Glycine Decarboxylase Activity Drives

Non-Small Cell Lung Cancer

Tumor-Initiating Cells and Tumorigenesis
Wen Cai Zhang,1,3 Ng Shyh-Chang,1 He Yang,5 Amit Rai,6 Shivshankar Umashankar,6,7 Siming Ma,1 Boon Seng Soh,1
Li Li Sun,1 Bee Choo Tai,11 Min En Nga,9 Kishore Kumar Bhakoo,12 Senthil Raja Jayapal,13 Massimo Nichane,1 Qiang Yu,2
Dokeu A. Ahmed,4 Christie Tan,4 Wong Poo Sing,10 John Tam,10 Agasthian Thirugananam,14 Monireh Soroush Noghabi,1
Yin Huei Pang,9 Haw Siang Ang,5 Wayne Mitchell,16,17 Paul Robson,1 Philipp Kaldis,13 Ross Andrew Soo,5,8
Sanjay Swarup,6,7 Elaine Hsuen Lim,3,8,15,* and Bing Lim1,18,*
1Stem

Cell and Developmental Biology

Biology and Pharmacology
Genome Institute of Singapore, 60 Biopolis Street, Singapore 138672
3Department of Respiratory Medicine
4Department of Thoracic Surgery
Tan Tock Seng Hospital, 11 Jalan Tan Tock Seng, Singapore 308433
5Cancer Science Institute of Singapore, National University of Singapore, 28 Medical Drive, Singapore 117456
6Metabolites Biology Lab, Department of Biological Sciences, National University of Singapore, 14 Science Drive 4, Singapore 117543
7NUS Environmental Research Institute (NERI), T-Lab Building (TL), National University of Singapore, 5A Engineering Drive 1, Singapore 117411
8Department of Haematology-Oncology
9Department of Pathology
10Department of Thoracic Surgery
National University Hospital, 5 Lower Kent Ridge Road, Singapore 119074
11Department of Community, Occupational and Family Medicine, Yong Loo Lin School of Medicine, National University of Singapore,
Singapore 117597
12Translational Molecular Imaging Group (TMIG), Laboratory of Molecular Imaging (LMI), Singapore BioImaging Consortium (SBIC),
11 Biopolis Way, Singapore 138667
13Cell Division and Cancer Laboratory (PRK), Institute of Molecular and Cell Biology, 61 Biopolis Drive, Singapore 138673
14Department of Thoracic Surgery
15Department of Medical Oncology
National Cancer Centre, 11 Hospital Drive, Singapore 169610
16Bioinformatics, Experimental Therapeutics Centre, 31 Biopolis Way, Singapore 138669
17Division of Information Sciences, School of Computer Engineering, Nanyang Technological University, 50 Nanyang Avenue, Singapore 639798
18Beth-Israel Deaconess Medical Center, Harvard Medical School, 330 Brookline Avenue, Boston, MA 02215, USA
*Correspondence: elainelim77@ymail.com (E.H.L.), limb1@gis.a-star.edu.sg (B.L.)
DOI 10.1016/j.cell.2011.11.050
2Cancer

SUMMARY

Identification of the factors critical to the tumor-initiating cell (TIC) state may open new avenues in cancer
therapy. Here we show that the metabolic enzyme
glycine decarboxylase (GLDC) is critical for TICs in
non-small cell lung cancer (NSCLC). TICs from primary NSCLC tumors express high levels of the oncogenic stem cell factor LIN28B and GLDC, which are
both required for TIC growth and tumorigenesis.
Overexpression of GLDC and other glycine/serine
enzymes, but not catalytically inactive GLDC, promotes cellular transformation and tumorigenesis.
We found that GLDC induces dramatic changes in
glycolysis and glycine/serine metabolism, leading
to changes in pyrimidine metabolism to regulate
cancer cell proliferation. In the clinic, aberrant activation of GLDC correlates with poorer survival in lung

cancer patients, and aberrant GLDC expression is

observed in multiple cancer types. This link between
glycine metabolism and tumorigenesis may provide
novel targets for advancing anticancer therapy.
INTRODUCTION
Despite numerous advances in our knowledge of cancer (Vogelstein and Kinzler, 2004; Hanahan and Weinberg, 2011), our
ability to develop clinically effective therapies based on this
understanding has met with limited success. Current therapies
can control tumor growth initially, but most patients ultimately
relapse. One prominent example is lung cancer, the leading
cause of cancer-related mortality with over 1 million deaths
each year (Jemal et al., 2011). Non-small cell lung cancer
(NSCLC) accounts for approximately 85% of all lung cancers.
Although NSCLC patients with epidermal growth factor receptor
(EGFR) mutations respond to EGFR inhibitors initially, most
patients experience a relapse within 1 year (Sequist et al.,
Cell 148, 259272, January 20, 2012 2012 Elsevier Inc. 259

2007). These findings underscore the urgent need for both combination therapies and also new approaches to treat cancerous
tumors. One such approach may be to target tumor-initiating
cells (TICs).
Data from leukemias, germ cell tumors, and a number of solid
tumors support the notion that cancers are maintained by
a subpopulation of self-renewing and evolving TICs. This is
also popularly known as the cancer stem cell (CSC) model
(Reya et al., 2001; Rosen and Jordan, 2009). Although the validity
of the CSC model is an issue of controversy in melanoma (Quintana et al., 2008, 2010; Boiko et al., 2010; Civenni et al., 2011),
many other solid tumors appear to follow the CSC model (Ishizawa et al., 2010). Recently it was proposed that at earlier stages
of tumorigenesis, rare TIC clones differentiate into nonmalignant
progeny to form the bulk of the tumor, whereas at advanced
stages, TIC clones constitute the bulk of the tumor (Boiko
et al., 2010). Studies with mouse models of lung cancer have
also begun to reconcile the connection between the evolving
genotype of TIC clones and the surface phenotype of TICs (Curtis et al., 2010). Thus accumulated findings suggest that targeting TICs may be a promising approach for eradicating tumors
early. However, progress in the targeting of TICs to improve
cancer therapy has been hindered by a lack of understanding
of the molecular pathways that are critical to TICs.
Recent studies have led to an emerging appreciation of the
importance of metabolic reprogramming in cancer (Hsu and Sabatini, 2008; Vander Heiden et al., 2009; Hanahan and Weinberg,
2011). Most recently, the embryonic isoform of pyruvate kinase
PKM2, in collaboration with phosphoglycerate mutase, was
found to regulate the shift from oxidative phosphorylation to glycolysis in cancer cells (Christofk et al., 2008; Vander
Heiden et al., 2010). These findings have led to a resurgence of
interest in the Warburg effectthe phenomenon whereby cancer
cells, like embryonic cells, preferentially use glycolysis even
under aerobic conditions (Warburg, 1956). Besides glycolysis,
an arm of metabolism that results in sarcosine production has
also been implicated in prostate cancer (Sreekumar et al.,
2009). These data suggest that metabolic reprogramming is
crucial for tumorigenesis, and much remains to be uncovered.
Here we show that glycine metabolism and the metabolic
enzyme glycine decarboxylase (GLDC) drive TICs and tumorigenesis in NSCLC. Using CD166 as a surface marker and
NOD/SCID Il2rg
/
mice as xenotransplantation recipients, we
isolated lung TICs from a broad range of primary NSCLC tumors
(stages IIII). Primary lung TICs express high levels of LIN28B,
GLDC, and many other glycine/serine metabolism enzymes.
Both LIN28B and GLDC were required for lung TIC proliferation
and tumor growth. Overexpression of GLDC alone, and other
glycine/serine enzymes, promotes cellular transformation both
in vitro and in vivo. Metabolomic analysis shows that GLDC overexpression induces dramatic changes in glycolysis and glycine
metabolism, leading to changes in pyrimidine metabolism for
cancer cell proliferation. In human patients, aberrant upregulation of GLDC is significantly associated with higher mortality
from lung cancer, and aberrant GLDC expression is observed
in multiple cancer types. Our findings establish a link between
glycine metabolism and tumorigenesis and may provide novel
targets for advancing anticancer therapy.
260 Cell 148, 259272, January 20, 2012 2012 Elsevier Inc.

