Death-receptor O-glycosylation controls tumor-cellsensitivity to the proapoptotic ligand Apo2L/TRAILKlaus W Wagner1,7,8, Elizabeth A Punnoose1,2,8, Thomas Januario1, David A Lawrence2, Robert M Pitti2,Kate Lancaster3, Dori Lee1, Melissa von Goetz1, Sharon Fong Yee4, Klara Totpal4, Ling Huw1,Viswanatham Katta3, Guy Cavet5, Sarah G Hymowitz6, Lukas Amler1 & Avi Ashkenazi2

Apo2L/TRAIL stimulates cancer cell death through the proapoptotic receptors DR4 and DR5, but the determinants of tumor

susceptibility to this ligand are not fully defined. mRNA expression of the peptidyl O-glycosyltransferase GALNT14 correlated

with Apo2L/TRAIL sensitivity in pancreatic carcinoma, non–small-cell lung carcinoma and melanoma cell lines, and up to 30%

of samples from various human malignancies showed GALNT14 overexpression. RNA interference of GALNT14 reduced cellular

Apo2L/TRAIL sensitivity, whereas overexpression increased responsiveness. Biochemical analysis of DR5 identified several

ectodomain O-(N-acetyl galactosamine–galactose–sialic acid) structures. Sequence comparison predicted conserved extracellular

DR4 and DR5 O-glycosylation sites; progressive mutation of the DR5 sites attenuated apoptotic signaling. O-glycosylation

promoted ligand-stimulated clustering of DR4 and DR5, which mediated recruitment and activation of the apoptosis-initiating

protease caspase-8. These results uncover a new link between death-receptor O-glycosylation and apoptotic signaling, providing

potential predictive biomarkers for Apo2L/TRAIL-based cancer therapy.

Apoptosis is critical for regulating cell number in normal metazoantissues, and its deregulation is a hallmark of malignancy1. Severalmolecular strategies designed to activate tumor cell apoptosis are inclinical investigation2. Since cancer is genetically diverse, biomarkersthat can help predict tumor sensitivity may be crucial for successfuldevelopment of apoptosis-targeted therapies3.

One approach aims to kill tumor cells by targeting the extrinsicapoptosis pathway through proapoptotic death receptors4,5. Apo2ligand (also known as tumor necrosis factor–related apoptosis-inducing ligand and herein referred to as Apo2L/TRAIL) triggersapoptosis through DR4 (TRAIL-R1) and/or DR5 (TRAIL-R2). Uponbinding, these receptors, which have apoptosis-inducing cytoplasmic‘death’ domains, bind the adaptor molecule Fas-associated deathdomain (FADD), which recruits the apoptosis-initiating proteasecaspase-8 into a death-inducing signaling complex (DISC)6,7. TheDISC stimulates autocatalytic caspase-8 processing, releasing activecaspase-8 into the cytoplasm, where it cleaves and activates effectorcaspases-3 and -7. Further stimulation is mediated by engagementof the intrinsic (mitochondrial) apoptosis pathway throughcaspase-8-dependent activation of the proapoptotic Bcl-2 familyprotein Bid1,8. Recombinant human Apo2L/TRAIL is in clinicalinvestigation because it can induce apoptosis in variouscancer cell types while sparing most normal cells5,9,10. Althoughgenetic or epigenetic changes such as Bax mutation or c-FLIP over-expression inhibit Apo2L/TRAIL activity in individual cancer cell

lines8,11, factors that may affect sensitivity in multiple cancers are notwell defined.

O-linked glycans regulate biochemical and functional properties ofcell surface proteins, including conformation, multimerization, traf-ficking and turnover; O-glycan deregulation contributes to Wiskott-Aldrich syndrome, hematological disorders and cancer12,13. O-glycanbiosynthesis involves generation and transport of nucleotide-sugardonors to the endoplasmic reticulum and Golgi apparatus, as well asactivity of glycosyltransferases and glycosidases14. The most commonform of protein O-glycosylation (mucin-type) is initiated bya-glycosidic linkage of N-acetyl galactosamine (GalNAc) to serine orthreonine side-chains, catalyzed by 24 peptidyl GalNAc transferase(GALNT) isoforms; further O-glycan processing is mediated by tenfucosyltransferases12,15. Here we describe a newly discovered mechan-ism that modulates Apo2L/TRAIL signaling in tumors throughdeath-receptor O-glycosylation.

RESULTS

O-glycosylation enzymes predict sensitivity to Apo2L/TRAIL

To identify potential determinants that broadly control tumor sensi-tivity to Apo2L/TRAIL, we investigated 119 human cancer cell lines,including 23 pancreatic adenocarcinomas, 42 non–small-cell lungcarcinomas (NSCLC), 18 malignant melanomas and 36 colorectaladenocarcinomas (Fig. 1). By measuring cell survival as a function ofApo2L/TRAIL concentration (see examples in Fig. 1a), we classified

Received 1 May; accepted 11 July; published online 2 September 2007; doi:10.1038/nm1627

1Departments of Molecular Diagnostics, 2Molecular Oncology, 3Analytical Development, 4Translational Oncology, 5Bioinformatics and 6Protein Engineering, Genentech,Inc., 1 DNA Way, South San Francisco, California 94080, USA. 7Present address: Indiana University School of Medicine, 1001 West 10th Street, Indianapolis, Indiana46202, USA. 8These authors contributed equally to this work. Correspondence should be addressed to A.A. (aa@gene.com).

1070 VOLUME 13 [ NUMBER 9 [ SEPTEMBER 2007 NATURE MEDICINE

ART ICL ES©

2007

Nat

ure

Pub

lishi

ng G

roup

ht

tp://

ww

w.n

atur

e.co

m/n

atur

emed

icin

e

100

75

50

25

4,000

3,000

2,000

1,000

750

0

Apo2L (ng/ml)

Pancreatic cancer NSCLC

Melanoma

0 200

500

1,00

0

1,50

0

2,00

0

3,00

0

2,50

0

P = 9 × 10–6

P = 0.026

P = 0.0013

P = 0.01

Cel

l via

bilit

y (%

)

0 10 100 1,000

2,000

DLD-1

PSN-1

BxPC-3 vehicleBxPC-3 Apo2LPSN-1 vehiclePSN-1 Apo2LHs294T

Panc0327T84

PATU8902928MEL

SW620

GA

LNT

14 m

RN

AG

ALN

T3

mR

NA

FU

T6

mR

NA

FU

T3

mR

NA

Mea

n pa

ncre

atic

tum

or v

olum

e (m

m3 )

Mea

n co

lore

ctal

tum

or v

olum

e (m

m3 )

