ARTICLE
doi:10.1038/nature20132
LKB1 loss links serine metabolism to
DNA methylation and tumorigenesis
Filippos Kottakis1,2,3, Brandon N. Nicolay1,3, Ahlima Roumane1,2,3, Rahul Karnik4,5,6, Hongcang Gu4,5,6, Julia M. Nagle1,2,3,
Myriam Boukhali1,3, Michele C. Hayward7, Yvonne Y. Li8,9, Ting Chen8,9,10, Marc Liesa11,12, Peter S. Hammerman8,9,13,
Kwok Kin Wong8,9,10, D. Neil Hayes7, Orian S. Shirihai11,12, Nicholas J. Dyson1,3, Wilhelm Haas1,3, Alexander Meissner4,5,6 &
Nabeel Bardeesy1,2,3
Intermediary metabolism generates substrates for chromatin modification, enabling the potential coupling of metabolic
and epigenetic states. Here we identify a network linking metabolic and epigenetic alterations that is central to oncogenic
transformation downstream of the liver kinase B1 (LKB1, also known as STK11) tumour suppressor, an integrator of
nutrient availability, metabolism and growth. By developing genetically engineered mouse models and primary pancreatic
epithelial cells, and employing transcriptional, proteomics, and metabolic analyses, we find that oncogenic cooperation
between LKB1 loss and KRAS activation is fuelled by pronounced mTOR-dependent induction of the serine–glycine–
one-carbon pathway coupled to S-adenosylmethionine generation. At the same time, DNA methyltransferases are
upregulated, leading to elevation in DNA methylation with particular enrichment at retrotransposon elements associated
with their transcriptional silencing. Correspondingly, LKB1 deficiency sensitizes cells and tumours to inhibition of serine
biosynthesis and DNA methylation. Thus, we define a hypermetabolic state that incites changes in the epigenetic landscape
to support tumorigenic growth of LKB1-mutant cells, while resulting in potential therapeutic vulnerabilities.
Substrates and inhibitors of chromatin-modifying enzymes are generated in intermediary metabolism, so changes in nutrient availability
and utilization can influence epigenetic regulation1,2. Importantly,
recent studies have indicated that the interplay between metabolism
and epigenetics can serve as a programmed switch in cell states. For
example, mouse embryonic stem cell differentiation is promoted by
succinate-mediated inhibition of histone demethylases (HDMs) and
TET DNA demethylases3, or by decreased S-adenosyl-methionine
(SAM) levels leading to loss of histone H3K4 methylation4. Moreover,
aberrant metabolic activity can produce pathological effects by altering
chromatin regulation. Most notably, mutations in the genes encoding
the isocitrate dehydrogenase (IDH)1 and IDH2 enzymes lead to the
generation of 2-hydroxyglutarate, which inhibits HDMs and TETs and
thereby alters DNA and histone methylation—changes that have been
implicated in overriding cell differentiation and promoting tumorigenesis5.
Whether this paradigm extends more generally to other oncogenic
mutations remains unclear, and this question has implications for
understanding cancer pathogenesis and developing improved treatments. Here, we demonstrate that dynamic exchange between metabolism and chromatin regulation contributes to pancreatic tumorigenesis
driven by mutation of the LKB1 serine–threonine kinase.
LKB1 is mutationally inactivated in a range of sporadic cancers,
including pancreatic carcinomas6–8. Additionally, germline mutations
in LKB1 cause Peutz-Jeghers syndrome, which comprises gastrointestinal polyps and a high incidence of gastrointestinal tract carcinomas (for
example, an approximately 100-fold increase in pancreatic cancer)9,10.
Cancers with LKB1 mutations tend to exhibit aggressive clinical features and different therapeutic sensitivity from cancers without these
mutations11–14. LKB1 directly activates a family of 14 kinases related
to AMP-activated protein kinase (AMPK), many of which are coupled
to nutrient sensing and broadly reprogram cell metabolism15. Thus,
metabolic rewiring is thought to be a driver of tumorigenesis after LKB1
loss. We now identify an LKB1-regulated program that links metabolic
alterations to control of the epigenome and is involved in malignant
growth. Our results provide evidence that coupled metabolic and epigenetic states have a more general role in cancer pathogenesis and suggest
therapeutic strategies that could target these intersecting processes.
Synergy between LKB1 and KRAS mutations
LKB1 inactivation frequently coincides with mutations in the RAS–RAF
pathway in human cancers and these genetic alterations cooperate
to drive tumorigenesis in genetically engineered mouse models
(GEMMs)6,11,14,16. We examined the interactions between oncogenic
KRASG12D and deletion of LKB1 in adult pancreatic ducts using a
tamoxifen-inducible GEMM (Extended Data Fig. 1a). The combined
alterations resulted in pancreatic cancers by 20–25 weeks, whereas the
individual mutations had no pathological effects at this age (Fig. 1a and
Extended Data Fig. 1b). To investigate the mechanisms of tumorigenesis,
we isolated primary pancreatic ductal epithelial cells from mice with
conditional KRASG12D and LKB1 alleles (n = 2 lines per genotype) and
transduced them with adenoviruses expressing Cre and/or Flp recombinase to generate KRASG12D/+, LKB1−/− and KRASG12D/+;LKB1−/−
cells (K, L and KL cells, respectively) as well as wild-type parental lines
(Extended Data Fig. 1c). Only KL cells were tumorigenic following
injection into severe combined immunodeficient (SCID) mice or
growth in soft agar, and tumorigenicity was blocked by restoration of
1
Cancer Center, Massachusetts General Hospital, 185 Cambridge Street, Boston, Massachusetts 02114, USA. 2Center for Regenerative Medicine, Massachusetts General Hospital, 185
Cambridge Street, Boston, Massachusetts 02114, USA. 3Department of Medicine, Harvard Medical School, Boston, Massachusetts 02114, USA. 4Broad Institute of MIT and Harvard, Cambridge,
Massachusetts 02142, USA. 5Harvard Stem Cell Institute, Cambridge, Massachusetts 02138, USA. 6Department of Stem Cell and Regenerative Biology, Harvard University, Cambridge,
Massachusetts 02138, USA. 7UNC, Lineberger Comprehensive Cancer Center, Chapel Hill, North Carolina 27599, USA. 8Department of Medicine, Brigham and Women’s Hospital and Harvard
Medical School, Boston, Massachusetts 02115, USA. 9Department of Medical Oncology, Dana Farber Cancer Institute, Boston, Massachusetts 02215, USA. 10Belfer Institute for Applied Cancer
Science, Dana Farber Cancer Institute, Boston, Massachusetts 02215, USA. 11Evans Center for Interdisciplinary Research, Department of Medicine, Mitochondria ARC, Boston University School of
Medicine, Boston, Massachusetts 02118, USA. 12Department of Medicine, Division of Endocrinology, Diabetes and Hypertension, UCLA David Geffen School of Medicine, Los Angeles, California
90095, USA. 13Cancer Program, Broad Institute of Harvard and MIT, Cambridge, Massachusetts 02142, USA.
3 9 0 | NAT U R E | VO L 5 3 9 | 1 7 NOV E M B E R 2 0 1 6
© 2016 Macmillan Publishers Limited, part of Springer Nature. All rights reserved.
ARTICLE RESEARCH
5
10
15 20
Days
**
200
0
0
2
One-carbon
metabolism
Pyruvate
TCA
NES = –1.72
P < 0.001
FDR = 0.027
400
200
0
0
5,000
i
40
K
KL
0
15,000
10
20
Time (h)
Serine M+3
30
60
KL cells
10,000
Time (s)
***
20
K
KL
***
***
80
0
K cells
600
Pyruvate M+3
100
A
**
800
4 6 8 10
Time (h)
h
Fractional
Catabolic
process
Pentose phosphate pathway
Lactate
25
400
SGOC (RNA)
Hexosamine synthesis
Serine
biosynthesis
2
600
g
Anabolic
processes
Glucose
4
13C-enrichment
0
6
O F
1,000
**
Fractional
50
**
K
KL
13C-enrichment
100
e
800
K
KL
150
d
8
OCR (pmol per min)
40K cells
KL
Rescue
50 μm
f
2NBDG/background
**
Enrichment score (ES)
Tumour volume (mm3)
KRASG12D/+
LKB1–/– (KL)
c
200
Lactate
(nmol per 40K cells)
b
KRASG12D/+ (K)
a
***
20
**
K
KL
10
0
30
0
10
20
Time (h)
30
Figure 1 | LKB1 inactivation synergizes with KRASG12D to potentiate
glycolysis, serine metabolism, and tumorigenesis. a, Representative
pancreas histology of the indicated genotypes of mice at 20–25 weeks
(n = 4 per genotype). b, Subcutaneous tumour growth of KL cells
expressing empty vector (KL) or LKB1 (rescue) (n = 8 per group).
c, d, Ductal cells tested for glucose uptake (c; n = 6, independent
replicates), and lactate secretion (d; n = 3). e, Oxygen consumption rates
in K and KL cells under nutrient-replete conditions (n = 21). f, Fates of
glycolytic intermediates. g, GSEA showing enrichment of serine–glycine–
one-carbon network18 (K cells, n = 3; KL cells, n = 4). NES, normalized
enrichment score. h, i, Isotopomer abundance of U[13C]glucose-derived
M+3 pyruvate (h) or serine (i; n = 3, biological replicates). Data pooled
from three (e) or representative of two (d) experiments. Error bars,
s.e.m. (b), s.d. (c, d, g, h). *P < 0.05, **P < 0.01, ***P < 0.001.
wild-type LKB1 (Fig. 1b and Extended Data Fig. 1d–g). In vitro, KL
cells showed greater proliferation than K cells, and both showed greater
proliferation than wild-type or L cells (Extended Data Fig. 1h–j). Thus,
primary ductal cells provide a tractable in vitro system to study mechanisms of epithelial cell transformation arising from LKB1 inactivation.
