By inhibiting PFKFB3, aspirin overcomes sorafenib resistance in hepatocellular carcinoma

Hepatocellular carcinoma (HCC) is one of the few cancers with a continuous increase in incidence and mortality. Drug resistance is a major problem in the treatment of HCC. In this study, two sorafenib‐resistant HCC cell lines and a nude mouse subcutaneously tumor model were used to explore the possible mechanisms leading to sorafenib resistance, and to investigate whether aspirin could increase the sensitivity of hepatoma cells to sorafenib. The combination of aspirin and sorafenib resulted in a synergistic antitumor effect against liver tumors both in vitro and in vivo. High glycolysis and PFKFB3 overexpression occupied a dominant position in sorafenib resistance, and can be targeted and overcome by aspirin. Aspirin plus sorafenib induced apoptosis in tumors without inducing weight loss, hepatotoxicity or inflammation. Our results suggest that aspirin overcomes sorafenib resistance and their combination may be an effective treatment approach for HCC.

However, emerging evidence shows that the clinical efficacy of sorafenib is severely limited with low response rates due to the development of resistance. 5 Since no agents evaluated in phase 3 trials as first-and second-line treatments for HCC could exceed the benefits of sorafenib, [6][7][8] investigation of the underlying mechanisms of resistance to sorafenib and identification of potential new strategies to enhance its efficacy are urgently needed.
HCC is frequently associated with enhanced glycolysis and increased expression of glycolytic enzymes. 9,10 This reprogrammed cancer metabolism is known as the Warburg effect. 11 There is extensive evidence to indicate that the dysregulated Warburg-like glucose metabolism correlates well with either inherent or acquired drug resistance in cancer cells, [12][13][14] and enhanced glycolysis contributes to sorafenib resistance in HCC cells. [15][16][17] Moreover, 6-phosphofructo-2kinase/fructose-2,6-biphosphatase 3 (PFKFB3) is found to be overexpressed in aggressive tumors compared with normal tissues, and this feature is highly linked to poor treatment effect and prognosis. 18 The PFKFB3 gene encodes both ubiquitous and inducible 6-phosphofructo-1-kinase 2 (PFK2), the latter stimulates fructose 2,6-bisphosphate (F2,6BP), which consequently allosteric activated PFK1 and an overall increase the glycolytic flux. Agents targeting glycolytic enzymes such as PFKFB3 (also called PFK2) can be regarded as promising efficacy in overcoming sorafenib resistance.
Aspirin (acetylsalicylic acid, ASA), as the most common nonsteroidal anti-inflammatory drugs (NSAIDs), is widely used as an analgesic, antipyretic and cardiovascular prophylaxis agent. 19 Since almost all HCC occurs with preexisting chronic liver disease, or chronic hepatic inflammation due to the hepatitis virus infection, 20 targeting inflammation or platelet function may be a therapeutic option in the carcinogenesis process. NSAIDs have been shown to reduce chronic inflammation and risk of many cancers, 21,22 but only aspirin was associated with reduced risk of developing HCC. 23 Various mechanisms such as the inhibition of cyclooxygenase-2 (COX-2) enzymes, and reduced production of prostaglandins and other inflammatory mediators, 24 as well as the important energy sensor AMPactivated protein kinase (AMPK), 25 and VEGF 26 may be involved in this process. Recently, it was reported that aspirin could target PFK, a major glycolysis regulatory enzyme, to inhibit cell proliferation. 27 Based on this idea, we proposed a hypothesis that, through combination treatment with aspirin and sorafenib, HCC sensitivity to sorafenib could be increased and HCC cells could be eradicated.

Reagents and cell culture
Aspirin was purchased from Sigma-Aldrich (St. Louis, MO), and sorafenib tosylate was purchased from Selleck (Selleck Chemicals, Shanghai, China), and was dissolved in dimethyl sulfoxide (Sigma-Aldrich).

Cell proliferation and combination analysis
Cells were cultured in a 96-well plate, and then, 10 lL CCK-8 solution (Peptide Institute, Osaka, Japan) was added to each well. The plate was maintained in the dark for 378C for 2 hr. The absorbance was measured at 450 nm using a microplate reader for further IC 50 and drug combination analysis.
Differences in potential synergistic inhibition between sorafenib and ASA were assessed using Chou-Talalay median effect analysis. 28 The combination indexes (CIs) and dose reduction indexes (DRIs) after drug treatments were calculated using ComboSyn software. The CI values of < 1, 5 1, and > 1 indicated synergistic, additive, and antagonistic effects, respectively, while DRI values > 1 demonstrated that combinations could reduce drug doses compared treatment with each drug alone.