1,500

1,000

500

00 5 10 15 20

2,000 HPAF II vehicleHPAF II Apo2LPATU8902 vehiclePATU8902 Ap02L1,500

1,000

500

00 5 10

Days15 20

2,000 Colo205 vehicleColo205 Apo2LDLD-1 vehicleDLD-1 Apo2L1,500

1,000

500

00

2,000

Colo320 vehicleColo320 Apo2LRKO vehicleRKO Apo2L

1,500

1,000

500

0

Days

5 10 15 20 25 30

0 5 10 15 20 25 30

0

2,000

4,000

6,000

8,000

0

1,400

1,200

1,000

800

600

400

200

0

3,500

3,000

2,500

2,000

1,500

1,000

500400

Normal

SCCMelanoma

NormalSCC

Normal

Panc Ca

Normal

Ductal Ca

Normal

Ovarian Ca

Pancreas

Lung

Skin

Breast

Ovary

Endometrium

Bladder

Lymph

Normal

Adenoca

NormalTCC

Normal

FLDLBCL

Lobular Ca

Adenoca

Large cellSmall cell

Pan

c 05

.04

CL-

11C

OLO

205

CO

LO 2

01C

OLO

-20

6FS

W94

8C

170

HC

T-1

5C

L-34

DLD

-1H

T55

LS10

34T

84S

K-C

O-1

HT

-29

CL-

40C

X-1

SW

837

KM

12S

W11

16S

W14

63LS

180

LoV

oC

OLO

-678

SW

403

Cac

o-2

HC

T-8

SW

1417

SW

480

HC

T 1

16C

OLO

741

SW

620

CO

LO 3

20D

MC

OLO

320

HS

RM

DS

T8

RK

O-A

S45

-1R

KO

Pan

c 03

.27

Pan

c 08

.13

PA

-TU

-898

8TH

s766

TP

anc

02.0

3

Pan

c 04

.03

Cap

an-2

Mia

PaC

a-2

PA

-TU

-890

2A

spc-

1H

2122

A42

7S

K-M

ES

-1H

647

H83

8H

2126

H35

8H

460

H16

66H

1568

H65

0H

322T

H17

93H

OP

18

A54

9H

1781

H23

H16

50H

1703

SW

1573

H24

05H

441

H11

55H

226

H20

30H

1651

H12

99ca

lu-1

LXF

L 52

9H

520

EK

VX

H19

75H

292

H59

6H

LFa

H18

38H

522

H66

1H

OP

92

H20

09H

1435

HO

P 6

2R

PM

I795

1H

s294

T

Hs6

95T

527M

EL

A20

58G

361

888

624

MA

LME

3MS

KM

EL2

892

8ME

LLO

XIM

VI

MeW

oS

K23

C32

CO

LO82

9

A37

5M

DA

-MB

-435

Pan

c 10

.05

Pan

c1H

PA

F-I

I

PL4

5

SU

.86.

86B

xPC

3H

uP-T

3P

SN

1C

FP

AC

-1C

apan

-1S

W 1

990

HP

AC

a

c

d e

b

Figure 1 Specific O-glycosylation enzyme expression correlates with sensitivity to Apo2L/TRAIL. (a) Effect of Apo2L/TRAIL on cell viability (mean ± s.d. of

triplicates, representative experiment of three shown). (b) Effect of Apo2L/TRAIL on growth of established tumor xenografts from cell lines either sensitive

(top panels) or resistant (bottom panels) in vitro to Apo2L/TRAIL (mean ± s.e.m., n = 10 mice per group). P values comparing time to tumor progression of

the vehicle and Apo2L/TRAIL groups: BxPC3 and PSN-1, P o 0.0001; HPAFII, P = 0.91, PA-TU-8902, P = 0.56; Colo205 and DLD-1, P o 0.0001;

Colo-320, P = 0.23; RKO, P = 0.18. (c) GALNT14 mRNA expression levels in pancreatic cancer, NSCLC and malignant melanoma cell lines. (d) GALNT3,

FUT6 and FUT3 mRNA expression levels in colorectal cancer cell lines. In c and d, black, gray, or open bars, respectively, depict cell lines in each tumor

type that are highly sensitive, moderately sensitive, or resistant to Apo2L/TRAIL; P values based on Fisher’s exact test for the correlation between cell line

sensitivity (high and moderate) and mRNA expression above cutoff (as indicated). (e) mRNA expression levels for GALNT14 in primary human tissue

samples. Gray, normal tissue; black, tumor tissue. SCC, squamous cell carcinoma; ca, carcinoma; lymph, lymphoma; adenoca, adenocarcinoma;

TCC, transitional cell carcinoma; FL, follicular lymphoma, DLBCL, diffuse large B-cell lymphoma. Gray horizontal bar for each class, median expressionof samples. Black bars, cutoff values separating high and low GALNT14 expressors.

ART ICL ES

NATURE MEDICINE VOLUME 13 [ NUMBER 9 [ SEPTEMBER 2007 1071

©20

07 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

urem

edic

ine

34% (40/119) as highly or moderately sensitive (see Methods), with50% cell death at 3–800 ng/ml (0.05–13 nM). We confirmed withseveral tumor xenografts that sensitivity was consistent in vitro andin vivo (Fig. 1b). Next, we used whole-genome profiling to identifygenes expressed most differentially between sensitive and resistantcells. Expression of GALNT14 mRNA, encoding the O-glycosylationinitiating enzyme GALNT14, was markedly higher in Apo2L/TRAIL-sensitive versus resistant pancreatic, NSCLC and melanoma cell lines(P o 9 � 10�6; Fisher’s exact test, n ¼ 83) (Fig. 1c). Exceptions were6/28 sensitive and 14/55 resistant lines with GALNT14 levels below orabove cutoff, respectively. Other GALNT isoforms on the gene chipwere less differentially expressed (Supplementary Table 1a online).Certain GALNT isoforms have distinct tissue expression but over-lapping substrate specificity15,16. In colorectal lines, levels of GALNT3rather than GALNT14 mRNA correlated significantly with sensitivity(P o 0.026, n ¼ 36) (Fig. 1d). Levels of FUT6 and FUT3 mRNAs,encoding O-glycan processing fucosyltransferase enzymes, also wereassociated with sensitivity in the colorectal lines (P o 0.0013 and0.01, respectively), more strongly than those encoding other fucosyl-transferases (Fig. 1d; Supplementary Table 1b). Exceptions were 2/12sensitive and 6/24 resistant colorectal lines expressing levels ofGALNT3, FUT6 and FUT3 below or above cutoff, respectively.

By contrast, cell surface expression of DR4 and DR5 or the relateddecoy receptors DcR1 (TRAIL-R3) and DcR2 (TRAIL-R4) across thepancreatic or colorectal panels did not generally correlate with Apo2L/TRAIL sensitivity (Supplementary Fig. 1a,b online). We also char-acterized expression of antiapoptotic proteins capable of inhibitingApo2L/TRAIL activity in specific cell lines: namely c-FLIP, Bcl-2, Mcl-1,Bcl-XL and XIAP, in 8 pancreatic and 13 colorectal cell lines (Supple-mentary Fig. 1c,d). Although in some individual lines one or more ofthese factors might contribute to resistance, none showed a consistentcorrelation, whereas GALNT14 or FUT6 protein was present in allsensitive but not in most resistant lines, consistent with the mRNA

levels (Fig. 1c,d). The growth rates of several sensitive and resistant celllines in vitro or in vivo were comparable (Supplementary Fig. 1e,f),indicating that proliferation capacity did not strongly affect Apo2L/TRAIL responsiveness. We analyzed the expression of GALNT14mRNA in normal and malignant tissue from skin, lung, pancreas,breast, ovary, endometrium, bladder and lymphoid cancers (Fig. 1e). Asubset of tumor samples, ranging from 10% in lobular breast cancer to30% in lung cancer and diffuse large B-cell lymphoma, showedGALNT14 mRNA overexpression. This rate was comparable to thefrequency of Apo2L/TRAIL sensitivity in Figure 1c,d and in a previouscell line panel9. Because GALNT14 expression in cancer is dynamic, itmay provide a useful biomarker for Apo2L/TRAIL sensitivity.