Focusing on the metabolic alterations provoked by loss of LKB1, we
found that KL cells exhibited an approximately 30% increase in glucose
uptake compared to K cells, and showed marked elevations in levels of
the GLUT1 transporter and ATP (Fig. 1c and Extended Data Fig. 1k, l).
Lactate levels were elevated in KL cells, whereas oxygen consumption
and citrate levels were reduced (Fig. 1d, e and Extended Data Fig. 1m).
Moreover, KL cells showed heightened sensitivity to acute glucose
deprivation and to inhibition of glycolysis using the glucose analogue
2-deoxyglucose, the pyruvate dehydrogenase kinase inhibitor dichloroacetate, or the lactate dehydrogenase inhibitor galloflavin (Extended
Data Fig. 1n–q). Importantly, neither KRASG12D nor LKB1 inactivation alone promoted significant alterations in glucose metabolism
(Extended Data Fig. 1r–t). Thus, these genetic lesions acted synergistically to potentiate glycolysis while rendering cells highly dependent
on glucose availability.
These data suggested that an increased supply of glycolytic intermediates was available for anabolic processes to support the growth
of KL cells (Fig. 1f). Notably, gene set enrichment analysis (GSEA) of
RNA-sequencing (RNA-seq) and quantitative proteomics17 data indicated that KL cells are enriched for glycolytic enzymes and for networks that connect glycolytic intermediates to one-carbon metabolism,
with serine–glycine–threonine and folate metabolism scoring highly
among the induced pathways (Supplementary Data Table 1). There
was particularly striking enrichment of a 64-gene signature defining
the entire serine–glycine–one carbon (SGOC) network18, indicating
strong coordinate activation of these pathways (Fig. 1g and Extended
Data Fig. 2a). Accordingly, the use of uniformly carbon-13-labelled
glucose (U[13C]glucose) demonstrated that KL cells show augmented
production of glucose-derived pyruvate and lactate and an even more
a
Glucose
4
2
***
+s
K er
KL –s
e
+ r
KL ser
–s
er
0
K
f
10
***
5
***
20
– –
+ –
– +
K
+ – –
– + –
– – +
KL
**
***
10
0
0
shControl +
shPSAT1-1 –
shPSAT1-2 –
shControl +
shPSAT1-1 –
shPSAT1-2 –
Figure 2 | Activation of de novo serine biosynthesis supports growth of
LKB1-deficient cells. a, Serine biosynthesis pathway. Red, upregulated
in KL cells. b, Serine pathway gene expression (n = 8 per genotype).
c, Isotopomer abundance of [15N]glutamine-derived M+1 serine and
glycine (n = 3, biological replicates). d, e, Three-day growth of ductal
cells cultured with or without 0.4 mM serine (d; n = 20) or transduced
with the indicated shRNAs (e; n = 6). f, Six-day proliferation of KL cells
– –
+ –
– +
Vector
Fractional
0
0
KL cells
30
5
+ – –
– + –
– – +
10
20
Time (h)
g
750
500
30
0
14
*
0
10
20
Time (h)
h
shControl
shPSAT1-1
shPSAT1-2
**
250
***
21
Days
***
**
28
PSAT1
PCNA+ cells (%)
6
Relative growth
Relative growth
8
10
0
15
* **
K
KL
10
30
40
30
20
*
**
10
0
sh
C
sh on
sh PS trol
PS AT
AT -1
12
0
e
10
*** ***
2
20
Glycine M+1
15
15N-enrichment
***
***
Fractional
***
Serine M+1
K
***
KL
30
Tumour volume (mm3)
CO2 + NH3
d
4
Relative growth
Glycine
GLDC
K
KL
15N-enrichment
3PS
PSPH
Serine
SHMT1/2
c
6
PS 1
SH PH
M
SH T1
M
T
G 2
LD
C
N5,N10
methyleneTHF
PSAT1
AT
Pyruvate
b
3PHP
Glu
α-KG
Relative expression
PHGDH
PS
3PG
transduced with the indicated shRNAs and expression constructs (n = 3).
g, Subcutaneous growth of tumours from KL cells transduced with the
indicated shRNAs (n = 12 tumours per group). h, Proportion of CK19+
tumours cells that are PCNA+ (shControl n = 4, shPSAT1-1 n = 4,
shPSAT1-2 n = 3, representative tumours). Data pooled from four (b) or
representative of two (e, f) or four (d) experiments. Error bars: s.d. (b–f),
s.e.m. (g, h). *P < 0.05, **P < 0.01, ***P < 0.001.
1 7 NOV E M B E R 2 0 1 6 | VO L 5 3 9 | NAT U R E | 3 9 1
© 2016 Macmillan Publishers Limited, part of Springer Nature. All rights reserved.
RESEARCH ARTICLE
N N mT
HF
N5mTHF
*****
0.5
+ – –
– + –
– – +
KL
DNMT1
K
NADPH
SAM
ATP
Nucleotides
DNA
methylation
DNMT3A
METTL20
TRMT2A
WHSC1
SETDB2
ALKBH8
SETD7
COMT
METTL5
OCLN
***
***
***
100
50
0
shControl + – –
shPSAT1 – + +
SAM (100 μM) – – +
f
K
***
K cells
d
% of normal growth
1.0
e
TCA
150
KL
200
**
***
150
**
100
+ – –
– + +
– – +
50
0
K+3DZA
KL+3DZA
0
100
5mC
2
4
6
Days
8
KL cells
5mC levels (a.u.)
K
0
+ – –
shControl
shPSAT1-1 – + –
shPSAT1-2 – – +
5 10
α-KG
1.5
Relative growth
Relative SAM levels
Serine pathway
shPSAT1-1 shPSAT1-2
c
b
Glycolysis
shPSAT1-2 shPSAT1-1 shControl
a
25 μm
80
60
K
KL
******
40
20
0
shControl + – – + – –
shPSAT1-1 – + – – + –
shPSAT1-2 – – + – – +
Figure 3 | Activation of the SGOC network promotes DNA methylation
in LKB1-deficient cells. a, SGOC network. b, SAM levels in ductal cells
transduced with the indicated shRNAs (n = 4). c, Three-day growth of
KL cells with or without SAM supplementation or PSAT1 knockdown
(n = 16). d, Proliferation of ductal cells treated with 3-deazaadenosine
(3-DZA) (n = 12). e, Heatmap of differentially regulated SAM-using
enzymes (proteomics). f, 5mC in ductal cells (77–177 cells). Data pooled
from two (c, d, f) or representative of two (b) experiments. Error bars:
s.d. (b–d), s.e.m. (f). *P < 0.05, **P < 0.01, ***P < 0.001.
pronounced increase in serine and glycine biosynthesis rates without
changes in total levels of these amino acids (Fig. 1h, i and Extended
Data Fig. 2b–d).
Similar results were obtained after treatment with aminooxyacetate
(AOA), an inhibitor of PSAT1 and other aminotransferases (Extended
Data Fig. 4e, f). Thus, the requirement of KL cells for PSAT1 appears
to be independent of its roles in redox homeostasis or maintenance of
nucleoside pools.
Serine metabolism can also fuel the methionine salvage pathway
(Fig. 3a), which is the main mechanism of production for the methyl
donor S-adenosyl methionine (SAM)21,22. Notably, multiple enzymes
that channel serine metabolism intermediates into the methionine
salvage pathway and that contribute to SAM biosynthesis were upregulated in KL cells (Extended Fig. 4g). Accordingly, PSAT1 knockdown
reduced SAM levels in KL cells but not in K cells (Fig. 3b). Furthermore,
SAM supplementation mitigated the proliferation defects caused by
PSAT1 ablation in KL cells (Fig. 3c). Conversely, suppression of SAM
biosynthesis by inhibition of S-adenosyl-homocysteine-hydrolase
(SAH) with 3-deazaadenosine (3DZA), or of methionine-adenosyltransferase (MAT) with cycloleucine, slowed the growth of KL cells,
whereas K cells remained unaffected (Fig. 3d and Extended Data
Fig. 4h). Thus, activation of de novo serine biosynthesis contributes to
SAM production to support growth of LKB1-deficient cells, possibly
by providing active methyl groups in the form of N5-methyl-THF, or by
promoting ATP generation (for example, via PSAT1-dependent TCA
cycle anaplerosis)22,23.
