Mechanism of Cancer Cell Adaptation to Metabolic Stress

Gastric cancer is the second most common cancer worldwide and has a poor prognosis. To determine the mechanism of adaptation to metabolic stress in cancer cells, we used gastric cancer as a model system to reveal the potential signaling pathways involved. Two-dimensional polyacrylamide gel electrophoresis coupled with ESI-Q-TOF MS/MS analysis was used to identify differentially expressed proteins between gastric tumor tissues and the corresponding noncancerous tissues. In total, 107 spots with significant alteration (±over 2-fold, p < 0.05) were positively identified by MS/MS analysis. Altered expression of representative proteins was validated by RT-PCR and Western blotting. Cluster analysis of the changed proteins revealed an interesting group of metabolic proteins, which suggested accumulation of triiodothyronine (T3; the major functional component of thyroid hormone) and overexpression of hypoxia-induced factor (HIF) in gastric carcinoma. These observations were further confirmed by electrochemiluminescence immunoassay and immunohistochemistry. T3-induced expression of HIF1-α and vascular endothelial growth factor was further verified using a gastric cancer cell line and in vivo mouse model. Because the early accumulation of HIF1-α was found to be independent of de novo transcription, we also found that the cytosolic cascade phosphatidylinositol 3-kinase/Akt pathway sensitive to T3 stimulus was involved. Furthermore we demonstrated that T3-induced overexpression of HIF1-α was mediated by fumarate accumulation and could be enhanced by fumarate hydratase inactivation but inhibited by 2-oxoglutarate. These results provide evidence for alteration of metabolic proteins and dysfunction of thyroid hormone regulation in gastric tumors, and a novel thyroid hormone-mediated tumorigenic signaling pathway is proposed. Our findings are considered a significant step toward a better understanding of adaptations to metabolic stress in gastric carcinogenesis.

Cellular hypoxia and metabolic stress have been observed in many cancer types (1)(2)(3). Hypoxia-induced factor (HIF) 1 is considered a central regulator of oxygen homeostasis. HIF1-␣ or HIF2-␣, together with HIF1-␤, forms an active transcription complex that controls numerous target genes, which have roles in glycolysis, angiogenesis, and tumor metastasis (4). Previous studies indicated that HIF1-␣ is overexpressed in many cancer types, such as colon, lung, renal, and thyroid gland, whereas HIF proteins mediate cell adaptation to hypoxia (1)(2)(3). Both HIF1-␣ and HIF2-␣ are labile under normoxic conditions; this is because of proteasomal degradation following their oxygen-dependent ubiquitination by ubiquitin ligase complex targeted to HIF by the von Hippel-Lindau protein (VHL) (5,6). VHL recognition of HIF1-␣ requires the enzymatic hydroxylation of two converted residues on HIF1-␣ mediated by HIF prolyl hydroxylase (HPH) (7,8). HPH activity requires the cofactors ascorbate and iron and the cosubstrates 2-oxoglutatrate and molecular oxygen. This elegant system is the basis for HIF stabilization under hypoxia as HPH function and subsequently VHL recognition of hypohydroxylated HIF is compromised in the absence of oxygen (9,10).
Thyroid hormone (TH) plays an important part in hormone homeostasis and is the major regulator of mitochondrial activity. TH affects gene expression through binding to TH receptor elements in promoter regions. A recent study indicated that triiodothyronine (T 3 ), the major functional component of TH, could also influence mitochondrial physical activities by activating the cytoplasmic pathway (11). T 3 -induced HIF overexpression was individually observed in a hepatoma cell line and primary fibroblasts. However, the underlying mechanism remains poorly defined (12,13). In view of the fact that a universal feature of mitochondrial respiratory confusion is the involvement of T 3 stimuli (11), identification of a rational signaling pathway explaining the relationship among T 3 dysregulation, HIF1-␣ overexpression, glycolysis elevation, and mitochondrial dysfunction is needed.
Both the tricarboxylic acid cycle and glycolysis play essential roles in cellular energy provision. A primary function of the tricarboxylic acid cycle is oxidation of pyruvate supplied by glycolysis. Energy released from the tricarboxylic acid cycle is finally fixed in ATP during oxidative phosphorylation (14,15). Enhanced aerobic glycolysis was observed in cancer cells as early as the 1930s (16), and evidence linking disordered energy metabolism and carcinogenesis is accumulating (17,18). However, there is still a lack of molecular evidence, for example on the profiling of altered enzymes to illustrate the carcinogenic mechanism underlying the metabolic shift.
Fumarate is an important intermediate in the tricarboxylic acid cycle. During the tricarboxylic acid cycle, fumarate is converted to malate by fumarate hydratase (FUMH), an essential metabolic enzyme. Previous studies have indicated that the intracellular level of fumarate is tightly controlled by FUMH and that silencing the expression of FUMH results in rapid fumarate accumulation (19). Recent studies in renal carcinoma suggested that accumulation of fumarate coupled with hereditary FUMH inactivation significantly abrogates VHL degradation of HIF1-␣ by impairing HPH activity (20). However, there are no data supporting a similar role for fumarate in gastric carcinoma.
