Thioredoxin-interacting Protein (Txnip) Is a Critical Regulator of Hepatic Glucose Production*

Thioredoxin-interacting protein (Txnip) has been recently described as a possible link between cellular redox state and metabolism; Txnip binds thioredoxin and inhibits its disulfide reductase activity in vitro, while a naturally occurring strain of Txnip-deficient mice has hyperlipidemia, hypoglycemia, and ketosis exacerbated by fasting. We generated Txnip-null mice to investigate the role of Txnip in glucose homeostasis. Txnip-null mice were hypoglycemic, hypoinsulinemic, and had blunted glucose production following a glucagon challenge, consistent with a central liver glucose-handling defect. Glucose release from isolated Txnip-null hepatocytes was 2-fold lower than wild-type hepatocytes, whereas β-hydroxybutyrate release was increased 2-fold, supporting an intrinsic defect in hepatocyte glucose metabolism. While hepatocyte-specific gene deletion of Txnip did not alter glucose clearance compared with littermate controls, Txnip expression in the liver was required for maintaining normal fasting glycemia and glucose production. In addition, hepatic overexpression of a Txnip transgene in wild-type mice resulted in elevated serum glucose levels and decreased ketone levels. Liver homogenates from Txnip-null mice had no significant differences in the glutathione oxidation state or in the amount of available thioredoxin. However, overexpression of wild-type Txnip in Txnip-null hepatocytes rescued cellular glucose production, whereas overexpression of a C247S mutant Txnip, which does not bind thioredoxin, had no effect. These data demonstrate that Txnip is required for normal glucose homeostasis in the liver. While available thioredoxin is not changed in Txnip-null mice, the effects of Txnip on glucose homeostasis are abolished by a single cysteine mutation that inhibits binding to thioredoxin.

Tight regulation of glucose homeostasis is fundamental to all higher life forms, and dysregulated glucose metabolism is a significant medical problem (1,2). The liver has a major role in determining glucose levels through its control of both hepatic glucose uptake and glucose release. During fasting states the liver is particularly important, as it is the primary organ for maintaining blood glucose levels by up-regulating gluconeogenesis and glycogenolysis to increase glucose release (3,4). These processes are coordinated by hormonal control, primarily through the reciprocal actions of glucagon and insulin, as well as through glucocorticoid and adipokine cues (5,6). Downstream effectors of hormonal inputs include transcription factors and coactivators that act as master regulators of metabolic transcriptional programs, such as PGC1␣, 3 CREB, TORC2, and Foxo1 (7)(8)(9)(10). Other mechanisms governing hepatic carbohydrate metabolism include changes in substrate availability (11), changes in mitochondrial respiration (12,13), and altered cellular redox state (14,15).
The thiol-disulfide redox state of the cell is emerging as an important regulator of diverse processes, including metabolism (16 -19). A key regulator is thioredoxin, a ubiquitous oxidoreductase found in nearly all life forms. Thioredoxin participates in maintaining a reduced cellular environment through the reversible oxidation-reduction of its two active-site cysteines (20). Although originally described as an antioxidant, thioredoxin associates with several transcription factors and signaling molecules, enabling it to serve as a potential mediator of redox signaling (21)(22)(23). The role of thioredoxin in carbohydrate synthesis in plant biology is well established. Light reduces thioredoxin via ferredoxin, and reduced thioredoxin then activates enzymes involved in the Calvin cycle by reduction of protein disulfides (24).
An initial report from Txnip-null mice generated by targeted deletion confirmed the importance of Txnip for the metabolic response to the fasting state, these mice die from multiple organ failure after 3 days of fasting. Fasted mice are rescued by a glucoseonly diet but not a fat-only diet (31), suggesting a defect in either fat utilization or in the generation and release of glucose. A link between thioredoxin redox state and metabolism has important implications, but it is not yet clear whether the Txnip-null phenotype is fully explained by the loss of inhibition of thioredoxin. It is possible that Txnip has other functions besides thioredoxin binding. Txnip has distant sequence homology to the arrestins, intracellular proteins that bind to phosphorylated receptors and then direct further signaling, and/or endocytosis (32). In addition to Txnip, mammals have three Txnip-homologous proteins of unknown function, named Arrestin domain containing 2-4 (Arrdc2-4), which do not bind thioredoxin (33). Interestingly, the gene encoding Arrdc3 has recently been linked to human obesity in males (34). This suggests the hypothesis that the Arrdc family of Txnip-like proteins is a family of metabolic regulators whose functions are not limited to thioredoxin binding (33).
We recently reported that mutation of a single Txnip cysteine (Cys-247) abolishes the ability of Txnip to bind thioredoxin and to inhibit thioredoxin activity when overexpressed in 3T3-L1 adipocytes (33). We have now generated total and hepatocyte-specific Txnip gene deletion in mice to assess the functions of Txnip in vivo. We used these mice to explore the role of Txnip in glucose homeostasis and demonstrated that in the absence of Txnip the liver is intrinsically defective in maintaining blood glucose levels through glucose production and release. We then demonstrated that the effects of Txnip on glucose regulation are abolished by a single cysteine mutation that is required for the Txnip-thioredoxin interaction. These results implicate Txnip as a key regulator of hepatic glucose production and global glucose homeostasis.
