TXNIP Interaction with GLUT1 Depends on PI(4,5)P₂

Abstract

GLUT1 is a major glucose facilitator expressed ubiquitously among tissues. Upregulation of its expression plays an important role in the development of many types of cancer and metabolic diseases. Thioredoxin-interacting protein (TXNIP) is an α-arrestin that acts as an adaptor for GLUT1 in clathrin-mediated endocytosis. It regulates cellular glucose uptake in response to both intracellular and extracellular signals via its control on GLUT1-4. In order to understand the interaction between GLUT1 and TXNIP, we generated GLUT1 lipid nanodiscs and carried out isothermal titration calorimetry and single-particle electron microscopy experiments. We found that GLUT1 lipid nanodiscs and TXNIP interact in a 1:1 ratio and that this interaction requires phosphatidylinositol 4,5-bisphosphate (PI(4,5)P₂ or PIP₂).

Graphical abstract

1. Introduction

Glucose is an important metabolic fuel and building block for life. There are many transporters that can facilitate cellular glucose uptake [1-3], Among the most studied are the ubiquitously expressed GLUT1; the dominant transporter in muscle and adipose tissue, GLUT4; the low-affinity transporter in liver, GLUT2; the high-affinity transporter in brain, GLUT3; and the sodium-dependent transporters SGLTs¹, found in the kidney and intestine. GLUT1-4 are uniporters, where glucose transport follows the concentration gradient. Once inside the cell, glucose is promptly phosphorylated to form glucose-6-phosphate, preventing it from exiting the cell. The glucose flux mediated by these uniporters depends on how many transporter molecules are on the cell surface, which is regulated by thioredoxin-interacting protein (TXNIP)-mediated clathrin-dependent endocytosis [4, 5]. TXNIP is an α-arrestin protein that serves as a signaling node for cellular control of glucose uptake via GLUT1-4. Both energy-stress-activated AMPK and insulin-activated AKT kinases can phosphorylate TXNIP and lead to its degradation, thus stopping the endocytosis of the GLUT transporters and leading to an immediate increase in glucose influx.

The GLUT1 structure has been solved via crystallography [6]. Its outward-open conformation allows glucose to bind to the protein, and the switch to inward-open conformation releases glucose to the cytoplasm. The TXNIP structure has also been solved by crystallography both in complex with thioredoxin [7] and as the truncated N-terminal domain alone [8]. The overall shape of TXNIP looks very much like β-arrestins, with two arrestin-fold β-sandwiches that have a high degree of flexibility relative to each other. Previously, we have shown that TXNIP can interact with phosphoinositides such as PIP₂, which is abundant on the plasma membrane [4]. PIP₂ is also the lipid marker for many other adaptors, such as AP2, in the clathrin-mediated endocytosis pathway [9]. The residues K233 and R238 in the TXNIP C-terminal domain are important for this lipid interaction. It is unclear whether TXNIP interaction with GLUT1 is dependent on its interaction with phosphoinositides. This interaction is difficult to study because the detergents used to purify GLUT1 and keep it in solution are not compatible with lipids. After all, detergents are used to dissolve away the lipids surrounding the membrane protein in order to extract the protein.

Recently, methods have been developed to enclose membrane proteins together with lipids to form nanodiscs [10-12]. The method that uses the engineered derivatives of apolipoprotein A1 scaffold proteins, also known as MSPs (membrane scaffold proteins), is relatively easy. Various length MSPs are available for generating nanodiscs of different sizes [13-15].

By enclosing purified human GLUT1 in PIP₂-containing MSP1E3 nanodiscs, we were able to study the GLUT1 interaction with TXNIP via size-exclusion chromatography, isothermal titration calorimetry (ITC), and electron microscopy (EM). We show that TXNIP interacts with GLUT1 in a 1:1 ratio and that this interaction requires the phospholipid PIP₂. In addition, we provide evidence that rat GLUT5, a fructose transporter, does not interact with TXNIP in vitro.

2. Materials and Methods

2.1. Lipid preparation

Brain PIP₂ (840046), egg phosphatidyl choline (PC) (131601), and soy PC (441601) were obtained from Avanti Polar lipids. Fifty milligrams of egg PC and 50 mg of soy PC were dissolved together in 2 mL of chloroform to give a 1:1 mixture of PC. One milligram of PIP₂ was dissolved in 1 mL of chloroform and mixed with 0.5 mL of the 1:1 PC mixture. This PIP₂/PC mixture was aliquoted 150 μL per glass vial (0.1 mg PIP₂ and 2.5 mg PC), dried with a speed vacuum concentrator, and stored at −80 °C for later use. The mixed PC was also dried and stored at 2.6 mg per vial.

