Thioredoxin-independent Regulation of Metabolism by the α-Arrestin Proteins*

Thioredoxin-interacting protein (Txnip), originally characterized as an inhibitor of thioredoxin, is now known to be a critical regulator of glucose metabolism in vivo. Txnip is a member of the α-arrestin protein family; the α-arrestins are related to the classical β-arrestins and visual arrestins. Txnip is the only α-arrestin known to bind thioredoxin, and it is not known whether the metabolic effects of Txnip are related to its ability to bind thioredoxin or related to conserved α-arrestin function. Here we show that wild type Txnip and Txnip C247S, a Txnip mutant that does not bind thioredoxin in vitro, both inhibit glucose uptake in mature adipocytes and in primary skin fibroblasts. Furthermore, we show that Txnip C247S does not bind thioredoxin in cells, using thiol alkylation to trap the Txnip-thioredoxin complex. Because Txnip function was independent of thioredoxin binding, we tested whether inhibition of glucose uptake was conserved in the related α-arrestins Arrdc4 and Arrdc3. Both Txnip and Arrdc4 inhibited glucose uptake and lactate output, while Arrdc3 had no effect. Structure-function analysis indicated that Txnip and Arrdc4 inhibit glucose uptake independent of the C-terminal WW-domain binding motifs, recently identified as important in yeast α-arrestins. Instead, regulation of glucose uptake was intrinsic to the arrestin domains themselves. These data demonstrate that Txnip regulates cellular metabolism independent of its binding to thioredoxin and reveal the arrestin domains as crucial structural elements in metabolic functions of α-arrestin proteins.

Txnip-deficient mice have fasting hypoglycemia and ketosis (8,9,11,12) with a striking enhancement of glucose uptake by peripheral tissues (8,9). We have proposed that Txnip inhibits thioredoxin by forming a mixed disulfide with thioredoxin at its catalytic active site cysteines in a disulfide exchange reaction (13). However, it is not known how Txnip metabolic functions relate to its ability to bind thioredoxin.
Structurally, Txnip belongs to the arrestin superfamily of proteins (14). The prototypical arrestins (the visual arrestins and the ␤-arrestins) are key regulators of receptor signaling. The ␤-arrestins, named for their interaction with the ␤-adrenergic receptor, are now known to control signaling through the multiple families of receptors (15). These arrestin proteins have two wing-like arrestin domains arranged around a central core that detects and binds selectively to the charged phosphates of activated receptors (16). The arrestin domains then act as multifunctional scaffolds that cannot only quench receptor signals by recruiting endocytotic machinery and ubiquitin ligases, but also start new signal cascades (15). Recently, arrestin-␤2 has also been shown to play a key role in metabolism as a controller of insulin receptor signaling that is deficient in diabetes (17).
In addition to the classical visual/␤-arrestins, a large number of arrestins more closely related to Txnip are present throughout multicellular evolution. These proteins have been termed the "␣-arrestins," as they are of more ancient origin than the visual/␤ family (14). Although no structures are known of the ␣-arrestins to date, they appear highly likely to share the overall fold: two ␤-sheet sandwich arrestin domains connected by a short linker sequence (14,18). Confidence in this prediction has been enhanced by the surprising finding that the vps26 family of proteins, even more distantly related to the classical arrestins than Txnip, also share the arrestin fold (19). The vps26 proteins are a component of the retromer complex that controls retrograde transport of recycling endosomes to the trans-Golgi network. This functional overlap with visual/␤-arrestin regulation of endocytosis suggests that control of endosome formation and transport may be a conserved function of the arrestin superfamily fold.
Given that other ␣-arrestins are not thioredoxin-binding proteins, we hypothesized that Txnip metabolic functions may be conserved in mammalian ␣-arrestins and independent of its interaction with thioredoxin. Overexpression of Txnip in vitro can decrease levels of available thioredoxin and increase levels of reactive oxygen species (1,3,26). However, in vivo studies of two different Txnip-deficient mouse models found no change in available thioredoxin levels (8,27). Txnip reportedly binds to other proteins including Jab1 (28) and Dnajb5 (29), but it is not clear to what extent these interactions are themselves independent of a Txnip-thioredoxin complex (30).