Apoptosis analysis
After different drug treatments, cells were stained with phycoerythrin (PE)-Annexin V/7-amino-actinomycin (7-AAD; BD BioSciences, San Jose, CA) according to the manufacturer's What's new? Sorafenib, a kinase inhibitor, is one of the most effective drugs available for the treatment of hepatocellular carcinoma (HCC). Its use, however, is limited by acquired resistance. The present study shows that the expression of PFKFB3, a gene involved in glycolytic flux that encodes 6-phosphofructo-1-kinase 2 (PFK2), is strongly associated with sorafenib resistance in HCC cells. PFK is a suspected target of aspirin, a drug associated with reduced HCC risk. Experiments in cells and animals reveal the existence of a synergistic antitumor effect between aspirin and sorafenib, suggesting that sorafenib-resistant HCC patients may benefit from combined treatment with aspirin.

JC-1 staining
JC-1 (BD Pharmingen) was used to measure mitochondrial membrane potential. Briefly, cells were cultured in six-well plates and then stained with 10 lg/mL JC-1 for 15 min. Cells were washed twice with PBS and then analyzed using a flow cytometer (Beckman Coulter, Villepinte, France).

2-DG uptake, lactate production and O 2 consumption
Cells were washed with uptake buffer twice, cultured in 1 lCi/mL [ 3 H]-2-DG at 378C for 30 min, and solubilized with 0.1% sodium dodecyl sulfate. The radioactivity was calculated and were normalized to the protein content and corrected for the zero-time uptake per mg protein. Lactate levels were measured using a fluorometric assay (BioVision, Milpitas, CA) according to the manufacturer's protocol. O 2 consumption was tested using the 110 Fiber optic oxygen monitor (Instech, Plymouth Meeting, PA), and results were expressed as nmol O 2 /million cells/min.

Reverse transcription (RT)-PCR and quantitative real time (qRT)-PCR
TRIzol reagent was used to extract total RNA. cDNA was synthesized using SuperScript II reverse transcriptase with Oligo (dT) (Invitrogen, Carlsbad, CA). The real-time PCR experiment was performed following the protocol of the realtime PCR kit (Takara, Dalian, China). The levels of the target genes were normalized to b-actin.

Protein extraction and western blotting
The cytosolic and mitochondrial fractions were separated and purified using a Mitochondrial Isolation Kit (Pierce, Rockford, IL) according to the manufacturer's protocol. Total cellular proteins were extracted using radio-immunoprecipitation assay buffer (Sigma-Aldrich) containing protease inhibitors.
The samples were resolved by sodium dodecyl sulfatepolyacrylamide gel electrophoresis and transferred to polyvinyl difluoride membranes. The membranes were sequentially blocked in PBS containing 0.1% Tween 20 (PBST) with 5% non-fat milk for 1 hr and probed with primary antibodies (Supporting Information Table S1). Membranes were then washed with PBST three times and incubated with the appropriate secondary antibodies for 1 hr at room temperature. Finally, the membranes were washed again and scanned using the Odyssey two-color infrared laser imaging system (fluorescence detection). HSP-70 and b-actin was used as an internal control.

Plasmid construction, lentivirus packaging and infection
A full-length cDNA encoding the PFKFB3 sequence was amplified from 293 T cDNA and then cloned into the pCDH-CMV-MCS-EF1-GFP vector (System Biosciences, Mountain View, CA). Empty lentiviral vector was used as control. All plasmid sequences were confirmed by DNA sequencing. Target cells were infected with empty vector or PFKFB3 in the presence of 8 lg/mL polybrene (Sigma-Aldrich) overnight. The transduction efficiency was measured by real-time PCR and western blotting.
For PFKFB3 siRNA knockdown, PFKFB3 expression in Huh7-R cells was ablated with small interfering RNAs (siR-NAs), and PFKFB3 scramble siRNA (scRNA) was used as control. All plasmid sequences were confirmed by DNA sequencing. The siRNAs were transfected into cells using Lipofectamine TM 2000 following the manufacturer's instructions. The transduction efficiency was measured by real-time PCR. The sorted cells were then characterized and used in further assays.