2.2. MSP1E3 purification

The expression plasmid pMSP1E3D1 was purchased from Addgene (20066) [16]. The plasmid was transformed into BL21 cells, and the culture was grown in Terrific broth at 37 °C until the OD₆₀₀ reached 0.6-0.7. Then the culture was induced with 0.5 mM isopropylthio-β-galactoside (IPTG) at 30 °C for 4 h before being spun down. The cell pellet from 2 L of culture was resuspended in 30 mL of lysis buffer (30 mM Tris pH 8.0, 10 mM imidazole, 200 mM NaCl, 2 mM MgSO₄, and 10% glycerol), before being frozen in liquid nitrogen and stored at −80 °C for later purification.

After thawing the frozen cell pellet, NP-40 detergent was added to a final concentration of 1%, and the cells were lysed by sonication. The cell lysate was cleared with a 1-h centrifugation at 34,000 × g. The cleared lysate was loaded onto a 5-mL Ni-NTA column (Cytiva) and washed successively with 250 mL of buffer 1 (40 mM Tris pH 8.0, 300 mM NaCl, and 1% NP40), 250 mL of buffer 2 (40 mM Tris pH 8.0, 300 mM NaCl, 50 mM sodium cholate, and 20 mM imidazole), and 250 mL buffer 3 (40 mM Tris pH 8.0, 300 mM NaCl, and 50 mM imidazole). The protein was then eluted with 40 mM Tris pH 8.0, 300 mM NaCl, and 400 mM imidazole. Homemade TEV protease was added to the protein and the solution was dialyzed overnight against 2 L of 20 mM Tris pH 7.5, 100 mM NaCl, and 12.5 mM β-mercaptoethanol. This step removes the His-tag on MSP1E3. The next day, the protein solution was passed through a 2-mL Ni-NTA column to remove the TEV and the residual MSP1E3 that still had His-tag. The protein was then concentrated, aliquoted, and stored at −80 °C for later use.

2.3. GLUT protein purification

Human GLUT1 (N45T) was cloned into pFastBac 1 vector (Thermo Fisher) with a C-terminal His 10 tag [6]. Baculovirus was generated using the Bac-to-Bac system (Thermo Fisher). High Five insect cells were used for protein expression. Rat GLUT5 was cloned and purified the same way as GLUT1.

A frozen cell pellet from 500 mL of cells was thawed and lysed in hypotonic buffer (10 mM Tris pH 7.5, 1.5 mM MgSO₄, 10 mM KCl, and 0.5 mM tris(2-carboxyethyl)phosphine [TCEP]) using a homogenizer. The lysate was spun at 40,000 rpm for 30 min in a Ti45 rotor to pellet the cell membrane, which was then solubilized with 2% n-dodecyl-β-D-maltoside (DDM) for 2 h at 4 °C (25 mM Tris pH 7.5, 150 mM NaCl, 10 mM MgSO₄, 5% glycerol, 2% DDM, and 0.5 mM TCEP). The solubilized membrane was clarified with another spin at 40,000 rpm for 30 min. The supernatant was mixed in a 1:1 volume ratio with binding buffer (50 mM Tris pH 7.5, 300 mM NaCl, 20 mM MgSO₄, 40 mM imidazole, and 10% glycerol) along with Ni-NTA beads for 1 h of batch-binding at 4 °C. Then the beads were packed into an empty column and washed with 100 mL of wash buffer (30 mM Tris pH 8.0, 30 mM imidazole, 300 mM NaCl, 5 mM MgSO₄, 10% glycerol, and 0.5 mM TCEP) and eluted in elution buffer (25 mM MES pH 6.0, 50 mM NaCl, 300 mM imidazole, 20 mM glucose, 3% glycerol, and 0.5 mM TCEP). The most concentrated fractions were immediately buffer-exchanged to a reduced salt buffer (20 mM MES pH 6.0, 50 mM NaCl, 5 mM glucose, 0.2% DDM, and 1 mM TCEP) on a PD10 column. PIP₂/PC mixed lipids (2.6 mg) were solubilized in 0.75% DDM and added to the protein. The protein solution was diluted to reduce the final DDM concentration to 0.2% and then was incubated at 4 °C for 2 h to equilibrate. Then purified MSP1E3 was added for 1 h (280 μL of MSP1E3 measured by A₂₈₀ = 6.16 into 2.6 mg of lipids). Because the nanodiscs are formed with excess lipids, the amount of MSP1E3 added was constant to the lipid amount instead of the GLUT1 amount. To remove detergent, 500 mg of Bio-Beads SM-2 (Bio-Rad 1523920) was then added to the mixture and rotated gently at 4 °C overnight. The next day, after removing the Bio-Beads, more Ni-NTA resin was added for a 1-h batch bind, packed in a column, washed with 50 mL of wash buffer without DDM, and eluted in elution buffer without DDM. The protein was then concentrated in a 30-kDa cut-off concentrator and loaded onto an analytical Superdex 200 column. The final buffer was 30 mM Tris pH 7.5, 50 mM NaCl, 0.1% glycerol, and 1 mM dithiothreitol (DTT).

