Transforming Growth Factor β Mediates Drug Resistance by Regulating the Expression of Pyruvate Dehydrogenase Kinase 4 in Colorectal Cancer*

Drug resistance is one of the main causes of colon cancer recurrence. However, our understanding of the underlying mechanisms and availability of therapeutic options remains limited. Here we show that expression of pyruvate dehydrogenase kinase 4 (PDK4) is positively correlated with drug resistance of colon cancer cells and induced by 5-fluorouracil (5-FU) treatment in drug-resistant but not drug-sensitive cells. Knockdown of PDK4 expression sensitizes colon cancer cells to 5-FU or oxaliplatin-induced apoptosis in vitro and increases the effectiveness of 5-FU in the inhibition of tumor growth in a mouse xenograft model in vivo. In addition, we demonstrate for the first time that TGFβ mediates drug resistance by regulating PDK4 expression and that 5-FU induces PDK4 expression in a TGFβ signaling-dependent manner. Mechanistically, knockdown or inhibition of PDK4 significantly increases the inhibitory effect of 5-FU on expression of the anti-apoptotic factors Bcl-2 and survivin. Importantly, studies of patient samples indicate that expression of PDK4 and phosphorylation of Smad2, an indicator of TGFβ pathway activation, show a strong correlation and that both positively associate with chemoresistance in colorectal cancer. These findings indicate that the TGFβ/PDK4 signaling axis plays an important role in the response of colorectal cancer to chemotherapy. A major implication of our studies is that inhibition of PDK4 may have considerable therapeutic potential to overcome drug resistance in colorectal cancer patients, which warrants the development of PDK4-specific inhibitors.

Colorectal cancer is the third most common cancer diagnosed and the second leading cause of cancer mortality in the United States. In addition to surgery, treatment for colorectal cancer patients, especially for patients with advanced disease, primarily relies on chemotherapy and radiation therapy. Fluorouracil (5-FU) 5 is one of the most commonly used chemotherapeutic agents for colorectal cancer treatment. It induces cell cycle arrest and apoptosis in cancer cells (1). Although adjuvant 5-FU treatment has a good success rate, the high recurrence is still a major hurdle of treating colorectal cancer because of the resistance to chemotherapeutic drugs. Therefore, it is important to understand the mechanisms of drug resistance and identify new targets to increase the effectiveness of chemotherapy in colorectal cancer. TGF␤ plays an important role in cancer development and progression. Upon ligand binding, TGF␤ type II receptor (RII) recruits and activates TGF␤ type I receptor (RI), which then activates Smad2 and Smad3. Activated Smad2 and Smad3 form complexes with Smad4 and translocate to the nucleus, where they regulate gene expression (2). We and others have demonstrated that TGF␤ suppresses tumorigenicity in a variety of cancers, including colorectal cancer, and that loss of TGF␤ signaling leads to malignancy (3)(4)(5)(6). Although many studies have shown that TGF␤ promotes metastasis (7), others have demonstrated that TGF␤ suppresses metastasis (8,9). Recently, studies have indicated that TGF␤ signaling is an emerging player in cancer drug resistance (10 -12). Although TGF␤ signaling may act as a tumor suppressor, it could have adverse effects in chemotherapy. More studies are needed to better understand the underlying mechanisms.
Metabolic abnormality is one of the hallmarks of cancer (13). Aberrant glucose metabolism is widespread in cancer cells and has been shown to play a role in drug resistance (14,15). Otto Warburg first reported that cancer cells use glycolysis rather than oxidative phosphorylation for energy, which is referred to as aerobic glycolysis or the "Warburg effect" (16). Pyruvate dehydrogenase (PDH) complex, which converts pyruvate to acetyl-CoA, is one of the important mediators of glucose oxidative metabolism. It interconnects glycolysis and the TCA cycle, therefore representing a key regulatory step in glucose metabolism. The activity of PDH is tightly controlled by reversible phosphorylation mediated by PDH kinases (PDKs) and PDH phosphatases (17). Phosphorylation and inactivation of PDH by PDKs favors the glycolytic phenotype, which confers resistance to apoptosis (18,19). Inhibition of PDKs switches glucose metabolism to aerobic oxidation, which is disadvantageous to tumor growth (20).