Using overexpression of a mutant Txnip that does not bind thioredoxin, we show here that a major metabolic function of Txnip, its inhibition of glucose uptake, does not require interaction with thioredoxin. Instead, we show that inhibition of glucose uptake is a conserved function of another human ␣-arrestin, Arrdc4. Studies of Txnip mutants and chimeric ␣-arrestins suggest that the metabolic functions of Txnip and Arrdc4 are intrinsic to the arrestin domains.
Cell Culture-Primary human skin fibroblasts (GM00498) were from Coriell Institute for Medical Research (Camden, NJ) and maintained in Eagle's minimum essential medium with 15% fetal calf serum plus penicillin and streptomycin. 293T cells were from ATCC (Manassas, VA) and were cultured in Dulbecco's modified Eagle's medium with 25 mM glucose, 4 mM L-glutamine, and 1 mM pyruvate, plus penicillin and streptomycin.
Detection of Txnip-Thioredoxin Complexes-To stabilize the disulfide-linked Txnip-thioredoxin complex, we prevented disulfide exchange by alkylating free sulfhydryls with 4-acetamido-4Ј-maleimidylstilbene-2,2Ј-disulfonate (AMS) (33). 293T cells were transfected with Txnip wild type and mutant constructs. Two days after transfection, cells were washed twice in ice-cold PBS, and proteins were precipitated with ice-cold 0.54 M trichloroacetate in PBS. Lysates were incubated on ice for 10 min, then centrifuged at 3000 ϫ g for 10 min at 4°C. The pellets were washed with acetone, then centrifuged at 3000 ϫ g for 10 min at 4°C. The resulting pellets were briefly air-dried on ice then dissolved in AMS-labeling buffer (62.5 mM Tris, pH 6.8, 1% SDS, 25 mM AMS) and incubated at 4°C with end-over-end rotation for 18 h. After adding nonreducing sample loading buffer and sonicating, the labeled proteins were subjected to SDS-PAGE and Western analysis.
Glucose Uptake-␣-Arrestin-mCherry fusion constructs were overexpressed in dermal fibroblasts by overnight incubation in lentivirus diluted 1:1 in medium with 5 g/ml polybrene. After 2-3 days, expression was verified by red epifluorescence. Four hours prior to labeling, medium was changed to 5.6 mM glucose to reduce endogenous Txnip expression. Cells were then washed three times in KRH-HEPES before addition of 100 M 2-deoxyglucose and 125 M 2-[ 3 H]deoxyglucose (1 Ci/ml, PerkinElmer Life Sciences) for 30 min. Unincorporated radiolabel was removed by three washes in ice-cold PBS with 25 mM glucose. Cells were lysed in 0.2 N NaOH then neutralized with 6 N HCl. The radiolabel remaining in the cell lysates was measured by liquid scintillation counting.
Extracellular Lactate Output-293T cells were transfected and conditioned medium was collected after 2-3 days with no medium changes. Lactate was measured by a lactate dehydrogenase-based assay from Eton Biosciences, Inc. (San Diego, CA).
Confocal Imaging-Cells were plated on glass-bottomed chambers coated with 2 g/ml fibronectin. Nuclei were stained with 1.6 M Hoechst dye 33342. Single confocal sections were obtained with a Zeiss LSM 710 confocal microscope by sequential excitation at 405 nm and 543 nm.
Statistical Analysis-Data are reported as means Ϯ S.E. Twosample comparisons were made by two-sided Student's t-tests.