2.4. TXNIP purification

Human TXNIP was cloned into a pPROEX HTa vector in a stepwise manner to generate a dual-expression cassette. TXNIP (C120S/C170S/C205S/C267S)-His6 [7] was expressed under the Trc promoter in pPROEX HTa, followed by T7 promoter–driven E. coli thioredoxin with a C36S mutation. The vector was then transformed into Rosetta-gami 2 cells for expression. The overnight starter culture was grown with carbenicillin, tetracycline, and chloramphenicol and was used to inoculate the large culture. The expression culture was grown with carbenicillin and tetracycline at 30 °C until the OD₆₀₀ reached 0.7. Then, 0.3 mM IPTG was added, and the temperature was lowered to 19 °C for overnight induction. The next day, cells were spun down and frozen until needed.

Cells were lysed on ice with 0.5% octylthioglucoside in 30 mM Tris pH 8.0, 10 mM imidazole, 200 mM NaCl, 2 mM MgSO₄, and 10% glycerol. Twenty-five milligrams of lysozyme was added per 3 L of culture. After 30 min, DNase was added and the lysate was cleared with a 1-h, 34,000 × g spin. TXNIP was purified using a 5-mL Ni-NTA column with 250 mL wash buffer (30 mM Tris pH 8.0, 30 mM imidazole, 300 mM NaCl, 5 mM MgSO₄, 10% glycerol, and 0.5 mM DTT) and eluted with 30 mM Tris pH 8.0, 250 mM imidazole, 100 mM NaCl, 5 mM MgSO₄, 10% glycerol, and 0.5 mM DTT. The eluted protein was then concentrated and frozen for later use.

When forming TXNIP-GLUT1 complexes, we purified the TXNIP further using an analytical Superdex 200 column in a buffer of 30 mM Tris pH 7.5, 50 mM NaCl, 5% glycerol, and 1 mM DTT. Only the monomeric portion of TXNIP was used for further assay.

The TXNIP K233A/R238A mutant was cloned by site-directed mutagenesis and purified similarly.

2.5. Isothermal titration calorimetry

Isothermal titration calorimetry was carried out with MicroCal PEAQ-ITC (Malvern). TXNIP and GLUT1/lipid/MSP1E3 particles were purified on the day of ITC with the same buffer (30 mM Tris pH 7.5, 50 mM NaCl, 5% glycerol, and 1 mM DTT). Concentrated nanodisc particles were placed in the syringe, and TXNIP was in the sample cell. On average, 0.25 mM nanodiscs and 30 μM TXNIP (both measured by BCA assay against BSA standards) were used in the titration. The measurement was carried out at 22 °C, with 750 rpm stirring. Three hundred microliters of TXNIP was placed in the cell, and approximately 70 μL of nanodiscs was placed in the syringe. Injection volume from the syringe was 2 μL and wait time between injections was 200 s. To cover the full range of titration, we typically had to carry out two series of injections of nanodiscs (total 38 injections). Analysis and graphs were done with the MicroCal PEAQ-ITC analysis software accompanying the instrument. The fit was adjusted by fixing N (number of binding sites) to 1 and varying K_D (binding affinity) and ΔH (binding enthalpy). The K_D value was reported as ± SD from three experiments.

