Download MYC activation on AMPK

Supplemental Information
Linglin Yu, Mingyang Lu, Dongya Jia, Jianpeng Ma, Eshel Ben-Jacob, Herbert Levine, Benny
Abraham Kaipparettu, José N. Onuchic
1. Metabolic interplays between glycolysis and OXPHOS
Overall, glycolysis, glucose oxidation and fatty acid oxidation inhibit each other. In detail,
aerobic glycolysis and glucose oxidation compete with each other as the same pyruvate
molecules are used in both pathways. Moreover, both glucose and fatty-acid oxidations produce
acetyl-CoA before they enter the TCA cycle. However, excessive acetyl-CoA prevents pyruvate
from becoming acetyl-CoA, and inhibits glycolysis and fatty-acid oxidation in cytosol. Thus,
glucose and fatty-acid oxidations inhibit each other as well, also known as the Randle cycle (1).
2. Details of the core AMPK:HIF-1:ROS circuit
In the core AMPK:HIF-1:ROS circuit, AMPK and HIF-1 are directly associated with
mitochondrial OXPHOS and glycolysis respectively. In response to various metabolic stresses,
AMPK, once activated through phosphorylation, induces fatty acid and glucose metabolism (2,3).
In particular, AMPK increases glucose uptake in mitochondria by increasing GLUT-4
translocation (4) and induces fatty acid oxidation by phosphorylating and inactivating ACC, as
well as activating PPARA and PGC-1 (5). HIF-1 activates enzymes that are required in the
glycolytic pathway and induces glycolysis (6,7).
AMPK and HIF-1 regulate each other through mutual inhibition because AMPK down-regulates
HIF-1 through inhibiting mTOR complex (8), and HIF-1 directly inhibits the transcription of
AMPK (9). The activation of AMPK also leads to enhanced ATP production through OXPHOS,
while excessive ATP inhibits the phosphorylation of AMPK, thus forming an indirect AMPK
self-inhibition. In addition, HIF-1 promotes glycolysis and the glycolytic products, such as
pyruvate and lactate, stabilize HIF-1, thus forming an indirect HIF self-activation (10-12).
Moreover, AMPK and HIF-1 also highly interact with ROS. HIF-1 promotes noxROS
production through targeting NOX, and inhibits mtROS production by decreasing OXPHOS and
activating glycolysis (13). On the other hand, AMPK promotes mtROS production by OXPHOS
(14,15) and increases mtROS scavenging activity through the AMPK-FOXO pathway that
upregulates thioredoxin (Trx) expression (16-19). In cytosol, AMPK inhibits noxROS production
by regulating NOX (20,21). ROS, including both mtROS and noxROS, stabilizes HIF-1α and
activates AMPK, respectively (22-26).
Abbreviations:
AMPK: 5' AMP-activated protein kinase
pAMPK: phosphorylated AMPK
HIF-1: Hypoxia-inducible factor 1
1
NOX: NADPH oxidase
GLUT1: Glucose transporter 1
HK: Hexokinase
FASN: Fatty Acid Synthase
PPAR𝜶: Peroxisome proliferator-activated receptor alpha
PGC1: Peroxisome proliferator-activated receptor gamma coactivator 1
3. Mathematical model for the core AMPK:HIF-1:ROS circuit
The core circuit was modeled by the following deterministic rate equations:
0
𝑅̇𝑚𝑡 = 𝑔𝑟𝑚𝑡 𝐻 𝑠− (𝐴𝑎𝑟 , 𝐴0𝑎𝑟 , 𝜆𝑎𝑟 , 𝑛𝑎𝑟 )𝐶𝑟𝑐𝑜𝑚𝑝
(𝛶, 𝑔𝑛 , ℎ, ℎℎ𝑟
, 𝑛ℎ𝑚𝑡 , 𝐴, 𝐴0𝑎𝑟𝑚𝑡 , 𝑛ℎ𝑚𝑡 ) − 𝑘𝑟𝑚𝑡 𝑅𝑚𝑡
𝑚𝑡
𝑚𝑡
(eq. S1.1)
0
𝑅̇𝑛𝑜𝑥 = 𝑔𝑟𝑛𝑜𝑥 𝐶𝑅𝑐𝑜𝑚𝑝
(𝑔0 , ℎ, ℎℎ𝑟
, 𝑛ℎ𝑛𝑜𝑥 , 𝑔1 , 𝐴, 𝐴0 , 𝑔2 , 𝑛𝑎𝑛𝑜𝑥 ) − 𝑘𝑟𝑛𝑜𝑥 𝑅𝑛𝑜𝑥
𝑛𝑜𝑥
𝑛𝑜𝑥
(eq. S1.2)
𝑅 = 𝑅𝑚𝑡 + 𝑅𝑛𝑜𝑥
(eq. S1.3)
0
0
𝐴̇ = 𝑔𝑎 𝐻 𝑠+ (𝑅𝑟𝑎 , 𝑅𝑟𝑎
, 𝜆𝑟𝑎 , 𝑛𝑟𝑎 )𝐻 𝑠− (ℎℎ𝑎 , ℎℎ𝑎
, 𝜆ℎ𝑎 , 𝑛ℎ𝑎 )𝐻 𝑠− (𝐴𝑎𝑎 , 𝐴0𝑎𝑎 , 𝜆𝑎𝑎 , 𝑛𝑎𝑎 ) − 𝑘𝑎 𝐴
(eq. S1.4)
0
0
ℎ̇ = 𝑔ℎ 𝐻 𝑠− (𝐴𝑎ℎ , 𝐴0𝑎ℎ , 𝜆𝑎ℎ , 𝑛𝑎ℎ ) − 𝑘ℎ ∙ ℎ ∙ 𝐻 𝑠− (ℎℎℎ , ℎℎℎ
, 𝜆ℎℎ , 𝑛ℎℎ )𝐻 𝑠− (𝑅𝑟ℎ , 𝑅𝑟ℎ
, 𝜆𝑟ℎ , 𝑛𝑟ℎ )
(eq. S1.5)
Here, we followed the computational modeling approach presented by Lu et al. (27). In the
equations, 𝑅𝑚𝑡 , 𝑅𝑛𝑜𝑥 , 𝐴 , ℎ represent the levels of mtROS, noxROS, pAMPK and HIF-1
respectively; 𝑔𝑟𝑚𝑡 , 𝑔𝑟𝑛𝑜𝑥 , 𝑔𝑎 , 𝑔ℎ are the synthesis rate of mtROS, noxROS, pAMPK and HIF-1
respectively; 𝑘𝑟𝑚𝑡 , 𝑘𝑟𝑛𝑜𝑥 , 𝑘𝑎 , 𝑘ℎ represent the degradation rates of mtROS, noxROS, pAMPK
and HIF-1 respectively. We used the shifted Hill-function 𝐻 𝑠+ for excitatory regulations, 𝐻 𝑠−
for inhibitory regulations, and 𝐶 𝑐𝑜𝑚𝑝 for competitive regulations by two genes/metabolites.
In the equations, the shifted Hill-function 𝐻 𝑠 and the competitive regulations 𝐶 𝑐𝑜𝑚𝑝 are
expressed by the following equations:
𝐻
𝑠 (𝑋
, 𝑋0 , 𝜆, 𝑛) =
, where {
𝑋 𝑛
)
𝑋0
𝑋 𝑛
1+( )
𝑋0
1+𝜆(
𝐻 𝑠− ≡ 𝐻 𝑆
𝐻 𝑠+ ≡ 𝐻 𝑆
(eq. S1.6)
𝑖𝑓 𝜆 < 1
𝑖𝑓 𝜆 > 1
2
𝛾(𝑔𝑛 +( 0
𝐴
0
𝐶𝑅𝑐𝑜𝑚𝑝
(𝛾, 𝑔𝑛 , ℎ, ℎℎ𝑟
, 𝑛ℎ𝑚𝑡 , 𝐴, 𝐴0𝑎𝑟𝑚𝑡 , 𝑛𝑎𝑚𝑡 ) =
𝑚𝑡
𝑚𝑡
1+( 0
ℎ
ℎ
ℎ𝑟𝑚𝑡
0
𝐶𝑅𝑐𝑜𝑚𝑝
(𝑔0 , ℎ, ℎℎ𝑟
, 𝑛ℎ𝑛𝑜𝑥 , 𝑔1 , 𝐴, 𝐴0 , 𝑔2 , 𝑛𝑎𝑛𝑜𝑥 ) =
𝑛𝑜𝑥
𝑛𝑜𝑥
𝑛𝑎𝑚𝑡
𝐴
)
𝑎𝑟𝑚𝑡
𝑛ℎ
𝑚𝑡
)
+( 0
𝐴
)
(eq. S1.7)
𝑛𝑎𝑚𝑡
𝐴
)
𝑎𝑟𝑚𝑡
𝑔0 +𝑔𝑟ℎ𝑛𝑜𝑥 ( 0
ℎ
ℎ
)
ℎ𝑟𝑛𝑜𝑥
1+( 0
ℎ
ℎ
)
𝑛ℎ
𝑛𝑜𝑥
𝑛ℎ
𝑛𝑜𝑥
ℎ𝑟𝑛𝑜𝑥
+𝑔𝑟𝑎𝑛𝑜𝑥 ( 0
𝐴
+( 0
𝐴
𝐴
𝐴
𝑛𝑎𝑛𝑜𝑥
)
𝑎𝑟𝑛𝑜𝑥
𝑛𝑎𝑛𝑜𝑥
)
𝑎𝑟𝑛𝑜𝑥
(eq. S1.8)
As described in Eq. S1, 𝐻 𝑠+ represents excitatory regulations and 𝐻 𝑠− represents inhibitory
regulations, which correspond to 𝜆 > 1 and 𝜆 < 1 respectively. In Eq.S1.6, 𝑋 represents the
concentration of the protein or metabolite, such as pAMPK, HIF-1, and ROS (𝐴, ℎ and 𝑅 in Eq.
S1). 𝑋0 represents the threshold of the Hill function, 𝜆 represents the maximum fold change of
the corresponding protein or metabolites, and 𝑛 represents the Hill coefficient. In Eq. S1.1 and
Eq. S1.2, we used a competitive model to describe the competitive regulations from AMPK and
HIF-1 to mtROS (Eq. S1.7) and noxROS (Eq. S1.8).
4. Mathematical model for the core AMPK:HIF-1:ROS circuit driven by MYC
To model the regulation of glycolysis and OXPHOS by MYC, we included two shifted Hill
0
0
functions - 𝐻 𝑠+ (𝑀𝑌𝐶, 𝑀𝑌𝐶𝑎𝑚𝑝𝑘
, 𝜆𝑀𝑌𝐶,𝑎 , 𝑛𝑀𝑌𝐶,𝑎 ) and 𝐻 𝑠− (𝑀𝑌𝐶 , 𝑀𝑌𝐶ℎ𝑖𝑓−1
, 𝜆𝑀𝑌𝐶,ℎ , 𝑛𝑀𝑌𝐶,ℎ ) to
the production term of pAMPK (Eq. S1.4) and degradation term of HIF-1 (Eq. S1.5)
respectively to represent the contribution of MYC to the production of pAMPK and the
stabilization of HIF-1. The modified equations for MYC are shown in (Eq. S1.9) and (Eq.
S1.10).
2
𝐴̇ = 𝑔𝑎 (1+𝜆
𝑀𝑌𝐶,𝑎
0
0
) 𝐻 𝑠+ (𝑀𝑌𝐶, 𝑀𝑌𝐶𝑎𝑚𝑝𝑘
, 𝜆𝑀𝑌𝐶,𝑎 , 𝑛𝑀𝑌𝐶,𝑎 )(𝑅𝑟𝑎 , 𝑅𝑟𝑎
, 𝜆𝑟𝑎 , 𝑛𝑟𝑎 ) ∙
0
𝐻 𝑠− (ℎℎ𝑎 , ℎℎ𝑎
, 𝜆ℎ𝑎 , 𝑛ℎ𝑎 )(𝐴𝑎𝑎 , 𝐴0𝑎𝑎 , 𝜆𝑎𝑎 , 𝑛𝑎𝑎 ) − 𝑘𝑎 𝐴
2
ℎ̇ = 𝑔ℎ 𝐻 𝑠− (𝐴𝑎ℎ , 𝐴0𝑎ℎ , 𝜆𝑎ℎ , 𝑛𝑎ℎ ) − 𝑘ℎ ℎ(1+𝜆
𝑀𝑌𝐶,ℎ
(eq. S1.9)
0
)𝐻 𝑠− (𝑀𝑌𝐶 , 𝑀𝑌𝐶ℎ𝑖𝑓−1
, 𝜆𝑀𝑌𝐶,ℎ , 𝑛𝑀𝑌𝐶,ℎ ) ∙
0
0
𝐻 𝑠− (ℎℎℎ , ℎℎℎ
, 𝜆ℎℎ , 𝑛ℎℎ )𝐻 𝑠− (𝑅𝑟ℎ , 𝑅𝑟ℎ
, 𝜆𝑟ℎ , 𝑛𝑟ℎ )
(eq. S1.10)
0
We rescaled the two shifted Hill functions - 𝐻 𝑠+ (𝑀𝑌𝐶, 𝑀𝑌𝐶𝑎𝑚𝑝𝑘
, 𝜆𝑀𝑌𝐶,𝑎 , 𝑛𝑀𝑌𝐶,𝑎 ) and
0
𝐻 𝑠− (𝑀𝑌𝐶 , 𝑀𝑌𝐶ℎ𝑖𝑓−1
, 𝜆𝑀𝑌𝐶,ℎ , 𝑛𝑀𝑌𝐶,ℎ ) by (1+𝜆
2
𝑀𝑌𝐶,𝑎
) and (1+𝜆
2
𝑀𝑌𝐶,ℎ
), so that when the level of
0
0
MYC (𝑀𝑌𝐶) is equal to the thresholds (𝑀𝑌𝐶𝑎𝑚𝑝𝑘
, 𝑀𝑌𝐶ℎ𝑖𝑓−1
), the behavior of the AMPK:HIF1:ROS circuit is exactly the same as shown in previous cancer case (Figure 3A). The MYC
relevant parameters can be found in Table S4.
