Glutamine synthetase activity fuels nucleotide ...ruppin/ncb_glutamine.pdf · glutamine synthetase...

ART ICLES

Glutamine synthetase activity fuels nucleotidebiosynthesis and supports growth ofglutamine-restricted glioblastomaSaverio Tardito1, Anaïs Oudin2, Shafiq U. Ahmed3, Fred Fack2, Olivier Keunen2, Liang Zheng1, Hrvoje Miletic4,Per Øystein Sakariassen4, AdamWeinstock5, Allon Wagner5, Susan L. Lindsay6, Andreas K. Hock1,Susan C. Barnett6, Eytan Ruppin5,7, Svein Harald Mørkve8, Morten Lund-Johansen8,9, Anthony J. Chalmers3,Rolf Bjerkvig2,4, Simone P. Niclou2,4 and Eyal Gottlieb1,10

L-Glutamine (Gln) functions physiologically to balance the carbon and nitrogen requirements of tissues. It has been proposed thatin cancer cells undergoing aerobic glycolysis, accelerated anabolism is sustained by Gln-derived carbons, which replenish thetricarboxylic acid (TCA) cycle (anaplerosis). However, it is shown here that in glioblastoma (GBM) cells, almost half of theGln-derived glutamate (Glu) is secreted and does not enter the TCA cycle, and that inhibiting glutaminolysis does not affect cellproliferation. Moreover, Gln-starved cells are not rescued by TCA cycle replenishment. Instead, the conversion of Glu to Gln byglutamine synthetase (GS; cataplerosis) confers Gln prototrophy, and fuels de novo purine biosynthesis. In both orthotopic GBMmodels and in patients, 13C–glucose tracing showed that GS produces Gln from TCA-cycle-derived carbons. Finally, the Glnrequired for the growth of GBM tumours is contributed only marginally by the circulation, and is mainly either autonomouslysynthesized by GS-positive glioma cells, or supplied by astrocytes.

Gln and Glu constitute a metabolic hub in cellular physiology. Anincreased demand for Gln by transformed cells has been recognizedby biochemists for almost a century, and it has been linked toits role as an abundant circulating respiratory fuel1. Notably, Glncarbons can support anabolism through entering the TCA cyclethrough glutaminolysis. Only specific tumour types exhibit Glndependency2–9, and its genetic and metabolic basis remains debatable.In certain cancer models, the inhibition of glutaminase (GLS), whichdeaminates Gln to Glu, reduces proliferation and tumorigenicity10.Conversely, GLS2 can be induced by the tumour suppressor p53(ref. 11), and in human hepatocellular carcinoma, β-catenin increasesthe expression of GS, which catalyses the reversed GLS reaction12.Originally, tuning of the Gln–Glu cycle was observed in the centralnervous system13 where Glu is the most abundant neurotransmitter14.Unlike astrocytes, glioma cells can release neuro-excitotoxic amountsof Glu, potentially promoting tumorigenesis15. Gln addiction has

been proposed as a mark of GBM, the most aggressive glioma4.Here, we dissected the differential metabolic roles of Gln-derivedcarbon and nitrogen atoms in sustaining anabolism and growth insix human established GBM cell lines, in primary GBM stem-likecells, and in normal astrocytes. Additionally, Gln-related metabolismwas investigated in both primary orthotopic murine xenografts andin GBM patients, leading to the identification of a GBM–astrocytemetabolic crosstalk.

RESULTSGln starvation reduces GBM cell proliferation unsystematicallyTo explore their growth response to different nutrient supplies, GBMcells were incubated either in DMEM containing supra-physiologicalconcentrations of glucose and lacking some of the non-essentialamino acids, or in a newly formulated SMEM, containing nutrientconcentrations comparable to human serum (Supplementary Table 1).

1Cancer Metabolism Research Unit, Cancer Research UK, Beatson Institute, Switchback Road, Glasgow G61 1BD, UK. 2NorLux Neuro-Oncology Laboratory,Department of Oncology, Luxembourg Institute of Health, L-1526 Luxembourg, Luxembourg. 3Institute of Cancer Sciences, University of Glasgow, Glasgow G12 8QQ,UK. 4Kristian Gerhard Jebsen Brain Tumour Research Center, Department of Biomedicine, University of Bergen, Bergen N-5009, Norway. 5The Blavatnik School ofComputer Science, Tel Aviv University, Tel Aviv 69978, Israel. 6Institute of Infection, Immunity and inflammation, College of Medical, Veterinary and Life Sciences,University of Glasgow, Glasgow G12 8TA, UK. 7The Sackler School of Medicine, Tel Aviv University, Tel Aviv 69978, Israel. 8Department of Neurosurgery, HaukelandUniversity Hospital, Bergen N-5021, Norway. 9Department of Clinical Medicine, University of Bergen, Bergen N-5020, Norway.10Correspondence should be addressed to E.G. (e-mail: [email protected])

Received 30 October 2014; accepted 19 October 2015; published online 23 November 2015; DOI: 10.1038/ncb3272

1556 NATURE CELL BIOLOGY VOLUME 17 | NUMBER 12 | DECEMBER 2015

© 2015 Macmillan Publishers Limited. All rights reserved

mailto:[email protected]

ART ICLES

20 25 30 350

20

40

60

80

100

Doubling time (h)

Gro

wth

inhi

biti

on (%

) Pearson r –0.45

U251SF188

GUVWLN229

U87

LN18

+Gln–Gln

0 200 400 600

FL2-A

Cou

nts

00 200 400 600

FL2-A

Cou

nts

00 200 400 600

FL2-A

Cou

nts

00 200 400 600

FL2-A

Cou

nts

00 200 400 600

FL2-A

Cou

nts

00 200 400 600

FL2-A

Cou

nts

0

GUVW

0 0.5 1.0 1.5 2.00

20

40

60

80

100

a

Gln (mM)

Cel

l num

ber

(per

cent

age

of m

ax. r

esp

.)

DMEMSMEM

0 0.5 1.0 1.5 2.00

20

40

60

80

100

Gln (mM)

DMEMSMEM

0 0.5 1.0 1.5 2.00

20

40

60

80

100

Gln (mM)

DMEMSMEM

0 0.5 1.0 1.5 2.00

20

40

60

80

100

Gln (mM)

DMEMSMEM

0 0.5 1.0 1.5 2.00

20

40

60

80

100

Gln (mM)

DMEMSMEM

0 0.5 1.0 1.5 2.00

20

40

60

80

100

Gln (mM)

DMEMSMEM

LN18 LN229 SF188 U251 U87

GUVW LN18 LN229 SF188 U251 U87



0 48 96 1440

200

400

600

800

1,000

Time (h)0 48 96 144

Time (h)0 48 96 144

Time (h)0 48 96 144

Time (h)0 48 96 144

Time (h)0 48 96 144

Time (h)

Time (h) Time (h) Time (h) Time (h) Time (h) Time (h)

Cel

l num

ber

(×10

3 ) –Gln +Gln

–Gln +Gln

–Gln +Gln

–Gln +Gln

–Gln +Gln

–Gln +Gln

0

200

400

600

800

1,000

0

200

400

600

0

200

400

600

0

100

200

300

0

300

600

900

1,200

Cel

l dea

th in

dex

(%)

0 24 48 72 0 24 48 72 0 24 48 72 0 24 48 72 0 24 48 72 0 24 48 720

20

40

60

80

100

0

20

40

60

80

100

0

20

40

60

80

100

0

20

40

60

80

100

0

20

40

60

80

100

0

20

40

60

80

100

–Gluc–Gln+Gln

–Gluc–Gln+Gln

–Gluc–Gln+Gln

–Gluc–Gln+Gln

–Gluc–Gln+Gln

–Gluc–Gln+Gln

0

20

40

60

80

100

GUVWLN

18

LN22

9

SF188

U251

U87

Gro

wth

inhi

biti

on (%

)

++ +++ ++ + + ++Sensitivity

+Gln–Gln

+Gln–Gln

+Gln–Gln

+Gln–Gln

+Gln–Gln

b

c

d

e

f

Figure 1 Gln starvation reduces GBM cell proliferation. (a) Dose–responsecurves for cell lines incubated for 3 days in DMEM or SMEM with theindicated concentrations of Gln. (b) Cells were incubated for the indicatedtimes in SMEM with/without Gln. (c) Cells were incubated for 72h inSMEM with/without Gln or glucose as indicated. Each dot represents thenumber of dead cells (with plasma membrane integrity loss) normalizedover a confluence index. The resulting cell death index was assessedevery hour. (d) Cell cycle distribution of cell lines incubated for 3 days

with/without Gln. The mean of 3 independent experiments is shown inSupplementary Fig. 1b. (e) Growth inhibition caused by Gln starvation. Mean± s.e.m. n=3 independent experiments. (f) Scatter plot of the doublingtime obtained for the cell lines in Gln-fed conditions, in relation to growthinhibition caused by Gln starvation. Mean ± s.e.m. n= 3 independentexperiments. In a–d, the data derive from one experiment performed twice(a–c), or three times (d). Raw data from independent repeats are provided inSupplementary Table 5.

Both media were supplemented with various concentrations of Gln(Fig. 1a). In serum-like medium, all cell lines grew comparably toor faster than cells cultured in DMEM. In both media, the minimalGln concentration required for maximal growth was below 0.65mM,hereafter used as the control concentration. In the absence of Gln,cells grew faster in SMEM, demonstrating that medium formulationaffects the response to Gln deprivation. Gln starvation hinderedproliferation to different extents (Fig. 1b and Supplementary Fig. 1a)without inducing cell death, contrary to previous reports3,4, and totheir response to glucose withdrawal (Fig. 1c). DNA flow-cytometryanalysis showed that Gln starvation did not cause cell cycle arrest atany particular phase (Fig. 1d and Supplementary Fig. 1b). Overall,Glnwithdrawal resulted in cell line-specific growth inhibition, ranging

from 20% for U251 and SF188, to 80% for LN18 cells (Fig. 1e),independently of the initial proliferation rate (Fig. 1f).

Gln-based anaplerosis is not essential for the proliferation ofGBM cell linesTo investigate cellular metabolic alterations on Gln starvation, theexchange rate of metabolites between cells and medium was analysedby high-performance liquid chromatography–mass spectrometry(HPLC–MS). Gln was the second most consumed nutrient by all celllines (Supplementary Fig. 2 and Supplementary Table 2). However,no clear relationship emerged between Gln consumption and Glndependency (Fig. 2a and Supplementary Fig. 3a). In contrast, allcell lines showed a net secretion of Glu despite its presence in the

NATURE CELL BIOLOGY VOLUME 17 | NUMBER 12 | DECEMBER 2015 1557


ART ICLES

Glna

+– +– +– +– +– +––5

–4

–3

–2

–1

0

Sec

retio

n an

d c

onsu

mp

tion

(nm

ol μ

g–1 p

rot.

per

day

)

Gln

+– +– +– +– +– +–Gln

+– +– +– +– +– +–Gln +– +– +– +– +– +–Gln +– +– +– +– +– +–Gln +– +– +– +– +– +–GlnU87

GUVW

U251SF188LN229LN18

13C5

13C0

13C1

13C0

13C5

13C013C113C213C313C4

13C5

13C013C113C213C313C4

13C5

13C013C113C213C313C4

13C5

13C013C113C2

13C013C113C2

13C313C4

Glu

0

1

2

3

Sec

retio

n an

d c

onsu

mp

tion

(nm

ol μ

g–1 p

rot.

per

day

)

Gln

0

20

40

60

Rel

ativ

e am

ount

(pea

k ar

ea/μ

g p

rot.

)

Rel

ativ

e am

ount

(pea

k ar

ea/μ

g p

rot.

)

Rel

ativ

e am

ount

(pea

k ar

ea/μ

g p

rot.

)

Glu

0

20

40

60

AcCoA

0

1

2

3

Oleate

0

10

20

30

Rel

ativ

e am

ount

(pea

k ar

ea/μ

g p

rot.

)

Glu

+ – + –+ + + +

0

0.5

1.0

1.5

2.0

Sec

retio

n(n

mol

μg–1

pro

t. p

er d

ay)

Sec

retio

n(n

mol

μg–1

pro

t. p

er d

ay)

0

0.5

1.0

1.5

2.0

Gln13C6–Gluc

LN18

SF188

Gln15N1–Ala

LN18

SF188

Glu

13N015N1

0 0.5 1.0 1.5 2.0 2.50

20

40

60

80

100

Glu secretion(nmol μg–1 prot. per day)

Gro

wth

inhi

biti

on (%

)

Pearson r 0.93

U251SF188

GUVW LN229U87

LN18

Glu

Sec

retio

n an

d c

onsu

mp

tion

(nm

ol μ

g–1 p

rot.

per

day

)

–

Glu 4

mM

dm-α

KG 4 m

M

SSZ 0.2

mM

Cystin

e 0.5

mM

–

Glu 4

mM

dm-α

KG 4 m

M

SSZ 0.2

mM

–

Glu 4

mM

dm-α

KG 4 m

M

SSZ 0.2

mM

Cystin

e 0.5

mM –

Glu 4

mM

dm-α

KG 4 m

M

SSZ 0.2

mM

Cystin

e 0.5

mM –

Glu 4

mM

dm-α

KG 4 m

M

SSZ 0.2

mM

Cystin

e 0.5

mM –

Glu 4

mM

dm-α

KG 4 m

M

SSZ 0.2

mM

Cystin

e 0.5

mM

–10123456789

10 +Gln–Gln

+Gln–Gln

+Gln–Gln

+Gln–Gln

+Gln–Gln

+Gln–Gln

Glu

Rel

ativ

e am

ount

(per

cent

age

of c

ontr

ol)

0

25

50

75

100

125

150Aspartate

0

50

100

150Citrate

0255075

100125150175

GSH

0

25

50

75

100

Cel

l num

ber

(per

cent

age

of c

ontr

ol)

0

25

50

75

100

P = 0.0005P = 0.025

P = 0.0003

LN18

Time (h)

13C

5–G

lu s

ecre

ted

(per

cent

age

of t

otal

)

0 3 6 9Time (h)

0 3 6 9Time (h)

0 3 6 9Time (h)

0 3 6 90

10

20

30

40

50 0 μM2.5 μM5 μM10 μM

0 μM2.5 μM5 μM10 μM



SF188

0

10

20

30

40

50LN18

Gln

in m

ediu

m(p

erce

ntag

e of

initi

al v

alue

)

60

70

80

90

100

110SF188

60

70

80

90

100

110

Cel

l num

ber

(per

cent

age

of c

ontr

ol)

– + – + – + – + – + – +0

25

50

75

100

BPTES(2.5 μM)

GUVWLN

18

LN22

9

SF188

U251

U87

SystemASC

SystemsA/N

Gln GluGLS CySH

CySH

CyS

CyS

CyS

CyS

O2

R-SH R-S

R-S

Glu

GS

SLC7A11 4F2HC

Gln

Gln

Cell

Medium

SystemASC

GSH

Prot.