2.6. Negative-stain electron microscopy

To assess GLUT1 incorporation into lipid nanodiscs and TXNIP binding to GLUTI/PIP₂/MSP1E3 nanodiscs, negative-stain grids were prepared and examined. The protein sample (2 μL) was applied to a freshly plasma-cleaned, 300-mesh, carbon-coated copper grid (EMS) for 1 min. The grid was then blotted to remove excess sample, briefly stained in a 20-μL droplet of 2% uranyl formate, blotted, and then stained in a second 20-μL droplet of 2% uranyl formate, and incubated for 1 min before blotting away excess stain. Negative-stain grids were imaged using a Thermo Fisher Scientific Tecnai Spirit G2 BioTWIN microscope operating at 120 kV with a nominal magnification of 30,000× at a pixel size of 2.1 Å and a defocus range of −0.8 μm to −2.1 μm.

2.7. Cryo-EM sample preparation and data acquisition

Three microliters of purified GLUTI/PIP₂/MSP1E3/TXNIP complex (about 2.5 mg/mL measured by A₂₈₀) was applied to Quantifoil R1.2/1.3 300-mesh Au grids. We observed a strongly preferred orientation of our sample after normal glow-discharging. To improve particle orientation, we glow-discharged in the presence of amylamine for 20 s to yield positively charged grids. After sample application, grids were blotted for 4 s, and then plunge-frozen in liquid ethane using a Thermo Fisher Scientific Vitrobot Mark IV at 100% RH and 4 °C.

Cryo-electron microscopy (cryo-EM) images were collected using the Thermo Fisher Scientific Talos Arctica electron microscope operating at 200 kV. Images were acquired with a Falcon 3EC direct electron detector operating in counting mode at a nominal 120,000x magnification (a pixel size of 1.215 Å) with a defocus range of −1.5 μm to −2.5 μm. Each micrograph was exposed for 75 s with 0.95 e⁻/pixel²/s dose rate (total accumulated dose, 48 e⁻/A²), and about 126 frames were captured per movie stack. EPU, an automated acquisition program (Thermo Fisher Scientific), was used for data collection.

2.8. Image processing

For negative-stain images, a dataset of 308 micrographs was collected and 100,514 particles were picked using EMAN2/BOXER [17], All particle extractions, two- and three-dimensional (2D and 3D) classifications, and refinements were performed using RELION 3.0 [18]. Particles were extracted and binned 2 X 2 (4.2 Å/pixel, 128-pixel box size) and subjected to reference-free 2D classification using a 196 Å mask diameter. After several rounds of 2D class averaging, the best classes were selected and re-extracted unbinned (2.1 Å/pixel, 128-pixel box size), resulting in a dataset of 54,207 particles. Starting with a low-resolution sphere as a model, 3D classification was performed in RELION using C1 symmetry. The particles from the most populated class (about 15,644) were used for 3D refinement resulting in an overall resolution of 18.4 Å for the final 3D reconstruction. The 3D map was adjusted to a threshold of 0.04 using UCSF Chimera [19]. The 3D figure was prepared using UCSF Chimera.

For cryo-EM, a modest dataset of 813 micrographs was collected over two collection dates. Movies were motion-corrected with dose-weighting using MotionCor2 [20]. Subsequent processing was performed in RELION 3.0. Contrast transfer function (CTF) was determined using CTFFIND4.1 [21].

Particles were automatically picked using RELION 3.0’s reference-free Laplacian-of-Gaussian [22]. A total of 674,263 particles were extracted and binned 4 x4 (4.86 Å/pixel, 104-pixel box size) from 813 dose-weighted micrographs and subjected to reference-free 2D classification using a 196 Å mask diameter. The best 2D class averages with the GLUT1/TXNIP complex in nanodiscs were selected for further processing and were recentered and reextracted unbinned (1.215 Å/pixel, 196-pixel box size). A total of 156,124 particles from the best 2D class averages showing secondary-structure and multiple views was selected after several rounds of 2D class averaging for 3D classification. However, due to the challenges of processing small, dynamic protein complexes, we were not able to determine 3D structure to high resolution.