Modulation of O-glycosylation affects sensitivity to Apo2L/TRAIL

The latter correlations suggested a possible functional link betweenactivity of O-glycosylation enzymes and Apo2L/TRAIL signaling.Supporting this, preincubation of Colo205 cells with benzyl-a-GalNAc—a general inhibitor of O-glycosylation that competitivelyblocks formation of the GalNAc-Gal (N-acetylgalactosamine-galactose) core 1 structure and its subsequent sialylation17,18—mark-edly reduced sensitivity to Apo2L/TRAIL (Fig. 2a). To examine thisfurther, we designed specific small interfering (si) RNAs targetingGALNT14, GALNT3, or FUT6. We tested multiple siRNAs per targetand verified mRNA suppression by quantitative RT-PCR and proteindepletion by immunoblot with GALNT14 and FUT6 antibodies(Supplementary Fig. 2a online). Knockdown of GALNT14 inpositive PSN-1 pancreatic cancer or Hs294T melanoma cells sub-stantially reduced sensitivity to Apo2L/TRAIL, while caspase-8 knock-down provided complete protection (Fig. 2b). Confirming specificity,GALNT14 siRNA was less effective at depleting the target protein andreducing sensitivity in PSN-1 cells stably transduced with a GALNT14overexpression vector (Supplementary Fig. 2b). GALNT3 or FUT6knockdown in DLD-1 or C170 colorectal cells also markedly dimin-ished sensitivity (Fig. 2c; Supplementary Fig. 2c). In sum, knock-down of GALNT14 in 6/7 and GALNT3 or FUT6 in 2/3 cell linesdecreased Apo2L/TRAIL responsiveness (Supplementary Table 2online). By contrast, knockdown of GALNT14 in PSN-1 or Hs294Tcells or GALNT3 in C170 cells did not alter susceptibility to theintrinsic-pathway activators etoposide and staurosporine19 (Supple-mentary Fig. 2d,e), suggesting that O-glycosylation of specific extrin-sic-pathway component(s) promotes Apo2L/TRAIL signaling.

Death-receptor overexpression induces apoptosis by spontaneousdeath domain self-association20. Cotransfection of HEK293 cells withGALNT14 substantially augmented apoptosis induced by overexpres-sion of DR4 or DR5, but not apoptosis induced by overexpression ofthe related receptors Fas and TNFR1 or the intrinsic-pathway agonistBax (Fig. 3a). Cotransfection of GALNT14 with lower DR5 amountssignificantly increased apoptosis stimulation by Apo2L/TRAIL(Supplementary Fig. 3a online). Furthermore, stable GALNT14overexpression in two resistant lines expressing low endogenous

100

75

50

25

ControlDMSO (2)DMSO (4)bGaINAc (2 mM)bGaINAc (4 mM)

0 10Apo2L (ng/ml)

Col

o205

cel

l via

bilit

y (%

)

100 1,000

100

75

50

25

NTCsiC8siGALNT14 (1)siGALNT14 (2)

0 10Apo2L (ng/ml)

PS

N-1

cel

l via

bilit

y (%

)

100 1,000

100

75

50

25

NTCsiC8siGALNT3 (1)siGALNT3 (2)

0 10Apo2L (ng/ml)

DLD

-1 c

ell v

iabi

lity

(%)

100 1,000

100

75

50

25

NTCsiC8siGALNT14 (1)siGALNT14 (2)

0 10Apo2L (ng/ml)

Hs2

94T

cel

l via

bilit

y (%

)

100 1,000

100

75

50

25NTCsiC8siFUT6 (1)siFUT6 (2)

0 10Apo2L (ng/ml)

DLD

-1 c

ell v

iabi

lity

(%)

100 1,000

a

b

c

Figure 2 Inhibition of O-glycosylation enzymes reduces sensitivity to

Apo2L/TRAIL. (a) Colo205 cells were preincubated with benzyl-a-GalNAc

(bGalNAc) or DMSO vehicle, treated with Apo2L/TRAIL or buffer control

for 24 h, and assayed for cell viability. (b) PSN-1 and Hs294T cells were

transfected with siRNAs against caspase-8 (siC8) or GALNT14 for 48 h,

followed by Apo2L/TRAIL treatment for 24 h and assayed for cell viability.

Nontargeting control (NTC) siRNAs were used for comparison. Two different

siRNAs are shown for each enzyme; see Supplementary Methods for

sequences. (c) DLD-1 cells were transfected with GALNT3 or FUT6 siRNA

and tested as in b. Graphs are representative mean ± s.d. data from one of

three experiments performed in triplicate.

ART ICL ES

1072 VOLUME 13 [ NUMBER 9 [ SEPTEMBER 2007 NATURE MEDICINE

©20

07 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

urem

edic

ine

enzyme levels, H1568 and PA-TU-8902, increased Apo2L/TRAILactivity (Fig. 3b,c), without affecting sensitivity to a related ligand,tumor necrosis factor (TNF)-a (Supplementary Fig. 3b). Similarsensitization was observed with PL-45 and Panc 10.05 cells (notshown). In vivo, xenografts based on untransfected or vector-transfected PA-TU-8902 cells showed weak, insignificant inhibitionby Apo2L/TRAIL, whereas two clones overexpressing GALNT14showed a marked, statistically significant response (Fig. 3d). Thus,overexpression of GALNT14 can selectively augment apoptosis activa-tion by Apo2L/TRAIL.

O-glycans control death-receptor activation by Apo2L/TRAIL

Comparison of several colorectal cancer cell lines indicated thatApo2L/TRAIL-resistant cells had defective stimulation of caspase-8processing and correspondingly less cleavage of Bid, caspase-9 andcaspase-3; furthermore, resistant cells showed weaker DISC recruit-ment of FADD and caspase-8 (Supplementary Fig. 4a,b online).Consistently, GALNT14 knockdown in PSN-1 cells attenuated ligand-induced processing of caspase-8, Bid, caspase-9, and caspase-3(Fig. 4a), and caspase-3/7 activation (Fig. 4b). FUT6 knockdown inDLD-1 cells similarly attenuated caspase-8 processing (Fig. 4a). Con-versely, stable GALNT14 overexpression in PA-TU-8902 cells substan-tially augmented DISC formation (Supplementary Fig. 5a online)and processing of caspase-8 and caspase-3 (Fig. 4c).

Knockdown of GALNT14 in PSN-1 cells reduced DISC recruitmentof FADD and caspase-8, processing of DISC-bound caspase-8, anddevelopment of DISC-associated caspase-8 activity (Fig. 4d,e). Simi-larly, FUT6 knockdown in DLD1 cells reduced DISC activation ofcaspase-8 (Supplementary Fig. 5b). GALNT14 or FUT6 knockdowndid not appreciably alter the amount of DR4 and DR5 in the DISC(Fig. 4d and Supplementary Fig. 5b), suggesting that any potentialchanges in ligand binding or receptor levels were not the main causefor decreased caspase-8 recruitment. Moreover, transfection with twosiRNAs each against GALNT14, GALNT3, or FUT6 did not substan-

tially alter cell surface DR4 or DR5 levels (Supplementary Fig. 5c).Two of the oligonucleotides, siGALNT14 (1) and siGALNT3 (1), wereassociated with a slight decrease in receptor levels; however, since theother siRNAs against these enzymes reduced ligand sensitivity withoutchanging receptor amounts, the effect on signaling was unlikely to bedue to substantial alteration in receptor expression. Further support-ing this conclusion, treatment of Colo205 cells with benzyl-a-GalNAcor stable GALNT14 overexpression in PA-TU-8902 cells had no majoreffect on surface DR4 and DR5 levels (Supplementary Fig. 5d,e).