SAM is the substrate for methylation of lipids, DNA, RNA,
metabolites and proteins. Examination of our RNA-seq and proteomics
data sets for the expression of the 183 annotated SAM-dependent
methyltransferase enzymes revealed significant upregulation of the
DNA methyltransferases DNMT1 and DNMT3A in KL cells, whereas
few of the other enzymes showed any changes (Fig. 3e, Extended Data
Fig. 5a and Supplementary Data Table 2). Quantitative PCR and immunoblot analyses verified the regulation of expression of these enzymes
by LKB1 (Extended Data Fig. 5b, c). Importantly, inhibition of DNA
methylation using 5-aza-2-deoxycytidine (decitabine) or the nonnucleoside DNMT inhibitor RG108 promoted SAM accumulation in KL
cells but not in K cells (Extended Data Fig. 5d), suggesting that DNMTs
are major SAM-utilizing enzymes in the context of LKB1 deficiency.
Accordingly, immunofluorescence and DNA dot blot analysis demonstrated that KL cells had a significant increase in 5-methylcytosine
(5mC) as compared to K cells, an effect that was reversed by LKB1
rescue (Extended Data Fig. 5e–g). By contrast, there were no consistent
changes in global levels of a series of histone methylation marks, in
5-hydroxy-methylcytosine, or in N6-methyladenosine (Extended Data
Fig. 5h–i and data not shown). Notably, PSAT1 knockdown suppressed
Serine pathway dependence of KL cells
Multiple serine pathway enzymes (PSAT1, PSPH, SHMT1 and
SHMT2) were upregulated in KL cells, whereas restoration of LKB1
expression reversed these changes, broadly suppressed the entire SGOC
network, and reduced serine biosynthesis (Fig. 2a, b and Extended
Data Fig. 2e–g). PSAT1 catalyses the transamination of 3-phosphohydroxy-pyruvate (3PHP) to 3-phosphoserine (3PS), with glutamate as
the nitrogen donor and α-ketoglutarate (α-KG) as a secondary product (Fig. 2a). Consistent with elevated PSAT1 activity, [15N]glutamine
labelling revealed a marked increase in nitrogen incorporation into
serine and glycine in KL cells (Fig. 2c). Thus, LKB1 restricts serine
metabolism.
Consistent with this finding, KL cells were unaffected by culturing
in serine-free medium, whereas K cells showed a roughly 40% decrease
in proliferation, as did KL cells in which LKB1 was expressed (Fig. 2d
and Extended Data Fig. 3a). On the other hand, KL cells were specifically sensitive to PSAT1 knockdown, exhibiting reduced proliferation
in normal medium, restored dependency on exogenous serine, and
impaired colony formation in soft agar (Fig. 2e and Extended Data
Fig. 3b–d). Introduction of short hairpin (sh)RNA-resistant human
PSAT1 cDNA rescued these phenotypes (Fig. 2f and Extended Data
Fig. 3e). Moreover, PSAT1 knockdown strongly inhibited subcutaneous
tumour growth by KL cells (Fig. 2g, h and Extended Data Fig. 3f, g).
Notably, PSAT1 knockdown had no effect on the proliferation or tumorigenicity of cell lines from pancreatic cancer GEMMs with wildtype LKB1 (KRASG12D-p53 null, KPC and KRASG12D-CDKN2A null,
KIC)19, supporting the idea that these effects result from LKB1 loss
rather than being secondary to high proliferation rates (Extended Data
Fig. 3b, h–k). Thus, increased serine biosynthesis pathway activity is
required to drive oncogenic transformation in KL cells.
Serine biosynthesis fuels DNA methylation
The serine biosynthesis pathway can fuel a range of anabolic processes, including generation of N5,N10-methylene-tetrahydrofolate
(THF), which supports NADPH production, redox homeostasis, and
nucleotide biosynthesis20. PSAT1 knockdown failed to increase levels
of reactive oxygen species (ROS) in KL cells and, correspondingly,
proliferation was not restored by the ROS scavenger N-acetylcysteine
(NAC) or by nucleoside supplementation (Extended Data Fig. 4a–d).
3 9 2 | NAT U R E | VO L 5 3 9 | 1 7 NOV E M B E R 2 0 1 6
© 2016 Macmillan Publishers Limited, part of Springer Nature. All rights reserved.
ARTICLE RESEARCH
K KL
b
Methylation
Methylation
shControl
shPSAT1-1
shPSAT1-2
1.0
* *
0.5
18
S
0
*
2
*
*** *
1
0
2
1.5
3
KL
N
E1
-
2.0
i
Relative expression
shControl
shPSAT1-1
shPSAT1-2
2.5
-1
0
K
1
Methylation
1
KL
LI
LI
N
2
h
LI
N
E1
0
*
**
Relative methylation
*
Relative expression
*
3
0
K
LI
N
E1
-1
LI
N
E1
-2
0.5
E1
LI -1
N
E1
-2
18
S
-2
-1
N
E1
LI
N
E1
Relative methylation
Methylation
Relative expression
1.0
g
0.438
0
1.5
0
EV
LKB1 WT
0.455
*
***
0.5
EV
LKB1 wt
0.364
1.0
Methylation
2.0
0.5
0.4
K
KL
0.556
f
1.5
0.556
e
0.471
KL
0.5
Hypomethylated
0.188
0.200
0.125
0.154
K
Methylation
1.00
0.75
0.50
0.25
0.00
0.5
Hypermethylated
0.50
0
1.0
Median
Median
0.25
Repeat LINE
(n = 1,783,715)
d
1.0
0.75
LI
Repeat LTR
(n = 1,479,183)
c
LI
Global methylation
1.00
N
E1
-
a
Figure 4 | The LKB1–PSAT1 pathway controls methylation and
expression of retrotransposons. a, Distribution of methylation levels
in ductal cells. Median values are indicated. b, Heatmap of differentially
methylated tiles. c, d, Distribution of methylation levels in LINE1 (c) and
LTR (d) repeats. e, Expression of LINE1 in ductal cells, tested using two
independent primer sets (n = 3). f, g, KL cells transduced with EV or LKB1
assessed for methylation of LINE1 and 18S (f; n = 6, LINE1-1 and 18S;
n = 4, LINE1-2), and LINE1 expression (g; n = 6). h, i, KL cells transduced
with the indicated shRNAs were assessed for LINE1 and 18S methylation
(h; n = 6) and LINE1 expression (i; n = 10). Data pooled from two (f, h) or
representative of two (e, g, i) experiments. Error bars: s.d. *P < 0.05,
**P < 0.01, ***P < 0.001.
DNA methylation in KL cells, and this effect was reversed by shRNAresistant PSAT1 cDNA, while K and KPC controls were unaffected.
(Fig. 3f and Extended Data Fig. 6a–c). Comparison of all four genotypes
using our set of isogenic ductal cells confirmed that serine pathway
activity and DNA methylation were specifically potentiated upon dual
KRASG12D expression and LKB1 loss (Extended Data Fig. 6d–g).
The expression of a kinase-dead (KD) LKB1 mutant confirmed that
the activation of key transcriptional and metabolic circuits in KL cells
was due to loss of LKB1 kinase activity (Extended Data Fig. 7a–e).
Accordingly, KL cells showed reduced activity of AMPK, an important
target of the LKB1 kinase, and consequent activation of mechanistic
target of rapamycin complex I (mTORC1) (Extended Data Fig. 7f, w).
Moreover, in K cells, AMPK suppression using the small molecule
inhibitor dorsomorphin (compound C) or shRNAs targeting AMPKα1
or AMPKα2 increased each of the metabolic and gene expression
parameters and upregulated global DNA methylation levels, mimicking
the effects of LKB1 loss (Extended Data Fig. 7g–q). Additionally,
AMPK silencing blocked the effects of LKB1 re-expression on the
growth and metabolic properties of KL cells (Extended Data Fig. 7r–u).
Finally, KL cells were hypersensitive to mTOR inhibition using Torin 1,
which suppressed the metabolic and transcriptional changes seen in
these cells (Extended Data Fig. 7v–z). Thus, deregulation of AMPK–
mTORC1 signalling contributes to the altered metabolic and epigenetic
network in KL cells.
repeats (LTRs), long interspersed nuclear elements (LINEs) and short
interspersed nuclear elements (SINEs)) compared to non-repeat elements (promoters, CpG islands and shores) (Fig. 4c, d and Extended
Data Fig. 8d–k). These elements comprise a large proportion of the
genome, have promoters that can be actively transcribed when unmethylated, and can influence regulation of host genes25. Consistent with
a functional role for the observed retrotransposon methylation, KL
cells had a > 50% reduction in LINE-1 expression compared to K cells
(Fig. 4e). Moreover, restoration of LKB1 suppressed LINE-1 methylation, as determined by 5mC–DNA immunoprecipitation–PCR analysis
(meDIP–qPCR), and led to an approximately twofold upregulation of
LINE-1 transcript expression (Fig. 4f, g). Importantly, PSAT1 knockdown also reduced LINE-1 methylation and induced its expression
in KL cells (Fig. 4h, i), indicating that increased serine metabolism
supports retrotransposon methylation and transcriptional silencing
downstream of LKB1 loss.
