Insulin Releases Glut4 from Static Storage Compartments into Cycling Endosomes and Increases the Rate Constant for Glut4 Exocytosis*

In adipocytes, insulin triggers the redistribution of Glut4 from intracellular compartments to the plasma membrane. Two models have been proposed to explain the effect of insulin on Glut4 localization. In the first, termed dynamic exchange, Glut4 continually cycles between the plasma membrane and intracellular compartments in basal cells, and the major effect of insulin is through changes in the exocytic and endocytic rate constants, kex and ken. In the second model, termed static retention, Glut4 is packaged in specialized storage vesicles (GSVs) in basal cells and does not traffic through the plasma membrane or endosomes. Insulin triggers GSV exocytosis, increasing the amount of Glut4 in the actively cycling pool. Using a flow cytometry-based assay, we found that Glut4 is regulated by both static and dynamic retention mechanisms. In basal cells, 75-80% of the Glut4 is packaged in noncycling GSVs. Insulin increased the amount of Glut4 in the actively cycling pool 4-5-fold. Insulin also increased kex in the cycling pool 3-fold. After insulin withdrawal, Glut4 is rapidly cleared from the plasma membrane (t½ of 20 min) by rapid adjustments in kex and ken and recycled into static compartments. Complete recovery of the static pool required more than 3 h, however. We conclude that in fully differentiated confluent adipocytes, both the dynamic and static retention mechanisms are important for the regulation of plasma membrane Glut4 content. However, cell culture conditions affect Glut4 trafficking. For example, replating after differentiation inhibited the static retention of Glut4, which may explain differences in previous reports.

In adipocytes, insulin triggers the redistribution of Glut4 from intracellular compartments to the plasma membrane. Two models have been proposed to explain the effect of insulin on Glut4 localization. In the first, termed dynamic exchange, Glut4 continually cycles between the plasma membrane and intracellular compartments in basal cells, and the major effect of insulin is through changes in the exocytic and endocytic rate constants, k ex and k en . In the second model, termed static retention, Glut4 is packaged in specialized storage vesicles (GSVs) in basal cells and does not traffic through the plasma membrane or endosomes. Insulin triggers GSV exocytosis, increasing the amount of Glut4 in the actively cycling pool. Using a flow cytometrybased assay, we found that Glut4 is regulated by both static and dynamic retention mechanisms. In basal cells, 75-80% of the Glut4 is packaged in noncycling GSVs. Insulin increased the amount of Glut4 in the actively cycling pool 4 -5-fold. Insulin also increased k ex in the cycling pool 3-fold. After insulin withdrawal, Glut4 is rapidly cleared from the plasma membrane (t1 ⁄ 2 of 20 min) by rapid adjustments in k ex and k en and recycled into static compartments. Complete recovery of the static pool required more than 3 h, however. We conclude that in fully differentiated confluent adipocytes, both the dynamic and static retention mechanisms are important for the regulation of plasma membrane Glut4 content. However, cell culture conditions affect Glut4 trafficking. For example, replating after differentiation inhibited the static retention of Glut4, which may explain differences in previous reports.
The rate of glucose uptake in muscle and adipose tissue is limited by the total number of facilitative glucose transport proteins inserted into the plasma membrane (1,2). Insulin regulates glucose uptake in fat and muscle by controlling the subcellular distribution of glucose transporters, predominantly the Glut4 isoform. In the absence of insulin, Glut4 is largely excluded from the plasma membrane and is retained within endosomes, the trans-Golgi network, and tubular-vesicular compartments termed Glut4 storage vesicles (GSVs) 2 (1,2). Insulin triggers the redistribution of Glut4 from intracellular compartments to the plasma membrane. This increases the total number of transporters localized to plasma membrane and thus the maximal rate of glucose transport into cells. When insulin is removed, the Glut4 is rapidly cleared from the plasma membrane back into intracellular compartments, where it is recycled for use in subsequent stimulations.
How insulin regulates the subcellular distribution of Glut4 is still debated (for reviews, see Refs. 1 and 3). Two competing models have emerged. In the first, termed dynamic exchange or dynamic equilibrium, plasma membrane Glut4 continually exchanges with Glut4 in the endosomes and GSVs (3). Under basal conditions, slow exocytosis from GSVs together with fast endocytosis from the plasma membrane minimizes the amount of Glut4 at the cell surface. Insulin redistributes Glut4 to the plasma membrane by increasing the rate of Glut4 exocytosis (through an increase in the exocytic rate constant k ex ) while simultaneously decreasing the rate of Glut4 endocytosis from the plasma membrane (through a decrease in the endocytic rate constant k en ). Key features of this model are that all of the Glut4 is actively cycling in both basal and insulin-stimulated cells, and the effect of insulin is completely accounted for by changes in k ex and k en . A second model, termed static retention, differs from dynamic equilibrium in one important aspect. In static retention, Glut4 is sequestered in GSVs and does not cycle through the plasma membrane under basal conditions (4,5). Insulin stimulates the fusion of GSVs with the plasma membrane, releasing Glut4 into the cycling membrane system (plasma membrane, endosomes, and trans-Golgi network) (4). In this model, the GSVs behave like regulated exocytic vesicles. Insulin increases cell surface Glut4 through an increase in the total amount of Glut4 in the actively cycling pool as well as through an increase in k ex .