Identification of potential O-glycosylation sites in DR4 and DR5

O-glycan attachment typically modifies proteins destined for the cellsurface12,13. We reasoned therefore that DR4, DR5 or both might bethe relevant O-glycosylation targets. Cotransfection with GALNT14augmented the apoptosis-inducing activity of overexpressed chimericreceptors containing the DR4 or DR5 extracellular domains (ECDs)fused to the transmembrane and intracellular region of Fas (Supple-mentary Fig. 6a online), consistent with potential O-glycanmodification of the DR4 and DR5 ECDs. We carried out a moredetailed analysis of DR5 by expressing its ECD in Chinese hamsterovary cells and analyzing the secreted protein by liquid chromato-graphy–mass spectrometry (LCMS). As well as the parent DR5 peak atthe expected mass, the spectrum showed peaks corresponding toattachment of one to four O-linked GalNAc-Gal–sialic acid moieties,each with up to two sialic acids (Fig. 5a, top). Sialidase treatmentcollapsed these peaks to masses corresponding to the four sites ofGalNAc-Gal addition (Fig. 5a, bottom).

The specificity of GALNTs depends on substrate sequence, second-ary structure and surface accessibility15,21. O-glycosylation sites areoften flanked by serine, threonine, alanine and proline, with negativelycharged residues disfavored at the –1 and +3 positions. A bioinfor-matics tool (NetOglyc) trained on verified protein O-glycosylationsites21 predicted two such regions in the common ECD sequence ofthe long (DR5L) and short (DR5S) human DR5 splice variants, and a

50100

75

50

25

250P = 0.36

P = 0.09

Par

GALNT14

Actin

Vec

Vec

C1

C1

GALNT14

C2

C2

P = 0.0002

P = 0.0039

Vehicle Apo2L

200

150

100

50

0

Vector

Vec

– + – + – + – + – +

DR5L DR4

Flag-GALNT14

100

75

50 V G

GT14

Actin

0 10 100Apo2L (ng/ml)

1,000

GALNT14

Vector25

FasFlag-GALNT14

DR5L DR4 Fas TNFR1 Bax

Bax

0 10

V G

GT14

Actin GALNT14

Apo2L (ng/ml)

Vector

100 1,000

**

VectorGALNT14

Ann

exin

V s

tain

ing

(%)

PA

-TU

-890

2 ce

ll vi

abili

ty (

%)

Tum

or v

olum

e (m

m3 )

H15

68 c

ell v

iabi

lity

(%)

40

30

20

10

0

a b

d

c*

*

Figure 3 Overexpression of O-glycosylation

enzymes promotes sensitivity to Apo2L/TRAIL.

(a) HEK293 cells were cotransfected with

plasmids encoding each indicated receptor or

Bax in combination with GALNT14 or a vector

control. Apoptosis was measured at 48 h by

Annexin V staining. The corresponding Flag

immunoblot for Flag-tagged GALNT14 is

indicated (bottom). *P o 0.05 based on t-test.

(b,c) H1568 or PA-TU-8902 cells were

transduced with retrovirus directing stable

GALNT14 (GT14) expression or control retrovirus;

resulting cell line pools were treated with Apo2L/

TRAIL for 24 h and cell viability determined.

Graphs, representative mean ± s.d. data fromone of two experiments performed in triplicate.

Immunoblot analyses of Flag-tagged GALNT14

are shown (insets). V, vector; G, GALNT14.

(d) Apo2L/TRAIL sensitivity of tumor xenografts.

Shown are day-10 tumor volumes (mean ±

s.e.m., n ¼ 10 mice per group) of xenografted

PA-TU-8902 cells (untransfected) or those

expressing an empty vector (vec) or two clones

overexpressing GALNT14 (C1, C2). Mice were

treated with vehicle or Apo2L/TRAIL starting at

day of implant. *P o 0.05, t-test comparing the

vehicle and Apo2L/TRAIL arms for each xenograft

type. Flag immunoblot indicates expression of

Flag-GALNT14 in stably transfected lines.

ART ICL ES

NATURE MEDICINE VOLUME 13 [ NUMBER 9 [ SEPTEMBER 2007 1073

©20

07 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

urem

edic

ine

third region within the alternatively spliced DR5 region (Fig. 5b). Thefirst region (amino acids 74–77) contains three serines; the second(amino acids 130–144) has five threonines; the third (amino acids184–212) has four threonines and three serines. Human DR4 showssequence similarity with the first two regions, containing one serineand five threonines, while mouse DR5 has two serines and fourthreonines, respectively. By contrast, human Fas and TNFR1 show

less conservation. To test whether the predicted DR5 sites might beimportant for post-translational modification, we generated a set ofDR5L and DR5S mutants, replacing by alanines either the fivethreonines of region 2 (amino acids 130–144) (DR5L-5T, DR5S-5T)or these same five threonines as well as the three serines of region 1(amino acids 74–77) (DR5L-5T3S, DR5S-5T3S). Immunoblot analysisof lysates from HEK293 cells transfected with wild-type DR5L or

siControl

PSN-1

Apo2L (h):p55, 53p43, 41IB: C8

Bid

C9

C3

Actin

p32

p50

0 4 8 0 4 8 0 4 8 0 4 8 0 4 8 0 4 8

DLD-1

siControlsiCasp8 siCasp8 siFUT6

500

Cas

pase

-3/7

activ

ity (

RF

U)

400300200100

00 125

siControl

siControlFlag-Apo2L

IP: Flagp55, 53

p43, 41

FADD

L

SDR5

DR4

IB: C8

IB: C8

min: 0 15 60 0 15 60

siGALNT14

siGALNT14

125

* *

1,000 1,000

P < 0.05

P < 0.05

0ng/ml:

siGALNT14

DIS

C c

aspa

se-8

activ

ity (

RF

U ×

10–3

)

600

400

200

0

p55, 53

p43, 41

p32

Actin

C3

Apo2L (h): 0 4Vec C1 C2

8 24 0 4 8 24 0 4 8 24

GALNT14

siGALNT14siControl

0 15 60 0 15 60min:

a b

d

c

e

**

Figure 4 Modulation of O-glycosylation enzymes affects caspase-8

activation. (a) Analysis of Apo2L/TRAIL–induced caspase signaling. PSN-1

and DLD-1 cells were transfected with siRNAs against GALNT14, FUT6 or

caspase-8 (siCasp8) for 48 h. The cells were treated with Apo2L/TRAIL

(1 mg/ml) for the indicated times and cell lysates were analyzed by

immunoblot (IB) with antibodies to caspase-8 (C8), Bid, caspase-9 (C9) and caspase-3 (C3), or to actin as a loading control. Number after ’p’ indicates

protein molecular mass (kDa). (b) PSN-1 cells were transfected with GALNT14 siRNA as in a, treated with Apo2L/TRAIL for 4 h, and enzymatic activity

of caspases 3 and 7 (mean ± s.d.) in cell lysates was measured in triplicate samples. *P o 0.05, Student’s t-test comparing siGALNT14 to siControl.

(c) PA-TU-8902 clones stably expressing GALNT14 (C1, C2) or empty vector (vec) were treated with Apo2L/TRAIL (1 mg/ml) for the indicated times and

lysates were analyzed by immunoblot as indicated. (d) Analysis of the Apo2L/TRAIL DISC. PSN-1 cells were transfected with GALNT14 siRNA as in a,

incubated with Flag-Apo2L/TRAIL (1 mg/ml) for 15 or 60 min or after lysis for the 0 time point, and lysates were subjected to DISC immunoprecipitation

with Flag antibody. DISC-associated FADD, caspase-8, DR4, and DR5 were detected by immunoblot. L and S, DR5 long and short splice variants,

respectively. (e) Same as d with immunoprecipitated sample then subjected to a caspase-8 enzymatic activity assay as previously described37.Replicates and statistics were done as in b.