Gastric cancer is a common malignant tumor, representing the second major cause of cancer-related deaths worldwide. The problem is particularly marked in Asia where gastric cancer comprises 23% of malignant tumor death (21). However, early diagnosis of gastric cancer is problematic, and most gastric cancers are naturally resistant to anticancer drugs. Although chronic gastritis and reiterative ulcer have been recognized as the major risks for the development of gastric cancer, research into alternative mechanisms of gastric carcinogenesis is still required (22).
The 2-DE-based proteomics approach provides a powerful tool to simultaneously analyze the expression levels of hundreds of proteins in tissue samples. This may enable the identification of cancer-related proteins for therapeutic intervention and the establishment of biomarkers for early diagnosis (23,24). Several groups have carried out proteomics studies of gastric carcinoma, and a number of proteins with altered expression levels have been identified. Although abnormalities in individual proteins have been intensively studied, a more detailed profiling of gastric cancer is still needed to identify signaling pathways involved in responses to metabolic stress and those underlying the multistep process of carcinogenesis itself (21,25). In the present study, we demonstrated that gastric carcinoma shows accumulation of T 3 and elevated expression of HIF1-␣ proteins as well as the protein product of the HIF-regulated gene VEGF. Furthermore we showed that T 3 accumulation results in HIF1-␣ overexpression both in vitro and in vivo. Finally we confirmed that T 3 may elevate HIF1-␣ through fumarate accumulation; this elevation is apparently enhanced by T 3 -mediated FUMH inactivation and inhibited by 2-oxoglutatrate treatment.

MATERIALS AND METHODS
Gastric Tumor Tissue Samples-40 pairs of gastric carcinoma and the adjacent noncancerous gastric samples were obtained from West China Hospital, Sichuan University (Chengdu, China). The specimens were diagnosed histologically after staining with hematoxylin and eosin, and the surgical-pathologic stage was determined according to the TNM classification system of the International Union against Cancer. Detailed information of the patients, such as age, sex, histodifferentiation, and surgical-pathologic stage, is listed in Table I. All pairs of samples were immediately frozen in liquid nitrogen prior to experiments. All 40 pairs of samples were subjected to the validation experiments, including (i) Western blotting detection of transthyretin (TTHY), hypoxia up-regulated protein 1 (HYOU1), FUMH, and pyruvate kinase muscle isozyme (KPYM); (ii) ECLI of T 3 ; (iii) HPLC assay of fumarate and citrate; and (iv) immunohistochemical analysis of HIF1-␣ and VEGF. In addition, a subsample comprising six pairs of samples was randomly selected for 2-DE analysis. This study was approved by the Institutional Ethics Committee of Sichuan University, and informed consents were obtained from all patients prior to analysis.
2-DE-100 mg of tissue sample was cut into pieces about 2 mm 3 , homogenized in liquid nitrogen, and lysed in 1 ml of lysis buffer (7 M urea, 2 M thiourea, 4% CHAPS; Bio-Rad) containing protease inhibitor mixture 8340 (Sigma). Samples were then kept on ice, sonicated in 10 cycles each consisting of 10-s sonication followed by a 30-s break, and finally held for 30 min on ice with occasional Vortex mixing. After centrifugation at 14,000 rpm for 1 h at 4°C, proteins were precipitated with cold acetone at 20°C for 1 h and then dissolved with rehydration buffer (8 M urea, 2 M thiourea, 4% CHAPS, 100 mM DTT, 2% ampholyte). Protein concentrations were determined using the DC protein assay kit (Bio-Rad). Individual sample concentrations were adjusted by dilution in the same rehydration buffer. Samples were either applied immediately to IEF or stored at Ϫ80°C in aliquots prior to analysis. Protein samples (2.5 mg, 300 l) were applied to IPG strips (17 cm, pH 3-10, nonlinear, Bio-Rad) using a passive rehydration method. After 12-16 h of rehydration, the strips were transferred to an IEF cell (Bio-Rad). IEF was performed as follows: 250 V for 30 min, linear; 1000 V for 1 h, rapid; linear ramping to 10,000 V for 5 h; and finally 10,000 V for 6 h (26). Once IEF was completed, the strips were equilibrated in equilibration buffer (25 mM Tris-HCl, pH 8.8, 6 M urea, 20% glycerol, 2% SDS, 130 mM DTT) for 15 min followed by the same buffer containing 200 mM iodoacetamide instead of DTT for another 15 min (27). The second dimension was performed using 12% SDS-PAGE at 30-mA constant current per gel. The gels were stained using CBB R-250 (Merck). For 2-DE analysis, each of the paired samples were run in triplicate to ensure the consistency of the data.