RESULTS
TICs in Lung Cancer
To assess the cellular heterogeneity within NSCLC, we obtained
freshly resected lung tumors from 36 human patients with a
broad range of stage IIII primary NSCLC (Table S1 available
online). Patient lung cancer cells were directly transplanted
subcutaneously into NOD/SCID Il2rg
/
mice with Matrigel
(Quintana et al., 2008). Using this maximally sensitive assay,
we estimated by limiting dilution analysis that lung TICs exist
with a low frequency of 1 in 4 3 105 cells in unsorted NSCLC
tumor cells (n = 36 patients; Figure 1A), consistent with published
findings (Ishizawa et al., 2010).
To profile the surface phenotype of this subpopulation of lung
TICs, we fractionated the NSCLC tumors by fluorescenceactivated cell sorting (FACS; Figure S1A). After excluding
hematopoietic and endothelial cells (Lin
), we tested a panel of
cell-surface markers, including CD166, CD44, CD133, and
EpCAM (Figure 1B). We found that CD166 was the most robust
marker for enriching the lung TIC subpopulation, compared to
CD133, CD44, or EpCAM, allowing us to reliably enrich lung
TICs by nearly 100-fold (Figures 1A and 1B). In 12 out of 12
NSCLC patient tumors (lung adenocarcinoma), the CD166+
Lin
fraction contained cells that consistently initiated lung tumor
formation in vivo. In contrast, CD166
Lin
tumor cells generally
failed to initiate lung tumor formation even after 8 months of
observation, although they also expressed carcinoembryonic
antigen (CEA), a tumor-specific marker not expressed in normal
adult lung cells (Figures 1A, 1B, and S1B). Similar results were
observed in lung squamous cell carcinoma and large cell carcinoma (Figure S1C). Although CD166 expression varied across
the NSCLC tumors we examined, CD166 was consistently
higher in lung tumors than in normal adjacent lung tissues
(n = 25 patients; Figures S1D and S1E).
CD166+ lung TICs demonstrate a capacity for self-renewal
and differentiation in vivo. Serial transplantations showed that
only the CD166+ fraction was able to self-renew and initiate primary and secondary xenograft tumors (Figures 1A and S1F).
Upon transplantation, CD166+ lung TICs differentiated to form
xenograft tumors that phenocopy the complex cytoarchitecture of their parental patient tumors, sharing similar histological morphology by hematoxylin-eosin (H&E) staining and
similar tissue distributions of CD166, cytokeratin, E-cadherin,
vimentin, smooth muscle actin, and synaptophysin (Figures
1C, S1G, and S1H). Furthermore, we found that transplants
with more TICs grow more rapidly, suggesting that lung TIC
frequency is correlated with tumor growth rate (Figures 1D
and S1I).
The self-renewal capacity of CD166+ lung TICs is further
corroborated by in vitro assays. We tested the CD166+ fraction
for the ability to form tumor spheres, a widely used in vitro technique for assessing self-renewal capacity (Dontu et al., 2003).
Although both primary CD166+ and CD166
cells remained
viable in vitro, only primary CD166+ but not CD166
cells were
able to form compact self-renewing spheres (n = 9 patients;
Figures 1E, 1F, and S1J). Using immunofluorescence and flow
cytometry, we found that the lung tumor spheres retained high
levels of CD166 expression but undetectable CD133 expression

150,000
200,000
300,000
500,000
1,000,000
2,000,000

2/8
2/3
7/9
2/2
4/4
1/1

210,000
260,000
500,000

0/1
1/3
0/3

CD166+Lin-

CD166-Lin-

2/4
1/10
4/5
3/5
9/9
2/2

1,000
5,000
10,000
23,000
25,000
50,000

Enrichment
factor (x)

1/403,971
(1/255,949 1/637,598)

1/5,175
(1/2,837 1/9,441)

78

1.37E-23

1/3,297,091
(1/460,464 1/23,608,384)

0.1

4.82E-03

No. tumors/
No. injections
12/12
0/10
6/8
6/8
6/11
7/13
10/13
8/14

Population
CD166+
CD166CD44+
CD44CD133+
CD133EpCAM+
EpCAM-

Xenograft tumor

Patient tumor

P value

3)
Tumor volume (mm
(

Unsorted

LTICs
frequency
(95% CI)

H&E

No. tumors
/ No.
injections

CD166

No. cells per

injection

Cells

pan-CK

CD166+Lin- (1st xenograft; 5,000)

CD166+Lin- (patient; 5,000)
Lin- (patient; 150,000)
CD166-Lin100,000))
i (patient;
( i

400
300

AdC

200
100
0
14

15

16

17

18

19

20

21

Weeks

CD166+Lin-

AdC

CD
D166-Lin-

Tumor sphere assay

SCC

LCC

No. of spheres / 10,000 cells

G
120
90
60
30
0

AdC

SCC

LCC

No. No. tumors/ No. injections

Cells 1st passage 10th passage
2500
12/12
N.D.
500
18/18
18/18
50
24/24
24/24
10
12/12
12/12
5
16/18
17/18
1
5/18
5/18

Figure 1. CD166+ Fraction Contains TICs from NSCLC Patients

(A) Frequency of TICs in unsorted, CD166+, and CD166
subpopulations of cells from 36 NSCLC patients. CI, confidence interval. Lin
cells, CD45
CD31
cells.
(B) Tumor initiation frequency by various FACS-purified fractions of Lin
lung cancer cells isolated from primary xenograft tumors shown in (A). 5 3 104 CD166,
CD44, CD133, and EpCAM cells were tested for tumor initiation in NOD/SCID Il2rg
/
mice (n = 12).
(C) Histological analysis of patient tumors and primary xenografts derived from patient tumor CD166+ Lin
lung cancer cells. Tumors were stained for H&E,
CD166, and pan-CK (cytokeratin). Scale bar, 50 mm.
(D) Representative tumor-growth curves of xenografts derived from different cell fractions in a lung adenocarcinoma (AdC) patient tumor and the primary
xenograft.
(E) Phase-contrast images of tumor spheres seeded with CD166+ Lin
(top) and CD166
Lin
(bottom) cells in lung adenocarcinoma (AdC), squamous cell
carcinoma (SCC), and large cell carcinoma (LCC). Scale bar, 50 mm.
(F) Quantification of tumor spheres formed by cells from NSCLC patient (AdC, SCC, or LCC) CD166+Lin
, CD166
Lin
, and Lin
populations.
(G) Frequency of tumorigenesis by single patient-derived tumor sphere cells (n = 3). N.D., not determined.
In all panels, error bars represent the standard error of the mean (SEM). See also Figure S1 and Table S1.

in contrast (Figures S1K and S1L). When primary lung tumor

spheres were dissociated into single cells and transplanted
into NOD/SCID Il2rg
/
mice in vivo, we found that as few as
15 single cells consistently initiated tumorigenesis (Figures 1G
and S1M).

The increased tumor-initiating frequency of lung tumor sphere

cells suggests that they are even more highly enriched for lung
TICs than the patient tumor CD166+ fraction, and that lung
TICs expanded during in vitro culture to form tumor spheres.
To test whether CD166 drives tumorigenicity in lung TICs, we
Cell 148, 259272, January 20, 2012 2012 Elsevier Inc. 261

S/XX+/X-

P+/PN+/N-

Candidate
genes

4
Log2 ratio

Log2 ratio

Enrichment
c e
X+
S

-4

-4
N+/X-

P+/X-

X+/X-

N+/X-

S/X-

P+/X-

X+/X_

S/X-

S: sphere; X, xenograft tumor; P, primary tumor;

N, normal tissue; +, CD166+; -, CD166-.

D
Percentage of genes in pathway
0%
TNFAIP3
IGLL1

2
1

HSPA1A

N+/NRRAD

P+/P-

CXCL2
GADD45G

X+/X-

PPP1R15A
RASD1
MMP9

S/X+

RENBP

FOXK2

REEP1

-2

2%

4%

6%

8%

10%

Cell cycle
DNA replication
G i serine
Glycine,
i and threonine
i
metabolism

SMC6
PEA3

-1

TMEM118

0
GLDC
NPAS1

Relative mRNA expressioon level (log2 ratio)

LIN28B
RG9MTD2
CCNB1

Pyrimidine metabolism
MAPK signaling pathway

-3

Experimental
Predicted

p53 signaling pathway

-4
-5

E
3P-Hydroxy pyruvate

PSAT1

Phosphoserine

PSPH

Serine

SHMT1

Glycine

GLDC

CO2 + NH3 + CH2

GCAT
L-2-Amino-acetoacetate

Figure 2. Lung TICs Express High Levels of GLDC and LIN28B

(A) Venn diagram showing strategy for enriching tumorigenic gene expression profile by genome-wide transcriptome analysis. A list of the differentially expressed
genes (cutoff threshold of 1.5-fold, p < 0.05) common between P+ versus P
, X+ versus X
, and S versus X
was derived, with the differentially expressed genes of
N+ versus N
excluded. The gene list was further filtered to select only genes further upregulated or downregulated in S versus X+. N, normal lung tissue (n = 3); P,
patient tumor (n = 1); S, tumor sphere (n = 4); X, xenograft tumor (n = 3); +, CD166+;
, CD166
. Total, n = 11.
(B) Graphs of relative expression of candidate lung TIC-associated genes in increasing (left, n = 194) or decreasing (right, n = 295) trends across different CD166+
fraction cells from normal lung tissue (N), primary tumor (P), xenograft tumor (X), and tumor spheres (S) versus nontumorigenic CD166
(X
) cells.
(C) Top-ranked genes differentially expressed in lung TICs.
(D) Enrichment of KEGG pathways by genes differentially expressed in lung TICs. *p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001.
(E) Schematic diagram of glycine, serine, and threonine metabolism genes significantly (p < 0.05) upregulated in lung TICs. [, upregulation.
See also Figure S2.