Serum Metabolite, Insulin, and Glucagon Assays-Tail vein blood samples were collected with heparinized capillary tubes and serum was obtained by centrifugation at 5000 ϫ g through a serum separating gel barrier at 4°C (Statspin, Inc.). Serum triglyceride, non-esterified free fatty acids, cholesterol, and ␤-hydroxybutyrate levels were assayed spectrophotometrically using commercially available kits (Wako and StanBio Labs). Serum insulin levels were measured by ELISA (Linco and ALPCO) and serum glucagon levels were measured by radioimmunoassay (Linco). Whole blood glucose levels were assayed from tail clipping venous blood samples using an Ascensia Elite XL glucometer (Bayer Co).
Glucose, Insulin, and Glucagon Tolerance Tests-Glucose (2 mg/g weight), insulin (0.25 milliunits/g weight), or glucagon (200 ng/g weight) was injected into the peritoneum of fasting mice (18 h fast for glucose tolerance, 4 h for insulin tolerance, and 48 h for glucagon tolerance). Tail vein whole blood glucose levels were assayed at intervals by glucometer, as above.
Northern Analysis-Total RNA was isolated from various tissues using the RNeasy mini kit (Qiagen). 10 g of total RNA was separated by 1% denaturing agarose gel electrophoresis and hybridized with a [␣-32 P]dCTP-labeled cDNA fragment corresponding to the full-length mouse Txnip coding region.
Immunohistochemistry and Immunofluorescence-Tissues were harvested from mice and fixed in ice-cold 4% paraformaldehyde in phosphate-buffered saline for 4 h at 4°C, then transferred to 70% ice-cold ethanol overnight at 4°C. Fixed tissues were then embedded in paraffin, sectioned onto slides, and rehydrated by standard means. Rehydrated tissue sections were immunostained without antigen-retrieval using primary antibodies to GFP (Rabbit polyclonal, Invitrogen) or insulin (guinea pig polyclonal, Zymed Laboratories Inc.). Immunohistochemistry was performed using the Vectastain ABC kit (Vector Labs) and immunofluorescence was performed using Rhodamine Red-X-AffiniPure secondary antibodies (Jackson Labs).
Quantitative Reverse Transcriptase-PCR-Liver total RNA was isolated using the RNeasy mini kit (Qiagen). cDNA was synthesized from 2 g of total RNA and random hexamers using the TaqMan Reverse Transcription kit (ABI). Real-time PCR was performed in 96-well plates using a 7300 Real-time PCR system (ABI). All reactions were performed in triplicate and verified by melting curve analysis. The relative amount of mRNA in each sample was normalized to 18 S transcript levels. Primer sequences are included in the supplemental "Methods." Hepatocyte Isolation and Culture-Mouse hepatocytes were isolated by a modified two-step perfusion procedure (37,38). In brief, mice were perfused retrograde through the hepatic vein via the inferior vena cava and out the severed portal vein, with the inferior vena cava clamped above the diaphragm. Hepatocyte Perfusion medium and Hepatocyte Collagenase solution (Invitrogen) were maintained at 37°C and perfused at 9 ml/min for 5 min each. The liver was then excised and gently disrupted to release the cells, which were strained and then washed once in ice-cold Hepatocyte Wash medium (Invitrogen) at 50 ϫ g and once with 27% Percoll in isotonic phosphate-buffered saline (Amersham Biosciences) at 100 ϫ g for 10 min, followed by a final wash with wash medium. Cells were incubated in Williams medium E with 10% fetal bovine serum, 100 nM insulin, and 100 nM dexamethasone for 4 h, and afterward maintained in serum-free Hepatozyme-SFM medium (Invitrogen) supplemented with 2% Me 2 SO for 48 h. Fao hepatoma cells, a gift of C. R. Kahn (Joslin Diabetes Center, Boston), were maintained in RPMI 1640 media supplemented with 10% fetal bovine serum, L-glutamine, and antibiotics.
Hepatocyte Glucose Production-Hepatocytes cultured on adsorbed collagen type 1 (BD Biosciences), or Fao hepatoma cells, were stimulated overnight with 250 M 8-bromo-cAMP and 100 nM dexamethasone (Sigma). The following day, the cells were washed twice with Dulbecco's modified Eagle's medium containing no glucose or phenol red, allowed to sit for 30 min in glucose-free media, then cultured in the same media with the addition of 20 mM lactate and 2 mM pyruvate or 20 mM glycerol (Sigma). After 3 h, the culture medium was removed and centrifuged at 15,000 ϫ g for 10 min. The glucose content of the supernatant was measured by a glucose oxidase colorimetric assay (Sigma).
Lentiviral Production and Tail Vein Injection-Lentiviral constructs for Txnip and Txnip C247S were made as described previously (33). Viral multiplicity of infection for liver infectivity was estimated based on in vitro primary hepatocyte transduction efficiency: 0.5 ml of undiluted viral stocks supplemented with Polybrene (7.5 g/ml) was added to 10 5 primary hepatocytes cultured in 12-well plates, and GFP-positive cells were counted 96 h after transduction. Anesthetized male C57Bl/6 mice, aged 9 -12 weeks, were injected with 150 l of undiluted viral stocks supplemented with Polybrene (5 g/ml) into the tail vein. Serum analysis or tissue harvesting was performed 7 days after injection.