Image Analysis-The images were scanned with a Bio-Rad GS-800 scanner (400 -750 nm), and the differentially expressed proteins were identified using the PDQuest 2-D analysis software (Bio-Rad). The quantity of each spot in a gel was normalized as a percentage of the total quantity in the map according to its OD value. Only those spots that changed consistently and significantly (more than 2-fold) were selected for MS/MS analysis.
Tryptic In-gel Digestion-In-gel digestion of proteins was carried out using mass spectrometry grade trypsin gold (Promega, Madison, WI) according to the manufacturer's instructions. Briefly spots were cut out of the gel (1-2-mm diameter) using a razor blade and destained twice with 100 mM NH 4 HCO 3 , 50% ACN at 37°C for 45 min in each treatment. After dehydration with 100% ACN and drying, the gels were preincubated in 10 -20 l of trypsin solution (10 ng/l) for 1 h. Then adequate digestion buffer (40 mM NH 4 HCO 3 , 10% ACN) was added to cover the gels, which were incubated overnight at 37°C (12-14 h). Tryptic digests were extracted using Milli-Q water followed by double extraction with 50% ACN, 5% TFA for 1 h each time. The combined extracts were dried in a SpeedVac concentrator (Thermo Scientific) at 4°C. The samples were then subjected to mass spectrometry.
ESI-Q-TOF-Mass spectra were acquired using a Q-TOF mass spectrometer (Micromass, Manchester, UK) fitted with an ESI source (Waters). Tryptic digests were dissolved in 18 l of 50% CAN. MS/MS was performed in a data-dependent mode in which the top 10 most abundant ions for each MS scan were selected for MS/MS analysis. Trypsin autolysis products and keratin-derived precursor ions were automatically excluded. The MS/MS data were acquired and processed using MassLynx software (Micromass), and MASCOT was used to search the database. Database searches were carried out using the following parameters: database, Swiss-Prot; taxonomy, Homo sapiens; enzyme, trypsin; mass tolerance, Ϯ0.1 Da; MS/MS tolerance, Ϯ0.05 Da; and an allowance of one missed cleavage. Fixed modifications of cysteine carbamidomethylation and variable modifications of methionine oxidation were allowed. The data format was selected as Micromass peak list, and the instrument was selected as ESI-Q-TOF. Proteins with probability-based MOWSE scores exceeding their threshold (p Ͻ 0.05) were considered to be positively identified. To eliminate the redundancy of proteins appearing in the database under different names or accession numbers, the one protein member with the highest MASCOT score and belonging to the species H. sapiens was further selected from the relevant multiple member protein family. Cell Culture and Drug Treatment-Human gastric cancer cell line MKN28 was purchased from ATCC. MKN28 was cultured in RPM I1640 medium (Invitrogen) supplemented with 10% fetal bovine serum. At 80% cell confluence, the medium was replaced with THdepleted bovine serum produced by treatment of fetal bovine serum with anion exchange resin. Treatment with T 3 , fumarate, and 2-oxoglutarate was performed at 48 h, respectively, after replacement of the TH-depleted cell culture medium.
Animal Treatment-8-week-old BALB/c mice were used for T 3 treatment. T 3 was mixed in a paste containing 5% carboxymethylcellulose and 1% Tween 20 and at a final concentration at 2 mg/ml. 300 l of paste/mouse was fed directly into the stomach with a feeding needle. Such treatment was performed every 24 h until sacrifice, and gastric tissue was immediately fixed in formalin or frozen in liquid nitrogen.
Semiquantitative RT-PCR-Total RNA was isolated using TRIzol reagent (Invitrogen) according to the manufacturer's instructions. First strand cDNA was reverse transcribed from 1 g of total RNA in a final volume of 20 l using reverse transcriptase and random hexamers from ExScript TM reagent kit (TaKaRa, Dalian, China) according to the manufacturer's instructions. Primers were designed using Primer Premier 5 software. Primers and annealing temperature are listed in Table  II. PCR was performed with rTaq (TaKaRa) in a DNA thermal cycler (Bio-Rad) according to a standard protocol as follows: one cycle of 95°C for 3 min; 23 cycles of 94°C for 45 s, annealing for 45 s, and 72°C for 1 min; a final extension at 72°C for 10 min; and holding at 4°C. The amount of cDNA used for each PCR was 20 ng in a 25-l reaction volume. The PCR products (5 l) were analyzed by electrophoresis through 2% agarose gels and visualized by SYBR Gold (Molecular Probes, Eugene, OR) staining.