Txnip Inhibits Glucose Uptake Independent of Ability to Bind
Thioredoxin-We have previously demonstrated that a single cysteine-to-serine mutation (C247S for human Txnip) abolishes the ability of Txnip to bind thioredoxin-GST in an in vitro binding assay (13). We therefore used the Txnip C247S mutant to test whether the interaction of thioredoxin with Txnip is required for Txnip metabolic functions. Specifically, we chose to characterize the inhibition of glucose uptake by Txnip because it is validated as an important Txnip function in vivo (7,8). We overexpressed wild type and C247S Txnip in mature 3T3-L1 adipocytes by lentiviral transduction and measured glucose uptake by incorporation of 2-[ 3 H]deoxyglucose. Uptake was expressed relative to the mean basal uptake of the cells infected with the empty vector control virus. Consistent with our previous studies (7), wild type Txnip inhibited adipocyte glucose uptake compared with an empty vector control under both basal (0.65 Ϯ 0.04 versus 1.00 Ϯ 0.05, p Ͻ 0.01) and insulin-stimulated (7.1 Ϯ 0.3 versus 9.2 Ϯ 0.2, p Ͻ 0.01) conditions ( Fig. 1A). Unexpectedly, we found that overexpression of mutant Txnip C247S also inhibited adipocyte glucose uptake under basal (0.74 Ϯ 0.05, p Ͻ 0.05 versus empty vector) and insulin-stimulated (7.78 Ϯ 0.14, p Ͻ 0.01 versus empty vector) conditions, indicating that this function of Txnip is independent of binding to thioredoxin.
To increase confidence that this result was generalizable to primary cells, we repeated the experiment with primary human skin fibroblasts. Lentiviral overexpression of wild type Txnip-Cherry inhibited glucose uptake compared with mCherry alone (Fig. 1B: 0.52 Ϯ 0.03 versus 1.00 Ϯ 0.07 respectively, p Ͻ 0.01), and overexpression of Txnip C247S-Cherry also significantly inhibited glucose uptake compared with mCherry alone (0.38 Ϯ 0.01, p Ͻ 0.001). Western analysis confirmed overexpression of both wild type and mutant Txnip.
To ensure that the C247S mutation abolishes the interaction of Txnip and thioredoxin in cells, we developed a new method for isolating intact Txnip-thioredoxin complexes. Because Txnip and thioredoxin may form a disulfide-linked complex (13), alkylation of free sulfhydryls should prevent loss of the complex by disulfide exchange. Using either N-ethylmaleimide or AMS as the alkylating agent, we confirmed that reaction of free sulfhydryls allows isolation of endogenous Txnip-thioredoxin complexes. We then tested whether overexpressed Txnip C63S and C247S interact with thioredoxin in cells. As an additional control, we also overexpressed Txnip C267S, a mutation of a highly conserved cysteine that does not affect binding to thioredoxin (13). The constructs were overexpressed in 293T cells, which have undetectable endogenous Txnip protein levels. Cell lysates were labeled with AMS and Western analysis was performed. Anti-thioredoxin antibodies detected a single anti-thioredoxin-reactive band at the expected size of a Txnip and thioredoxin complex (ϳ62 kDa) with wild type Txnip as well as Txnip C63S and Txnip C267S, whereas no band was observed with Txnip C247S (Fig. 1C). Probing with anti-Txnip antibody JY2 verfied that the 62-kDa band was a Txnip-thiore- These data demonstrate that mutation of Txnip Cys-247 leads to loss of interaction with thioredoxin in cells and that inhibition of glucose uptake by Txnip is not affected by the C247S mutation despite loss of ability to bind to thioredoxin.
A Second ␣-Arrestin Strongly Inhibits Lactate Output and Glucose Uptake-Txnip appears to be unique among the ␣-arrestins in its ability to bind thioredoxin: we and others have shown that the 3 ␣-arrestins most closely related to Txnip (Arrdc4, Arrdc3, and Arrdc2, in order of homology) do not bind thioredoxin (13,20). Because thioredoxin binding was not required for regulation of glucose uptake by Txnip, we hypothesized that regulation of glucose metabolism might be a function conserved in other ␣-arrestins.