TCAcycle

αKGGDHGPT

Glucose

Pyr

Xc–

f

pq r s

k l m n o

g h i j

b c d e

+ – + –+ + + +

Figure 2 The effects of Glu secretion and GLS inhibition on GBM cell growthand metabolism. (a,b) Cells were incubated for 24h with/without 13C5–Gln.Secretion (positive bars) and consumption (negative bars) rates of Gln andGlu isotopologues are shown. Mean ± s.e.m. n=3 independent experiments.(c–f) Cells were incubated as in a,b and the levels of intracellular Gln,Glu, acetyl-CoA and oleate isotopologues are shown. Mean ± s.e.m. n=3independent experiments. (g,h) LN18 and SF188 cells were incubated for24h with/without Gln in media where glucose (g) or alanine (h) was fullyreplaced by 13C6–glucose or 15N1–alanine, respectively. The isotopologuedistributions of Glu released in the medium are shown. Mean ± s.e.m.n=3 independent experiments. (i) Scatter plot of Glu secretion observedin the absence of Gln, in relation to the growth inhibition caused by Glnstarvation. Mean ± s.e.m. n=3 independent experiments. (j) A schematicrepresentation of the X−c activity in the context of Glu metabolism. (k) LN18cells were incubated for 24h with/without Gln in media supplemented or notwith Glu, α-ketoglutarate dimethylester (dm-αKG), sulphasalazine (SSZ) or

cystine, at the indicated concentrations, and the secretion and consumptionrates of Glu are shown. (l–o) LN18 cells were incubated as in k and theintracellular levels of Glu (l), aspartate (m), citrate (n) and the reducedform of glutathione (o) are shown as a percentage of the untreated control.(p) LN18 cells were incubated for 72h as described for k. Cell numberis shown as a percentage of the untreated control. Mean ± s.e.m. n=4independent experiments. P values refer to a two-tailed t-test for unpairedsamples. (q,r) Cells were pre-incubated in medium with 0, 2.5, 5, 10, 15,30 µM BPTES for 3 h. At t=0, the medium was replaced with one containing13C5–Gln. The abundance of 13C5–Glu (q) or 13C5–Gln (r) in the medium wasmonitored over time. In all conditions, cells were exposed to 0.3% DMSO.(s) Cells were incubated in medium with/without 2.5 µM BPTES for 72h, andcounted. DMSO was 0.3% in all conditions. Mean± s.e.m. n=3 independentexperiments. In k–o,q,r, the data derive from one experiment performed once(k–o) or twice (q,r). Raw data from independent repeats are provided inSupplementary Table 5.

medium (Fig. 2b). Tracing 13C5-labelled Gln revealed that 38± 8% to60 ± 19% (GUVW and U87 cells, respectively) of the Gln consumedwas deamidated and secreted as 13C5–Glu (Fig. 2a,b). Unexpectedly,

even Gln-starved cells discharged Glu (Fig. 2b, empty bars). Here,intracellularGlnwas almost exhausted (Fig. 2c) andGlu concentrationfell by more than 50% (Fig. 2d), yet neither acetyl-CoA (Fig. 2e)



ART ICLES

GS expression (a.u.)

Gro

wth

inhi

biti

on (%

)

Pearson r –0.58

U251 SF188

GUVW

LN229U87

LN18

c-Myc

GLS

GS

Gln

LN18 LN229 SF188 U251 U87GUVW

β-Tub

Gln

a

c

f h

g

d e

b

15N015N1

012345

75

100R

elat

ive

amou

nt(p

erce

ntag

e of

con

trol

)

Gln + – + –

GlnGlu + NH4

++ –– – + +

+ – GlnGlu + NH4

+

GlnGlu + NH4

+

LN18 SF188

SF188

Cel

l num

ber

(per

cent

age

of c

ontr

ol)

Cel

l num

ber

(per

cent

age

of c

ontr

ol)

0

25

50

75

100

GluMSO

–Gln+GlnP = 0.0016

U251

–Gln+GlnP = 0.0017

SF188 NTC shRNA

Cel

l num

ber

(per

cent

age

of c

ontr

ol)

Cel

l num

ber

(per

cent

age

of c

ontr

ol)

0

25

50

75

100

0

25

50

75

100

SF188 GS shRNA1 SF188 GS shRNA2

U251 NTC shRNA

0

25

50

75

100

0

25

50

75

100

0

25

50

75

100

U251 GS shRNA1 U251 GS shRNA2

Glutamine synthetase(GS)

ATP ADP

(Pi)

Glutaminase(GLS)

Glu

NH3

O O O O

O O

GlnNH4

+

NH4+

H2N

0

20

40

60

80

100

0 10 20 30

–– + – +

– + + GluMSO

–– + – +

– + +0

25

50

75

100

0

25

50

75

100

+ NH3+

–– O–

+ – + – + – + – + – + –

SF188

U251

NTCshRNA

GSshRNA1

GSshRNA2

NTCshRNA

GSshRNA1

GSshRNA2

Gln

Glu + NH4+

+

– – + +

– + –

SF188NTC

shRNA

Gln

GS

Actin

NTCshRNA

GSshRNA1

GSshRNA2

GSshRNA1

GSshRNA2

U251

+ – + – + – + – + – + –

+ – + –– – + +

GlnGlu + NH4

+

+ – + –– – + +

GlnGlu + NH4

++ – + –– – + +

+ – + –– – + +

GlnGlu + NH4

++ – + –– – + +

Figure 3 GS sustains cell growth during Gln starvation. (a) GS- and GLS-catalysed reactions. (b) Cells were incubated for 24h with/without Gln andprotein expression was assessed. (c) Scatter plot of GS protein expressionobserved in the Gln-fed condition (arbitrary units, a.u.) in relation to thegrowth inhibition caused by Gln starvation. Mean ± s.e.m. n=3 independentexperiments. (d) LN18 and SF188 cells were incubated for 24h with/withoutGln in the presence of 0.8mM 15NH4

+. The intracellular levels of Glnisotopologues are shown as a percentage of 15N0–Gln in LN18 cells. Dataderive from one experiment performed twice. Raw data from independentrepeats are provided in Supplementary Table 5. (e) SF188 and U251 cellswere incubated for 72h with/without Gln in medium supplemented with 4mMGlu and 1mM MSO as indicated. Cell numbers are shown as a percentage

of the untreated control. Mean ± s.e.m. n=3 independent experiments.(f) SF188 and U251 cells stably expressing a non-targeting control shRNA(NTC shRNA) or two sequences targeting GS (GS shRNA1 and GS shRNA2)were incubated for 24h with/without Gln. (g) Cells were incubated for 72hwith/without Gln in medium supplemented with 4mM Glu and 0.8mM NH4

+,as indicated. Cell number is shown as a percentage of the respective Gln-fed control. Mean ± s.e.m. n=3 independent experiments. (h) Cells wereincubated for 12–17 days with/without Gln in medium supplemented with4mM Glu and 0.8mMNH4

+ as indicated. Colonies obtained in representativewells are shown. n=4 independent experiments, quantified as shown inSupplementary Fig. 4b. Unprocessed original scans of western blots areshown in Supplementary Fig. 8.

nor oleate (Fig. 2f) levels were reduced, re-affirming that Gln doesnot sustain fatty acids biosynthesis under normoxic condition16–20.Consistently, in all cell lines, less than 15% of the citrate was

derived from reductive carboxylation (13C5–citrate; SupplementaryFig. 3b), and labelled acetyl-CoA and oleate were barelydetectable (Fig. 2e,f).



ART ICLES

To identify the carbon source for Glu synthesis under Glnstarvation, Gln-deprived LN18 and SF188 cells were incubated with13C6–glucose. In both cases, the contribution of glucose carbons toGluwas markedly increased (Fig. 2g). Accordingly, alanine consumptionwas increased (Supplementary Fig. 2, inset), providing the nitrogenrequired for Glu production (Fig. 2h). Moreover, a strong directcorrelation between Glu efflux and growth inhibition was observed(Fig. 2i), suggesting that on Gln withdrawal, Glu efflux limits theintracellular Glu available for reactions essential for growth.

To test whether Glu efflux indeed limited both its availabilityand the proliferation of Gln-starved cells, LN18 cells were incubatedwith sulphasalazine, an inhibitor of X−c , a Glu/cystine antiporter(Fig. 2j) that is active in glioma cells15, or with 4mM Glu, largelyexceeding the Ki for X−c (ref. 21). In both conditions Glu releasewas largely inhibited (Fig. 2k). Consistently, the activation of X−cby increasing the extracellular concentrations of cystine, boostedGlu efflux (Fig. 2k), firmly associating X−c with the escape of Glufrom GBM cells. Inhibiting X−c prevented the drop in intracellularGlu, aspartate and citrate, caused by Gln withdrawal (Fig. 2l–n).The levels of glutathione, an alternative metabolic fate for Glu,were also substantially decreased on Gln starvation, and significantlyreplenished by Glu and sulphasalazine (Fig. 2o). However, the drop inglutathione on Gln starvation was not accompanied by its increasedoxidation (Supplementary Fig. 3c), indicating that Gln delimits Gluavailability for glutathione biosynthesis without causing oxidativestress. Furthermore, adding the membrane-permeable dimethylesterof α-ketoglutarate (dm-αKG) to Gln-starved cells doubled the Gluefflux (Fig. 2k) and replenished Glu, aspartate, citrate and glutathioneintracellular pools (Fig. 2l–o). Overall, in the absence of Gln, X−cinhibition or dm-αKG supplementation restored the cellular Glu,aspartate, citrate and glutathione levels to that of non-starved cells.However, maintaining the levels of these metabolites only moderatelyrescued the proliferation of Gln-starved cells (Fig. 2p). Together withthe observations that, under Gln starvation, intracellular oleate wasunaffected (Fig. 2f), and glucose-dependent Glu production increased(Fig. 2g–h), these results imply that the contribution of Gln to growthis largely independent of anaplerosis.

To evaluate this directly, BPTES, a GLS inhibitor, was employed.The kinetics of Gln-derived Glu secretion from LN18 and SF188 cellsshowed that BPTES inhibited GLS activity (Fig. 2q). Accordingly,the rate of Gln consumption was reduced (Fig. 2r). Nevertheless,at the minimum effective concentration, BPTES did not affect thegrowth of the GBM lines (Fig. 2s and Supplementary Fig. 3d–i)confirming that here, GLS and Gln-based anaplerosis are dispensablefor maximal growth.

Glutamine synthetase sustains purine availability and cellgrowth under Gln starvationNext, a genome-scale constraint-based metabolic modelling approachwas employed22, searching for reactions that become essential whenGln is removed from the SMEMmedium. According to the model, GSwas the only enzyme essential for sustaining biomass production afterGln starvation (Supplementary Note).

Together, GS and GLS control Gln homeostasis by catalysingopposite reactions (Figs 2j and 3a). The messenger RNA levels ofGS and GLS in all cell lines revealed no pattern of Gln dependency

(Supplementary Fig. 4a). Although it has been proposed that c-Mycdetermines Gln addiction by increasing GLS expression3,23,24, MYC-induced lung tumours were shown to increase GS expression2. Inline with this, SF188 cells, harbouring MYC amplification25, andexpressing high levels of c-Myc (Fig. 3b and Supplementary Fig. 4a),showed the highest levels of GS mRNA and protein (Fig. 3b andSupplementary Fig. 4a). In most cell lines, GS protein levels rose onGln deprivation (Fig. 3b). Furthermore, GS protein levels tend toincrease in cells with decreased sensitivity to Gln withdrawal (Fig. 3c).Nevertheless, GS did not match the residual low amount of Gln foundin Gln-starved cells (Fig. 2c). To investigate this apparent discrepancy,the metabolic flux through GS was assessed by the incorporation of15N-labelled ammonia (15NH4

+) into Gln in LN18 and SF188 cells,which exhibit low and high levels of GS, respectively. Significant levelsof 15N-labelled Gln were indeed detected in SF188 but not in LN18cells (Fig. 3d). Next, SF188 and U251 cells, exhibiting high levelsof GS and low sensitivity to Gln withdrawal, were incubated withL-methionine sulphoximine (MSO), a selective irreversible inhibitorof GS. MSO sensitized cells to Gln starvation and, further, abolishedthe protective effect of Glu supplementation (Fig. 3e). To complementthis approach, GS expression was stably silenced in these cells by twoshort hairpin RNA (shRNA) sequences (Fig. 3f). On Gln starvation,cell proliferation (Fig. 3g) as well as colony formation (Fig. 3h andSupplementary Fig. 4b) was lowered byGS silencing. Supplementationwith the GS substrates Glu and ammonia rescued Gln-deprivedcontrol cells more effectively than GS-silenced cells (Fig. 3g,h).

To corroborate the causal link between Gln biosynthesis and Glndependency, GS was overexpressed in LN18 cells, which exhibitlow GS levels and high sensitivity to Gln deprivation. To this end,LN18-derived clones stably expressing infrared fluorescent proteinalone26,27 (iRFP) or iRFP and GS were established. To eliminateintrinsic clonal variability, GS expression and its effect on growthunder Gln starvation were evaluated in multiple clones (6 iRFP and9 iRFP–GS). After five days of starvation, growth of iRFP controlcells reached, on average, 16 ± 5% of control Gln-fed cells, andiRFP–GS clones reached, on average, 54 ± 12% (Fig. 4a). Under Glnsupplementation, the iRFP–GS5 clone proliferated slower than iRFPcontrols. This was not rectified by GS inhibition with MSO (Fig. 5b),consistent with a reported non-metabolic, anti-proliferative role ofGS (ref. 28). Nevertheless, iRFP–GS5, but not control iRFP4 cells,proliferated and formed colonies in Gln-freemedium, and this growthadvantage was blocked by MSO (Fig. 4b,c). These results imply thatunder Gln starvation the amidation of Glu catalysed by GS sustainscell growth. In linewith this, when supplementedwith 15N1–ammonia,GS-expressing cells exhibited 15N incorporation into ∼50% of thetotal Gln pool, even when Gln fed (Fig. 4d). When incubated with15N1–ammonia without Gln, residual intracellular Gln was higherin iRFP–GS5 cells compared with control iRFP4 cells, and producedentirely by GS as judged by 15N incorporation (Fig. 4d).