Because retrotransposons can function as important modulators
of host gene expression25, we explored the correlation between differentially expressed genes in K versus KL cells and the presence of
linked retrotransposons. Retrotransposon elements were significantly
enriched (χ2 test, P < 0.0001) in the gene bodies of differentially
expressed genes as compared to the distribution of these elements
across all genes (Extended Data Fig. 8l, m). By contrast, minimal
enrichment was observed when promoter regions were considered.
Such specificity is notable in light of evidence that intragenic methylation (including repeat methylation) can modify the activity of linked
promoters and affect RNA processing25,26. Thus, the LKB1–SGOC
pathway alters the epigenetic landscape and dynamically regulates
retrotransposon methylation and transcriptional activity, changes that
appear to be associated with differences in host gene transcription.
LKB1 loss silences retrotransposons
DNA methylation can contribute to transcriptional regulation and
maintenance of genomic stability24. We mapped methylation changes
by conducting whole-genome bisulphite sequencing (WGBS) of K
and KL cells (two independently derived lines per genotype). The data
demonstrated a marked increase in mean CpG methylation in KL
cells (Fig. 4a and Extended Data Fig. 8a). Comparison of methylation
levels at non-repetitive 100-bp tiles revealed 3,395 hypermethylated
and 1,270 hypomethylated regions in KL cells (FDR < 0.05, methylation difference >0.1; Fig. 4b and Extended Data Fig. 8b). However,
the genes associated with differentially methylated regions had only
modest overlap with those showing differential expression, suggesting
that methylation of these elements may not prominently affect transcriptional regulation (Extended Data Fig. 8c). Thus, we focused on
the repetitive portion of the genome. Notably, methylation in KL cells
was particularly enriched at retrotransposon repeats (long terminal
KL cells are sensitive to DNMT inhibition
Consistent with the functional relevance of increased DNA methylation, KL cells were sensitive to shRNA-mediated knockdown of
DNMT1 or DNMT3A, whereas K or KPC cells remained largely unaffected (Extended Data Fig. 9a–c). Similar specific sensitivity of KL
cells was observed in vivo using doxcycline (dox)-inducible shRNAs
to acutely deplete DNMT1 or DNMT3 (or shGFP control) once subcutaneous tumours reached 50 mm3 in size (Fig. 5a–c and Extended Data
Fig. 9d–f). These data indicate that inhibition of DNA methylation is a
potential vulnerability of KL cells. Indeed, KL cells were hypersensitive
1 7 NOV E M B E R 2 0 1 6 | VO L 5 3 9 | NAT U R E | 3 9 3
© 2016 Macmillan Publishers Limited, part of Springer Nature. All rights reserved.
RESEARCH ARTICLE
Tumour volume (mm3)
Tumour volume (mm3)
–Dox
+Dox
1,000
Dox
500
0
0
10
20
Days
c
KL shDNMT1-2
1,000
–Dox
+Dox
*
750
500
Dox
250
0
30
0
10
20
Days
–Dox
+Dox
Dox
0
0
100
1,500
1,000
500
0
Tumour volume (mm3)
2,000
8988T
YAPC
8902
PANC1
SW1990
MIAPACA2
KP3L
PL45
SNU324
COLO357
Decitabine IC50 (nM)
i
2,500
0
1,500
0
–4
–2
0
log10(Conc)
2
Decitabine
0.5
**
0
25
h
20
50 μm
PCNA CK19 DAPI
3+
15 20
Days
1.0
50
C
10
1.5
40
A+
5
60
N
0
Vehicle
** 2.0
100
Ratio
200
Vehicle
Decitabine
30
150
C
300
10
20
Days
g
80
*
PC
Control
Decitabine
Percentage
Tumour volume (mm3)
400
*
250
30
Decitabine IC50
K (IC50 = 138 nM)
KL (IC50 = 2 nM)
KPC (IC50 = 258 nM)
200
500
f
e
d
KL shDNMT3A-2
750
Relative growth
b
KL shControl
1,500
Tumour volume (mm3)
a
Vehicle
Decitabine
***
1,000
CC3 CK19 DAPI
500
0
0
5 10 15 20 25
Days
100 μm
Figure 5 | LKB1 deficiency confers hypersensitivity to DNA
methylation inhibitors. a–c, Volume of subcutaneous tumours derived
from KL cells transduced with Dox-inducible shRNAs (n = 6 tumours
per condition). d, Decitabine IC50 for K, KL or KPC cells (n = 4).
e–g, Decitabine treatment of KL xenografts (n = 20 tumours per condition).
e, Tumour growth. f, Proportion of CK19+ tumour cells that are PCNA+
(n = 4 representative tumours), and ratio of cleaved-caspase 3 (CC3) to
DAPI staining (n = 4 representative tumours). g, Haematoxylin and eosin
staining of representative tumours (top; arrows: tumour glands, dotted
line: hyalinization), and staining for the indicated markers (quantified in f).
Insets: threefold magnification. h, Decitabine IC50 in human pancreatic
cancer cell lines (red, LKB1 mutant; black, LKB1 wild type). i, Growth
of COLO357 xenografts treated with decitabine (1 mg kg−1) or vehicle
(n = 6 per group). Data pooled from three (h) or representative of two (d)
experiments. Error bars: s.e.m. (a–c, i), s.d. (d–f). *P < 0.05, **P < 0.01,
***P < 0.001.
to the clinically approved DNMT inhibitor decitabine, as compared
to K and KPC cell lines (IC50 KL: 2 nM; K: 138 nM; KPC: 258 nM)
(Fig. 5d and Extended Data Fig. 9g–i). Similar results were obtained
using additional DNMT inhibitors (RG108, EGCG and SGI1027;
Extended Data Fig. 9j–m). Importantly, decitabine treatment caused
striking regression of subcutaneous KL tumours, associated with
necrosis, hyalinization, replacement of tumour glands with fibrosis,
decreased tumour cell proliferation, and pronounced apoptosis
(Fig. 5e–g). By contrast, KPC xenografts were unaffected by decitabine
(Extended Data Fig. 9n–q). Thus, LKB1 deficiency confers specific
hypersensitivity to inhibition of DNA methylation in vitro and in vivo.
Available human LKB1 mutant pancreatic cancer cell lines
(COLO357 and SNU324) showed similar vulnerabilities. They exhibited 70–90% inhibition of proliferation following PSAT1 knockdown,
whereas a set of LKB1 wild-type pancreatic cancer lines showed
modest responses (Extended Data Fig. 10a), consistent with prior
results27 in LKB1 wild-type lines. PSAT1 inactivation also inhibited
the growth of LKB1 mutant COLO357 xenografts but not LKB1
wild-type MIAPACA2 xenografts (Extended Data Fig. 10b). Moreover,
LKB1 restoration or PSAT1 silencing decreased global methylation in
COLO357 cells but not in PANC1 or PATU-8988T cells (Extended Data
Fig. 10c, d). LKB1 mutant lines also exhibited an increased response
to decitabine in vitro and in vivo, and to the methionine salvage pathway inhibitors 3DZA and cycloleucine (Fig. 5h, i and Extended Data
Fig. 10e). Thus, both mouse and human LKB1 mutant pancreatic
tumour cells are sensitized to inhibition of the serine pathway and to
DNMT inhibition.
intermediates are channelled towards the SGOC network leading to
increased DNA methylation (Extended Data Fig. 10f). Such interplay
between metabolic control and epigenetic reprogramming has been
proposed as a key mechanism for cancer development in the context
of mutations in metabolic enzymes1,2,5. The present work provides
evidence for a broader role of metabolic and epigenetic crosstalk in
cancer pathogenesis, revealing that LKB1 mutant pancreatic cancer
cells have a marked dependency on pathways linking glycolysis, serine
metabolism and DNA methylation.
It appears probable that the influence of aberrant metabolism on epigenetics contributes to cancer in the context of other oncogenic mutations.
For instance, PI3K/AKT signalling controls acetyl- coenzyme A
metabolism to support histone acetylation and regulate growthpromoting genes28. Numerous oncogenic pathways rewire metabolism,
and the circuits that are activated vary widely, with different genetic
lesions favouring distinct fates of glucose, glutamine, fatty acids, and
other nutrients. Thus, the resulting alterations in the levels of metabolites
affecting chromatin regulation (for example, NAD+/NADH, FAD,
O-linked N-acetylglucosamine, free fatty acids, SAM and acetyl-CoA)
may be fundamental to cancer-promotion. Varying conditions in the
tumour microenvironment may alter tumour phenotypes via similar
processes29,30.
The gain in DNA methylation at retrotransposons that we observed
in KL cells is notable in light of the emerging view that these abundant
repeat elements serve critical roles in host gene regulation25. Moreover,
there is increasing evidence that reactivation of silenced retrotransposons
may underlie the therapeutic benefit of DNMT inhibitors, serving to
induce a type I interferon anti-viral defence program31,32. The basis for
the widespread differences in response to DNMT inhibitors observed
in vitro and in the clinic remains unclear. In this regard, our studies
in KRAS mutant pancreatic ductal cells suggest that LKB1 status is a
Discussion
In summary, we have defined a metabolic state central to tumorigenesis
resulting from LKB1 loss in which glucose and glutamine-derived
3 9 4 | NAT U R E | VO L 5 3 9 | 1 7 NOV E M B E R 2 0 1 6
© 2016 Macmillan Publishers Limited, part of Springer Nature. All rights reserved.