Therefore, we made lentiviral overexpression constructs for the ␣-arrestins, all with C-terminal mCherry fusions: human Txnip-Cherry, mouse Arrdc4-Cherry, and human Arrdc3-Cherry. First, we characterized the localization of the ␣-arrestin-cherry fusion proteins by confocal imaging of transfected 293T cells (Fig. 2, A-D). As previously reported, Txnip localized almost exclusively to the nucleus (Fig. 2B) (34). Arrdc3 localized to the membrane and to numerous small vesicles in the cytoplasm, consistent with an association of Arrdc3 with endocytotic machinery, as suggested by Oka et al. (20). Localization of Arrdc4 was similar to that of Arrdc3, to both the cell membrane and cytoplasmic vesicles. No membrane localization was observed in the only previous characterization of Arrdc4 localization (in COS7 cells) (35), which may be attributable to either cell type differences or to the use of live cell imaging here. Therefore, localization of both Arrdc4 and Arrdc3 is suggestive of a conserved arrestin-like involvement in vesicle endocytosis and trafficking for the mammalian ␣-arrestins.
On transfection of the ␣-arrestins into 293T cells, we noted a striking change in the color of the culture medium: while the medium of cells transfected with mCherry alone and Arrdc3 turned orange-yellow as expected, the medium of cells transfected with Txnip and Arrdc4 remained red, suggesting an effect on either glucose uptake or metabolism. Normally, cell metabolism acidifies the medium, changing phenol red absorbance to yield yellow tones. In cell culture with glucose as the primary carbon source, medium is acidified largely by lactate (36). Lactate is the major endproduct of glucose uptake in immortalized cells in culture (37) and is exported to the medium by facilitated diffusion (38). We therefore quantified the lactate concentration in the medium (Fig. 2E). Compared with cells overexpressing mCherry alone (57.5 Ϯ 4.8 mM), cells overexpressing Txnip had greatly reduced lactate levels (30.6 Ϯ 1.9 mM, p Ͻ 0.001). Overexpression of the ␣-arrestin Arrdc4 also strongly inhibited lactate release to the medium (28.9 Ϯ 3.2 mM, p Ͻ 0.001 versus mCherry alone). In contrast, overexpression of Arrdc3 had no effect on lactate levels (62.2 Ϯ 6.4, p ϭ 0.58 versus mCherry alone). Overexpression of a fourth human ␣-arrestin, Arrdc2, also had no effect on lactate release (supplemental Fig. S1). Uptake was normalized to total protein content and expressed relative to the mean value of uptake for mCherry alone. 2-Deoxyglucose uptake was significantly inhibited by both Txnip and Arrdc4 but not Arrdc3 compared with mCherry alone (n ϭ 5; **, p Ͻ 0.01; ***, p Ͻ 0.001). SEPTEMBER 11, 2009 • VOLUME 284 • NUMBER 37

JOURNAL OF BIOLOGICAL CHEMISTRY 24999
To show that Arrdc4 inhibits glucose uptake, we determined the effect of overexpression of ␣-arrestins in primary human skin fibroblasts. 2-[3H]deoxyglucose uptake was divided by total protein content of the cell lysates to yield uptake per weight of total protein, then normalized to the level of the mCherry control. Overexpression of Txnip decreased glucose uptake to 63 Ϯ 6% of control levels (p Ͻ 0.01 versus mCherry only) (Fig. 2F). Overexpression of Arrdc4 also strongly decreased glucose uptake, to 29 Ϯ 3% of control levels (p Ͻ 0.001 versus mCherry only), while overexpression of Arrdc3 had no significant effect (111 Ϯ 11%; p ϭ 0.42 versus mCherry only). These data confirm that Txnip and Arrdc4, unlike Arrdc3, share the ability to regulate glucose uptake.