3
5. Mathematical model for the core AMPK:HIF-1:ROS circuit driven by c-SRC
To model the regulation of glycolysis and OXPHOS by c-SRC, we included two shifted Hill
0
0
functions - 𝐻 𝑠+ (𝑆𝑅𝐶, 𝑆𝑅𝐶𝑎𝑚𝑝𝑘
, 𝜆𝑆𝑅𝐶,𝑎 , 𝑛𝑆𝑅𝐶,𝑎 ) and 𝐻 𝑠+ (𝑆𝑅𝐶, 𝑆𝑅𝐶𝑛𝑜𝑥
, 𝜆𝑆𝑅𝐶,𝑛𝑜𝑥 , 𝑛𝑆𝑅𝐶,𝑛𝑜𝑥 ) to the
production term of AMPK (eq. S1.4) and production term of noxROS (eq. S1.2) respectively, to
represent the contribution of c-SRC to the production of pAMPK and noxROS. The modified
equations for c-SRC are shown as (eq. S1.11) and (eq. S1.12).
2
𝐴̇ = 𝑔𝑎 (1+𝜆
𝑆𝑅𝐶,𝑎
0
0
) 𝐻 𝑠+ (𝑆𝑅𝐶, 𝑆𝑅𝐶𝑎𝑚𝑝𝑘
, 𝜆𝑆𝑅𝐶,𝑎 , 𝑛𝑆𝑅𝐶,𝑎 )(𝑅𝑟𝑎 , 𝑅𝑟𝑎
, 𝜆𝑟𝑎 , 𝑛𝑟𝑎 ) ∙
0
𝐻 𝑠− (ℎℎ𝑎 , ℎℎ𝑎
, 𝜆ℎ𝑎 , 𝑛ℎ𝑎 )𝐻 𝑠− (𝐴𝑎𝑎 , 𝐴0𝑎𝑎 , 𝜆𝑎𝑎 , 𝑛𝑎𝑎 ) − 𝑘𝑎 𝐴
2
0
𝑅̇𝑛𝑜𝑥 = 𝑔𝑟𝑛𝑜𝑥 (1+𝜆
)𝐻 𝑠+ (𝑆𝑅𝐶, 𝑆𝑅𝐶𝑛𝑜𝑥
, 𝜆𝑆𝑅𝐶,𝑛𝑜𝑥 , 𝑛𝑆𝑅𝐶,𝑛𝑜𝑥 ) ∙
(eq. S1.11)
𝑆𝑅𝐶,𝑛𝑜𝑥
0
𝐶𝑅𝑐𝑜𝑚𝑝
(𝑔0 , ℎ, ℎℎ𝑟
, 𝑛ℎ𝑛𝑜𝑥 , 𝑔1 , 𝐴, 𝐴0 , 𝑔2 , 𝑛𝑎𝑛𝑜𝑥 )
𝑛𝑜𝑥
𝑛𝑜𝑥
− 𝑘𝑟𝑛𝑜𝑥 𝑅𝑛𝑜𝑥
(eq. S1.12)
0
We rescaled the two shifted Hill functions - 𝐻 𝑠+ (𝑆𝑅𝐶, 𝑆𝑅𝐶𝑎𝑚𝑝𝑘
, 𝜆𝑆𝑅𝐶,𝑎 , 𝑛𝑆𝑅𝐶,𝑎 ) and
2
0
𝐻 𝑠+ (𝑆𝑅𝐶, 𝑆𝑅𝐶𝑛𝑜𝑥
, 𝜆𝑆𝑅𝐶,𝑛𝑜𝑥 , 𝑛𝑆𝑅𝐶,𝑛𝑜𝑥 ) by (1+𝜆
𝑆𝑅𝐶,𝑎
) and (1+𝜆
2
𝑆𝑅𝐶,𝑛𝑜𝑥
) respectively, so that when
0
0
the level of SRC (𝑆𝑅𝐶 ) is equal to the thresholds (𝑆𝑅𝐶𝑎𝑚𝑝𝑘
, 𝑆𝑅𝐶𝑛𝑜𝑥
), the behavior of the
AMPK:HIF-1:ROS circuit is exactly the same as shown in previous cancer case (Figure 3A).
The c-SRC relevant parameters can be found in Table S5.
6. Mathematical model for the core AMPK:HIF-1:ROS circuit driven by RAS
To model the regulation of glycolysis and OXPHOS by RAS, we considered the contribution of
RAS to the production of pAMPK, HIF-1 and mTOR, which is inhibited by AMPK and
promotes the production of HIF-1. The rate equation for mTOR is shown as (eq. S1.13).
𝑚̇ 𝑇 = 𝑔𝑚𝑇 (1+𝜆
2
𝑅𝐴𝑆,𝑚𝑇
0
)𝐻 𝑠+ (𝑅𝐴𝑆, 𝑅𝐴𝑆𝑚𝑇
, 𝜆𝑅𝐴𝑆,𝑚𝑇 , 𝑛𝑅𝐴𝑆,𝑚𝑇 )𝐻 𝑠+ (𝐴, 𝐴0𝑎,𝑚𝑇 , 𝜆𝑎,𝑚𝑇 , 𝑛𝑎,𝑚𝑇 ) −
𝑘𝑚𝑇 𝑚 𝑇
(eq. S1.13)
When considering the effects of RAS, the rate equations for AMPK and HIF-1 are modified as
follows:
2
𝐴̇ = 𝑔𝑎 (1+𝜆
𝑅𝐴𝑆,𝑎
0
0
)𝐻 𝑠+ (𝑅𝐴𝑆, 𝑅𝐴𝑆𝑎𝑚𝑝𝑘
, 𝜆𝑅𝐴𝑆,𝑎 , 𝑛𝑅𝐴𝑆,𝑎 )𝐻 𝑠+ (𝑅𝑟𝑎 , 𝑅𝑟𝑎
, 𝜆𝑟𝑎 , 𝑛𝑟𝑎 ) ∙
0
𝐻 𝑠− (ℎℎ𝑎 , ℎℎ𝑎
, 𝜆ℎ𝑎 , 𝑛ℎ𝑎 )𝐻 𝑠− (𝐴𝑎𝑎 , 𝐴0𝑎𝑎 , 𝜆𝑎𝑎 , 𝑛𝑎𝑎 ) − 𝑘𝑎 𝐴
2
ℎ̇ = 𝑔ℎ (1+𝜆
𝑅𝐴𝑆,ℎ
(eq. S1.14)
0
)𝐻 𝑠+ (𝑅𝐴𝑆, 𝑅𝐴𝑆ℎ𝑖𝑓−1
, 𝜆𝑅𝐴𝑆,ℎ , 𝑛𝑅𝐴𝑆,ℎ )𝐻 𝑠+ (𝑚 𝑇 , 𝑚 𝑇 0𝑎ℎ , 𝜆𝑚𝑇,ℎ , 𝑛𝑚𝑇 ,ℎ ) −
0
0
𝑘ℎ ℎ ∙ 𝐻 𝑠− (ℎℎℎ , ℎℎℎ
, 𝜆ℎℎ , 𝑛ℎℎ )𝐻 𝑠− (𝑅𝑟ℎ , 𝑅𝑟ℎ
, 𝜆𝑟ℎ , 𝑛𝑟ℎ )
(eq. S1.15)
4
0
We rescaled the three shifted Hill functions - 𝐻 𝑠+ (𝑅𝐴𝑆, 𝑅𝐴𝑆𝑚𝑇
, 𝜆𝑅𝐴𝑆,𝑚𝑇 , 𝑛𝑅𝐴𝑆,𝑚𝑇 ) ,
0
𝐻 𝑠+ (𝑅𝐴𝑆, 𝑅𝐴𝑆𝑎𝑚𝑝𝑘
, 𝜆𝑅𝐴𝑆,𝑎 , 𝑛𝑅𝐴𝑆,𝑎 ) and
2
(1+λ
RAS,a
) and (1+λ
2
RAS,nox
𝐻 𝑠+ (𝑅𝐴𝑆, 𝑅𝐴𝑆ℎ0 , 𝜆𝑅𝐴𝑆,ℎ , 𝑛𝑅𝐴𝑆,ℎ ) by (1+𝜆
2
𝑅𝐴𝑆,𝑚𝑇
),
) respectively, so that when the level of RAS (𝑅𝐴𝑆) is equal to the
0
0
thresholds (𝑅𝐴𝑆𝑎𝑚𝑝𝑘
, 𝑅𝐴𝑆ℎ𝑖𝑓−1
), the behavior of the AMPK:HIF-1:ROS circuit is similar as
shown in previous cancer case (Figure 3B). The RAS relevant parameters can be found in Table
S6. In this case, we included the details that AMPK inhibits HIF-1 by first inhibiting mTOR,
which in turn activates HIF-1. Thus we rescaled the basal production rate of HIF-1 from 15 nM/h
to 8 nM/h as shown in Table S6.
7. Principal component analysis (PCA) of RNA-seq data from the TCGA
In this section, we explain how we quantified the activities of AMPK and HIF-1 by using RNAseq data from the TCGA database. Unlike genes in a transcription regulatory network, AMPK is
typically regulated by phosphorylation, while HIF-1 is regulated by protein degradation.
Therefore, the transcription levels of AMPK and HIF-1 may not be effective in indicating the
activities of these genes. Here, instead of directly using their expressions, we examined the
expressions of the downstream targets of AMPK and HIF-1, and quantified their activities by a
special PCA-based method, as shown below.
(1) AMPK and HIF-1 signatures
pAMPK can up-regulate three major transcriptional factors – PGC1𝛼, CREB and FOXO (28),
whose downstream target genes (29-31) are chosen to evaluate the activity of AMPK. HIF-1
itself is a transcriptional factor, therefore its downstream target genes (32) are also chosen to
evaluate the activity of HIF-1.
AMPK downstream gene list (84 genes): ACADL, ACADM, ACOX1, ACOX3, ACSL1,
ACSL3, ACSL4, ACSL5, ANGPTL4, APOC3, APOE, ATF4, ATF5, BAX, BCL2L11, BTG2,
CAT, CCND1, CCND2, CLU, CPT1A, CPT1B, CPT2, CREM, CYP27A1, CYP4A11, CYP7A1,
DDB1, DGAT1, DNMT1, EGR1, EHHADH, ELN, FADS2, FASLG, FOS, FOSB, FOSL1,
FOXA2, G6PC, G6PC2, G6PC3, GADD45A, GADD45B, GADD45G, GK, HES1, HNF4A,
HSPD1, ID1, ID3, JUNB, JUND, KLF5, LPL, MAFA, MAPK9, MECR, MMP9, NCOA2,
NR1H3, NR4A1, NR4A2, NUP155, NUP88, NUP98, OLR1, ONECUT1, ONECUT2, PCK1,
PCK2, PDK4, PDX1, PRMT1, PRMT3, RUVBL1, SCD, SLC2A4, SOD2, SORBS1, SP1,
STAT5A, TOB1, TXNIP.
HIF-1 downstream gene list (59 genes): ADM, ALDH4A1, ALDOA, ALDOC, ARHGEF1,
ASPH, BHLHE40, BNIP3, BTAF1, CA9, CCNB1, CITED2, CP, DDIT4, EDN1, EGLN1,
EGLN3, ENO1, EPRS, ETS1, FN1, GLCCI1, GRK6, HACD3, HSP90B1, IVNS1ABP, KDM3A,
MECOM, MET, MXD1, MYLK, NAMPT, NDRG1, NOS2, NOS3, P4HA1, P4HA2, P4HTM,
PDGFA, PFKL, PGAM1, PGK1, PLOD1, RARA, RBPJ, RNF165, RRAGD, RSBN1,
SERPINE1, SLC31A1, SLC7A6, SOX6, SSRP1, STC2, TFRC, TGFB3, TMEFF1, TMEM45A,
VEGFA.
We first examined the RNA-seq data of liver hepatocellular carcinoma from the TCGA database
(373 samples). The liver cancer was first picked because liver is known to play a crucial role in
5
metabolism of human body. The RNA-seq data of the aforementioned AMPK and HIF-1
downstream genes were extracted and compiled into two separate datasets. For each dataset, the
RSEM expression level of each gene first undergoes log2 transformation and standardization, i.e.
𝑥→
̅̅̅̅̅̅̅̅̅̅
𝑙𝑜𝑔2 (𝑥) − 𝑙𝑜𝑔
2 (𝑥)
,
𝜎(𝑙𝑜𝑔2 (𝑥))
̅̅̅̅̅̅̅̅̅̅
where 𝑥 represents the expression for the gene, 𝑙𝑜𝑔
2 (𝑥) is the mean of the log2 transformed
value and 𝜎(𝑙𝑜𝑔2 (𝑥)) is the standard deviation of the log2 transformed value.