ARTICLE RESEARCH
genetic marker for DNMT inhibitor responsiveness. Because repeat
element methylation does not strictly correlate with sensitivity to these
treatments33, the precise mechanisms of the specific response of KL cells
to DNMT inhibitors remain to be defined. In this setting, it will be of
interest to determine the relative contributions of IFN pathway activation and of modulation of host gene expression. Together, our data point
to novel therapeutic vulnerabilities in the context of LKB1 mutations,
suggesting that it might be possible to use agents targeting nodes of this
network in defined patient subsets, although additional studies will be
needed to broadly establish these associations in human cancers.
Our findings and other recent reports highlight the importance
in cancer of the deregulation of serine biosynthesis, which can result
from amplification of genes encoding pathway enzymes23,34 or their
transcriptional activation downstream of oncogenic drivers, including
NRF235, c-MYC36–38, and mTOR39. Serine metabolism can fuel
diverse biosynthetic pathways and contribute to energy production.
Correspondingly, its functional role in aberrant growth appears context
dependent, relating variously to supporting nucleotide biosynthesis,
NADPH production, TCA cycle anaplerosis, and DNA or histone
methylation. These findings underscore the emerging interest in developing new approaches to targeting components of the SGOC network
as cancer therapeutics.
Online Content Methods, along with any additional Extended Data display items and
Source Data, are available in the online version of the paper; references unique to
these sections appear only in the online paper.
Received 27 May 2015; accepted 27 September 2016.
Published online 31 October 2016.
1.
2.
3.
4.
5.
6.
7.
8.
9.
10.
11.
12.
13.
14.
15.
16.
17.
18.
Gut, P. & Verdin, E. The nexus of chromatin regulation and intermediary
metabolism. Nature 502, 489–498 (2013).
Etchegaray, J.-P. & Mostoslavsky, R. Interplay between metabolism and
epigenetics: a nuclear adaptation to environmental changes. Mol. Cell 62,
695–711 (2016).
Carey, B. W., Finley, L. W. S., Cross, J. R., Allis, C. D. & Thompson, C. B.
Intracellular α-ketoglutarate maintains the pluripotency of embryonic stem
cells. Nature 518, 413–416 (2015).
Shyh-Chang, N. et al. Inluence of threonine metabolism on
S-adenosylmethionine and histone methylation. Science 339, 222–226
(2013).
Losman, J.-A. & Kaelin, W. G. Jr. What a diference a hydroxyl makes: mutant
IDH, (R)-2-hydroxyglutarate, and cancer. Genes Dev. 27, 836–852 (2013).
Waddell, N. et al. Whole genomes redeine the mutational landscape of
pancreatic cancer. Nature 518, 495–501 (2015).
Witkiewicz, A. K. et al. Whole-exome sequencing of pancreatic cancer deines
genetic diversity and therapeutic targets. Nat. Commun. 6, 6744 (2015).
Su, G. H. et al. Germline and somatic mutations of the STK11/LKB1
Peutz-Jeghers gene in pancreatic and biliary cancers. Am. J. Pathol. 154,
1835–1840 (1999).
Giardiello, F. M. et al. Very high risk of cancer in familial Peutz-Jeghers
syndrome. Gastroenterology 119, 1447–1453 (2000).
Korsse, S. E. et al. Pancreatic cancer risk in Peutz-Jeghers syndrome patients:
a large cohort study and implications for surveillance. J. Med. Genet. 50, 59–64
(2013).
Ji, H. et al. LKB1 modulates lung cancer diferentiation and metastasis. Nature
448, 807–810 (2007).
Chen, Z. et al. A murine lung cancer co-clinical trial identiies genetic modiiers
of therapeutic response. Nature 483, 613–617 (2012).
Wingo, S. N. et al. Somatic LKB1 mutations promote cervical cancer
progression. PLoS One 4, e5137 (2009).
Liu, W. et al. LKB1/STK11 inactivation leads to expansion of a prometastatic
tumor subpopulation in melanoma. Cancer Cell 21, 751–764 (2012).
Shackelford, D. B. & Shaw, R. J. The LKB1-AMPK pathway: metabolism and
growth control in tumour suppression. Nat. Rev. Cancer 9, 563–575 (2009).
Cancer Genome Atlas Research Network. Comprehensive molecular proiling
of lung adenocarcinoma. Nature 511, 543–550 (2014).
Ting, L., Rad, R., Gygi, S. P. & Haas, W. MS3 eliminates ratio distortion in
isobaric multiplexed quantitative proteomics. Nat. Methods 8, 937–940
(2011).
Mehrmohamadi, M., Liu, X., Shestov, A. A. & Locasale, J. W. Characterization of
the usage of the serine metabolic network in human cancer. Cell Reports 9,
1507–1519 (2014).
19. Bardeesy, N. et al. Both p16(Ink4a) and the p19(Arf)-p53 pathway constrain
progression of pancreatic adenocarcinoma in the mouse. Proc. Natl Acad. Sci.
USA 103, 5947–5952 (2006).
20. Locasale, J. W. Serine, glycine and one-carbon units: cancer metabolism in full
circle. Nat. Rev. Cancer 13, 572–583 (2013).
21. Mentch, S. J. et al. Histone methylation dynamics and gene regulation occur
through the sensing of one-carbon metabolism. Cell Metab. 22, 861–873
(2015).
22. Maddocks, O. D. K., Labuschagne, C. F., Adams, P. D. & Vousden, K. H. Serine
metabolism supports the methionine cycle and DNA/RNA methylation
through de novo ATP synthesis in cancer cells. Mol. Cell 61, 210–221 (2016).
23. Possemato, R. et al. Functional genomics reveal that the serine synthesis
pathway is essential in breast cancer. Nature 476, 346–350 (2011).
24. Schübeler, D. Function and information content of DNA methylation. Nature
517, 321–326 (2015).
25. Elbarbary, R. A., Lucas, B. A. & Maquat, L. E. Retrotransposons as regulators of
gene expression. Science http://dx.doi.org/10.1126/science.aac7247 (2016).
26. Kulis, M., Queirós, A. C., Beekman, R. & Martín-Subero, J. I. Intragenic DNA
methylation in transcriptional regulation, normal diferentiation and cancer.
Biochim. Biophys. Acta 1829, 1161–1174 (2013).
27. Son, J. et al. Glutamine supports pancreatic cancer growth through a
KRAS-regulated metabolic pathway. Nature 496, 101–105 (2013).
28. Lee, J. V. et al. Akt-dependent metabolic reprogramming regulates tumor cell
histone acetylation. Cell Metab. 20, 306–319 (2014).
29. Shim, E.-H. et al. L-2-Hydroxyglutarate: an epigenetic modiier and putative
oncometabolite in renal cancer. Cancer Discov. 4, 1290–1298 (2014).
30. Intlekofer, A. M. et al. Hypoxia induces production of L-2-hydroxyglutarate.
Cell Metab. 22, 304–311 (2015).
31. Roulois, D. et al. DNA-demethylating agents target colorectal cancer cells by
inducing viral mimicry by endogenous transcripts. Cell 162, 961–973 (2015).
32. Chiappinelli, K. B. et al. Inhibiting DNA methylation causes an interferon
response in cancer via dsRNA including endogenous retroviruses. Cell 162,
974–986 (2015).
33. Treppendahl, M. B., Kristensen, L. S. & Grønbæk, K. Predicting response to
epigenetic therapy. J. Clin. Invest. 124, 47–55 (2014).
34. Locasale, J. W. et al. Phosphoglycerate dehydrogenase diverts glycolytic lux
and contributes to oncogenesis. Nat. Genet. 43, 869–874 (2011).
35. DeNicola, G. M. et al. NRF2 regulates serine biosynthesis in non-small cell lung
cancer. Nat. Genet. 47, 1475–1481 (2015).
36. Sun, L. et al. cMyc-mediated activation of serine biosynthesis pathway is
critical for cancer progression under nutrient deprivation conditions. Cell Res.
25, 429–444 (2015).
37. Ye, J. et al. Serine catabolism regulates mitochondrial redox control during
hypoxia. Cancer Discov. 4, 1406–1417 (2014).
38. Zhang, W. C. et al. Glycine decarboxylase activity drives non-small cell lung
cancer tumor-initiating cells and tumorigenesis. Cell 148, 259–272 (2012).
39. Ben-Sahra, I., Howell, J. J., Asara, J. M. & Manning, B. D. Stimulation of de novo
pyrimidine synthesis by growth signaling through mTOR and S6K1. Science
339, 1323–1328 (2013).
Supplementary Information is available in the online version of the paper.
Acknowledgements We thank A. Kimmelman, K. Patra, L. J. Etchegaray, and
R. Mostoslavsky for comments on the manuscript, and P. Foltopoulou,
B. Martinez and Bardeesy laboratory members for advice. N.B. holds the Gallagher
Endowed Chair in Gastrointestinal Cancer Research and received support
from the Granara-Skerry Trust, the Linda J. Verville Foundation and the Begg
Family, and grants from the NIH (P01 CA117969-07, R01 CA133557-05). F.K. is
supported by a Hirshberg Foundation Career Development Award. F.K. and N.B.
were supported by NIH grant P50CA1270003 and are members of the Andrew
Warshaw Institute.