knocked down CD166 in two lines of NSCLC patient-derived

tumor spheres by retroviral shRNA (Figure S1N). We found that
the tumorigenicity of lung TICs in the tumor spheres was not
significantly affected by CD166 shRNA, demonstrating that
CD166 is an inert cell-surface marker that enriches for lung
TICs (Figures S1OS1Q).
Lung TICs Express High levels of Glycine/Serine
Metabolism Enzymes
To gain a deeper understanding of the molecular basis for the
TIC state and its tumorigenic capacity, we sought to obtain
262 Cell 148, 259272, January 20, 2012 2012 Elsevier Inc.

a molecular signature for lung TICs. To do this, we performed

genome-wide transcriptome analysis on CD166
Lin
tumor
cells, CD166+ Lin
tumor cells, and lung tumor spheres, in
increasing order of lung TIC frequency (Figure 2A). As a negative
control, we also profiled CD166+ versus CD166
cells from
normal adjacent lung tissues (n = 3 patients; Table S1). This
led us to a profile of genes that are upregulated and downregulated in lung TICs, compared to non-TICs (Figure 2B).
Lung TIC-associated genes include the oncogenic stem cell
factor LIN28B, embryonic lung transcription factors like PEA3
and the trachealess homolog NPAS1 (Viswanathan et al., 2009;

Liu et al., 2003; Levesque et al., 2007), as well as cell-cycle regulators like CCNB1 and GADD45G (Figure 2C). The highestranking genes were validated by qRT-PCR (Figure S2A). KEGG
pathway analysis of the lung TIC-gene profile showed that the
top enriched pathways were cell cycle, DNA replication,
glycine, serine, and threonine metabolism, pyrimidine metabolism, MAPK signaling pathway, and p53 signaling pathway (Figure 2D). Within the glycine, serine, and threonine
metabolism pathway, we found that glycine/serine metabolism
enzymes like GLDC, glycine C-acetyltransferase (GCAT), serine
hydroxymethyltransferase (SHMT1), phosphoserine phosphatase (PSPH), and phosphoserine aminotransferase (PSAT1)
were all upregulated in lung TICs (Figures 2E and S2BS2D). In
particular, GLDC was one of the most highly upregulated genes
in multiple analyses of lung TIC-enriched populations, at both the
mRNA and protein levels (Figures 2C and S2C). GLDC is a key
component of the highly conserved glycine cleavage system
in amino acid metabolism that catalyzes the breakdown of
glycine to form CO2, NH3, and 5,10-methylene-tetrahydrofolate
(CH2-THF) to fuel one-carbon metabolism (Kume et al., 1991).
GLDC Is an Oncogene that Promotes Tumorigenesis
and Cellular Transformation
High expression of GLDC and LIN28B in lung TIC-enriched populations, but not in CD166
lung cancer cells and CD166+ normal
lung cells, suggests that these two genes drive tumorigenicity in
lung TICs. To test this hypothesis, we knocked down GLDC and
LIN28B in lung tumor spheres with shRNAs (Figure S3A) and
compared their growth both in vitro and in vivo. We found that
both GLDC and LIN28B were necessary for cell proliferation in
sphere culture, as well as anchorage-independent colony formation in soft agar (Figures 3A and S3B). Importantly, tumorigenicity
was also significantly reduced upon knockdown of either GLDC
or LIN28B (Figures 3B and S3C). A549 lung adenocarcinoma
cells showed similar results (Figures S3DS3G). Our results
suggest that lung TICs and lung tumorigenesis are dependent
on GLDC. This led us to ask what oncogenes upregulate
GLDC. Because the E2F pathway upregulates many metabolic
genes during cell proliferation, we examined the expression of
GLDC over the course of the cell cycle in both normal human
lung fibroblasts (HLFs) and the transformed A549 cells after
synchronization by serum starvation. Our results showed that
GLDC is insensitive to cell-cycle progression in both normal
HLFs and transformed A549 cells, suggesting that GLDC is not
regulated by cell-cycle or E2F signals (Figure 3C). We then
examined GLDC levels in MCF10A cells after transformation by
oncogenic KRASG12D, PIK3CAE545K, and MYCT58A. Our results
show that all three oncogenes induce GLDC by 20-fold, suggesting that oncogene-induced GLDC transcription is common
to the cellular transformation process mediated by oncogenic
Ras, PI3K, and Myc (Figure 3D).
To test whether aberrant GLDC upregulation is sufficient
to drive cellular transformation, as has been shown for LIN28B
(Viswanathan et al., 2009), we overexpressed GLDC in NIH/
3T3 cells (Figure S3H). We found that GLDC overexpression
significantly increased colony formation by 3T3 cells under normal culture conditions (Figures 3E and 3F). To test for cellular
transformation in vitro, we cultured the 3T3 cells overexpressing

GLDC under anchorage-independent conditions and found that

GLDC transforms 3T3 cells readily with a rate exceeding that of
LIN28B (Figures 3G and S3I). Upon transplantation into NOD/
SCID Il2rg
/
mice, 3T3 cells overexpressing GLDC consistently
formed tumors in 6/6 transplants, and 3T3 cells overexpressing
LIN28B formed tumors in 3/6 transplants, whereas 3T3 cells
overexpressing the empty control vector never formed tumors
(Figures S3JS3L).
To test whether GLDC can also transform normal primary HLF
and normal primary human bronchial epithelial (NHBE) cells, we
overexpressed GLDC in HLF and NHBE cells (Figures S3M
and S3O). Both HLF and NHBE cells showed a dramatic increase
in cell proliferation upon overexpression of GLDC alone (Figures 3H3J and S3P). Surprisingly we found that GLDC also
transforms HLF and NHBE cells readily in vitro (Figures 3K
and S3Q). However, perhaps because primary adult HLF and
NHBE cells are not immortalized, GLDC-overexpressing HLF
and NHBE cells do not form tumors upon transplantation (Figures S3N and S3R). In contrast, CD166
lung tumor cells, which
also could not form tumors in vivo, could now initiate tumorigenesis at a low frequency upon overexpression of GLDC (Figure 3L).
Collectively, our results show that GLDC is an oncogene that is
both necessary and sufficient to promote tumorigenesis.
GLDC Promotes Tumorigenesis through Its Metabolic
Activity
Although GLDC is a metabolic enzyme, it remained unclear
whether GLDC promotes tumorigenesis through a metabolismdependent or -independent mechanism. To address this question, we engineered a series of four point mutations within or
near the evolutionarily conserved catalytic active site of the
GLDC enzyme to disrupt its metabolic activity (Figure 4A). These
point mutations comprised three nonlethal GLDC mutations
found in human patients with nonketotic hyperglycinemia
(H753P, P769L, G771R; Figures S4A and S4B) and one mutation
K754A that is predicted to abrogate the covalent bond with the
critical pyridoxal-50 -phosphate cofactor (Nakai et al., 2005;
Kure et al., 2006). When we overexpressed these four GLDC
mutants in 3T3 cells, none of them could lead to tumorigenesis
in vivo, whereas wild-type GLDC could, even though all of
them were expressed at high levels similar to that in transformed
A549 cells (Figure 4B). Thus the metabolic activity of GLDC is
required for its tumorigenic function.
In addition, the upregulation of many other upstream enzymes
in the glycine/serine pathway in lung TICs further supports the
idea that metabolic activity in the glycine/serine pathway is responsible for promoting tumorigenesis (Figure 2E). To test this
idea, we also overexpressed PSAT1, PSPH, SHMT1, SHMT2,
and GCAT in 3T3 cells and transplanted them in vivo to test
for cellular transformation and tumorigenesis (Figure 4C). By
2 months, we found that three other glycine/serine enzymes
PSAT1, PSPH, and SHMT2could also transform 3T3 cells to
form tumors in vivo (Figure 4D). Interestingly, we noted that
PSAT1, PSPH, and SHMT2 overexpression only led to a slight
upregulation of GLDC protein (Figure 4E), suggesting that their
tumorigenic activity is due to increased glycine/serine metabolism, rather than indirect upregulation of GLDC. These findings
indicate that increased metabolism in the glycine/serine pathway
Cell 148, 259272, January 20, 2012 2012 Elsevier Inc. 263

Tumor spheres

Number of cells

80,000
60,000
40,000
20,000

488 h

4h

8h

244 h

0.6

Time after release in serum

488 h

4h

166 h

2h

0h

0.8

A549

Time after release in serum

1h

TS-Ctrl-sh
TS-28B-sh
TS-GD-sh

100,000

HLF

Ctrl-sh
GD-sh
28B h
28B-sh

cyyc

120,000

Mass of xenograft tumourrs (g)