GSSG:GSH Assay-Glutathione levels were assayed spectrophotometrically using a Bioxytech GSH/GSSG-412 kit, with modifications (Oxis International Inc., Foster City, CA). Over-night fasted mouse livers were harvested and immediately homogenized in the presence or absence of 3 mM 1-methyl-2-vinylpyridinium trifluoromethanesulfonate at 4°C. Uncomplexed glutathione was acidextracted from the homogenates with ice-cold 5% metaphosphoric acid (Sigma) and assayed according to the manufacturer's protocol.
Thioredoxin Activity Assay-Total available thioredoxin in mouse liver homogenates was measured by the insulin-disulfide reducing assay, modified from Holmgren and Bjornstedt (20). To increase the sensitivity of the assay to unavailable thioredoxin (through formation of either mixed disulfides at the active site or noncovalent protein complexes), we omitted the incubation at 70°C and the incubation with dithiothreitol (33). Livers were homogenized in a Tris buffer with 0.5% Triton X-100 and clarified at 16,000 ϫ g. Portions of the fresh clarified homogenate containing 50 g of total protein were incubated with insulin, rat liver thioredoxin reductase (Sigma), and NADPH for 15 min at 37°C. 5,5Ј-Dithiobis(nitrobenzoic acid) in guanidine was then added and the final reduced product was quantified by optical density at 420 nm.
Statistical Analysis-All data are presented as mean Ϯ S.E. Statistical analysis was performed with Student's t test. Data were considered statistically significant for p Ͻ 0.05.

Txnip-null Mice Are Hypoglycemic and Have Marked Fasting
Ketosis, but Normal Lipid Levels-We generated mice with conditional gene deletion of Txnip using a Cre-lox strategy for targeted deletion as previously described in detail (35). Homozygous Txnip-floxed mice (Txnip fl/fl ) were crossed with Protamine-Cre mice to generate offspring heterozygous for Txnip deletion, which were bred to homozygosity for use in this study. Northern and Western analysis demonstrated absence of gene expression in various tissues (Fig. 1, A and B). Txnip-null mice were fertile and outwardly normal in appearance and behavior. No significant differences were seen in animal weight compared with wild-type littermate controls (supplemental Fig. S1).
cose than wild-type littermates after an overnight fast, and 65 Ϯ 4% less after a 48-h fast (p Ͻ 0.01 for both comparisons, Fig. 1C). In addition, fasted Txnip-null mice had a striking ketosis with levels 7.1 Ϯ 0.8-fold higher than littermate controls (p Ͻ 0.001). A trend toward increased fasting non-esterified fatty acids was also observed (28 Ϯ 11% increase, p ϭ 0.08). The hypoglycemia, ketosis, and mild increase in non-esterified fatty acids exacerbated by fasting are consistent with the phenotypes reported previously for the HcB-19 mouse (29,30) and the Txnip-null mice characterized by Oka et al. (31).
Txnip-null Mice Have Increased Glucose Clearance and Impaired Response to Glucagon-To investigate the defect in glucose homeostasis in the Txnip-null mice, we first examined the kinetics of serum glucose clearance after an intraperitoneal glucose bolus in overnight fasted mice. The rise in serum glucose levels was significantly decreased for the Txnip-null mice compared with wild-type littermates, in terms of both absolute change and change relative to baseline ( Fig. 2A), suggesting an enhanced rate of glucose clearance by peripheral tissues. To determine whether the more rapid glucose clearance was the result of greater insulin-mediated glucose uptake, we next compared the blood glucose levels in response to an intraperitoneal insulin injection. Using a submaximal insulin dose, Txnip-null mice exhibited significantly augmented glucose clearance (Fig.  2B), consistent with enhanced insulin-stimulated glucose disposal by peripheral tissues, enhanced insulin-mediated suppression of hepatic glucose production, or both.
To determine whether a liver defect contributed to the increased glucose clearance, mice fasted for 48 h were challenged with an intraperitoneal glucagon injection. Following the injection, serum glucose levels were markedly blunted in the Txnip-null mice (Fig. 2C). This may reflect a primary defect in glycogen metabolism or a failure to replete glycogen stores due to a primary defect in gluconeogenesis. Hyperinsulinemia and a brisk insulin secretory response to glucose are possible hormonal mechanisms that would both augment glucose clearance as well as suppress appropriate hepatic glucose production. However, fed serum insulin levels were depressed by 61 Ϯ 10% (p Ͻ 0.01) in the Txnip-null mice compared with wild-type littermates, and by 76 Ϯ 10% (p Ͻ 0.05) in the fasting state (Fig.  2D), excluding this possibility. Hypoinsulinemia in our Txnipnull model is a departure from the prior observations of the HcB-19 mice and the global knock out reported by Oka et al. (31), which described either a trend toward higher serum insulin levels or frank hyperinsulinemia when compared with control mice (29 -31). Given this, we verified our findings using two different commercially available insulin ELISA, performed on two different experimental groups of mice, and further corroborated the findings with results from an independent commercial laboratory (IDEXX Laboratories, Westbrook, ME).