T 3 ECLI-Proteins were extracted in RIPA buffer and quantified by the DC protein assay kit (Bio-Rad). The T 3 ECLI kit was purchased from Roche Applied Science, and the experimental procedures were as follows. 30 l of sample and a T 3 -specific antibody labeled with a ruthenium complex were used. Bound T 3 was released from the binding proteins in the sample by 8-anilino-1-naphthalenesulfonic acid. After addition of streptavidin-coated microparticles and biotinylated T 3 , the remaining free binding sites of the labeled antibody become occupied with formation of an antibody-hapten complex. This complex then bound to the solid phase via interaction of biotin and streptavidin. The reaction mixture was aspirated into the measuring cell where the microparticles were magnetically captured onto the surface of the electrode. Unbound substances were then removed with ProCell. Application of a voltage to the electrode then induced chemiluminescent emission that was measured by a photomultiplier. Results were determined via a calibration curve that is instrumentspecifically generated by two-point calibration and a master curve provided via the reagent barcode. T 3 ECLI was performed in triplicate for each sample.
HPLC-The Waters 2695-2487 LC system (Waters Corp.) was utilized with an Atlantis C 18 column operated at a flow rate of 1 ml/min. LC was used in the isocratic mode; the mobile phase consisted of 95% potassium dihydrogen phosphate (20 mM), 5% acetonitrile. The retention time was normalized using a standard substance.
The raw data were analyzed by Empower software, and the concentration was calculated according to a predetermined formula corresponding to the area of the peak.
FUMH Enzyme Assay-Measurement of fumarate hydratase activity has been described elsewhere (28). Briefly cell lysate containing a defined amount of total protein (30 -100 g) was added to a final volume of 200 l of assay buffer containing 25 mM HEPES-KOH (pH 7.5; Sigma), 0.4 mM NADP (Sigma), 4 mM MgCl 2 , 5 mM KH 2 PO 4 , 0.4 unit/ml NADP:malic enzyme (Sigma), 10 mM fumarate. Increases in absorbance at 340 nm, due to formation of NADPH, were monitored at room temperature in UV light-transparent 96-well flat bottom microtiter plates (Costar) in an ELϫ808 microplate reader (Bio-Tek). Fumarate hydratase activity is expressed as nmol of NADPH formed/g of protein/min. Samples lacking either cellular protein or fumarate were used to determine blank values.
Immunohistochemistry-Immunohistochemistry was performed using the Dako EnVision Systems (Dako Cytomation GmbH, Hamburg, Germany). Consecutive paraffin wax-embedded tissue sections (3-5 m) were dewaxed and rehydrated. Antigen retrieval was performed by pretreatment of the slides in citrate buffer (pH 6.0) in a microwave oven for 12 min. Thereafter slides were cooled to room temperature in deionized water for 5 min. Endogenous peroxidase activity was quenched by incubating the slides in methanol containing 0.6% hydrogen peroxide followed by washing in deionized water for 3 min after which the sections were incubated for 1 h at room temperature with normal goat serum and subsequently incubated at 4°C overnight with the primary antibodies rabbit anti-TTHY (diluted 1:1000; Santa Cruz Technology) and rabbit anti-VEGF (diluted 1:1000; Santa Cruz Biotechnology). Next the sections were rinsed with washing buffer (TBS with 0.1% bovine serum albumin) and incubated with horseradish peroxidase-linked goat anti-rabbit antibodies followed by reaction with diaminobenzidine and counterstaining with Mayer's hematoxylin. Immunostaining was detected using 3,3Ј-diaminobenzidine substrate solution (Dako Cytomation GmbH) according to the manufacturer's instructions. RESULTS

2-DE Profiling of Differentially Expressed Proteins in Gastric
Cancer-The protein expression profiles in gastric carcinomas and adjacent normal tissues were obtained by 2-DE. Gel images and representative 2-DE maps for a subsample of six pairs of samples, which were unambiguously matched by the PDQuest software, are shown in Fig. 1A. Approximately 1500 -1600 protein spots were detected by CBB R-250 staining in a single 2-DE gel. The quantity of each spot in a gel was normalized as a percentage of the total quantity of all spots in the gel. In comparison with 2-DE patterns, differentially expressed proteins were defined as statistically meaningful (p Ͻ 0.05) based on both of the following two criteria: 1) intensity alterations Ͼ2.0-fold and 2) recurrence more than three times in the six pairs of samples examined. By applying these criteria, a total of 107 spots were identified as differentially expressed, and of these 57 proteins were up-regulated, whereas 50 proteins were down-regulated in gastric carcinoma. Master gel images were generated by PDQuest software (Fig. 1A), and representative spots, labeled with arrows, are shown in Fig. 1B. 20 representative proteins (10 up-regulated and 10 down-regulated in tumor samples) with most significant alteration are shown in Fig. 1D, and their corresponding spots are boxed and enlarged in the surrounding area (Fig. 1C).

Mass Spectrum Identification of Differentially Expressed
Proteins-Differentially expressed protein spots were subsequently subjected to MS/MS analysis. The MS/MS data were retrieved using the search algorithm MASCOT against the ExPASy protein sequence database. The proteins were identified using a number of criteria including pI, molecular weight, the number of matched peptides, sequence coverage, and MOWSE score; all the information is listed in Table III. Proteins were classified into different groups based on their functions and subcellular localization (Fig. 1, E and F).