There are four different human PDKs (PDK1-4) (17). These different PDKs respond to various stimuli, including hypoxia and nutrient deprivation, to regulate glucose metabolism (21,22). Among the four PDKs, PDK4 is a principal regulator responsible for the modulation of PDH activity and fuel selection between glucose and fatty acids for energy utilization (23). Expression of PDK4 is selectively induced in most tissues and organs in response to metabolic stress (23). In this study, we have identified a novel function of PDK4 that mediates the response of colon cancer cells to 5-FU treatment. We show that PDK4, but not PDK1-3, is differentially expressed in colon cancer cells and that its expression positively correlates with the resistance to 5-FU treatment. Knockdown of PDK4 sensitizes colon cancer cells to 5-FU-or oxaliplatin-induced apoptosis.
Remarkably, experiments in tumor xenograft mouse models demonstrate that knockdown of PDK4 significantly increases the effectiveness of 5-FU-mediated inhibition of tumor growth in vivo. Furthermore, we have discovered a novel cross-talk between TGF␤ signaling and PDK4. TGF␤ signaling enhances PDK4 expression, and elevated PDK4 expression contributes to TGF␤-mediated drug resistance in colon cancer cells. Studies of patient specimens show that PDK4 expression and TGF␤ activation positively correlate with each other and with chemoresistance in colorectal cancer, indicating the clinical relevance of our studies. Collectively, our studies unveil a novel function of TGF␤/PDK4 in mediating drug resistance in colorectal cancer.
Knockdown of Expression of PDK1-3 by siRNAs-Individual pools of siRNAs against PDK1, PDK2, or PDK3 were purchased from Dharmacon/GE Healthcare Inc. (Lafayette, CO) and transfected into the cells using Dharmaconfect-2 (Dharmacon/GE Healthcare Inc.) following the protocol of the manufacturer.
Apoptosis Assays-Apoptosis was detected using a DNA fragmentation ELISA kit (Roche) or Annexin V staining. DNA fragmentation ELISA assays were performed according to the protocol of the manufacturer. Briefly, cells were seeded in 96-well

TGF␤/PDK4 Confers Drug Resistance
plates and treated with different concentrations of 5-FU for 72 h. The cells were stained with 3-(4,5-dimethylthiazol-2-yl)-2,5-diphenyltetrazolium bromide to determine cell numbers or lysed for the ELISA assays to determine apoptosis. The relative apoptosis was determined by dividing ELISA values by 3-(4,5dimethylthiazol-2-yl)-2,5-diphenyltetrazolium bromide values of each sample. -Fold changes were calculated by dividing relative apoptosis of 5-FU-treated samples by that of the controls.
For Annexin V staining, cells were seeded and treated with 2.5 g/ml 5-FU for 72 h. Cells were trypsinized and washed with PBS before staining with a cell apoptosis Annexin V FITC/ propidium iodide kit (Fisher Scientific). Cells were then analyzed by flow cytometry. The percentage of Annexin V-positive and PI-negative cells was calculated.
In Vivo Xenograft Model-Experiments involving animals were approved by the University of Nebraska Medical Center Institutional Animal Care and Use Committee. HCT116 cells (2.5 ϫ 10 6 ) infected with a lentiviral vector encoding PDK4 sh-3 or a scrambled shRNA were injected into the flanks of male athymic nude mice (4 -5 weeks old). One week after injection, 5-FU (25 mg/kg/day) or carrier was administered by i.p. injection for 5 consecutive days/week for 2 weeks (26,27). Tumor volumes were measured at the beginning of the treatment and every other day afterward until the mice were euthanized. The estimated tumor volumes (V) were calculated by the formula V ϭ W 2 ϫ L ϫ 0.5, where W represents the largest tumor diameter and L represents the next-largest tumor diameter. The relative tumor volumes (RTV) were calculated by RTV ϭ V x /V 0 , where V x is the volume at a given time, and V 0 is the volume at the beginning of the treatment (26).