We next compared fed and fasting serum glucagon levels between the Txnip-null mice and littermate controls by radioimmunoassay. No significant differences were detected (Fig.  2E); however, because the Txnip-null mice were hypoinsuline-FIGURE 2. Txnip-null mice rapidly clear a glucose load and fail to augment glucose production. A, intraperitoneal glucose tolerance test following an overnight fast. A dextrose bolus (2 mg/g weight) was injected and whole blood glucose levels were assayed from the tail vein (n ϭ 6 each). B, intraperitoneal insulin tolerance test after a 4-h fast. Regular insulin (0.25 milliunits/g weight) was injected followed by tail vein blood glucose assay (n ϭ 6 each). C, intraperitoneal glucagon tolerance test following a 48-h fast. Tail vein blood glucose was assayed following injection of 200 ng/g weight glucagon (n ϭ 6 each). For panels A-C, asterisk refers to a significant difference with p Ͻ 0.01. D, serum insulin levels for fed and fasted mice by colorimetric ELISA (n ϭ 8 -10 per group). E, serum glucagon levels for fed and fasted mice assayed by radioimmunoassay (n ϭ 7-10 per group). NS, no significant difference. F, pancreas sections were stained for insulin by immunofluorescence (top panels, nuclei were counterstained with 4Ј,6-diamidino-2-phenylindole) or stained with H&E (bottom panels). The bar corresponds to 50 m. mic, the glucagon to insulin ratio in these mice was persistently elevated in both the fed and fasting states. The elevated glucagon:insulin ratio is consistent with a physiological response to persistent hypoglycemia by the pancreatic alpha and beta cells (39). Pancreas histology and insulin-specific immunofluorescent imaging suggested grossly intact islet architecture and beta cell mass (Fig. 2F). In addition, automated image analysis detected no difference in the proportion of beta cell area to total exocrine pancreas area between wild-type and Txnip-null mice (1.9 Ϯ 0.3 versus 1.8 Ϯ 0.3%, respectively, p ϭ 0.7, n ϭ 4). Taken together, these data suggest an appropriate beta cell response to persistent hypoglycemia, although a defect in beta cell insulin secretion remains a possibility.
Isolated Txnip-null Hepatocytes Have Decreased Glucose Release-Because the in vivo observations suggested Txnipnull mice had a defect in hepatic glucose output and ketogenesis, we tested whether hepatocytes isolated from the Txnip-null mice exhibited an intrinsic defect in glucose and ketone production. Primary hepatocytes were isolated from Txnip-null mice and wild-type littermate mice, cultured in the presence of the gluconeogenic precursor lactate, and the levels of glucose and ketones released into the cell culture media were measured. Primary hepatocytes from the Txnip-null mice produced half as much glucose as hepatocytes isolated from wild-type mice (52 Ϯ 6% relative to wild-type, p Ͻ 0.01, Fig. 3A). In addition, Txnip-null primary hepatocytes produced more than 2-fold the levels of ␤-hydroxybutyrate than wild-type cells (2.4 Ϯ 0.5-fold, p Ͻ 0.05, Fig. 3B). Thus, isolated hepatocytes, free from the influence of circulating hormones and substrates, recapitulated the observed in vivo metabolic abnormalities, and this strengthened the evidence for an intrinsic defect in regulation of hepatocyte metabolism.
Glycogen Metabolism Is Intact in the Txnip-null Hepatocyte-Having established that Txnip deficiency results in an intrinsic defect in hepatocyte glucose production, we next investigated whether the mechanism resides in defective glycogen metabolism, defective gluconeogenesis, or both. Liver glycogen levels from 24-h fasted Txnip-null mice were equivalent to their wildtype controls (33.6 Ϯ 6.3 versus 28.7 Ϯ 4.3 g/g of protein, respectively, p ϭ 0.53, n ϭ 5; supplemental Fig. S2A). Rates of glycogenolysis were measured in primary cultured hepatocytes from wild-type and Txnip-null mice, with glycogen content and glucose release measured following a glycogenolytic stimulus of 10 nM glucagon and 1 mM cAMP. Txnip-null hepatocytes had higher levels of basal and glucagon-stimulated glucose release relative to wild-type hepatocytes (basal 162 Ϯ 5% and stimulated 156 Ϯ 2%, p Ͻ 0.01, n ϭ 6, supplemental Fig. S2B). With respect to glycogen content, unstimulated Txnip-null hepatocytes had significantly elevated levels of total glycogen content relative to wild-type hepatocytes (219 Ϯ 24%, p Ͻ 0.01, n ϭ 6; supplemental Fig. S2C). As cAMP was used to promote glycogenolysis in these assays, we next examined whether glucagonstimulated cAMP production was defective. Levels of cAMP following a maximal (100 nM) and half-maximal (0.5 nM) glucagon stimulation were equivalent for wild-type and Txnip-deficient hepatocytes, as was the estimated EC 50 for glucagon stimulation (1.4 Ϯ 0.5 versus 2.0 Ϯ 0.4 nM, respectively, p ϭ 0.4, n ϭ 5; supplemental Fig. S2D). These data suggest that defective hepatic glucose production following Txnip deletion is not the result of deficient glycogen storage, decreased glycogenolysis, or aberrant glucagon signaling.