Author Contributions F.K. and N.B. conceived and designed the study. F.K., A.R.
and J.M.N. performed cell-based and mouse experiments. T.C. assisted with
mouse experiments. F.K. and B.N.N. performed and interpreted the tracing
experiments. M.B and W.H. performed proteomics. F.K. and M.L. performed the
OCR measurements. M.C.H. and D.N.H. provided essential samples and data
analysis. Y.Y.L. performed computational analysis. H.G. prepared WGBS libraries.
R.K. and A.M. analysed and interpreted the WGBS data. P.S.H., K.K.W., O.S.S. and
N.J.D. assisted with data interpretation. F.K. and N.B. wrote the manuscript with
feedback from all authors.
Author Information Reprints and permissions information is available at
www.nature.com/reprints. The authors declare no competing financial
interests. Readers are welcome to comment on the online version of the
paper. Correspondence and requests for materials should be addressed to
N.B. (bardeesy.nabeel@mgh.harvard.edu).
Reviewer Information Nature thanks M. Rehli and the other anonymous
reviewer(s) for their contribution to the peer review of this work.
1 7 NOV E M B E R 2 0 1 6 | VO L 5 3 9 | NAT U R E | 3 9 5
© 2016 Macmillan Publishers Limited, part of Springer Nature. All rights reserved.
RESEARCH ARTICLE
Extended Data Figure 1 | LKB1 suppresses KRASG12D-driven
tumorigenesis and limits glycolysis in primary pancreatic ductal
epithelial cells. a, Schematic of GEM models. Sox9-CreER, LKB1L/L,
and LSL-KRASG12D/+ mice were crossed to generate four cohorts: WT
(Sox9-CreER), L (Sox9-CreER;LKB1L/L), K (Sox9-CreER;LSL-KRASG12D/+)
and KL (Sox9-CreER;LSL-KRASG12D/+;LKB1L/L). Genetic lesions were
induced by intraperitoneal injections of tamoxifen at 6 weeks of age,
after which mice were observed for signs of disease and killed when
KL animals were moribund (20–25 weeks of age). The WT, K, and L
mice had no signs of illness or other abnormalities at this time point.
b, Haematoxylin and eosin-stained sections of representative pancreata
from WT and L mice (n = 4 mice per group). Scale bars, 50 µm.
c, Schematic of primary pancreatic ductal epithelial cell system.
Pancreatic ductal epithelial cells were isolated from LSL-KRASG12D/+
and LSL-KRASG12D/+;LKB1L/L mice and infected with Adeno-Cre
to generate K and KL cells. For studies comparing K, L, KL and WT
genotypes, cells were isolated from FSF-KRASG12D/+;LKB1L/L mice and
infected with Adeno-Flipase and/or Adeno-Cre, or neither. d, Volume
and e, weight of subcutaneous tumours derived from ductal cells of the
indicated genotypes (n = 4 tumours per group). Error bars show s.e.m.
For source data on tumour volume, see Supplementary Data Table 4.
f, Weight of subcutaneous tumours from K (n = 6), KL (n = 8) and KL
cells transduced with retrovirus expressing LKB1 cDNA (rescue, n = 8).
g, Number of colonies formed in soft agar by K (none detected) or KL
cells (n = 6 independent biological replicates). Error bars show s.e.m.
h, Proliferation of K and KL cells in nutrient-replete medium (n = 4).
i, Proliferation of KL cells transduced with retroviruses expressing
empty vector or LKB1 (rescue; n = 3). j, Proliferation of wild-type (WT),
KRASG12D/+ (K), LKB1−/− (L) and KRASG12D/+;LKB1−/− (KL) cells
(n = 6). k–t, In vitro studies of K and KL cells. k, Detection of GLUT1
(SLC2A1) by immunofluorescence (scale bar, 20 µm) or immunoblot.
2-(4-amidinophenyl)-1H-indole-6-carboxamidine (DAPI) was used to
visualize nuclei. Actin was used as the loading control. For gel source
images see Supplementary Data Fig. 1. l, Steady-state ATP levels under
nutrient-replete conditions measured by CellTiterGlo (Promega),
normalized to cell number and expressed as relative to ATP levels in
K cells (n = 4). m, Intracellular levels of pyruvate, lactate, and TCA cycle
metabolites as detected by GC–MS. Values are normalized to cell number.
Data are expressed as relative to the levels in K cells (n = 6 biological
replicates). n, Three-day proliferation of cells in 25 mM glucose or acutely
switched to media with the indicated reduced glucose concentrations
(n = 6). Data are expressed as relative to day 0. o–q, Proliferation of cells
treated with 5 mM 2-deoxyglucose (2DG) (o), 5 mM dichloroacetate
(DCA) (p) or 20 µM galloflavin (gallo) (q). Values are expressed as
percentage of normal growth (2DG n = 6, DCA n = 8, gallo n = 6).
r–t, WT, K, L and KL cells were measured for glucose uptake using
2NBDG (r, n = 6), lactate release into the medium (s, n = 4), and expression
of glycolytic genes (t, n = 6). Data are pooled from two (j, l, n–r) or three (t)
experiments or representative of two (h, i, s) experiments. For all panels,
error bars show s.d. unless otherwise stated; *P < 0.05, **P < 0.01,
***P < 0.001.
© 2016 Macmillan Publishers Limited, part of Springer Nature. All rights reserved.
ARTICLE RESEARCH
Extended Data Figure 2 | LKB1 loss induces the serine–glycine–onecarbon network. a, GSEA showing enrichment of proteins involved in
the serine–glycine–one-carbon network18 in KL cells compared to K cells
using global proteomics (n = 4 samples per group). b, Plot of isotopomer
abundance of U[13C]glucose-derived intracellular M+3 lactate over time
(n = 3 independent biological replicates). c, Plot of isotopomer abundance
of U[13C]glucose-derived intracellular M+2 glycine over time (n = 3
independent biological replicates). d, Intracellular levels of serine, glycine
and glutamine as detected by GC–MS. Values normalized to cell number.
Data are expressed as relative to the levels in K cells (n = 6 biological
replicates). e, PSAT1 and GLDC expression determined by immunoblot
in K and KL cells and in KL cells transduced with LKB1 cDNA. Actin was
used as loading control. For gel source images see Supplementary Data
Fig. 1. f, GSEA of RNA-seq data showing suppression of genes involved in
glycolysis, serine biosynthesis, folate cycle and the serine–glycine–onecarbon network upon re-expression of LKB1 cDNA in KL cells (rescue,
n = 2 samples) compared to parental KL cells (n = 4 samples). g, Plot
of isotopomer abundance of U[13C]glucose-derived intracellular M+3
serine 6 h after addition of U[13C]glucose (n = 3 independent biological
replicates). For all panels, error bars show s.d. unless otherwise stated;
*P < 0.05, **P < 0.01, ***P < 0.001.
© 2016 Macmillan Publishers Limited, part of Springer Nature. All rights reserved.
RESEARCH ARTICLE
Extended Data Figure 3 | See next page for caption.
© 2016 Macmillan Publishers Limited, part of Springer Nature. All rights reserved.
ARTICLE RESEARCH
Extended Data Figure 3 | The de novo serine biosynthesis pathway is
required for KL but not KPC tumour growth. a, Proliferation of KL
cells transduced with vector or LKB1 cDNA and cultured in the presence
or absence of 0.4 mM serine. Growth is expressed as relative to day 0
(n = 3). LKB1 re-expression slows growth and results in sensitivity to
serine deprivation. b, qRT–PCR showing effective knockdown of PSAT1
in K, KL, KPC and KIC cells transduced with shControl or two different
shRNAs against PSAT1 (n = 2). Data are expressed as relative to shControl
for each cell line. 18S rRNA was used for normalization. c, Number of
colonies formed in soft agar by KL cells transduced with shControl or two
different shRNAs against PSAT1 (n = 6). d, Proliferation of K or KL cells
transduced with shControl or two independent shRNAs against PSAT1
in the absence of serine. Data are expressed as percentage of growth in
the presence of 0.4 mM serine (n = 6). e, qRT–PCR showing effective
knockdown of endogenous mouse PSAT1 (mPSAT1, endogenous) and
forced expression of human PSAT1 (hPSAT1, exogenous) in KL cells
transduced with shControl, shPSAT1-1, shPSAT1-2, vector or hPSAT1
cDNA. Data are normalized to 18s rRNA (n = 2). f, Weight at the time
of harvesting of subcutaneous tumours from KL cells transduced
with shControl (n = 8), shPSAT1-1 (n = 12), or shSPAT1-2 (n = 12).
Error bars show s.e.m. g, Haematoxylin and eosin-stained sections
and immunofluorescence analysis of representative tumours derived
from subcutaneous injections of KL cells transduced with shControl,
shPSAT1-1, or shPSAT1-2. Note that PSAT1 knockdown tumours have
a reduction in malignant glands (arrows) relative to the fibrotic stroma.