To explore the essentiality and metabolic fate of de novo syn-thesized Gln under Gln starvation, we employed flux imbalanceanalysis29. In silico, nucleotide biosynthesis delimited cell growth,with a higher weighted cost for purine biosynthesis (SupplementaryNote). Indeed, regardless of GS status, Gln removal only marginallyaffected the levels of the pyrimidine nucleotide uridine monophos-phate (UMP; Fig. 4e). Moreover, the contribution of GS-derived



ART ICLES

0

20

40

60

80

100

Cel

l gro

wth

(μg

pro

t. p

er w

ell,

per

cent

age

of c

ontr

ol)

Gln

iRFP iRFP–GS

β-tubGS

0

100

200

Cel

l num

ber

(×10

3 )

300

400

500a

c

h

m n

i j k l

d e f g

b iRFP4 + GlniRFP4 + Gln + MSOiRFP4 – GlniRFP4 – Gln + MSO

iRFP–GS5 + GlniRFP–GS5 + Gln + MSOiRFP–GS5 – GlniRFP–GS5 – Gln + MSO

Time (h)

Gln

Rel

ativ

e am

ount

(per

cent

age

of c

ontr

ol)

UMP

0

25

50

75

100

125

150

Rel

ativ

e am

ount

(per

cent

age

of c

ontr

ol)

Rel

ativ

e am

ount

(per

cent

age

of c

ontr

ol)

Rel

ativ

e am

ount

(per

cent

age

of c

ontr

ol)

Rel

ativ

e am

ount

(per

cent

age

of c

ontr

ol)

Rel

ativ

e am

ount

(per

cent

age

of c

ontr

ol)

AICAR

ND

IMP

ND

AMP

0

25

50

75

100

125

150ATP

0

25

50

75

100

125

Rel

ativ

e am

ount

(per

cent

age

of c

ontr

ol) GTP IMP

0

100

200

300

400

Rel

ativ

e am

ount

(pea

k ar

ea/μ

g p

rot.

)U87

GUVW

U251

SF188

LN229

LN18

–3 –2 –1 0 1 20

20406080

100

Δpeak area/μg prot.

Gro

wth

inhi

biti

on (%

) Pearson r –0.83

U251SF188

GUVW

LN229U87

LN18

Cel

l num

ber

(per

cent

age

of c

ontr

ol)

– A G C T U

AGCTU Glu

Glu + M

SO

A + G

lu

A + G

lu + M

SOM

SO – A G C T U

AGCTU Glu

Glu + M

SO

A + G

lu

A + G

lu + M

SOM

SO0

25

50

75

100

125

Cel

l num

ber

(per

cent

age

of c

ontr

ol)

0

25

50

75

100

125+Gln–Gln +Gln

–Gln

+ – + – + – + – + – + – + – + – + – + – + – + – + – + – + –

+ – + – + – + – + – + –

1 2 3

–

MSO

4 5 6 1 2 3 4 5 6 7 8 9

+ – + – + – + – + – + – + – + – + –0 24 48 72 96 120 144

iRFP4

+Gln –Gln +Gln –GlniRFP–GS5

iRFP4 iRFP–GS5

0Gln + – + –

iRFP4

iRFP4 iRFP–GS5

iRFP–GS5

Gln

iRFP4 iRFP–GS5

Gln + – + –iRFP4 iRFP–GS5


Gln + – + –

25

50

75

100

125

150

0

50

100

250

150

200

15N015N1

15N015N1

15N2

15N015N115N2

15N015N1

15N3

15N2

15N015N115N2

15N015N115N2

15N015N115N2

0

250

500

750

1,000

1,250

+ – + –iRFP4 iRFP–GS5


Gln + – + – Gln +–+–+–+–+–+–0

25

50

75

100

125

1 2 3 4 5 6 1 2 3 4 5 6 7 8 9

Figure 4 GS activity regulates cell growth and purine availability under Glnstarvation. (a) Top: LN18 clones stably expressing iRFP or iRFP–GS wereincubated with/without Gln for 5 days and protein expression was assessed.Unprocessed original scans of western blots are shown in SupplementaryFig. 8. Bottom: For each clone, growth was determined from the proteinamount and presented as a percentage of the respective control. Dashedlines show the mean percentage values obtained without Gln. (b) iRFP4

and iRFP–GS5 cells were incubated for the indicated times in mediumwith/without Gln and MSO (1mM), and counted. (c) iRFP4 and iRFP–GS5

cells were incubated with/without Gln and MSO (1mM) for 21 days,and colonies in representative wells are shown. Data derive from oneexperiment performed twice. (d–j) iRFP4 and iRFP–GS5 cells were incubatedwith/without Gln for 24 h in medium supplemented with 0.8mM 15NH4

+.The intracellular isotopologues of Gln, UMP, AICAR, IMP, AMP, ATP andGTP are shown as a percentage of the values obtained for the 15N0

metabolites in iRFP4 cells in the presence of Gln. (k) Cell lines wereincubated with/without Gln for 24 h and the relative amount of intracellularIMP is shown. Mean ± s.e.m. n=3 independent experiments. (l) Scatterplot of the changes in intracellular IMP levels in relation to the growthinhibition caused by Gln starvation. Mean ± s.e.m. n= 3 independentexperiments. (m,n) iRFP4 (m) and iRFP–GS5 (n) cells were incubatedfor 72h with/without Gln in medium containing adenosine (A), guanosine(G), cytidine (C), thymidine (T) and uridine (U), each at 0.2mM, or incombination (AGCTU) at 0.2mM each, Glu (4mM) and MSO (1mM), asindicated. Cell numbers are shown as percentage of untreated control.Dashed lines show percentage values obtained in the absence of Glnwithout any further supplementation. Mean ± s.e.m. n= 3 independentexperiments. In a,b,d–j, the data derive from one experiment performed once(a,b) or twice (d–j). Raw data from independent repeats are provided inSupplementary Table 5.



ART ICLES

E2 Diff. E2 GSC

R10 Diff. R10 GSC

R24 Diff. R24 GSC

0 48 96 1440

50

100

150

200

E2

GS

Sox2

CD133

OLIG2

GFAP

Actin

R10 R24

Diff. GSCDiff. GSC Diff. GSC

Time (h) Time (h)

Time (h) Time (h)

Time (h) Time (h)

Cel

l num

ber

(×10

3 )C

ell n

umb

er (×

103 )

Cel

l num

ber

(×10

3 )

Cel

l num

ber

(×10

3 )C

ell n

umb

er (×

103 )

Cel

l num

ber

(×10

3 )

+Gln + MSO+Gln

–Gln–Gln + MSO

+Gln + MSO+Gln

–Gln–Gln + MSO

+Gln + MSO+Gln

–Gln–Gln + MSO

+Gln + MSO+Gln

–Gln–Gln + MSO

+Gln + MSO+Gln

–Gln–Gln + MSO

+Gln + MSO+Gln

–Gln–Gln + MSO

0 48 96 1440

50

100

150

200

0 48 96 1440

50

100

150

200

250

0 48 96 1440

50

100

150

200

250

0 48 96 1440

50

100

150

0 48 96 1440

50

100

150

Gln

Glu Glu Glu

Gln Gln

Gln Gln Gln

Glu Glu Glu

+ – + ––4–3–2–1012

Sec

retio

n/co

nsum

ptio

n(n

mol

µg–

1 p

rot.

per

day

)S

ecre

tion/

cons

ump

tion

(nm

ol µ

g–1

pro

t. p

er d

ay)

Sec

retio

n/co

nsum

ptio

n(n

mol

µg–

1 p

rot.

per

day

)S

ecre

tion/

cons

ump

tion

(nm

ol µ

g–1

pro

t. p

er d

ay)

Sec

retio

n/co

nsum

ptio

n(n

mol

µg–

1 p

rot.

per

day

)S

ecre

tion/

cons

ump

tion

(nm

ol µ

g–1

pro

t. p

er d

ay)

Gln

E2GSCDiff.

+ – + –

E2GSCDiff.

+ – + –Gln

Gln Gln

GSCDiff.+ – + –Gln

GSCDiff.

–4–3–2–1012

R10

+ – + –GSCDiff.

R10

–4–3–2–1012

R24

+ – + –GSCDiff.

R24

+ – + –

E2GSCDiff.

Gln Gln+ – + –GSCDiff.

R10

+ – + –GSCDiff.

R24

+ – + –

E2GSCDiff.


R10

+ – + –GSCDiff.

R24

+ – + –

E2GSCDiff.


R10

+ – + –GSCDiff.

R24

–2

–1

0

1

2

Gln–2

–1

0

1

2

–2

–1

0

1

2

0

50

100

150

200

Rel

ativ

e am

ount

(per

cent

age

of c

ontr

ol)

Rel

ativ

e am

ount

(per

cent

age

of c

ontr

ol)

Rel

ativ

e am

ount

(per

cent

age

of c

ontr

ol)

Rel

ativ

e am

ount

(per

cent

age

of c

ontr

ol)

Rel

ativ

e am

ount

(per

cent

age

of c

ontr

ol)

Rel

ativ

e am

ount

(per

cent

age

of c

ontr

ol)

Rel

ativ

e am

ount

(per

cent

age

of c

ontr

ol)

Rel

ativ

e am

ount

(per

cent

age

of c

ontr

ol)

Rel

ativ

e am

ount

(per

cent

age

of c

ontr

ol)

Rel

ativ

e am

ount

(per

cent

age

of c

ontr

ol)

Rel

ativ

e am

ount

(per

cent

age

of c

ontr

ol)

Rel

ativ

e am

ount

(per

cent

age

of c

ontr

ol)

Gln0

50

100

150

0

100

200

300

0

50

100

150

Gln0

50

100

150

0

50

100

150

0

25

50

75

100

Gln0

25

50

75

100

0

25

50

75

100

+ – + –

E2GSCDiff.


R10

+ – + –GSCDiff.

R24

0

25

50

75

100

Gln0

25

50

75

100

AMPAMPAMP

0

25

50

75

100

15N015N1

15N015N1

15N2

15N015N1

15N015N1

15N2

15N015N115N2

Citrate Citrate Citrate

a c

b

d

e

f

g

h

Figure 5 Gln metabolism in differentiated and GBM stem-like (GSC)primary human GBM cells. (a) Protein expression was assessed in E2,R10 and R24 cells maintained in DMEM/F-12 and supplemented asdescribed in the Methods. Arrow points to a Sox2 specific band. Arepresentative experiment repeated twice is shown. Unprocessed originalscans of western blots are shown in Supplementary Fig. 8. (b) Cellswere incubated in SMEM supplemented as described in the Methods,

with/without 0.65mM Gln and 1mM MSO as indicated. Mean ± s.e.m. n=3independent experiments. (c,d) The exchange rates of Gln (c) and Glu (d)isotopologues in cells incubated for 24h in SMEM with/without 0.65mMGln supplemented with of 0.8mM 15NH4

+. Mean ± s.e.m. n=3 independentexperiments. (e–h) The intracellular content of Gln (e), Glu (f), citrate (g) andAMP (h) isotopologues in cells incubated as in c,d. Mean ± s.e.m. n=3independent experiments.

15N1–Gln to 15N1–UMP showed that the low GS activity in iRFP4

cells could maintain UMP production. Similarly, the biosynthe-sis of UDP-N -acetylglucosamine, an intermediate of hexosamine

biosynthesis, which requires Gln-derived nitrogen, was also sustainedin iRFP4 cells during Gln starvation (Supplementary Fig. 4c). In con-trast, 5-aminoimidazole-4-carboxamide ribotide (AICAR), a purine



ART ICLES

precursor for inosinemonophosphate (IMP), dropped to undetectablelevels in Gln-starved iRFP4 cells (Fig. 4f). The ammonia-derivedAICAR (15N2-isotopologue) in Gln-starved iRFP–GS5 cells demon-strates that GS contributes the two Gln nitrogen atoms required for itsbiosynthesis (Fig. 4f). Thus, 15N2–IMP accumulated in Gln-deprivediRFP–GS5 cells, but not in iRFP4 cells (Fig. 4g). The IMP found iniRFP–GS5 starved cells corroborates the reported inhibitory effect ofGln deprivation on IMP dehydrogenase30, and suggests that underthese conditions over-activity ofGS exceeds the rate of IMP conversionto AMP and GMP. Indeed, AMP levels were not significantly affectedby either Gln presence or GS overexpression (Fig. 4h). Moreover, ATPandGTP levels, indices of the cell’s bioenergetic state, were comparablebetween iRFP–GS5 and iRFP4 starved cells (Fig. 4i,j). Conversely, thefractions of 15N-labelled AMP, ATP and GTP found in iRFP–GS5cells exceeded those of iRFP4 cells, demonstrating that, under Glnstarvation, GS sustains de novo biosynthesis of purine nucleotides.

Notably, the reduction in IMP caused by Gln starvation in thesix GBM cell lines correlated with Gln dependency (Fig. 4k,l).Accordingly, adenosine, but not guanosine or pyrimidine nucleosides,partially restored Gln-independent growth of iRFP4 cells (Fig. 4m).Furthermore, the combined addition of adenosine and Glucompensated for the lack of exogenous Gln in iRFP4 cells (Fig. 4m),and Glu alone completely restored proliferation of iRFP–GS5starved cells (Fig. 4n). In both lines, MSO prevented the Glu rescue,confirming that under Gln starvation, Glu availability determinesGln production rather than anaplerosis. The effect of adenosineon Gln-starved cells was MSO-independent (Fig. 4m,n), because itsupported proliferation downstream of GS.

Primary human GBM stem-like cells are self-sufficient for GlnrequirementsThe clinical relevance of studying established glioma cell lines has beenquestioned owing to their inability to form tumours that recapitulatehuman pathology. Therefore, we used three primary patient-derivedGBM cell lines (E2, R10, R24), generating paired populations ofdifferentiated cells (DIFF) and glioma stem-like cells31 (GSC). Stemcell markers such as CD133, Olig2 and Sox2 were predominantlyexpressed in GSC but not DIFF; however, the astrocytic markerglial fibrillary acidic protein (GFAP) was not consistently associatedwith the DIFF population (Fig. 5a). The expression of GS wasmarkedly higher in all GSC comparedwithDIFF (Fig. 5a) andwhereasDIFF proliferation was attenuated in the absence of Gln, GSC grewindependently of Gln supplementation (Fig. 5b). Once more, thegrowth of Gln-starvedDIFF andGSCwas abolished byMSO (Fig. 5b).

Next, the exchange rates (Fig. 5c,d) and intracellular compositionof metabolites (Fig. 5e–h) were analysed in these primary Gln-starvedor control cells, in the presence of 15N1–ammonia. The net Glnconsumption was consistently higher in DIFF compared with GSC,whereby R24-GSC demonstrated no net Gln consumption (Fig. 5c).As the intracellular 15N1–Gln fraction shows, GSC have higher GSactivity compared with DIFF and sustain higher residual Gln levelson starvation (Fig. 5e). The differences in Glu exchange rates werealso striking: GSC exhibited a net uptake of Glu, whereas the reverseoccurred in DIFF (Fig. 5d). Also, on Gln withdrawal, intracellularlevels of citrate (Fig. 5g) decreased in DIFF, but remained unalteredin GSC, showing that Gln-derived anaplerosis was redundant to the

stem-like population. Finally, ammonia-derived 15N incorporationinto purine nucleotides (15N2–AMP) under Gln starvation was greaterin GSC compared with the paired DIFF.