Gluconeogenesis Is Defective at the Level of the Mitochondria in Txnip Deficiency-As no apparent defects were observed in glycogen metabolism, we next focused on a potential dysregulation of hepatic gluconeogenesis. To determine whether the dysregulation in hepatocyte metabolism was due to a change in transcriptional regulation, we performed real-time PCR on mRNA isolated from fasted Txnip-null and wild-type livers. Hepatic glucose metabolism is controlled by the coordinated transcriptional regulation of rate-limiting enzymes such as phosphoenolpyruvate carboxykinase (PEPCK), glucose-6phosphatase (Glc-6-Pase), and pyruvate carboxylase (40), as well as master regulatory factors, such as peroxisome proliferator-activated receptor-␥ co-activator 1␣ and cyclic AMP response element-binding protein (CREB) (7)(8)(9). No statistically significant differences were detected in the mRNA levels of these molecules in Txnip-null compared with wild-type livers (Table 1). We further examined Pck1 (PEPCK) and G6pc (Glc-6-Pase) transcriptional responses in primary cultured hepatocytes stimulated with glucagon and cAMP. Both wild-

Quantification of mRNA expression levels in 24-h fasted liver of wildtype and Txnip ؊/؊ mice by real-time PCR
Livers were harvested from Txnip-null mice and littermate controls after a 24-h fast and total RNA was isolated. Quantitative real-time RT-PCR for each transcript was performed in triplicate with n ϭ 6 per group. Expression levels were normalized to 18 S transcript levels and expressed as the mean -fold change Ϯ S.E. relative to wild-type expression levels.

Metabolic pathway Gene
Relative mRNA expression levels  . Hepatocytes lacking Txnip are intrinsically defective in glucose production. Hepatocyte glucose production and ketone production was obtained from primary hepatocytes isolated from Txnip-null and littermate control mice. Hepatocytes were harvested by retrograde perfusion of liver with collagenase and then maintained in monolayer culture. A, glucose release into the media from primary hepatocytes following incubation in glucose production media for 1 h. B, medium was assayed for ␤-hydroxybutyrate levels after overnight incubation in a defined serum-free medium; n ϭ 8 per group.
type and Txnip-deficient hepatocytes exhibited a robust upregulation in G6pc transcripts after a 5-h stimulation, although Txnip-null hepatocytes had a mildly lower increase (46 Ϯ 1 versus 40 Ϯ 1-fold increase, p Ͻ 0.05, n ϭ 6). Pck1 transcript levels were also increased 60-fold in both groups (supplemental Fig. S3A). Liver ketogenesis occurs following ␤-oxidation of free fatty acids in the mitochondria to generate acetyl-CoA, a process that is reciprocally inhibited by lipogenesis. The initial steps of lipogenesis are catalyzed by acetyl-CoA carboxylase (ACC1 and -2) and produce malonyl-CoA. Malonyl-CoA levels decrease ␤-oxidation and ketogenesis by inhibiting carnitine palmitoyltransferase-1, which mediates transport of free fatty acids into the mitochondria. Therefore, a reduction in lipogenic gene expression might be expected in the Txnip-null livers. However, no statistically significant differences were detected in the key metabolic enzymes regulating triglyceride, lipid, and ketone biosynthesis: fatty acid synthetase, ACC1 and ACC2, carnitine palmitoyltransferase-1, and ketothiolase, or the upstream regulator, sterol response element-binding protein 1c (SREBP1c). A trend toward decreased transcript levels of the hepatic glucose transporter GLUT2 in Txnip-null livers and a trend toward increased transcript levels of hepatic lipoprotein lipase were seen, but the differences were not statistically significant. The absence of any significant differences in the transcriptional program for gluconeogenesis, fatty acid metabolism, or ketogenesis suggests that the primary metabolic defect in Txnip-null hepatocytes resides in translational regulation, enzyme turnover, post-translational regulation of enzymatic activity, the presence of enzymatic inhibitors, or in substrate availability derived from upstream metabolic pathways, such as mitochondrial respiration.
We next examined enzyme activity for two key gluconeogenic enzymes, Glc-6-Pase and PEPCK, following a 24-h fast. We observed a 166 Ϯ 8% increase in Glc-6-Pase activity from Txnip KO liver microsomes relative to wild-type controls (p Ͻ 0.001, n ϭ 6, supplemental Fig. S3B), indicating that Glc-6-Pase activity is not only intact, but up-regulated in Txnip-deficient hepatocytes. PEPCK activity was measured from both cytoplasmic and mitochondrial subfractions; we observed no difference in either the cytoplasmic fractions (1024 Ϯ 80 versus 1031 Ϯ 47 nmol of NADH ox /mg min, p ϭ 0.96, n ϭ 6) or the mitochondrial fractions (486 Ϯ 48 versus 435 Ϯ 45 nmol NADH ox /mg min, p ϭ 0.84, n ϭ 6; supplemental Fig. S3D) from wild-type and Txnip-null fasted livers. This data suggests that Txnip deficiency does not regulate gluconeogenesis by affecting post-translational changes or inhibition of these two key enzymes.