100

Theoretical mass:

18,521.44

0 sites

2 sitesSA

GalNAc-Gal

3 sites

4 sites

4 Sites

GalNAc-GalGALNT14:

30

25

20

15

10

5

Vector GALNT14

0– + – + – + – + – + – + – +

GALNT14

DR5 LSS*

*L

3 Sites2 Sites

0 Sites 1 Site

1 site

1.82 1.86 1.90 1.94 1.98 2.02 2.06 2.10 2.14

1.80 1.961.921.88Mass, Da × 104

1.84 2.00 2.04

90

80

70

60

50

Inte

nsity

, c.p

.s.

40

30

20

180

120

80

40

10

0

0

Inte

nsity

, c.p

.s.

hDR5L 72 hDR5S 72 hDR4 123

mDR5 65 hFAS 64

hTNFR1 75

hTNFR1 207 hFAS 168

mDR5 173hDR4 233

hDR5S 182hDR5L 182

hDR5L 128hDR5S 128

hDR4 178hDR5 120hFAS 118

hTNFR1 157

DR

5L

DR

5L-5

T

DR

5L-5

T3S

DR

5S

DR

4

Ann

exin

V s

tain

ing

(%)

Vec

tor

mD

R5

mD

R5

DR

4

DR

5L

DR

5S

DR

4

DR

5L

DR

5S

DR

5L-5

T3S

DR

5S-5

T3S

DR

5L-5

T3S

DR

5S-5

T3S

Vec

tor

DR

5S-5

T

DR

5S-5

T3S

a b

d

c

Figure 5 Identification of potential O-glycosylation sites in the ectodomain of DR4 and DR5. (a) Top, reconstructed mass spectrum for DR5L ECD. The

theoretical mass of the protein (18,521.44 Da) was observed for the parent ion (zero sites). The other labeled peaks correspond to the addition of one to

four O-linked glycans of the structure GalNAc-Gal (365 Da) with zero to two sialic acids (SA) (293 Da) each. Mass differences are illustrated by gray arrows

corresponding to one sialic acid and black arrows for one GalNAc-Gal. Bottom, reconstructed mass spectrum of sialidase-treated DR5L ECD. The primary

mass difference, 365 Da, corresponds to the GalNAc-Gal moiety and the ions observed correspond to one to four sites of glycosylation. The minor peaks

correspond to individual differences of GalNAc or Gal in the oligosaccharide structure. c.p.s., counts per second. (b) Sequence comparison of human Apo2L/TRAIL receptors (hDR5L, hDR5S and hDR4), mouse Apo2L/TRAIL receptor (mDR5), human Fas (hFas) and human TNFR1 (hTNFR1). Residue numbers are

based on immature polypeptides. Boxes, putative DR4/5 O-glycosylation sites; red, amino acids predicted to be modified. (c,d) HEK293 cells were

transfected with vector or GALNT14 expression plasmid, together with the indicated receptor constructs, for 48 h. Cells were subsequently lysed and

analyzed by DR5 or DR4 immunoblot (c) or stained with Annexin V to quantify apoptosis (d). Asterisks in c, bands with higher molecular weights

corresponding to modified forms of the indicated receptor. L and S, DR5 long and short splice variants, respectively.

ART ICL ES

1074 VOLUME 13 [ NUMBER 9 [ SEPTEMBER 2007 NATURE MEDICINE

©20

07 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

urem

edic

ine

DR5S or DR4 revealed bands with expected molecular masses as wellas bands with higher mass (Fig. 5c); cotransfection with GALNT14enriched the higher molecular mass bands. The DR5L-5T and DR5S-5T mutants had notable diminution of, and DR5L-5T3S and DR5S-5T3S showed near absence of, the high molecular mass bands(Fig. 5c), supporting the relevance of these sites for DR5 modification.Concurrently, we assessed the importance of the modification sites forapoptotic signaling by DR5. Overexpression of human DR4, DR5L orDR5S or the single mouse homolog, mDR5 (TRAIL-R), inducedconsiderable cell death (Fig. 5d); each DR5 mutant showed lessapoptosis than its corresponding wild-type construct, with weakestactivity for DR5S-5T3S, which harbors mutations in all three regions.GALNT14 cotransfection markedly enhanced apoptosis induction byall the constructs; however, DR5S-5T3S showed considerably lessactivity. By contrast, DR5L-5T3S, which retains the unmutated alter-natively spliced region, was not affected, suggesting some functionalredundancy between DR5 O-glycan sites. Treatment with sialidaseplus O-glycanase but not with N-glycanase decreased abundance ofthe high molecular mass DR5 band (Supplementary Fig. 6b), furtherconfirming DR5 modification by O-GalNAc-Gal–sialic acid. Incuba-tion of a synthetic DR5 peptide (amino acids 195–210) with purifiedGALNT14, followed by tandem mass spectrometric analysis, identifiedSer201 as the primary site of modification in this region (Supple-mentary Fig. 6c), indicating that GALNT14 may directly modify DR5.

O-glycosylation promotes ligand-induced receptor clustering

The crystal structure of the Apo2L/TRAIL–DR5 complex22 indicatedthat the serines and threonines in DR5 regions 1 (amino acids 74–77)and 2 (amino acids 130–144) are external to the ligand binding site(Supplementary Fig. 7a online), consistent with the notion thatmodification at these sites might regulate events distinct from ligandbinding. Further supporting this, Apo2L/TRAIL bound with similar

affinity constants to glycosylated or deglycosylated DR5, generated byDR5 cotransfection with GALNT14 or by treatment of the latter withO-glycanase and sialidase (Supplementary Fig. 7b–d).

Death receptors can undergo pre-ligand association in homo-oligomeric complexes and further clustering upon ligand binding23,24.Ligand-induced clustering of Fas, which can be detected by gelelectrophoresis as formation of SDS-insoluble aggregates, is importantfor caspase-8 activation25. Apo2L/TRAIL stimulation of PSN-1 cellspromoted the formation of SDS-stable, high molecular mass forms ofDR4 and DR5 (Fig. 6a). Coimmunoprecipitation showed predominantassociation of caspase-8 with these high molecular mass receptorforms, selectively upon Apo2L/TRAIL stimulation (Fig. 6b);GALNT14 knockdown decreased this association (Fig. 6c). To examinereceptor clustering further, we analyzed by size-exclusion columnchromatography PSN-1 cells lysed with the milder detergent CHAPS.DR4 and DR5 eluted at oligomeric sizes, consistent with pre-ligandassociation (Fig. 6d). Apo2L/TRAIL shifted DR4 and DR5 into highermolecular mass fractions, indicating further receptor clustering. Ligandstimulation recruited a small amount of FADD and caspase-8 into thehigh molecular mass fractions that contained DR4 and DR5 (Fig. 6d).Most of the caspase-8 in these fractions was processed, indicating itsfull activation, as confirmed by measurement of receptor-associatedcaspase-8 activity (Fig. 6e). Similar results were observed in Jurkat andH460 cells (data not shown). Notably, GALNT14 knockdownimpaired the ligand-induced translocation of DR4 and DR5 as wellas FADD and caspase-8 into the high molecular mass fractions, ascompared to control siRNA (Fig. 6f). Furthermore, benzyl-a-GalNAcpretreatment attenuated the formation of ligand-induced, SDS-stableDR4 and DR5 oligomers as detected by electrophoresis (Fig. 6g).These results indicate that receptor O-glycosylation facilitates ligand-induced clustering of DR4 and DR5, which in turn promotes DISCrecruitment and caspase-8 activation.