Early kinetic work used photoactivated glucose analogs to label Glut4 at the cell surface (6,7). Glut4 labeled at the cell surface at 4°C redistributed to intracellular compartments when cells were warmed to 37°C, under both basal and insulinstimulated conditions. Insulin increased k ex and decreased k en . * This work was supported by National Institutes of Health Grant DK56197.
These observations suggested that the distribution of Glut4 was controlled through dynamic exchange and that the effect of insulin on Glut4 localization was due to effects on the rate constants for exocytosis and endocytosis. In these studies, it was explicitly assumed that all of the Glut4 actively cycled between the plasma membrane and intracellular compartments in both basal and insulin-stimulated cells. The size of the actively cycling pool was not directly determined.
More recently, exofacial epitope-tagged Glut4 reporter constructs have been used to study Glut4 traffic in live cells (3, 4, 8 -11). Unlike endogenous Glut4, these reporters can be labeled with antibody when they are inserted into the plasma membrane. When expressed at low levels, the reporters traffic with the endogenous Glut4 (3,8,12). Both the rate constant of exocytosis and the size of the total cycling pool can be determined from the kinetics of antibody binding/uptake during continuous labeling at 37°C. Using this method, Martin et al. (3) concluded that the distribution of Glut4 in 3T3-L1 adipocytes was controlled by dynamic exchange. In their experiments, ϳ90% of the total Glut4 was cycling in both basal and insulin-stimulated cells. The major effect of insulin was to increase k ex by 7-10fold. Insulin also decreased k en by 70%. However, using the same cells, the same reporter constructs, and the same antibodies, Govers et al. (4,5) found that the distribution of Glut4 was controlled by static retention. The major effect of insulin was to increase the size of the actively cycling pool by ϳ7-fold (from 10% of the total to 70% of the total Glut4 in the cell). Insulin also increased k ex by 2-fold.
These two models suggest very different mechanisms regulating Glut4 traffic. To determine the physiological relevance of these observations, it is important to determine how differences in the experimental methods used in the two studies influence Glut4 traffic. Previous reports using anti-HA uptake to measure the kinetics of Glut4 traffic have either relied on single cell analysis (3) or whole plate assays (4). We developed a flow cytometry based assay to analyze the kinetics of Glut4 exocytosis (we had previously used flow cytometry to measure the kinetics of endocytic trafficking of transferrin and ␣-2-macroglobulin) (13)(14)(15). Flow cytometry combines the ability of microscopy to analyze individual cells with the ability of the whole plate assay to analyze large numbers of cells in each sample. We evaluated the effects of methodological differences between the studies of Martin et al. (3) and Govers et al. (4) on Glut4 trafficking in differentiated 3T3-L1 cells. Specifically, we tested the effect of replating quiescent, confluent differentiated cells, as is commonly done in electroporation and/or microscopy experiments using the 3T3-L1 cell culture model.

EXPERIMENTAL PROCEDURES
Cell Culture-3T3-L1 fibroblasts (CCL 92.1; American Type Culture Collection) were maintained in DMEM (high glucose, with 2 mM L-glutamine, 50 units/ml penicillin, and 50 g/ml streptomycin) supplemented with 10% calf serum (CS; Hyclone). For experiments, cells were plated into individual wells of a 96-well plate (0.12 ϫ 10 5 cells/well). At confluence (3 days after plating), the cells were switched to DMEM supplemented with 10% fetal bovine serum (FBS; Hyclone; characterized) and 5 M Troglitazone (Calbiochem). 1 day postconflu-ence (day 0), the cells were induced to differentiate into adipocytes essentially as described (16). Briefly, cells were incubated for 3 days in differentiation medium (DMEM, 10% FBS, 5 M Troglitazone, dexamethasone, isomethylbutylxanthine, and insulin), 3 days in insulin medium (DMEM, 10% FBS, Troglitazone, and insulin), and then 4 -5 days in DMEM, 10% FBS. Cells were used 8 -11 days after the initiation of differentiation. 3T3-L1 fibroblasts were used 2-3 days postconfluence. To test the effect of confluence on Glut4 traffic, differentiated 3T3-L1 cells were lifted from the culture plates by collagenase digestion and replated at lower density. Briefly, cells were differentiated in 96-well plates as described. Upon removal of insulin-containing medium, cells were incubated with collagenase (1 g/l in PBS plus 1% BSA, 20 l/well, 5 min, 37°C), fresh culture medium (DMEM, 10% FBS) was added to each well (200 l/well), and cells were suspended by gentle pipetting. 100 l from each well was transferred to a new 96-well culture plate. Cells were allowed to settle for 2 h. After settling, cells were refed with 120 l of culture medium, and cells were further cultured for 1, 2, or 3 days.