Animal experiments
Four-week-old male athymic BALB/C nu/nu mice with free access to water and food were housed in a standard animal laboratory with a 12-hr light-dark cycle and constant environmental conditions. All experiments were performed in accordance with ethical standards and in compliance with the Declaration of Helsinki, and according to national and international guidelines. The study was approved by the Animal Care and Use Committee of Shanghai Tongji University. Serum-free culture medium (200 lL) containing HCC-LM3 cells (5 3 10 6 ) was subcutaneously injected into the upper flank region of 54 mice. When the tumor volume was 100 mm 3 , the animals were randomly divided into nine groups: NC, ASA alone (20, 50 and 100 mg/kg), sorafenib alone (10 and 20 mg/kg) and combination treatment (sorafenib 10 mg/kg 1 ASA 20 mg/kg, sorafenib 10 mg/kg 1 ASA 50 mg/kg and sorafenib 10 mg/kg 1 ASA 100 mg/kg). Both sorafenib and ASA were given by oral gavage once a day for 30 days. Tumor volume was calculated using the following formula: volume (mm 3 ) 5 (width) 2 3 length/2. Body weight of the mice was measured every 5 days. Mice were euthanized 24 hr after the last treatment. Tumors were resected and imaged using a high-definition digital camera. The terminal deoxynucleotidyl transferase-mediated dUTP nick-end labeling (TUNEL) assay was used to measure the degree of tumor cell apoptosis. The percentage of TUNEL-positive tumor cells was determined by counting the average cell number in each of four high-power fields in each sample.

Immunohistochemistry
Slices (3 lm) were dewaxed, rehydrated and treated with 3% H 2 O 2 . The nonspecific sites were blocked with 10% goat serum for 30 min at room temperature. Then, slices were incubated overnight with primary antibody. On the following day, after incubation with secondary antibody, the slides were counterstained with hematoxylin and imaged under a light microscope.

Statistical analysis
All results are expressed as means 6 standard deviation. Statistical analysis was performed using a two-tailed unpaired Student's t test and SPSS 17.0 software (IBM, Armonk, NY). Quantitative data are representative of at least three independent experiments. Values of *p < 0.05, **p < 0.01 and ***p < 0.001 were considered statistically significant.