Human GBM tumours rely on in situ de novo Gln synthesisAs shown above, GS activity, largely determining Gln dependency,varies between established and primary human GBM cells. Similarly,tissue microarray (TMA) analysis showed that GS expression variesbetween human GBM patients (n = 209), resembling a Gaussiandistribution ranging from tumours with low GS levels, comparable toneurons (25% of patients), to high-expression tumours comparableto astrocytes (15%; Fig. 6a,b). However, GS expression did notpredict patient median survival (Fig. 6c). Of twenty biopsies, fromwhich core TMA were sampled, five showed substantial intratumoralGS immunostaining heterogeneity (three examples are reported inSupplementary Fig. 5). Most GBMs showed either GS uniformity,or a mosaic infiltration of GS-positive cells, suggesting autonomousintratumoral Gln biosynthetic capacity. To assess this hypothesis inhuman tumours, seven GBMpatients were injected with 13C6–glucosebefore surgery, and metabolites were extracted from the resectedtumours and their oedematous margins. The metabolic analysisreliably discriminated between tumour and adjacent tissues by usingthe choline to creatine ratio, a parameter for classifying brain tumoursby magnetic resonance spectroscopy32 (Fig. 6d). No significantdifference in Gln content was observed between tumour and adjacenttissues (Fig. 6e). At the time of resection, 13C6–glucose enrichment insera ranged between 16 and 50% (Fig. 6f, and Supplementary Fig. 6a).Glucose-derived 13C–Gln was detected in 6/7 tumours and in 7/7adjacent oedematous tissues with an enrichment ranging between 1and 12% (Fig. 6g). In 3/5 patients the fraction of glucose-derived Glnin the tumour was higher than in the serum sample, and so not inequilibrium with the circulating Gln, suggesting that the tumour Glnpool is synthesized in situ and/or provided by adjacent normal brain.

To complement the analysis in patients, mice were orthotopicallytransplantedwith aGS-positive humanGBM(P3, Fig. 6h) and injectedwith 13C-labelled glucose or Gln ∼20min before tissue extraction.An enrichment of 44 ± 3% of 13C6–glucose was found in the bloodat the time of tissue sampling. The intracellular hexose phosphatepool derived from 13C6–glucose was ∼10% and 5% in tumour andcontralateral brain tissue, respectively (Fig. 6i). Concomitantly,∼10%and 15% of the total Gln was labelled (13C2) from 13C6–glucosein tumours and contralateral brain, respectively (Fig. 6i), consistentwith GS activity in those tissues. After 13C5–Gln injection, theenrichment in circulating 13C5–Gln at the time of tissue sampling was17 ± 1%. Isotopologue distribution analysis showed an enrichmentin 13C5–Gln of <5% in both tumour and contralateral brain tissues(Fig. 6j), whereas in the liver, 13C5–Gln presented 12% of the total(Supplementary Fig. 6e). Products of glutaminolysis, such as Glu andα-ketoglutarate, were labelled below 1% in both tumour and braintissues but ∼5% in liver (Fig. 6j and Supplementary Fig. 6e). Theseresults suggest slow kinetics both for Gln uptake from the blood, andfor glutaminolysis in GBM and brain tissues, compared with liver.Similar results were obtained on constant carotid artery 13C5–Glninfusion: within two hours, circulating 13C5–Gln levels plateaued at∼20% enrichment, with an overall ∼70% increase in steady-statelevels of Gln (Supplementary Fig. 6f). Here too, 13C5–Gln and



ART ICLES

Histoscore

0 30 60 90 120

150

180

210

240

270

300

0

5

10

15

20Low

500 μm

50 μm

Medium High

N

A

Normal brain25 Low Medium High

Months

Per

cent

age

of s

urvi

val

0 24 48 72 96 1200

20

40

60

80

100 GS lowGS mediumGS high

Creatine

Rel

ativ

e am

ount

(×10

9 p

eak

area

)

Rel

ativ

e am

ount

(×10

8 p

eak

area

)

Rel

ativ

e am

ount

(×10

8 p

eak

area

)

Tumour Adjacentoedematous

brain


brain


brain


brain

0

2

4

6 P = 0.0002 P = 0.002 P = 0.89P = 0.037

Choline

0

1

2

3

4

5

Choline/creatine

Pea

k ar

ea r

atio

0

0.05

0.10

0.15

0.20

Gln

0

2

4

6

8

10

GBM patients GBM patients

13C

6–gl

ucos

e(p

erce

ntag

e of

tot

al)

13C

–Gln

from

glu

cose

(per

cent

age

of t

otal

)

1 2 3 4 5 6 70

102030405060 Adjacent oedematous brain

Tumour

SerumAdjacent oedematous brainTumour

Serum

NA

1 2 3 4 5 6 70

3

6

9

12

15

NA ND ND

0

25

50

75

100Cataplerosis

Cataplerosis

P3

tum

our

Isot

opol

ogue

dis

trib

utio

n (%

)

Con

tral

ater

al b

rain

Isot

opol

ogue

dis

trib

utio

n (%

)

Con

tral

ater

al b

rain

Isot

opol

ogue

dis

trib

utio

n (%

)

P3

tum

our

Isot

opol

ogue

dis

trib

utio

n (%

)13C6–glucose injection

0

25

50

75

100

0

25

50

75

100Anaplerosis

13C5–glutamine injection

0

25

50

75

100

Freq

uenc

y (%

)

P = 0.89N

estin

GS

GS

P3 tumour

Contralateralbrain

Contralateralbrain

Tumour

Tumour

1 mm

1 mm

50 μm 50 μm

V

A

AFN

V

Gln GlnGlu GluαKG

αKG

Citrat

e

Citrat

e

Hex-P

Hex-P

GlnGluαKG

Citrat

e

Hex-P Gln Glu

αKG

Citrat

e

Hex-P

Gln biosynthesis

Gln biosynthesis

13C013C113C213C313C413C513C6

13C013C113C213C313C413C513C6

13C013C113C213C313C413C513C6

13C013C113C213C313C413C513C6

Glutaminolysis

Anaplerosis

Glutaminolysis

a b c

d e

f g

h i j

Figure 6 Gln metabolism in GBM patients and primary orthotopic xenografts.(a) GBM tissue microarray. GS immunostaining of representative tissue coresat low and high magnification (top and bottom, respectively). A, astrocyte;N, neuron. (b) Frequency distribution of GBM patients (n=209) dividedaccording to their histoscore for GS, and categorized as low, medium andhigh. Normal astrocytes were used as a reference for defining maximalimmunoreactivity. (c) Kaplan Meier curves for GBM patients divided into low,medium and high GS expression. P value refers to a log-rank (Mantel–Cox)test. (d,e) Creatine, choline, choline to creatine ratio (d) and Gln (e) levelsin tumour tissue and adjacent oedematous brain of GBM patients injectedwith 13C6–glucose before surgical intervention. n= 7 patients; P valuesrefer to a two-tailed t-test for paired samples. (f,g) 13C6–glucose (f), and13C–Gln (g) enrichment in serum at time of tumour resection, in tumourtissue, and in adjacent oedematous tissue. Gln isotopologues incorporating

one or more 13C atoms, over the total amount of Gln detected (percentageof total) are shown. NA, not available; ND, not detectable. Values werecorrected for the natural abundance of 13C. (h) Coronal section of ahuman P3 GBM xenograft grown in the brain of immunocompromised mice,and stained for human nestin and GS. Lower panels are a magnificationof the respective framed regions. A, astrocytes; AF, astrocytic end-feet;N, neuron; V, blood vessel. (i,j) Isotopologue distribution of metabolites(hexose phosphates, citrate, αKG, Glu, Gln) obtained in mice orthotopicallyxenografted with human P3 GBM, and injected in the tail vain with abolus of 13C6–glucose (i) or 13C5–Gln (j). Tissues were sampled 22min afterinjection. The values are mean ± s.e.m. n= 3 mice for all conditions,except for contralateral brain of mice injected with glucose, where 2 micewere used. For i,j, raw data from independent repeats are provided inSupplementary Table 5.



ART ICLES

glutaminolysis products were scarce in tumour and contralateral brain(Supplementary Fig. 6h,i) indicating that physiologically, circulatingGln does not significantly supply Gln to the brain or the tumourwithin it.

Next, Erwinase, an enzyme that temporarily depletes circulatingasparagine (Asn) and significantly reduces Gln (refs 9,33; Fig. 7a),was injected daily 5 times per week into mice bearing GS-negativeorthotopic GBM xenografts (T101, Fig. 7b). Erwinase reducedintratumoral and intracerebral Asn but not Gln (Fig. 7c). Thediffuse morphology of GS-negative tumours impaired the volumetricquantification of tumours by magnetic resonance imaging (MRI)(Fig. 7d).Histological reconstruction to assess tumour burden (Fig. 7e)revealed no significant differences between control and Erwinase-treated groups (Fig. 7f). Together, these results indicate that Glnis not provided to GBM by the blood, and so the proximity ofGS-positive astrocytes and GS-negative glioma cells (Fig. 7b,g asrepresentatives) suggests that astrocytes may be a source of Gln.Indeed, when mice bearing GS-negative GBM xenografts (T407,Fig. 7g) were infused with 15N1–ammonia into the carotid arteryfor 4 h, the fraction of 15-N-Gln was ∼5% in both tumour andcontralateral brain tissues (Fig. 7h) indicating that they are a secluded,autonomous compartment for Gln biosynthesis and utilization, whereGS-expressing cells supply Gln to GS-negative ones.

Gln-starved GBM cells feed on astrocyte-derived GlnTo assess this potential interaction, rat primary cortical astrocyteswere cultured and their Gln requirement andmetabolism investigated.Similar to GBM cells, the minimal Gln concentration required formaximal astrocyte growth was ∼0.65mM (Fig. 8a). Nonetheless,astrocyte proliferation was barely affected by Gln deprivation(Fig. 8b). As observed in the human TMA (Fig. 6a), GS proteinlevels in astrocytes and in the highest expressing GBM cells werecomparable (Fig. 8c). However, only astrocytes demonstrated no netGln consumption but rather, rapid Glu uptake (Fig. 8d,e), in linewith the expression of excitatory amino acid transporters in thiscell type34. Under Gln starvation, Glu consumption was unaffectedand paralleled by an equimolar net Gln efflux (Fig. 8d,e). Theabsence of Gln in the medium reduced intracellular Gln, but not Glu(Fig. 8f,g). Moreover, 13C6–glucose tracing showed that only 30–40%of both intracellular Glu and Gln (Fig. 8f,g) was glucose-derived.Astrocytes maintained ∼30% of the control level of intracellular Glnunder Gln starvation (Fig. 8f,h), fitting with high GS expression.Gln maintenance depended on GS activity, as seen from both15N1–ammonia tracing andGS inhibition byMSO (Fig. 8h).Moreover,GS inhibition markedly elevated the intracellular amounts of itssubstrate, Glu (Fig. 8i), without changing the steady-state levels ofthe Gln product, AMP (Fig. 8j). However, combined Gln withdrawaland GS inhibition significantly reduced the labelled fraction of AMPderived from de novo synthesis (15N2 and 15N3; Fig. 8j), and hinderedproliferation (Fig. 8k).

Finally, in co-culture, astrocytes enabled the proliferation ofGS-negative LN18 iRFP4 cells without Gln supplementation (Fig. 8l).Moreover, Transwell co-culturing of these cells showed that thefactor conveying growth was diffusible (Fig. 8m,n). The addition ofErwinase, which depletes both Asn and Gln, prevented the rescue ofGln-deprived cells by astrocytes (Fig. 8m,n). As Asn was present in the

media during all Gln-starvation experiments, and because astrocytesconsumeAsn but produce and secrete Gln (Fig. 8d and SupplementaryFig. 6k), these results designate astrocyte-derived Gln as the growth-supporting factor for Gln-starved GBM cells.

DISCUSSIONGln plays multiple metabolic and non-metabolic roles. Consequently,the dependency of cancer cells on Gln is difficult to discern.Nevertheless, it was demonstrated here that the Gln requirementin GBM goes beyond anaplerosis (Fig. 2q–s). We identified twoalternative metabolic determinants for Gln sensitivity: Glu releasethrough the X−c antiporter; and GS-dependent conversion of Gluto Gln. GS and X−c seemingly compete for cytoplasmic Glu, whichduring Gln starvation becomes limiting (Fig. 2d,l). Nevertheless, therescuing effect achieved by maintaining intracellular Glu through X−cinhibition depends on GS activity (Fig. 3e,h,g and Fig. 4m,n). Theseresults demonstrate that on Gln starvation, Glu conversion to Glnconstitutes a critically limiting reaction required for growth. Thismetabolic trait applies to established GBM lines, and to naive primarycells. Indeed, when primary GBM cells were maintained in a stem-like state, GS expression was markedly increased (Fig. 5a) and Glu wastaken up rather than released (Fig. 5d). Both responses enable growthof glioma stem-like cells independent of extracellular Gln (Fig. 5b).The fate of newly synthesized Gln in both established and primaryGBM cells was followed by 15N1–ammonia tracing, identifying theAMP biosynthesis pathway as a significant player in Gln dependency(Fig. 4d–j and Fig. 5h).