3. Results

3.1. GLUT1 reconstitution into nanodiscs

The N45T mutant of human GLUT1 with a C-terminal His-tag was purified from insect cells as described [6](Fig. 1A). The eluted protein was quickly buffer-exchanged to reduce the imidazole concentration and was mixed with excess lipids, in this case, a 1:1 mix of soy and egg PC. After a 2-h incubation to allow lipid association with the GLUT1 protein, purified MSP1E3 was added to enclose the protein and lipids. Detergent was then removed using Bio-Beads overnight. A second Ni-NTA column was used to collect the GLUT1-containing nanodiscs (Fig. 1B). During the final purification step using a Superdex 200 size-exclusion column, we typically saw a sharp peak around 12 mL, plus some larger species that were excluded from the downstream analysis (Fig. 1C, 1D).

Figure 1.

Insertion of GLUT1 into lipid nanodiscs

(A) Ni-NTA elution of GLUT1 fractions 1 to 6 plus PD10 buffer-exchanged GLUT1. (B) Ni-NTA elution of GLUT1/PIP₂/MSP1E3 nanodisc fractions. (C) Superdex 200 elution profile (standard molecular weight markers are indicated by dashed lines) and (D) SDS-PAGE gel (fractions 8-21) of GLUT1/PIP₂/MSP1E3 nanodiscs.

To characterize GLUT1 nanodiscs, we carried out negative-stain EM. As a control, we first looked at empty PIP₂/MSP1E3 nanodiscs. The negative-stain micrographs revealed good particle density and showed top views of the empty PIP₂/MSP1E3 nanodiscs as well as side views, which are sometimes seen in a stacked formation (Fig. 2A). In our experiments we found that the negatively stained, empty nanodiscs often formed such stacks, which have been observed in other nanodisc studies as well [23-26]. We could see the characteristics of the empty PIP₂/MSP1E3 nanodiscs from the negative-stain reference-free 2D classes (Fig. 2B). The upper row of Fig. 2B shows the top view of the empty PIP₂/MSP1E3 nanodisc and the lower row shows the side view, including one class with the nanodisc stack formation. In contrast to the empty PIP₂/MSP1E3 nanodiscs, the reference-free 2D class averages of the GLUT1/PIP₂/MSP1E3 nanodiscs showed GLUT1 protein density in the center of the nanodisc (upper row) and a protruding GLUT1 density showed in the side and angled views (lower row) (Fig. 2D). The minimal nanodisc stacking in the raw micrograph (Fig. 2C) further confirmed the incorporation of GLUT1 in nanodiscs, because bound protein on the nanodisc has been demonstrated to interfere with stack formation [24].

Figure 2.

Electron microscopy characterization of GLUT1 nanodiscs

(A) A representative negative-stain EM micrograph of PIP₂/MSP1E3 nanodiscs; some particles display the stacked formation. (B) 2D class averages from negative-stain PIP₂/MSP1E3 nanodisc images, showing top views (upper row) and side views (lower row). Scale bar represents 10 nm. (C) A representative negative-stain EM micrograph of GLUT1/PIP₂/MSP1E3 nanodiscs. With bound GLUT1, the nanodisc stacking is reduced. (D) 2D class averages from negative-stain images of GLUT1/PIP₂/MSP1E3 nanodiscs, showing top views (upper row) and side views (lower row). Scale bar represents 10 nm.

3.2. TXNIP interacts with GLUT1/PIP₂/MSP1E3 nanodiscs in a 1:1 ratio

To investigate GLUT1 binding to TXNIP, we purified human TXNIP from bacteria using a homemade dual expression vector that co-expresses thioredoxin to stabilize TXNIP. This TXNIP construct has four of its 11 cysteines (C120S, C170S, C205S, and C267S) mutated to serine to reduce protein aggregation during expression and purification [7].

GLUT1 is normally located on the plasma membrane where PIP₂ is abundant. PIP₂ is an important localization signal for many clathrin-associated endocytosis adaptors, including TXNIP. Using a floating assay, we have previously shown that TXNIP interacts with liposomes containing PIP₂ but not with liposomes containing only PC/PS (phosphatidyl serine) [4].

To ensure that we mimicked the endogenous conditions as closely as possible, we started by examining TXNIP interaction with GLUT1 nanodiscs containing 4% PIP₂ (w) in a 50:50 mixture (w:w) of soy and egg PC. First, we compared the size-exclusion column FPLC profile of TXNIP by itself (Fig. 3A) and TXNIP in complex with GLUT1/PIP₂/MSP1E3 nanodiscs (Fig. 3B). We observed that a significant amount of TXNIP co-migrated together with the GLUT1 nanodiscs; this complex also eluted earlier than GLUT1 nanodiscs alone (Fig. 3C, 3D). This result suggests a direct interaction of TXNIP with the nanodiscs.