Each of the processed expression datasets (AMPK-related data or HIF-1-related data) was
subject to PCA. Here, the cancer dataset was assumed to be diverse enough to cover samples
with both strong and weak metabolisms. Therefore, we can use first principal components (PC1s)
of the AMPK and HIF-1 datasets to quantify the activities of AMPK and HIF-1, respectively.
For this liver cancer TCGA data, we found the contribution of the AMPK PC1 is about 18%, and
the contribution of the HIF-1 PC1 is about 12%. Nevertheless, we already observed a significant
anti-correlation between the values of these two axes (Figure S4B), even through the two PC1s
were obtained independently.
We further improve the axes by picking the most relevant genes and performed PCA again. To
do this, for each dataset (AMPK or HIF-1), we selected top 40% of the genes according to the
absolute values of the corresponding components in the previously obtained PC1 (Figure S4A).
With the updated sets of genes for both AMPK and HIF-1 (Table S9), we redid PCA for the
RNA-seq data of liver hepatocellular carcinoma. By using the updated genes, we found the
contribution of the new AMPK PC1 is dramatically increased to 41%, and the contribution of the
new HIF-1 PC1 is about 26%. The eigenvalues of the PC1 (13.5 for AMPK and 5.9 for HIF-1)
is significantly larger than the eigenvalues of the PC2 (2.7 for AMPK and 2.8 for HIF-1). As
shown in Figure S4C, the downstream genes of AMPK and HIF-1 from the refined gene lists
have similar contribution to the PC1s and there is no dominant case, which further validates our
approach to include many downstream genes instead of only a few of them. Using these updated
gene sets, we got a similar anti-correlation between the values of the two new axes (Figure S4D).
We propose that the AMPK and HIF-1 activities can be quantified by the AMPK and HIF-1
PC1s, respectively.
For the other cancer types, we also applied the PCA-based method to obtain the AMPK and HIF1 signatures using the updated gene sets. As shown in Table S7 and Table S8, the PC1s
obtained from different cancers are mostly similar to one another. In addition, we evaluated lung
adenocarcinoma data using the AMPK and HIF-1 axes obtained from liver hepatocellular
carcinoma (Figure S3). Similar anti-correlations were observed for the lung adenocarcinoma
data using both sets of axes. Moreover, the three clusters (Figure S3A) obtained from k-mean
clustering using the axes derived from the lung adenocarcinoma data can also be observed in the
plot using the axes derived from the liver hepatocellular carcinoma data (Figure S3B). All of
these suggest that the PCA-based method is robust in quantifying the activities of AMPK and
HIF-1.
6
(2) Evaluation of AMPK and HIF-1 activities for different subtypes of invasive breast carcinoma.
To evaluate the AMPK and HIF-1 activities for Luminal A, Luminal B, Her2+ and triple
negative subtypes of invasive breast carcinoma from TCGA, we revised the aforementioned
standardization step to taking into account the possible bias caused by different number of
samples in each subtype. The RSEM expression level of each gene first undergoes log2
transformation and then is standardized by the weighted mean and the weighted standard
̅̅̅̅̅̅̅̅̅̅̅̅̅̅
𝑙𝑜𝑔 (𝑥 )−𝑙𝑜𝑔
2 (𝑥𝑖 )𝑤
̅̅̅̅̅̅̅̅̅̅̅
deviation, i.e. 𝑥𝑖 → 2 𝑖
, where 𝑥 represents the expression for the gene, 𝑙𝑜𝑔
2 (𝑥𝑖 )𝑤
𝜎𝑤 (𝑙𝑜𝑔2 (𝑥𝑖 ))
is the weighted mean of the log2 transformed value and 𝜎𝑤 (𝑙𝑜𝑔2 (𝑥𝑖 )) is the weighted standard
∑𝑛 𝑤 𝑙𝑜𝑔 (𝑥 )
deviation of the log2 transformed value, i.e. ̅̅̅̅̅̅̅̅̅̅̅
𝑙𝑜𝑔2 (𝑥𝑖 )𝑤 = 𝑖=1∑𝑛𝑖 𝑤2 𝑖 , 𝜎𝑤 (𝑙𝑜𝑔2 (𝑥𝑖 )) =
̅̅̅̅̅̅̅̅̅̅̅̅ 2
∑𝑛
𝑖=1 𝑤𝑖 (𝑙𝑜𝑔2 (𝑥𝑖 )−𝑙𝑜𝑔2 (𝑥𝑖 )𝑤 )
∑𝑛
𝑖=1 𝑤𝑖
𝑖=1
𝑖
1
, where 𝑤𝑖 is the weight factor, 𝑤𝑖 = 𝑛, and 𝑛 is the total number of
samples in that subtype.
(3) Comparison of the AMPK and HIF-1 activities of different types of tumors.
Here we uniformed the averaged expression level of ACTB (Figure 5E), TUBB (Figure S6A)
and GAPDH (Figure S6B) from different types of tumors and then rescaled the other gene
expression accordingly. We combined RNA-seq data from different types of tumors together and
analyzed their relative AMPK and HIF-1 activities following the aforementioned procedure.
(4) Evaluation of the AMPK and HIF-1 activity using single-cell RNA-seq data of lung
adenocarcinoma.
The RNA-seq data is analyzed following the same procedure as explained in (1). Details of the
data processing are presented in the caption of Fig S10.
8. Quantification of the activities of MYC, c-SRC and AKT.
We applied an existing set of MYC, c-SRC and AKT signatures from a previous study (33).
Each of the signatures contains a list of upregulated genes (Table S12) identified in both human
breast cell cultures and prostate tumors. The RNA-seq data of the genes from each signature
were extracted from TCGA for each sample and underwent log2 transformation and
standardization. For each oncogene pathway, we assigned an oncogenic score by the number of
genes whose normalized expression is positive. The oncogenic score was applied to evaluate
the activities of the MYC, c-SRC and AKT pathways in samples with different metabolic states –
W, W/O and O (Figure 5). Here, we identified the samples whose oncogenic scores are among
the top 20% and evaluated their enrichments in each metabolic state (W, W/o and O) (Figure 5).
Similarly, we also identified the samples whose oncogenic scores are among the bottom 20% and
evaluated their enrichments in each metabolic state (Figure S7).
9. Modeling the effects of metabolic therapy to the core metabolic regulatory circuit
In the main text, we showed the one-parameter and two-parameter bifurcations for the response
of various treatment strategies on the dynamic behaviors of the AMPK:HIF-1:ROS core circuit.
Specifically, we modeled the administration of metformin, 3BP, hyperbaric oxygen, the
combination of metformin and 3BP, and the combination of metformin and the hyperbaric
7
oxygen. Here, we describe how we modeled the effects of the therapy to the circuit dynamics as
follows.
Hyperbaric oxygen therapy: The hyperbaric oxygen mainly influences the HIF-1 degradation
by directly binding to Prolyl hydroxylases (PHDs), which is crucial for HIF-1α
hydroxylation(34). However, ROS (such as H2O2), can decrease the binding of oxygen and
PHDs, thus stabilizes the HIF-1 (35). Therefore, we can simplify the chemical reactions as
follows:
𝑟𝑂+
→
𝑛𝑜ℎ O2 +HIF-1 ← HO
𝑟𝑂−
𝑟𝑅+
→
HR
←
𝑛𝑟ℎ ROS+HIF-1
𝑟𝑅−
[HIF-1]+[HO]+[HR] = 1 ,
where HO represents the HIF-1 that binds to oxygen, and HR represents the HIF-1 that binds to
𝑟
ROS. Here, we assume the sum of the HIF-1 related products is a constant. By assuming 𝑟𝑂− =
[𝑂20 ]𝑛𝑜ℎ
and
𝑟𝑅−
𝑟𝑅+
𝑂+
=
[𝑅 0 ]𝑛𝑟ℎ
at equilibrium, we obtain
(
𝑛𝑜ℎ
[𝑂2 ]
𝑂20
)
=
[𝐻𝑂]
[HIF-1]
=
[𝐻𝑅]
[HIF-1]
𝑛𝑟ℎ
[𝑅]
( )
𝑅0
[𝑂 ] 𝑛𝑜ℎ
Therefore, [HIF-1] + ( 𝑂20 )
2
[𝑅] 𝑛𝑟ℎ
[HIF-1] + ( 𝑅 )
0
[HIF-1] = 1. Hence, under the influence of
oxygen, the dynamics of HIF-1 levels can be expressed as:
𝑑[HIF-1]
𝑑𝑡
= −𝑘ℎ 𝐶(𝑘1 , 𝑘2 , 𝑘3 , 𝑅, 𝑂, 𝑅ℎ0 , 𝑂ℎ0 , 𝑛𝑟ℎ , 𝑛𝑜ℎ )[HIF-1]
[𝑅]
where 𝐶(𝑘1 , 𝑘2 , 𝑘3 , 𝑅, 𝑂, 𝑅ℎ0 , 𝑂ℎ0 , 𝑛𝑟ℎ , 𝑛𝑜ℎ , ) =
𝑛𝑟ℎ
𝑘1 +𝑘2 ( 0 )
𝑅
ℎ
[𝑅]
1+( 0 )
𝑅
ℎ
,
𝑛𝑟ℎ
[𝑂 ]
+𝑘3 ( 02 )
(eq. S2.1)
𝑛𝑜ℎ
𝑂ℎ
𝑛𝑜ℎ
[𝑂 ]
+( 02 )
𝑂ℎ
, 𝑘ℎ is the basal
degradation rate of HIF-1, k1, k2, k3 together represent the contributions of the oxygen and ROS
for HIF-1 degradation.
Hence, the dynamics of HIF-1 levels in the circuit can be expressed as:
0
ℎ̇ = 𝑔ℎ 𝐻 𝑠− (𝐴𝑎ℎ , 𝐴0𝑎ℎ , 𝜆𝑎ℎ , 𝑛𝑎ℎ ) − 𝑘ℎ ∙ ℎ ∙ 𝐻 𝑠− (ℎℎℎ , ℎℎℎ
, 𝜆ℎℎ , 𝑛ℎℎ ) ∙ 𝛼 ∙
0
0
𝐶(𝑘1 , 𝑘2 , 𝑘3 , 𝑅, 𝑂, 𝑅ℎ , 𝑂ℎ , 𝑛𝑟ℎ , 𝑛𝑜ℎ )
(eq. S2.2)
8
Here, α is a constant parameter, which is chosen so that Eq. S2.2 is equivalent to Eq.S1.5.
0
Therefore, 𝛼 ∙ 𝐶(𝑘1 , 𝑘2 , 𝑘3 , 𝑅, 𝑂, 𝑅ℎ0 , 𝑂ℎ0 , 𝑛𝑟ℎ , 𝑛𝑜ℎ ) = 𝐻 𝑠− (𝑅𝑟ℎ , 𝑅𝑟ℎ
, 𝜆𝑟ℎ , 𝑛𝑟ℎ ) for any level of
ROS. As a result, there are only two non-redundant parameters among 𝛼, 𝑘1 , 𝑘2 and 𝑘3 .
Metformin-based therapy: As we mentioned in the main text, metformin not only activates
AMPK but also inhibits HIF-1 in an AMPK-independent way(36). Therefore, we used a shifted
𝑠+
𝑠+
Hill functions 𝐻𝑚𝑎
to model the pAMPK generation and 𝐻𝑚ℎ
to model the HIF-1 degradation
under the influences of metformin. In addition, metformin can also induce high mtROS
production by altering the mitochondrial potential(37). Hence, we modeled the increase in the
mtROS production rate to be proportional to the level of metformin.
(𝑔𝑟𝑚𝑡 )
𝑚𝑒𝑡
= 𝑔𝑟𝑚𝑡 + 𝛼𝑚𝑒𝑡 ∙ [𝑀𝑒𝑡𝑓𝑜𝑟𝑚𝑖𝑛]
(eq. S3)
Here, (𝑔𝑟𝑚𝑡 )
is the generation rate of mtROS under the influence of metformin, and 𝛼𝑚𝑒𝑡 is
𝑚𝑒𝑡
a weighting factor for the mtROS production by metformin.
Therefore, we used the following equations to model the dynamics of pAMPK and HIF-1levels
under the influence of metformin,
0
0
𝐴̇ = 𝑔𝑎 𝐻 𝑠+ (𝑅𝑟𝑎 , 𝑅𝑟𝑎
, 𝜆𝑟𝑎 , 𝑛𝑟𝑎 )𝐻 𝑠− (ℎℎ𝑎 , ℎℎ𝑎
, 𝜆ℎ𝑎 , 𝑛ℎ𝑎 )
𝑠− (𝐴
0
𝑠+
0
∙𝐻
𝑎𝑎 , 𝐴𝑎𝑎 , 𝜆𝑎𝑎 , 𝑛𝑎𝑎 )𝐻 (𝑀𝑚𝑎 , 𝑀𝑚𝑎 , 𝜆𝑚𝑎 , 𝑛𝑚𝑎 ) − 𝑘𝑎 𝐴
(eq. S4)
0
ℎ̇ = 𝑔ℎ 𝐻 𝑠− (𝐴𝑎ℎ , 𝐴0𝑎ℎ , 𝜆𝑎ℎ , 𝑛𝑎ℎ )𝐻 𝑠− (𝑀𝑚ℎ , 𝑀𝑚ℎ
, 𝜆𝑚ℎ , 𝑛𝑚ℎ ) − 𝑘ℎ ∙ ℎ ∙
0
0
𝑠− (ℎ
𝑠−
𝐻
ℎℎ , ℎℎℎ , 𝜆ℎℎ , 𝑛ℎℎ )𝐻 (𝑅𝑟ℎ , 𝑅𝑟ℎ , 𝜆𝑟ℎ , 𝑛𝑟ℎ )
(eq. S5)
3BP (or 2DG) therapy: Since 3BP inhibits glycolysis, it decreases the strength of HIF-1 selfactivation. Thus, we modeled the effects of 3BP as
(𝜆ℎℎ )3𝐵𝑃 = 𝜆ℎℎ + 𝛼3𝐵𝑃 ∙ [3𝐵𝑃]
(eq. S6)
Here, (𝜆ℎℎ )3𝐵𝑃 is the maximum fold change due to HIF-1 self-activation under the influence of
3BP (or 2DG), and 𝛼3𝐵𝑃 is a constant representing the extent of the glycolysis reduction.