244 h

GLDC

GLDC

FOS

CDK1

CDK1

HSP90

*
0.4
**
0.2

HSP90

0
0

Days

MCF10A

3T3

30

100

3T3-GD

10

3T3-28B

**

**

75
50
25

3T3
3T3-GD
3T3-28B

80
No. of agar colonies / 55,000
3T3

20

3T3
3T3-GD
3T3-28B

125
No. of adherent colon
nies /
500 3T3

3T3

Relative GLDC eexpression

70
60

**

**

50
40
30
20
10

HLF

40,000

HLF-28B

20,000
0
0

140
No. of adherent colonies /
1,000 HL
LF

HLF

60,000

HL
LF-GD

Numbeer of cells

HLF

HLF-28B

120
100
80
60
40
20
0

Days

**

HLF-GD

80,000

K
HLF
HLF-GD
HLF-28B
**

100,000

35
No. of agar colon
nies / 5,000
HLF

30
25

HLF
HLF-GD
HLF-28B
**

**

20
15
10
5
0

g
Xenograft

L
Group

No. cells /
injection

No. tumors/
No. injections

200,000

0/12

CD166100,000

0/12

200,000

4/12

100,000
,

1/12

CD166- GD

Figure 3. GLDC and LIN28B Are Necessary and Sufficient for Malignant Growth
(A) Proliferation curve of tumor sphere (TS) cells with shRNA knockdown of either GLDC (GD-sh) or LIN28B (28B-sh).
(B) Quantitative mass analysis of xenograft tumors formed 30 days after transplanting 100,000 tumor sphere cells with either GLDC knockdown (GD-sh) or
LIN28B knockdown (28B-sh).
(C) Western blot analysis of endogenous GLDC during the cell cycle in synchronized normal human HLFs and transformed A549 cells. HLF or A549 cells were
serum-starved for 72 hr followed by release into serum-containing medium with samples collected at indicated time points. Expression of GLDC, FOS (early

264 Cell 148, 259272, January 20, 2012 2012 Elsevier Inc.

K754A

P769L

G771R

Vector

8/
8

0/
8

0/
8

0/
8

0/
8

0/
8

A549

H753P

WT

GLDC

No. tumors/
No. injections

R
Relative
RNA expression level

8/
8

100,000
10,000
1,000
100
10
1

Group
3T3-GLDC
3T3-PSAT1
3T3-PSPH
3T3-SHMT2
3T3-SHMT1
3T3-GCAT
3T3

No. tumors/
No. injections
6/6
3/6
2/6
1/6
0/6
0/6
0/6

GLDC-GFP
GLDC
-Actin

Figure 4. GLDC Promotes Tumorigenesis through Its Metabolic Activity

(A) Crystal structure of T. thermophilus GLDC near the catalytic active site. The labeled bacterial residues H703b, K704b (purple), P729b, and G731b are
homologous to H753, K754, P769, and G711 of human GLDC. Residues implicated in human nonketotic hyperglycinemia are shown (red). PLP, pyridoxal-50 phosphate cofactor (green).
(B) GLDC protein expression and tumor formation efficiency of 3T3 cells overexpressing wild-type or mutant GLDC. Four point mutations were tested: H753P,
K754A, P769L, and G771R. Incidence of tumor formation was determined 2 months after injection with 1.5 3 106 cells per mouse (n = 8). A549 cells served as
a positive control. WT, wild-type.
(C) Gene expression in 3T3 cells overexpressing GLDC, PSPH, PSAT1, GCAT, SHMT1, and SHMT2, relative to 3T3 cells with the empty vector, as determined by
qRT-PCR.
(D) Tumor formation efficiency of 3T3 cells overexpressing GLDC, PSPH, PSAT1, GCAT, SHMT1, SHMT2, or the empty vector. Incidence of tumor formation was
determined 2 months after injection with 1.5 3 106 cells per mouse (n = 6).
(E) GLDC protein expression of 3T3 cells overexpressing GLDC, PSPH, PSAT1, GCAT, SHMT1, and SHMT2. A549 cells served as a positive control. b-actin was
used as a loading control.
In all panels, error bars represent SEM. See also Figure S4.

due to GLDC or other glycine/serine enzymes can exert a potent

tumorigenic effect.
GLDC Regulates Glycine Metabolism, with Effects on
Glycolysis and Pyrimidines
Given that GLDC promotes tumorigenesis through a metabolism-dependent mechanism, we performed metabolomic anal-

ysis to gain deeper mechanistic insights into the GLDC-driven

metabolism changes that lead to tumorigenesis. We used liquid
chromatography-mass spectrometry (LC-MS) to perform metabolomics profiling of HLF cells and 3T3 cells overexpressing
GLDC, as well as A549 lung adenocarcinoma cells with retroviral
knockdown of GLDC, relative to empty vector controls. We
found that glycine-related metabolites, glycolysis intermediates,

serum response), and CDK1 (E2F target) were tested. Normal growing, unsynchronized cells (Cyc) were used as a control. HSP90 was used as a loading control.
CDK1, cyclin-dependent kinase 1; HSP90, heat shock protein 90.
(D) Expression of endogenous GLDC in MCF10A cells transformed by oncogenic KRASG12D, PIK3CAE545K, MYC, or MYCT58A, by qRT-PCR.
(E) Colony formation assay in adherent conditions by seeding 500 3T3 cells overexpressing either GLDC (3T3-GD) or LIN28B (3T3-28B).
(F) Quantitative analysis of colony formation efficiency under adherent conditions as shown in (E).
(G) Quantitative analysis of soft agar colony formation by 5,000 3T3 cells overexpressing either GLDC (3T3-GD) or LIN28B (3T3-28B). Colonies were stained with
INT on day 28.
(H) Proliferation curve of HLF cells overexpressing GLDC (HLF-GD), LIN28B (HLF-28B), or the empty vector (HLF).
(I) Colony formation assay in adherent conditions seeding 1,000 HLF cells overexpressing GLDC (HLF-GD), LIN28B (HLF-28B), or the empty vector (HLF).
(J) Quantitative analysis of colony formation efficiency under adherent conditions as shown in (I).
(K) Quantitative analysis of soft agar colony formation by 5,000 HLF cells overexpressing either GLDC (HLF-GD), LIN28B (HLF-28B), or the empty vector (HLF).
(L) Tumor formation by CD166
lung tumor cells 3 months after overexpression of GLDC. CD166
tumor cells from xenografts were sorted by FACS and infected
with retrovirus expressing either the empty vector (CD166
) or GLDC (CD166
GD), followed by transplantation into mice 24 hr after infection (n = 12 for each
group).
In all panels, error bars represent SEM. *p < 0.05, **p < 0.01. See also Figure S3.

Cell 148, 259272, January 20, 2012 2012 Elsevier Inc. 265

and many pyrimidines were significantly perturbed by both

GLDC overexpression and knockdown (p < 0.05; Figures
5A5D). For glycine-related metabolites, we found that sarcosine
(N-methylglycine) levels increased significantly upon GLDC
overexpression and dropped significantly upon GLDC knockdown, indicating that GLDC promotes sarcosine synthesis or
accumulation (Figure 5A). Consistent with this observation,
betaine aldehyde in the betaine-sarcosine-glycine pathway for
glycine synthesis also showed the same pattern of changes (Figure 5A). Glycine levels decrease with GLDC overexpression and
increase with GLDC knockdown, in agreement with the fact that
GLDC breaks down glycine irreversibly. In contrast, serine levels
increase with GLDC overexpression and decrease with GLDC
knockdown, suggesting that GLDC promotes serine synthesis
or uptake (Figure 5A).
Surprisingly, GLDC perturbation also led to dramatic changes
in glycolysis and other amino acids (Figures 5B, 5C, and S5A).
Our data suggest that GLDC promotes glycolysis, leading
to the increased synthesis or accumulation of glucose-1phosphate, phosphoenolpyruvate, pyruvate, and lactate (Figures 5B and 5C). In fact many of the upstream glycine/serine
metabolism enzymes that we found upregulated in lung TICs,
such as PSAT1, PSPH, and SHMT1/2, channel glycolytic intermediates into de novo serine and glycine biosynthesis (Figure 2E), suggesting that GLDC is working in a concerted fashion
with these enzymes to promote the glycolysis-serine-glycine
flux. This is supported by our finding that GLDC does not significantly promote glycine uptake (Figures S5B and S5C) but
promotes glycolysis instead (Figures 5B and 5C).
Finally, our metabolomics analysis also revealed that GLDC
promotes the synthesis or accumulation of pyrimidines, including thymidine, deoxyuridine, thymine, uracil, and cytosine
(Figure 5D). The GLDC-catalyzed reaction converts glycine into
CH2-THF (Kume et al., 1991). CH2-THF contains the methylene
group that fuels de novo thymidine synthesis from deoxyuridine
in concert with pyrimidine biosynthesis and hence nucleotide
synthesis during cell proliferation (Tibbetts and Appling, 2010).
Recent studies suggest that early oncogenesis involves aberrant
activation of cell proliferation, which then leads to a crisis of
nucleotide deficiency and replication stress (Bester et al., 2011).
Our observations on pyrimidine synthesis suggest that upregulation of GLDC could promote cellular transformation by overcoming this deficiency to progress onward in early oncogenesis.
To test whether any of the metabolite changes induced by
GLDC can mimic GLDCs effects on cancer cells, we analyzed
whether an increased exogenous supply of specific metabolites
could rescue GLDC retroviral knockdown in A549 cells. We
found that 10 mM of sarcosine could significantly rescue the
proliferation defect upon GLDC knockdown, with little effect on
control A549 cells (Figure 5F), indicating that increased sarcosine-glycine metabolite flux can rescue the effects of reduced
GLDC enzyme. To further test whether the production of
CH2-THF is necessary for GLDCs effects on proliferation, we
tested whether the antifolate drug methotrexate could specifically abrogate GLDC-induced proliferation by reducing the tetrahydrofolate (THF) cofactor needed to produce CH2-THF for
pyrimidine synthesis. Our results show that low doses of methotrexate specifically abrogated GLDC-induced proliferation in 3T3
266 Cell 148, 259272, January 20, 2012 2012 Elsevier Inc.