a b c d f

15 16 17 18 19 20Fraction

bGal

NA

c

bGal

NA

c

– +Apo2L:

Apo2L min:

: Apo

2L

+ – + +

21 22 23 24 25 26

Control

siGALNT14

siGALNT14

siGALNT14

siGALNT14

siControl

siControl

siControl

siGALN

T14

siCon

trol

siGALN

T14

siCon

trol

siControl3028262422201816Fraction:

e g

DR5

DR4

p55, 53p43, 41

p55, 53p43, 41

P23

p23

FADD+

+

+

+

–

–

–

–

C8

C8

LSLS

KDa: 5385135346553438 274 171 107 67700

217

3028262422201816Fraction:

DR5IB: IB:

DR5

39

DR4IB:

Caspase-8

DR5DR4

0 15 60 0 15 60

IB:DR5DR4IB:

Caspase-8

DR4

p55, 53p43, 41

p55, 53p43, 41

p23

p23

FADD

C8

C8

LSLS

KDa: 5385135346553438 274 171 107 67700

217

IB: DR4 DR5

Apo2L

Cas

pase

-8 a

ctiv

ity

(lum

ines

cenc

e un

its)

0

10,000

20,000

30,000

40,000

50,000

516497

***

191kDa

IP: C8

IP: C8

Apo2L min: 0 15 60 0 15 60

IP: Flag

Figure 6 Inhibition of O-glycosylation impairs Apo2L/TRAIL-induced

receptor clustering. (a) PSN-1 cells were stimulated with Flag-Apo2L/

TRAIL for the indicated time, or treated after lysis (0 min). Cell lysateswere immunoprecipitated (IP) with Flag antibody, resolved by non-

reducing SDS-PAGE (SDS-PAGE) and immunoblotted (IB) with DR4 or

DR5 antibodies. (b) Cells were treated as in a, but subjected to IP with

caspase-8 (C8) antibody followed by IB as in a. Arrows in a and b,

monomeric or oligomeric receptor forms; asterisks, IgG background

bands. Molecular mass markers in c apply also to a and b. (c) PSN-1

cells were transfected with siRNA against GALNT14 or a control siRNA

for 48 h. The cells were treated with Apo2L/TRAIL and analyzed as

in b. (d) Cell lysates from untreated or Apo2L/TRAIL-treated PSN-1

cells were resolved on a Superdex 200 size-exclusion chromatography

column and the eluted fractions analyzed by immunoblotting as

indicated. Protein standards were used to approximate molecular mass (kDa) of proteins in each fraction. L, S and p are defined as in Figure 4. (e) Fractions

containing DR4 and DR5 (15-26) were immunoprecipitated with a DR4 antibody and associated caspase-8 activity measured (representative experiment is

shown). (f) PSN-1 cells were transfected with control or GALNT14 siRNA as in c, then stimulated with Apo2L/TRAIL and analyzed as in d. (g) Colo205 cells

were treated with benzyl-a-GalNAc (b-Gal) or vehicle (DMSO), stimulated with Flag-Apo2L/TRAIL and analyzed as in a.

ART ICL ES

NATURE MEDICINE VOLUME 13 [ NUMBER 9 [ SEPTEMBER 2007 1075

©20

07 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

urem

edic

ine

DISCUSSION

Our results provide the first example of apoptosis modulationby death-receptor O-glycosylation. Elevated expression of specificO-glycosyltransferases, which may support receptor modification incertain cancer cells, significantly correlated with sensitivity to theproapoptotic ligand Apo2L/TRAIL. The O-glycosylation initiatingenzyme GALNT14 showed a strong link to Apo2L/TRAIL sensitivityin pancreatic carcinoma, NSCLC and melanoma, whereas expressionof another initiating enzyme, GALNT3, along with the O-glycanprocessing enzymes FUT3 and FUT6, correlated with responsivenessin colorectal cancer cells. Up to 30% of samples from various humancancers showed relative GALNT14 mRNA overexpression at levelscomparable to those in Apo2L/TRAIL-sensitive cell lines. Together,these findings suggest that levels of specific O-glycosylation enzymesmay be helpful for identifying cancer patients who are more likely torespond to Apo2L/TRAIL-based therapy. Combined expression of thefour enzymes strongly correlated with sensitivity (P o 8.3 � 10�7,n ¼ 119), reinforcing the link between O-glycans and apoptoticsignaling. Several cell lines were refractory to Apo2L/TRAIL despitehigh expression of the relevant O-glycosylation enzymes, perhapsreflecting secondary resistance mechanisms. Nonetheless, based onthe cell line data, the presence of these enzymes would correctlypredict sensitivity (positive predictive value) 61% of the time, whileabsence would correctly predict resistance (negative predictive value)in 88% of cases, providing potentially useful criteria for patientinclusion or exclusion in clinical trials with Apo2L/TRAIL–basedtherapy. It is noteworthy that well established biomarkers, such asBcr-Abl fusion in chronic myeloid leukemia or Her2 (also known asNeu) amplification in breast cancer, also do not fully predict respon-siveness to corresponding therapies, probably because of the complexunderlying biology.

The observed association suggested that specific O-glycosylationenzymes might be involved in post-translational modification ofapoptosis components. We confirmed a functional connection betweenthe two cellular processes both by loss- and gain-of-function experi-ments. Pharmacologic inhibition of O-glycosylation enzymes or siRNAknockdown of GALNT14, GALNT3, or FUT6 inhibited, whereasGALNT14 overexpression enhanced, Apo2L/TRAIL sensitivity. Knock-down of GALNT14 or FUT6 attenuated apoptosis activation by Apo2L/TRAIL but not by other stimuli, including TNF-a, suggesting modula-tion of dedicated Apo2L/TRAIL pathway component(s). Detailedanalysis of DR5 confirmed modification by up to four GalNAc-Gal–sialic acid moieties with up to two sialic acids per O-glycan. A syntheticpeptide based on the alternatively spliced DR5 region was O-glycosy-lated by GALNT14, indicating that DR5 may be a direct target for thisenzyme, and raising the possibility that alternative mRNA splicingfurther modulates DR5 O-glycan modification. Sequence analysispredicted highly conserved O-glycosylation sites in both DR4 andDR5 but not in Fas or TNFR1, consistent with the observed selectiveeffects on Apo2L/TRAIL signaling. Further mechanistic studies showedthat O-glycosylation did not substantially alter DR4 or DR5 cell surfacelevels or Apo2L/TRAIL binding affinity; rather, it promoted ligand-induced receptor clustering, an event that was associated with efficientDISC recruitment and caspase-8 activation.

Among various aspects of cellular function, O-glycans can regulatesignaling pathways. Examples include the inhibition of T-cell activa-tion through increased CD45 dimerization26 and enhancement ofNotch ligand-receptor interactions27. O-glycans also can influenceT-cell fate: an increase in unsialylated core 1 O-glycans promotesapoptosis of CD8+ T effector cells while supporting survival ofmemory cells28. Cancers often show marked alterations in glycan

profiles, creating unique tumor-associated carbohydrate antigens orsupporting metastatic homing of tumor cells29–34. Our results suggestthat O-glycosylation of death receptors in cancer cells modulatessensitivity to Apo2L/TRAIL by promoting ligand-induced receptorclustering and consequent caspase-8 activation. These findings providenew insight into the control of cell-extrinsic apoptotic signaling andsuggest the potential utility of specific O-glycosylation enzymes ortheir modified targets as predictive biomarkers for Apo2L/TRAIL-based cancer therapy.