Lentivirus-cDNA encoding HA-Glut4 (from Dr. Samuel Cushman, National Institutes of Health, Bethesda, MD) (8) was amplified by PCR and inserted at the Age1 site in frame upstream of the GFP reporter of the pRRL-PGK-GFP vector (from Dr. Rob C. Hoeben, Leiden University Medical Center, Leiden, The Netherlands) (17). Lentivirus was prepared using the ViraPower lentiviral packaging mix and 293FT packaging cells following the manufacturer's recommended protocol (Invitrogen). 3T3-L1 fibroblasts were infected with recombinant lentivirus essentially following recommended protocols (Invitrogen). Cells were plated at low density in 35-mm culture dishes and allowed to recover overnight in 2 ml of DMEM, 10% CS. The following day, the growth medium was replaced with infection medium (1 ml of HA-Glut4/GFP lentiviral supernatant diluted to 2 ml in DMEM, 10% CS, supplemented with 6 g/ml Polybrene). Cells were incubated in infection medium for 24 h. Infection medium was then replaced with DMEM, 10% CS. Infected cells were allowed to reach 80% confluence and then cultured as described above. Infection efficiency was determined by flow cytometry. Under these conditions, ϳ80% of the cells expressed the HA-Glut4/GFP reporter construct.
Flow Cytometry-3T3-L1 cells (fibroblasts and differentiated adipocytes) were suspended by collagenase digestion (1 g/l in PBS plus 1% BSA, 20 l/well, 5 min at 37°C or 30 min at 4°C) and gentle pipetting in a final volume of 200 l of PBS plus 1% BSA. Suspended cells were filtered through a 100-m cell strainer. Samples were analyzed by flow cytometry using a FACScan (BD Biosciences) equipped with argon (488 nm) and helium/neon (633 nm) lasers. Log intensities of scattered light (forward scatter (FSC) and side scatter (SSC); 488 nm excitation) and fluorescence (FL1, 488 nm excitation/525 nm emission; FL2, 488 nm excitation/595 nm emission; FL3, 488 nm excitation/Ͼ640 nm emission; FL4, 633 nm excitation/Ͼ670 nm emission) were collected. Data collection thresholds were set based on FSC and SSC profiles for undifferentiated fibroblasts ( Fig. 1A) to omit small light scattering debris and machine noise from the analysis. Fluorescence detector gains were set so that background autofluorescence in noninfected cells was collected in the first log decade of the detector range.
Gating-Differentiated 3T3-L1 cells contain three populations of light scattering objects: fully differentiated adipocytes, nondifferentiated fibroblasts, and necrotic cells that have died during the differentiation process (debris). The nondifferentiated fibroblasts were separated from the adipocytes and dead cell debris by setting an elliptical gate in SSC and FSC (low scatter cells (LSC); Fig. 1, B and D). Differentiated adipocytes were resolved from residual fibroblasts by setting an elliptical gate in SSC and FSC (high scatter cells (HSC); Fig. 1B). A second elliptical gate in SSC and FL3 autofluorescence was used to separate adipocytes from necrotic cell debris (Fig. 1C). Cells expressing the HA-Glut4/GFP reporter construct (LG4) were separated from cells that do not express the reporter (NV) using a gate in SSC and FL1 (Fig. 2). Gated populations (adipocytes, fibroblasts, and debris) were sorted using a Beckman Coulter EPICS Elite ESP sorter. Cells were collected on glass slides and imaged by differential interference contrast microscopy ( Fig. 1E).
Anti-Glut4 Labeling in Fixed Cells-Differentiated control (NV) and HA-Glut4/GFP lentivirus-infected (LG4) 3T3-L1 cells were suspended by collagenase digestion in PBS plus 2% BSA. The cell suspension was brought to 4% gluteraldehyde (methanol-free EM grade; Polysciences) by the addition of 16% gluteraldahyde, mixed by gentle pipetting, and incubated at room temperature for 5 min. Fix was quenched by the addition of PBS plus 5% BSA (5 times the volume of the gluteraldahydefixed cell suspension). Cells were washed three times by centrifugation (1000 ϫ g for 3 min) and resuspended in PBS plus 2% BSA. Washed cells were permeabilized with IntraPrep (Beckman Coulter) according to the manufacturer's instructions. Permeabilized cells were labeled with polyclonal anti-Glut4 antibody (raised against the C-terminal 30 amino acids) for 1 h at room temperature, pelleted by centrifugation, and resuspended in PBS plus 2% BSA. Pelleting/resuspension was repeated three times to remove excess free anti-Glut4 antibody. Cells were then labeled with Alexa647-conjugated anti-rabbit IgG secondary. Control cells were labeled with Alexa647 antirabbit IgG antibody alone. After labeling, cells were analyzed by flow cytometry (supplemental Fig. 1A).