GS is found in most human GBM (Fig. 6a,b), and its expressionis associated with poor prognosis35, although this is not supportedby our TMA study. However, we show that GS expression variesgreatly between tumours, ranging from negative, comparable to GSexpression in neurons, to high, as in normal astrocytes. This variationaccords with a Gln-rich tumour microenvironment, which alleviatesthe need to synthesize Gln. Nevertheless, in agreement with previousreports36,37, it is shown here that most human GBM, as well asGS-proficient orthotopic GBM xenografts, withdraw carbons fromthe TCA cycle (cataplerosis) to synthesize Gln by means of GS.When compared with the liver, GBM are inclined towards net Glnsynthesis, rather than glutaminolysis (Fig. 6i and SupplementaryFig. 6d). Accordingly, the circulation provides minimal amountsof Gln to normal brain38 and GBM (Fig. 6l and SupplementaryFig. 6f,h,i). Moreover, a marked decrease in circulating Gln levels didnot affect tumour growth (Fig. 7a–h), and GBM expressing low GShad levels of ammonia-derived Gln comparable to contralateral brain(Fig. 7j). The stability of Gln levels may explain the heterogeneity inGS expression between patients (Fig. 6b) and within some tumours(Supplementary Fig. 5): GS-positive astrocytes and/or GBM cells(potentially GSC) excrete Gln that supports the growth of GS-negative GBM cells. Indeed, primary astrocytes in culture retainthe metabolic traits of Glu uptake, GS-dependent Gln synthesisand Gln secretion, and support the proliferation of GS-negative,Gln-auxotrophic GBM cells (Fig. 8l–n). Thus, this brain-to-tumourmetabolic communication portrays a scenario in which Gln, providedby astrocytes, feeds GS-negative cancer cells (Fig. 3g). Indeed,GS-positive cellular protrusions of astrocytes surround GS-negativeGBM cells in orthotopic xenografts and human tumours (Fig. 7b,g



ART ICLES

Time (h)

Blo

od c

once

ntra

tion

(μM

)

0 240

1020304050

300400500600

Asn

Gln

6

∗ P = 0.0096

Erwinase

Nes

tin

Tumour

T101 tumour

Tumour

Contralateralbrain

Contralateral brain

GS

1 mm

1 mm

50 μm

A

50 μm

A

N

ADC Tumour mask IHC

Asn

Rel

ativ

e am

ount

(per

cent

age

of c

ontr

ol)

Con

trol

Rel

ativ

e am

ount

(per

cent

age

of c

ontr

ol)

0255075

100125150

Erwinase

Erw

inas

e

Gln

0255075

100125150

T101

tum

our

Contra

later

al bra

in

1 mm 1 mm

1 mm 1 mm

– + – + Erwinase

T101

tum

our

Contra

later

al bra

in– + – +

T101 tumour, control

T101 tumour, Erwinase

1 mm

1 mm

Posterior

Posterior

Anterior

Anterior

Tum

our

volu

me

(mm

3 )

Ctr Erw0

20

40

60

80P = 0.80

T101 tumour

Nes

tinG

S

Tumour

Tumour Contralateralbrain

1 mm

1 mm

T407 tumour Contralateral brain

50 μm 50 μm

A

V

A0

2

4

6

8

10

T407

tum

our

Contra

later

al bra

in

P = 0.11

15N

1–G

ln fr

om 1

5 NH

4+

(per

cent

age

of t

otal

)

a b

c d

e

f g h

Figure 7 Glutamine supply for GBM tumours with low GS expression.(a) Gln and Asn levels were measured by HPLC–MS in peripheral bloodsamples obtained at the indicated time points from immunocompromisedmice intraperitoneally injected with Erwinase (5Ug−1 of body weight).Mean ± s.e.m. n=5 mice. P values refer to a two-tailed t-test for pairedsamples. (b) Coronal section of a human T101 GBM xenograft grown inthe brain of immunocompromised mice, and stained for human nestin andGS. Right panels are a magnification of the respective outlined regions.A, astrocytes; N, neuron. (c–f) Immunocompromised mice were orthotopicallyimplanted with T101 GBM tumours and treated with Erwinase for 6 weeksas described in the Methods. (c) Asn and Gln were assessed in the tumourand contralateral brain 6 h after the last Erwinase injection. Mean ± s.e.m.n= 3 mice. (d) MRI-based apparent diffusion coefficient (ADC) maps ofT101 GBM tumours. The tumour mask has been manually delineated to

highlight the tumour region. IHC staining of brain sections correspondingto the MRI scans is shown. T101 tumours were stained with an anti-human EGFR antibody. (e) Two representative series of coronal sectionsof T101 brain xenografts were stained for human EGFR. (f) Volumes ofT101 orthotopic tumours obtained thorough quantitative imaging of EGFR-stained serial sections of brains. Mean ± s.e.m. n=7 mice. P value refersto a two-tailed t-test for unpaired samples. (g) Coronal section of a humanT407 GBM xenograft grown in the brain of immunocompromised mice, andstained for human nestin and GS. Right panels are a magnification of therespective framed regions. V, blood vessel. (h) 15N1–Gln enrichment in T407GBM tumours and in contralateral brains, after a 4 h intracarotid infusionwith 15NH4

+. The dashed lines correspond to the natural abundance of15N1–Gln. Mean ± s.e.m. n=4 mice. P value refers to a two-tailed t-testfor paired samples.



ART ICLES

0 1 2 3 40

20

40

60

80

100

0 2 4 6 8 100

10

20

30

40

50

60

70

80

+Gln

–Gln

Days

Cel

l num

ber

(×10

3 )

Cel

l num

ber

(per

cent

age

of m

ax r

esp

)

Gln (mM)

Gln + – + – + – + –

GS

GFAP

β-Tub

LN18 SF188 Astro extr.1 Astro extr.2

Gln

0

25

50

75

100

Rel

ativ

e am

ount

(per

cent

age

of c

ontr

ol)

Rel

ativ

e am

ount

(per

cent

age

of c

ontr

ol)

Rel

ativ

e am

ount

(per

cent

age

of c

ontr

ol)

MSO – –

–Gln+ ++ –

+ +

+ +

+15NH4

+

MSO – –

–Gln+ ++ –

+ +

+ +

+15NH4

+

MSO – –

–Gln+ ++ –

+ +

+ +

+15NH4

+ MSO – ––Gln + –

+

Glu

0

100

200

300

400

500

600

0

25

50

75

100

Cel

l num

ber

(per

cent

age

of c

ontr

ol)

AMP

0

25

50

75

100

125

15015N015N1

15N015N1

15N2

15N015N115N215N3

Gln

–2

–1

0

1

2

Sec

retio

n an

d c

onsu

mp

tion

(nm

ol μ

g–1

pro

t. p

er d

ay)

Gln

–Tracer

Glu

–1.0

–0.5

0S

ecre

tion

and

con

sum

ptio

n(n

mol

μg–

1 p

rot.

per

day

)Gln

0

25

50

75

100

Rel

ativ

e am

ount

(per

cent

age

of c

ontr

ol)

Rel

ativ

e am

ount

(per

cent

age

of c

ontr

ol)

Glu

0

25

50

75

100

+ – + –

13 C 5–G

ln

13 C 6–G

luc

13C013C113C213C313C413C5

13C013C113C213C313C413C5

13 C 6–G

lucGln

–Tracer

+ – + –

13 C 5–G

ln

13 C 6–G

luc

13 C 6–G

lucGln

–Tracer

+ – + –

13 C 5–G

ln

13 C 6–G

luc

13 C 6–G

lucGln

–Tracer

+ – + –

13 C 5–G

ln

13 C 6–G

luc

13 C 6–G

luc

0 24 48 72 96 1200

5

10

15

20

25

30

35

Time (h)

Cel

l via

bili

ty(fl

uore

scen

ce, a

.u.)

iRFP4 + GlniRFP4 – GlniRFP4 – Gln + ErwiRFP4 + Astro + GlniRFP4 + Astro – GlniRFP4 + Astro – Gln + Erw

Days

0 1 2 3 4 5 5

iRFP4

+Gln

+Gln

–Gln

–Gln+Erw

–Gln+Erw

–Gln

–Astro

+Astro

Astro

Days

0 1 2 3 4

iRFP4

+Gln

–Gln–Astro

+Astro+Gln

–Gln

a b c

d e f g

h i j k

l m n

Figure 8 Astrocytes provide GBM cells with Gln. (a,b) Astrocytes wereincubated in SMEM for 6 days with 0, 0.1, 0.3, 0.65, 1, 2 and 4mM Gln(a) or for the indicated times with 0 and 0.65mM Gln (b). (c) Astrocytesderived from two independent extractions, and cell lines, were incubated for3 days with/without Gln and protein expression was assessed. Unprocessedscans of western blots are shown in Supplementary Fig. 8. (d–j) Astrocyteswere incubated in SMEM for 24h in the presence of 5.56mM glucose(13C6 or 13C0), 0.65mM Gln (13C5 or 13C0), 0.8mM 15NH4

+, and 1mMMSO as indicated. Secretion and consumption rates (positive and negativebars, respectively) are shown for Gln (d) and Glu (e). Intracellular levelsof Gln (f) and Glu (g) isotopologues are reported as a percentage ofthe control (total of isotopologues in Gln-fed conditions). The intracellularisotopologues of Gln (h), Glu (i) and AMP (j) are shown as a percentageof the control (total isotopologues in the presence of Gln and 15NH4).

(k) Astrocytes were incubated for 6 days with/without 0.65mM Gln, and1mM MSO and counted. (l) Astrocytes were grown to confluence in multi-well plates. iRFP4 cells were seeded in wells with/without astrocytes,and with/without Gln. The fluorescence of iRFP4 cells in representativewells is shown. The experiment was performed twice with comparableresults. (m) Astrocytes were grown to confluence in multi-well plates.iRFP4 cells seeded in Transwell inserts were co-cultured with/withoutastrocytes, with/without Gln, and with/without Erwinase (Erw) as indicated.Fluorescence of iRFP4 cells in representative inserts is shown. At day 5astrocytes were stained with sulphorodamine-B and the fluorescence ofrepresentative wells is shown. (n) Quantification of the iRFP4 fluorescenceas described for m. In a,b,d–k,n, the data derive from one experimentperformed twice. Raw data from independent repeats are provided inSupplementary Table 5.



ART ICLES

and Supplementary Fig. 5p). This ‘parasitic’ behaviour of cancer cellscould divert the physiological Gln–Glu cycle in the brain and, indeed,an increase in epileptic seizures has been reported for patients withGBM expressing low GS (ref. 39). �

METHODSMethods and any associated references are available in the onlineversion of the paper.

Note: Supplementary Information is available in the online version of the paper

ACKNOWLEDGEMENTSThis study has been supported by Cancer Research UK. S.T. is a recipient of anAIRC/Marie Curie International Fellowship for Cancer Research. The human andanimal metabolomic studies were supported by The Norwegian Cancer Society,The Norwegian Research Council, Helse Vest, Haukeland University Hospitaland the K.G-Jebsen Foundation. We acknowledge A. Golebiewska, V. Baus-Talko,N. Van Den Broek, G. MacKay, C. Nixon and E. MacKenzie for excellent technicalassistance and A. King for excellent editorial work.

AUTHOR CONTRIBUTIONSS.T. conceived the study, designed and performed most experiments, interpretedthe data, and wrote the manuscript, A.O. performed the experiments in orthotopicxenograft models, S.U.A. and A.J.C. provided the differentiated and stem-likeprimary glioblastoma cells, L.Z. supervised the analysis of LC-MS samples, O.K.performed theMRI analysis, F.F. processed the orthotopic and clinical GBMsamples,H.M. provided the tissue microarray, A.K.H. designed and provided the iRFP andiRFP–GS constructs, A.Weinstock, A.Wagner and E.R. generated and employed themetabolic modelling, S.C.B. and S.L.L. provided the primary astrocytes, M.L.-J.,S.H.M. and P.Ø.S. provided the surgical specimens from the patients, S.P.N. andR.B. conceived and supervised the experiments in orthotopic models and humanpatients, and E.G. conceived and supervised the study, interpreted the data, andrevised the manuscript.

COMPETING FINANCIAL INTERESTSThe authors declare no competing financial interests.

Published online at http://dx.doi.org/10.1038/ncb3272Reprints and permissions information is available online at www.nature.com/reprints

1. Moreadith, R. W. & Lehninger, A. L. The pathways of glutamate and glutamineoxidation by tumor cell mitochondria. Role of mitochondrial NAD(P)+-dependentmalic enzyme. J. Biol. Chem. 259, 6215–6221 (1984).

2. Yuneva, M. O. et al. The metabolic profile of tumors depends on both the responsiblegenetic lesion and tissue type. Cell Metab. 15, 157–170 (2012).

3. Wise, D. R. et al. Myc regulates a transcriptional program that stimulatesmitochondrial glutaminolysis and leads to glutamine addiction. Proc. Natl Acad. Sci.USA 105, 18782–18787 (2008).

4. DeBerardinis, R. J. et al. Beyond aerobic glycolysis: transformed cells can engagein glutamine metabolism that exceeds the requirement for protein and nucleotidesynthesis. Proc. Natl Acad. Sci. USA 104, 19345–19350 (2007).

5. Tardito, S. et al. L-Asparaginase and inhibitors of glutamine synthetase discloseglutamine addiction of β-catenin-mutated human hepatocellular carcinoma cells.Curr. Cancer Drug Targets 11, 929–943 (2011).

6. Son, J. et al. Glutamine supports pancreatic cancer growth through a KRAS-regulatedmetabolic pathway. Nature 496, 101–105 (2013).

7. Gameiro, P. A. et al. In vivo HIF-mediated reductive carboxylation is regulated bycitrate levels and sensitizes VHL-deficient cells to glutamine deprivation. Cell Metab.17, 372–385 (2013).

8. Willems, L. et al. Inhibiting glutamine uptake represents an attractive new strategyfor treating acute myeloid leukemia. Blood 122, 3521–3532 (2013).

9. Chiu, M. et al. Glutamine depletion by crisantaspase hinders the growth of humanhepatocellular carcinoma xenografts. Br. J. Cancer 111, 1159–1167 (2014).

10. Wang, J. B. et al. Targeting mitochondrial glutaminase activity inhibits oncogenictransformation. Cancer Cell 18, 207–219 (2010).

11. Suzuki, S. et al. Phosphate-activated glutaminase (GLS2), a p53-inducible regulatorof glutamine metabolism and reactive oxygen species. Proc. Natl Acad. Sci. USA107, 7461–7466 (2010).

12. Dal Bello, B. et al. Glutamine synthetase immunostaining correlates with pathologicfeatures of hepatocellular carcinoma and better survival after radiofrequency thermalablation. Clin. Cancer Res. 16, 2157–2166 (2010).

13. Krebs, H. A. Metabolism of amino-acids: the synthesis of glutamine from glutamicacid and ammonia, and the enzymic hydrolysis of glutamine in animal tissues.Biochem. J. 29, 1951–1969 (1935).

14. van den Berg, C. J. & Garfinkel, D. A stimulation study of brain compartments.Metabolism of glutamate and related substances in mouse brain. Biochem. J. 123,211–218 (1971).

15. Takano, T. et al. Glutamate release promotes growth of malignant gliomas. Nat. Med.7, 1010–1015 (2001).

16. Scott, D. A. et al. Comparative metabolic flux profiling of melanoma cell lines: beyondthe Warburg effect. J. Biol. Chem. 286, 42626–42634 (2011).

17. Metallo, C. M. et al. Reductive glutamine metabolism by IDH1 mediates lipogenesisunder hypoxia. Nature 481, 380–384 (2012).

18. Kamphorst, J. J. et al. Hypoxic and Ras-transformed cells support growth byscavenging unsaturated fatty acids from lysophospholipids. Proc. Natl Acad. Sci.USA 110, 8882–8887 (2013).

19. Fan, J., Kamphorst, J. J., Rabinowitz, J. D. & Shlomi, T. Fatty acid labeling fromglutamine in hypoxia can be explained by isotope exchange without net reductiveisocitrate dehydrogenase (IDH) flux. J. Biol. Chem. 288, 31363–31369 (2013).

20. Mullen, A. R. et al. Reductive carboxylation supports growth in tumour cells withdefective mitochondria. Nature 481, 385–388 (2012).

21. Lewerenz, J. et al. The cystine/glutamate antiporter system x(c)(-) in health anddisease: from molecular mechanisms to novel therapeutic opportunities. Antioxid.Redox Signal. 18, 522–555 (2013).

22. Duarte, N. C. et al. Global reconstruction of the human metabolic network based ongenomic and bibliomic data. Proc. Natl Acad. Sci. USA 104, 1777–1782 (2007).

