The Interaction of Thioredoxin with Txnip

The thioredoxin system plays an important role in maintaining a reducing environment in the cell. Recently, several thioredoxin binding partners have been identified and proposed to mediate aspects of redox signaling, but the significance of these interactions is unclear in part due to incomplete understanding of the mechanism for thioredoxin binding. Thioredoxin-interacting protein (Txnip) is critical for regulation of glucose metabolism, the only currently known function of which is to bind and inhibit thioredoxin. We explored the mechanism of the Txnip-thioredoxin interaction and present evidence that Txnip and thioredoxin form a stable disulfide-linked complex. We identified two Txnip cysteines that are important for thioredoxin binding and showed that this interaction is consistent with a disulfide exchange reaction between oxidized Txnip and reduced thioredoxin. These cysteines are not conserved in the broader family of arrestin domain-containing proteins, and we demonstrate that the thioredoxin-binding property of Txnip is unique. These data suggest that Txnip is a target of reduced thioredoxin and provide insight into the potential role of Txnip as a redox-sensitive signaling protein.

Thioredoxin is a ubiquitous disulfide oxidoreductase that, along with the glutathione system, plays a major role in maintaining the cytoplasm in a reducing environment. Thioredoxin activity is mediated by a pair of cysteine thiols at its active site (human thioredoxins C32 and C35) that are oxidized during reduction of the substrate. By maintaining this reducing environment, thioredoxin is a critical defense against excess concentrations of reactive oxygen species, which are deleterious to cells and implicated in the pathophysiology of diseases such as atherosclerosis (1,2), diabetes (3), and arthritis (4,5). The reducing environment of the cell is also important for keeping protein thiols reduced, so that under normal conditions proteins contain many free sulfhydryl groups and relatively rare accessible disulfides (6).
In addition to the classic reducing activity of thioredoxin, recent evidence suggests that the redox state of thioredoxin may itself be an important component of redox signaling pathways. Thioredoxin reportedly binds a number of transcription factors and signaling molecules, including NF-B p50 subunit (7), , Jab-1 (9), Oct-4 (10), and PTEN (11). One of the better characterized interactions is that of reduced thioredoxin with apoptosis signal-regulating kinase 1 (ASK1), 2 which plays a key role in promoting stress-induced apoptosis (12). Because only reduced thioredoxin is thought to bind ASK1, the interaction can be controlled by the redox state of the cell: during conditions of oxidative stress, ASK1 is released from thioredoxin and promotes apoptosis. The identification of thioredoxin and ASK1 cysteines required for the interaction (13) supports the hypothesis that thioredoxin forms a mixed disulfide complex with ASK1. However, the implications of this interaction are not clear in part due to (i) the lack of direct evidence that thioredoxin can form a stable mixed disulfide and (ii) the lack of a plausible mechanism for formation of this disulfide in the normal reducing environment of the cell.
Txnip (thioredoxin-interacting protein, also called "vitamin-D3-up-regulated protein-1" (14) and "thioredoxin-binding protein 2"), is a 50-kDa protein with structural homology to the arrestins. Despite the arrestin homology, currently the only known function of Txnip is to bind thioredoxin and inhibit thioredoxin-reducing activity (15)(16)(17). Txnip, like ASK1, appears to form a disulfide bond with thioredoxin, because the interaction requires the presence of the thioredoxin active-site cysteines (17). Consistent with this role, overexpression of Txnip decreases thioredoxin-reducing activity, increases redox stress, inhibits growth and hypertrophy, and causes increased apoptosis (16,18,19), whereas loss of Txnip expression is associated with tumor growth and metastasis (20 -22). In addition, increasing evidence suggests major physiological roles for Txnip in glucose metabolism and cell differentiation. Both a naturally occurring mouse strain "hyplip1" with a truncation in Txnip (23)(24)(25) and mice with targeted deletion of Txnip (26) have low blood glucose, hyperlipidemia, and a severely dysregulated response to fasting. It is possible that these roles for Txnip are not mediated by inhibition of thioredoxin function alone. Understanding Txnip function will require that the mechanism for the interaction of Txnip and thioredoxin be better defined. * This work was supported by grants from the National Institutes of Health (NIH) (to R. T. L.) and an NIH National Research Service Award (to P. P.). The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked "advertisement" in accordance with 18 U.S.C. Section 1734 solely to indicate this fact. In the present work we explored the mechanism of the Txnip-thioredoxin interaction and present evidence that Txnip and thioredoxin form a stable disulfide-liked complex. We identified two Txnip cysteines that are important for thioredoxin binding and showed that this interaction is consistent with a disulfide exchange reaction between oxidized Txnip and reduced thioredoxin. These two critical cysteines are not conserved in the broader family of arrestin domain-containing proteins, suggesting that Txnip is a unique redox-sensitive signaling protein.