Metabolic characterization change of HCC cells is associated with their resistance to sorafenib
The five sorafenib-na€ ıve HCC cell lines tested showed different half maximal inhibitory concentration (IC 50 values) for sorafenib (range: 4.79-26.10 lM; Fig. 1a). Among them, HCC-LM3 displayed the highest IC 50 for sorafenib, and was regarded as a typical innate sorafenib-resistant HCC cell line, which was consistent with the conclusions of previous reports. 29,30 The IC 50 of the acquired sorafenib-resistant Huh7-R cells was 26.10 6 1.73 lM, much higher than for sorafenib-na€ ıve Huh7 cells (p < 0.001). The prominent sorafenib resistance of HCC-LM3 and Huh7-R cells was also confirmed by flow cytometry compared with Huh7 cells when treated with 10 lM sorafenib (apoptosis rate of Huh7-R vs. Since previous studies indicated that a bioenergetic propensity for using glycolysis is highly associated with resistance of HCC cells to sorafenib, 15 the metabolic characteristics, including glucose uptake and lactate production, along with O 2 consumption and byproducts of OXPHOS of two sorafenibresistant HCC cell lines were compared with the sorafenibsensitive Huh7 cells. When cultured under normoxic conditions for 24 hr, glucose uptake and lactate production in Huh7-R cells were approximately five times more than Huh7 cells (Fig. 1c), and those in HCC-LM3 cells were increased 4fold over Huh7, demonstrating a higher rate of glycolysis. In opposite to high glycolytic flux, energy generation from oxidative phosphorylation (OXPHOS) was impaired, evidenced by decreased oxygen consumption and disrupted OXPHOS capacity. Our data showed the low efficiency of O 2 consumption and down-regulated expression of OXPHOS metabolismcorrelated proteins (complexes I/II/III/IV/V in the electron transport chain) in Huh7-R and HCC-LM3 cells (Figs. 1d and 1e). Based on these results, there is a shunt directed toward glycolysis from oxidative phosphorylation in the two sorafenibresistant HCC cell lines.
To examine the mechanism underlying the effect of elevated glycolysis in sorafenib-resistant HCC cells, we measured the expression of several key glycolytic enzymes, including GLUT1/2/3/4, HK1/2, PFK1, PFKFB3, PKM2 and LDH-A by qRT-PCR. Among them, the PFKFB3 mRNA level was increased the most in both Huh7-R and HCC-LM3 cells, followed by GLUT1 and HK2 (Fig. 1f). Results of western blotting further confirmed the up-regulated expression of PFKFB3, GLUT1 and HK2 in untreated resistant cells (Fig.  1e). The tremendous heterogeneity suggests a positive correlation between enhanced glycolysis, especially the ascendant expression of PFKFB3, GLUT1 and HK2, and resistance to sorafenib.
Combination treatment with aspirin and sorafenib in vitro increased cell inhibition and apoptosis rate in sorafenibresistant HCC cells Previous studies indicated ASA could induce cancer cell death through inhibition of glycolysis. 27 We hypothesized that ASA could reverse the sorafenib-resistance induced by elevated glycolysis in HCC cells. Huh7-R and HCC-LM3 cells were treated with ASA or sorafenib alone, or ASA plus sorafenib, for 24 hr, and the inhibitory effect on cell viability with combination treatment was found to be drastically boosted compared with either ASA or sorafenib alone (Supporting Information Fig. S1).
Fraction affected (Fa) values (the fraction of cells inhibited with drug exposure) were obtained after exposure of HCC cells to a series of drug concentrations. To indicate the effects at different Fa values, the CI and DRI values were calculated for each Fa. Fa-CI plots illustrate the effects of ASA and sorafenib at different fixed drug ratios, and demonstrate synergism (CI < 1) at Fa > 0.5 in both cell lines (Fig. 2a). As expected, synergism corresponding to CI < 1 always yields a favorable DRI (> 1) for both drugs (Fig. 2b), indicating that dosage of sorafenib can be significantly reduced when combined with ASA. We then treated Huh7-R cells with inconstant drug ratios, and the results also indicated synergistic interactions, evidenced by CI < 1 (Fig. 2c) and DRI > 1 (Fig. 2d). In addition, the drug ratio of option to exhibit the best synergistic effect is 1: 500 (sorafenib: ASA), followed by 1: 200 (sorafenib: ASA) in both Huh7-R and HCC-LM3 cells (Figs. 2a and 2b). Thus, sorafenib at 10 lM and ASA at 5 mM was chosen for the following in vitro study.
Next, the combination effect of ASA and sorafenib on apoptosis was evaluated by flow cytometry. The results showed that apoptosis with combination therapy increased significantly in two sorafenib-resistant HCC cell lines (Fig.  2e). Western blot analysis showed that combination treatment significantly decreased the expression of PCNA, and induced the activation of caspase 3 and 9 and cleavage of PARP (Fig. 2f), which was not obviously seen when cells were treated with sorafenib alone.
To explore the safety of the drug dose, two normal hepatic cell lines, QSG-7701 31 and LO2, 32 were treated with ASA or sorafenib alone, or ASA plus sorafenib for 24 hr. The IC 50 values of both ASA and sorafenib in two normal hepatic cell lines were evidently higher than in the sorafenib-resistant Huh7-R cells (Supporting Information Fig. S2A,B). Meanwhile the combination treatment, at the same dose that significantly increase, the apoptosis rate in two sorafenib-resistant HCC cell lines, had no obvious pro-apoptosis effect on QSG-7701 and LO2 cells (Supporting Information Fig. S2C), indicating a rather safe situation for the combination chemotherapy. To investigate the effects of the combination of ASA and sorafenib in vivo, we established a mouse xenograft model using HCC-LM3 cells. Saline, sorafenib (10 or 20 mg/kg), ASA (20,50 or 100 mg/kg), and sorafenib (10 mg/kg) 1 ASA (20, 50 or 100 mg/kg) were used for the in vivo experiments.
Mice treated with ASA or sorafenib alone showed a relatively smaller tumor size than untreated mice after 30 days of treatment (Figs. 3a and 3b). Alternatively, sorafenib combined with ASA at 50 and 100 mg/kg significantly suppressed tumor size compared with sorafenib alone (0.145 6 0.025, 0.024 6 0.008 vs. 0.308 6 0.022, p 5 0.0027, p < 0.0001, respectively; Figs. 3a and 3b). Although the combination of ASA at 20 mg/kg with sorafenib showed no significant difference compared with sorafenib alone, there was a tendency towards a reduction in tumor size (0.166 6 0.078 vs. 0.308 6 0.022, p 5 0.1316). It is notable that, tumor size in the sorafenib (10 mg/kg) 1 ASA (100 mg/kg) group was smaller than in the sorafenib (20 mg/kg) alone Group (0.024 6 0.008 vs. 0.209 6 0.053, p 5 0.0139). These findings indicate that with the addition of ASA, the antitumor effect of sorafenib can still be maintained, or even improved, even at a reduced dose.
The results of dynamic observations of the antitumor effects over 30 days of treatment showed the same pattern (data not shown), while no significant loss of body weight, or elevation of serum AST or ALT was observed over the 30 days of treatment (Supporting Information Fig. S3), demonstrating no obvious deleterious effect on body weight or hepatic function in mice treated with ASA or sorafenib alone or in combination.
Compared with treatment with sorafenib alone, combination treatment with ASA and sorafenib significantly increased the rate of apoptosis in vivo (65.97 6 4.52% vs. 32.206 5.95%, p 5 0.0107; Figs. 3c and 3d). These data further confirmed the hypothesis that ASA could sensitize cells to sorafenib in sorafenib-resistant HCC cells in a xenograft mouse model.