TUNEL and Ki67 Staining-Formalin-fixed, paraffin-embedded tissue blocks of xenograft tumors were stained for TUNEL and Ki67 using a procedure described previously (28). Three xenograft tumors from each group were analyzed. Ten histologically similar fields were randomly selected from each slide for analysis. Apoptosis and proliferation of tumor cells were determined quantitatively by calculating the percentage of positively stained cells for TUNEL and Ki67 at ϫ20 magnification, respectively.
Determination of PDK4 Expression and Smad2 Phosphorylation in Human Tissue Samples-Formalin-fixed, paraffin-embedded blocks of human colorectal adenocarcinomas were obtained from the files of the Department of Pathology and Microbiology at the University of Nebraska Medical Center. The ages of all patients (including both men and women) were between 55-85 years. The cancer patients received neoadjuvant chemoradiotherapy prior to surgical removal of the tumors. Moderate responders refers to patients who have residual cancer with evident tumor regression but with more than single cells or rare small groups of cancer cells remaining; poor or non-responders refers to those who have extensive residual cancer with no evident tumor regression after neoadjuvant therapy. The study was performed with the approval of the ethics committee (institutional review board).
Immunohistochemistry (IHC) staining was performed to examine PDK4 expression and Smad2 phosphorylation in tumor samples following the Novolink TM Min Polymer Detection System kit protocol (Leica). Briefly, slides were subjected to antigen retrieval using Novocastra Epitope Retrieval Solutions (pH 6), followed by incubation with an anti-PDK4 or anti-p-Smad2 antibody overnight at 4°C. The next day, slides were developed with 3,3Ј-diaminobenzidine (DAB) after incubation with Novolink polymer for 30 min. Finally, the sections were counterstained with hematoxylin. For each slide, 10 randomly chosen tumor fields were captured at ϫ40 magnification. The staining intensity was quantified with Imagescope software (Aperio Technologies, Inc.) as described previously (29).
Statistical Analysis-Statistical analyses were performed using two-way ANOVA or Student's t test.

Expression of PDK4 Positively Correlates with 5-FU Resistance in Colon Cancer
Cells-To understand the mechanisms and identify the determinants underlying drug resistance, we examined a panel of colon cancer cell lines for their response to 5-FU treatment. Among the cell lines tested, RKO and HCT116 cells were much more sensitive to 5-FU-induced apoptosis than FET and CBS cells. DNA fragmentation assays revealed that the induction of apoptosis by 5-FU treatment was much higher in RKO and HCT116 cells than in FET and CBS cells (Fig. 1A). Annexin V/PI staining was performed to determine the percentage of early apoptotic cells after 5-FU treatment. The 5-FUmediated increase of early apoptotic cells represented by the Annexin Vϩ/PIϪ cell population was markedly higher in RKO and HCT116 cells than in FET and CBS cells (Fig. 1B). Consistent with these results, 5-FU induced a significant increase of poly(ADP-ribose) polymerase (PARP) cleavage in RKO and HCT116 cells, whereas FET and CBS cells were relatively resistant to this induction (Fig. 1C).
Aberrant glucose metabolism has been shown to play a role in drug resistance (14,15). To identify the determining factors that mediate 5-FU response, we examined the expression of some key regulators of glucose metabolism. We found that PDK4 was differentially expressed in those cell lines. As shown in Fig. 1, D and E, 5-FU resistant FET and CBS cells expressed higher levels of PDK4 than 5-FU-sensitive RKO and HCT116 cells at both the mRNA and protein levels, indicating a positive correlation between PDK4 expression and drug resistance. However, expression of other PDKs, PDK1-3, was similar among those cell lines (Fig. 1D, right panel). Of note, PDK4 expression was much lower in immortalized human colon epithelial cells than in colon cancer cell lines (Fig. 1D, left panel). In addition, 5-FU treatment led to increased PDK4 mRNA and protein expression in drug-resistant FET and CBS cells but not in sensitive RKO and HCT116 cells (Fig. 1, F, left panel, and G). Expression of PDK1-3 was not affected in either resistant or sensitive cells (Fig. 1F, right panel). These results suggest that elevated PDK4 expression may contribute to 5-FU resistance in colon cancer cells.
