Survey
* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
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-1Ξ± 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 PPARπΆ: 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 π Μππ‘ = ππππ‘ π» π β (π΄ππ , π΄0ππ , πππ , πππ )πΆπππππ (πΆ, ππ , β, ββπ , πβππ‘ , π΄, π΄0ππππ‘ , πβππ‘ ) β ππππ‘ π ππ‘ ππ‘ ππ‘ (eq. S1.1) 0 π Μπππ₯ = πππππ₯ πΆπ ππππ (π0 , β, ββπ , πβπππ₯ , π1 , π΄, π΄0 , π2 , πππππ₯ ) β πππππ₯ π πππ₯ πππ₯ πππ₯ (eq. S1.2) π = π ππ‘ + π πππ₯ (eq. S1.3) 0 0 π΄Μ = ππ π» π + (π ππ , π ππ , πππ , πππ )π» π β (ββπ , ββπ , πβπ , πβπ )π» π β (π΄ππ , π΄0ππ , πππ , πππ ) β ππ π΄ (eq. S1.4) 0 0 βΜ = πβ π» π β (π΄πβ , π΄0πβ , ππβ , ππβ ) β πβ β β β π» π β (βββ , βββ , πββ , πββ )π» π β (π πβ , π πβ , ππβ , ππβ ) (eq. S1.5) Here, we followed the computational modeling approach presented by Lu et al. (27). In the equations, π ππ‘ , π πππ₯ , π΄ , β represent the levels of mtROS, noxROS, pAMPK and HIF-1 respectively; ππππ‘ , πππππ₯ , ππ , πβ are the synthesis rate of mtROS, noxROS, pAMPK and HIF-1 respectively; ππππ‘ , πππππ₯ , ππ , πβ represent the degradation rates of mtROS, noxROS, pAMPK and HIF-1 respectively. We used the shifted Hill-function π» π + for excitatory regulations, π» π β for inhibitory regulations, and πΆ ππππ for competitive regulations by two genes/metabolites. In the equations, the shifted Hill-function π» π and the competitive regulations πΆ ππππ are expressed by the following equations: π» π (π , π0 , π, π) = , where { π π ) π0 π π 1+( ) π0 1+π( π» π β β‘ π» π π» π + β‘ π» π (eq. S1.6) ππ π < 1 ππ π > 1 2 πΎ(ππ +( 0 π΄ 0 πΆπ ππππ (πΎ, ππ , β, ββπ , πβππ‘ , π΄, π΄0ππππ‘ , ππππ‘ ) = ππ‘ ππ‘ 1+( 0 β β βπππ‘ 0 πΆπ ππππ (π0 , β, ββπ , πβπππ₯ , π1 , π΄, π΄0 , π2 , πππππ₯ ) = πππ₯ πππ₯ ππππ‘ π΄ ) ππππ‘ πβ ππ‘ ) +( 0 π΄ ) (eq. S1.7) ππππ‘ π΄ ) ππππ‘ π0 +ππβπππ₯ ( 0 β β ) βππππ₯ 1+( 0 β β ) πβ πππ₯ πβ πππ₯ βππππ₯ +ππππππ₯ ( 0 π΄ +( 0 π΄ π΄ π΄ πππππ₯ ) πππππ₯ πππππ₯ ) πππππ₯ (eq. S1.8) As described in Eq. S1, π» π + represents excitatory regulations and π» π β represents inhibitory regulations, which correspond to π > 1 and π < 1 respectively. In Eq.S1.6, π represents the concentration of the protein or metabolite, such as pAMPK, HIF-1, and ROS (π΄, β and π in Eq. S1). π0 represents the threshold of the Hill function, π represents the maximum fold change of the corresponding protein or metabolites, and π 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 - π» π + (πππΆ, πππΆππππ , ππππΆ,π , ππππΆ,π ) and π» π β (πππΆ , πππΆβππβ1 , ππππΆ,β , ππππΆ,β ) 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 π΄Μ = ππ (1+π πππΆ,π 0 0 ) π» π + (πππΆ, πππΆππππ , ππππΆ,π , ππππΆ,π )(π ππ , π ππ , πππ , πππ ) β 0 π» π β (ββπ , ββπ , πβπ , πβπ )(π΄ππ , π΄0ππ , πππ , πππ ) β ππ π΄ 2 βΜ = πβ π» π β (π΄πβ , π΄0πβ , ππβ , ππβ ) β πβ β(1+π πππΆ,β (eq. S1.9) 0 )π» π β (πππΆ , πππΆβππβ1 , ππππΆ,β , ππππΆ,β ) β 0 0 π» π β (βββ , βββ , πββ , πββ )π» π β (π πβ , π πβ , ππβ , ππβ ) (eq. S1.10) 0 We rescaled the two shifted Hill functions - π» π + (πππΆ, πππΆππππ , ππππΆ,π , ππππΆ,π ) and 0 π» π β (πππΆ , πππΆβππβ1 , ππππΆ,β , ππππΆ,β ) by (1+π 2 πππΆ,π ) and (1+π 2 πππΆ,β ), so that when the level of 0 0 MYC (πππΆ) is equal to the thresholds (πππΆππππ , πππΆβππβ1 ), 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 - π» π + (ππ πΆ, ππ πΆππππ , πππ πΆ,π , πππ πΆ,π ) and π» π + (ππ πΆ, ππ πΆπππ₯ , πππ πΆ,πππ₯ , πππ πΆ,πππ₯ ) 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 π΄Μ = ππ (1+π ππ πΆ,π 0 0 ) π» π + (ππ πΆ, ππ πΆππππ , πππ πΆ,π , πππ πΆ,π )(π ππ , π ππ , πππ , πππ ) β 0 π» π β (ββπ , ββπ , πβπ , πβπ )π» π β (π΄ππ , π΄0ππ , πππ , πππ ) β ππ π΄ 2 0 π Μπππ₯ = πππππ₯ (1+π )π» π + (ππ πΆ, ππ πΆπππ₯ , πππ πΆ,πππ₯ , πππ πΆ,πππ₯ ) β (eq. S1.11) ππ πΆ,πππ₯ 0 πΆπ ππππ (π0 , β, ββπ , πβπππ₯ , π1 , π΄, π΄0 , π2 , πππππ₯ ) πππ₯ πππ₯ β πππππ₯ π πππ₯ (eq. S1.12) 0 We rescaled the two shifted Hill functions - π» π + (ππ πΆ, ππ πΆππππ , πππ πΆ,π , πππ πΆ,π ) and 2 0 π» π + (ππ πΆ, ππ πΆπππ₯ , πππ πΆ,πππ₯ , πππ πΆ,πππ₯ ) by (1+π ππ πΆ,π ) and (1+π 2 ππ πΆ,πππ₯ ) respectively, so that when 0 0 the level of SRC (ππ πΆ ) is equal to the thresholds (ππ πΆππππ , ππ πΆπππ₯ ), 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). πΜ π = πππ (1+π 2 π π΄π,ππ 0 )π» π + (π π΄π, π π΄πππ , ππ π΄π,ππ , ππ π΄π,ππ )π» π + (π΄, π΄0π,ππ , ππ,ππ , ππ,ππ ) β πππ π π (eq. S1.13) When considering the effects of RAS, the rate equations for AMPK and HIF-1 are modified as follows: 2 π΄Μ = ππ (1+π π π΄π,π 0 0 )π» π + (π π΄π, π π΄πππππ , ππ π΄π,π , ππ π΄π,π )π» π + (π ππ , π ππ , πππ , πππ ) β 0 π» π β (ββπ , ββπ , πβπ , πβπ )π» π β (π΄ππ , π΄0ππ , πππ , πππ ) β ππ π΄ 2 βΜ = πβ (1+π π π΄π,β (eq. S1.14) 0 )π» π + (π π΄π, π π΄πβππβ1 , ππ π΄π,β , ππ π΄π,β )π» π + (π π , π π 0πβ , πππ,β , πππ ,β ) β 0 0 πβ β β π» π β (βββ , βββ , πββ , πββ )π» π β (π πβ , π πβ , ππβ , ππβ ) (eq. S1.15) 4 0 We rescaled the three shifted Hill functions - π» π + (π π΄π, π π΄πππ , ππ π΄π,ππ , ππ π΄π,ππ ) , 0 π» π + (π π΄π, π π΄πππππ , ππ π΄π,π , ππ π΄π,π ) and 2 (1+Ξ» RAS,a ) and (1+Ξ» 2 RAS,nox π» π + (π π΄π, π π΄πβ0 , ππ π΄π,β , ππ π΄π,β ) by (1+π 2 π π΄π,ππ ), ) respectively, so that when the level of RAS (π π΄π) is equal to the 0 0 thresholds (π π΄πππππ , π π΄πβππβ1 ), 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 β PGC1πΌ, 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. π₯β Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ πππ2 (π₯) β πππ 2 (π₯) , π(πππ2 (π₯)) Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ where π₯ represents the expression for the gene, πππ 2 (π₯) is the mean of the log2 transformed value and π(πππ2 (π₯)) 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 Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ πππ (π₯ )βπππ 2 (π₯π )π€ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ deviation, i.e. π₯π β 2 π , where π₯ represents the expression for the gene, πππ 2 (π₯π )π€ ππ€ (πππ2 (π₯π )) is the weighted mean of the log2 transformed value and ππ€ (πππ2 (π₯π )) is the weighted standard βπ π€ πππ (π₯ ) deviation of the log2 transformed value, i.e. Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ πππ2 (π₯π )π€ = π=1βππ π€2 π , ππ€ (πππ2 (π₯π )) = Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ 2 βπ π=1 π€π (πππ2 (π₯π )βπππ2 (π₯π )π€ ) βπ π=1 π€π π=1 π 1 , where π€π is the weight factor, π€π = π, and π 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 β 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-1Ξ± 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: ππ+ β ππβ O2 +HIF-1 β HO ππβ ππ + β HR β ππβ ROS+HIF-1 ππ β [HIF-1]+[HO]+[HR] = 1 , where HO represents the HIF-1 that binds to oxygen, and HR represents the HIF-1 that binds to π