PFKFB3 is essential for combination treatment-inhibition of glycolysis and proliferation in HCC cells both in vitro and in vivo
Huh7-R cells treated with ASA (5 mM) showed significantly lower (p 5 0.023) glucose uptake (514.3 pmol/mg/min) than the untreated controls (1029.0 pmol/mg/min). Combination treatment (sorafenib 10 lM 1 ASA 5 mM) also led to markedly lower (p 5 0.0198) glucose uptake (402.7 pmol/mg/min) compared with sorafenib alone (627.3 pmol/mg/min; Fig. 4a). Lactate production showed a similar pattern to that of glucose uptake (Fig. 4a), indicating a reduced glycolytic rate in sorafenib-resistant HCC cells with combination treatment.
To examine the mechanism underlying the effect of ASA plus sorafenib on inhibition of glycolysis, the effects of combination treatment on the expression of key glycolytic enzymes were assessed in Huh7-R cells using qRT-PCR. Among these, PFKFB3 was the most down-regulated enzyme at the mRNA level (Fig. 4b). Western blot analysis also indicated a significant decrease in PFKFB3 and PFK1 protein expression with combination treatment (Fig. 4c). In addition, analysis of the PFKFB3 immunohistochemical changes in xenograft mouse tumors confirmed these results in vivo (IOD of combination treatment vs. sorafenib: 37.00 6 3.61 vs. 71.33 6 2.03, P 5 0.0012; Figs. 4d and 4e). Taking these data together, we can infer that PFKFB3 plays a vital role in combination treatment-inhibition of HCC cells glycolysis and proliferation.
To verify the effect on PFKFB3, 3-(3-pyridinyl)-1-(4-pyridinyl)-2-propen-1-one (3PO), a specific PFKFB3 inhibitor was used as a positive control. 33 The reduced glycolysis and cell proliferation and elevated apoptosis effect induced by the combination treatment were similar to those induced by the PFKFB3 inhibitor (Figs. 4fh). These results indicate that both treatments may function through the same mechanism: inhibition of PFKFB3.

Overexpression of PFKFB3 mimics the sorafenib resistance effect in Huh7 cells
To further prove the status of PFKFB3, lentivirus was used to infect the sorafenib-na€ ıve Huh7 cells, causing them to overexpress PFKFB3 (Supporting Information Fig. S4). The increased glucose uptake and lactate production in PFKFB3-overexpressing Huh7 cells indicated the ascendancy of PFKFB3 in the glycolysis pathway (Fig. 5a). In addition, the IC 50 of PFKFB3-overexpressing Huh7 cells was higher than the empty virusinfected Huh7 cells (11.67 6 0.78 vs. 4.97 6 0.49, p 5 0.0019; Fig.  5b). Sorafenib at a concentration of 10 lM combined with Lenti-PFKFB3 significantly attenuated sorafenib-induced apoptosis in Huh7 cells (24.10% for combination vs. 37.03% for sorafenib alone; Fig. 5c) by inhibiting the activation of caspases 3 and 9 (Fig. 5d). These results indicated a sorafenib-resistance effect in PFKFB3-overexpressing Huh7 cells, showing that PFKFB3 is essential for both glycolysis and sorafenib resistance.
Furthermore, the siRNA-induced knockdown of PFKFB3 in Huh7-R cells also restored their sensitivity to sorafenib (Supporting Information Fig. 5). Taken together, the loss of PFKFB3 in the sorafenib-resistant cells restored their sensitivity, while the over-expression of PFKFB3 in sorafenib-sensitive cells created sorafenib-resistance, indicating the dominant status of PFKFB3 in the sorafenib-resistance mechanism of HCC cells. ASA combined with sorafenib showed synergism in PFKFB3overexpressing Huh7 cells, indicating the cell death promoting effect of the combination treatment is targeting PFKFB3.