Purification and Labeling of Anti-HA-Anti-HA.11 antibody (Covance) was purified from ascites on a fast flow protein A column (GE Health Sciences) following manufacturer's recommended protocols. Briefly, ascites was diluted 10-fold in high salt loading buffer (3 M NaCl, 1.5 M glycine, pH 8.9) and injected onto the protein A column (pre-equilibrated in high salt loading buffer). The antibody-loaded protein A column was washed with 10 ml of high salt loading buffer. Bound anti-HA was eluted in 100 mM citric acid, pH 4.0. Eluted antibody was collected in 0.5-ml fractions into tubes containing 100 l of 150 mM Tris, pH 8.1. Protein-containing fractions (absorbance 280 nm) were pooled and exchanged into PBS using a 5-ml desalting column (GE Health Sciences). Purified anti-HA was then labeled for 2 h at room temperature with constant stirring with succinimidyl ester Alexa647 (Invitrogen) following the manufacturer's recommended protocols. Free dye was removed by exchange into PBS using a 5-ml desalting column (GE Health Sciences). Labeled antibody was stored in PBS at 4°C. Absorbance spectroscopy was used to determine both the concentration of labeled antibody ([P]) and the concentration of Alexa647 ([Alexa]) in the eluate as follows.
Anti-HA Labeling in Live Cells-3T3-L1 cells differentiated in 96-well plates were serum-starved for 2 h in 50 l of DMEM, 0.5% FBS and then incubated with or without insulin (100 nM or as indicated) for the indicated times. For surface labeling experiments, the cells were chilled to 4°C and then incubated at 4°C in medium containing 50 g/ml Alexa647-conjugated anti-HA. For uptake experiments, the starvation medium was replaced with prewarmed medium (with or without insulin) containing Alexa647-conjugated anti-HA antibody at the indicated concentrations. Cells were incubated in the continuous presence of anti-HA antibody at 37°C for the indicated times. To stop labeling, cells were placed on ice, antibody-containing medium was removed, and plates were washed two times with 100 l of ice-cold PBS. For the recovery experiments, the insulin-containing medium was removed, and cells were washed and incubated for 5 min total (two quick washes, two 2-min incubations) in low pH buffer (18), rinsed twice in medium, and then incubated in medium without insulin at 37°C for the indicated times (measured from time of insulin removal) prior to surface labeling or the antibody uptake experiment. After labeling, cells were suspended by collagenase digestion at 4°C and analyzed by flow cytometry. Collagenase digestion does not affect the anti-HA antibody bound to the cell surface, even after long incubation times (12).
Data Analysis-The flow cytometric data were analyzed using Flowjo software (Tree Star Inc.). For each population of cells (adipocytes, fibroblasts, all cells) within each sample the geometric mean of FL4 was determined for both the HA-Glut4/ GFP-positive and -negative (uninfected) cells. The mean fluorescence of the uninfected cells measures autofluorescence and nonspecific binding (BG). The background fluorescence varied with the time of incubation with anti-HA but was always less than 15% of total FL4 (supplemental Fig. 2D). The backgroundcorrected mean fluorescence intensity (MFI) was calculated as MFI ϭ FL4 Ϫ BG. The geometric mean of FL1 was determined for the HA-Glut4/GFP-positive cells. The FL1 MFI of uninfected cells (autofluorescence) was less than 5% of the FL1 MFI of infected cells (Fig. 2C); therefore, uncorrected FL1 MFIs were used in analysis. To correct for HA-Glut4/GFP expression levels, mean fluorescence ratios (MFRs) were calculated as MFR ϭ MFI FL4/MFI FL1. This parameter is similar to calculations made by Martin et al. (3) and Govers et al. (4). To correct for differences in the Alexa647 labeling efficiencies in different batches of labeled anti-HA antibody, the MFRs were standardized to the calculated maximum MFR (MFR max ) for cells treated with 100 nM insulin in each experiment. To calculate MFR max , the anti-HA uptake time course for insulin-stimulated cells was fit by a single exponential relaxation using Origin 7.5 software.
Replicates-The data points shown in the figures are the average MFR Ϯ S.E. from 2-10 separate experiments (as indicated). Curves are exponential fits of the averaged data. The calculated k obs values obtained from fits of the average MFRs were similar to the average k obs from fits of the individual experiments (i.e. analysis of eight uptake experiments, fit of average MFRs: basal k obs ϭ 0.012 Ϯ 0.003 min Ϫ1 and insulin k obs ϭ 0.036 Ϯ 0.003; average k obs from fits of eight individual experiments: basal k obs ϭ 0.013 Ϯ 0.003 min Ϫ1 and insulin k obs ϭ 0.040 Ϯ 0.005).