Genetic or Pharmacological Inhibition of PDK4 Sensitizes Colon Cancer Cells to 5-FU-or Oxaliplatin-induced Apoptosis in Vitro-To further investigate the possibility that PDK4 plays an important role in drug resistance, expression of PDK4 was knocked down in HCT116 and CBS cells. Two shRNAs against PDK4 (sh-2 and sh-3) significantly reduced PDK4 expression in both cell lines ( Fig. 2A). Expression of PDK1-3 was not affected by PDK4 knockdown (data not shown). Knockdown of PDK4 expression sensitized CBS cells to 5-FU-induced apoptosis, as reflected by a 2-to 3-fold increase in apoptosis induction in PDK4 knockdown cells compared with control cells (Fig. 2B,  bottom panel). Consistently, the increase in PARP cleavage was enhanced in PDK4 knockdown cells (Fig. 2C, bottom panel). Although HCT116 cells were sensitive to 5-FU treatment, knockdown of PDK4 further enhanced its sensitivity to the treatment (Fig. 2, B and C, top panels). These results suggested that down-regulation of PDK4 expression, even in sensitive cells, could increase their response to 5-FU treatment. In contrast, knockdown of expression of PDK1, PDK2, or PDK3 individually had little effect on 5-FU sensitivity (Fig. 2D). As a complementary approach, PDK4 cDNA was introduced into HCT116 cells. Consequently, PDK4 expression was markedly increased (Fig. 2E, left panel). When treated with 5-FU, PDK4overexpressing cells displayed enhanced resistance to 5-FU-induced apoptosis compared with control cells (Fig. 2E, right  panel). Taken together, these results indicate that PDK4 mediates the 5-FU response of colon cancer cells.
To determine whether knockdown of PDK4 sensitizes colon cancer cells to other chemotherapeutic drugs, HCT116 cells were treated with oxaliplatin, an alkylating agent commonly used in combination with 5-FU for treating advanced colon cancer (30,31). Similar to 5-FU, oxaliplatin treatment induced more apoptosis in PDK4 knockdown cells than in control cells, as shown by DNA fragmentation assays and PARP cleavage (Fig. 3, A and B). Although 5-FU and oxaliplatin function through different mechanisms, these data suggest that PDK4 up-regulation may be a common mechanism for their resistance in colon cancer cells.
Dichloroacetate (DCA) is a nonspecific pharmacological inhibitor of mitochondrial PDK isoforms (32). DCA has been shown to attenuate 5-FU resistance in gastric cancer cells (14). In preclinical studies, different cancer cells showed different responses to DCA-induced apoptosis (32)(33)(34). We next investigated whether DCA would increase the effectiveness of 5-FU against colon cancer cells. Low concentrations of DCA or 5-FU alone showed a slight increase in apoptosis in FET and CBS cells, respectively. However, combined treatment with both significantly increased apoptosis compared with either one alone (Fig. 3C). The effect on PARP cleavage further confirmed the results (Fig. 3D). These data indicated that pharmacological inhibition of PDK4 potentiated the apoptotic effect of 5-FU in colon cancer cells. Taken together, these findings suggest that PDK4 is a promising target for overcoming the drug resistance of colon cancer cells.
To further understand the mechanisms by which knockdown of PDK4 expression or treatment with DCA sensitized colon cancer cells to 5-FU-inudced apoptosis, we examined the expression of effectors involved in apoptosis of colon cancer cells. Among the effectors tested, expression of both Bcl-2 and survivin was reduced by PDK4 knockdown, which was further decreased by 5-FU treatment (Fig. 2C). Similarly, combined treatment by 5-FU and DCA significantly decreased expression of Bcl-2 and survivin compared with either treatment alone (Fig. 3D). In addition, overexpression of PDK4 increased Bcl-2 and survivin expression in HCT116 cells (Fig. 2E, center panel).