PFKFB3 was suppressed through inhibition of HIF-1a in HCC cells
To determine a possible mechanism by which combination treatment induced apoptosis and reduced glycolysis in HCC cells, real-time PCR was used to test mRNA levels of 27 apoptosis-and glycolysis-related molecules after Huh7-R cells were treated with ASA plus sorafenib for 24 hr. The values of log relative mRNA compared with the negative control group are shown in Supporting Information Fig. S6A. Among those, HIF-1a mRNA level altered the most, with an 85% reduction in combination-treated cells (p < 0.001), followed by the tumor suppressor PTEN and P53, which increased by 4.4-and 4.9-fold respectively (p < 0.001 for PTEN; p < 0.01 for P53). Expression of COX-2 was also down-regulated, while AMPKa1/2 and AMPKb1/2 was up-regulated with the combination treatment (Supporting Information Fig. S6A).
To test the possible connection between AMPK and PFKFB3, Compound C, a potent AMPK inhibitor 34 was used to test glycolysis and PFKFB3 alteration. As is shown in Supporting Information Fig. S6B, effects of ASA plus sorafenib on the expression of PFKFB3 and glycolytic rate were not altered with the inhibited AMPK, suggesting that the effects of PFKFB3 on HCC cell lines is AMPK-independent. Meanwhile, roxadustat (FG-4592), a prolyl-4-hydroxylases inhibitor and HIF-1a stabilizer, 35 stabilized HIF-1a and reversed the glycolysis-reducing effect induced by ASA combined with sorafenib through up-regulating the expression of PFKFB3 and PFK1 (Supporting Information Fig. S6B,C). These data show that, ASA plus sorafenib reduce glycolysis via inhibition of HIF-1a/PFKFB3. The mitochondrial apoptosis pathway is involved in combination treatment-induced apoptosis and is partly caspase-dependent Mitochondria play a pivotal role in two integral components of cellular transformation, cellular metabolism and apoptosis. 36 The major form of apoptosis in most cancer cells is the mitochondrial pathway, defined by a pivotal event in the process: mitochondrial outer membrane permeabilization, which can be detected as mitochondrial inner transmembrane potential (Dwm). The loss of Dwm in combination treatment provides evidence to this particular mechanism (Fig. 6a).
During the process of mitochondrial apoptosis, caspase activation is closely linked to mitochondrial outer membrane permeabilization. 37 To test whether the combination treatment-induced  . (a, b) EV or Huh7-PFKFB3 cells harvested after 24 hr of culture were used to detect relative lactate production, 2-DG uptake, and half maximal inhibitory concentration. (c) EV or Huh7-PFKFB3 cells cultured with or without sorafenib (10 lM) for 24 hr were used to detect cell apoptosis using flow cytometry. (d) The expression of caspase 3 and cleaved caspase 9 was monitored by western blot analysis in EV or Huh7-PFKFB3 cells cultured with or without sorafenib (10 lM) after 24 hr. (e, f) Illustrative Fa-CI (e) and Fa-DRI (f) plots for the combination of sorafenib and ASA using different constant drug ratios in Huh7-PFKFB3 cells. CI < 1 and DRI > 1 denotes synergistic interactions. (g) Huh7-PFKFB3 cells were treated with sorafenib (10 lM) or/and ASA (5 mM) for 24 hr, then apoptosis was monitored by flow cytometry. Plotted values represent the mean 6 standard error of three independent experiments (n 5 3; *p < 0.05, **p < 0.01, ***p < 0.001). [Color figure can be viewed at wileyonlinelibrary.com] Figure 6. Apoptosis regulatory effects of combination treatment in Huh7-R cells. (a) To measure changes in the Dwm, Huh7-R cells (5 3 104) treated with sorafenib (10 lM) or/and ASA (5 mM) for 24 hr were stained with JC-1 (10 lg/mL) and analyzed by flow cytometry. (b) Huh7-R cells were treated with combination treatment (sorafenib 10 lM and ASA 5 mM) or/and Z-VAD-FMK (100 lM) for 24 hr, and the apoptosis rate was examined by flow cytometry. (c) The expression of mitochondrial apoptotic regulatory proteins in cytoplasm or mitochondria was monitored by western blot analysis in Huh7-R cells treated with sorafenib (10 lM) or/and ASA (5 mM) for 24 hr. Actin and hsp-70 served as loading controls. Results are the means of three experiments. (d) Mode of aspirin reveres sorafenib resistance in HCC cells: With aspirin treatment, high PFKFB3 expression in sorafenib-resistant HCC cells is suppressed, leading to restrained aerobic glycolysis by PFK1, resulting in the suppression of ATP production. This leads to mitochondrial membrane potential breakdown and causes direct caspases activation, resulting in cell death. In addition, the inhibition of PFKFB3 induces the suppression of PFK1, thus activating Bad by dephosphorylation, which interacts with the anti-apoptotic Bcl-2 family proteins Bcl-2 and Bcl-xl to relieve their inhibition of pro-apoptotic proteins Bax and Bak. Oligomerization and activation of Bax and Bak leads to the increase of mitochondrial outer membrane permeabilization, and releases apoptogenic substrates from the mitochondria, such as Cyto c into the cytoplasm and AIF to the nucleus, activating caspases, resulting in nuclear DNA fragmentation and cell apoptosis, and restoring the sensitivity to sorafenib.
One link between apoptosis inhibition and metabolic reprogramming may be provided by the association of PFK with the proapoptotic protein Bad. 39 As shown in Figures 6c  and 6d, Bad was dephosphorylated in response to the combination treatment, and its de-phosphorylation encouraged the translocation of Bax and Bak from the cytosol to the mitochondria and of Bcl-2 and Bcl-xl from mitochondria, leading to the release of apoptogenic substrates from the mitochondria, such as cytochrome c (Cyto c) and apoptosis-inducing factor (AIF). Taken together, these data show that, Bcl-2 family proteins are involved in the regulation of combination treatment-induced mitochondrial apoptosis.

