci U S A 2003100(14)8372ndash8377

23 Powles T Eder JP Fine GD et al MPDL3280A (anti-PD-L1) treatment leads toclinical activity in metastatic bladder cancer Nature 2014515(7528)558ndash562

24 Kluger H Sznol M Callahan M et al Survival Response Duration andActivity by Braf Mutation (Mt) Status in a Phase 1 Trial of Nivolumab (Anti-Pd-1 Bms-936558 Ono-4538) and Ipilimumab (Ipi) Concurrent Therapy inAdvanced Melanoma (Mel) Ann Oncol 201425

25 Robert L Tsoi J Wang X et al CTLA4 blockade broadens the peripheral T-cellreceptor repertoire Clin Cancer Res 201420(9)2424ndash2432

26 Cancer Genome Atlas Network Electronic address imo Cancer GenomeAtlas N Genomic Classification of Cutaneous Melanoma Cell 2015161(7)1681ndash1696

27 Cancer Genome Atlas Research N Comprehensive genomic characterizationof squamous cell lung cancers Nature 2012489(7417)519ndash525

28 Damrauer JS Hoadley KA Chism DD et al Intrinsic subtypes of high-gradebladder cancer reflect the hallmarks of breast cancer biology Proc Natl AcadSci U S A 2014111(8)3110ndash3115

29 Cancer Genome Atlas N Comprehensive molecular portraits of humanbreast tumours Nature 2012490(7418)61ndash70

30 Cancer Genome Atlas N Comprehensive molecular characterization of hu-man colon and rectal cancer Nature 2012487(7407)330ndash337

31 Cancer Genome Atlas N Comprehensive genomic characterization of headand neck squamous cell carcinomas Nature 2015517(7536)576ndash582

32 Brooks SA Brannon AR Parker JS et al ClearCode34 A prognostic risk pre-dictor for localized clear cell renal cell carcinoma Eur Urol 201466(1)77ndash84

33 Cancer Genome Atlas Research N Comprehensive molecular profiling oflung adenocarcinoma Nature 2014511(7511)543ndash550

34 Ojala KA Kilpinen SK Kallioniemi OP Classification of unknown primary tu-mors with a data-driven method based on a large microarray reference data-base Genome Med 20113(9)63

35 Iglesia MD Vincent BG Parker JS et al Prognostic B-cell signatures usingmRNA-seq in patients with subtype-specific breast and ovarian cancer ClinCancer Res 201420(14)3818ndash3829

36 Fan C Prat A Parker JS et al Building prognostic models for breast cancer pa-tients using clinical variables and hundreds of gene expression signaturesBMC Med Genomics 201143

37 Palmer C Diehn M Alizadeh AA et al Cell-type specific gene expression pro-files of leukocytes in human peripheral blood BMC Genomics 20067115

38 Schmidt M Bohm D von Torne C et al The humoral immune system has akey prognostic impact in node-negative breast cancer Cancer Res 200868(13)5405ndash5413

39 Rody A Karn T Liedtke C et al A clinically relevant gene signature in triplenegative and basal-like breast cancer Breast Cancer Res 201113(5)R97

40 Rody A Holtrich U Pusztai L et al T-cell metagene predicts a favorable prog-nosis in estrogen receptor-negative and HER2-positive breast cancers BreastCancer Res 200911(2)R15

AR

TIC

LE

10 of 11 | JNCI J Natl Cancer Inst 2016 Vol 108 No 11

at University of N

orth Carolina at C

hapel Hill on June 22 2016

httpjncioxfordjournalsorgD

ownloaded from

41 Beck AH Espinosa I Edris B et al The macrophage colony-stimulating factor1 response signature in breast carcinoma Clin Cancer Res 200915(3)778ndash787

42 Team RDC R A Language and Environment for Statistical Computing In2011

43 Therneau T A Package for Survival Analysis in S In 201244 Hansen MH Nielsen H Ditzel HJ The tumor-infiltrating B cell response in

medullary breast cancer is oligoclonal and directed against the autoantigenactin exposed on the surface of apoptotic cancer cells Proc Natl Acad Sci U S A200198(22)12659ndash12664

45 Coronella JA Spier C Welch M et al Antigen-driven oligoclonal expansion oftumor-infiltrating B cells in infiltrating ductal carcinoma of the breast JImmunol 2002169(4)1829ndash1836

46 Sainz-Perez A Lim A Lemercier B et al The T-cell receptor repertoire oftumor-infiltrating regulatory T lymphocytes is skewed toward public se-quences Cancer Res 201272(14)3557ndash3569

47 Buchbinder EI McDermott DF Cytotoxic T-Lymphocyte Antigen-4 Blockadein Melanoma Clin Ther 201537(4)755ndash763

48 Massari F Santoni M Ciccarese C et al PD-1 blockade therapy in renal cellcarcinoma current studies and future promises Cancer Treat Rev 201541(2)114ndash121

49 Mahmoud SM Paish EC Powe DG et al Tumor-infiltrating CD8thorn lympho-cytes predict clinical outcome in breast cancer J Clin Oncol 201129(15)1949ndash1955

50 Mahmoud SM Lee AH Paish EC et al The prognostic significance of B lym-phocytes in invasive carcinoma of the breast Breast Cancer Res Treat 2012132(2)545ndash553

51 Tosolini M Kirilovsky A Mlecnik B et al Clinical impact of different classesof infiltrating T cytotoxic and helper cells (Th1 th2 treg th17) in patientswith colorectal cancer Cancer Res 201171(4)1263ndash1271

52 Dalerba P Maccalli C Casati C et al Immunology and immunotherapy of co-lorectal cancer Crit Rev Oncol Hematol 200346(1)33ndash57

53 Pages F Berger A Camus M et al Effector memory T cells early metastasisand survival in colorectal cancer N Engl J Med 2005353(25)2654ndash2666

54 Le DT Uram JN Wang H et al PD-1 Blockade in Tumors with Mismatch-Repair Deficiency N Engl J Med 2015372(26)2509ndash2520

55 Hitchcock ER Morris CS Mononuclear cell infiltration in central portions ofhuman astrocytomas J Neurosurg 198868(3)432ndash437

56 Morantz RA Wood GW Foster M et al Macrophages in experimental and hu-man brain tumors Part 1 Studies of the macrophage content of experimentalrat brain tumors of varying immunogenicity J Neurosurg 197950(3)298ndash304

AR

TIC

LE

M D Iglesia et al | 11 of 11

at University of N

orth Carolina at C

hapel Hill on June 22 2016

httpjncioxfordjournalsorgD

ownloaded from