Notably significant alterations were found in a group of metabolic proteins. These proteins function in diverse metabolic processes, such as thyroid hormone regulation, glycolysis, tricarboxylic acid cycle, oxidative phosphorylation, electron transport, and fatty metabolism. Cluster maps (Fig.  1G) showing altered expression of these proteins were generated by Cluster software, and detailed information, such as subcellular location and biological function, is listed in Table IV.
Validation of the Altered Proteins by Semiquantitative RT-PCR and Western Blotting Analysis-To examine whether the proteomics identification of these metabolic proteins corresponded to changes at the transcriptional level and translational level, seven proteins (HYOU1, TTHY, KPYM, 78-kDa glucose-regulated protein (GRP78), FUMH, ALDOA, and LDHA) with significant expression changes were chosen for validation by semiquantitative RT-PCR, and some of them (FUMH, TTHY, KPYM, and HYOU1) were further confirmed by Western blot. As shown in Fig. 2, A (RT-PCR) and B (Western blotting), the expression levels were found to be consistent with the observations in 2-DE analysis.
Accumulation of T 3 , Fumarate, and Citrate in Gastric Carcinoma-Of those changed proteins found to be differentially expressed in tumor tissues, TTHY has high affinity with TH and functions in TH transport into tissues (29). In addition, alteration of energy metabolic enzymes known to be THregulated (KPYM, F16P1, etc.) was observed (30). Consequently the hypothesis that a disturbed TH homeostasis is involved in gastric carcinoma was considered. To evaluate this hypothesis, 40 pairs of gastric tumor and normal tissues were subjected to ECLI to quantify their total T 3 . All of the donor patients had a normal plasma TH level by clinical examination. In total, 27 tumor samples displayed higher T 3 levels than their corresponding normal samples, and 19 tumor samples showed more than 1.5 times elevation.
As shown in Table IV, alteration of several tricarboxylic acid cycle-related enzymes was observed. Notably fumarate hydratase and citrate synthase were markedly repressed in gastric tumor. Therefore, the samples were examined to determine whether their respective substrates, fumarate and citrate, were relatively changed. HPLC was performed to measure the contents of fumarate and citrate in tissues. Interestingly fumarate and citrate levels were elevated in almost all gastric tumor samples. The -fold change (cancer/normal) of FIG. 1. A, representative 2-D gel images of human gastric carcinoma and adjacent normal tissues. Total protein extracts were separated on pH 3-10 nonlinear IPG strips in the first dimension followed by 12% SDS-PAGE in the second dimension and visualized by CBB staining. B, reference maps were generated from A using PDQuest software. 65 differentially expressed spots (31 up-regulated and 34 down-regulated) are marked with numbers. Information for each altered spot is reported in Table III. C, expression profile of the 20 significantly altered proteins with recurrence more than three times. The selected area was symmetrically boxed, and arrows indicate each protein spot or its theoretical location. The upper portion shows 10 proteins up-regulated in gastric carcinoma, and the lower portion shows 10 proteins down-regulated in gastric carcinoma. D, expression profile of the 20 altered proteins as shown in Fig. 1C. The intensities of spots were quantified using PDQuest 2-D analysis software. E, 107 identified proteins were functionally classified into 10 groups. Many were involved in metabolism (23%), protein folding (16%), proliferation and apoptosis (5%), cell skeleton (4%), and other functions (52%). F, the identified proteins were categorized into several protein groups according to their subcellular locations. 79% of the total proteins were located in the cytoplasm, and the remainder were situated in the nuclear or cell membrane. G, protein cluster map generated by Cluster software. Expression of proteins in the normal group was constant at 0, whereas proteins up-regulated in cancer tissue are in red, and the down-regulated proteins are in green. The intensity of the color green or red corresponds to the degree of alteration, respectively, according to the color strip at the bottom of the figure. T, tumor; N, normal. All the data were shown as mean Ϯ S.D.  Fig. 3A.

HIF1-␣ and VEGF Are Overexpressed in Gastric
Carcinoma-Two hypoxia stress proteins, HYOU1 and GRP78, were up-regulated in the proteomics analysis. Additionally clusters of enzymes involved in glycolysis were also elevated. As HIF-1␣ has been reported as a potential activator of those proteins (4), it was of particular interest to examine whether expression of HIF-1␣ was altered in gastric carcinoma, although a role for HIF-1␣ has not been reported previously for gastric tumor. 40 pairs of gastric tumor tissues, which were used previously in the T 3 assay, were fixed and prepared for immunohistochemical analysis. HIF1-␣ and its target oncoprotein VEGF (31) were significantly overexpressed in tumor tissues, especially in gastric glands. Continuous sections showed that the positive restrictions of HIF1-␣ and VEGF were well matched, and representative staining from tumor and normal tissues is shown in Fig. 3B. sion in hepatoma cell line HepG2 (12). Consequently assays were performed in the present study to determine whether or not such a relationship exists in gastric tumor. MKN28 cells were preadapted in a TH-deleted medium for 48 h to avoid interference by serum TH. MKN28 cells were subsequently treated with T 3 at a final concentration of 2 nM, and cells were harvested on a time course basis from 1 to 12 h. Expression of HIF1-␣ was monitored at both the mRNA and protein levels. As shown in Fig. 3C, expression of HIF1-␣ increased in a time-dependent manner. Detectable elevation appeared as early as 2 h and peaked at 5 h. Similar changes at the mRNA level were observed, but detectable elevation was delayed until 5 h (Fig. 3B).