These results indicate that PDK4 regulates the expression of Bcl-2 and survivin. It has been shown that Bcl-2 and survivin protect colon cancer cells from stress-induced apoptosis (35,36) and that 5-FU induces apoptosis in colon cancer cell lines with modulation of Bcl-2 family proteins (37). To demonstrate the potential role of Bcl-2 and survivin in 5-FU-induced apoptosis in PDK4 knockdown cells, Bcl-2 and survivin were ectopically expressed individually in CBS/PDK4 sh-2 cells. Increased expression of Bcl-2 or survivin reduced 5-FU-induced PARP cleavage in PDK4 sh-2 cells, whereas it had little effect in control cells (Fig. 3E). DNA fragmentation assays confirmed those results (Fig. 3F). Therefore, inhibition of PDK4 expression or activity increases 5-FU-induced apoptosis by suppressing the expression of anti-apoptotic proteins such as Bcl-2 and survivin.
Knockdown of PDK4 Expression Increases the Efficiency of 5-FU-induced Inhibition of Tumor Growth in Vivo-To test whether increased sensitivity to 5-FU-induced apoptosis by knockdown of PDK4 expression translates to enhanced drug efficacy in vivo, we utilized a colon tumor xenograft model. Because high concentrations of 5-FU can be toxic to mice, we chose to use HCT116 cells that are relatively sensitive to 5-FU so that a low dose of 5-FU could be used for the study, which would have limited adverse effect on mice. Knockdown of PDK4 in HCT116 cells has been shown to further sensitize the cells to 5-FU-inducd apoptosis (Fig. 2, B and C). Therefore, athymic nude mice were subcutaneously inoculated with 2.5 million control HCT116 cells or cells expressing a PDK4 shRNA. After 7 days, mice were randomly divided into two groups. One group was treated with 5-FU administered by i.p. injection and the other with the carrier. A 5-FU dose of 25 mg/kg/day was used because the control tumors showed little response to this dose, which would provide a good window to observe enhanced drug efficacy in vivo by knockdown of PDK4 expression. The treatment was for 5 consecutive days/week for 2 weeks (26,27). Throughout the treatment, the weight of the mice remained stable. Tumor growth and therapeutic sensitivity were monitored during the course of 5-FU treatment.  AUGUST 12, 2016 • VOLUME 291 • NUMBER 33

JOURNAL OF BIOLOGICAL CHEMISTRY 17409
Xenograft tumor growth curves showed that tumors with control cells (designated as control tumors) and those with PDK4 shRNA-expressing cells (designated as PDK4 KD tumors) grew at similar rates (Fig. 4A). However, there was a marked difference in the response of the tumors to 5-FU treatment. PDK4 KD tumors did not grow under drug treatment, whereas control tumors continued to grow at a steady rate (Fig.  4A). The difference in the degree of inhibition by 5-FU treatment between control and PDK4 KD tumors was significant (p Ͻ 0.001). These results indicate that 5-FU treatment was more effective in inhibiting the growth of PDK4 KD tumors than that of control tumors.
To examine whether PDK4 knockdown-mediated sensitization to apoptosis in vitro was associated with an increased 5-FU effect in vivo, TUNEL assays were performed to determine the apoptotic index of the tumors. TUNEL staining showed that, although the percentage of apoptotic cells was similar in control and PDK4 KD tumors, the increase in apoptotic cells induced by 5-FU treatment was significantly higher in PDK4 KD tumors than in control tumors (8.4-fold versus 2.6-fold, Fig. 4B). In addition, Ki67 staining showed that 5-FU treatment decreased the percentage of proliferative cells in PDK4 KD tumors by 75% and in control tumors by 38% (Fig. 4C). These studies indicated that the effect of PDK4 knockdown on drug efficacy was a combined result of both increased sensitivity to 5-FU-induced apoptosis and inhibition of proliferation. Therefore, these in vitro and in vivo results demonstrate an important role for PDK4 in mediating the drug resistance of colon cancer cells.