The parameters and the corresponding references involved in the treatment modeling are shown
in Table S11.
10. Statistical test for the association of oncogenic pathways with metabolic states
In Figure 5, we compared the fraction of the top 20% samples with high activity of either MYC,
c-SRC or AKT among all the samples of each metabolic group (‘W’, ‘W/O’ and ‘O’). To
evaluate the statistical significance of the comparison, we constructed a 2x2 contingency table
for any pair of metabolic groups (‘W’ versus ‘W/O’, ‘W’ versus ‘O’ and ‘W/O’ versus ‘O’). The
table lists the number of samples for each metabolic group within the top 20% category or the
rest samples. Chi-square test was applied to the contingency table to calculate the p-values.
9
Only those pairs whose p-values are below 0.05 are labeled by a line in the figure. Similarly, we
also performed the statistical test for the fraction of the bottom 20% samples, as shown in Figure
S8.
10
Supplemental Tables
Table S1. Experimental evidences for the requirement of OXPHOS for cancer metastasis.
References
(10)
(38)
(39)
(40)
(41)
(42)
(43)
(44)
(45)
Cell Lines
SiHa human cervix squamous cell carcinoma, WiDr human colorectal
adenocarcinoma, FaDu human pharynx squamous cell carcinoma, HeLa human
epithelial cervix cancer, and HCT116 human colorectal carcinoma cancer cells
SiHa-F3 human cervix squamous cell carcinoma
4T1 mouse mammary adenocarcinoma, B16F10 mouse melanoma, MDA-MB231 human mammary adenocarcinoma, and MDA-MB-435 human melanoma
cell lines
A459 non-small cell lung carcinoma cells
MDA-MB-231 human metastatic breast cancer cells, HepG2 human
hepatocellular carcinoma cells, PC3 human metastatic prostate cancer cells,
B16F10 mouse metastatic melanoma cells and NIH-3T3 mouse embryonic
fibroblast cells
Mice and human pancreatic cancer cells
Melanoma cells
Breast cancer MCF-7, glioblastoma U87, colon cancer HCT116 cell lines
Human breast cancer cell lines (MDA-MB-231, MDA-MB-468, MDA-MB-453)
11
Table S2. Experimental evidences for the metabolic regulatory network and core AMPK:HIF1:ROS regulatory circuit.
Regulations
HIF-1 inhibits AMPK
AMPK inhibits HIF-1
HIF-1 self-activation
AMPK self-inhibition
AMPK promotes mtROS
Involved
Molecules/
Metabolisms
ROS
References
(9)
Caenorhabditis elegans
ROS
Raptor
mTOR
(9)
(8)
(46)
Caenorhabditis elegans, liver
cells, human cervical epithelial
adenocarcinoma cells
Glycolysis
Pyruvate
Lactate
ATP
Fatty acid
oxidation
Fatty acid
oxidation
FOXO3
Glutathione
Thioredoxin
(11)
(12)
(10)
(47)
(5)
(14)
Human vascular muscle cells,
human gliomas cells
Escherichia coli cells, skeletal
muscle cells
(15)
Kidney cortical tubules
Human aortic endothelial cells,
AMPK inhibits nox ROS
NADPH oxidases
HIF-1 inhibits mtROS
Mitochondrial
metabolism
HIF-1 promotes noxROS
NADPH oxidase-2
mtROS activates AMPK
Mitochondrial
metabolism
AMP/ATP
(48)
(17)
(16)
(20)
(21)
(49)
(13)
(50)
(24)
(25)
(22)
(51)
mtROS stabilizes HIF-1
Oxygen
Mitochondria
(52)
(53)
noxROS activates AMPK
H2O2
(26)
noxROS stabilizes HIF-1
NOX
PDH
(23)
HIF-1 activates SNAIL
Direct
transcriptional
regulation
AMPK inhibits mtROS
Cell lines
(54)
(55)
HIF-1 activates TWIST
Direct
transcriptional
regulation
(56)
AMPK inhibits TGF-β
p300
(57)
(58)
TWIST inhibits AMPK
mTOR
(59)
MEFs, HT1080, and HeLa cells
Vascular smooth muscle cells,
blood cells
Lung cancer cells, UT-7/TPO
cells
PC12 cells, human umbilical
endothelial cells
Rat heart cells, vascular smooth
muscle cells
Breast cancer cells, human
immortalized hepatocyte cells,
human lung cells
Human embryonic kidney cells
(HEK 293)
HL-7702 immortalized human
hepatocyte cells
Human pancreatic cancer cell
lines, human HCC cell lines
HepG2 and SMMC-7721, human
embryonic kidney cell line
HEK293
Human hypopharyngeal
carcinoma (FaDu), embryonic
kidney (293T) cell lines, breast
cancer (MCF-7), lung cancer
(H1299) and tongue cancer
(SAS) cell lines
Human primary mesangial cells
(HMC), Hepatic stellate cell
(HSC)
H1299 and A549 non-small cell
12
lung carcinoma cells
TGF-β elevates OXPHOS
MYC stabilizes HIF-1
MYC activates AMPK
c-SRC activates noxROS
c-SRC activates AMPK
RAS activates HIF-1
RAS activates AMPK
RAS activates mTOR
AMPK inhibits mTOR
mTOR activates HIF-1
Fatty acid synthesis
Fatty acid
oxidation
VHL complex
posttranslational
modifications
ATP, ROS
NADPH
oxidase/Rac
pathway
PKCα, PLCγ,
LKB1
Raf/MEK/ERK
pathway
(40)
A459 non-small cell lung
carcinoma cells
(60)
(61)
IMEC epithelial cells MCF7
cells
(62)
MEF cells
(63)
293 and HeLa cells
(64)
OVCAR3 and A431 cells
(65)
(66)
HEK 293 cells
B-raf, Erk, LKB1
(66)
PI3K/AKT
pathway
Tuberous sclerosis
complex (TSC) 1
and 2
Glucose- and
reoxygenationdependent manner
(66)
(67)
Lung cancers and cervical
cancers
Pancreatic ductal
adenocarcinomas (PDAC)
(68)
HEK293 cells
(46)
HeLa cells
13
Table S3. Parameters for modeling the AMPK:HIF-1:ROS regulatory circuit.
Parameters
Description
Reference(s)
Value
Unit
𝒈𝒂
30
nM/h
Production rate of AMPK
𝒌𝒂
0.2
ℎ−1
Degradation rate of AMPK
𝑨𝟎𝒂𝒂
350
nM
Threshold for self-inhibition
(19)
𝑨𝟎𝒂𝒉
250
nM
Threshold for HIF-1 inhibition
(19)
𝑨𝟎𝒂𝒓
350
nM
Threshold for mtROS inhibition
(19)
𝑨𝟎𝒂𝒓𝒎𝒕
150
nM
Threshold for mtROS activation
(19)
𝑨𝟎𝒂𝒓𝒏𝒐𝒙
150
nM
Threshold for noxROS inhibition
(19)
𝝀𝒂𝒂
0.2
-
Fold change for self-inhibition
𝝀𝒂𝒉
0.1
-
Fold change for HIF-1 inhibition
𝝀𝒂𝒓−
0.25
-
Fold change for mtROS inhibition
γ*
3
-
Fold change for mtROS activation
𝒈𝟐
0.2
-
Fold change for noxROS inhibition
𝒏𝒂𝒂
2
-
Hill coefficient for self-inhibition
𝒏𝒂𝒉
1
-
Hill coefficient for HIF-1 inhibition
𝒏𝒂𝒓−
2
-
Hill coefficient for mtROS inhibition
𝒏𝒂𝒓_𝒎𝒕
4
-
Hill coefficient for mtROS activation
𝒏𝒂𝒓_𝒏𝒐𝒙
2
-
Hill coefficient for noxROS inhibition
𝒈𝒉
15
nM/h
Production rate of HIF-1
(69)
𝒌𝒉 *
0.25
ℎ−1
Degradation rate of HIF-1
(69)
𝒉𝟎𝒉𝒉
80
nM
Threshold for HIF-1 self-activation
(35)
𝒉𝟎𝒉𝒂
250
nM
Threshold for AMPK inhibition
(35)
AMPK
(9,48)
HIF-1
14
𝒉𝟎𝒉𝒓𝒎𝒕
200
nM
Threshold for mtROS inhibition
(35)
𝒉𝟎𝒉𝒓𝒏𝒐𝒙
250
nM
Threshold for noxROS activation
(35)
𝝀𝒉𝒉
0.1
-
Fold change for HIF-1 self-activation
𝝀𝒉𝒂
0.1
-
Fold change for AMPK inhibition
𝒈𝟏
5
-
Fold change for noxROS activation
𝒏𝒉𝒉
4
-
Hill coefficient for HIF-1 self-activation
𝒏𝒉𝒂
1
-
Hill coefficient for AMPK inhibition
𝒏𝒉_𝒎𝒕
2
-
Hill coefficient for mtROS inhibition
𝒏𝒉_𝒏𝒐𝒙
2
-
Hill coefficient for noxROS inhibition
𝒈𝒓𝒎𝒕
150
𝜇M/min
Production rate of mitochondrial ROS
(35,70)
𝒈𝒓𝒏𝒐𝒙
40
𝜇M/min
Production rate of NOX derived ROS
(35,70)
𝒈𝒏
0.2
-
𝒌𝒓𝒎𝒕
5.0
/min
Degradation rate of mitochondrial ROS
(35,70)
𝒌𝒓𝒏𝒐𝒙
5.0
/min
Degradation rate of NOX derived ROS
(35,70)
𝑹𝟎𝒓𝒂
100
𝜇M
Threshold for AMPK activation
(26,35,70)
𝑹𝟎𝒓𝒉
300
𝜇M
Threshold for HIF-1 activation
(26,35,70)
𝝀𝒓𝒂 *
8
-
Fold change for AMPK activation
𝝀𝒓𝒉
0.2
-
Fold change for HIF-1 activation
𝒏𝒓𝑎
4
-
Hill coefficient for AMPK activation
𝒏𝒓𝒉
4
-
Hill coefficient for HIF-1 activation
ROS
Basal cytosol ROS
(26)
* 𝛾 and 𝑘ℎ are different for normal cells and cancer cells, which has been discussed in details in
the main text part (see Methods). The values in the table are corresponding to the cancer cells.
For normal cells, the values of these two parameters are 3, 0.3/min respectively.
15
Table S4. Parameters for modeling the regulation of MYC on the AMPK:HIF-1:ROS regulatory
circuit. The other parameters in SI section 4 are taken from Table S3.
Parameters
Value
Unit
Description
MYC activation on AMPK
𝑴𝒀𝑪𝟎𝒂𝒎𝒑𝒌
500
nM
𝝀𝑴𝒀𝑪,𝒂
2
-
Fold change for MYC regulation on AMPK production
𝒏𝑴𝒀𝑪,𝒂
3
-
Hill coefficient for MYC regulation on AMPK production
Threshold for MYC regulation on AMPK production
MYC stabilization on HIF-1
𝑴𝒀𝑪𝟎𝒉𝒊𝒇−𝟏
500
nM
𝝀𝑴𝒀𝑪,𝒉
0.2
-
Fold change for MYC regulation of HIF-1 degradation
𝒏𝑴𝒀𝑪,𝒉
3
-
Hill coefficient for MYC regulation of HIF-1 degradation
Threshold for MYC regulation of HIF-1 degradation
Table S5. Parameters for modeling the regulation of c-SRC on the AMPK:HIF-1:ROS
regulatory circuit. The other parameters in SI section 5 are taken from Table S3.
Parameters
Value
Unit
Description
c-SRC activation on noxROS
𝑺𝑹𝑪𝟎𝒏𝒐𝒙
500
nM
𝝀𝑺𝑹𝑪,𝒏𝒐𝒙
6
-
Fold change for c-SRC regulation on noxROS production
𝒏𝑺𝑹𝑪,𝒏𝒐𝒙
3
-
Hill coefficient for c-SRC regulation on noxROS production
Threshold for c-SRC regulation on noxROS production
c-SRC activation on AMPK
𝑺𝑹𝑪𝟎𝒂𝒎𝒑𝒌
500
nM
𝝀𝑺𝑹𝑪,𝒂
3
-
Fold change for c-SRC regulation on AMPK production
𝒏𝑺𝑹𝑪,𝒂
1
-
Hill coefficient for c-SRC regulation on AMPK production
Threshold for c-SRC regulation on AMPK production
16
Table S6. Parameters for modeling the regulation of RAS on the AMPK:HIF-1:ROS regulatory
circuit. The other parameters in SI section 6 except for 𝒈𝒉 are taken from Table S3.