MATERIALS AND METHODS
293 Cell Culture and Transfection-HEK 293 cells were maintained in Dulbecco's modified Eagle's medium with high glucose, no sodium pyruvate, 10% fetal bovine serum, and antibiotics (100 units/ml penicillin and streptomycin). Plasmid-encoded proteins were expressed by transfection with FuGENE 6 (Roche Applied Science) using 3% FuGENE and a 1:3 ratio of micrograms of DNA to microliters of FuGENE. Three days after transfection, cells were washed in ice-cold phosphatebuffered saline, harvested in lysis buffer (Tris-buffered saline with 0.5% Triton X-100, 1 mM phenylmethylsulfonyl fluoride, and protease inhibitor mixture (Sigma)), and clarified at 16,000 ϫ g for 10 min.
Adipocyte Culture-3T3-L1 mouse fibroblasts (ATCC, Manassas, VA) were cultured in Dulbecco's modified Eagle's medium with high glucose (25 mM), penicillin, streptomycin, and 10% newborn calf serum (Invitrogen). Differentiation was induced by adding 5 g/ml insulin, 0.25 M dexamethasone, and 0.5 mM 3-isobutyl-1-methylxanthine to the media according to a standard protocol (27). By 8 days after induction, greater than 95% of cells were mature adipocytes based on morphology and Oil Red O uptake.
Adipocyte Transduction-Human Txnip constructs were overexpressed in mature adipocytes by infection with a vector based on replication-deficient human immunodeficiency virus. TXNIP was subcloned into the pCDH1-MCS1-EF1-Puro vector (System Biosciences, Mountain View, CA), and the TXNIP C247S mutant was made by site-directed mutagenesis. Pseudoviral particles were made by transfection of TXNIP, TXNIP C247S, or an empty vector into 293TN cells (ATCC). A vector expressing green fluorescent protein (GFP) (pSIH1-H1-copGFP, System Biosciences) was co-transfected with each vector in a 1:100 ratio. The multiplicity of infection of the resulting particles was determined by the number of cells expressing GFP on fluorescence microscopy 96 h after transduction. For experimental studies, 8 days after induction cells were cultured in low glucose (5.6 mM), transduced with a multiplicity of infection of 1-2, and allowed to express viral proteins for 96 h. Real-time PCR and Western analysis verified about a 2-fold overexpression of Txnip and Txnip C247S in transduced cells (data not shown).
Site-directed Mutagenesis-Cysteine-to-serine mutants were introduced into TXNIP (in pcDNA3.1ϩ) and TXN (in pGEX) by whole plasmid PCR with PfuTurbo polymerase, followed by digestion of template DNA with DpnI (New England Biolabs, Ipswich, MA). Constructs were verified by bidirectional sequencing.
GST Binding Assays-TXN (thioredoxin) was cloned from HeLa cell cDNA and subcloned between the EcoRI and XhoI sites of pGEX-4T3 (Amersham Biosciences), carboxyl-terminal to the glutathione S-transferase (GST). GST fusion proteins were induced in Escherichia coli BL21 cells by 0.5 mM isopropyl-␤-D-thiogalactopyranoside for 3 h. Cell pellets were lysed in B-PER (Pierce) and 40 units/ml DNase, and the cleared lysates were incubated with glutathione-Sepharose 4B beads (Bio-World, Dublin, OH) for 30 min at 4°C followed by three washes with 0.5% Triton X-100 in Tris-buffered saline. For GST binding, 293 cell lysates were incubated with beads containing equal amounts of GST protein. Binding proceeded overnight with rotation at 4°C followed by three washes with lysis buffer. Bound proteins were released by boiling in gel-loading sample buffer. All experiments were replicated at least once, and essentially identical results were obtained to those shown.
Western Analysis-Protein samples were electrophoresed through 10% SDS-polyacrylamide Bis-Tris gels (Invitrogen). Immunodetection of Txnip was performed with a custommade monoclonal antibody (JY2, available from MBL Intl., Woburn, MA). Monoclonal antibodies to the V5 and Xpress epitopes were from Invitrogen.