23. Gao, P. et al. c-Myc suppression of miR-23a/b enhances mitochondrial glutaminaseexpression and glutamine metabolism. Nature 458, 762–765 (2009).

24. Yuneva, M., Zamboni, N., Oefner, P., Sachidanandam, R. & Lazebnik, Y. Deficiency inglutamine but not glucose induces MYC-dependent apoptosis in human cells. J. CellBiol. 178, 93–105 (2007).

25. Bax, D. A. et al. Molecular and phenotypic characterisation of paediatric glioma celllines as models for preclinical drug development. PLoS ONE 4, e5209 (2009).

26. Filonov, G. S. et al. Bright and stable near-infrared fluorescent protein for in vivoimaging. Nat. Biotechnol. 29, 757–761 (2011).

27. Hock, A. K. et al. iRFP is a sensitive marker for cell number and tumor growth inhigh-throughput systems. Cell Cycle 13, 220–226 (2013).

28. Yin, Y. et al. Glutamine synthetase functions as a negative growth regulator in glioma.J. Neurooncol. 114, 59–69 (2013).

29. Reznik, E., Mehta, P. & Segre, D. Flux imbalance analysis and the sensitivityof cellular growth to changes in metabolite pools. PLoS Comput. Biol. 9,e1003195 (2013).

30. Calise, S. J. et al. Glutamine deprivation initiates reversible assembly of mammalianrods and rings. Cell. Mol. Life Sci. 71, 2963–2973 (2014).

31. Carruthers, R. et al. Abrogation of radioresistance in glioblastoma stem-like cells byinhibition of ATM kinase. Mol. Oncol. 9, 192–203 (2015).

32. Meyerand, M. E., Pipas, J. M., Mamourian, A., Tosteson, T. D. & Dunn, J. F.Classification of biopsy-confirmed brain tumors using single-voxel MR spectroscopy.AJNR Am. J. Neuroradiol. 20, 117–123 (1999).

33. Parmentier, J. H. et al. Glutaminase activity determines cytotoxicity ofL-asparaginases on most leukemia cell lines. Leuk. Res. 39, 757–762 (2015).

34. Had-Aissouni, L. Toward a new role for plasma membrane sodium-dependentglutamate transporters of astrocytes: maintenance of antioxidant defenses beyondextracellular glutamate clearance. Amino Acids 42, 181–197 (2012).

35. Rosati, A. et al. Glutamine synthetase expression as a valuable marker of epilepsyand longer survival in newly diagnosed glioblastoma multiforme. Neuro-oncol. 15,618–625 (2013).

36. Marin-Valencia, I. et al. Analysis of tumor metabolism reveals mitochondrial glucoseoxidation in genetically diverse human glioblastomas in the mouse brain in vivo. CellMetab. 15, 827–837 (2012).

37. Maher, E. A. et al. Metabolism of [U-13 C]glucose in human brain tumors in vivo.NMR Biomed. 25, 1234–1244 (2012).

38. Bagga, P. et al. Characterization of cerebral glutamine uptake from blood in the mousebrain: implications for metabolic modeling of 13C NMR data. J. Cereb. Blood FlowMetab. 34, 1666–1672 (2014).

39. Rosati, A. et al. Epilepsy in glioblastoma multiforme: correlation with glutaminesynthetase levels. J. Neurooncol. 93, 319–324 (2009).



http://dx.doi.org/10.1038/ncb3272





DOI: 10.1038/ncb3272 METHODS

METHODSCell cultures. The human glioblastoma cell lines MOG-G-UVW, LN-18, LN-229,SF-188, U-251 MG and U-87 MG (hereafter referred to as GUVW, LN18,LN229, SF188, U251 and U87, respectively) were obtained from the followingsources: GUVW: frozen stock originally isolated by I. Freshney at the Universityof Glasgow, UK; LN18 and LN229: ATCC; SF188: Brain Tumour Tissue BankUniversity of California San Francisco, USA; U251 and U87: kindly provided byK. Ryan at the Cancer Research UK—Beatson Institute, UK and A. Chalmersat the University of Glasgow, UK, respectively. The cell lines were authenticatedusing Promega GenePrint 10 system (STR multiplex assay that amplifies 9tetranucleotide repeat loci and the Amelogenin gender determiningmarker). No cellline used in the manuscript has been found in the ICLAC database of commonlymisidentified cell lines. Cortical astrocytes were generated from the cortex ofP7 Sprague-Dawley rat pups (unclassified gender). Briefly, both cortices wereremoved from each pup cleared of meninges and enzymatically dissociated andpurified as previously described. Tissue from 2 pups was added to each 75 cm2

tissue culture poly-L-lysine-coated flask (13 µgml−1, Sigma). Astrocytes preparedin this way were over 95% pure as assessed by immunoreactivity to glial fibrillaryacidic protein40.

All of the cell lines tested negative for mycoplasma with MycoAlert when testedwith the Mycoplasma Detection Kit (Lonza). Cells were maintained in DMEM(Invitrogen 21969-035) supplemented with 2mM glutamine and 10% FBS. All ofthe experiments were performed on cells seeded and incubated for at least 24 h in aserum-like medium, SMEM, formulated as reported in Supplementary Table 1, andsupplementedwith 10%FBS, dialysed against a semi-permeable cellulosemembrane(3.5 kDa MW cutoff).

For co-culture experiments 30,000 astrocytes per well were plated in a 24-wellplate and allowed to grow to confluence, and then an LN18-derived clone stablyexpressing infrared fluorescent protein, iRFP4, was seeded at 25,000 cells per well or5,000 iRFP4 cells per Transwell (Costar, CLS3413) in complete medium for direct orTranswell co-culture experiments, respectively. The next day, iRFP4 cell cultureswereextensivelywashedwith PBS, Transwells were inserted inwells containing astrocytes,and the medium was replaced with the conditions indicated in Fig. 8l,m. Half ofthe medium volume was renewed daily for the whole duration of the experiments.Growth was assessed by measuring fluorescence with a Licor Odyssey scanner andquantified using Image Studio 2.0 software.

E2, R10 and R24 human glioblastoma cells were obtained and cultured aspreviously reported with minor modifications31. In brief, differentiated (DIFF) cellswere maintained in advanced DMEM/F-12 (Gibco) supplemented with 10% FBSand 2mM Gln, and glioma stem-like cells (GSC) were maintained in advancedDMEM/F-12 supplemented with B27 (1%, Invitrogen), N2 (0.5%, Invitrogen),4 µgml−1 heparin, 20 ngml−1 fibroblast growth factor 2 (bFGF, Sigma), 20 ngml−1epidermal growth factor (EGF, Sigma), 1mM pyruvate and 2mM Gln. For all ofthe experiments stem cells were seeded and incubated in SMEM supplementedas described above, and DIFF cells were seeded and incubated in advancedDMEM/F-12 supplemented with 10% dialysed FBS. Glutamine (0.65mM) wasadded as indicated in Fig. 5.

Cell number assessment. Cells were plated at a density of 20,000 cells per wellin 24-well plates. Cell number was assessed at the specified experimental time asfollows: each well was washed twice with PBS, trypsinized, cells were re-suspendedin a Casyton solution and counted with a Casy cell counter. The doubling time (DT)was obtained using the exponential growth equation with GraphPad prism 5.0. Thepercentage of growth inhibition was calculated according to the following equation:[1−(DT(+Gln))/(DT(−Gln))]×100.

Cell death assessment. Cells were plated at a density of 15,000 cells per well in12-well plates. After 2 days, cells were placed in medium with or without glucoseand Gln, and supplemented with CellTox Green Dye (1:1,000, Promega, G8731).Cells were incubated in an IncuCyte FLR (EssenBioScience) for 72 h, during whichtime bright-field and fluorescence pictures were taken every hour. For each timepoint a cell death index was calculated by dividing the number of dead cells on theconfluency index.

Cell cycle analysis.Culture media and cells were collected, centrifuged and washedwith PBS. Cells were then fixed with cold methanol and stained with a PBSsolution containing propidium iodide (100 µgml−1) and RNAase A (250 µgml−1).DNA nuclear content was measured by using a Beckton Dickinson FACScan flowcytometer and data were analysed with FlowJo software.

Tissue microarray (TMA) and immunohistochemistry (IHC). The collectionof human biopsy tissuewas approved by the regional ethical committee atHaukelandUniversity Hospital, Bergen, Norway (REK 013.09). All patients gave a writteninformed consent for tumour biopsy collection and signed a declaration permitting

the use of their biopsy specimens for research. Core tissues of glioblastomacancer tissue (n = 209) were identified by a pathologist, and collected fromrepresentative areas of tumours derived from patients’ surgical resection. Tissuecores were collected in triplicate for each patient. Follow-up details includingdate and cause of death were collected. For IHC analysis of the TMA, andmouse brain slices, the following procedure was followed. Epitope retrieval wasperformed at 98 ◦C for 15′ in citrate buffer, pH 6. The immunostaining wasperformed using an EnVision System-HRP DAB (Dako, K4006). The cores wereincubated overnight at 4 ◦C with primary antibody. Cores were counterstained withhaematoxylin, dehydrated and coverslipped. The staining was quantified using aweighted histoscore method with a final value range 0–300. The intensity of theimmunostaining observed for normal astrocytes was classified as 300 and used asreference. Low-, medium- and high-glutamine synthetase (GS)-expressing groupswere defined as reported in Fig. 6b. For TMA and IHC the following antibodieswere diluted in a solution of 5% BSA in TBS–Tween and employed at the specifieddilutions: purified mouse anti-human monoclonal GS antibody (BD TransductionLaboratories, Cat. 610518, clone 6 at a final concentration of 250 ngml−1); mousemonoclonal anti-human nestin antibody (Abcam ab6320, clone 3k1, 1:500); mousemonoclonal anti-human EGFR antibody (Dako 7239 clone E30, 1:200, 1 h at roomtemperature). For tumour volume quantification by IHC the mouse brain wasparaffin embedded and entirely sliced into sections of 8 µm. Two sections wereloaded on each slide, and one slide in every 10 was stained for EGFR. Digitalimages were acquired with a Leica SCN400 Slide Scanner, loaded into Matlab,calibrated and pixels positive for the staining were identified and used to quantifytumour volume. All IHC staining images shown are representative of three ormore sections.

Metabolic study in GBM patients. Before undergoing surgical tumour resection,seven GBM patients were infused for 15min with 20 g of 13C6–glucose. Duringthe operation, tissue fragments from the core and peripheral region of the tumourwere stereotactically sampled (∼200mg for each region). On average the specimenswere snap-frozen 47min after the end of the glucose injection (range 30–80min).Peripheral blood samples were also collected at the same time of tissue sampling.Sub-fragments of frozen tissues (15–25mg) were further dissected from peripheraland core regions of the tumour, and from oedematous adjacent tissue. The sampleswere extracted on dry ice and processed as described below for human orthotopicGBM xenografts.

Informed consent was obtained from all human subjects involved in the study;the heavy-isotope administration and the collection of tissue from surgical resectionswere approved by the regional ethical committee at Haukeland University Hospital,Bergen, Norway (REK 2010/130-2).

Primary orthotopic human GBM xenografts. Patient-derived GBM xenograftswere generated as previously reported41,42. Briefly, GBM patient-derived spheroidswere stereotactically implanted into the right frontal cortex of 8–12-week-old, maleand female, NOD SCID mice (Charles River) using a Hamilton syringe. After 4–6weeks a P3 GBM tumour was fully established in the brain, and U-13C6–glucose orU-13C5–glutamine (Cambridge Isotope Laboratories) was injected as a bolus in thetail vein at a dose of 1mg g−1 and 0.15mg g−1 of body weight respectively. Twenty-two minutes after injection, the animals were euthanized, organs were dissected,and tumour and contralateral brain fragments were snap-frozen. For intracarotidinfusion the mice were kept under anaesthesia with ketamine medetomidineisoflurane and underwent a surgical procedure to expose the right carotid artery.Ammonia incorporting 15N1 was infused at 4.7 µmol kg−1 min−1, 0.833 µl min−1 for4 h inNODSCIDmice (Charles River) xenografted orthotopicallywith aT407GBM.Glutamine (13C5) was infused at 13 µmol kg−1 min−1, 0.125 µl min−1 for 4 h in 8-week-old, female, SwissNu/Numice (Charles River) xenografted orthotopically withthe GS-positive T16 GBM. To assess the effects of Erwinase, 8-week-old, female,Swiss Nu/Nu mice (Charles River) were implanted with T101 tumour spheroids(six per mouse). Forty-two days after implantation mice were randomly assignedto Erwinase treatment or control group. Erwinase was injected intraperitoneally at5U g−1 of body weight, five times per week, where 1 international unit is definedas the amount of enzyme required to generate 1 µmol of ammonia per minuteat pH 7.3 and 37 ◦C. The treatment continued for 6 weeks. Mice of the controlgroup were intraperitoneally injected with saline solution. Mice were euthanized sixhours after the last injection and tumour and contralateral brain fragments weresnap-frozen for metabolic analysis. Whole brains were immediately fixed in 4%paraformaldehyde for ex vivo MRI. For metabolic analysis, 10–20mg of tissue wasextracted with a beads mill (Qiagen, Tissuelyser) in a solution containing 20%water,50%methanol, 30% acetonitrile (25 µl of extraction solution permilligram of tissue).The extracts were spun down at 16,000g for 10min and supernatants were storedat −80 ◦C for LC–MS analysis. Human glioblastoma biopsies were obtained fromthe Neurosurgery Department of the Centre Hospitalier in Luxembourg (CHL). Allpatients had providedwritten informed consent, with procedures thatwere approved

NATURE CELL BIOLOGY


METHODS DOI: 10.1038/ncb3272

for the project (project number: REC-LRNO-20110708) by the National ResearchEthics Committee for Luxembourg (CNER). The handling of the animals and thesurgical procedures were performed in accordance with the European Directive onanimal experimentation (2010/63/EU) and the national regulations of Luxembourgand the local ethical committee (theAnimalWelfare Structure (AWS) LIH) approvedthe protocol (protocol number: LRNO-2014-06).

MRI analysis.At the end of Erwinase treatment the brains of the 20-week-old SwissNu/Nu mice with T101 tumours (7 female mice per group) were collected and fixedin 4% paraformaldehyde. Ex vivo scans were acquired on a 7T micro-MRI system(Bruker PharmaScan) using a mouse brain volume coil. The T2-weighted MRIprotocol used a fast spin echo sequence with TE= 36ms, TR= 4,300ms, and 63 µmin plane resolution. The diffusion-weighted imaging sequence used EPI readout withTE = 28.5ms, TR = 7,000ms and b-values ranging from 500 to 6,000 smm−2 in 3directions. The tumours were independently delineated by two users (A.O. andO.K.)using Paravision 5.0 (Bruker).