T 3 Induces Up-regulation of HIF1-␣ in Vitro and in Vivo-Previous studies indicated that T 3 enhanced HIF1-␣ expres-
To further confirm T 3 -dependent HIF1-␣ induction in vivo, we treated BALB/c mice with T 3 , and the alterations of HIF1-␣ were monitored. T 3 -containing paste, at an appropriate dose, was fed directly into the stomach of mice every 24 h. Immunohistochemical analysis of the mouse stomach tissues revealed dramatic enhancement of HIF1-␣ in the gastric glands. Fig. 3B, elevated expression of HIF1-␣ at the protein level preceded that at the RNA level, suggesting that a nucleus-independent signaling pathway was probably responsible for its early accumulation. Additionally recent studies in a colon cancer cell line and fibroblast cells indicated that HIF1-␣ could be activated by both the PI3K/Akt pathway and the MAPK pathway under intracellular stimuli, such as reactive oxygen species and hypoxia, and extracellular stimuli, such as insulin and T 3 (12,13). In view of these observations, the response of the two pathways to T 3 stimuli was examined. MKN28 cells were treated with T 3 through a rapid time course, and phosphorylation site-specific antibodies were used to monitor activation of these pathways by Western blotting. As shown in Fig. 4, the phosphorylated form of PI3K and its downstream receptor Akt were dramatically elevated compared with their background form. However, the ratio between phosphorylated MAPK and background MAPK was almost unchanged. T 3 Induces Intracellular Fumarate Accumulation by Affecting the Expression and Activity of FUMH-T 3 is considered the major regulator of mitochondrial respiration. Consequently T 3 accentuation results in accumulation of intermediates during the tricarboxylic acid cycle (11). As fumarate is a major intermediate in the tricarboxylic acid cycle, work was next performed to examine whether T 3 affects the fumarate level. Toward this goal, T 3 -treated MKN28 cells were lysed by iterative sonication, and the intracellular fumarate level was monitored by HPLC. As expected, fumarate was rapidly elevated as early as 2 h post-treatment with T 3 , and the elevation was held at a level of ϳ2.5-fold for at least 48 h compared with untreated control. Representative HPLC maps for 3 h, 6 h, and untreated control are shown in Fig. 5A.

PI3K/Akt Pathway Is Involved in T 3 Signal Cytosolic Cascade-As shown in
Next the effect of T 3 treatment on FUMH expression was examined, especially in the context of determining the mechanism underlying T 3 -induced fumarate accumulation. Moderate repression of FUMH at the RNA level was detected at 24 h (Fig. 5B), but no significant changes were observed within 12 h (data not shown). These data indicate that FUMH repression may contribute to stabilizing a fumarate rich environment but was not responsible for early accumulation of fumarate.
Next the possibility that T 3 affects FUMH activity rather than its expression was considered. As shown in Fig. 5C, 2 nM T 3 reduced FUMH activity by ϳ30%, and such inhibitory behavior was rapid. Very strikingly, lysate of T 3 -untreated MKN28 cells showed a normal FUMH activity even in the presence of T 3 , which may suggest that a complete cell structure was necessary for T 3 -induced FUMH inactivation.
T 3 -induced HIF1-␣ Overexpression Is Fumarate-dependent-Because fumarate functioned as an inhibitor of HIF1-␣specific degradation by inactivating HPH in fibroblasts (20), it was necessary to examine whether T 3 -induced overexpression of HIF1-␣ is dependent on fumarate in gastric tumorigenesis. MKN28 cells were treated with fumarate alone or in combination with T 3 . Expression of HIF1-␣ was monitored by Western blotting and RT-PCR. As shown in Fig. 5D, rapid accumulation of HIF1-␣ protein was observed in as few as 30 min. However, the transcripts of HIF1-␣ were not changed at the transcriptional level, and this was consistent with the previous findings in this study (Fig. 3C).