Lower panels: anti-CK19 (green) was used to visualize the neoplastic
epithelium and anti-PCNA (red) was used to mark proliferation. DAPI
was used to stain nuclei (blue). h, Proliferation of KPC and KIC pancreatic
cancer cells transduced with shControl or two independent shRNAs
against PSAT1. Growth is expressed as relative to day 0 (n = 6 for KPC and
n = 4 for KIC). i–k, KPC cells were transduced with shControl (n = 6),
shPSAT1-1 (n = 6) or shSPAT1-2 (n = 4) and injected into SCID mice.
i, Volume (left) and weight (right) of subcutaneous tumours. Error bars
show s.e.m. For source data on tumour volume, see Supplementary Data
Table 4. j, k, Tumours in i were stained using anti-CK19 antibody (green)
to visualize the neoplastic epithelium and anti-PCNA staining (red)
to mark proliferating cells. DAPI was used to stain nuclei (blue).
The proportion of stained CK19+ cells is quantified in j (n = 4,
representative tumours) and CK19+ cells with nuclear PCNA staining
are quantified in j. There are no significant effects on any of these
parameters. k, Haematoxylin and eosin-stained sections (top) and
immunofluorescence analysis (bottom) of representative tumours. Scale
bars, 100 µm. Insets show threefold magnification. Data pooled from two
(c, d) or representative of two (a, b, e) or three (h) experiments. For all
panels, error bars show s.d. unless otherwise stated; *P < 0.05,
**P < 0.01, ***P < 0.001.
© 2016 Macmillan Publishers Limited, part of Springer Nature. All rights reserved.
RESEARCH ARTICLE
Extended Data Figure 4 | Characterization of serine–glycine–onecarbon pathway in KL cells. a, Detailed graph of SGOC network.
Enzymatic inhibitors used in this study are marked in red. b, ROS in K or
KL cells transduced with shControl, shPSAT1-1 or shPSAT1-2 measured
by DCFDA (left) and CellRox staining (right). Data are normalized to cell
number (n = 4). c, Six-day proliferation assay of KL cells transduced with
shControl, shPSAT1-1 or shPSAT1-2, showing lack of growth rescue by
N-acetylcysteine (NAC). Data are expressed as relative to day 0 (n = 6).
d, Three-day growth assay of KL cells transduced with shControl,
shPSAT1-1 or shPSAT1-2, showing the lack of rescue by excess
nucleosides (adenosine, guanosine, thymidine, uridine, cytidine; 1 mM
each). Data are presented as percentage of the growth of shControl cells
(n = 16). e, Proliferation of K or KL cells treated with aminooxyacetate
(AOA). Data are expressed as relative to day 0 (n = 8). f, Five-day
proliferation of K or KL cells treated with AOA and/or NAC. Data are
expressed as relative to day 0 (n = 6). g, Data from RNA-seq (left) and
quantitative proteomics (right) showing levels of genes involved in
the production of SAM in K and KL. The data plotted are expressed as
mean-centred values. h, Proliferation of K or KL cells treated with 2 mM
cycloleucine. Data are expressed as percentage of the growth of vehicle
treated cells (n = 12). Data are pooled from two (b–f, h) experiments.
For all panels, error bars show s.d. unless otherwise stated; *P < 0.05,
**P < 0.01, ***P < 0.001.
© 2016 Macmillan Publishers Limited, part of Springer Nature. All rights reserved.
ARTICLE RESEARCH
Extended Data Figure 5 | Deletion of LKB1 induces DNMT1 and
DNMT3A expression and increases global DNA methylation. a, Heat
map of RNA-seq data showing levels of the differentially regulated SAMusing enzymes. Plotted data are expressed as mean centred values.
b, Expression of DNMT1 and DNMT3A in K, KL, and KL cells transduced
with LKB1 cDNA (rescue) was measured by qRT–PCR. Levels were
normalized to 18S rRNA. Data are expressed as relative to K cells (n = 4,
representative of two experiments). c, Immunoblots of lysates from K,
KL or rescue cells were probed for DNMT1 or DNMT3A. Actin was
used as loading control. For gel source images see Supplementary Data
Fig. 1. d, Measurement of SAM in K and KL cells treated with 5-aza-2deoxycytidine (decitabine) or RG108 for 3 days. In each case, data are
expressed as relative to the amount of SAM in vehicle-treated K cells,
which was arbitrarily set to 1 (n = 6 independent replicates).
e, Immunofluorescence staining and quantitation of 5mC in K or KL cells
(77–130 cells). Scale bar, 25 µm. f, Dot blot of DNA isolated from K or KL
cells probed with anti-5mC antibody. Quantified signal was normalized
to total DNA as measured by methylene blue staining (n = 4, independent
replicates). g, Dot blot of DNA isolated from KL cells transduced with
empty vector or LKB1 cDNA probed with anti-5mC antibody. Quantified
signal was normalized to total DNA as measured by methylene blue
staining (n = 3, independent replicates). h, Immunoblot analysis of
histone 3 (H3) methyl marks from K or KL cells. Data are normalized
to total H3 (K4me3, n = 2; K27me3, n = 5; K36me3, n = 5, independent
replicates). For gel source images see Supplementary Data Fig. 1. i, Dot
blot of DNA isolated from K or KL cells probed with anti-5hmC antibody.
Quantified signal was normalized to total DNA as measured by methylene
blue staining (n = 4, independent replicates). For all panels, error bars
show s.d. unless otherwise stated; *P < 0.05, **P < 0.01, ***P < 0.001.
© 2016 Macmillan Publishers Limited, part of Springer Nature. All rights reserved.
RESEARCH ARTICLE
Extended Data Figure 6 | Serine pathway activity sustains DNA
methylation in KL cells. a, Dot blot of DNA isolated from K or KL cells
transduced with shControl, shPSAT1-1 or shPSAT1-2 probed with anti5mC antibody. Total DNA was visualized by methylene blue staining.
Graph shows quantified signal normalized to total DNA as measured
by methylene blue staining (K cells, n = 4; KL cells, n = 8, independent
replicates). b, Dot blot of DNA isolated from KPC cells transduced with
shControl, shPSAT1-1 or shPSAT1-2 probed with anti-5mC antibody.
The graph shows quantified signal normalized to total DNA (n = 4,
independent replicates). c, Dot blot of DNA probed with anti-5mC
antibody. DNA was isolated from KL cells first transduced with vector or
human PSAT1 then transduced with shControl, shPSAT1-1 or shPSAT-2.
Graph shows quantified signal normalized to total DNA as measured by
methylene blue staining (n = 3, independent replicates). d, Expression
of serine pathway genes and DNMT genes in WT, K, L or KL cells by
qRT–PCR. Data are normalized to 18S and expressed as relative to K
cells (n = 6, pooled data from two experiments). e, Plots of isotopomer
abundance of U[13C]glucose-derived M+3 serine and M+2 glycine, 6 h
after addition of U[13C]glucose (n = 6 independent biological replicates).
f, Quantified DNA dot blot signal of DNA isolated from WT, K, L or
KL cells probed with anti-5mC antibody normalized to total DNA as
measured by methylene blue staining (n = 4, independent replicates).
g, Five-day growth of WT, K, L or KL cells transduced with shControl or
two shRNAs against PSAT1. Data are expressed as relative to day 0 (n = 4,
pooled from two experiments). For all panels, error bars show s.d. unless
otherwise stated; *P < 0.05, **P < 0.01, ***P < 0.001.
© 2016 Macmillan Publishers Limited, part of Springer Nature. All rights reserved.
ARTICLE RESEARCH
Extended Data Figure 7 | See next page for cation.
© 2016 Macmillan Publishers Limited, part of Springer Nature. All rights reserved.
RESEARCH ARTICLE
Extended Data Figure 7 | LKB1-mediated regulation of glycolysis–
SGOC–DNMT pathway involves the AMKP–mTOR axis. a–e, KL cells
were transduced with vector, wild-type LKB1 or kinase-dead LKB1.
a, Proliferation, expressed as relative to day 0 (n = 4). b, Glucose uptake
measured using 2NBDG followed by fluorimetry (data normalized to cell
number and expressed as relative to KL-vector cells (n = 8)). c, Lactate
levels measured by fluorimetry 3 h after medium change, normalized to
cell number (n = 8). d, Oxygen consumption rates measured in normal
duct medium, followed by injections with 4 µM oligomycin (O), 4 µM
FCCP (F), or 4 µM antimycin A (A) (n = 3). e, Expression of the indicated
genes measured by qRT–PCR (n = 4), with levels normalized to 18S
rRNA. Data are expressed as relative to KL-empty vector (EV) cells.
f, Immunoblot of K, KL or KL cells transduced with LKB1 cDNA (rescue).