3T3-L1 Adipocyte Cultures Are a Heterogeneous Population
of Cells-Cultures of 3T3-L1 fibroblasts or differentiated adipocytes were analyzed by flow cytometry (Fig. 1). Dual parameter histograms of side scatter versus forward scatter and side scatter versus autofluorescence (FL3) were analyzed. Cultures of confluent fibroblasts contain a single population of cells with low scatter and low autofluorescence ( Fig. 1A) (data not shown). In contrast, the cultures of differentiated adipocytes are mixed populations of cells (Fig. 1, B and C). Two populations can be resolved based on scatter profiles: HSC and LSC. The high scatter objects can be further resolved into two populations, one with low autofluorescence and one with high autofluorescence (Fig. 1C). The high scatter, low autofluorescence cells are differentiated adipocytes. They contain lipid droplets (Fig. 1E) and express Glut4 (supplemental Fig. 1). The highly autofluorescent, high scatter objects are necrotic cell debris. They are very granular, with rough surfaces (Fig. 1E) and are highly autofluorescent in all detector channels (supplemental Fig. 1). The low scatter cells are indistinguishable in optical properties from confluent nondifferentiated 3T3-L1 cells (Fig.  1D). These are residual nondifferentiated fibroblasts. They contain no lipid droplets (Fig. 1E), express no Glut4 (supplemental Fig. 1), and spread and divide when replated on tissue culture dishes (data not shown). Propidium iodide staining showed that the adipocytes and fibroblasts are viable intact cells, whereas many of the necrotic cells are permeabilized (data not shown). Highly autofluorescent necrotic cells could also be detected by microscopy in intact monolayers of differentiated cells and were not solely the result of cell breakage during collagenase digestion and suspension of the cells (data not shown). The relative proportion of fibroblasts and necrotic debris to adipocytes was similar in samples within a single set of differentiated cells but varied from one batch of differentiated cells to the next. The number of residual fibroblasts tended to decrease with days post-initiation of differentiation, whereas the necrotic cells tended to increase. For analysis, elliptical gates in side scatter versus forward scatter were used to separate adipocytes from the residual fibroblasts. A second elliptical gate in side scatter versus FL3 was used to eliminate the highly autofluorescent necrotic cell debris from the adipocyte population.
3T3-L1 Adipocytes Are a Homogenous Population of Cells-An epitope-tagged HA-Glut4/GFP reporter construct was used to follow the traffic of Glut4 in differentiated adipocytes (8,12). This construct is similar to that used by Martin et al. (3). The reporter was stably expressed in 3T3-L1 fibroblasts using recombinant lentivirus. Differentiated lentivirus-transduced cells were analyzed by flow cytometry (Fig. 2). Cells expressing the reporter are best resolved from uninfected cells in dual parameter histograms of GFP fluorescence (FL1) versus side scatter. Diagonal gates were used to resolve HA-Glut4/GFPexpressing cells from noninfected cells as shown. Cells were infected with recombinant lentivirus at a titer sufficient to transduce ϳ80% of the cells. The uninfected cells serve as internal controls for nonspecific binding and autofluorescence in each sample. The FL1 fluorescence intensities of the infected cells (LG4) are 10 -100 times greater than the mean fluorescence intensity of the noninfected cells (NV).
Overexpression of Glut4 has been reported to inhibit its basal intracellular retention (19,20). To compare the level of expression of HA-Glut4/GFP with endogenous Glut4, infected and noninfected 3T3-L1 cells were differentiated, suspended by collagenase digestion, fixed, permeabilized, and labeled with an antibody that recognizes the carboxyl terminus of Glut4 (supplemental Fig. 1A). Infected and noninfected adipocytes (left) and fibroblasts (middle) were analyzed. Uninfected adipocytes express Glut4, whereas uninfected fibroblasts do not. Adipocytes expressing HA-Glut4/GFP have the same level of anti-Glut4 labeling as the uninfected adipocytes, indicating that the reporter is expressed at low levels relative to endogenous Glut4. Unlike the uninfected fibroblasts, the infected fibroblasts are positive for anti-Glut4 labeling, confirming that the anti-Glut4 antibody used in this analysis can recognize the epitope-tagged  Glut4/GFP fusion protein. There is a broad distribution of expression levels of endogenous Glut4 in adipocytes. There is also a broad distribution of HA-Glut4/GFP expression per cell.
Both show approximately a log-normal distribution of fluorescence intensities (21). 3 In infected adipocytes, there is no correlation between the anti-Glut4 labeling and GFP fluorescence, as would be expected if the HA-Glut4/GFP were expressed at low levels relative to endogenous Glut4. In contrast, there is a strong correlation between anti-Glut4 labeling and GFP fluorescence in infected fibroblasts, where the HA-Glut4/GFP accounts for all of the Glut4 expressed. These results are consistent with our previous experiments using recombinant adenovirus to express HA-Glut4/GFP in adipocytes (12).