METHODSReagents and cell lines. We obtained reagents and cell lines from established

sources (Supplementary Methods online).

Cell viability and apoptosis assays. Cell viability was determined by MTT or

CellTiter-Glo assay and apoptosis was measured by Annexin V staining or

caspase activity (Supplementary Methods). We defined a cell line as highly

sensitive if less than 50% of the cells were viable at an Apo2L/TRAIL

concentration of 1 mg/ml in a 72 h assay, when grown in the presence of low

(0.5%) and high (10%) serum concentration. Cells were called moderately

sensitive if they showed less than 50% viability in at least one but not all three

biological repeat experiments or at only one of the two serum concentrations.

We defined a cell line as resistant if greater than 50% of the cells were viable in

response to an Apo2L/TRAIL concentration of 1 mg/ml in the presence of low

and high serum concentration.

Microarray hybridization and data analysis. We prepared total cellular RNA

from untreated cells and hybridized labeled cRNA to oligonucleotide micro-

arrays consisting of 54,613 gene probes (U133P GeneChip; Affymetrix) as

described previously35,36 (for details see Supplementary Methods). Probe set

219271_at was used to examine mRNA expression of GALNT14 in pancreatic

cancer, NSCLC and malignant melanoma cell lines. To qualify high versus low

expressors, cutoff values were assigned at 750 arbitrary units (AU) for

pancreatic carcinoma and NSCLC and 300 AU for melanoma. mRNA expres-

sion levels of GALNT3, FUT6 and FUT3 in colorectal cancer cell lines were

examined with the following probe sets: 211885_x_at, GALNT3; 203397_s_at,

FUT6; 214088_s_at, FUT3. Cutoffs were set as follows: GALNT3, 2,000 AU;

FUT6, 200 AU; FUT3, 400 AU. To determine the expression of GALNT14 in

primary tumors, we extracted gene expression profiles from the commercially

available database BioExpress (GeneLogic) as described36. Cutoff values were

set at 500 AU for most tumor samples and 200 AU for skin cancers. Positive

predictive value was measured using the formula (number of sensitive marker-

positive samples / total marker-positive samples); negative predictive value by

(number of resistant marker-negative samples / total marker-negative samples).

Mice and xenograft studies. All animals used for the study were female

athymic nu/nu mice (Charles River Laboratory).

Established subcutaneous xenograft model. Mice were inoculated subcuta-

neously with 5 � 106 cells per mouse, with ten mice per treatment group. Mice

with established tumors of B150 mm3 were treated intraperitoneally with

Apo2L/TRAIL (60 mg per kilogram body weight per day) or vehicle for one or

two cycles as indicated. Each treatment cycle constituted 5 d of consecutive

treatment followed by 2 d of no treatment.

Adjuvant subcutaneous xenograft model. We inoculated mice (as described

above) with PA-TU-8902 untransfected or stably transfected cell lines expres-

sing GALNT14 or an empty vector. Treatment was administered starting on the

day of cell implant with vehicle or with Apo2L/TRAIL and continued for two

treatment cycles. See Supplementary Methods for detail on tumor measure-

ment and statistical analysis.

Expression constructs and retroviral transduction. We generated cDNA and

retroviral constructs and mutant DR5 variants using standard methodology

(Supplementary Methods).

siRNA and benzyl-a-GalNAc studies. The siRNAs against GALNT14,

GALNT3, FUT6 and caspase-8 as well as the nontargeting control siRNA were

from Dharmacon. Cells were transfected with siRNA at a concentration of

ART ICL ES

1076 VOLUME 13 [ NUMBER 9 [ SEPTEMBER 2007 NATURE MEDICINE

©20

07 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

urem

edic

ine

25 nM using Lipofectamine 2000 (Invitrogen). Following a 48-h incubation,

cells were harvested for mRNA and protein analysis, or incubated with Apo2L/

TRAIL, etoposide or staurosporine for a further 24–72 h for viability assays,

15 min–1 h for DISC experiments, or 4–24 h for western blot analysis. (For

detail on transfection and siRNA sequences see Supplementary Methods). For

inhibition of O-glycosylation with benzyl-a-GalNAc, Colo205 cells were grown

in the presence of benzyl-a-GalNAc (Calbiochem) at a concentration of 4 mM

for 72 h. The cells were then replated and allowed to adhere for 24 h, while still

in the presence of the inhibitor, for follow up analysis.

Glycosidase treatment and carbohydrate analysis by mass spectrometry.

Deglycosylation by glycosidase treatment was carried out on recombinant, CHO

cell–derived DR5L ECD or on DR5 immunoprecipitates from HEK293 cell

lysates, using the enzymatic deglycosylation kit (Prozyme). Samples were

incubated with glycosidase(s) for 12 h at 37 1C under nondenaturing conditions,

as per the manufacturer’s protocol. Mass spectrometric analysis of recom-

binant DR5L ECD was performed using liquid chromatography/mass

spectrometry on an Applied Biosystems Q-Star Pulsar I with an Agilent

1100 liquid chromatograph.

Receptor clustering and gel filtration analysis. To detect ligand-induced high

molecular mass receptor complexes by immunoprecipitation, 1 � 107 cells were

stimulated with Flag-Apo2L/TRAIL (1 mg/ml, for 1h), lysed with a 1% Triton-

X100 lysis buffer followed by immunoprecipitation with a Flag antibody or

caspase-8 antibody and run on an SDS-PAGE gel under nonreducing condi-

tions. Ligand-induced high molecular mass receptor complexes were also

analyzed by gel filtration. 1 � 108 cells were stimulated with Apo2L/TRAIL

(1 mg/ml, for 1 h) followed by cell lysis (in Tris-buffered saline containing

0.82% CHAPS, and protease inhibitor cocktail (Roche)). Lysates were spun at

14,000g and supernatants filtered through a 0.8-mm filter and loaded onto a

Superdex 200 10/300 GL column (GE Healthcare). Fractions were collected in

0.5-ml volumes and concentrated by the chloroform-methanol method for

western blot analysis by SDS-PAGE or immunoprecipitated with a DR4

antibody (4G7) for caspase-8 activity assay.

Statistical analysis. We assessed the statistical significance of the difference

between two sets of data using an unpaired, two-tailed t-test or a paired t-test

for control and experimental data groups that could be paired. Differences were

considered to be statistically significant at the 5% level. For details on statistics

for microarray expression analysis, see Supplementary Methods.

Accession numbers. GEO microarray data, GSE8332.

Note: Supplementary information is available on the Nature Medicine website.

ACKNOWLEDGMENTSWe thank S. Marsters and M. Nagel for plasmids and purification of recombinantDR5 respectively; S. Ross and M. Go for execution of xenograft studies; andW. Forrest for statistical analysis.

AUTHOR CONTRIBUTIONSK.W.W., D.L. and M.V.G. performed the cell line characterization. K.W.W.and G.C. performed the microarray analysis. E.A.P., T.J., D.A.L., R.M.P. and K.T.performed the functional and mechanistic studies. E.A.P., L.H., K.L. and V.K.performed the glycosylation and mass spectrometric analyses. S.F.Y. conductedthe in vivo experiments. S.G.H. carried out the structural modeling. K.W.W.,E.A.P., L.A. and A.A. guided the project and contributed to the experimentaldesign and to data interpretation. K.W.W., E.A.P. and A.A. wrote the manuscript.

COMPETING INTERESTS STATEMENTThe authors declare competing financial interests: details accompany the full-textHTML version of the paper at http://www.nature.com/naturemedicine/.