For gel source images see Supplementary Data Fig. 1. g, Immunoblot
of K cells treated overnight with 10 µΜ Compound C. For gel source
images see Supplementary Data Fig. 1. h, i, Glucose uptake (h) and 3 h
lactate production (i) in the same cells as in h. Data are normalized to cell
number and expressed as relative to control K cells (glucose uptake, n = 3;
lactate, n = 4). j, Glucose uptake in K or KL cells treated with vehicle or
10 µΜ Compound C. Data are normalized to cell number and expressed as
relative to control K cells (n = 4). Note the blunted response of KL cells.
k, qRT–PCR analysis of the indicated genes in the same cells as in h. Levels
are normalized to 18S rRNA. Data are expressed as relative to control
K cells (n = 6). l, Immunostaining for 5-mC (left) and quantification of
staining (right) in K cells treated with vehicle or Compound C for 4 days
(158–163 cells). m–q, K cells were transduced with shControl or shRNAs
against AMPKa1 or AMPKa2. m, Glucose uptake. Data are normalized to
cell number and expressed as relative to shControl-treated cells (n = 4).
n, Steady-state ATP levels measured with CellTiterGlo (Promega),
normalized to cell number and expressed as relative to ATP levels in
shControl-treated cells (n = 4). o, Lactate levels measured by fluorimetry
3 h after medium change. Data are normalized to cell number (n = 3).
p, Immunostaining for 5-mC (left) and quantification of staining (right)
(159–296 cells). q, Expression of the indicated genes determined by
qRT–PCR. Levels are normalized to 18S rRNA. Data are expressed as
relative to shControl cells (n = 4). r–u, KL cells were transduced with
vector or an LKB1 cDNA and shControl or shRNAs against AMPKa1
or AMPKa2. r, Glucose uptake. Data are normalized to cell number and
expressed as relative to vector-shControl cells (n = 4). s, Steady-state ATP
levels normalized to cell number and expressed as relative to ATP levels in
EV-shControl cells (n = 4). t, Lactate levels 3 h after medium change. Data
are normalized to cell number (n = 4). u, Proliferation expressed as relative
to day 0 (n = 3). v, Impact of Torin 1 treatment on growth of K and KL cells
(n = 4). w, Immunoblot of K and KL cells treated with vehicle or 25 nM
Torin 1. For gel source images see Supplementary Data Fig. 1. x, Glucose
uptake and 3 h lactate production in K or KL cells treated overnight with
25 nM Torin 1. Data are normalized to cell number (n = 4). y, Isotopomer
abundance of U[13C]glucose-derived serine and glycine in the same
cells as in r (n = 3 independent replicates). Cells were labelled with
U[13C]glucose for 6 h. z, Expression of the indicated genes in Torin-treated
K (left) or KL cells (right) determined by qRT–PCR. Levels are normalized
to 18S rRNA. Data are expressed as relative to control-treated cells (n = 4).
Data are pooled from two (b, c, e, i–n, p–t, v, x, z) or representative of two
(a, d, h, o) or three (u) experiments. Error bars show s.e.m. in d, s.d. in all
other panels. *P < 0.05, **P < 0.01, ***P < 0.001.
© 2016 Macmillan Publishers Limited, part of Springer Nature. All rights reserved.
ARTICLE RESEARCH
Extended Data Figure 8 | See next page for caption.
© 2016 Macmillan Publishers Limited, part of Springer Nature. All rights reserved.
RESEARCH ARTICLE
Extended Data Figure 8 | Loss of LKB1 increases DNA methylation
in retrotransposon elements. a, Hierarchical clustering of K and KL
samples based on methylation levels at 100-bp autosomal genomic tiles
as measured by whole-genome bisulfite sequencing. b, Two-dimensional
density plot for methylation at non-repetitive 100-bp autosomal genomic
tiles in KL versus K cells. The yellow line shows the mean methylation of
tiles. The dots represent differentially methylated tiles (FDR q-value
< 0.05, methylation change > 0.1). There are 3,395 hypermethylated tiles
and 1,270 hypomethylated tiles in KL cells versus K cells. c, Overlap of
genes associated with differentially methylated regions and differentially
regulated genes (hyper DMRs, hypermethylated regions; hypo DMRs,
hypomethylated regions; DRGs, differentially regulated genes).
d–j, Distribution of methylation density within the low-CpG-promoters
(LCPs), high-CpG-promoters (HCPs), islands, shores, exons, introns and
LTRs. Numbers reflect median values. k, Average percentile change in
methylation in the same elements as in d–j as well as LINEs and LTRs.
l, Number of genes containing retrotransposon repeat elements (sum
of LINEs, SINEs, LTRs) among the set of differentially regulated genes
(DRGs) in K versus KL cells and among all genes. Note that the DRGs are
enriched for the presence of retrotransposons in their gene bodies (top),
but not in their promoters (middle) or 3′UTRs (bottom). m, Specific
enrichment of LINE, SINE and LTR elements in the gene bodies of DRGs
when compared to all genes in the genome.
© 2016 Macmillan Publishers Limited, part of Springer Nature. All rights reserved.
ARTICLE RESEARCH
Extended Data Figure 9 | See next page for caption.
© 2016 Macmillan Publishers Limited, part of Springer Nature. All rights reserved.
RESEARCH ARTICLE
Extended Data Figure 9 | LKB1 deficiency confers hypersensitivity to
inhibitors of DNA methylation. a–c, Proliferation of K (a), KL (b) and
KPC (c) cells transduced with shControl or two shRNAs against each of
DNMT1 (D1) or DNMT3A (D3A). Data are expressed as relative to day
0 (K and KL n = 6, KPC n = 4). d–f, Volume of subcutaneous tumours
derived from KPC cells transduced with doxcycline (Dox)-inducible
shRNAs against DNMT1 or DNMT3A (n = 4). Doxcycline was introduced
to the drinking water when tumours reached 125 mm3. Error bars show
s.e.m. For source data on tumour volume, see Supplementary Data Table 4.
g, Apoptosis measured by caspase 3/7 activity in K or KL cells treated
with decitabine for 48 h. Values are normalized to cell number (n = 3).
h, i, Proliferation of KL cells transduced with empty vector (h) or LKB1
cDNA (i) treated with decitabine. Data are expressed as relative to day 0
(n = 3). j, Proliferation of KL cells transduced with empty vector or LKB1
cDNA treated with RG108. Data are expressed as percentage of growth
of untreated cells (n = 6). k–m, Proliferation of K or KL cells treated with
RG108 (n = 6) (k), EGCG (n = 12) (l) or SGI1027 (n = 12) (m). Data are
expressed as percentage of growth of untreated cells. n–q, Mice bearing
subcutaneous KPC tumours were treated with decitabine (n = 12) or
vehicle (n = 12) when tumours reached 125 mm3. n, o, Tumour volume (n)
and final tumour weight (o). Error bars show s.e.m. For source data on
tumour volume, see Supplementary Data Table 4. p, Haematoxylin and
eosin-stained slides from representative tumours (top). Bottom, anti-CK19
(green) was used to visualize the neoplastic epithelium and anti-PCNA
(red) was used to mark proliferating cells. DAPI was used to stain nuclei
(blue). q, Quantification of the CK19+ neoplastic epithelial compartment
(%CK19+ cells/total cells; top). Quantification of CK19+ cells with nuclear
PCNA staining (bottom; n = 6). Scale bars, 100 µm. Insets are threefold
magnification. Data pooled from two (a–c, j) or four (k–m) experiments
or representative of two (h) or three (g) experiments. Error bars show s.d.
unless otherwise stated; *P < 0.05, **P < 0.01, ***P < 0.001.
© 2016 Macmillan Publishers Limited, part of Springer Nature. All rights reserved.
ARTICLE RESEARCH
Extended Data Figure 10 | See next page for caption.
© 2016 Macmillan Publishers Limited, part of Springer Nature. All rights reserved.
RESEARCH ARTICLE
Extended Data Figure 10 | Vulnerabilities of human LKB1 mutant
pancreatic cancer cell lines. a, Three-day growth of LKB1 wild-type
(black) or LKB1 mutant (red) human pancreatic cancer cells. Data are
expressed as relative to shControl-transduced cells, which is arbitrarily
set to 100 (n = 3). b, Quantification of tumour volume of subcutaneous
tumours derived from implantation of the indicated cells transduced with
shControl or two shRNAs against PSAT1 (n = 4). Error bars show s.e.m.
For source data on tumour volume, see Supplementary Data Table 4.
c, Immunofluorescence staining and quantification of 5mC in COLO357,
PANC1 and PATU-8988T cells transduced with shControl or two shRNAs
against PSAT1 (607–760 cells for COLO357, 305–342 cells for PANC1 and
623–889 cells for PATU-8988T). Scale bar, 25 µm. Error bars show s.e.m.
Data are expressed as fluorescence per nucleus. d, Detection of 5mC by
immunofluorescence in COLO357 (LKB1-deficient) cells transduced with
vector or wild-type LKB1. Quantification is presented as fluorescence
per nucleus (438–618 cells). Scale bar, 25 µm. Error bars show s.e.m.
e, Three-day growth of LKB1 wild-type (black) or LKB1 mutant (red)
human pancreatic cancer cells treated with 10 µM 3-deazaadenosine
(3DZA) (top) or 2 mM cycloleucine (CycloLeu) (bottom). f, Metabolic and
epigenetic changes promoting transformation upon deletion of the tumour
suppressor LKB1. Data pooled from (c, d) or representative of (a) two
experiments. Error bars show s.d. unless otherwise stated; *P < 0.05,
**P < 0.01, ***P < 0.001.
© 2016 Macmillan Publishers Limited, part of Springer Nature. All rights reserved.