TGF␤/PDK4 Confers Drug Resistance
TGF␤ Signaling Mediates Drug Resistance by Regulating PDK4 Expression-Based on the in vitro and in vivo studies described above, PDK4 contributes to the drug resistance of colon cancer cells. Therefore, it is critical to elucidate how its expression is regulated, which would provide important information to increase the efficacy of drug treatment. One important difference between 5-FU-sensitive and -resistant cells is TGF␤ signaling. Although 5-FU-sensitive RKO and HCT116 cells are defective in TGF␤ signaling because of the mutations in TGF␤ RII (3), 5-FU resistant FET and CBS cells are responsive or partially responsive to TGF␤ signaling, respectively (36,38). This suggests that the TGF␤ signaling pathway may play a role in the 5-FU response. To determine whether this is the case, a dominant negative RII (DNRII) construct was transfected into FET cells to inactivate TGF␤ signaling (6,36). Complementarily, wild-type RII cDNA was introduced into HCT116 cells to restore TGF␤ signaling (5). As shown in Fig.  5A, FET-DNRII cells displayed higher sensitivity to 5-FU than FET-Vec control cells, as indicated by increased apoptosis induced by 5-FU treatment in DNA fragmentation assays. Consistently, 5-FU induced a significant increase in PARP cleavage in FET-DNRII but not in FET-Vec cells (Fig. 5A, right panel). In contrast, expression of RII in HCT116 cells conferred them increased resistance to 5-FU treatment, as reflected by reduced apoptosis in DNA fragmentation assays and decreased PARP cleavage in response to 5-FU treatment (Fig. 5B). These results indicate that TGF␤ signaling contributes to 5-FU resistance in colon cancer cells.
Given that FET and CBS cells with active TGF␤ signaling express higher levels of PDK4 than RKO and HCT116 cells with defective TGF␤ signaling (Fig. 1, D and E), we next investigated whether TGF␤ regulates PDK4 expression. Treatment with TGF␤ increased PDK4 expression at both the protein and mRNA levels in FET-Vec and HCT116-RII cells with active TGF␤ signaling, whereas it had no effect in FET-DNRII and HCT116-Vec cells with an inactivated TGF␤ pathway (Fig. 5C). Next we determined whether Smad2 and Smad3 play a role in TGF␤-mediated PDK4 expression. Expression of Smad2 and Smad3 was knocked down individually by shRNAs in FET cells (Fig. 5D, left panel). Reduced expression of Smad2 or Smad3 attenuated or abrogated TGF␤-induced PDK4 expression (Fig.  5D, right panel), indicating that Smad2/3 contributes to TGF␤mediated PDK4 expression.
As shown in Fig. 1, F and G, 5-FU treatment led to increased PDK4 expression in FET and CBS cells but not in RKO and HCT116 cells. To determine whether TGF␤ signaling contributes to 5-FU-induced PDK4 expression, the effect of 5-FU treatment on PDK4 expression was examined in FET-Vec, FET-DNRII, HCT116-Vec, and HCT116-RII cells. Although inactivation of TGF␤ signaling by DNRII in FET cells abrogated 5-FU-induced PDK4 expression, restoration of TGF␤ signaling by expressing TGF␤ RII in HCT116 cells led to increased PDK4 expression by 5-FU treatment (Fig. 5E). These results indicate that 5-FU-induced PDK4 expression is dependent on the active TGF␤ signaling pathway. Taken together, it suggests that 5-FU may increase PDK4 expression by activating TGF␤ signaling.  AUGUST 12, 2016 • VOLUME 291 • NUMBER 33

JOURNAL OF BIOLOGICAL CHEMISTRY 17411
To test whether this is the case, FET cells were treated with 5-FU, and Smad2 phosphorylation, an indicator of TGF␤ activity, was determined. TGF␤ treatment was used as a control. As shown in Fig. 5F, 5-FU increased p-Smad2 to a similar level as TGF␤ treatment, indicating that 5-FU activates the TGF␤ signaling pathway.
These results indicate that one of the mechanisms by which TGF␤ mediates 5-FU resistance is through the induction of PDK4 expression and that elevated PDK4 expression contributes to 5-FU resistance in colon cancer cells with active TGF␤ signaling.