Discussion
HCC is one of the most common malignancies worldwide, with an incidence and mortality that are increasing yearly. [40][41][42] Most modern medicines currently available for treating HCC are expensive, toxic, and not effective enough. [43][44][45] The recent approval of sorafenib as the first effective oral drug for HCC marks a significant milestone. However, sorafenib has been proven to limit survival benefits with very low response rates because of drug resistance. 46 The Warburg effect is reported to be closely associated with either inherent or acquired sorafenib resistance. 15,29 Glycolysis in sorafenib resistant HCC cells increases to generate macromolecules such as lactate, etc., and the subsequent usage of which is required for cancer cell growth. This is clearly observed in our constructed sorafenib-resistant Huh7 cell line (Huh7-R) and in the innate sorafenib-resistant HCC-LM3 cells compared with the parental sensitive cells (Fig. 1c), indicating that drug resistance in HCC cells is directly linked to an increase in glycolysis. Further exploration of the key glycolysis enzymes reveals that upregulated PFKFB3 is strongly associated with sorafenib resistance in HCC cells. This is proven by three key lines of evidence: first, the expression of PFKFB3 is higher in sorafenib-resistant cells compared to the parental sensitive cells (Figs. 1e and 1f); second, overexpression of PFKFB3 in sorafenib-sensitive Huh7 cells mimics the sorafenib resistance effect (Figs. 5ad); and third, siRNA-knockdown of PFKFB3 in sorafenib-resistant Huh7-R cells restores its sensitivity to sorafenib (Supporting Information Fig. S5). All these imply that bioenergetic changes toward increased PFKFB3 and glycolysis are linked to sorafenib-resistance. Therefore, in the search for a combination of chemotherapeutic drugs with tumor glycolysis inhibitors, inhibitors of PFKFB3 in particular may represent a promising strategy to overcome sorafenib resistance.
Recently, the preventive effects of ASA on cancers have been studied extensively, including in prostate cancer, 47 colorectal cancer, 24 pancreatic cancer 48 and HCC, 23 etc. However, most laboratory experiments and clinical trials were carried out to explore its preventive effects on tumorigenesis; its therapeutic effect has rarely been mentioned. This study illustrated the therapeutic effect of aspirin in HCC cells (Figs. 2e and 2f and Supporting Information Fig. S1) and on tumor size in an established-mouse subcutaneous tumor model (Figs. 3a and 3b). In addition, our research also demonstrated that high glycolysis and PFKFB3 overexpression in sorafenibresistant HCC cells could be overcome by aspirin (Figs. 4ae and 5 g), which explained the possible molecular mechanism of its cancer-suppressing effects.
Given the characteristics of aspirin, in the present study, two sorafenib-resistant HCC cell lines (Huh7-R and HCC-LM3 cells) and a nude mouse subcutaneous tumor model were used to determine whether the combination of aspirin and sorafenib could enhance the sorafenib sensitivity of HCC cells. Results showed that the combination therapy had a synergistic antitumor effect against liver tumors, evidenced by CI < 1 in Chou-Talalay median effect analysis (Figs. 2a and 2c); and the DRI > 1 (Figs. 2b and 2d) illustrated that to reach the same therapeutic effect, both sorafenib dose and aspirin dose could be significantly reduced with combination treatment, but especially the dose of sorafenib; besides, the best drug ratio is explored: 1: 500 for sorafenib vs. aspirin in vitro. These finding have also been confirmed in vivo: tumor size in the sorafenib (10 mg/kg) 1 ASA (100 mg/kg) group was smaller than in the sorafenib Group (20 mg/kg; Figs. 3a and 3b), further indicating that with the addition of aspirin, although sorafenib dosage was reduced by half, the antitumor effect was still maintained, or even improved. A similar synergistic effect was also observed in the PFKFB3 overexpression-induced sorafenib-resistant Huh7 cells (Figs. 5eg). Analysis of apoptosis in vitro and in vivo also confirmed the synergistic antitumor effect, as evidenced by alterations in the number of positive apoptotic cells (Fig. 2e), expression of proteins that regulate apoptosis (Figs. 2f and 6c) and positive TUNEL cells (Figs. 3c and 3d) after combination treatment.