Thioredoxin Insulin Reduction Activity Assay-Thioredoxin activity in cell lysates was measured by the insulin-reduction assay of Holmgren et al. (28) with modifications. Briefly, cell lysates were incubated in excess insulin, thioredoxin reductase, and NADPH at 37°C for 20 min. The reaction was terminated by the addition of guanidine and 5,5Ј-dithiobis-(2-nitrobenzoic acid). Reduction of 5,5Ј-dithiobis-(2-nitrobenzoic acid) to 5-thio-2-nitrobenzoic acid was detected by optical density at 412 nm. Recombinant human thioredoxin-C73S-GST was used as a standard.
Statistical Analysis-Differences between groups were tested by t-tests with unequal variance assumed. All tests were corrected for multiple comparisons using ␣ ϭ 1 Ϫ (1 Ϫ ␣ 0 ) 1/n , where ␣ 0 ϭ 0.05 and n ϭ total number of comparisons.

Wild-type Txnip and Thioredoxin Form a DTT-sensitive
Complex-Previous studies have shown that Txnip does not interact with the thioredoxin active-site double-cysteine mutant (C32S,C35S) (15,17), suggesting that Txnip and thiore-doxin interact by formation of a disulfide bond with one of the thioredoxin active-site cysteines. However, there is little direct evidence that wild-type thioredoxin forms stable mixed disulfides with its putative binding partners under physiological conditions. Indeed, thioredoxin is thought to be an active disulfide reductase in large part due to the instability of the intermediate mixed disulfide (29,30). NMR studies have identified interactions of NFB and Ref-1 peptides with thioredoxin, but because the peptides were cross-linked to a thioredoxin C35A mutant, they do not provide evidence regarding the physiological relevance of the mixed disulfide complex (7,8). Indirect evidence supports the generation of a mixed disulfide between thioredoxin and ASK1: ASK1 binds selectively to reduced thioredoxin; the interaction requires one thioredoxin active-site cysteine and an ASK1 cysteine; and the interaction is stabilized by single mutation of either thioredoxin Cys-35 or Cys-32 to serine (31). However, despite the structural similarity of the reduced and oxidized thioredoxin forms (32,33), there is a precedent for selective binding to reduced thioredoxin solely by noncovalent interactions in the E. coli T7 polymerase complex (34).
We therefore tested whether we could identify a stable disulfide complex of Txnip and thioredoxin by a thioredoxin-GST binding assay. Human Txnip was overexpressed in 293 cells, which have nearly undetectable levels of endogenous Txnip, and the cell lysates were incubated overnight with thioredoxin-GST bound to glutathione-Sepharose beads. Bound proteins were released by boiling in non-reducing sample buffer, incubated in a range of dithiothreitol (DTT) concentrations, and subjected to Western analysis with a monoclonal antibody against Txnip (JY2) (Fig. 1). On short exposure, primarily monomeric Txnip at ϳ45 kDa was observed (left panel). The intensity of this band increased with increasing concentrations of DTT, consistent with release of the monomeric Txnip from disulfide-linked complexes by DTT. Fainter bands were also visible at the expected size of the Txnip-thioredoxin-GST complex (ϳ83 kDa). On longer exposure (middle panel), this 83-kDa band was clearly visible with 0 -0.1 mM DTT but disappeared in a dose-dependent manner from 0.1 to 10 mM DTT. The 83-kDa band was not visible in the input lysates (right panel). The identification of an 83-kDa JY2-reactive band that is sensitive to DTT is thus consistent with the identification of a stable Txnip-thioredoxin disulfidelinked complex.
We also observed that the Txnip monomer showed a mobility shift with increasing levels of DTT (Fig. 1), suggesting that Txnip contains one or more intramolecular disulfides. This observation is similar to the mobility shift seen for the redoxactive disulfide of PTEN (35). However, this evidence was limited by the inability of the Western analysis to resolve the Txnip bands that would be expected for separate dithiol and disulfide forms of Txnip.