and HLF cells, with little effect on control 3T3 and HLF cells
(Figure 5E). Furthermore, methotrexate in combination with
GLDC shRNA killed transformed A549 cells much more effectively than either alone (Figure 5E), suggesting that a combination
of antifolates with a GLDC inhibitor could completely shut off
glycine catabolism to treat cancer cells more effectively. Using
these metabolic data, we constructed a model of how aberrant
GLDC expression might reprogram glycolysis and glycine
metabolism fluxes in cancer cells to promote cancer cell proliferation and tumorigenesis (Figure 5G).
Prognostic Significance of Aberrant GLDC Expression
in NSCLC Patients
To assess whether our experimental findings on GLDC are relevant to human lung cancer patients in the clinic, we used tissue
microarray immunohistochemistry to examine the prognostic
significance of GLDC expression, tumor size, tumor grade, and
cancer stage in clinical tumor samples from cohorts of NSCLC
patients (n = 143) (Figures 6A and S6; Table S2). Subdistribution
hazard ratio (SHR) analysis showed that patients with high or
grade 3+ GLDC expression have a 3-fold higher risk of lung
cancer mortality compared to patients with low or grade 0
GLDC expression, even when adjusted for cancer stage (SHR =
3.01, 95% confidence interval [CI]: 1.486.10, p = 0.002) (Figure 6B). Cumulative mortality analysis also showed that high
GLDC expression (grade 3+) is significantly associated with
higher cumulative incidence of mortality across 143 NSCLC
lung cancer patients, even when adjusted for cancer stage
(p = 0.005) (Figure 6C). CD166 expression was not significantly
associated with higher mortality in lung cancer patientswhich
was not unexpected given that only 1 in 5 3 103 CD166+ cells
are tumorigenic (Figures S6AS6C). Indeed, coimmunostaining
of lung tumors revealed that GLDC+ cells mostly form a subset
of CD166+ cells, and that not all CD166+ cells are GLDC+ (Figures
6D and S6D). LIN28B immunohistochemistry was also not significantly correlated with lung cancer patient mortality (Figures
S6AS6C), although western blots revealed that lung TICs
express a second LIN28B isoform that is indiscernible from
immunohistochemistry staining, thus rendering our LIN28B staining results inconclusive (Figure S6E). Our immunohistochemistry
results on clinical tumor samples are consistent with the idea that
lung TICs constitute the bulk of the tumors in late stages of
malignancy and demonstrate that aberrant activation of GLDC
is significantly associated with human mortality in NSCLC
patientsfurther supporting its role as a metabolic oncogene in
human NSCLC.
Aberrant GLDC Expression in Other Cancers
To check whether aberrant GLDC expression is specific to
NSCLC, we examined a variety of other cancers. Surprisingly
GLDC is also aberrantly upregulated in subsets of primary
tumors from other cancers, especially ovarian and germ cell
tumors (Figure 7A; Table S3). Further analysis of 606 human
cancer cell lines showed that 158 (26.1%) cancer cell lines overexpress GLDC, including lines from ovarian, germ cell, cervical,
lung, lymphoma, prostate, bladder, and colon cancer (Figure 7B;
Table S4). To test whether GLDC is also required for growth by
one of these GLDC-overexpressing cell lines, we knocked

B
4.00

4.00

1
2
3
4

1.00
1

0.50

Glycine-related metabolites
S
Sarcosine
i
Betaine aldehyde
Serine
Glycine

Fold change (Log2 scalle)

2.00
Fold change (Log2 scale)

Cells

Cells

0.25
0.13
3T3-GD/Ctrl

0.06

1
2
3

2 00
2.00

1.00
1

0.50

3T3-GD/Ctrl
HLF-GD/Ctrl

HLF-GD/Ctrl

0.03

Glycolysis
Glucose 11-phosphate
phosphate
Phosphoenolpyruvate
Pyruvate

A549-GD-sh/Ctrl

A549-GD-sh/Ctrl

0.25

0 02
0.02

2.5

2.5
3T3

A549-GD-sh

2.0

1.5

1.5

1.0

1.0

0.5

0.5

0.0

0.0

Fold changee (Log2 scale)

Lacta
ate OD

2.0

4.00

A549

3T3-GD

2.00

1
2
3
4
5

1 00
1.00
1

0.50

Pyrimidines
Thymidine
Deoxyuridine
Thymine
y
Cytosine
Uracil

3T3-GD/Ctrl
HLF-GD/Ctrl

24 Hr

A549-GD-sh/Ctrl
0

0.25

24 Hr

No. of adherent colonies

/ 200 cells

120

3T3
Agar colony number

No. of adherent colonies

/ 200 cells

F
60
40
20
0
0 0.1 0.2 0.5 1

5 10 20

0 0.1 0.2 0.5 1

Control

5 10 20

GLDC

100

0 uM Sarcosine
10 uM Sarcosine

80
60
40
20
0

Methotrexate ( M)
20

HLF

15

Glycolysis

10
5
Phosphoserine

0
0 0.1 0.2 0.5 1

10

0 0.1 0.2 0.5 1

Control

Methotrexate

10

GLDC

Serine

Tetrahydrofolate (THF)

Methotrexate ( M)

No. of soft agar colonies

/ 1000 cells

100

Betaine
Aldehyde

A549

Sarcosine

Glycine
GLDC

CH2-THF
C

Pyrimidine
metabolism
t b li

CO2
NH3
NADH

50
0
0

10 20 50
Control

10 20 50

GLDC shRNA1
Methotrexate ( M)

10 20 50

GLDC shRNA2

Figure 5. Metabolomics of Cells upon GLDC Overexpression and Knockdown

(AD) Relative fold change in levels of (A) glycine-related metabolites, (B) glycolysis intermediates, and (D) pyrimidines in 3T3 cells with GLDC overexpression
(3T3-GD/Ctrl), HLF cells with GLDC overexpression (HLF-GD/Ctrl), and A549 cells with GLDC knockdown (A549-GD-sh/Ctrl), as determined by LC-MS
metabolomics. (C) Lactate production by 3T3 cells with GLDC overexpression or A549 cells with GLDC knockdown.
(E) Effects of the antifolate drug methotrexate on colony formation after GLDC overexpression or knockdown. 3T3 and HLF cells overexpressing GLDC were
plated at clonal density and exposed to varying concentrations of methotrexate for 8 days. A549 cells with GLDC knockdown were plated in soft agar at clonal
density and exposed to varying concentrations of methotrexate for 14 days.
(F) Effects of sarcosine on soft agar colony formation after GLDC knockdown in A549 cells. One thousand cells were seeded in soft agar at clonal density and
exposed to 10 mM sarcosine for 14 days.
(G) Model of metabolic flux changes induced by GLDC.
In all panels, error bars represent SEM. See also Figure S5.

Cell 148, 259272, January 20, 2012 2012 Elsevier Inc. 267

GLDC

1+

2+

3+

AdC
C

C
SHR (95% CI)

P-value
0.005

1.28 (0.61 2.71)

0.514

1.19 (0.50 2.84)

0.700

3.01 (1.48 6.10)

0.002

GLDC

3+

0.8
Cumula
ative incidence

GLDC Intensity

1.0

1+
2+
0.6

04
0.4
P = 0.005

0.2

n=143
0
0

10

12

14

Time to lung cancer death (years)

D
CD166

DAPI

GLDC

CD166 GLDC DAPI

CD166

GLDC

DAPI

CD166 GLDC DAPI

Figure 6. GLDC Is a Prognostic Indicator for Mortality in NSCLC Patients

(A) GLDC immunohistochemistry staining in a NSCLC tumor microarray (n = 143). Representative images shown for human primary lung adenocarcinomas (AdC)
immunostained with GLDC. Staining intensity grade is indicated in the upper right corner. The boxed regions in the upper images are shown at higher magnification in the lower images. Scale bar, 100 mm.
(B) Subdistribution hazard ratios for each GLDC staining intensity grade, adjusted for American Joint Committee on Cancer (AJCC) staging. CI, confidence
interval.
(C) Cumulative incidence of lung cancer mortality adjusted for AJCC staging, for patients with each GLDC staining intensity grade.
(D) Coimmunofluorescence staining of CD166 (red) and GLDC (green) on primary lung cancer patient tumors, counterstained with DAPI (blue). Representative
cases with coexpression of high levels of CD166 and high levels of GLDC (left panel) and low levels of CD166 and low levels of GLDC (right panel) are shown.
Higher magnification inset is shown in bottom left corner. Scale bar, 50 mm.
See also Figure S6 and Table S2.