Another intermediate in the tricarboxylic acid cycle, 2-oxoglutarate, is a competitive substrate against fumarate for HPH (20). Therefore, it was considered relevant to examine whether the induction of HIF1-␣ mediated by T 3 could be abolished by increasing the level of 2-oxoglutarate. MKN28 cells were treated with either a combination of 5 mM fumarate and 5 mM 2-oxoglutarate or a combination of 2 nM T 3 and 5 mM 2-oxoglutarate. As expected, 2-oxoglutarate completely abrogated HIF1-␣ induction mediated either by T 3 or fumarate (Fig. 6E). Indeed even low concentrations of 2-oxoglutarate significantly inhibit HIF1-␣ accumulation (data not shown). HPLC was performed to monitor intracellular accumulation of exogenetic fumarate in MKN28 cells, which indicated that the abolition of HIF1-␣ induction was not caused by blockade of fumarate absorption (data not shown).  (Fig. 6A). Similar alterations were observed in Western blot analysis. In the mouse model, VEGF was found to be significantly elevated after T 3 treatment, and up-regulation of VEGF was coincident with high expression of HIF1-␣ (Fig. 6B).   FIG. 3. A, ECLI and HPLC validation were performed to verify total T 3 (red), fumarate (brown), and citrate (blue) in 40 pairs of gastric tumor tissues. The intensity is presented in -fold change (cancer/ normal (C/N)). B, expression of HIF1-␣ and VEGF in gastric tumor was verified by immunohistochemistry. C, MKN28 cells were treated with 2 nM T 3 from 1 to 6 h and for 12 h, whereas normal MKN28 cells were used as untreated controls. RT-PCR and Western blotting were used to validate expression of HIF1-␣ at the RNA level and protein level. ␤-Actin was used as an equal loading control. D, BALB/c mice were fed with T 3 every 24 h until sacrificed. Expression of HIF1-␣ in mouse gastric tissue was examined by immunohistochemistry. All the data were shown as mean Ϯ S.D.

DISCUSSION
A 2-DE-and mass spectrum-based proteomics strategy provides high throughput simultaneous identification of hundreds of proteins (32,33). Earlier studies have aimed to profile altered proteins in gastric cancer, and several potential biomarkers have been obtained, for example MAD1L1 (33)(34)(35). In the present study, 107 proteins showed altered expression levels, and these proteins function in diverse metabolic processes. The work focused on four groups of proteins: those associated with TH regulation, hypoxia stress, tricarboxylic acid cycle, and glycolysis. This is the first report that T 3 , HIF1-␣, and fumarate are simultaneously elevated in gastric tumors.
Glycolysis has been shown to be elevated in almost all cancer types, even under normoxia, a phenomenon termed the Warburg effect (16). Blockade of the tricarboxylic acid cycle by hypoxia, uncoupling of oxidative phosphorylation, and genetic inactivation of key enzymes such as succinate dehydrogenase should enhance cellular reliance upon glycolysis and select for cells demonstrating activation of this pathway (18,37,38). Indeed our proteomics analysis provided a detailed profiling of these changes at the molecular level, and some of the altered proteins were consistent with previous studies. Of these, overexpression of glycolytic enzymes such as ␣-enolase, fructose-bisphosphate aldolase C, fructosebisphosphate aldolase A, and glyceraldehyde-3-phosphate dehydrogenase indicated elevated rates of glycolysis. Repression of the key enzyme fumarate hydratase and direct enhancement of its substrate, fumarate, coupled with abnormal expression of electron transport-associated proteins, such as AKR7A3, suggested an impaired tricarboxylic acid cycle and blocked oxidative phosphorylation. Meanwhile altered expressions of pyruvate kinase isozymes M1/M2 and transaldolase and especially the overexpression of L-lactate dehydrogenase A chain and L-lactate dehydrogenase B chain suggested the selective diversion of pyruvate metabolism from the tricarboxylic acid cycle to lactic acid. In addition, down-regulation of fructose-1,6-bisphosphatase 1, aldehyde dehydrogenase X, and lactoylglutathione lyase together with up-regulation of enoyl-CoA hydratase, short chain-specific acyl-CoA dehydrogenase, and hydroxyacyl-coenzyme A dehydrogenase suggested changes in gluconeogenesis, glucuronic acid synthesis, and fatty acid ␤ oxidation. Such changes may be caused by altered intermediates during glycolysis and the tricarboxylic acid cycle.
Although the Warburg effect was observed as early as the 1930s, the precise carcinogenic mechanism explaining cancer cell adaptation to this metabolic stress remains poorly defined. Recently several lines of evidence have pointed to HIF as a potential control element. HIF plays an essential role in sustaining glycolytic metabolism by activating several key substrates and enzymes in this pathway (39 -41). The present data provide new evidence for HIF-1␣ overexpression in gastric tumors. It was also demonstrated in the present study that the HIF target oncogene VEGF was significantly up-regulated in vivo and in vitro corresponding to the accumulation of HIF; however, it remains to be determined whether overexpression of HIF is required and/or sufficient for the development of gastric carcinoma.