ROS. Here, we assume the sum of the HIF-1 related products is a constant. By assuming ππβ = [π20 ]ππβ and ππ β ππ + π+ = [π 0 ]ππβ at equilibrium, we obtain ( ππβ [π2 ] π20 ) = [π»π] [HIF-1] = [π»π ] [HIF-1] ππβ [π ] ( ) π 0 [π ] ππβ Therefore, [HIF-1] + ( π20 ) 2 [π ] ππβ [HIF-1] + ( π ) 0 [HIF-1] = 1. Hence, under the influence of oxygen, the dynamics of HIF-1 levels can be expressed as: π[HIF-1] ππ‘ = βπβ πΆ(π1 , π2 , π3 , π , π, π β0 , πβ0 , ππβ , ππβ )[HIF-1] [π ] where πΆ(π1 , π2 , π3 , π , π, π β0 , πβ0 , ππβ , ππβ , ) = ππβ π1 +π2 ( 0 ) π β [π ] 1+( 0 ) π β , ππβ [π ] +π3 ( 02 ) (eq. S2.1) ππβ πβ ππβ [π ] +( 02 ) πβ , πβ 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 βΜ = πβ π» π β (π΄πβ , π΄0πβ , ππβ , ππβ ) β πβ β β β π» π β (βββ , βββ , πββ , πββ ) β πΌ β 0 0 πΆ(π1 , π2 , π3 , π , π, π β , πβ , ππβ , ππβ ) (eq. S2.2) 8 Here, Ξ± is a constant parameter, which is chosen so that Eq. S2.2 is equivalent to Eq.S1.5. 0 Therefore, πΌ β πΆ(π1 , π2 , π3 , π , π, π β0 , πβ0 , ππβ , ππβ ) = π» π β (π πβ , π πβ , ππβ , ππβ ) for any level of ROS. As a result, there are only two non-redundant parameters among πΌ, π1 , π2 and π3 . 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 π + π + Hill functions π»ππ to model the pAMPK generation and π»πβ 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. (ππππ‘ ) πππ‘ = ππππ‘ + πΌπππ‘ β [πππ‘ππππππ] (eq. S3) Here, (ππππ‘ ) is the generation rate of mtROS under the influence of metformin, and πΌπππ‘ is πππ‘ 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 π΄Μ = ππ π» π + (π ππ , π ππ , πππ , πππ )π» π β (ββπ , ββπ , πβπ , πβπ ) π β (π΄ 0 π + 0 βπ» ππ , π΄ππ , πππ , πππ )π» (πππ , πππ , πππ , πππ ) β ππ π΄ (eq. S4) 0 βΜ = πβ π» π β (π΄πβ , π΄0πβ , ππβ , ππβ )π» π β (ππβ , ππβ , ππβ , ππβ ) β πβ β β β 0 0 π β (β π β π» ββ , βββ , πββ , πββ )π» (π πβ , π πβ , ππβ , ππβ ) (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 (πββ )3π΅π = πββ + πΌ3π΅π β [3π΅π] (eq. S6) Here, (πββ )3π΅π is the maximum fold change due to HIF-1 self-activation under the influence of 3BP (or 2DG), and πΌ3π΅π 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 (βWβ, βW/Oβ and βOβ). To evaluate the statistical significance of the comparison, we constructed a 2x2 contingency table for any pair of metabolic groups (βWβ versus βW/Oβ, βWβ versus βOβ and βW/Oβ versus βOβ). 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-Ξ² 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-Ξ² 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 PKCΞ±, PLCΞ³, 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 ππ 30 nM/h Production rate of AMPK ππ 0.2 ββ1 Degradation rate of AMPK π¨πππ 350 nM Threshold for self-inhibition (19) π¨πππ 250 nM Threshold for HIF-1 inhibition (19) π¨πππ 350 nM Threshold for mtROS inhibition (19) π¨πππππ 150 nM Threshold for mtROS activation (19) π¨ππππππ 150 nM Threshold for noxROS inhibition (19) πππ 0.2 - Fold change for self-inhibition πππ 0.1 - Fold change for HIF-1 inhibition πππβ 0.25 - Fold change for mtROS inhibition Ξ³* 3 - Fold change for mtROS activation ππ 0.2 - Fold change for noxROS inhibition πππ 2 - Hill coefficient for self-inhibition πππ 1 - Hill coefficient for HIF-1 inhibition πππβ 2 - Hill coefficient for mtROS inhibition πππ_ππ 4 - Hill coefficient for mtROS activation πππ_πππ 2 - Hill coefficient for noxROS inhibition ππ 15 nM/h Production