To examine this in more detail, anti-HA binding was used to measure cell surface Glut4 content in adipocytes (Fig. 3). Differentiated lentivirus-transduced 3T3-L1 cells were incubated with or without insulin (100 nM, 30 min) and then labeled with Alexa647-conjugated anti-HA antibody at 4°C and analyzed by flow cytometry. If overexpression of the HA-Glut4/GFP reporter were interfering with intracellular retention of Glut4, then basal cells expressing more HA-Glut4/GFP would retain less of the total transporter, and the surface-to-total ratio would increase as the expression level of the reporter increased. This was not observed. The basal surface-to-total ratio (calculated on a cell-by-cell basis) tends to decrease as the amount of HA-Glut4/GFP expressed increases (Fig. 3A). Therefore, even at the highest level of expression, the HA-Glut4/GFP does not saturate the basal retention machinery. This analysis also shows that although the differentiated cell culture is a mixed population of cells, the gated adipocytes behave as a single population. All adipocytes exhibit low surface-to-total ratios of Glut4 in basal cells and elevated surface-to-total ratios after insulin stimulation (mean FL4/FL1 increases 10 -15-fold). In contrast, the transfected fibroblasts exhibit a much broader distribution of basal surface HA-Glut4/GFP, with many cells showing elevated levels (supplemental Fig. 1B, middle).
Anti-HA Antibody Uptake in 3T3-L1 Adipocytes-To measure the kinetics of Glut4 traffic, differentiated 3T3-L1 cells expressing HA-Glut4/GFP were incubated at 37°C in the continuous presence of Alexa647-conjugated anti-HA antibody for 10 -240 min and analyzed by flow cytometry (3, 4, 9 -11). The HA epitope is inserted into the first extracellular loop of the Glut4 reporter constructs. When anti-HA is added to the culture medium, cells expressing the reporter accumulate the antibody with time ( Fig. 4). Although Glut4 resides in multiple compartments within the cell, whole cell anti-HA uptake can be fit by a single exponential relaxation of the following form, where Y (t) represents the total cell-associated anti-HA at time t; Y min is the amount of antibody bound at the earliest time point (10 min in our experiments); Y max is the calculated maximum amount of antibody bound at t ϭ ∞; and k is the relaxation rate constant. 4 If antibody binding is saturating, then Y max is a measure of the size of the total cycling pool of Glut4. C.
LG4 NV were calculated for the gated adipocytes in each sample. Plots of MFI versus time for each antibody concentration were fit by single exponential relaxations (Fig. 4A). The S.E. values for fit parameters were less than 10% of the predicted parameter val-ues, the fit residuals were randomly distributed, and the R 2 for the fits were Ͼ0.98 in all cases ( Fig. 4A and data not shown), indicating that the single exponential treatment is appropriate. The data could not be resolved into multiple relaxations by fitting with additional exponentials. MFI max and the observed relaxation rate constant (k obs ) were calculated for each antibody concentration from these fits.
To calculate the EC 50 of antibody binding, a plot of MFI max versus concentration of antibody (Fig. 4B) was fit by an equation for a hyperbola of the following form, where B [Ab] is the calculated maximal amount of antibody bound at each antibody concentration ([Ab]; MFI max ), and B max is the calculated maximum amount of antibody bound at [Ab] ϭ ∞. Again, the S.E. values for the fit parameters were less than 10% of the predicted parameter values, the residuals were randomly distributed, and the R 2 value was Ͼ0.99, indicating an appropriate fit. The data could not be fit by multiple hyperbolic binding isotherms, indicating that a single antibody/antigen interaction controlled the amount of anti-HA associated with the cells. The EC 50 for antibody binding was 3 g/ml (20 nM). At 50 g/ml, ϳ95% of the cycling HA-Glut4/GFP is bound by the antibody when steady state is reached. 5 In these experiments, anti-HA accumulation is a function of both the rate of antibody binding and the rate of exchange of intracellular and cell surface Glut4. Both Martin et al.  the concentration of incubating antibody (the observed rate constant for antibody binding to antigen at the cell surface will increase linearly with [Ab]; k obs ϭ k on [Ab] ϩ k off , based on the second order kinetics of antibody antigen binding). However, no data were presented to verify that this assumption can be made at 50 g/ml anti-HA antibody. To demonstrate the effect of antibody concentration on the uptake kinetics, the amount of antibody associated with the cells was expressed as a fraction of the calculated maximal amount of antibody that will bind at steady state at each antibody concentration (% MFI max ) and plotted versus time (Fig. 4C). This analysis clearly shows that the rate of accumulation depends on the concentration of antibody in solution. To look at this in more detail, a plot of k obs versus antibody concentration was analyzed (Fig. 4D). Between 1.25 and 40 g/ml, k obs increases approximately linearly with respect to the concentration of incubating antibody. Thus, at 40 g/ml antibody or less, accumulation is rate-limited by antibody binding. In contrast, uptake at 80 g/ml is relatively insensitive to the concentration of incubating antibody. The rate of uptake at 50 g/ml lies between the values at 40 and 80 g/ml.
A second explicit assumption made by both Martin et al.