Immunoblotting. Cells were washed twice with cold PBS and lysed in RIPAbuffer (20mM Tris-HCl, pH 7.5, 150mM NaCl, 1mM EDTA, 1mM EGTA, 1%Triton, 2,5mMsodiumpyrophosphate, 1mMglycerophosphate) supplementedwith1mMNa3VO4, 1mMNaF and protein inhibitor cocktail. Protein concentration wasdetermined with the bicinchoninic acid assay (Thermo scientific) using BSA as astandard. Equal amounts of protein were mixed with Laemmli buffer 4× (250mMTris-HCl, pH 6.8, 8% SDS, 40% glycerol, 0.4 M dithiothreitol), warmed at 95 ◦Cfor 5min, and loaded on 10% gels for SDS–PAGE. After electrophoretic separation,proteins were blotted onto 0.22mm nitrocellulose membrane (Millipore), blockedwith 5% non-fat milk in TBS–Tween, and incubated at 4 ◦C overnight with thefollowing antibodies: glutamine synthetase (BD Transduction Labs, 610517, clone6, 1:1,000), β-tubulin (Sigma Aldrich, T5201, clone AA2, 1:5,000), GLS (Abcam,ab60709, 1:1,000), c-Myc (Cell Signaling Technology, 5605, clone D84C12, 1:1,000),glial fibrillary acidic protein (Dako, Z0334, 1:10,000), SOX2 (Abcam, ab75485, clone57CT23.3.4, 1:1,000), CD133 (Miltenyi Biotec, W6B3C1, clone W6B3C1, 1:1,000),Olig2 (R&D system, AF2418, 1:500), actin (Santa Cruz Biotechnology, sc-1616-R,1:1,000). Membranes were then washed and incubated with secondary donkey anti-rabbit (Licor 926 32213, 1:10,000) or donkey anti-mouse (Licor 926 32212, 1:10,000).The IR scanning was performed using a Licor Odyssey scanner and acquired usingImage Studio 2.0. All western blots shown are representative of two experimentsunless otherwise indicated.

qPCR analysis. For qPCR analysis 1 µg of total RNAwas isolated from cells with theRNeasy Mini Kit (Qiagen) and retro-transcribed into cDNA using M-Mulv ReverseTranscriptase and random hexamers. qPCR reactions were performed using a FastSybr Green Master Mix (Applied Biosystems), 0.5mM primers, ROX dye, and 25 ngof cDNA in a final volume of 20 µl according to the manufacturer’s instructions.The following primers were used: MYC_F 5′-CACCAGCAGCGACTCTGA-3′,MYC_R 5′-GATCCAGACTCTGACCTTTTGC-3′; GS_F 5′-TCATCTTGCATCGTGTGTGTG-3′, GS_R 5′-CTTCAGACCATTCTCCTCCCG-3′; GLS_F 5′-GAAATTCGGAACAAGACTGTG-3′, GLS_R 5′-AACTTCGATGTGTCCTTCAC-3′,β-actin_F 5′-TCCATCATGAAGTGTGACGT-3′, β-actin_ R 5′-TACTCCTGCTTGCTGATCCAC-3′. The following PCR steps were performed on the 7500 FastReal-Time PCR System (Life Technologies Corporation): 15min at 95 ◦C followedby 40 cycles of 10 s at 94 ◦C, 30 s at 55 ◦C, 30 s at 72 ◦C. A 10min final step at 72 ◦Cwas performed to visualize the PCR products by means of the melting curves. Theexpression levels of the indicated genes were calculated using the 11CT methodusing actin for normalization.

Metabolites extraction and HPLC–MS. The established cell lines were platedin 6-well plates in complete DMEM from 125,000 to 250,000 cells per well. After24 h the medium was replaced with SMEM supplemented with 10% dialysed FBS.Stem-like and differentiated primary glioma stem cells were plated at 150,000 cellsper well in 6-well plates in SMEM supplemented as described in the cell culturesection. When optimal confluence was reached (1–3 days) 3 wells were extractedas described below, and the protein was determined (µg prott0 ). The medium inthe remaining wells was changed with 2 or 7ml per well of complete SMEM forthe analysis of metabolites present in the medium or in the cells, respectively.U-13C5–Gln (0.65mM, Cambridge Isotopes), U-13C6–glucose (5.56mM, CambridgeIsotopes) or 15N1–ammonia (0.8mM, Sigma) was supplemented as indicated inthe figures. Cells, as well as parallel cell-free wells, were incubated for 24 h at37 ◦C. At the end of the incubation, an aliquot of the media was diluted 1:50in a solution containing 20% water, 50% methanol and 30% acetonitrile. Themedium derived from wells devoid of cells was used to estimate the exchangerates of metabolites. For the profiling of intracellular metabolites, monolayerswere rapidly washed 3 times with ice-cold PBS and extracted with an aqueoussolution of 50% methanol, and 30% acetonitrile. Both media and cell extracts

were centrifuged at 16,000g for 10min at 4 ◦C and the supernatants were analysedby means of HPLC–MS. The extracted cell monolayer was used for proteindetermination (µg prott24 ) with a Lowry assay. Aliquots of freshly preparedSMEM, without FBS, and spiked with known concentrations of lactate wereprocessed in parallel and used as a reference for metabolite quantification.The secretion/consumption rate for a specific metabolite (x) was obtainedaccording to the equation x = (2× (1metabolite)/µg prott0 + µg prott24 ), where1metabolite= ((x)nmol medium with no cell− (x)nmol medium with cells). Forthe LC separation of medium samples a ZIC-HILIC (SeQuant) with a guard column(Hichrom) was used. Mobile phase A was a 0.1% formic acid solution in water andmobile phase B was 0.1% formic acid in acetonitrile. The flow rate of the mobilephases was kept at 100 µl min−1 and the gradient was as follows: 0min 80% of B,12min 50% of B, 26min 50% of B, 28min 20% of B, 36min 20% of B and 37 to45min 80% of B. For the separation, cell extracts were injected on a ZIC-pHILICcolumn with a guard column. (Mobile phase C: 20mM ammonia carbonate plus0.1% ammonia hydroxide in water. Mobile phase D: acetonitrile.) The flow rate waskept at 100 µl min−1 and gradient as follows: 0min 80%ofD, 30min 20%ofD, 31min80% of D and 45min 80% of D. The Exactive Orbitrap mass spectrometer (ThermoScientific) was operated in a polarity switching mode.

Cell transfection and infection. For stable transfection 10 cm dishes of80% confluent LN18 cells were incubated for 9 h with 10ml of FBS-freetransfection medium containing 1 µgml−1 of DNA (vector containing P2A-iRFPIRES puro alone, or GS-P2A-iRFP IRES puro), and 2 µl ml−1 of Lipofectamine2000(Invitrogen). The transfection mixture was than replaced with 10ml of completeDMEM. The following day medium was changed and cells were cultured for 3weeks in selectionmedium containing 2 µgml−1 of puromycin. Fluorescent colonieswere visualized with the Licor Odyssey scanner, picked and amplified in selectionmedium to obtain stable cultures. For the establishment of cell lines expressingshRNA against glutamine synthetase (GS), four human unique shRNA sequencesof 29 nucleotides, and a scrambled negative control non-effective shRNA cassette(NTC) in a lentiviral GFP vector, were employed according to the manufacturer’sinstructions (TL312740a: 5′-GGTGAGAAAGTCCAGGCCATGTATATCTG-3′;TL312740b: 5′-GAATGGTGCAGGCTGCCATACCAACTTCA-3′; TL312740c: 5′-TGGTACTGGAGAAGGACTGCGCTGCAAGA-3′; TL312740d: 5′-GGCACACCTGTAAACGGATAATGGACATG-3′; TR30021ntc: 5′-GCACTACCAGAGCTAACTCAGATAGTACT-3′, Origene). TL312740b, TL312740d and TR30021ntc arereported in themanuscript asGS shRNA1GS shRNA2 andNTC shRNArespectively.Briefly, lentiviruses were produced by co-transfecting HEK293T cells with eachshRNA plasmid, lentiviral plasmid psPAX2-1, and packaging plasmid pVSV-G-1,according to a standard calciumphosphate procedure. Supernatants containing viruswere collected, filtered and frozen in stocks at−80 ◦C. For infection of recipient cellsthe supernatant was thawed, mixed with hexadimethrine bromide (Sigma, H9268,8 µgml−1) and incubated for 24 h with the cells. Infected cells were selected for 2weeks in medium containing 1–3 µgml−1 puromycin.

Colony-forming assay. iRFP and iRFP–GS-expressing cells were plated at 300cells per well in 6-well plates in complete SMEM, in the absence or presence of0.65mM Gln and 1mM MSO as indicated. After 3 weeks, colonies were fixedwith trichloroacetic acid, stained with SRB and plates were scanned using a LicorOdyssey scanner. Imageswere acquired using Image Studio 2.0. SF188 andU251 cellsexpressing NTC shRNA and GS shRNA sequences were seeded in 6-well plates, at500 and 350 cells perwell for SF188 andU251 cells, respectively. Cells were incubatedin complete SMEM and supplemented as indicated in the legend of Fig. 3. Twelve toseventeen days after seeding, colonies were fixed, stained, and representative imagesobtained as described above. For the quantification of colony surface area an ImageJmacro was designed and run. The values obtained from the macro were normalizedfor the exact number of days cells were incubated, and presented as percentage ofrelative control.

Reagents. U-13C5–L-Glutamine and U-13C6–D-Glucose (CLM-1822 andCLM-1396, respectively) were obtained from Cambridge Isotopes Laboratories,15N1 ammonium chloride (299251), sulphasalazine (S0883), MSO (M5379),BPTES (SML0601) and dm-αKG (349631) were obtained from Sigma Aldrich.Compound 968 (352010) was obtained from Calbiochem. BPTES and compound968 were dissolved 10mM in DMSO. Erwinase was obtained from Eusa Pharma,an international division of Jazz Pharmaceuticals. All amino acids, nucleosides, andnitrogenous bases used in the study were obtained from Sigma Aldrich.

Genome-scale constraint-basedmetabolic modelling.A detailed description ofthemetabolicmodellingmethods and results is provided in the SupplementaryNote.

Statistical methods. No statistical method was used to predetermine sample size.For the animal studies, the experiments were not randomized, and the investigators

NATURE CELL BIOLOGY


DOI: 10.1038/ncb3272 METHODS

were not blinded to allocation during experiments. The number of independentexperiments performed is reported in the figure legends. For experiments performedonce or twice, the raw data of independent repeats are provided in the statisticssource data. Error bars represent standard error of the mean (s.e.m.). Two-tailedStudent’s t-tests, Pearson correlation tests, and Log-rank (Mantel–Cox) tests wereperformed with Graph Pad Prism 5.0.

40. Noble, M. &Murray, K. Purified astrocytes promote the in vitro division of a bipotentialglial progenitor cell. EMBO J. 3, 2243–2247 (1984).

41. Keunen, O. et al. Anti-VEGF treatment reduces blood supply and increasestumor cell invasion in glioblastoma. Proc. Natl Acad. Sci. USA 108,3749–3754 (2011).

42. Sanzey, M. et al. Comprehensive analysis of glycolytic enzymes as therapeutic targetsin the treatment of glioblastoma. PLoS ONE 10, e0123544 (2015).

NATURE CELL BIOLOGY


S U P P L E M E N TA RY I N F O R M AT I O N

WWW.NATURE.COM/NATURECELLBIOLOGY 1

DOI: 10.1038/ncb3272

Supplementary Figure 1 Glutamine starvation reduces GBM cell proliferation. (a) Representative microscopic fields of cells incubated for 6 days with or without Gln. (b) Cell cycle distribution of cells incubated for 3 days with or

without Gln. Mean ± S.E.M. n=3 independent experiments. p value refers to a two-tailed t test for unpaired samples. Raw data of independent repeats are provided in the statistics source data Supplementary Table 5.


http://www.nature.com/doifinder/10.1038/ncb3272


2 WWW.NATURE.COM/NATURECELLBIOLOGY

Supplementary Figure 2 Exchange rates of metabolites in GBM cell lines. Cells were incubated for 24h +/- U-13C5 Gln, secretion rates (positive bars) and consumption rates (negative bars) of the indicated metabolites are

shown. The sum of all isotopologues is reported for clarity. Alanine (Ala) exchange rates are magnified in the inset. Mean ± S.E.M. n=3 independent experiments.




Supplementary Figure 3 Effects of Gln availability and Glutaminase inhibition on GBM cell lines growth and redox state. (a) Correlation between Gln consumption and growth inhibition caused by Gln starvation is shown in six cell lines. Mean ± S.E.M. n=3 independent experiments. (b) Citrate isotopologues distribution of 13C5-Gln derived atoms. 13C6-Citrate was not detected. Mean ± S.E.M. n=3 independent experiments. (c) Oxidized (GSSG) to reduced (GSH) glutathione ratio. LN18 cells were incubated for 24h +/- Gln and supplemented with Glu, α-ketoglutarate dimethylester (dm-

αKG), sulfasalazine (SSZ), or cystine, as indicated. GSSG/GSH ratio was assessed by HPLC-MS, and shown as % of untreated control. Data derive from one experiment performed once. Raw data of independent repeats are provided in the statistics source data Supplementary Table 5. (d-i) Cells were incubated with 0, 2.5, 5, 10, 15, 30μM BPTES for 72h, in the absence (solid line) or presence (dotted line) of 4mM dm-αKG and counted. In all conditions cells were exposed to 0.3%DMSO. Mean ± S.E.M. n=3 independent experiments.




Supplementary Figure 4 GS expression and its effects on colony formation capacity, and glucosamine biosynthetic pathway in GBM cells. (a) Cells were incubated for 24h +/- Gln and GLUL (GS), GLS, and MYC mRNA relative expression was assessed by qPCR. Actin expression was used for normalization. Data derive from one experiment performed once. Raw data of independent repeats are provided in the statistics source data Supplementary Table 5. (b) SF188 and U251 cells stably expressing a non-targeting control shRNA (shNTC) and two sequences targeting GS (shGS-1 and shGS-2) were cultured for 12-17 days +/- Gln in medium

supplemented with glutamate (4mM), and ammonia (0.8mM) as indicated. The quantification of colonies surface area was obtained as described in the Methods section. Mean ± S.E.M., n=4 independent experiments. (c) iRFP4 and iRFP-GS5 expressing LN18 cells were incubated +/- Gln for 24h in medium supplemented with 0.8 mM 15NH4

+. The intracellular isotopologues

distribution of UDP N-acetylglucosamine is shown as % of control (value obtained for the metabolites in iRFP4 cells + Gln). Data derive from one experiment performed twice. Raw data of independent repeats are provided in the statistics source data Supplementary Table 5.