To test whether a defect in gluconeogenesis might be localized to the mitochondria, we repeated hepatic glucose production assays in primary hepatocyte cultures from wild-type and Txnip-deficient mice using lactate and pyruvate, gluconeogenic precursors that transit through the mitochondria, compared with glycerol, which does not transit through the mitochondria and enters the gluconeogenic pathway at the triose phosphate level (41). As before, Txnip-null hepatocytes had significantly less glucose production from a lactate and pyruvate (75 Ϯ 6% relative to wild-type, p Ͻ 0.01); however, no significant difference was seen when glycerol was used as a precursor (95 Ϯ 4% relative to wild-type, p ϭ 0.3, Fig. 4A). A reciprocal study was performed under similar conditions, using the Fao hepatoma cell line and forced overexpression of a Txnip transgene by lentiviral transduction. Hepatocytes overexpressing Txnip produced significantly more glucose (176 Ϯ 10% relative to mock transduced, p Ͻ 0.001) from lactate and pyruvate, but did not produce more glucose when supplied with glycerol (108 Ϯ 9% relative to wild-type, p ϭ 0.5, Fig. 4B). Combined with the observation that Txnip-null hepatocytes produce more ketones, the defective gluconeogenesis with lactate but not glycerol suggests Txnip deletion affects mitochondrial function, with preferential shunting of acetyl-CoA toward ketogenesis rather than gluconeogenesis.
Mice with Hepatocyte-specific Deletion of Txnip Also Have Fasting Hypoglycemia and a Blunted Response to Glucagon-Because isolated Txnip-null hepatocytes had defects in glucose production, we further investigated whether the in vivo defects in glucose homeostasis resulted from a liver defect in gluconeogenesis in mice carrying a hepatocyte-specific deletion of Txnip. Txnip fl/fl mice were mated with mice expressing a Cre recombinase transgene under the control of the albumin promoter (36) to generate mice with hepatocyte-specific gene deletion of Txnip (liver KO mice). Txnip fl/fl mice had equivalent hepatic Txnip expression levels by Northern and Western analysis, as well as normal fed and fasted glucose levels, when compared with wild-type littermates, and hence were used as littermate controls (data not shown). Northern analysis confirmed that Txnip gene disruption was restricted to the liver in liver KO mice (Fig. 5A). As with total Txnip-null mice, liver KO mice had a 34 Ϯ 7% lower blood glucose level in the fasting state (p Ͻ 0.001), although the degree of hypoglycemia was not as marked as for total Txnip-null mice (total KO, Fig. 5B). Similarly, the liver KO mice also had elevated fasting ketone levels (161 Ϯ 77% relative to wild-type, p Ͻ 0.05, Fig. 5C), although again not as marked as for the total KO mice. In contrast, after an intraperitoneal glucose bolus, the rise in blood glucose was not different between liver KO and littermate controls (Fig. 5D). Fed and fasting insulin and glucagon levels for liver Txnip KO mice were unchanged from wild-type controls (supplemental Fig. S4). These data suggest that Txnip expression in the liver is required for appropriate regulation of the baseline fasting glucose levels, whereas Txnip in peripheral tissues influences glucose clearance.
Hepatic glucose production in liver KO mice was then assessed by glucagon challenge in mice fasted for 48 h. Compared with control littermates, liver KO mice showed a markedly blunted increase in glucose levels in response to intraperitoneal glucagon, similar to the blunted response of the total Txnip-null mice (Fig. 5E).
We then tested whether primary hepatocytes isolated from liver-specific Txnip-null mice shared the same intrinsic defect in glucose production seen in the total Txnip-null mouse hepatocyte, free from the influence of circulating hormones and substrates. Primary hepatocytes from the liver KO mice produced half as much glucose as hepatocytes isolated from wildtype controls (57 Ϯ 10% relative to wild-type, p Ͻ 0.01, Fig. 5F), a relative reduction in glucose output comparable with that observed in total knock-out hepatocytes.
Liver-specific Overexpression of Txnip Raises Fasting Blood Glucose and Reduces Fasting Ketogenesis-We next studied the role of Txnip in liver glucose homeostasis in vivo by overexpressing a Txnip transgene in the liver of wild-type mice. 9 -12week-old C57Bl/6 mice were injected in their tail vein with lentivirus particles expressing either human Txnip (hTxnip) or an empty vector (empty), both mixed with lentiviral particles expressing GFP. Delivery of retroviral and lentiviral transgenes by tail vein injection results in specific tropism to liver and allows the in vivo study of transgenic overexpression specific to the liver (9, 42, 43). Following injection of hTxnip or empty lentivirus at an estimated multiplicity of infection of 1-2, over half of the hepatocytes were transduced as determined by immunohistochemical detection of the GFP (Fig. 6A). Expression of viral transgenes was restricted to the liver as demonstrated by reverse transcriptase-PCR from various tissues harvested from the infected mice (Fig. 6B). Mice expressing the hTxnip transgene in the liver had a 39 Ϯ 8% increase in fasting blood glucose levels compared with mice expressing the empty vector (p Ͻ 0.05, Fig. 6C). In addition, fasting serum ketone levels were reduced by 45 Ϯ 11% (p Ͻ 0.01, Fig. 6D).