Parameters
Value
Description
Unit
RAS activation on AMPK
𝑹𝑨𝑺𝟎𝒂𝒎𝒑𝒌
500
nM
𝝀𝑹𝑨𝑺,𝒂
2
-
Fold change for RAS regulation on AMPK production
𝒏𝑹𝑨𝑺,𝒂
3
-
Hill coefficient for RAS regulation on AMPK production
Threshold for RAS regulation on AMPK production
RAS activation on HIF-1
𝑹𝑨𝑺𝟎𝒉𝒊𝒇−𝟏
500
nM
𝝀𝑹𝑨𝑺,𝒉
2.5
-
Fold change for RAS regulation on HIF-1 production
𝒏𝑹𝑨𝑺,𝒉
3
-
Hill coefficient for RAS regulation on HIF-1 production
Threshold for RAS regulation on HIF-1 production
RAS activation on mTOR
𝑹𝑨𝑺𝟎𝒎𝑻
500
nM
𝝀𝑹𝑨𝑺,𝒎𝑻
2
-
Fold change for RAS regulation on mTOR production
𝒏𝑹𝑨𝑺,𝒎𝑻
1
-
Hill coefficient for RAS regulation on mTOR production
𝒈𝒎𝑻
100
-
Production rate of mTOR
𝒌𝒎𝑻
0.5
ℎ−1
Degradation rate of mTOR
8
nM/h
Threshold for RAS regulation on mTOR production
mTOR
HIF-1
𝒈𝒉
Production rate of HIF-1 (69)
AMPK inhibition on mTOR
𝑨𝟎𝒂,𝒎𝑻
150
nM
𝝀𝒂,𝒎𝑻
0.3
-
Fold change for AMPK regulation on mTOR production
𝒏𝒂,𝒎𝑻
3
-
Hill coefficient for AMPK regulation on mTOR production
Threshold for AMPK regulation on mTOR production
mTOR activation on HIF-1
𝒎𝑻 𝟎𝒉
150
nM
𝝀𝒎𝑻,𝒉
2
-
Fold change for mTOR regulation of HIF-1 production
𝒏𝒎𝑻,𝒉
3
-
Hill coefficient for mTOR regulation of HIF-1 production
Threshold for mTOR regulation of HIF-1 production
17
Table S7. Comparison of the first AMPK principal components (PC1s) derived from the TCGA
data of different types of cancer.
The first principal component (PC1) of 33 AMPK downstream genes expression is used to
evaluate the activity of AMPK for each cancer type. To evaluate the similarity between different
PC1s, we calculated the dot product of any pair of PC1s derived from different cancer types. For
most of the cancer types, the AMPK PC1s are quite consistent with dot product between most
pairs of PC1s close to 1.
18
Table S8. Comparison of the first HIF-1 principal components (PC1s) derived from the TCGA
data of different types of cancer.
The first principal component (PC1) of 33 HIF-1 downstream genes expression is used to
evaluate the activity of HIF-1 for each cancer type. To evaluate the similarity between different
PC1s, we calculated the dot product of any pair of PC1s derived from different cancer types. For
most of the cancer types, the HIF-1 PC1s are quite consistent with dot product between most
pairs of PC1s close to 1.
*Since the RNA-seq data of gene TMEFF1 in stomach adenocarcinoma is not available from the
TCGA database, we removed the components for the gene TMEFF1 in the PC1 eigenvectors
from other cancer types. The PC1s were normalized again before we computed the dot products.
19
Table S9. Downstream gene lists for both AMPK and HIF-1 and the reference to support the upand down-regulations.
Downstream genes of
AMPK
Up-regulated genes
ACADL
ACADM
ACOX1
ACSL1
ACSL5
ANGPTL4
APOC3
APOE
CAT
CPT1A
CPT2
CYP27A1
CYP4A11
CYP7A1
EHHADH
FOXA2
G6PC
Reference(s)
(72)
(73)
(74)
(75)
Downstream genes of
HIF-1
Up-regulated genes
ALDOA
BHLHE40
CA9
CCNB1
DDIT4
EGLN3
EPRS
ETS1
IVNS1ABP
(76,77)
KDM3A
(79)
MECOM
MXD1
PGK1
SERPINE1
SSRP1
STC2
TFRC
GADD45A
TGFB3
GADD45G
HNF4A
ONECUT2
PCK1
PCK2
PDK4
TOB1
Down-regulated genes
TMEFF1
TMEM45A
VEGFA
Down-regulated genes
ALDH4A1
BNIP3
CCND2
DNMT1
G6PC3
MMP9
PRMT1
RUVBL1
ATF4
BAX
(72)
(76)
Reference(s)
(71)
(78)
(80)
(81)
(82)
(83)
(84,85)
(86)
(87)
(88)
20
Table S10. Cancer treatment experiments targeting cancer cellular metabolism.
Drug
FTS
Effects to metabolic genes/pathways
Activating AMPK
Activating AMPK, inhibiting HIF-1,
promoting ROS production
Inhibiting HIF-1
2DG
Inhibiting glycolysis
3BP
Inhibiting glycolysis
Tiron
Scavenging ROS
mitoTEMPO
Scavenging mtROS
FTS+2DG
Inhibiting HIF-1 and glycolysis
Inhibiting glycolysis, Activating AMPK,
promoting ROS production
AICAR
Metformin
Metformin+2DG
Cell Lines
Cervix, prostate, liver (89)
Breast cancer (90)
Mice pancreatic tumor (91)
Bovine aortic endothelial cells, human
umbilical vein endothelial cells (92),
Colon carcinoma cell lines (93), mive
hepatocellular carcinoma (94)
Mice hepatocellular carcinoma (94,95),
colon carcinoma cell lines (93)
Calu-6 human pulmonary
adenocarcinoma cell line (96)
B16F10 murine melanoma tumor cells
(38)
Human pancreatic cancer cell line (97)
LNCap and P69 cells (98), human
gastric and esophageal cell lines (37)
21
Table S11. Parameters for modeling the therapeutic treatment.
Parameters
Value
Unit
Description
Reference(s)
Hyperbaric Oxygen
𝒌𝟐
0.2
-
HIF-1 fold change due to ROS
(99)
𝒌𝟑
1.8
-
HIF-1 fold change due to oxygen
(35,99)
𝒏𝒐𝒉
2
-
Hill coefficient for HIF-1 inhibition
𝑶𝟎𝒉
5
%
Threshold for AMPK activation
(69,99,100)
Metformin
𝝀𝒎𝒂
2
-
Fold change for AMPK activation
(19)
𝝀𝒎𝒉
0.5
-
Fold change for HIF-1 inhibition
(36)
𝒏𝒎𝒂
2
-
Hill coefficient for AMPK activation
𝒏𝒎𝒉
2
-
Hill coefficient for HIF-1 inhibition
𝑴𝟎𝒎𝒂
300
μM
Threshold for AMPK activation
(19,36,37,98,101)
𝑴𝟎𝒎𝒉
300
μM
Threshold for HIF-1 inhibition
(19,36,37,98,101)
𝜶𝒎𝒆𝒕
0.025
/μM
Influence on mtROS maximum fold
change
3BP
𝜶𝟑𝑩𝑷
0.0005
/μM
Influence on HIF-1 fold change due to
self-activation
22
Table S12. Downstream upregulated gene list for MYC, c-SRC and AKT (33).
MYC
Up-regulated genes
ADSL
BYSL
CCT5
CCT7
CYC1
DKC1
EBNA1BP2
EIF2S1
EIF3D
FRAT2
GSPT1
HK2
HNRNPAB
HNRNPF
HSPA8
HSPD1
HTRA2
NCL
NME1
NOP16
NOP56
PABPC4
PPIF
PPRC1
PWP2
RAN
RSL1D1
SERBP1
SLC19A1
SLC29A2
SMPDL3B
SRM
SRSF1
c-SRC
Up-regulated genes
ARAP1
ARPC4
ASIP
BRD4
CAMK2B
CCL7
CLDN9
DDX51
GNAT1
GP9
GSK3A
GTSE1
HNF4A
KIFC1
NOVA2
PNMT
PRG2
PRRC2A
PTPRS
RASGRP2
RPS6KB2
SLC25A42
ZFPL1
AKT
Up-regulated genes
ACLY
ACOT13
ALDOA
APLP2
ATP5J2
BSG
CAMK1
CIB1
CLDN3
COX6A1
DDR1
ENTPD5
ERGIC3
GDE1
GUK1
H2AFY
LDHA
LSR
LTBR
MIF
MLF2
MLX
MRPL9
NDUFB7
NR2F6
NUTF2
PDLIM5
PRDX2
PRPS1
PSMB7
PSMD11
PSMD8
PSMD9
PTGES3
RAB14
RANGAP1
RNF126
SC5D
SEC63
SEPW1
SPINT2
SRSF9
TKT
TMPRSS2
TSPAN1
YWHAQ
YWHAZ
23
Supplemental Figures
(A)
(B)
pAMPK (! M)
ppppp
pAMPK (! M)
ppppp
Figure S1. The parameter sensitivity analysis. 20000 sets of parameters were generated by
randomly choosing all parameters (except for 𝛶 and 𝑘ℎ ) within the range of ±25% of the
original values. The plots show the distributions of the stable steady states for the normal cells
(panel A) and the cancer cells (panel B). The color bar on the right side represents the gene
expression level in the logarithm scale. The proportions of the ‘W’ state, ‘W/O’ state and the ‘O’
state in normal cells are 49.02%, 2.13% and 48.84% respectively (panel A). The proportions of
the ‘W’ state, ‘W/O’ state and the ‘O’ state in cancer cells are 36.16%, 13.19% and 50.42%
respectively (panel B). In future, the landscape approach (102-106) could be utilized to quantify
the stabilities of and transition processes among OXPHOS, glycolysis and hybrid metabolism
phenotypes.
24
(B)
W/O
W
pAMPK(nM)
(A)
600
O
W/O
W
W/O
O
500
O
400
W/O
300
200
100 W
O
0
0.1
0.2
0.3
−1
0.4
0.5
γ
kh (hour )
(C)
W
W
O
kh (hour −1)
HIF-1(nM)
600
W
500
400
W/O
300
200
100
0
0.1
O
0.2
0.3
−1
0.4
0.5
kh (hour )
Figure S2. (A) Two-parameter bifurcation diagram of the metabolic circuit, as the function
of 𝜸 and 𝒌𝒉 . The circuit allows different phases for different values of 𝛾 and 𝑘ℎ . Each phase
(represented by different colors) allows a different combination of co-existing metabolic
phenotypes. For example, the yellow area is the phase allowing all three possible metabolic
phenotypes (‘W’, ‘W/O’ and ‘O’). (B-C) The one-parameter bifurcation diagrams of the core
AMPK:HIF-1:ROS circuit in response to the changes in 𝑘ℎ . The y-axis is the level of pAMPK
and HIF-1 respectively. All the other notations are the same as those in Figure 4.
25
HIF− 1 downstream genes
W
W/O
O
5
0
−5
Pearson: -0.68
P < 0.001
−5
0
5
AMPK downstream genes
HIF− 1 downstream genes
(B)
(A)
W
W/O
O
5
0
−5
Pearson: -0.33
P < 0.001
−5
0
5
AMPK downstream genes
Figure S3. Evaluation of the AMPK and HIF-1 activities using TCGA data from lung
adenocarcinoma by the AMPK axis and HIF-1 axis generated from liver hepatocellular
carcinoma. (A) The anti-correlation between AMPK and HIF-1 activity in lung adenocarcinoma.
Using the RNA-seq data from lung adenocarcinoma, we calculated the AMPK and HIF-1 axes of
lung by PCA (detailed gene list is shown in Table S9 and detailed procedure in shown in SI
section 5). (B) The anti-correlation between AMPK and HIF-1 activities in lung adenocarcinoma
evaluated by the AMPK axis and HIF-1 axis generated from liver hepatocellular carcinoma.
26
HIF− 1 downstream genes
(B)
0.3
0.2
Coefficient in PC1
Coefficient in PC1
(A)
0.1
0
-0.1
-0.2
0
20
40
60
0.2
0.1
0
-0.1
-0.2
0
80
AMPK downstream gene list ID
10
20
30
40
50
60
10
5
0
−5
Pearson: -0.62
− 10 P < 0.001
− 15 − 10 − 5
(C)
0
5
AMPK downstream genes
HIF-1 downstream gene list ID
(D)
Coefficient in PC1
Coefficient in PC1
0.2
0.1
0
0.2
0.1
0
-0.1
-0.1
-0.2
0
H IF − 1 do w n s tream gen es
0.3
0.3
5
10
15
20
25
30
AMPK downstream gene list ID
-0.2
0
5
10
15
20
HIF1 downstream gene list ID
10
5
0
−5
Pearson: -0.59
− 10 P < 0.001
− 15
− 10 − 5
0
5
AMPK downstream genes
Figure S4. AMPK and HIF-1 PC1s from liver hepatocellular carcinoma. Using 84 AMPK
downstream genes and 59 HIF-1 downstream genes: (A) Corresponding coefficient of each
downstream gene of AMPK and HIF-1 in the PC1 eigenvectors. Top 40% of genes were
selected according to the absolute values (cutoff values indicated by the read dotted lines) of the
corresponding components in the PC1 eigenvector. (B) Anti-correlation between the AMPK and
HIF-1 activities. Using sorted 33 AMPK downstream genes and 23 HIF-1 downstream genes:
(C) Corresponding coefficient of each downstream gene of AMPK and HIF-1 in the PC1
eigenvectors. (D) Anti-correlation between the AMPK and HIF-1 activities.