Supplementary Figure 5 Heterogeneous GS expression in human GBM biopsies. The whole bioptic tissues of three patients included in the tissue microarray were stained for GS, and Hematoxilin and Eosin [H&E], as indicated. Low magnifications of whole biopsies are reported in panels a-b, h-i, n-o. For each biopsy, tissue regions framed in blue and green are magnified in d-g, j-m, q-t. Biopsy 1 shows necrotic tissue [nec] surrounded by a solid tumour core [TC] with uniformly low GS staining (d,f). (d-e) Invasive tumour areas [TI] show infiltrative cells with heterogeneous

immunoreactivity for GS. (c) Normal astrocytes [black arrows] in the adjacent normal brain [Br] were positive for GS, while neurons [white arrows] were GS-negative. Biopsy 2 shows distinct areas with low/absent (k) and high (j) immunoreactivity for GS. GS-negative inflammatory tissue [i] surrounding a necrotic region [nec] is evident in panels h-i, k-m. Biopsy 3 shows distinct areas with low/absent (q) and high (r) immunoreactivity for GS. (p) GS-positive reactive astrocytes [arrows] with extensive cellular processes forming a network of GS positive fibers surrounding GS-negative tumour cells.




Supplementary Figure 6 Glucose and glutamine metabolic fates in GBM patients, and orthotopic mouse models. (a) Serum levels of 13C6 and 13C0 glucose in three GBM patients injected with 13C6 glucose before surgical intervention for tumour removal. The dashed line represents the time at which glucose infusion starts. Black arrow points at the time of surgical resection. Values are reported as % of the basal values obtained before infusion. (b) 13C Lactate and (c) 13C Glutamate enrichment in serum at time of tumour resection, in tumour tissue, and in adjacent edematous tissue, of seven (1-7) GBM patients injected with 13C6-glucose. The % of lactate or glutamate incorporating one or more 13C carbons over the total amount of lactate or glutamate detected is reported; na: not available, nd: not detectable. Values were corrected for the natural abundance of

13C. (d-e) Isotopologue distribution of metabolites obtained in the liver of mice orthotopically xenografted with human P3 GBM and injected with 13C6 Glucose (d) or 13C5-Gln (e). The values are mean ± S.E.M. n=3 mice. (f-g) Isotopologues of Gln (f) and glucose (g) found in the serum of mice orthotopically xenografted with the human GS-positive T16 GBM, and infused for 4h into the carotid artery with 13C5-Gln. Mean ± S.E.M. n=3 mice. (h-j) Isotopologue distributions of Gln, Glu and α-kg in T16 tumours, contralateral brain, and liver tissues of mice infused as in (f-g). Mean ± S.E.M n=3 mice. (k) Astrocytes were incubated for 24hours +/- Gln, and asparagine consumption is reported. Data derive from one experiment performed twice. Raw data of independent repeats are provided in the statistics source data Supplementary Table 5.




Supplementary Figure 7 A toy network used for metabolic modelling. A toy network (a) and its optimal flux distribution (b).




Supplementary Figure 8 Unprocessed scans of western blots accompanied by size markers. Red dotted lines delineate the region presented in the respective Figures as indicated. Images were obtained using a Licor Odyssey scanner and acquired using Image Studio 2.0.




Supplementary Table 1 Formulation of DMEM and Serum-like Modified Eagles Medium (SMEM).




Supplementary Table 2 Exchange rates of metabolites in GBM cell lines. Cells were incubated for 24h -/+ U-13C5-Gln, and secretion (potitive values) and consumption (negative values) rates of the

indicated metabolites are reported as nmol/μg prot/day. The sum of all isotopologues is reported for clarity. Mean ± S.D. n=3 independent experiments.




Supplementary Table 3 Representation of SMEM growth medium in the metabolic model Recon1.




Supplementary Table 4 Weighted costs of growth limiting metabolites.




Supplementary Table 5 Statistics Source Data.


SUPPLEMENTARY NOTE

Genome-scale constraint-based metabolic modeling.

To predict the effects of Gln deprivation on cellular growth we simulated the

experiment in a computational model. Our simulations were run on the generic

model of human metabolism, Recon11. Flux Balance Analysis (FBA)2 and Flux

Imbalance Analysis (FIA)3 were used to query the model and predict cellular growth

rate, essential genes and growth limiting metabolites.

1. Simulated experimental conditions:

In order to simulate the in vitro conditions we set the concentrations of nutrients

available to the model to match those of the SMEM growth medium. The list of

components used for the model growth medium and their concentrations are

reported in Supplementary Table 3. To account for biomass production, we added a

growth reaction representing the steady-state consumption of biomass compounds

required for cellular proliferation. The stoichiometric coefficients of the growth

reaction represent the relative molecular concentrations of 42 essential metabolites.

The addition of a biomass reaction is common practice when modeling proliferating

cells, and was described in detail in previous work 4,5

2. Searching for conditionally essential reactions:

We queried the model for conditionally essential reaction knockouts – reactions that

are non-essential on SMEM medium, yet become essential when Gln is removed

from the medium. The flux through a conditionally essential reaction is thus predicted

to modulate the cell’s sensitivity to Gln deprivation. We used an FBA approach to

identify conditionally essential reactions. For each reaction 𝑟, we simulated the

knockout of 𝑟 and used FBA to simulate the maximal cellular growth rate (denoted


𝐺𝑟!!) when Gln is present\absent from the growth medium (denoted 𝐺𝑙𝑛+,𝐺𝑙𝑛 −,

respectively). 𝑟 was defined as conditionally essential if 𝐺𝑟!!,!"#! > 0 and

𝐺𝑟!!,!"#! = 0.

Our analysis revealed a single conditionally essential reaction: Glutamine synthetase

(GS). The knockout of GS was also found to completely abolish the model’s ability to

replenish intra-cellular Gln.

3. Finding growth limiting metabolites:

We next explored which metabolites became growth limiting during Gln deprivation.

FIA allows for the prediction of growth limiting metabolites by calculating a shadow

price for each metabolite 𝑚. The shadow price of 𝑚 (denoted 𝑆𝑃(𝑚)) represents the

potential increase in cellular growth resulting from the synthesis or secretion of a

small quantity of 𝑚. A negative 𝑆𝑃(𝑚) signifies that 𝑚 is growth limiting, whereas

𝑆𝑃(𝑚) = 0 signifies that 𝑚 is not growth limiting. Positive shadow prices are rare and

theoretically signify toxic metabolites, whose secretion or detoxification would require

resources that could otherwise be used for growth.

Gln deprivation was simulated by removing Gln from the model’s growth medium

and restricting the flux through GS to half the amount necessary to sustain maximal

growth. FIA was used to calculate shadow prices for all biomass constituents,

revealing a negative shadow price for all nucleotides. The shadow prices were seen

to directly represent the number of Gln-derived amide nitrogens necessary for the

synthesis of each nucleotide – 1 amide nitrogen for TMP and UMP, 2 for AMP and

CMP, 3 for GMP. Calculated shadow prices are shown in Supplementary Table 4.

We then calculated the total flux passing through each metabolite 𝑚, denoted by

𝐹(𝑚) and weighted the shadow prices by multiplying 𝑆𝑃 𝑚 by 𝐹(𝑚). This weighting


was performed in order to determine which nucleotide imposes the greatest burden

on cellular growth. The weighting has two justifications:

(1) One caveat of FIA is that the shadow price of 𝑚 is unrelated to the amount of

limiting resource (Gln-derived nitrogen in our case) invested in the synthesis of 𝑚. At

the end of this section we provide a toy network example which better illustrates this

property of FIA.

(2) The total flux 𝐹(𝑚) passing through metabolite 𝑚 represents 𝑚’s direct

contribution to biomass production as well as 𝑚‘s involvement in other growth related

pathways, where 𝑚 participates as an intermediate metabolite or a co-factor. Hence,

𝐹(𝑚) provides a proxy to the significance of 𝑚 to the cellular growth process. We

expect some of the cellular effort to be vested in maintaining the steady state

concentrations of metabolites with high 𝐹(𝑚) values.

4. Detailed calculation of the weighted shadow prices:

For each metabolite 𝑚, 𝑆𝑃(𝑚) was multiplied by 𝐹(𝑚), the minimal absolute flux

passing through 𝑚 under FBA formulation. A flux distribution with minimal absolute

fluxes was calculated using the following linear problem:

𝑚𝑖𝑛𝑖𝑚𝑖𝑧𝑒 𝑣!!

(1)

𝑠. 𝑡.

𝑆𝑣 = 0 (2)

𝑣𝑚𝑖𝑛 ≤ 𝑣 ≤ 𝑣𝑚𝑎𝑥 (3)

𝑣𝑏𝑖𝑜𝑚𝑎𝑠𝑠 = 𝑣𝑚𝑎𝑥𝐵𝑖𝑜𝑚𝑎𝑠𝑠 (4)

Where 𝑣𝑚𝑎𝑥𝐵𝑖𝑜𝑚𝑎𝑠𝑠 is the maximal attainable growth rate calculated using FBA. The

absolute flux 𝐹(𝑚) through metabolite 𝑚 was then calculated as:


𝐹 𝑚 = 𝑆!,!𝑣!!

(5)

Finally, the weighted cost of 𝑚 was estimated by 𝑆𝑃(𝑚) ∙ 𝐹(𝑚). The results of this

calculation are given in Supplementary Table 4.

Following weighting AMP was shown to incur the heaviest cost to the cell, in spite of

its synthesis requiring the same amount of Gln-derived nitrogens as CMP, and less

than GMP. Specifically, AMP is the most limiting, due to its use as a precursor for

ATP, the most abundant nucleotide in the cell, and in other AMP containing cofactors

like NADH, NADPH, FADH, S-Adenosyl methionine, and Coenzyme A. This

indicates that in the absence of exogenous Gln, the availability of the Gln amide

group limits purine biosynthesis. Further, the reported Km for Gln for the enzyme

responsible for the first step in purine biosynthesis is at least one order of magnitude

higher than that for pyrimidine and protein synthesis (EC2.4.2.14: Human

amidophosphoribosyltransferase, Km 1.0 - 4.5 mM; EC6.3.5.5: mammalian

carbamoyl-phosphate synthase, Km 0.1 mM; EC6.1.1.18: S. Cerevisiae Glutamine-

tRNA ligase, Km 0.03 mM; http://www.brenda-enzymes.org, and Human glutaminyl-

tRNA Synthetase catalytically active at 0.05 mM Gln6). The differences in affinity for

Gln suggest that low intracellular Gln concentrations that permit protein and

pyrimidine synthesis, may be limiting for purine biosynthetic pathways.

5. Toy network example

For the sake of clarity, we repeated the analysis of growth limiting metabolites on a

toy network. Supplementary Figure 7A depicts a toy network with five intracellular

metabolites and a growth medium comprised of two extracellular metabolites. The

numbers on the edges signify the stoichiometric coefficients. The growth medium, in

our example, contains 36 (arbitrary) units of metabolite 𝐴 and 64 units of metabolite


𝐵. The aim of the analysis is to determine which metabolite incurs the greatest cost

to the cell in terms of the resources spent on its synthesis.

Supplementary Figure 7B shows an optimal flux distribution across the toy network.

Reaction fluxes are shown in red. Note that in order to calculate the total turnover of

a metabolite by a reaction one must multiply the reaction flux by the metabolite’s

stoichiometric coefficient (i.e. multiply the red and black numbers). The flux

distribution reveals that 𝐴 is the limiting resource whereas 𝐵 is abundant. In order to

sustain maximal growth, the cell invests 24 units of 𝐴 in synthesizing metabolite 𝐶

and 12 units in synthesizing 𝐵, making 𝐶 the most costly investment. The synthesis

of 𝐸 does not incur any cost in terms of the limiting resource 𝐴.

The shadow price 𝑆𝑃(𝑚) of a metabolite 𝑚 in our toy network would be 𝑆𝑃 𝑚 =

−𝑠/𝑧, where 𝑠 is the amount of limiting resource (metabolite 𝐴) necessary for

synthesizing 𝑚 and 𝑧 is the total amount of limiting resource necessary for producing

one unit of biomass. One unit of biomass requires 6 molecules of 𝐶, 2 molecules of

𝐷 and 3 molecules of 𝐸. Multiplying each metabolite by its respective cost in 𝐴 we

get:

𝑧 = 2 ∙ 6+ 3 ∙ 2+ 0 ∙ 3 = 18 (6)

The shadow prices are hence 𝑆𝑃 𝐶 = !!!", 𝑆𝑃 𝐷 = !!

!", 𝑆𝑃 𝐸 = !

!" . It can be seen

that 𝐷 has the highest shadow price, even though most of the limiting resource is

invested in the synthesis of 𝐶.

This behavior of FIA can be corrected by taking into account the absolute flux

passing through each metabolite. In the optimal flux distribution, the absolute fluxes

passing through the biomass constituents are:

𝐹 𝐶 = 1 ∙ 12+ 6 ∙ 2 = 24 (7)


𝐹 𝐷 = 1 ∙ 4+ 2 ∙ 2 = 8 (8)

𝐹 𝐸 = 1 ∙ 6+ 3 ∙ 2 = 12 (9)

(note that in the toy network all product metabolites have a stoichiometric coefficient

of 1). By weighting the shadow prices according to the total flux we get 𝐶𝑜𝑠𝑡 𝐶 =

!!"!", 𝐶𝑜𝑠𝑡 𝐷 = !!"

!", 𝐶𝑜𝑠𝑡 𝐸 = !

!". The weighted cost is directly proportional to the

amount of limiting resource invested in the synthesis of each metabolite and shows

that 𝐶 poses the heaviest limitation on cellular growth, as desired.


SUPPLEMENTARY NOTE REFERENCES

1 Duarte, N. C. et al. Global reconstruction of the human metabolic network

based on genomic and bibliomic data. Proc Natl Acad Sci U S A 104, 1777-

1782, doi:10.1073/pnas.0610772104 (2007).

2 Varma, A. & Palsson, B. O. Predictions for oxygen supply control to enhance

population stability of engineered production strains. Biotechnol Bioeng 43,

275-285, doi:10.1002/bit.260430403 (1994).

3 Reznik, E., Mehta, P. & Segre, D. Flux imbalance analysis and the sensitivity

of cellular growth to changes in metabolite pools. PLoS computational biology

9, e1003195, doi:10.1371/journal.pcbi.1003195 (2013).

4 Folger, O. et al. Predicting selective drug targets in cancer through metabolic

networks. Mol Syst Biol 7, 501, doi:10.1038/msb.2011.35 (2011).

5 Frezza, C. et al. Haem oxygenase is synthetically lethal with the tumour

suppressor fumarate hydratase. Nature 477, 225-228,

doi:10.1038/nature10363 (2011).

6 Zhang, X. et al. Mutations in QARS, encoding glutaminyl-tRNA synthetase,

cause progressive microcephaly, cerebral-cerebellar atrophy, and intractable

seizures. American journal of human genetics 94, 547-558,

doi:10.1016/j.ajhg.2014.03.003 (2014).


Glutamine synthetase activity fuels nucleotide ...ruppin/ncb_glutamine.pdf · glutamine synthetase...

Documents

Transcript of Glutamine synthetase activity fuels nucleotide ...ruppin/ncb_glutamine.pdf · glutamine synthetase...