Thioredoxin Availability and Glutathione Redox State Are Not Affected by Txnip Deficiency-Having isolated a defect in liver glucose metabolism in the Txnip-null mice, we next investigated whether this defect could be due to an increase in amount of available thioredoxin in the absence of Txnip. The Txnip-thioredoxin interaction requires the active site thioredoxins cysteines, and overexpression of Txnip in vitro decreases the amount of available thioredoxin as measured by the insulin reduction assay (25)(26)(27). Surprisingly, Sheth et al. (30) found no differences in amount of thioredoxin in liver homogenates from the Txnip-deficient Hyplip1 mouse using the assay of Holmgren and Bjornstedt (20). We modified the assay to increase its sensitivity to inhibition of thioredoxin due to complex formation (33), and tested whether available thioredoxin was increased in liver preparations from the Txnipnull mice. We found no difference in available thioredoxin between livers of Txnip-null and wild-type control mice after an overnight fast (Fig. 7A). Furthermore, no differences were seen in hepatocellular ratios of oxidized glutathione (GSSG) to reduced glutathione (GSH) between the two groups of mice (Fig. 7B). These findings suggest that the changes in metabolism in the Txnip-null mice are not due to global changes in cellular redox state or in levels of available thioredoxin.
Gene Transfer of Wild-type Txnip but Not Txnip C247S Rescues Hepatocyte Glucose Production-We have previously demonstrated that a single cysteine-to-serine mutation in Txnip (C247S) inhibits the ability of Txnip to bind to thioredoxin (33). We therefore tested whether Txnip C247S gene transfer could rescue the Txnip-null defects in glucose metabolism. First, primary hepatocytes from Txnip-null mice were transduced with lentiviral particles expressing an empty vector, wild-type human Txnip, or the Txnip C247S. Equivalent transduction efficiency was confirmed by real-time PCR expression levels of the co-expressed GFP transcript, while endogenous thioredoxin expression levels were unaffected (data not shown). Glucose production from Txnip-null hepatocytes was significantly increased by expression of Txnip compared with empty vector-transduced hepatocytes (20 Ϯ 4% increase, p Ͻ 0.01, Fig. 8A). In contrast, expression of Txnip C247S failed to rescue hepatocyte glucose production.
To confirm the role of Txnip cysteine 247 on glucose metabolism in vivo, we overexpressed the Txnip species in the liver of wild-type C57Bl/6 mice by tail vein injection. As demonstrated previously, the hTxnip transgene increased fasting blood glu-cose levels 7 days after infection (by 32 Ϯ 8%, p Ͻ 0.05); however, the C247S mutant species had no effect on fasting blood glucose levels compared with the mice injected with the empty vector control (Fig. 8B). These data demonstrate the critical role of Txnip in hepatocyte glucose production; furthermore, these data suggest that the interaction between Txnip and thioredoxin is pivotal in mediating the effect of Txnip on hepatocyte gluconeogenesis.

DISCUSSION
From these studies we demonstrate that Txnip is a key regulator of glucose metabolism, and that hepatic gluconeogenesis is intrinsically dependent on Txnip function. Both systemic and liver-specific Txnip deletion in mice exhibited fasting hypoglycemia and blunted glucagon-stimulated glucose production, and isolated Txnip-null hepatocytes had diminished glucose release, suggesting an intrinsic hepatocyte defect, possibly at the level of proper mitochondrial function, rather than secondary defects due to hormonal deficiencies or limited circulating precursors. Importantly, reintroduction of Txnip into Txnipnull hepatocytes rescued glucose production, confirming its critical role. Deletion of Txnip did not affect global thioredoxin reducing activity or the glutathione oxidation state, but the effect of Txnip on glucose metabolism was dependent on the presence of the Txnip Cys-247, which is required for binding of Txnip to thioredoxin. FIGURE 7. Liver thioredoxin availability and glutathione redox state are unchanged in the absence of Txnip. A, available thioredoxin was measured by the insulin reduction activity assay in liver homogenates taken from Txnipnull and wild-type littermate control mice after overnight fasting; n ϭ 4 per group. B, oxidized glutathione (GSSG) to reduced glutathione (GSH) ratios were obtained from Txnip-null and littermate control liver homogenates following an overnight fast; n ϭ 4 per group. NS, no significant difference.

FIGURE 8. Exogenous expression of wild-type Txnip but not Txnip C247S rescues glucose production in Txnip-deficient hepatocytes in vitro and increases glucose production in vivo.
A, primary hepatocytes were isolated from Txnip-null mice, cultured on collagen type I gel, and transduced with empty vector lentivirus, wild-type human Txnip, or C247S Txnip. 96 h after transduction, glucose production was assayed by a 1-h incubation in glucose production media. n ϭ 8 per group. B, fasting blood glucose levels 7 days after tail vein injection of lentivirus into wild-type C57BL/6 mice. n ϭ 5 per group. The impact of Txnip on liver glucose regulation may prove relevant to diabetes pathogenesis. One of the hallmark features of type 2 diabetes is resistance to the suppressive effect of insulin on hepatic glucose production (44,45). Txnip is one of the most abundantly up-regulated genes in response to glucose and its expression is also regulated by glucocorticoids and inflammatory markers such as tumor necrosis factor-␣ (46 -49), which are associated with obesity and insulin resistance (50,51). That Txnip can increase glucose output when its expression is increased suggests a pathologic mechanism: insults that increase Txnip expression would accelerate hepatic glucose production, contributing to hyperglycemia, and hyperglycemia in turn increases Txnip expression, further augmenting hepatic glucose production.