Figure 3.

TXNIP interaction with GLUT1/PIP₂/MSP1E3 nanodiscs

(A) SDS-PAGE gel of TXNIP elution fractions 8-21 from Superdex 200. (B) SDS-PAGE gel of GLUT1/PIP₂/MSP1E3 nanodiscs with TXNIP elution fractions 8-18 from Superdex 200. (C) The elution profile from panel B. Standard molecular weight markers are indicated by dashed lines. (D) Overlay of elution profiles of GLUT1/PIP₂/MSP1E3 nanodisc (dashed line) and GLUT1/PIP₂/MSP1E3/TXNIP (solid line).

Next, we carried out negative-stain EM to visualize the complex. The micrographs showed a monodisperse sample with no nanodisc stacking (data not shown), and the resulting reference-free 2D classes showed a clear TXNIP density in both the top and side views (Fig. 4A). The best 2D classes representing both side and top views were further refined with 3D classification/refinement. The final 3D reconstruction reached 18.4 Å overall resolution with C1 symmetry applied (Fig. 4B). The 3D reconstruction model calculated from negatively stained single particles revealed for the first time the overall architecture of the GLUT1/PIP₂/MSP1E3/TXNIP complex: it showed a 1:1 ratio of GLUT1/PIP₂/MSP1E3 nanodiscs and TXNIP.

Figure 4.

Interaction between TXNIP and GLUT1/PIP₂/MSP1E3 nanodiscs is 1:1 by electron microscopy

(A) 2D class averages from negative-stain GLUT1/PIP₂/MSP1E3/TXNIP images. Scale bar represents 10 nm. (B) Side and top views of the GLUT1/PIP₂/MSP1E3/TXNIP negative-stain 3D reconstruction. (C) Representative cryo-EM 2D class averages of GLUT1/PIP₂/MSP1E3/TXNIP showing secondary structure elements. Scale bar represents 10 nm.

To study the interaction of the complex at higher resolution, we examined our sample via cryo-EM. Gaussian-picked particles were subjected to reference-free 2D classification, which resulted in 2D class averages with top and side views showing clear secondary structure (Fig. 4C). The 2D images showed a poorly defined TXNIP C-terminal domain relative to the rest of the complex, as was also seen in the negative-stain data, indicating that this is likely a highly flexible region. Due to the variability in this flexible TXNIP region and because the relative position of MSP1E3 and GLUT1 is not fixed, we could not obtain a confident 3D cryo-EM model. The disordered molecules from the PIP₂/MSP1E3 nanodisc obscured the identification of the GLUT1 helices. However, the cryo-EM 2D classification corroborated the 1:1 ratio of GLUT1/PIP₂/MSP1E3 nanodisc and TXNIP that we saw in the negative-stain data.

3.3. Interaction of GLUT1 with TXNIP is dependent on TXNIP interaction with PIP₂

It was technically challenging to obtain a high-resolution structure of the TXNIP–T nanodisc complex. Therefore, we studied this interaction further by ITC, which is more sensitive and quantitative than analyzing complex formation on the size-exclusion column. Using a Superdex 200 column, we could not confidently detect the interaction between TXNIP and empty PIP₂/MSP1E3 nanodiscs (Fig. 3A, 5A and 5B), but we could measure the equilibrium dissociation constant (K_D) to be 43.3 ± 2.0 μM by ITC (Fig. 5C) (average of 3 experiments). The K_D between TXNIP and GLUTI/PIP₂/MSP1E3 was 3.8 ± 0.7 μM (average of 3 experiments), roughly a tenfold tighter interaction (Fig. 5D). This indicates that the presence of GLUT1 significantly increases TXNIP affinity for the nanodiscs and implies there is specific, direct protein interaction between GLUT1 and TXNIP.

Figure 5.