Txnip Cysteine 267 Is Conserved in Four Human/Mouse Homologs-We then hypothesized that if Txnip formed a disulfide bond with thioredoxin, at least one Txnip cysteine would be required for the interaction. Furthermore, if thioredoxin interaction were conserved among Txnip homologs, this cysteine should be conserved. A search of the PubMed mouse and human genome data base revealed five proteins homologous to Txnip of unknown function ( Fig. 2A). Txnip and these homologs all contain structural homology to the arrestins,  which are intracellular scaffolding proteins that have important roles in modulating receptor endocytosis, ubiquitination, and signaling (36). The homologs have been assigned official gene names arrestin domain-containing (Arrdc) 1 through 5. Three have so far been cloned and briefly characterized: human Arrdc4 as "down-regulated in hepatocellular carcinoma 1 (DRH1)" (37), rat Arrdc2 as "induced by lysergic acid diethylamide-1 (ILAD-1)" (38), and most recently, human Arrdc3 as "thioredoxin-binding-protein-2-like inducible membrane protein (TLIMP)" (39). The most distantly related protein, Arrdc5, is notable for its sperm-specific expression. The functions of these proteins and the significance of the arrestin domains remain unknown.
Alignment of the three proteins (Arrdc2-4) with the most identity to Txnip demonstrated a single conserved cysteine (Txnip C267) ( Fig. 2B; full alignments in supplemental Fig.  S1A). This cysteine is also largely conserved in arrestin domain-containing homologs through evolution to Caenorhabditis elegans and Drosophila melanogaster (Fig. S1B). This suggested that the Txnip cysteine 267 would be required for interaction with thioredoxin and that the arrestin domain-containing proteins describe a family of thioredoxin-interacting proteins.

Txnip Cysteine 247, Not Cysteine 267, Is Critical for the
Txnip-Thioredoxin Interaction-Using yeast two-hybrid screening, Yamanaka et al. (15) identified that deletion of the amino-terminal arrestin domain (residues 1-154) did not affect binding to thioredoxin. We therefore generated a Txnip cysteine-to-serine mutant for each of the 7 cysteine residues in the remainder of the protein (residues 155-391) and overexpressed them in 293 cells. Thioredoxin-GST bound wild-type Txnip and six of the mutant Txnips, including the C267S mutant. Surprisingly, thioredoxin-GST instead failed to bind a mutation of Txnip cysteine 247 (Cys-247). Thus Txnip Cys-247, not the conserved Cys-267, is critical for the interaction with thioredoxin.
Whereas Txnip Cys-267 is highly conserved through evolution, Txnip Cys-247 was not present in any of the arrestin domain-containing proteins other than Txnip orthologs (Fig.  3B, bottom, and Fig. S1). Even within Txnip orthologs it was not present in the sequenced avian (Gallus gallus) and amphibian (Xenopus laevis) species (Fig. 3B). This suggests that the family of Arrdc proteins is not defined by interaction with thioredoxin.
Txnip-like Proteins Do Not Bind to Thioredoxin-To test the hypothesis that the Txnip-like proteins do not interact with thioredoxin, we cloned the three mouse homologs with closest homology to Txnip (Arrdc2, -3, and -4), as well as mouse Txnip, into tagged expression vectors. The vectors were then overexpressed in 293 cells, and detection of the tagged proteins was confirmed in cell lysates (Fig. 4, left panels). Thioredoxin-GST failed to detect an interaction with either Arrdc2 or Arrdc4 under conditions that bound Txnip (Fig. 4, right panels). Con-  Lysates were incubated overnight with thioredoxin-C73S-GST. After washing in lysis buffer, the bound proteins were subject to Western analysis. A, Western analysis shows that Arrdc2 was detected in the input lysate with anti-V5, but no binding to thioredoxin-GST was detected. As a positive control, under these conditions binding of mouse Txnip was successfully detected. B, mouse Arrdc4 was expressed with an N-terminal Xpress epitope. Input lysates and bound proteins were therefore subjected to Western analysis with anti-Xpress and anti-V5. Arrdc4 was detected in the lysate, but no binding to thioredoxin was observed under conditions of successful Txnip binding. The experiments were replicated with essentially identical results as those shown. sistent with these observations, Oka et al. (39) have recently reported that human Arrdc3 also does not bind thioredoxin.
These results confirm that the arrestin domain-containing Txnip-like protein family is not defined by thioredoxin binding. The structural homology with the arrestins suggests a role as scaffolding proteins for protein-protein interaction and signaling, as described for the ␤-arrestins (36). It may be that each homolog has evolved for interaction with different substrates and that studies of their binding partners will be important for defining their functions. In addition, because the ability of Txnip to bind thioredoxin is unique, Txnip may retain other conserved scaffolding and/or signaling functions.