The synergistic antitumor effect of aspirin combined with sorafenib provides several advantages for clinical HCC patients: first, through combining with aspirin, the sensitivity of sorafenib is increased. Sorafenib, with its gifted and unique advantages in HCC, can benefit the increasing number of HCC patients that were originally sorafenib resistant; second, since there is a subgroup of HCC patients who suffer from deleterious side effects of sorafenib, after an initial satisfactory response, who then have to reduce the drug dosage or even terminate the therapy. 49 With combination treatment, aspirin ameliorates the side effects of sorafenib, allowing more patients who could not tolerate sorafenib previously to re-gain the benefits; third, sorafenib is an expensive oral medicine, it costs about $US 4,079 per month and about $US 40,639 6 $US 3,052 over a patient's lifetime according to research by Carr et al. in Canada in 2010. 50 This is a huge economic burden for most families. Through combination with aspirin, the dosage of sorafenib can be reduced, which may provide a financial opportunity for more patients to try the treatment, since aspirin is relatively cheap. In addition, the safety of combination treatment has been proven both in vitro and in vivo, given the fact that the drug dose shown to be effective in cancer inhibition has no obvious influence on normal hepatic cells in vitro (Supporting Information Fig.  S2), or on hepatic function or body weight in vivo (Supporting Information Fig. S3), indicating its practicable application in the clinic.
In addition, the possible mechanism of aspirin combined with sorafenib was explored, and results showed that both aspirin alone and the combination treatment decreased the expression of activated PFKFB3 as well as glycolysis in sorafenibresistant HCC cells (Fig. 4), proving our hypothesis that inhibition of the tumorigenic/proliferative ability of HCC by aspirin was linked to sensitization of HCC cells to sorafenib, through reversing the overexpression of PFKFB3 and glycolysis. Further exploration of the possible pathways, which regulate glycolysis and apoptosis showed that HIF-1a changed the most obviously with the combination treatment (Supporting Information Fig.  S6A). After HIF-1a was stabilized by roxadustat (FG-4592), inhibition of PFKFB3 expression and glycolysis by combination treatment was attenuated (Supporting Information Fig. S6B,C). Taken together, these data suggest that the combination treatment acts through the HIF-1a/PFKFB3/PFK1 pathway to regulate glycolysis and apoptosis.
The mechanism of cross-talk between glycolysis and apoptosis is still unclear. Through analyzing the results of our research, we proposed that the inhibition of glycolysis by aspirin increased mitochondrial outer membrane permeabilization, and eventually activated downstream mitochondrial apoptotic signaling, leading to the death of sorafenib-resistant cancer cells. This may occur through two approaches: first, the decrease in glycolysis would reduce the production of metabolites such as lactate, which provides advantages for cancer cell growth and metastasis, that eventually leading to apoptosis; second, PFK1, which is activated by PFKFB3, was recently identified as a novel Bad-associated proapoptotic protein. The interaction of PFK1/Bad also contributes to the process of apoptosis. In this study, results showed that the increase in mitochondrial permeabilization and expression of caspase proteins play a decisive part in combination treatment-induced apoptosis, which is regulated by the phosphorylation of Bad (Fig. 6).
In addition to PFKFB3, other glycolysis-regulating enzymes such as glucose transporter 1 (GLUT1), hexokinase 2 (HK2) and lactate dehydrogenase-A (LDHA) are also activated in sorafenib-resistant HCC cells (Figs. 1e and 1f). And agents that target glycolysis such as resveratrol, 29 inhibitors of HK2, and epigallocatechin-3-gallate, 30 inhibitors of PFK, have also shown promising efficacy in overcoming drug resistance in several in vitro models. What's more, expression of MAP kinase signaling, VEGF receptor and PDGF receptor is positively correlated with sorafenib resistance (Supporting Information Fig. S7). Therefore, further in vitro and in vivo studies identifying the specific molecular targets, signaling and metabolic pathways that affect sorafenib sensitivity, the therapeutic effect of aspirin and other selective NSAIDs alone or combined with sorafenib are required.
In conclusion, this study is the first to demonstrate that PFKFB3 overexpression occupied a dominant position in sorafenib resistance, and can be targeted and overcome by aspirin. In addition, our research is the first to exhibit that aspirin could enhance the effect of conventional cancer therapies through synergistic effects as well as through amelioration the deleterious side effects of sorafenib, providing effective treatment strategies for HCC patients.