Txnip-and Arrdc4-specific Motifs Do Not Regulate Lactate Output-In contrast to the ␤-arrestins, almost no structurefunction information is currently available for the ␣-arrestins (14,24). Because Txnip and Arrdc4 had similar metabolic regulatory effects while Arrdc3 did not, we hypothesized that motifs conserved in Txnip and Arrdc4 but not Arrdc3 or Arrdc2 might identify molecular mechanisms that control ␣-arrestin function. Alignments of Txnip, Arrdc4, Arrdc3, and Arrdc2 revealed only 5 potential motifs of more than one residue (Fig. 3A). Two sites were particularly interesting because they were part of a PXXP sequence, a known binding motif for SH3-domain-containing proteins (39). We constructed point mutations in all five candidate motifs by replacing the conserved Txnip residue with the homologous residue in Arrdc3 or Arrdc2. A Txnip-mCherry fusion in a CMV-driven vector was used as the template to allow verification of expression and localization for each mutant.
To screen the effects of the candidate motifs on metabolic regulatory function, we tested whether each mutant inhibited lactate output when transfected into 293T cells. Compared with mCherry alone, Txnip significantly inhibited lactate release to the medium (12.8 Ϯ 0.3 mM versus 17.9 Ϯ 0.6 mM; Fig. 2B). Overexpression of each mutant also inhibited lactate release (P135L: 11 Ϯ 0.3 mM; Y221F: 13.2 Ϯ 0.9 mM; H241S: 13.5 Ϯ 0.4 mM; K286S: 11.6 Ϯ 0.3; P347A: 14.7 Ϯ 0.2 mM; all comparisons to mCherry alone were significant with p Ͻ 0.01). Thus, none of the candidate motifs appeared to have a role in inhibiting lactate release.
However, we did observe that mutation of Txnip P135L caused loss of the nearly exclusive nuclear localization of Txnip-Cherry (Fig.  3C). The P135L mutation did not prevent nuclear accumulation but greatly enhanced cytoplasmic localization of the protein. Txnip accumulates in the nucleus by interaction with an importin-␣ (34), and Txnip has two classical monopartite nuclear localization sequences (residues 5-8: KKIK and residues 163-7: KKEKK). Proline 135 is therefore not close in primary structure to the nuclear localization sequences. Alternatively, the PXXP motif from residues 132 to 135 may be a functional SH3-binding motif with a nuclear binding partner, leading to enhanced nuclear accumulation of Txnip. Nevertheless, this change in localization did not affect Txnip inhibition of lactate output.
These results demonstrate that while a single cysteine is required for interaction with thioredoxin, a different single cysteine, partly conserved but not Txnip/Arrdc4-specific, is required for Txnip metabolic function in this setting. To confirm that Txnip C170S can still bind to thioredoxin, we performed the AMS-alkylation binding assay with overexpression of Txnip wild type and C170S in 293T cells. Western analysis confirmed the presence of a Txnip C170S-thioredoxin complex (supplemental Fig. S2A).
The Cys-170 of Txnip is notable for its predicted location just adjacent to the linker or "hinge" sequence that connects the two arrestin domains (14). In the ␤-arrestins, the flexibility of the hinge sequence is thought to be important for allowing a conformational shift on binding to a phosphorylated receptor (16). Across ␣-arrestin evolution, sequences homologous to Txnip Cys-170 are also remarkable for having a short stretch of residues enriched in vicinal cysteines in diverse combinations (supplemental Fig. S2B). We therefore hypothesized that the homologous mouse Arrdc4 cysteine (Cys-184) would also participate in regulating Arrdc4 metabolic function. However, overexpression of both wild type and C184S Arrdc4 in 293T cells inhibited lactate release to the medium compared with empty vector controls (supplemental Fig. S2C). Therefore, the Txnip Cys-170 appears to be a critical regulator of Txnip metabolic function but does not explain the conservation of function between Txnip and Arrdc4.