Effect of Redox Conditions on the Txnip-Thioredoxin Interaction Suggests Txnip Is
Redox-active-Our evidence that Txnip forms a DTT-sensitive complex with thioredoxin, together with the identification of a Txnip cysteine critical for the interaction, strongly suggested that the interaction is mediated by formation of a disulfide bond. However, in the reducing environment of the normal eukaryotic cytosol, oxidation by de novo formation of protein disulfides is uncommon in the absence of an electron acceptor to complete the redox reaction (40). Most protein disulfides are generated in the endoplasmic reticulum in a process that requires both protein disulfide isomerase and endoplasmic reticulum oxidoreductase-1 (Ero1) to catalyze electron transfer to molecular oxygen (41,42).
We therefore hypothesized that if Txnip formed a disulfide bond with thioredoxin, a disulfide exchange mechanism would be most likely. This suggested two possibilities: either Txnip interacts with oxidized thioredoxin, or Txnip contains a disulfide bond and interacts with reduced thioredoxin. Nishiyama et al. (17) have previously reported that the interaction of Txnip with thioredoxin is abolished by oxidation of thioredoxin, suggesting that Txnip interacts with reduced thioredoxin. However, oxidation of human thioredoxin in vitro can cause inactivation of thioredoxin by dimerization at cysteine 73, thus preventing binding to Txnip (43,44). To eliminate this possibility, we tested the effect of redox state on the interaction of thioredoxin-C73S with Txnip.
Thioredoxin-C73S-GST beads were preincubated in lysis buffer and either reduced by addition of 100 M DTT, left untreated, or oxidized by addition of 100 M H 2 O 2 . The DTT or H 2 O 2 was then removed by extensive washing prior to incubation with cell lysates containing overexpressed Txnip. As a negative control, thioredoxin-C73S did not bind Txnip C247S (Fig.  5A, left). Binding of wild-type Txnip was not affected by reduction of thioredoxin-C73S with DTT (Fig. 5A, right). In contrast, oxidation of thioredoxin-C73S strongly blocked binding to Txnip, confirming that thioredoxin must be reduced for efficient interaction with Txnip.
We also tested whether manipulating the redox conditions during binding affected the Txnip-thioredoxin interaction. Thioredoxin-GST beads were incubated with 293 cell lysate in the presence of DTT, no additive, or H 2 O 2 (Fig. 5B). Again, Txnip C247S served as a negative control and no Txnip C247S binding was detected (Fig. 5B, left). However, redox manipulation during binding had the opposite effect from redox manipulation of thioredoxin alone: addition of 100 M H 2 O 2 had no effect while addition of 100 M DTT inhibited binding (Fig. 5B,  right).
The results of these two experiments are consistent with a disulfide exchange mechanism and suggest that Txnip is present as either an intramolecular or intermolecular disulfide. Because Txnip interacts with reduced thioredoxin, the disulfide does not come from thioredoxin. Furthermore, addition of DTT during binding inhibits the reaction, suggesting that DTT reduces the putative Txnip disulfide and prevents disulfide exchange.
We therefore considered evidence that Txnip might have an intramolecular or intermolecular disulfide. Western analysis for Txnip in fresh lysate from 293 cells overexpressing Txnip (Fig. 1, input lysates) or endogenous Txnip from NIH 3T3 and Thp1 monocytic cell lines (not shown) showed little evidence for JY2 reactivity at sizes larger than the monomer under nonreducing conditions. The absence of evidence for an intermolecular disulfide, taken together with the redox-sensitive mobility shift of Txnip (Fig. 1), was thus most consistent with the hypothesis that Txnip contains one or more redox-active intramolecular disulfides.

A Second Txnip Cysteine Is Critical For Efficient Thioredoxin
Binding-If an intramolecular Txnip disulfide were required for interaction with thioredoxin, the simplest hypothesis would be that the disulfide involves cysteine 247 and that its partner Txnip cysteine would also be required for interaction with thioredoxin. We therefore generated cysteine-to-serine mutants of the four cysteines in the Txnip N-terminal arrestin domain and tested them for interaction with thioredoxin-GST (Fig. 6). The Txnip C63S mutant nearly abolished binding by thioredoxin-GST, identifying a second Txnip cysteine that is important for thioredoxin interaction. This result therefore supports the hypothesis that Txnip contains an intramolecular disulfide between Cys-63 and Cys-247 that allows interaction with reduced thioredoxin by disulfide exchange.