TXNIP interaction with GLUT1 requires PIP₂

(A) Fractions 8-21 of PIP₂/MSP1E3 nanodiscs. (B) Fractions 8-21 of PIP₂/MSP1E3/TXNIP. (C) Representative ITC of PIP₂/MSP1E3 nanodisc interaction with TXNIP. Differential power (DP) resulting from the injections is plotted against time, and binding enthalpy (ΔH) is plotted against the molar ratio of the proteins. The equilibrium dissociation constant (K_D) = 43.3 ± 2.0 μM. ΔH = −3.97 ± 0.62 kcal/mol. (D) Representative ITC of GLUT1/PIP₂/MSP1E3 nanodisc interaction with TXNIP. K_D = 3.8 ± 0.7 μM. ΔH = −6.94 ± 0.86 kcal/mol. The molar ratio of GLUT1:TXNIP is set at 1 for graph fitting. (E) Representative ITC of GLUT1/PC/MSP1E3 nanodiscs with TXNIP. (F) Representative ITC of GLUT1/PIP₂/MSP1E3 nanodiscs with the TXNIP K233A-R238A mutant.

To assess this GLUT1-specific TXNIP interaction, we inserted GLUT1 into lipid nanodiscs in the absence of PIP₂ and carried out ITC. To our surprise, we did not detect an interaction between GLUT1/PC/MSP1E3 nanodiscs with TXNIP (Fig. 5E). To confirm this, we purified the TXNIP K233A-R238A mutant, which does not interact with lipids [4]. This time, we used the GLUT1/PIP₂/MSP1E3 nanodiscs for isothermal titration. Again, there was no detectable interaction (Fig. 5F), which suggests that the electrostatic attraction between the basic residues K233/R238 and the anionic lipid, PIP₂, contributes strongly to the TXNIP–GLUT1 complex formation. This is supported by Gibbs free energy of interaction calculated from the K_D, ΔG (= RT ln K_D) = −7.27 ± 0.1 kcal/mol, and the large enthalpy contribution ΔH = −6.94 ± 0.86 kcal/mol measured by ITC. When compared to the negligible enthalpy if this electrostatic interaction is either eliminated by replacing PIP₂ with PC or by using the K233A-R238A mutant, the TXNIP-PIP₂ interaction appears to be the major contributor to enthalpy (ΔH). Because ΔG = ΔH - TΔS, where ΔS is the binding entropy, this TXNIP-PIP₂ interaction also contributes significantly to the total binding energy while entropy has a minor contribution. Without this TXNIP–lipid interaction, the energy released from the TXNIP–GLUT1 interaction alone falls below the detection limit of ITC.

3.4. TXNIP does not interact with GLUT5

Given the seemingly low affinity of the TXNIP-GLUT1 protein–protein interaction, we wanted to know whether that interaction is nonspecific, in the sense that localization of TXNIP to the plasma membrane by PIP₂ may allow nonspecific interaction between TXNIP and any other membrane protein present. To address this question, we cloned rat GLUT5, expressed it, and prepared nanodiscs as done using GLUT1. GLUT5 transports fructose instead of glucose, but the two share such a similar structure that a single point mutation on GLUT5 is sufficient to switch its specificity from fructose to glucose [27]. We reasoned that if TXNIP interaction with the protein component is nonspecific, TXNIP would also interact with GLUT5 due to the structural similarity. Both a size-exclusion column (Fig. 6A) and ITC (Fig. 6B) showed no interaction between rGLUT5/PIP₂/MSP1E3 nanodiscs and TXNIP. This means that the protein–protein interaction between TXNIP and the glucose transporters is specific, supporting the role of TXNIP in the regulation of glucose metabolism in vivo [28, 29].

Figure 6.

TXNIP does not interact with GLUT5

(A) Fractions 8-21 of rGLUT5/PIP₂/MSP1E3 in complex with TXNIP. (B) ITC of rGLUT5/PIP₂/MSP1E3 nanodiscs with TXNIP. Differential power (DP) resulting from the injections is plotted against time, and binding enthalpy (ΔH) is plotted against the molar ratio of the proteins.

4. Discussion

Glucose transporters such as GLUT1 play a crucial role in glucose metabolism, and their dysregulation is related to cancer and metabolic syndromes. TXNIP negatively regulates glucose uptake by facilitating GLUT1 endocytosis. However, the specific details of the TXNIP-GLUT1 interaction have been poorly understood. By means of incorporation into a lipid nanodisc, we show here that GLUT1 interacts with TXNIP in a 1:1 ratio and that this interaction depends on the presence of PIP₂.