rate of HIF-1 (69) ππ * 0.25 ββ1 Degradation rate of HIF-1 (69) ππππ 80 nM Threshold for HIF-1 self-activation (35) ππππ 250 nM Threshold for AMPK inhibition (35) AMPK (9,48) HIF-1 14 ππππππ 200 nM Threshold for mtROS inhibition (35) πππππππ 250 nM Threshold for noxROS activation (35) πππ 0.1 - Fold change for HIF-1 self-activation πππ 0.1 - Fold change for AMPK inhibition ππ 5 - Fold change for noxROS activation πππ 4 - Hill coefficient for HIF-1 self-activation πππ 1 - Hill coefficient for AMPK inhibition ππ_ππ 2 - Hill coefficient for mtROS inhibition ππ_πππ 2 - Hill coefficient for noxROS inhibition ππππ 150 πM/min Production rate of mitochondrial ROS (35,70) πππππ 40 πM/min Production rate of NOX derived ROS (35,70) ππ 0.2 - ππππ 5.0 /min Degradation rate of mitochondrial ROS (35,70) πππππ 5.0 /min Degradation rate of NOX derived ROS (35,70) πΉπππ 100 πM Threshold for AMPK activation (26,35,70) πΉπππ 300 πM Threshold for HIF-1 activation (26,35,70) πππ * 8 - Fold change for AMPK activation πππ 0.2 - Fold change for HIF-1 activation πππ 4 - Hill coefficient for AMPK activation πππ 4 - Hill coefficient for HIF-1 activation ROS Basal cytosol ROS (26) * πΎ and πβ 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 π΄ππͺπππππ 500 nM ππ΄ππͺ,π 2 - Fold change for MYC regulation on AMPK production ππ΄ππͺ,π 3 - Hill coefficient for MYC regulation on AMPK production Threshold for MYC regulation on AMPK production MYC stabilization on HIF-1 π΄ππͺππππβπ 500 nM ππ΄ππͺ,π 0.2 - Fold change for MYC regulation of HIF-1 degradation ππ΄ππͺ,π 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 πΊπΉπͺππππ 500 nM ππΊπΉπͺ,πππ 6 - Fold change for c-SRC regulation on noxROS production ππΊπΉπͺ,πππ 3 - Hill coefficient for c-SRC regulation on noxROS production Threshold for c-SRC regulation on noxROS production c-SRC activation on AMPK πΊπΉπͺπππππ 500 nM ππΊπΉπͺ,π 3 - Fold change for c-SRC regulation on AMPK production ππΊπΉπͺ,π 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 ππ are taken from Table S3. Parameters Value Description Unit RAS activation on AMPK πΉπ¨πΊπππππ 500 nM ππΉπ¨πΊ,π 2 - Fold change for RAS regulation on AMPK production ππΉπ¨πΊ,π 3 - Hill coefficient for RAS regulation on AMPK production Threshold for RAS regulation on AMPK production RAS activation on HIF-1 πΉπ¨πΊππππβπ 500 nM ππΉπ¨πΊ,π 2.5 - Fold change for RAS regulation on HIF-1 production ππΉπ¨πΊ,π 3 - Hill coefficient for RAS regulation on HIF-1 production Threshold for RAS regulation on HIF-1 production RAS activation on mTOR πΉπ¨πΊπππ» 500 nM ππΉπ¨πΊ,ππ» 2 - Fold change for RAS regulation on mTOR production ππΉπ¨πΊ,ππ» 1 - Hill coefficient for RAS regulation on mTOR production πππ» 100 - Production rate of mTOR πππ» 0.5 ββ1 Degradation rate of mTOR 8 nM/h Threshold for RAS regulation on mTOR production mTOR HIF-1 ππ Production rate of HIF-1 (69) AMPK inhibition on mTOR π¨ππ,ππ» 150 nM ππ,ππ» 0.3 - Fold change for AMPK regulation on mTOR production ππ,ππ» 3 - Hill coefficient for AMPK regulation on mTOR production Threshold for AMPK regulation on mTOR production mTOR activation on HIF-1 ππ» ππ 150 nM πππ»,π 2 - Fold change for mTOR regulation of HIF-1 production πππ»,π 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 ππ 0.2 - HIF-1 fold change due to ROS (99) ππ 1.8 - HIF-1 fold change due to oxygen (35,99) πππ 2 - Hill coefficient