268 Cell 148, 259272, January 20, 2012 2012 Elsevier Inc.

GLDC expression across primary tissue cancers

3
2
1

Colon

Bladder

Esophagus

Prostate

Head & Neck

-3

Lymphoma

Cervix

-2

Germ cell

-1

Brain

0
Ovary

Log2 fold change of G

GLDC in
tumor / normal tisssues

Lung

-4
10000
Normalized GLDC intensity

1000
250
100
10
1

CACO2
C

Ctrl-sh
GD-sh

0.5
00
0.0

Relative LIN28B
expression

Ctrl-sh

1.0
0.5

Ctrl-sh
28B-sh

GD-sh

28B-sh

Ctrl-sh
GD-sh
28B-sh

800,000
Number of cells

Relative GLDC
expression

D
1.0

600,000
400,000
200,000
0
0

0.0

4
Days

Figure 7. GLDC Is Aberrantly Expressed in Other Cancers

(A) Log2-transformed fold changes in GLDC expression of patient tumors versus normal adjacent tissues across different cancers. Data are normalized and
aggregated from 51 GEO datasets containing 84 sets of tumor expression data (tumor versus normal) with 2,020 tumor samples and 671 normal samples. Fold
change cutoff was set at 1.5 by yellow line.
(B) Normalized GLDC expression across various cancer cell lines (n = 606). Yellow line indicates the mean value (250) of GLDC expression intensities between
patient lung tumors and normal lung tissues.
(C) qRT-PCR for GLDC and LIN28B in CACO2 colon cancer cells expressing shRNA against GLDC (GD-sh) or LIN28B (28B-sh).
(D) Proliferation curve of CACO2 cells expressing shRNA against GLDC (GD-sh) or LIN28B gene (28B-sh) described in (C). Cell numbers were measured on days
2, 4, and 6.
(E) Tumor formation in mice upon injecting 2.5 3 104 CACO2 colon cancer cells with GLDC knockdown (CACO2-GD-sh) or LIN28B knockdown (CACO2-28B-sh).
Mice were assessed by week 13 (n = 6).
In all panels, error bars represent SEM. See also Figure S7 and Table S3 and S4.

down GLDC in CACO2 cells. Indeed we found that GLDC knockdown reduced their proliferation and tumorigenic potential upon
transplantation, suggesting that GLDC may act as an oncogene
in other cancer cells as well (Figures 7C7E). To examine the
possibility that GLDC is a housekeeping gene for cell proliferation, we also knocked down GLDC in normal HLFs (Figure S7A).
We found that HLF proliferation was unaffected by retroviral
knockdown of GLDC (Figures S7B and S7C). Furthermore we
observed that GLDC is highly expressed only in a few normal
tissues, including postmitotic liver cells, kidney cells, placenta
cells, and olfactory bulb neurons (Figure S7D). Altogether our
observations in both experimental and clinical settings suggest

that human GLDC is not a housekeeping gene required for cell

proliferation but rather an oncogenic metabolic enzyme aberrantly upregulated in NSCLC and possibly several other human
cancers.
DISCUSSION
TIC State in Lung Cancer
Our work sheds new light on the nature of the TIC state and the
role of metabolic reprogramming in tumorigenesis. In this study,
we isolated a subpopulation of TICs from NSCLC patients using
the marker CD166 and showed that both the oncogenic stem cell
Cell 148, 259272, January 20, 2012 2012 Elsevier Inc. 269

factor LIN28B and the glycine metabolism enzyme GLDC drive

the tumorigenicity of lung cancer TICs.
Our data showed that CD166 enriched for TICs in primary
NSCLC, and that CD166 served as an inert surface marker. In
contrast, our results on CD133 are different from the results reported by Eramo et al. (2008) even though both studies used
the same CD133 antibody. This is most likely due to differences
in the xenotransplantation assays, which tend to underestimate
the true frequency of TICs. We employed a more sensitive
mouse xenotransplantation assay using NOD/SCID Il2rg
/

mice instead of SCID mice, and we directly transplanted primary
tumor cells with Matrigel instead of expanding the tumor cells
in vitro. Previous studies have demonstrated that using a more
sensitive mouse xenotransplantation assay dramatically improves our understanding of TICs (Quintana et al., 2008). Our
present study supports this notion, leading us to CD166 as a
new marker for the lung TIC-containing fraction. In normal physiology, CD166 is expressed predominantly during embryonic
development, including the embryonic upper airway, primitive
cardiac cells, and mesenchymal stem cells (Avril-Delplanque
et al., 2005; Murakami et al., 2007; Hennrick et al., 2007; Sabatini
et al., 2005). Expression of CD166 in the embryonic lung is
consistent with our observation that CD166+ lung TICs express
high levels of embryonic lung transcription factors like PEA3
and the trachealess homolog NPAS1, as well as the oncogenic
stem cell factor LIN28B (Liu et al., 2003; Levesque et al., 2007;
Viswanathan et al., 2009). Interestingly, mouse Lin28 is also
expressed in the embryonic lung during normal development
(Yang and Moss, 2003). These observations suggest that the
TIC state in lung cancer is similar to the embryonic lung progenitor state in many aspects.
GLDC Is a Metabolic Oncogene
Our results demonstrate that multiple components in the glycine/
serine pathway are also oncogenes. In addition to embryonic
lung factors, lung TICs also express high levels of GLDC,
GCAT, SHMT1/2, PSPH, and PSAT1, suggesting that TICs rely
on glycine/serine metabolism for tumorigenesis. Overexpression
of catalytically active GLDC, as well as PSAT1, PSPH, and
SHMT2, could induce cellular transformation in 3T3 cells to
form tumors, whereas retroviral knockdown of GLDC significantly reduces the tumorigenicity of lung cancer cells. We further
observed that GLDC+ cells mostly form a subset of CD166+ cells
in lung tumors.
PSAT1, PSPH, and SHMT1/2 lie upstream of GLDC in
the glycine/serine pathway, diverting glycolytic flux from
3-phosphoglycerate through serine to glycine. GLDC is an
oxidoreductase that catalyzes the irreversible rate-limiting step
of glycine catabolism, by breaking down each glycine molecule
in the glycine cleavage system to produce NADH, CO2, NH3, and
CH2-THF (Kume et al., 1991). CH2-THF fuels the one-carbon/
folate metabolism pool, which in turn supplies methylene groups
for biosynthesis (Tibbetts and Appling, 2010). Consistent with
these facts, we found that GLDC regulates many metabolites
in glycolysis and the glycine/serine pathway, leading to specific
changes in pyrimidine synthesis. Pyrimidine derivatives like
thymidine, in turn, are required for nucleotide synthesis in cell
proliferation. Recent studies suggest that early oncogenesis
270 Cell 148, 259272, January 20, 2012 2012 Elsevier Inc.