Understanding the structural relationship between GLUT1 and TXNIP is challenging due to the necessity of retaining the lipids that are relevant to the interaction. In this study, we successfully enclosed GLUT1 in a lipid environment using MSP1E3 protein. There are many MSP constructs of differing lengths that form discs of various diameters. We experimented with ΔH5 and 1D1, but these two seemed to be too small, and we had more protein loss during the procedure. In addition, we had to use excess lipids to fill the nanodiscs in order to create enough lipid surface for TXNIP interaction.

We characterized TXNIP interaction with GLUTI/PIP₂/MSP1E3 nanodiscs using a size-exclusion column, EM, and ITC. We found the interaction to be 1:1, and the dissociation constant was in the micromolar range. Because the protein concentration of the nanodiscs was estimated with a BCA assay, including MSP1E3, it is higher than the true GLUT1 concentration; therefore, the dissociation constant appears underestimated by ITC. However, under these conditions, we can clearly detect tighter interaction of TXNIP with nanodiscs containing GLUT1 than with empty nanodiscs. Therefore, protein-to-protein interactions certainly contribute to the affinity and biological specificity. This is verified by the absence of interaction between TXNIP and GLUT5 in the same lipid nanodiscs.

Intriguingly, this protein-to-protein interaction does not seem strong enough to hold GLUT1 and TXNIP together if TXNIP interaction with the phospholipids is abolished, either by using PC alone or by using the K233A-R238A mutant. There is the possibility that TXNIP association with the lipid is the first step for the TXNIP–GLUT1 interaction and that this association induces some conformational change on TXNIP that facilitates further interaction with GLUT1. In addition, the basic residues K233 and R238 may not be needed only for membrane localization, but also for interaction with GLUT1 residues. Unfortunately, we could not obtain a high-resolution 3D reconstruction to verify the structural details. To reduce the obscuring density from the MSP1E3 nanodisc, maybe a polymer such as SMALP (styrene–maleic acid copolymer) can be used [11]. SMALP can enclose the membrane protein in its native lipids during membrane extraction, and affinity purification can be used to enrich the protein of interest. However, due to the heterogeneity of polymers, this technology requires much trial and error for each protein of interest.

The interaction with phospholipids has also been demonstrated for the extensively studied β-arrestins [30-33]. β-arrestins, located in the cytoplasm under basal conditions, are recruited to the activated and phosphorylated G-protein-coupled receptors (GPCRs) at the plasma membrane to modulate their desensitization and to facilitate their endocytosis. Recent cryo-EM structures show that β-arrestins can interact with the lipid membrane in two ways: the C-terminal domain loops on the edge of the arrestin fold insert into the membrane (or detergent micelle) and more specific PIP₂ interacts with residues K232, R236 and K250 (hARRB1 numbering) in the same C-terminal domain [30, 31]. Even though these β-arrestin interactions with lipids strengthens the GPCR–β-arrestin complex, the β-arrestin binding to phosphorylated GPCR seems to be sufficient to initiate the complex formation. In contrast, TXNIP is normally located on the plasma membrane and cytoplasmic vesicular structures [4, 5]. Its interaction with GLUT1 is independent of GLUT1 phosphorylation. TXNIP K233/R238 binding to PIP₂ seems to be the major determining factor for TXNIP localization to membranes, without which TXNIP–GLUT1 interaction does not happen. Multiple sequence comparison does not align TXNIP K233/R238 with the PPI₂-interacting basic residues in β-arrestins. However, the basic patch formed by K233/R238 is located on the TXNIP C-terminal domain surface similarly as the basic patch on β-arrestins [7]. At the low resolution of our EM structures, we could not determine if the edge of the TXNIP C-terminal domain was interacting with the membrane or not. In conclusion, arrestins are highly plastic proteins and binding to the anionic membrane via a basic patch seems to be a shared characteristic.

The results of this study provide insight into the requirements for TXNIP interaction with GLUT1 and expand our understanding of this facet of glucose metabolism. From our present evidence, together with data in the literature, we conclude that TXNIP recruitment to the plasma membrane and interaction with GLUT1 on the plasma membrane requires its interaction with PIP₂.

Highlights.

• GLUT1 incorporates into lipid nanodisc using MSP1E3 protein

• Electron microscopy shows TXNIP interaction with GLUT1 in a 1:1 ratio

• TXNIP-GLUT1 interaction depends on TXNIP interaction with the phospholipid PIP₂

• TXNIP-GLUT1 interaction is specific; TXNIP does not interact with GLUT5