for HIF-1 inhibition πΆππ 5 % Threshold for AMPK activation (69,99,100) Metformin πππ 2 - Fold change for AMPK activation (19) πππ 0.5 - Fold change for HIF-1 inhibition (36) πππ 2 - Hill coefficient for AMPK activation πππ 2 - Hill coefficient for HIF-1 inhibition π΄πππ 300 ΞΌM Threshold for AMPK activation (19,36,37,98,101) π΄πππ 300 ΞΌM Threshold for HIF-1 inhibition (19,36,37,98,101) πΆπππ 0.025 /ΞΌM Influence on mtROS maximum fold change 3BP πΆππ©π· 0.0005 /ΞΌM 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 πΆ and πβ ) 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 βWβ state, βW/Oβ state and the βOβ state in normal cells are 49.02%, 2.13% and 48.84% respectively (panel A). The proportions of the βWβ state, βW/Oβ state and the βOβ 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 β1 0.4 0.5 Ξ³ kh (hour ) (C) W W O kh (hour β1) HIF-1(nM) 600 W 500 400 W/O 300 200 100 0 0.1 O 0.2 0.3 β1 0.4 0.5 kh (hour ) Figure S2. (A) Two-parameter bifurcation diagram of the metabolic circuit, as the function of πΈ and ππ . The circuit allows different phases for different values of πΎ and πβ . 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 (βWβ, βW/Oβ and βOβ). (B-C) The one-parameter bifurcation diagrams of the core AMPK:HIF-1:ROS circuit in response to the changes in πβ . 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 HIFβ 1 downstream genes W W/O O 5 0 β5 Pearson: -0.68 P < 0.001 β5 0 5 AMPK downstream genes HIFβ 1 downstream genes (B) (A) W W/O O 5 0 β5 Pearson: -0.33 P < 0.001 β5 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 HIFβ 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 β5 Pearson: -0.62 β 10 P < 0.001 β 15 β 10 β 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 β 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 β5 Pearson: -0.59 β 10 P < 0.001 β 15 β 10 β 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 β5 β5 0 5 Pearson: -0.22 5 P < 0.01 0 β5 β5 10 AMPK downstream genes 0 β5 β 10 Pearson: 0.08 P = 0.072 β 10 β5 0 5 (E) Prostate adenocarcinoma 5 β 15 0 5 AMPK downstream genes 10 10 Pearson: 0.36 P < 0.001 5 0 β5 β 10 β 10 β5 0 5 AMPK downstream genes Pearson: -0.20 P < 0.01 5 0 β5 β 10 β 10 0 10 AMPK downstream genes AMPK downstream genes HIFβ 1 downstream genes HIFβ 1 downstream genes (D) Kidney renal clear cell carcinoma HIFβ 1 downstream genes Pearson: -0.63 P < 0.001 (C) Pancreatic adenocarcinoma (F) Colorectal adenocarcinoma HIFβ 1 downstream genes 5 (B) Acute myeloid leukemia HIFβ 1 downstream genes HIFβ 1 downstream genes (A) Stomach adenocarcinoma 10 Pearson: 0.18 P < 0.001 5 0 β5 β 10 β 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 βWβ, βW/Oβ and βOβ 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 (βWβ, βW/Oβ and βOβ). Chi-Square test (See SI section 10) is used to analyze the significance of difference among βWβ, βOβ and βW/Oβ state. β*β represents P value < 0.05, β**β represents P value < 0.01, β***β 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 βWβ, βOβ and βW/Oβ state. β**β represents P value < 0.01, β***β 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 β 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 β 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 β the fitted curve matches worse to the histogram in the region of positive PC1 values; (C) an alternative fitting result with two Gaussian