PDK4 Expression and TGF␤ Signaling Positively Correlate with Chemoresistance in Colorectal Cancer Specimens-To determine the clinical relevance of PDK4 expression in human cancer, we extended our analyses by assaying PDK4 expression in human colorectal adenocarcinoma specimens using IHC staining. We first verified an anti-PDK4 antibody used for IHC staining. IHC analysis of PDK4 expression was performed in control tumors and PDK4 KD tumors (Fig. 4). The intensity of PDK4 staining was much lower in PDK4 KD tumors than in control tumors, indicating the specificity of the anti-PDK4 antibody (Fig. 4D). IHC staining was then performed on paraffin sections prepared from 18 patients who had received neoadjuvant chemoradiotherapy. Four patients had sigmoid colon cancer and the rest rectal cancer. The patients were at different stages, and none of them had microsatellite instability. Among them, eight patients had a moderate response, and the other 10 had no response or a poor response. PDK4 staining varied significantly in samples from the moderate response and non-or poor response groups (Fig. 6A, top panels). Quantification of the staining showed that the average intensity of PDK4 staining was ϳ5.0-fold higher in non-or poor responders than in moderate responders (Fig. 6B, top panel; ***, p Ͻ 0.001). These results indicate that expression of PDK4 positively correlates with chemoresistance in colorectal cancer patients.
Because TGF␤ signaling enhances 5-FU resistance in colon cancer cells (Fig. 5, A and B), Smad2 phosphorylation (p-Smad2) was determined by IHC staining in the same set of tissue samples. p-Smad2 staining was significantly higher in non-or poor responders than in moderate responders (Fig. 6A,  bottom panels). Quantification of the staining showed that the average intensity of p-Smad2 staining was ϳ2.7-fold higher in the non-or poor responder groups than in the moderate responder group (Fig. 6B, bottom panel; ***, p Ͻ 0.001), indicating that the activation of the TGF␤ pathway is associated with chemotherapy resistance in colorectal cancer.
Given that TGF␤ increases PDK4 expression in 5-FU-resistant colon cancer cells (Fig. 5C), we next determined the relationship between PDK4 expression and Smad2 phosphorylation in patient samples. The study revealed a strong and positive correlation between PDK4 and p-Smad2 staining intensity ( Fig.  6C; Pearson r ϭ 0.8545; ***, p Ͻ 0.001). These results indicate that TGF␤-mediated up-regulation of PDK4 expression is relevant to the chemoresistance of colorectal tumors. Taken together with the in vitro and in vivo results in colon cancer cells, our studies demonstrate that the TGF␤/PDK4 signaling axis plays an important role in the drug resistance of colorectal cancer.

Discussion
The switch of glucose metabolism for energy production from oxidative phosphorylation to aerobic glycolysis is a hallmark of cancer (13,16). This metabolic switch is an important step to acquire aberrant survival capacity under stress conditions such as starvation or hypoxia (18,19). Although this switch has also been implicated in drug resistance, the molecular mechanisms are not well understood. PDK1-4 are a group of enzymes that control this metabolic switch by phosphorylating and thus inactivating PDH. In this study, we have made the novel findings that PDK4 mediates the response of colon cancer cells to the chemotherapeutic agent 5-FU and that TGF␤ signaling confers drug resistance through up-regulation of PDK4 expression. Our studies indicate that PDK4 is expressed at higher levels in 5-FU-resistant cells than in 5-FU-sensitive cells. 5-FU induces PDK4 expression in a TGF␤ signaling-dependent manner. Knockdown of PDK4 expression increases the sensitivity to 5-FU-induced apoptosis in vitro and to 5-FU-mediated inhibition of tumor growth in vivo. Importantly, studies of patient samples indicate that PDK4 expression and TGF␤ activation are positively correlated with each other and with chemoresistance in colorectal cancer. Because resistance to 5-FU treatment is one of the major causes for the failure of chemotherapy in treating advanced colorectal cancer (39 -41), the discovery of the TGF␤/PDK4 signaling axis as a contributing factor of drug resistance would identify TGF␤ signaling and PDK4 as potential targets of cancer therapy and warrant the development of novel PDK4-specific inhibitors to overcome chemoresistance and increase the efficacy of chemotherapeutic treatment of colorectal cancer.