27
0
−5
−5
0
5
Pearson: -0.22
5 P < 0.01
0
−5
−5
10
AMPK downstream genes
0
−5
− 10
Pearson: 0.08
P = 0.072
− 10
−5
0
5
(E) Prostate adenocarcinoma
5
− 15
0
5
AMPK downstream genes
10
10
Pearson: 0.36
P < 0.001
5
0
−5
− 10
− 10
−5
0
5
AMPK downstream genes
Pearson: -0.20
P < 0.01
5
0
−5
− 10
− 10
0
10
AMPK downstream genes
AMPK downstream genes
HIF− 1 downstream genes
HIF− 1 downstream genes
(D) Kidney renal clear
cell carcinoma
HIF− 1 downstream genes
Pearson: -0.63
P < 0.001
(C) Pancreatic adenocarcinoma
(F) Colorectal adenocarcinoma
HIF− 1 downstream genes
5
(B) Acute myeloid leukemia
HIF− 1 downstream genes
HIF− 1 downstream genes
(A) Stomach adenocarcinoma
10
Pearson: 0.18
P < 0.001
5
0
−5
− 10
− 10
0
10
AMPK downstream genes
Figure S5. Evaluation of the AMPK and HIF-1 activities using TCGA data from stomach
adenocarcinoma (n=265), acute myeloid leukemia (n=173), pancreatic adenocarcinoma
(n=179), kidney renal clear cell carcinoma (n=534) and prostate adenocarcinoma (n=498).
The activities of AMPK and HIF-1 were quantified by the RNA-seq data of their downstream
gene targets (See Table S9 for details). Each point in the figures represents the AMPK and HIF1 activities for one sample from the corresponding cancer type. For each data set, the Pearson
correlation coefficient was calculated. The anti-correlation of the AMPK and HIF-1 activities has
been observed in stomach adenocarcinoma (A), acute myeloid leukemia (B) and pancreatic
adenocarcinoma (C), but not so clear in clear cell renal cell carcinoma (CCRCC) (D), and there
is a positive-correlation in prostate adenocarcinoma (E). We found that most of the ccRCC
samples maintain a high HIF-1 activity, which is consistent with the experimental observation
that the upregulation of HIF-1 contributes to aerobic glycolysis in ccRCC (107). For prostate
adenocarcinoma, it is unclear why the HIF-1 activity correlates positively to the AMPK activity.
One possible explanation is that the activation of an oncogenic pathway forces the up-regulation
of both glycolysis and OXPHOS. The hypothesis might be worth to investigate in the future. For
colorectal adenocarcinoma, both AMPK and HIF-1 activity are relatively low, which is
consistent with the experimental observation that fatty acid metabolism is dominant over glucose
in colorectal adenocarcinoma (108).
28
Figure S6. Comparison of the AMPK and HIF-1 activities of liver hepatocellular
carcinoma (HCC, n=373, purple stars), clear cell renal cell carcinoma (CCRCC, n=534,
grey stars) and pancreatic adenocarcinoma (PAC, n=179, red stars). To do the comparison,
the expression of ACTB (Figure 5E), TUBB (A) and GAPDH (B) levels were rescaled to the
same level and the other gene expression were rescaled accordingly. The comparison results here
are consistent with experimental observation that glycolysis is pathognomonic in pancreatic
adenocarcinoma (PAC) given low vascularity (109) and clear cell renal cell carcinoma (CCRCC)
(110) and liver hepatocellular carcinoma (HCC) can vary metabolism phenotypes given various
microenvironment (111).
29
(A)
MYC
0.4
-5
Pearson: -0.59
P < 0.001
-10
-15 -10 -5
0
5
AMPK downstream genes
W
W/O
O
5
0
Pearson: -0.68
P < 0.001
-5
0
5
AMPK downstream genes
Breast invasive carcinoma
W
W/O
O
5
SRC liver
AKT liver
AKT
0.4
*
*
0.3
0.3
0.2
0.2
0.1
0.1
0.1
0
0
0
MYC
0.4
***
***
0.2
**
-5
0
5
AMPK downstream genes
c-SRC
0.4
***
SRC lung
AKT
***
0.4
***
0.3
0.2
0.1
0
0
MYC
0.5
0.4
***
***
0.1
0
0
SRC breastAKT
0.5
***
***
0.4
0.4
***
AKT breast
***
***
0.3
0.2
0.2
*
0
**
0.1
MYC breast c-SRC
0.6
AKT lung
0.5
**
***
0.2
0.1
0.2
Pearson: -0.33
P < 0.001
MYC lung
0.3
0.3
0.3
0
-5
c-SRC
0.4
***
***
0.2
-5
MYC liver
0.3
0
Lung adenocarcinoma
HIF-1 downstream genes
(C)
0.5
W
W/O
O
5
HIF-1 downstream genes
(B)
10
Fraction of samples with low oncogene score
HIF-1 downstream genes
Liver hepatocellular carcinoma
0.1
0
Figure S7. The fraction of samples with low oncogene score in liver hepatocellular
carcinoma (n=373) (A), lung adenocarcinoma (n=517) (B) and breast invasive carcinoma
(n=971) (C). Left panels: each point represents the AMPK and HIF-1 activities for each cancer
case. There are consistent anti-correlations between the AMPK and HIF-1 downstream gene
expressions across liver hepatocellular carcinoma (panel A), lung adenocarcinoma (panel B),
breast invasive carcinoma (panel C). For each data set, the standard k-mean analysis was applied
to group the cases into the ‘W’, ‘W/O’ and ‘O’ states. Right panels: the MYC, c-SRC and AKT
scores were applied to identify the bottom 20% of samples with low activities in these oncogenic
pathways. each panel shows the fraction of these bottom 20% samples among all the samples of
each metabolic group (‘W’, ‘W/O’ and ‘O’). Chi-Square test (See SI section 10) is used to
analyze the significance of difference among ‘W’, ‘O’ and ‘W/O’ state. ‘*’ represents P value <
0.05, ‘**’ represents P value < 0.01, ‘***’ represents P value < 0.001.
30
0.8
(B)
***
0.6
0.4
***
**
0.2
0
SRC
low RS
MYC
AKT
high RS
Fraction of samples with
high oncogene score
Fraction of samples with
low oncogene score
(A)
0.6
0.5
0.4
0.3
0.2
0.1
0
**
SRC
low RS
MYC
AKT
high RS
Figure S8. The fraction of single LUAD cells (n = 77) with low (A) and high (B) oncogene
score. The MYC, c-SRC and AKT scores were applied to identify the top 20% of single cells
with low activities (A) and high activities (B) in these oncogenic pathways. (A) and (B) show the
fraction of these top 20% cells in the high risk group and low risk group respectively (112). ChiSquare test (See SI section 10) is used to analyze the significance of difference among ‘W’, ‘O’
and ‘W/O’ state. ‘**’ represents P value < 0.01, ‘***’ represents P value < 0.001.
31
Figure S9. (A) Evaluation of the metabolism state of the bulk tumor, that was the source for
single cells analysis, using the AMPK and HIF-1 axes generated by TCGA lung adenocarcinoma
data. To do the evaluation, the expression of TUBB levels in bulk tumor were rescaled to the
same averaged level of TUBB of TCGA data. The bulk tumor for single cells has a mixed
glycolysis/OXPHOS metabolism phenotype. (B-D) Comparison of the metabolism state of bulk
cells with corresponding single cells. To do the comparison, the average levels of TUBB (B),
ACTB(C) and GAPDH (D) of the single cells were rescaled to the same levels as that in bulk
cells. LC-PT-45-Re represents the repeated experiment for single-cell isolation and RNA-seq.
The single-cell data were obtained from (112).
32
Figure S10 Evaluation of the AMPK and HIF-1 activities of single LUAD cells (n=77). Two
groups of single cells – LC-PT-45 (n=34) and LC-PT-45-Re (n=43), from the same cell source
were isolated and sequenced (112). Considering the potential batch effect, we normalized the
averaged expression of ACTB (B), GAPDH (C) and TUBB (D) to the same scale respectively
and then rescale the expression of the other genes accordingly. Consistently, we observed the
anti-correlation between the AMPK and HIF-1 activities in these single LUAD cells, as shown in
(A) (no normalization by ACTB, GAPDH or TUBB), (B) and (C). There is no significant
correlation when the averaged expression of TUBB was used to do the normalization (D).
Histogram represents the distribution of single cells projected to the first principal component of
their AMPK and HIF-1 signatures.
33
(A)
(B)
(C)
Figure S11 Fitting of the histogram of single LUAD cells with multiple Gaussian
distributions. (A) the histogram is best fitted by three Gaussian distributions – two of them have
similar mean values but distinct variances, and the other one has larger mean; (B) a possible
fitting result with two Gaussian distributions – the fitted curve matches worse to the histogram in
the region of positive PC1 values; (C) an alternative fitting result with two Gaussian distributions
– the fitted curve does not capture the long tail in the region of negative PC1 values.
34
(B)
Drug 2
Drug 2
(A)
Drug 1
Figure S12. The bifurcation diagram of the AMPK:HIF-1:ROS circuit in response to the
levels of two external signals for combinatorial therapies. (A) Two-parameter bifurcation
diagram of the circuit in response to the signal of metformin (y-axis) and hyperbaric oxygen (xaxis) therapies. The red lines and blue dashed lines with arrows illustrate three different
treatment strategies for the patients with cancer cells in the (‘W’, ‘W/O’, ‘O’) phase, and three
different treatment strategies for the patients with cancer cells in the (‘W’) phase respectively:
the line r1 or b1 is the treatment with metformin only; the line r2 or b2 is the treatment with
hyperbaric oxygen only; the line r3 or b3 is the treatment with a combination of both hyperbaric
oxygen and metformin. The administration of metformin alone could be risky for the patients in
this case, as it cannot reduce cancer metabolic plasticity. Here, metformin inhibits ETC but
induces beta-oxidation, therefore inducing high mtROS production, which drives cancer cells
into the hybrid phenotype. (B) Two-parameter bifurcation diagram of the circuit in response to
the signal of metformin (y-axis) and 3BP (x-axis) therapies. Different colors represent the
corresponding phases that allow different combinations of co-existing phenotypes (as denoted by
texts in the panels).
35
Figure S13. Assessment of different therapeutic strategies with the AMPK:HIF-1:ROS
model. (A) hyperbaric oxygen therapy; (B) 3BP; (C) metformin; (D) combined therapy of both
hyperbaric oxygen and metformin; (E) Combined therapy of both metformin and 3BP. In both D
36
and E, the two drugs are administered simultaneously. See Figure S9 for results on the treatment
where two drugs are administered alternatively in different days. In each figure, the left diagram
shows the dynamical trajectories in the phase plane, two axes are pAMPK (nM) and HIF-1(nM).
As in Figure 3, the light red line shows the nullcline of 𝑑ℎ⁄𝑑𝑡 = 0, and the light blue line shows
the nullcline of 𝑑𝐴⁄𝑑𝑡 = 0 (see SI Eq.S1). The steady states are shown as solid green circles
(stable) and empty green circles (unstable). The black, red, navy and brown lines with arrows
represent the trajectories of the treatment on the phase planes. The diagrams on the upper right
show the levels of drugs administered every day. The session of different therapies is shown in
different colors. Hyperbaric oxygen therapy is presented in light blue (panel A); 3BP is presented
in light red (panel C and E) and metformin is presented in light green (panel B D and E). The
diagrams on the lower right show the time evolutions of the levels of the activated AMPK (blue
line) and HIF-1 (red line). The dashed vertical lines separate the regions correspond to the states
‘W’ (left), ‘W/O’ (middle) and ‘O’ (right), respectively.
37
(A)
(B)
Figure S14. Assessment of two-drug therapeutic strategies using the AMPK:HIF-1:ROS
model. (A) The treatments where metformin and 3BP are administrated alternatively each day.
(B) The treatments where metformin and hyperbaric oxygen are administrated alternatively each
day. In the main text, we have shown the cases in which patients are administrated by either
single therapy or two therapies simultaneously in the same day (Fig. 6). We found that the
strategy of alternative administration is less efficient than the strategy of simultaneous
administration. As we showed in the main text (Figure S13), 3BP and hyperbaric oxygen are
less efficient in driving the phenotype from ‘W’ to ‘W/O’, and metformin is much less efficient
in driving the phenotype from the ‘W/O’ to ‘O’ compared to the therapies targeting glycolysis.
Therefore, cells stay in the hybrid ‘W/O’ state longer.
38
Figure S15. Final outcomes from different therapeutic strategies and administration
lengths. After a complete session of a treatment (x-axis) of certain length (y-axis), cells could
eventually stay in the ‘W’ state (magenta), the ‘W/O’ state (green), or the ‘O’ state (blue). There
are five groups of treatment strategies, each of which contains three different doses. From left to
right, group A: hyperbaric oxygen therapy (15%, 20%, 25%); group B: metformin-based therapy
(300μM, 350μM, 400μM); group C: 3BP-based therapy (250μM, 300μM, 350μM); group D:
combined therapy using both hyperbaric oxygen and metformin (10% and 250μM, 15% and
300μM, 20% and 350μM); group E: combined therapy using both 3BP and metformin (both are
250μM, 300μM and 350μM).
39
Reference
1.
2.
3.
4.
5.
6.
7.
8.
9.
10.
11.
12.
13.
14.
15.
16.
Hue L, Taegtmeyer H. The Randle cycle revisited: a new head for an old hat. American
Journal of Physiology-Endocrinology and Metabolism 2009;297:E578-E91.