(3) and Govers et al. (4) is that antibody binding is irreversible. To test this directly, the stability of cell-associated anti-HA in cells pulse-labeled with the antibody was measured (supplemental Fig. 2A). The amount of antibody associated with labeled cells does not decrease after cells were incubated for 120 min at 37°C in medium without antibody. Thus, antibody accumulation is apparently irreversible on the time scale analyzed. As a final control, it was verified that antibody accumulation requires ATP synthesis in live cells and is inhibited by the metabolic poisons 2-deoxyglucose and azide (supplemental Fig. 2B).
Cell Surface Glut4 Levels Are Regulated by Static Retention in 3T3-L1 Adipocytes-Anti-HA uptake was measured in basal and insulin-stimulated cultures of 3T3-L1 cells (Fig.  5). MFIs of both cell-associated Alexa647 and GFP were calculated for each sample. To correct for cellto-cell variation in HA-Glut4/GFP expression, MFRs were calculated (MFR ϭ MFI Alexa647/MFI GFP). This is similar to calculations made by Martin et al. (3) and Govers et al. (4). Plots of MFR versus time were fit by single exponential relaxations. This analysis yields both the size of the actively cycling pool of HA-Glut4/GFP (MFR max ) and the relaxation rate constant (k obs ). Insulin increased MFR max by 5-fold and increased k obs by 4-fold (from 0.01 to 0.039 min Ϫ1 ). The S.E. values for each parameter were less than 10% of the predicted values, and the residuals were small and randomly distributed (data not shown). Therefore, as observed by Govers et al. (4), our analysis indicates that Glut4 is regulated by static retention.
In the dynamic equilibrium model, all of the Glut4 is actively cycling under both basal and insulin-stimulated conditions. Martin et al. (3) reported that the basal maximum binding was 90% of the insulin maximal binding. To verify that the static retention model most accurately describes our data, our free fit of the basal data was compared with a constrained fit in which basal MFR max was set to 90% of the insulin MFR max (Fig. 5B). Doing this dramatically increased the residuals and the S.E. values for the fit parameters and decreased the R 2 value of the fit. In addition, the most accurate fit of our data with a dynamic equilibrium model forces the observed rate constant of anti-HA accumulation to 0.001 (t1 ⁄ 2 of ϳ600 min; Fig. 5C), significantly slower than that observed by Martin et al. (3) (k ϭ 0.007, t1 ⁄ 2 ϭ 100 min. In our experiments, uptake was analyzed only out to 240 min, whereas in the work of Martin et al. (3), uptake was analyzed out to 600 min. However, even when analyzed after 10 h of uptake, the basal cells in our assay accumulated less than 30% of the anti-HA antibody that insulin-stimulated cells accumulated (Fig. 5D). In contrast, in the work of Martin et al. (3), the basal cells had accumulated 90% of the insulin-stimulated anti-HA antibody at this time. Govers et al. (4) reported that insulin stimulates the quantal release of Glut4 from statically retained compartments. We also observe a dose-dependent increase in both cell surface Glut4 and the size of the actively cycling pool (Fig. 6, A and B). There is a 10 -15-fold increase in cell surface labeling and a 5-fold increase in pool size with maximal insulin stimulation (100 nM) compared with basal cells. There is also a dose-dependent increase in the percentage of the total cycling pool that is found at the cell surface. In basal cells, 10 -15% of the cycling pool is found at the plasma membrane (2-3% of the maximum insulin-stimulated pool), whereas in fully insulin-stimulated cells, 30 -35% of the cycling pool is at the cell surface (data not shown). This shift reflects the 3-4-fold increase in the trafficking relaxation rate constant, k obs . To determine the dose dependence of k obs , the kinetics of antibody uptake were measured at 0, 1, 10, and 100 nM insulin (Fig. 6, C-E). Plots of MFR versus time were fit by single exponential relaxations. As uptake experiments were done using cells that were replated on glass coverslips after completing an abbreviated differentiation protocol (2 days in differentiation medium, 2 days in insulin medium). The cells were analyzed 1 or 2 days after replating. It is common practice to replate 3T3-L1 cells after differentiation for microscopy or electroporation experiments. However, the effect of this widely used technique on Glut4 traffic has not been reported.