The C-terminal Tail Is Not Required for Regulation of Glucose Uptake and Lactate Output-Because differences in primary sequence did not appear to explain the conservation of function between Txnip and Arrdc4, we next determined whether specific ␣-arrestin domains were required for metabolic function. Like the ␤-arrestins, the ␣-arrestins have two arrestin domains and a C-terminal tail domain. However, the ␣-arrestin C-terminal tail is proline-rich with highly conserved PPXY motifs, a defining difference from the ␤-arrestins (14). PPXY motifs interact with WW domain-containing proteins, which include several ubiquitin ligases. Two groups have very recently shown that yeast ␣-arrestins mediate receptor endocytosis and ubiquitination through the PPXY motifs in the C-terminal tail (25,24).
It is unknown whether the C-terminal tail contributes to ␣-arrestin metabolic function. We therefore tested C-terminal truncations of Txnip for ability to inhibit lactate output. V5-tagged full-length Txnip (1-391) and 3 truncations (  ␣-Arrestins Regulate Metabolism SEPTEMBER 11, 2009 • VOLUME 284 • NUMBER 37 truncated Txnip proteins. These results demonstrate that the ␣-arrestin C-terminal tail is not required for metabolic function; they suggest instead that this function is intrinsic to the arrestin domains themselves. Regulation of Lactate Output and Glucose Uptake Is Intrinsic to the Arrestin Domains-As truncation mutants may abolish function by interfering with secondary and/or tertiary structure, we used a domain swap strategy to test whether one or both arrestin domains are required for metabolic regulation. Extensive work with chimeric ␤-arrestins suggests that the arrestin domains tolerate swaps well and that the chimeras have yielded reliable structure-function insights (40).
Therefore, in addition to regulation of lactate output, Txnip also inhibits cellular glucose uptake without need for its C-terminal tail. However, swaps of the N-terminal arrestin domain suggest that at least a partial effect on glucose uptake is retained with either one of the Txnip arrestin domains. Because only a trend toward inhibited lactate output was seen with one of the N-arrestin swaps ("3-T-T"), it is possible that the chimeric proteins are less robust in their function. On the other hand, inhibition of glucose uptake was a reproducible and sensitive assay for Txnip/Arrdc4 function.
These results support the concept that conserved aspects of the arrestin domains beyond primary sequence motifs contribute to the metabolic functions of Txnip and Arrdc4. Further-FIGURE 6. Effects of Txnip-Arrdc3 chimeras on lactate output and glucose uptake. A, human Txnip and Arrdc3 domains were interchanged to make four chimeric constructs. All were made with C-terminal mCherry fusions in a lentiviral transfer vector. B, 293T cells were transfected with mCherry alone, Txnip, Arrdc3, and the 4 Txnip-Arrdc3 chimeras. The conditioned medium was collected and assayed for lactate (n ϭ 5; ***, p Ͻ 0.001 versus mCherry alone). C, primary human skin fibroblasts were infected with lentiviral particles from the same constructs. 2-[ 3 H]Deoxyglucose uptake was measured, normalized to total protein content, and expressed relative to the mean value of uptake for mCherry alone. (n ϭ 5-6; **, p Ͻ 0.01; ***, p Ͻ 0.001 versus mCherry alone). more, even though Arrdc3 does not inhibit glucose uptake, chimeric Txnip-Arrdc3 proteins with only one Txnip arrestin domain appear functional, suggesting that the arrestin domains of Arrdc3 retain similar structure-function relationships as Txnip and Arrdc4. Although very little is known of Arrdc3 function, it has been linked to the risk of developing obesity in the DeCODE study of the Icelandic population (41), suggesting it may also have a role in regulation of metabolism.
These studies therefore show that Txnip inhibits glucose uptake and lactate production independent of thioredoxin binding. This suggests that, while previously thought to function primarily as an inhibitor of thioredoxin, Txnip may actually function primarily as an ␣-arrestin that is regulated by thioredoxin. Identification of both Txnip and the related ␣-arrestin Arrdc4 as inhibitors of glucose uptake suggest that the animal ␣-arrestins are a new family of metabolic regulators. As the mammalian ␣-arrestin functions are further defined, we speculate that Txnip may be unique in its ability to regulate metabolism in response to both glucose and intracellular redox state, as it interacts with reduced but not oxidized thioredoxin.