Furthermore, the Western analysis suggested that, although some binding of the Txnip C63S to thioredoxin was still possible, no binding at all of the Txnip C247S was detected. We therefore hypothesized that Txnip Cys-247 is more critical for thioredoxin binding, because it forms the disulfide with thioredoxin, whereas Txnip Cys-63 is required for efficient reaction by disulfide exchange, but mutation of this cysteine does not prevent the interaction completely. This hypothesis is also consistent with the yeast two-hybrid analysis that found a Txnip truncation construct with deletion of the amino-terminal 155 residues still bound thioredoxin (15).
Txnip Cys-247 Is Critical For Inhibition of Thioredoxin Activity in Adipocytes-To test whether the Txnip mixed disulfide is important for the inhibition of thioredoxin activity by Txnip, we overexpressed wild-type and Txnip C247S in mature 3T3-L1 adipocytes by lentiviral transduction. Because these cells are post-mitotic, any confounding effect of proliferation rate on thioredoxin activity is avoided. Four days after transduction, thioredoxin-mediated insulin-reducing activity in the cell lysates was measured, normalized to total protein content, and then normalized to the mean activity of the control cells (transduced with virus expressing an empty vector) (Fig. 7). In cells transduced with wild-type Txnip, thioredoxin activity was reduced to 48 Ϯ 13% (mean Ϯ S.D.; n ϭ 5) of control levels (p Ͻ 0.001). In contrast, in cells transduced with Txnip C247S, thioredoxin activity was 120 Ϯ 2% of control levels (p Ͻ 0.001 for comparison with wild-type Txnip; not significantly different for comparison with controls). Differentiated 3T3-L1 adipocytes were transduced with lentiviral pseudoparticles expressing human wild-type Txnip ("Txnip WT"), Txnip C247S, or an empty vector ("Empty"). After 4 days the cells were lysed and the insulinreducing activity of thioredoxin was measured. Data were expressed as activity per total protein content and then normalized to the mean level of the empty vector control lysates. Bars represent the mean for each group. Significant differences between groups are indicated with p values.  These data provide evidence that Txnip-thioredoxin mixed disulfide formation is functionally important for the inhibition of the reducing activity of thioredoxin by Txnip. However, because these experiments may involve more than physiological levels of Txnip expression, the effects of the Txnip-thioredoxin interaction in vivo may be different (though the overexpression obtained here was only ϳ2-fold). In addition, other functions of Txnip and thioredoxin may not be dependent on or affected by mixed disulfide formation.
The evidence shown here for a stable mixed disulfide between Txnip and thioredoxin also strengthens the more general hypothesis that the role of thioredoxin in redox signaling is not limited to involvement in reduction reactions but may also include protein-protein interactions that are regulated by the state of its redox-active cysteines. In particular, several aspects of the Txnip-thioredoxin interaction are similar to the interaction of ASK1 with thioredoxin. Like Txnip, ASK1 does not bind the thioredoxin active-site mutant (C32S,C35S) and binds preferentially to reduced thioredoxin, suggesting the thioredoxin active-site cysteines are involved. Furthermore, mutation of a single thioredoxin active-site cysteine appears to stabilize the complex, consistent with the technique of "trapping" the mixed disulfide complex by C35A mutation (7). Finally, mutagenesis of ASK1 has identified a single cysteine-to-serine mutation that is critical for the interaction with thioredoxin. Thus based on the results here we speculate that ASK1 may also bind thioredoxin by a disulfide exchange reaction. However, support for this hypothesis would require evidence to demonstrate either an intramolecular or intermolecular ASK1 disulfide.
In summary, the results shown here provide evidence that Txnip and thioredoxin form a stable mixed disulfide by disulfide exchange. We propose that oxidized Txnip contains a disulfide linkage between cysteines 63 and 247, and the oxidized Txnip reacts with thioredoxin to form a disulfide between Txnip cysteine 247 and thioredoxin cysteine 32 (Fig. 8). In this model, either reduction of Txnip or oxidation of thioredoxin would prevent the interaction. We speculate that identification of the mechanism and generation of mutants that do not bind thioredoxin may allow identification of thioredoxin-independent functions of Txnip, based on its arrestin domain structure. These results also support the hypothesis that thioredoxin, in addition to its reducing activity, can participate in redox signaling through the formation of mixed disulfide complexes with other signaling proteins in a redox-dependent manner.