Sambandam N, Lopaschuk GD. AMP-activated protein kinase (AMPK) control of fatty
acid and glucose metabolism in the ischemic heart. Progress in Lipid Research
2003;42:238-56.
Yamauchi T, Kamon J, Minokoshi Ya, Ito Y, Waki H, Uchida S, et al. Adiponectin
stimulates glucose utilization and fatty-acid oxidation by activating AMP-activated
protein kinase. Nature Medicine 2002;8:1288-95.
Zheng D, MacLean PS, Pohnert SC, Knight JB, Olson AL, Winder WW, et al.
Regulation of muscle GLUT-4 transcription by AMP-activated protein kinase. Journal of
Applied Physiology 2001;91:1073-83.
Lee WJ, Kim M, Park H-S, Kim HS, Jeon MJ, Oh KS, et al. AMPK activation increases
fatty acid oxidation in skeletal muscle by activating PPARα and PGC-1. Biochemical and
Biophysical Research Communications 2006;340:291-95.
Chen C, Pore N, Behrooz A, Ismail-Beigi F, Maity A. Regulation of glut1 mRNA by
hypoxia-inducible factor-1 Interaction between H-ras and hypoxia. Journal of Biological
Chemistry 2001;276:9519-25.
Ke Q, Costa M. Hypoxia-inducible factor-1 (HIF-1). Molecular Pharmacology
2006;70:1469-80.
Gwinn DM, Shackelford DB, Egan DF, Mihaylova MM, Mery A, Vasquez DS, et al.
AMPK phosphorylation of raptor mediates a metabolic checkpoint. Molecular Cell
2008;30:214-26.
Hwang AB, Ryu E-A, Artan M, Chang H-W, Kabir MH, Nam H-J, et al. Feedback
regulation via AMPK and HIF-1 mediates ROS-dependent longevity in Caenorhabditis
elegans. Proceedings of the National Academy of Sciences 2014;111:E4458-E67.
De Saedeleer CJ, Copetti T, Porporato PE, Verrax J, Feron O, Sonveaux P. Lactate
activates HIF-1 in oxidative but not in Warburg-phenotype human tumor cells. PloS One
2012;7:e46571.
Luo W, Hu H, Chang R, Zhong J, Knabel M, O'Meally R, et al. Pyruvate kinase M2 is a
PHD3-stimulated coactivator for hypoxia-inducible factor 1. Cell 2011;145:732-44.
Lu H, Forbes RA, Verma A. Hypoxia-inducible factor 1 activation by aerobic glycolysis
implicates the Warburg effect in carcinogenesis. Journal of Biological Chemistry
2002;277:23111-15.
Kirito K, Hu Y, Komatsu N. HIF-1 prevents the overproduction of mitochondrial ROS
after cytokine stimulation through induction of PDK-1. Cell Cycle 2009;8:2844-49.
Steinberg GR, Schertzer JD. AMPK promotes macrophage fatty acid oxidative
metabolism to mitigate inflammation: implications for diabetes and cardiovascular
disease. Immunology and Cell Biology 2014;92:340-45.
Rosca MG, Vazquez EJ, Chen Q, Kerner J, Kern TS, Hoppel CL. Oxidation of fatty acids
is the source of increased mitochondrial reactive oxygen species production in kidney
cortical tubules in early diabetes. Diabetes 2012;61:2074-83.
Aon MA, Stanley BA, Sivakumaran V, Kembro JM, O'Rourke B, Paolocci N, et al.
Glutathione/thioredoxin systems modulate mitochondrial H2O2 emission: an
experimental-computational study. The Journal of General Physiology 2012;139:479-91.
40
17.
18.
19.
20.
21.
22.
23.
24.
25.
26.
27.
28.
29.
30.
31.
Ameri K, Jahangiri A, Rajah AM, Tormos KV, Nagarajan R, Pekmezci M, et al.
HIGD1A Regulates Oxygen Consumption, ROS Production, and AMPK Activity during
Glucose Deprivation to Modulate Cell Survival and Tumor Growth. Cell Reports
2015;10:891-99.
Kukidome D, Nishikawa T, Sonoda K, Imoto K, Fujisawa K, Yano M, et al. Activation
of AMP-activated protein kinase reduces hyperglycemia-induced mitochondrial reactive
oxygen species production and promotes mitochondrial biogenesis in human umbilical
vein endothelial cells. Diabetes 2006;55:120-27.
Hou X, Song J, Li X-N, Zhang L, Wang X, Chen L, et al. Metformin reduces intracellular
reactive oxygen species levels by upregulating expression of the antioxidant thioredoxin
via the AMPK-FOXO3 pathway. Biochemical and Biophysical Research
Communications 2010;396:199-205.
Song P, Zou M-H. Regulation of NAD (P) H oxidases by AMPK in cardiovascular
systems. Free Radical Biology and Medicine 2012;52:1607-19.
Wang S, Zhang M, Liang B, Xu J, Xie Z, Liu C, et al. AMPKα2 Deletion Causes
Aberrant Expression and Activation of NAD (P) H Oxidase and Consequent Endothelial
Dysfunction In Vivo Role of 26S Proteasomes. Circulation Research 2010;106:1117-28.
Emerling BM, Weinberg F, Snyder C, Burgess Z, Mutlu GM, Viollet B, et al. Hypoxic
activation of AMPK is dependent on mitochondrial ROS but independent of an increase
in AMP/ATP ratio. Free Radical Biology and Medicine 2009;46:1386-91.
Li Y-N, Xi M-M, Guo Y, Hai C-X, Yang W-L, Qin X-J. NADPH oxidase-mitochondria
axis-derived ROS mediate arsenite-induced HIF-1α stabilization by inhibiting prolyl
hydroxylases activity. Toxicology Letters 2014;224:165-74.
Diebold I, Petry A, Sabrane K, Djordjevic T, Hess J, Görlach A. The HIF1 target gene
NOX2 promotes angiogenesis through urotensin-II. Journal of Cell Science
2012;125:956-64.
Lamberts RR, Onderwater G, Hamdani N, Vreden MJA, Steenhuisen J, Eringa EC, et al.
Reactive Oxygen Species–Induced Stimulation of 5′ AMP-Activated Protein Kinase
Mediates Sevoflurane-Induced Cardioprotection. Circulation 2009;120:S10-S15.
Zmijewski JW, Banerjee S, Bae H, Friggeri A, Lazarowski ER, Abraham E. Exposure to
hydrogen peroxide induces oxidation and activation of AMP-activated protein kinase.
Journal of Biological Chemistry 2010;285:33154-64.
Lu M, Jolly MK, Ben-Jacob E. Toward decoding the principles of cancer metastasis
circuits. Cancer Research 2014;74:4574-87.
Cantó C, Auwerx J. AMP-activated protein kinase and its downstream transcriptional
pathways. Cellular and Molecular Life Sciences 2010;67:3407-23.
Charos AE, Reed BD, Raha D, Szekely AM, Weissman SM, Snyder M. A highly
integrated and complex PPARGC1A transcription factor binding network in HepG2 cells.
Genome Research 2012;22:1668-79.
Impey S, McCorkle SR, Cha-Molstad H, Dwyer JM, Yochum GS, Boss JM, et al.
Defining the CREB regulon: a genome-wide analysis of transcription factor regulatory
regions. Cell 2004;119:1041-54.
Glauser DA, Schlegel W. The emerging role of FOXO transcription factors in pancreatic
β cells. Journal of Endocrinology 2007;193:195-207.
41
32.
33.
34.
35.
36.
37.
38.
39.
40.
41.
42.
43.
44.
45.
46.
47.
Benita Y, Kikuchi H, Smith AD, Zhang MQ, Chung DC, Xavier RJ. An integrative
genomics approach identifies Hypoxia Inducible Factor-1 (HIF-1)-target genes that form
the core response to hypoxia. Nucleic Acids Research 2009;37:4587-602.
Creighton CJ. Multiple oncogenic pathway signatures show coordinate expression
patterns in human prostate tumors. PloS One 2008;3:e1816.
Pan Y, Mansfield KD, Bertozzi CC, Rudenko V, Chan DA, Giaccia AJ, et al. Multiple
factors affecting cellular redox status and energy metabolism modulate hypoxia-inducible
factor prolyl hydroxylase activity in vivo and in vitro. Molecular and Cellular Biology
2007;27:912-25.
Qutub AA, Popel AS. Reactive oxygen species regulate hypoxia-inducible factor 1α
differentially in cancer and ischemia. Molecular and Cellular Biology 2008;28:5106-19.
Kalender A, Selvaraj A, Kim SY, Gulati P, Brûlé S, Viollet B, et al. Metformin,
independent of AMPK, inhibits mTORC1 in a rag GTPase-dependent manner. Cell
Metabolism 2010;11:390-401.
Cheong J-H, Park ES, Liang J, Dennison JB, Tsavachidou D, Nguyen-Charles C, et al.
Dual inhibition of tumor energy pathway by 2-deoxyglucose and metformin is effective
against a broad spectrum of preclinical cancer models. Molecular Cancer Therapeutics
2011;10:2350-62.
Porporato PE, Payen VL, Perez-Escuredo J, De Saedeleer CJ, Danhier P, Copetti T, et al.
A Mitochondrial Switch Promotes Tumor Metastasis. Cell Reports 2014;8:754-66.
LeBleu VS, O’Connell JT, Herrera KNG, Wikman H, Pantel K, Haigis MC, et al. PGC1α mediates mitochondrial biogenesis and oxidative phosphorylation in cancer cells to
promote metastasis. Nature Cell Biology 2014;16:992-1003.
Jiang L, Xiao L, Sugiura H, Huang X, Ali A, Kuro-o M, et al. Metabolic reprogramming
during TGFβ1-induced epithelial-to-mesenchymal transition. Oncogene 2014;34:3908-16.
Zhou H, Zhang B, Zheng J, Yu M, Zhou T, Zhao K, et al. The inhibition of migration and
invasion of cancer cells by graphene via the impairment of mitochondrial respiration.
Biomaterials 2014;35:1597-607.
Viale A, Pettazzoni P, Lyssiotis CA, Ying H, Sánchez N, Marchesini M, et al. Oncogene
Ablation-resistant Pancreatic Cancer Cells Depend on Mitochondrial Function. Nature
2014;514:628-32.
Roesch A, Vultur A, Bogeski I, Wang H, Zimmermann KM, Speicher D, et al.
Overcoming intrinsic multidrug resistance in melanoma by blocking the mitochondrial
respiratory chain of slow-cycling JARID1B high cells. Cancer Cell 2013;23:811-25.
Lu C-L, Qin L, Liu H-C, Candas D, Fan M, Li JJ. Tumor Cells Switch to Mitochondrial
Oxidative Phosphorylation under Radiation via mTOR-Mediated Hexokinase II
Inhibition-A Warburg-Reversing Effect. PloS One 2015;10.
Xu Q, Biener-Ramanujan E, Yang W, Ramanujan VK. Targeting metabolic plasticity in
breast cancer cells via mitochondrial complex I modulation. Breast Cancer Research and
Treatment 2015;150:43-56.
Harada H, Itasaka S, Kizaka-Kondoh S, Shibuya K, Morinibu A, Shinomiya K, et al. The
Akt/mTOR pathway assures the synthesis of HIF-1α protein in a glucose-and
reoxygenation-dependent manner in irradiated tumors. Journal of Biological Chemistry
2009;284:5332-42.
Xiao B, Sanders MJ, Underwood E, Heath R, Mayer FV, Carmena D, et al. Structure of
mammalian AMPK and its regulation by ADP. Nature 2011;472:230-33.
42
48.
49.
50.
51.
52.
53.
54.
55.
56.
57.
58.
59.
60.
61.
62.
Li X-N, Song J, Zhang L, LeMaire SA, Hou X, Zhang C, et al. Activation of the AMPKFOXO3 pathway reduces fatty acid–induced increase in intracellular reactive oxygen
species by upregulating thioredoxin. Diabetes 2009;58:2246-57.
Zhao T, Zhu Y, Morinibu A, Kobayashi M, Shinomiya K, Itasaka S, et al. HIF-1mediated metabolic reprogramming reduces ROS levels and facilitates the metastatic
colonization of cancers in lungs. Scientific Reports 2014;4:3793.
Yuan G, Khan SA, Luo W, Nanduri J, Semenza GL, Prabhakar NR. Hypoxia‐inducible
factor 1 mediates increased expression of NADPH oxidase‐2 in response to intermittent
hypoxia. Journal of Cellular Physiology 2011;226:2925-33.
Zhang M, Dong Y, Xu J, Xie Z, Wu Y, Song P, et al. Thromboxane receptor activates the
AMP-activated protein kinase in vascular smooth muscle cells via hydrogen peroxide.
Circulation Research 2008;102:328-37.
Ma J, Zhang Q, Chen S, Fang B, Yang Q, Chen C, et al. Mitochondrial dysfunction
promotes breast cancer cell migration and invasion through HIF1α accumulation via
increased production of reactive oxygen species. PloS One 2013;8:e69485.
Brunelle JK, Bell EL, Quesada NM, Vercauteren K, Tiranti V, Zeviani M, et al. Oxygen
sensing requires mitochondrial ROS but not oxidative phosphorylation. Cell Metabolism
2005;1:409-14.
Zhu G-h, Huang C, Feng Z-z, Lv X-h, Qiu Z-j. Hypoxia-induced snail expression through
transcriptional regulation by HIF-1α in pancreatic cancer cells. Digestive Diseases and
Sciences 2013;58:3503-15.