involves aberrant activation of cell proliferation, which then leads

to a crisis of nucleotide deficiency and replication stress (Bester
et al., 2011)a crisis that GLDC upregulation could overcome
for continued progression in tumorigenesis. Interestingly, we
found that GLDC also increases the levels of N-methylglycine
or sarcosine, an oncometabolite implicated in prostate cancer
(Sreekumar et al., 2009). Furthermore, we observed that GLDC
promotes glycolysis. Combined with our findings on LIN28,
which has been shown to promote glucose uptake and glycolysis (Zhu et al., 2010, 2011), GLDC might be cooperating with
LIN28 as well as PSAT1, PSPH, and SHMT1/2 to divert the
glycolytic flux to glycine and produce CH2-THF. These observations support the notion that the Warburg effect promotes
biosynthesis for tumorigenesis (Hsu and Sabatini, 2008; Vander
Heiden et al., 2009).
GLDC and Glycine Metabolism Are Relevant to Human
Cancer Patients
From the prognostic perspective, aberrant GLDC expression is
significantly correlated with the survival rates of NSCLC patients.
This is consistent with the model that TIC clones expand to
constitute the bulk of the tumor in advanced stages of malignancy (Boiko et al., 2010). Aberrantly increased GLDC is also
widespread in many other human cancers, including lymphoma,
ovarian, germ cell, cervical, prostate, bladder, and colon cancer,
whereas most normal adult human tissues express very low
levels of GLDC. Our experimental data further suggest that in
cancers that rely on GLDC and glycine metabolism, the highly
toxic antifolate drug methotrexate might be initially effective
because it targets TICs, although our data suggest an even
more effective chemotherapy could be potentially achieved by
combining an antifolate drug with a GLDC inhibitor or by searching for a glycine cleavage complex-specific antifolate drug
much like the search for kinase-specific inhibitors in targeted
cancer therapy.
Our study links a glycine metabolism enzyme to lung cancer
and tumorigenesis. Recently several metabolic enzymes have
been linked to cancer in patients, supporting the status of metabolic reprogramming as a new hallmark of cancer (Hanahan and
Weinberg, 2011). In particular, the pyruvate kinase M2 isoform
PKM2, isocitrate dehydrogenase IDH1/2, and phosphoglycerate
dehydrogenase PHGDH have been implicated in multiple
cancers (Christofk et al., 2008; Parsons et al., 2008; Dang et al.,
2009; Locasale et al., 2011; Possemato et al., 2011). Regardless
of the controversy over the frequency of TICs at different stages
of malignancy, our approach shows that characterizing the
unique molecular basis that defines cancer cells with tumorigenic capacity may nevertheless provide novel drug targets for
advancing cancer therapy.
EXPERIMENTAL PROCEDURES
Tumor Cell Preparation
NSCLC tumors were collected from patients according to protocols approved
by the Ethics Committee of the National University of Singapore. Samples
were washed, dissociated, and incubated in DNase and collagenase/dispase.
After incubation, cell clusters and red blood cells were removed. Then single
cells were resuspended and ready for transplantation. See the Extended
Experimental Procedures for more details.

Flow Cytometry
A list of antibodies used can be found in Table S5. Cells were FACS-sorted
using a FACSAria (BD). Flow cytometry was performed using a LSR II flow
cytometer, and data were analyzed with CELLQuest Pro software (BD).
Animals and Transplantation of Tumor Cells
NOD.Cg-Prkdcscid Il2rgtm1Wjl/SzJ mice (Jackson Laboratories) at 46 weeks
old were subcutaneously transplanted with single-cell suspensions in
serum-free medium and Matrigel (BD) (1:1).
Tumor Sphere Culture
Cells were grown in DMEM/F12 containing ITS (BD Biosciences) and supplemented with 50 ng/ml EGF and 20 ng/ml basic fibroblast growth factor (bFGF)
(Invitrogen), using nontreated cell culture plates (Nunc). Fresh medium was
replenished every 3 days.

(E.H.L.). We gratefully acknowledge assistance from Y.P.R. Yip, S.L. Long, and
J. Xu for collection of lung tissues. We are grateful to the Biopolis Shared Facilities Histopathology Laboratory staff for their support with immunohistochemistry and image analysis. We thank Z.M. Li for assistance with making transformed breast cell lines. We greatly acknowledge V. Gaddemane, C.S. Gan,
and T. Hennessy from Agilent Technologies, Singapore for their support in
acquiring and analyzing the mass spectrometry data for the differential analysis of the metabolites. We thank W.L. Tam and F. McKeon for comments
on the manuscript.
Received: February 22, 2011
Revised: August 11, 2011
Accepted: November 17, 2011
Published online: January 5, 2012
REFERENCES

cDNA Microarray Analysis

Total RNA was extracted by Trizol (Invitrogen) and purified by RNeasy Mini
Kit (QIAGEN). Lung primary tumors (one patient), tumor xenografts (three
patients), tumor spheres (four patients), and normal human adult lung
tissues (three patients) were used. RNA was processed and hybridized to
HumanRef-8 v3.0 Beadarrays (Illumina), and the microarray data were normalized and analyzed as described previously (Chua et al., 2006). A fold-change
cut-off threshold of 1.5 was applied to generate the lung TIC gene signature
after four comparisons: primary tumor CD166+ versus CD166
(P+/P
), xenograft tumor CD166+ versus CD166
(X+/X
), spheres versus xenograft tumor
CD166+ (S/X+), and normal lung CD166+ versus CD166
(N+/N
). After intersecting the differentially expressed genes (DEGs) of P+/P
, X+/X
, and S/X+
and excluding DEGs intersecting with N+/N
, DAVID Bioinformatics
Resources 6.7 was applied for KEGG pathway analysis of the final list of
DEGs (Huang et al., 2009).
Metabolomics
Metabolites were extracted by centrifugation of culture media at 14,000 rpm
for 30 min at 4 C. Metabolomic profiling was performed through UPLC/MS
using a Zorbax Eclipse Plus-C18 column on the Agilent 1200 RRLC and an
Agilent 6530 Accurate Mass QTOF. Mass spectrometry was performed on
an Agilent 6530 Accurate Mass Q-TOF mass spectrometer operating in positive ion mode with 2 GHz extended dynamic range mode. See Extended
Experimental Procedures for more details.
Statistical Analysis
Differences were compared using two-tailed Students t test. p values < 0.05
were considered statistically significant. All analyses were performed with
SPSS 18.0 (SPSS). Lung TIC frequencies were estimated using ELDA software
(Hu and Smyth, 2009). Fishers exact test was used to assess the association
between GLDC, CD166, or LIN28B and clinicopathological parameters. The
effect of GLDC, CD166, or LIN28B expressions on lung cancer mortality was
modeled using competing risks regression and quantified based on the SHR
(Fine and Gray, 1999).
ACCESSION NUMBERS
The GEO accession number for human datasets is GSE33198.

Avril-Delplanque, A., Casal, I., Castillon, N., Hinnrasky, J., Puchelle, E., and
Peault, B. (2005). Aquaporin-3 expression in human fetal airway epithelial
progenitor cells. Stem Cells 23, 9921001.
Bester, A.C., Roniger, M., Oren, Y.S., Im, M.M., Sarni, D., Chaoat, M., Bensimon, A., Zamir, G., Shewach, D.S., and Kerem, B. (2011). Nucleotide deficiency promotes genomic instability in early stages of cancer development.
Cell 145, 435446.
Boiko, A.D., Razorenova, O.V., van de Rijn, M., Swetter, S.M., Johnson, D.L.,
Ly, D.P., Butler, P.D., Yang, G.P., Joshua, B., Kaplan, M.J., et al. (2010).
Human melanoma-initiating cells express neural crest nerve growth factor
receptor CD271. Nature 466, 133137.
Christofk, H.R., Vander Heiden, M.G., Harris, M.H., Ramanathan, A., Gerszten,
R.E., Wei, R., Fleming, M.D., Schreiber, S.L., and Cantley, L.C. (2008). The M2
splice isoform of pyruvate kinase is important for cancer metabolism and
tumour growth. Nature 452, 230233.
Chua, S.W., Vijayakumar, P., Nissom, P.M., Yam, C.Y., Wong, V.V.T., and
Yang, H. (2006). A novel normalization method for effective removal of systematic variation in microarray data. Nucleic Acids Res. 34, e38.
Civenni, G., Walter, A., Kobert, N., Mihic-Probst, D., Zipser, M., Belloni, B.,
Seifert, B., Moch, H., Dummer, R., van den Broek, M., and Sommer, L.
(2011). Human CD271-positive melanoma stem cells associated with metastasis establish tumor heterogeneity and long-term growth. Cancer Res. 71,
30983109.
Curtis, S.J., Sinkevicius, K.W., Li, D.A., Lau, A.N., Roach, R.R., Zamponi, R.,
Woolfenden, A.E., Kirsch, D.G., Wong, K.K., and Kim, C.F. (2010). Primary
tumor genotype is an important determinant in identification of lung cancer
propagating cells. Cell Stem Cell 7, 127133.
Dang, L., White, D.W., Gross, S., Bennett, B.D., Bittinger, M.A., Driggers, E.M.,
Fantin, V.R., Jang, H.G., Jin, S., Keenan, M.C., et al. (2009). Cancer-associated
IDH1 mutations produce 2-hydroxyglutarate. Nature 462, 739744.
Dontu, G., Abdallah, W.M., Foley, J.M., Jackson, K.W., Clarke, M.F., Kawamura, M.J., and Wicha, M.S. (2003). In vitro propagation and transcriptional
profiling of human mammary stem/progenitor cells. Genes Dev. 17, 1253
1270.

SUPPLEMENTAL INFORMATION

Eramo, A., Lotti, F., Sette, G., Pilozzi, E., Biffoni, M., Di Virgilio, A., Conticello,
C., Ruco, L., Peschle, C., and De Maria, R. (2008). Identification and expansion
of the tumorigenic lung cancer stem cell population. Cell Death Differ. 15,
504514.

Supplemental Information includes Extended Experimental Procedures, seven

figures, and six tables and can be found with this article online at doi:10.1016/
j.cell.2011.11.050.