distributions β 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 (βWβ, βW/Oβ, βOβ) phase, and three different treatment strategies for the patients with cancer cells in the (βWβ) 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 πββππ‘ = 0, and the light blue line shows the nullcline of ππ΄βππ‘ = 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 βWβ (left), βW/Oβ (middle) and βOβ (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 βWβ to βW/Oβ, and metformin is much less efficient in driving the phenotype from the βW/Oβ to βOβ compared to the therapies targeting glycolysis. Therefore, cells stay in the hybrid βW/Oβ 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 βWβ state (magenta), the βW/Oβ state (green), or the βOβ 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 (300ΞΌM, 350ΞΌM, 400ΞΌM); group C: 3BP-based therapy (250ΞΌM, 300ΞΌM, 350ΞΌM); group D: combined therapy using both hyperbaric oxygen and metformin (10% and 250ΞΌM, 15% and 300ΞΌM, 20% and 350ΞΌM); group E: combined therapy using both 3BP and metformin (both are 250ΞΌM, 300ΞΌM and 350ΞΌM). 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 PPARΞ± 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. AMPKΞ±2 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-1Ξ± 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 SpeciesβInduced Stimulation of 5β² 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 Ξ² 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 1Ξ± 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, OβConnell JT, Herrera KNG, Wikman H, Pantel K, Haigis MC, et al. PGC1Ξ± 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 TGFΞ²1-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-1Ξ± 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 acidβinduced 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. Hypoxiaβinducible factor 1 mediates increased expression of NADPH oxidaseβ2 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 HIF1Ξ± 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-1Ξ± 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-1Ξ± 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-1Ξ± 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-Ξ²-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. AMPβactivated protein kinase inhibits TGFβΞ²βinduced 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, dβSouza 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-1Ξ± not through direct phosphorylation of HIF-1Ξ±-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-1Ξ±785 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-1Ξ± 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-1Ξ±. Cancer Research 2010;70:10351-61. Lehman JJ, Barger PM, Kovacs A, Saffitz JE, Medeiros DM, Kelly DP. Peroxisome proliferatorβactivated receptor Ξ³ 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-1Ξ± and HNF4Ξ± 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 Ξ± 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-1Ξ± and hypoxia-inducible factor-2Ξ± in HEK293T cells. Cancer Research 2005;65:3299-306. Lin J, Tarr PT, Yang R, Rhee J, Puigserver P, Newgard CB, et al. PGC-1Ξ² 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 TGFΞ² 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 KRABβA domain protein to repress HIFβ1Ξ± 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