Published online at http://www.nature.com/naturemedicine

Reprints and permissions information is available online at http://npg.nature.com/

reprintsandpermissions

1. Danial, N.N. & Korsmeyer, S.J. Cell death: critical control points. Cell 116, 205–219(2004).

2. Fesik, S.W. Promoting apoptosis as a strategy for cancer drug discovery. Nat. Rev.Cancer 5, 876–885 (2005).

3. Dalton, W.S. & Friend, S.H. Cancer biomarkers—an invitation to the table. Science312, 1165–1168 (2006).

4. Ashkenazi, A. & Dixit, V.M. Death receptors: signaling and modulation. Science 281,1305–1308 (1998).

5. Ashkenazi, A. Targeting death and decoy receptors of the tumour-necrosis factorsuperfamily. Nat. Rev. Cancer 2, 420–430 (2002).

6. Kischkel, F.C. et al. Cytotoxicity-dependent APO-1 (Fas/CD95)-associated proteinsform a death-inducing signaling complex (DISC) with the receptor. EMBO J. 14,5579–5588 (1995).

7. Kischkel, F.C. et al. Apo2L/TRAIL-dependent recruitment of endogenous FADD andcaspase-8 to death receptors 4 and 5. Immunity 12, 611–620 (2000).

8. Igney, F. & Krammer, P. Death and anti-death: tumour resistance to apoptosis. Nat. Rev.Cancer 2, 277–288 (2002).

9. Ashkenazi, A. et al. Safety and antitumor activity of recombinant soluble Apo2 ligand.J. Clin. Invest. 104, 155–162 (1999).

10. Kelley, S. & Ashkenazi, A. Targeting death receptors in cancer with Apo2L/TRAIL. Curr.Opin. Pharmacol. 4, 333–339 (2004).

11. LeBlanc, H. et al. Tumor-cell resistance to death receptor–induced apoptosis throughmutational inactivation of the proapoptotic Bcl-2 homolog Bax. Nat. Med. 8, 274–281(2002).

12. Hang, H. & Bertozzi, C. The chemistry and biology of mucin-type O-linked glycosyla-tion. Bioorg. Med. Chem. 13, 5021–5034 (2005).

13. Hanisch, F. O-glycosylation of the mucin type. Biol. Chem. 382, 143–149 (2001).14. Ohtsubo, K. & Marth, J.D. Glycosylation in cellular mechanisms of health and disease.

Cell 126, 855–867 (2006).15. Ten Hagen, K.G., Fritz, T.A. & Tabak, L.A. All in the family: the UDP-GalNAc:

polypeptide N-acetylgalactosaminyltransferases. Glycobiology 13, 1R–16R (2003).16. Wang, H. et al. Cloning and characterization of a novel UDP-GalNAc:polypeptide

N-acetylgalactosaminyltransferase, pp-GalNAc-T14. Biochem. Biophys. Res.Commun. 300, 738–744 (2003).

17. Delannoy, P. et al. Benzyl-N-acetyl-a-D-galactosaminide inhibits the sialylation and thesecretion of mucins by a mucin secreting HT-29 cell subpopulation. Glycoconj. J. 13,717–726 (1996).

18. Kuan, S.F., Byrd, J.C., Basbaum, C. & Kim, Y.S. Inhibition of mucin glycosylation byaryl-N-acetyl-a-galactosaminides in human colon cancer cells. J. Biol. Chem. 264,19271–19277 (1989).

19. Wei, M. et al. Proapoptotic BAX and BAK: a requisite gateway to mitochondrialdysfunction and death. Science 292, 727–730 (2001).

20. Boldin, M.P. et al. Self-association of the ‘‘death domains’’ of the p55 tumor necrosisfactor (TNF) receptor and Fas/APO1 prompts signaling for TNF and Fas/APO1 effects.J. Biol. Chem. 270, 387–391 (1995).

21. Hansen, J.E. et al. NetOglyc: prediction of mucin type O-glycosylation sites based onsequence context and surface accessibility. Glycoconj. J. 15, 115–130 (1998).

22. Hymowitz, S. et al. Triggering cell death: the crystal structure of Apo2L/TRAIL in acomplex with death receptor 5. Mol. Cell 4, 563–571 (1999).

23. Chan, F.K. et al. A domain in TNF receptors that mediates ligand-independent receptorassembly and signaling. Science 288, 2351–2354 (2000).

24. Clancy, L. et al. Preligand assembly domain-mediated ligand-independent associationbetween TRAIL receptor 4 (TR4) and TR2 regulates TRAIL-induced apoptosis. Proc.Natl. Acad. Sci. USA 102, 18099–18104 (2005).

25. Feig, C., Tchikov, V., Schutze, S. & Peter, M.E. Palmitoylation of CD95 facilitatesformation of SDS-stable receptor aggregates that initiate apoptosis signaling. EMBO J.26, 221–231 (2007).

26. Xu, Z. & Weiss, A. Negative regulation of CD45 by differential homodimerization of thealternatively spliced isoforms. Nat. Immunol. 3, 764–771 (2002).

27. Haines, N. & Irvine, K.D. Glycosylation regulates Notch signalling. Nat. Rev. Mol. CellBiol. 4, 786–797 (2003).

28. Priatel, J.J. et al. The ST3Gal-I sialyltransferase controls CD8+ T lymphocyte home-ostasis by modulating O-glycan biosynthesis. Immunity 12, 273–283 (2000).

29. Brockhausen, I. Pathways of O-glycan biosynthesis in cancer cells. Biochim. Biophys.Acta 1473, 67–95 (1999).

30. Dube, D. & Bertozzi, C. Glycans in cancer and inflammation—potential for therapeu-tics and diagnostics. Nat. Rev. Drug Discov. 4, 477–488 (2005).

31. Fuster, M., Brown, J., Wang, L. & Esko, J. A disaccharide precursor of sialylLewis X inhibits metastatic potential of tumor cells. Cancer Res. 63, 2775–2781(2003).

32. Fuster, M. & Esko, J. The sweet and sour of cancer: glycans as novel therapeutictargets. Nat. Rev. Cancer 5, 526–542 (2005).

33. Ohyama, C., Tsuboi, S. & Fukuda, M. Dual roles of sialyl Lewis X oligosaccharides intumor metastasis and rejection by natural killer cells. EMBO J. 18, 1516–1525(1999).

34. Takada, A. et al. Contribution of carbohydrate antigens sialyl Lewis A and sialyl Lewis Xto adhesion of human cancer cells to vascular endothelium. Cancer Res. 53, 354–361(1993).

35. Hoffman, E., Awad, T. & Palma, J. Expression profiling—best practices for datageneration and interpretation in clinical trials. Nat. Rev. Genet. 5, 229–237(2004).

36. Yauch, R. et al. Epithelial versus mesenchymal phenotype determines in vitrosensitivity and predicts clinical activity of erlotinib in lung cancer patients. Clin.Cancer Res. 11, 8686–8698 (2005).

37. Sharp, D.A., Lawrence, D.A. & Ashkenazi, A. Selective knockdown of the longvariant of cellular FLICE inhibitory protein augments death receptor-mediatedcaspase-8 activation and apoptosis. J. Biol. Chem. 280, 19401–19409 (2005).

ART ICL ES

NATURE MEDICINE VOLUME 13 [ NUMBER 9 [ SEPTEMBER 2007 1077

©20

07 N

atur

e P

ublis

hing

Gro

up

http

://w

ww

.nat

ure.

com

/nat

urem

edic

ine