Fine, J.P., and Gray, R.J. (1999). A proportional hazards model for the subdistribution of a competing risk. J. Am. Stat. Assoc. 94, 496509.

ACKNOWLEDGMENTS

Hanahan, D., and Weinberg, R.A. (2011). Hallmarks of cancer: the next generation. Cell 144, 646674.

This work is supported by grants from Biomedical Research Council (BMRC)

and Agency for Science, Technology and Research (A*STAR) (B.L.) and grants
from the Singapore Cancer Syndicate and Singapore Lung Cancer Consortium

Hennrick, K.T., Keeton, A.G., Nanua, S., Kijek, T.G., Goldsmith, A.M., Sajjan,
U.S., Bentley, J.K., Lama, V.N., Moore, B.B., Schumacher, R.E., et al.
(2007). Lung cells from neonates show a mesenchymal stem cell phenotype.
Am. J. Respir. Crit. Care Med. 175, 11581164.

Cell 148, 259272, January 20, 2012 2012 Elsevier Inc. 271

Hsu, P.P., and Sabatini, D.M. (2008). Cancer cell metabolism: Warburg and
beyond. Cell 134, 703707.
Hu, Y.F., and Smyth, G.K. (2009). ELDA: extreme limiting dilution analysis for
comparing depleted and enriched populations in stem cell and other assays.
J. Immunol. Methods 347, 7078.
Huang, W., Sherman, B.T., and Lempicki, R.A. (2009). Systematic and integrative analysis of large gene lists using DAVID bioinformatics resources. Nat.
Protoc. 4, 4457.
Ishizawa, K., Rasheed, Z.A., Karisch, R., Wang, Q.J., Kowalski, J., Susky, E.,
Pereira, K., Karamboulas, C., Moghal, N., Rajeshkumar, N.V., et al. (2010).
Tumor-initiating cells are rare in many human tumors. Cell Stem Cell 7,
279282.
Jemal, A., Bray, F., Center, M.M., Ferlay, J., Ward, E., and Forman, D. (2011).
Global cancer statistics. CA Cancer J. Clin. 61, 6990.
Kume, A., Koyata, H., Sakakibara, T., Ishiguro, Y., Kure, S., and Hiraga, K.
(1991). The glycine cleavage system. Molecular cloning of the chicken and
human glycine decarboxylase cDNAs and some characteristics involved in
the deduced protein structures. J. Biol. Chem. 266, 33233329.
Kure, S., Kato, K., Dinopoulos, A., Gail, C., DeGrauw, T.J., Christodoulou, J.,
Bzduch, V., Kalmanchey, R., Fekete, G., Trojovsky, A., et al. (2006). Comprehensive mutation analysis of GLDC, AMT, and GCSH in nonketotic hyperglycinemia. Hum. Mutat. 27, 343352.
Levesque, B.M., Zhou, S.T., Shan, L., Johnston, P., Kong, Y.P., Degan, S., and
Sunday, M.E. (2007). NPAS1 regulates branching morphogenesis in embryonic lung. Am. J. Respir. Cell Mol. Biol. 36, 427434.
Liu, Y.R., Jiang, H.Y., Crawford, H.C., and Hogan, B.L.M. (2003). Role for ETS
domain transcription factors Pea3/Erm in mouse lung development. Dev. Biol.
261, 1024.
Locasale, J.W., Grassian, A.R., Melman, T., Lyssiotis, C.A., Mattaini, K.R.,
Bass, A.J., Heffron, G., Metallo, C.M., Muranen, T., Sharfi, H., et al. (2011).
Phosphoglycerate dehydrogenase diverts glycolytic flux and contributes to
oncogenesis. Nat. Genet. 43, 869874.
Murakami, Y., Hirata, H., Miyamoto, Y., Nagahashi, A., Sawa, Y., Jakt, M., Asahara, T., and Kawamata, S. (2007). Isolation of cardiac cells from E8.5 yolk sac
by ALCAM (CD166) expression. Mech. Dev. 124, 830839.
Nakai, T., Nakagawa, N., Maoka, N., Masui, R., Kuramitsu, S., and Kamiya, N.
(2005). Structure of P-protein of the glycine cleavage system: implications for
nonketotic hyperglycinemia. EMBO J. 24, 15231536.
Parsons, D.W., Jones, S., Zhang, X.S., Lin, J.C.H., Leary, R.J., Angenendt, P.,
Mankoo, P., Carter, H., Siu, I.M., Gallia, G.L., et al. (2008). An integrated
genomic analysis of human glioblastoma multiforme. Science 321, 1807
1812.
Possemato, R., Marks, K.M., Shaul, Y.D., Pacold, M.E., Kim, D., Birsoy, K.,
Sethumadhavan, S., Woo, H.K., Jang, H.G., Jha, A.K., et al. (2011). Functional
genomics reveal that the serine synthesis pathway is essential in breast
cancer. Nature 476, 346350.
Quintana, E., Shackleton, M., Sabel, M.S., Fullen, D.R., Johnson, T.M., and
Morrison, S.J. (2008). Efficient tumour formation by single human melanoma
cells. Nature 456, 593598.

272 Cell 148, 259272, January 20, 2012 2012 Elsevier Inc.

Quintana, E., Shackleton, M., Foster, H.R., Fullen, D.R., Sabel, M.S., Johnson,
T.M., and Morrison, S.J. (2010). Phenotypic heterogeneity among tumorigenic
melanoma cells from patients that is reversible and not hierarchically organized. Cancer Cell 18, 510523.
Reya, T., Morrison, S.J., Clarke, M.F., and Weissman, I.L. (2001). Stem cells,
cancer, and cancer stem cells. Nature 414, 105111.
Rosen, J.M., and Jordan, C.T. (2009). The increasing complexity of the cancer
stem cell paradigm. Science 324, 16701673.
Sabatini, F., Petecchia, L., Tavian, M., Jodon de Villeroche, V., Rossi, G.A., and
Brouty-Boye, D. (2005). Human bronchial fibroblasts exhibit a mesenchymal
stem cell phenotype and multilineage differentiating potentialities. Lab. Invest.
85, 962971.
Sequist, L.V., Joshi, V.A., Janne, P.A., Muzikansky, A., Fidias, P., Meyerson,
M., Haber, D.A., Kucherlapati, R., Johnson, B.E., and Lynch, T.J. (2007).
Response to treatment and survival of patients with non-small cell lung cancer
undergoing somatic EGFR mutation testing. Oncologist 12, 9098.
Sreekumar, A., Poisson, L.M., Rajendiran, T.M., Khan, A.P., Cao, Q., Yu, J.D.,
Laxman, B., Mehra, R., Lonigro, R.J., Li, Y., et al. (2009). Metabolomic profiles
delineate potential role for sarcosine in prostate cancer progression. Nature
457, 910914.
Tibbetts, A.S., and Appling, D.R. (2010). Compartmentalization of mammalian
folate-mediated one-carbon metabolism. Annu. Rev. Nutr. 30, 5781.
Vander Heiden, M.G., Cantley, L.C., and Thompson, C.B. (2009). Understanding the Warburg effect: the metabolic requirements of cell proliferation.
Science 324, 10291033.
Vander Heiden, M.G., Locasale, J.W., Swanson, K.D., Sharfi, H., Heffron, G.J.,
Amador-Noguez, D., Christofk, H.R., Wagner, G., Rabinowitz, J.D., Asara,
J.M., and Cantley, L.C. (2010). Evidence for an alternative glycolytic pathway
in rapidly proliferating cells. Science 329, 14921499.
Viswanathan, S.R., Powers, J.T., Einhorn, W., Hoshida, Y., Ng, T.L., Toffanin,
S., OSullivan, M., Lu, J., Phillips, L.A., Lockhart, V.L., et al. (2009). Lin28
promotes transformation and is associated with advanced human malignancies. Nat. Genet. 41, 843848.
Vogelstein, B., and Kinzler, K.W. (2004). Cancer genes and the pathways they
control. Nat. Med. 10, 789799.
Warburg, O. (1956). On the origin of cancer cells. Science 123, 309314.
Yang, D.H., and Moss, E.G. (2003). Temporally regulated expression of Lin-28
in diverse tissues of the developing mouse. Gene Expr. Patterns 3, 719726.
Zhu, H., Shah, S., Shyh-Chang, N., Shinoda, G., Einhorn, W.S., Viswanathan,
S.R., Takeuchi, A., Grasemann, C., Rinn, J.L., Lopez, M.F., et al. (2010). Lin28a
transgenic mice manifest size and puberty phenotypes identified in human
genetic association studies. Nat. Genet. 42, 626630.
Zhu, H., Shyh-Chang, N., Segre`, A.V., Shinoda, G., Shah, S.P., Einhorn, W.S.,
Takeuchi, A., Engreitz, J.M., Hagan, J.P., Kharas, M.G., et al; DIAGRAM
Consortium; MAGIC Investigators. (2011). The Lin28/let-7 axis regulates
glucose metabolism. Cell 147, 8194.