Intriguingly, the absence of hyperlipidemia and hyperinsulinemia in this Txnip-null model is different from both the HcB-19 mutant mouse and the Txnip targeted deletion mouse described by Oka et al. (31). Hypoinsulinemia likely best accounts for the absence of hypertriglyceridemia in these mice: insulin stimulates lipid synthesis in part through SREBP-1cmediated up-regulation of a lipogenic transcriptional program (52). Excess acetyl-CoA, typically converted to ketone bodies during fasting, would be synthesized by the liver into triglycerides with insulin stimulation. Hui et al. (29) confirmed this physiology, as streptozotocin ablation of the HcB-19 insulin secretion effectively eliminated hypertriglyceridemia. The mechanism for the observed difference in insulin response between the two models is unclear. A persistently expressed N-terminal portion of Txnip, possibly present in HcB-19, cannot alone explain this. The targeted deletion of Txnip presented by Oka et al. (31) removed the initial 4 exons encompassing the potentially expressed Hyplip1 gene product, yet they also reported fasting hyperinsulinemia. We suspect either modifier gene effects resulting from strain differences between these models or contributions specific to the different Txnip deletions may play a role in the apparent difference seen in beta cell function. Apart from the hormonal differences, fasting hypoglycemia and ketosis are highly consistent among all models, and strongly support a primary defect in the proper use of acetyl-CoA for gluconeogenesis.
Importantly, hepatocyte metabolism only accounts for a portion of the total in vivo effect of Txnip on glucose homeostasis. The rapid glucose clearance of the total Txnip-null mice, but not for the liver KO mice, suggests glucose clearance by peripheral tissues is accelerated in the absence of Txnip. This is supported by enhanced insulin-mediated glucose disposal observed during a low-dose insulin tolerance test. Our observations for Txnipnull mice are consistent with a primary defect in both insulin and non-insulin-mediated basal glucose uptake. In agreement with this, we have observed a significant increase in both basal and insulin-stimulated adipocyte and skeletal muscle glucose uptake in vitro following Txnip RNA interference gene silencing (53), blunted basal and insulin-stimulated adipocyte glucose uptake with forced overexpression of Txnip (53), and augmented basal myocardial glucose uptake in cardiac-specific Txnip-null mice (35).
Our investigation of the importance of the Txnip-thioredoxin interaction also suggests new hypotheses for the mecha-nism of Txnip action. Evidence from in vitro overexpression experiments had suggested that Txnip acts to inhibit thioredoxin and increase cellular redox stress. However, we did not observe any difference in the glutathione redox potential (GSS-G:GSH) in our study of Txnip-null mouse livers. In addition, neither we nor Sheth et al. (30) observed any significant changes in the amount of available thioredoxin. Therefore, our data suggest that Txnip does not function in vivo to change the overall redox state of the cell, possibly because physiological levels of Txnip are not sufficient to bind enough thioredoxin to significantly affect thioredoxin availability. It is still possible that Txnip affects thioredoxin activity in vivo: the Txnip-thioredoxin interaction may be limited to subcellular compartments, such as the nucleus or mitochondria (50); alternatively, other thioredoxin binding partners may gain access to thioredoxin in the absence of Txnip, effectively maintaining the levels of available thioredoxin that we observe. Finally, Txnip may change the thioredoxin redox state, because the insulin reduction activity assay is not affected by thioredoxin redox state.
However, the evidence that the effects of Txnip require Cys-247, but not changes in the amount of available thioredoxin, suggests that the Txnip-thioredoxin interaction may be important for other reasons. For example, the interaction of Txnip with thioredoxin may be most important for altering the interactions of thioredoxin with its other binding partners, such as Ask-1 (28). Alternatively, Txnip may have other binding partners that require Cys-247 for association, such as other thioredoxin-fold family members. It is also possible that, rather than Txnip inhibiting thioredoxin, thioredoxin is necessary for an as-yet undefined signaling pathway through Txnip.
Further mutational analysis may provide more insights into the role of the Txnip-thioredoxin interaction. For example, overexpression of Txnip C63S appears to have an intermediate phenotype in terms of rescuing glucose production, 4 consistent with its diminished but not abolished ability to bind thioredoxin. However, more knowledge about the Txnip structure, the nature of the Txnip-thioredoxin interaction, and other binding partners will be necessary.
In conclusion, we provide evidence that Txnip is a significant regulator of global glucose homeostasis and demonstrate its central role in hepatic glucose metabolism. In addition, our data supports the concept that Txnip functions through its interaction with thioredoxin, but not by directly sequestering thioredoxin. Further characterization of mice carrying the Txnip conditional deletion in other tissues that regulate glucose, such as skeletal muscle, fat, and the ␤-cell, will help delineate the contribution of Txnip to total in vivo glucose metabolism.