It has been reported that metabolic stress such as starvation, fasting, and glucose deprivation markedly up-regulates PDK4 levels and that expression of PDK4 is induced by dietary lipids and hormones and inhibited by insulin (23). Transcription of PDK4 is regulated by many transcription factors, including forkhead box protein Os, estrogen-related receptors, peroxisome proliferator-activated receptors and the PPAR␥ coactivator 1 (42). Therefore, PDK4 expression is regulated by distinct pathways in different cells, highlighting cell type differences in the control of this key metabolic enzyme. In this study, we show, for the first time, that TGF␤ increases PDK4 mRNA and protein expression in colon cancer cells and that expression of PDK4 is induced by 5-FU in a TGF␤-dependent manner. Our study thus unveils another mechanism by which PDK4 expression is regulated.
Previous studies reported that DCA effectively reduced tumor growth of lung and breast cancer cells both in vitro and in vivo (32,33). It was suggested that DCA could rapidly translate to cancer clinical trials because it has been used to treat lactic acidosis and found to be relatively safe in humans (43,44). However, the doses of DCA required to induce apoptosis in cancer cells are very high and unlikely to be achieved clinically without causing significant side effects (45). When combined with 5-FU, low doses of DCA and 5-FU could significantly induce apoptosis in colon cancer cells. Therefore, DCA may not be an appropriate single agent for colon cancer treatment but could be used to potentiate the effect of 5-FU. Our study pro-vides proof of concept that the combination of DCA and 5-FU could be a potentially effective therapy for colorectal cancer. Moreover, because PDK1-3 appear to be not involved in the drug resistance of colon cancer cells, development of PDK4specific inhibitors would further reduce cytotoxicity and increase the efficacy of chemotherapy in colorectal cancer treatment.
It has been shown that TGF␤ signaling is associated with chemotherapy resistance in different types of cancer (10 -12). One of the proposed mechanisms is through the induction of epithelial-mesenchymal transition by TGF␤ (11). In addition, loss or inactivation of Smad4 has been shown to be a predictive marker for resistance to 5-FU-based chemotherapy in colorectal cancer (46,47). TGF␤ activates non-canonical signaling (i.e. PI3K, ERK, JNK, etc.) in Smad4-deficient cells (48), which could contribute to drug resistance. In our studies, the colon cancer cells used express wild-type Smad4, and TGF␤ does not induce EMT in those cells (data not shown). Instead, TGF␤ increases PDK4 expression, which confers resistance to drug treatment.
PDK4 has been reported to suppress cell proliferation by inhibiting the PDH-mediated TCA cycle (49). In lung cancer cells, inhibition of PDK4 promotes EMT associated with erlotinib resistance (50). On the other hand, PDK4 has been shown to contribute to anoikis resistance in cancer cells (51). A recent study reported that PDK4 promotes tumorigenesis through activation of the cAMP-response element-binding protein-ras homolog enriched in brain-mTORC1 signaling cascade (52). Therefore, like TGF␤, the function of PDK4 is cell context-dependent, and PDK4 may also function as a double-edged sword in cancer. Although it could inhibit cell proliferation or sensitize cancer cells to drug treatment, it could also confer drug resistance or promote anoikis and tumorigenesis. We have shown in this study that PDK4 inhibits 5-FU-induced apoptosis by increasing the expression of Bcl-2 and survivin. Future studies will determine the mechanisms by which PDK4 regulates their expression.
In summary, our discovery of the TGF␤/PDK4 axis as a mediator of the drug response in colorectal cancer provides a potential therapeutic target against drug resistance. Inhibition of TGF␤ signaling or PDK4 expression/activity could increase chemosensitivity in colorectal cancer treatment. Therefore, development of novel PDK4-specific inhibitors would greatly increase anti-tumor specificity and enhance the efficacy of chemotherapy in colorectal cancer treatment.