Zhang L, Huang G, Li X, Zhang Y, Jiang Y, Shen J, et al. Hypoxia induces epithelialmesenchymal transition via activation of SNAI1 by hypoxia-inducible factor-1α in
hepatocellular carcinoma. BMC Cancer 2013;13:108.
Yang M-H, Wu M-Z, Chiou S-H, Chen P-M, Chang S-Y, Liu C-J, et al. Direct regulation
of TWIST by HIF-1α promotes metastasis. Nature cell biology 2008;10:295-305.
Mishra R, Cool BL, Laderoute KR, Foretz M, Viollet B, Simonson MS. AMP-activated
protein kinase inhibits transforming growth factor-β-induced Smad3-dependent
transcription and myofibroblast transdifferentiation. Journal of Biological Chemistry
2008;283:10461-69.
Lim JY, Oh MA, Kim WH, Sohn HY, Park SI. AMP‐activated protein kinase inhibits
TGF‐β‐induced fibrogenic responses of hepatic stellate cells by targeting
transcriptional coactivator p300. Journal of Cellular Physiology 2012;227:1081-89.
Jin H, Hong S, Woo S, Lee J, Choe T, Kim E, et al. Silencing of Twist1 sensitizes
NSCLC cells to cisplatin via AMPK-activated mTOR inhibition. Cell Death & Disease
2012;3:e319.
Faubert B, Boily G, Izreig S, Griss T, Samborska B, Dong Z, et al. AMPK is a negative
regulator of the Warburg effect and suppresses tumor growth in vivo. Cell Metabolism
2013;17:113-24.
Doe MR, Ascano JM, Kaur M, Cole MD. Myc posttranscriptionally induces HIF1 protein
and target gene expression in normal and cancer cells. Cancer Research 2012;72:949-57.
Edmunds LR, Sharma L, Wang H, Kang A, d’Souza S, Lu J, et al. C-Myc and AMPK
control cellular energy levels by cooperatively regulating mitochondrial structure and
function. PloS One 2015;10:e0134049.
43
63.
64.
65.
66.
67.
68.
69.
70.
71.
72.
73.
74.
75.
76.
77.
Lee H-Y, Lee T, Lee N, Yang EG, Lee C, Lee J, et al. Src activates HIF-1α not through
direct phosphorylation of HIF-1α-specific prolyl-4 hydroxylase 2 but through activation
of the NADPH oxidase/Rac pathway. Carcinogenesis 2011;32:703-12.
Mizrachy-Schwartz S, Cohen N, Klein S, Kravchenko-Balasha N, Levitzki A. Upregulation of AMP-activated protein kinase in cancer cell lines is mediated through c-Src
activation. Journal of Biological Chemistry 2011;286:15268-77.
Lim J-H, Lee E-S, You H-J, Lee JW, Park J-W, Chun Y-S. Ras-dependent induction of
HIF-1α785 via the Raf/MEK/ERK pathway: a novel mechanism of Ras-mediated tumor
promotion. Oncogene 2004;23:9427-31.
Mihaylova MM, Shaw RJ. The AMPK signalling pathway coordinates cell growth,
autophagy and metabolism. Nature Cell Biology 2011;13:1016-23.
Hubbard PA, Moody CL, Murali R. Allosteric modulation of Ras and the
PI3K/AKT/mTOR pathway: emerging therapeutic opportunities. Frontiers in Physiology
2014;5.
Mulukutla BC, Khan S, Lange A, Hu W-S. Glucose metabolism in mammalian cell
culture: new insights for tweaking vintage pathways. Trends in Biotechnology
2010;28:476-84.
Chua YL, Dufour E, Dassa EP, Rustin P, Jacobs HT, Taylor CT, et al. Stabilization of
hypoxia-inducible factor-1α protein in hypoxia occurs independently of mitochondrial
reactive oxygen species production. Journal of Biological Chemistry 2010;285:31277-84.
Antunes F, Cadenas E. Estimation of H2O2 gradients across biomembranes. FEBS
Letters 2000;475:121-26.
Qing G, Skuli N, Mayes PA, Pawel B, Martinez D, Maris JM, et al. Combinatorial
regulation of neuroblastoma tumor progression by N-Myc and hypoxia inducible factor
HIF-1α. Cancer Research 2010;70:10351-61.
Lehman JJ, Barger PM, Kovacs A, Saffitz JE, Medeiros DM, Kelly DP. Peroxisome
proliferator–activated receptor γ coactivator-1 promotes cardiac mitochondrial biogenesis.
The Journal of Clinical Investigation 2000;106:847-56.
Portilla D, Dai G, Mcclure T, Bates L, Kurten R, Megyesi J, et al. Alterations of
PPARalpha and its coactivator PGC-1 in cisplatin-induced acute renal failure. Kidney
International 2002;62:1208-18.
Fan C-Y, Pan J, Usuda N, Yeldandi AV, Rao MS, Reddy JK. Steatohepatitis,
spontaneous peroxisome proliferation and liver tumors in mice lacking peroxisomal fatty
acyl-CoA oxidase. Implications for peroxisome proliferator-activated receptor alpha
natural ligand metabolism. Journal of Biological Chemistry 1998;273:15639-45.
Rhee J, Ge H, Yang W, Fan M, Handschin C, Cooper M, et al. Partnership of PGC-1α
and HNF4α in the regulation of lipoprotein metabolism. Journal of Biological Chemistry
2006;281:14683-90.
Ma K, Zhang Y, Elam MB, Cook GA, Park EA. Cloning of the rat pyruvate
dehydrogenase kinase 4 gene promoter: activation of pyruvate dehydrogenase kinase 4 by
the peroxisome proliferator-activated receptor gamma coactivator. Journal of Biological
Chemistry 2005;280:29525-32.
Vega RB, Huss JM, Kelly DP. The coactivator PGC-1 cooperates with peroxisome
proliferator-activated receptor α in transcriptional control of nuclear genes encoding
mitochondrial fatty acid oxidation enzymes. Molecular and Cellular Biology
2000;20:1868-76.
44
78.
79.
80.
81.
82.
83.
84.
85.
86.
87.
88.
89.
90.
91.
92.
93.
Wang V, Davis DA, Haque M, Huang LE, Yarchoan R. Differential gene up-regulation
by hypoxia-inducible factor-1α and hypoxia-inducible factor-2α in HEK293T cells.
Cancer Research 2005;65:3299-306.
Lin J, Tarr PT, Yang R, Rhee J, Puigserver P, Newgard CB, et al. PGC-1β in the
regulation of hepatic glucose and energy metabolism. Journal of Biological Chemistry
2003;278:30843-48.
Caniggia I, Mostachfi H, Winter J, Gassmann M, Lye SJ, Kuliszewski M, et al. Hypoxiainducible factor-1 mediates the biological effects of oxygen on human trophoblast
differentiation through TGFβ 3. The Journal of Clinical Investigation 2000;105:577-87.
Li Z, Wang D, Na X, Schoen SR, Messing EM, Wu G. The VHL protein recruits a novel
KRAB‐A domain protein to repress HIF‐1α transcriptional activity. The EMBO
Journal 2003;22:1857-67.
Lam SH, Gong Z. Modeling liver cancer using zebrafish: a comparative oncogenomics
approach. Cell Cycle 2006;5:573-77.
Lee H, Paik S-G. Regulation of BNIP3 in normal and cancer cells. Molecules and Cells
2006;21:1.
Essaghir A, Dif N, Marbehant CY, Coffer PJ, Demoulin J-B. The transcription of FOXO
genes is stimulated by FOXO3 and repressed by growth factors. Journal of Biological
Chemistry 2009;284:10334-42.
Kayampilly PP, Menon K. AMPK activation by dihydrotestosterone reduces FSHstimulated cell proliferation in rat granulosa cells by inhibiting ERK signaling pathway.
Endocrinology 2012;153:2831-38.
Morizane Y, Thanos A, Takeuchi K, Murakami Y, Kayama M, Trichonas G, et al. AMPactivated protein kinase suppresses matrix metalloproteinase-9 expression in mouse
embryonic fibroblasts. Journal of Biological Chemistry 2011;286:16030-38.
Lee Y-S, Silva AJ. The molecular and cellular biology of enhanced cognition. Nature
Reviews Neuroscience 2009;10:126-40.
Kim S-J, Lee J-H, Chung H-S, Song J-H, Ha J, Bae H. Neuroprotective effects of ampactivated protein kinase on scopolamine induced memory impairment. The Korean
Journal of Physiology & Pharmacology 2013;17:331-38.
Jose C, Hébert-Chatelain E, Bellance N, Larendra A, Su M, Nouette-Gaulain K, et al.
AICAR inhibits cancer cell growth and triggers cell-type distinct effects on OXPHOS
biogenesis, oxidative stress and Akt activation. Biochimica et Biophysica Acta (BBA)Bioenergetics 2011;1807:707-18.
Hirsch HA, Iliopoulos D, Tsichlis PN, Struhl K. Metformin selectively targets cancer
stem cells, and acts together with chemotherapy to block tumor growth and prolong
remission. Cancer Research 2009;69:7507-11.
Haklai R, Elad-Sfadia G, Egozi Y, Kloog Y. Orally administered FTS (salirasib) inhibits
human pancreatic tumor growth in nude mice. Cancer Chemotherapy and Pharmacology
2008;61:89-96.
Wang Q, Liang B, Shirwany NA, Zou M-H. 2-Deoxy-D-glucose treatment of endothelial
cells induces autophagy by reactive oxygen species-mediated activation of the AMPactivated protein kinase. PloS One 2011;6:e17234.
Ihrlund LS, Hernlund E, Khan O, Shoshan MC. 3-Bromopyruvate as inhibitor of tumour
cell energy metabolism and chemopotentiator of platinum drugs. Molecular Oncology
2008;2:94-101.
45
94.
95.
96.
97.
98.
99.
100.
101.
102.
103.
104.
105.
106.
107.
108.
109.
110.
111.
Ko YH, Pedersen PL, Geschwind J. Glucose catabolism in the rabbit VX2 tumor model
for liver cancer: characterization and targeting hexokinase. Cancer Letters 2001;173:8391.
Ko YH, Smith BL, Wang Y, Pomper MG, Rini DA, Torbenson MS, et al. Advanced
cancers: eradication in all cases using 3-bromopyruvate therapy to deplete ATP.
Biochemical and Biophysical Research Communications 2004;324:269-75.
Han YH, Park WH. Tiron, a ROS scavenger, protects human lung cancer Calu-6 cells
against antimycin A-induced cell death. Oncology Reports 2009;21:253-61.
Goldberg L, Israeli R, Kloog Y. FTS and 2-DG induce pancreatic cancer cell death and
tumor shrinkage in mice. Cell Death & Disease 2012;3:e284.
Sahra IB, Laurent K, Giuliano S, Larbret F, Ponzio G, Gounon P, et al. Targeting cancer
cell metabolism: the combination of metformin and 2-deoxyglucose induces p53dependent apoptosis in prostate cancer cells. Cancer Research 2010;70:2465-75.
Jiang B-H, Semenza GL, Bauer C, Marti HH. Hypoxia-inducible factor 1 levels vary
exponentially over a physiologically relevant range of O2 tension. American Journal of
Physiology-Cell Physiology 1996;271:C1172-C80.
D'Ippolito G, Diabira S, Howard GA, Roos BA, Schiller PC. Low oxygen tension inhibits
osteogenic differentiation and enhances stemness of human MIAMI cells. Bone
2006;39:513-22.
Hawley SA, Gadalla AE, Olsen GS, Hardie DG. The antidiabetic drug metformin
activates the AMP-activated protein kinase cascade via an adenine nucleotideindependent mechanism. Diabetes 2002;51:2420-25.
Yu C, Wang J. A physical mechanism and global quantification of breast cancer. PloS
One 2016;11:e0157422.
Chen C, Wang J. A physical mechanism of cancer heterogeneity. Scientific Reports
2016;6:20679.
Lu M, Onuchic J, Ben-Jacob E. Construction of an effective landscape for multistate
genetic switches. Physical Review Letters 2014;113:078102.
Zhou P, Li T. Construction of the landscape for multi-stable systems: Potential landscape,
quasi-potential, A-type integral and beyond. The Journal of Chemical Physics
2016;144:094109.
Ao P, Kwon C, Qian H. On the existence of potential landscape in the evolution of
complex systems. arXiv preprint q-bio/0703044 2007.
Shuch B, Linehan WM, Srinivasan R. Aerobic glycolysis: a novel target in kidney cancer.
Expert review of Anticancer Therapy 2013;13:711-19.
Liu Y, Zuckier LS, Ghesani NV. Dominant uptake of fatty acid over glucose by prostate
cells: a potential new diagnostic and therapeutic approach. Anticancer Research
2010;30:369-74.
Hosoki T. Dynamic CT of pancreatic tumors. American Journal of Roentgenology
1983;140:959-65.
Pinthus JH, Whelan KF, Gallino D, Lu J-P, Rothschild N. Metabolic features of clear-cell
renal cell carcinoma: mechanisms and clinical implications. Canadian Urological
Association Journal 2011;5:274-82.
Lee M, Yoon J-H. Metabolic interplay between glycolysis and mitochondrial oxidation:
The reverse Warburg effect and its therapeutic implication. World Journal of Biological
Chemistry 2015;6:148.
46
112.
Kim K-T, Lee HW, Lee H-O, Kim SC, Seo YJ, Chung W, et al. Single-cell mRNA
sequencing identifies subclonal heterogeneity in anti-cancer drug responses of lung
adenocarcinoma cells. Genome Biology 2015;16:127.
47