TH, the major effect of which is through the T 3 -mediated signaling pathway, regulates a great many cellular physical activities including respiration (42)(43)(44). In the present study, accumulation of cellular fumarate was confirmed as a phenomenon characteristic of T 3 -induced tricarboxylic acid cycle dysfunction. In addition, FUMH inactivation at the enzymatic activity level was also observed as early as 2 h after T 3 administration. Considering previous findings that RNA interference caused FUMH silencing and sharply elevated fumarate levels (19), it is reasonable to propose that T 3 induces fumarate accumulation by inhibiting FUMH.
Stabilization of HIF1-␣ by fumarate through inhibition of the VHL-dependent degradation pathway has been observed previously in renal tumors (20). The data suggest that a similar mechanism operates in gastric tumors. Furthermore fumarate-induced HIF1-␣ induction could be completely abolished by its competitive cosubstrate 2-oxoglutarate even in the presence of T 3 , indicating that fumarate accumulation is necessary for T 3 -induced HIF1-␣ induction.
Previous studies indicated that cellular T 3 functions in promoting gene expression through binding at its nuclear receptor (44). In contrast, recent studies suggested that cytosolic cascade pathways are also responsible for T 3 (13,45). Data from the present study demonstrate that the early accumulation of HIF induced by T 3 was probably dependent on the PI3K/Akt signaling pathway and therefore was nucleus-independent.
Efforts have been made to explore the relationship between TH dysfunction and carcinogenesis, but no rational signaling pathway has been established thus far (46 -48). The proteomics identification described here provides new clues pointing to TH imbalance, HIF overexpression, and tricarboxylic acid cycle impairment. Consequently a novel thyroid hormone- mediated carcinogenic signaling pathway is proposed in gastric carcinoma on the basis of the proteomics identification and function studies presented here (Fig. 7). Overexpression of TTHY primarily results in cellular accumulation of T 3 . In response to T 3 , cytosolic cascade PI3K/Akt is rapidly activated to deliver signals to mitochondria, and processes of the tricarboxylic acid cycle are consequently blocked. As a result, FUMH is repressed primarily at the enzymatic activity level and later at the expression level, both leading to fumarate accumulation. Elevated fumarate reduces VHL-dependent degradation of HIF1-␣ by inhibiting HDH. Stabilized HIF transfers into the nucleus through binding to the Hsp90-p300/CBP complex and subsequently activates expression of its target genes (48), which include VEGF. VEGF functions in tumorigenic angiogenesis (49). Additionally HYOU1, another hypoxia-induced protein, promotes prematurational folding and modification of VEGF (50). HIF also promotes expression of several key enzymes in glycolysis and switches energy metabolism from the tricarboxylic acid cycle to glycolysis, protecting cells from energy shortage. FIG. 5. A, MKN28 cells were treated with 2 nM T 3 for 1-6 and 12 h. Intracellular accumulation of fumarate was monitored by HPLC. Representative HPLC maps for 3 h, 6 h, and untreated control are shown in the left panel. The changing pattern is shown in the right panel. B, MKN28 cells were treated with 2 nM T 3 for 24 h. Expression of FUMH was validated by RT-PCR. C, the enzymatic activity was tested as indicated. Tbl, MKN28 cells treated with 2 nM T 3 for 2 h before lysis; Tal, normal MKN28 lysate with T 3 added to a final concentration of 10 nM. D, MKN28 cells were treated with 5 mM fumarate for 30 and 60 min. Expressions of HIF1-␣ were examined by both RT-PCR and Western blotting. E, MKN28 cells were selectively treated with T 3 (2 nM), fumarate (5 mM), and 2-oxoglutarate (5 mM), and expression of HIF1-␣ was monitored by Western blotting. AU, absorbance units. All the data were shown as mean Ϯ S.D.
Many of the proteins identified as modified in expression in this study belong to families, the members of which have been cited by Petrak et al. (36) as modified in many 2-DE proteomics studies regardless of experimental objectives, tissues, or species. These authors suggested that the modification of these proteins may be due to a general response to stress or technical aberrations. It is necessary to be mindful of these points in the interpretation of the present results. The present study examined proteins not listed by Petrak et al. (36), such as TTHY associated with thyroid hormone regulation and FUMH in the tricarboxylic acid cycle. In addition, the work presented here has provided new insights into the stressrelated signaling pathways involved in this cancer, thereby giving insights into the mechanisms leading to the expression modifications found. In view of this we have increased confidence in the validity of the findings even for some of the proteins listed by Petrak et al. (36). Further work is required to determine exactly the implications of their findings for studies such as ours. CONCLUSION In this study, comparative proteomics analysis of gastric carcinoma and the adjacent normal tissues revealed 107 differentially expressed proteins. Cluster analysis revealed dysregulation of a group of metabolic enzymes associated with aberrant repression of the tricarboxylic acid cycle, accumulation of thyroid hormone, and overexpression of hypoxia-inducible factor. Molecular analysis of the intrinsic interplay of the metabolic alterations suggested a potential novel gastric carcinogenic pathway, which was induced by thyroid hormone and mediated by hypoxia-inducible factor.