To test the effect of replating on anti-HA uptake in 3T3-L1 adipocytes, differentiated cells were suspended by collagenase digestion, replated 1 day after the removal of insulin medium, and then cultured for an additional 2 days (Fig. 7). This treatment is similar to that used by Martin et al. (3). Parallel samples of control differentiated cells were maintained in confluent culture. Replating increased the size of the basal cycling pool of Glut4 to 80%, which is very similar to the 90% reported by Martin et al. In contrast, the relaxation rate constant for Glut4 cycling was not significantly affected (k obs ϭ 0.014 min Ϫ1 in replated cells versus 0.012 min Ϫ1 in confluent adipocytes). These data indicate that the differences observed in trafficking  Glut4 Is Recycled to a Static Pool after Insulin Withdrawal-Under physiological conditions, insulin levels fluctuate as glucose levels fluctuate. In response to this fluctuation, Glut4 is rapidly inserted into the plasma membrane after insulin stimulation and is rapidly cleared from the plasma membrane when insulin is withdrawn. The Glut4 is recycled and is reused for subsequent stimulations with insulin (6). The increase in surface Glut4 to a new steady state level after insulin stimulation follows a single exponential relaxation, with a t1 ⁄ 2 of 6.5 min (Fig.  8A). The surface Glut4 returns to basal levels after insulin withdrawal, with a t1 ⁄ 2 of 20 min. Upon restimulation, the kinetics of transition from the basal to the insulin-stimulated state is very similar to the preinsulin kinetics (t1 ⁄ 2 ϭ 5 min). To determine if Glut4 is recycled into the static pool, the size of the actively cycling pool of Glut4 was examined after insulin withdrawal (Fig. 8B). The total amount of Glut4 cycling through the plasma membrane in 1 h decreased rapidly upon removal of insulin. The kinetics of recovery were best fit by two concurrent expo-nential relaxations, one with a t1 ⁄ 2 ϭ 5 min, and a second that relaxed much more slowly. To better characterize Glut4 trafficking after recovery, the full kinetics of Glut4 uptake were examined 3 h after insulin withdrawal (Fig. 8C). Glut4 is regulated by static retention 3 h after recovery from insulin stimulation. However, the total amount of cycling Glut4 in recovered cells remained elevated (0.48 versus 0.26 in the control cells). Interestingly, the k obs was also decreased by 2-fold in the recovered cells (0.0077 versus 0.015), so the total amount of Glut4 cycling through the plasma membrane in 1 h was similar in the control and recovered cells. In contrast, 12 h after removal of insulin, the kinetics of anti-HA uptake in recovered cells were identical to uptake in basal control cells (Fig. 8D).

DISCUSSION
We developed a flow cytometric assay to study Glut4 traffic in 3T3-L1 adipocytes. This was a modification of the anti-HA antibody uptake assays described by Govers et al. (3) and Martin et al. (4). In the flow assay, large numbers of cells are easily analyzed on a cell-by-cell basis. Thus, flow cytometry combines advantages of the whole plate assay described by Govers et al. (4) and the microscopy assay described by Martin et al. (3). Additional advantages of flow cytometry include sensitivity of detection, accurate noise correction, and ease of analyzing the large numbers of samples required for time course and dose response experiments. This method could easily be applied to other cycling surface antigens or receptors (13)(14)(15).
Using this assay, we found that cell surface Glut4 is regulated by static retention in confluent, quiescent 3T3-L1 adipocytes, in agreement with the results of Govers et al. (4). In contrast, Martin et al. (3) concluded that Glut4 was regulated solely by dynamic equilibrium. To understand the difference between our results and those of Martin et al. (3), we compared the kinetics of Glut4 trafficking in confluent differentiated cells and cells replated after differentiation. We found that Glut4 traffics differently under these two culture conditions. In confluent cultures, 80% of the Glut4 in basal 3T3-L1 adipocytes is retained in noncycling compartments (Fig. 5). This is most consistent with static retention, where GSVs behave like regulated secretory vesicles (4). In contrast, 80% of the Glut4 is cycling under basal conditions in replated cells (Fig. 7), consistent with dynamic exchange (3). We conclude that differences in the cell Lines, linear or exponential fits of the data (the recovery data were best fit by two concurrent exponential relaxations; all others were single exponential fits). Data points are average MFR Ϯ S.E. from two (A) or three separate experiments.
culture conditions can largely account for the differences in trafficking behavior observed by Martin et al. (3) and Govers et al. (4). In addition to increasing the size of the actively cycling pool, insulin also increased the observed relaxation rate constant (k obs ) 3-fold, from 0.012 to 0.036 min Ϫ1 . This is consistent with previous reports. Govers et al. (4) found that insulin increased the rate constant of Glut4 exocytosis from 0.014 to 0.03 min Ϫ1 . Martin et al. (3) found that insulin increased k ex from 0.007 to 0.05-0.07 min Ϫ1 (3). Using photoaffinity probes on primary adipocytes, Sato et al. (6) found that insulin increased k ex from 0.012 to 0.032 min Ϫ1 , whereas Jhun et al. (7) found that insulin increased k ex from 0.024 to 0.078 min Ϫ1 . Insulin caused a modest (2-7-fold) increase in k ex in all studies.
The effect of insulin on endocytosis can also be estimated from the uptake data, if the partitioning coefficient, P, can be accurately estimated (P ϭ [p]/[i] ϭ k ex /k en , where [p] is the proportion of cycling Glut4 in the plasma membrane and [i] is the proportion in intracellular compartments). At 50 g/ml anti-HA, the amount of antibody associated with cells after 10 min of uptake (MFR 10 min ) is approximately equal to [p] (within 10%; calculated using the equations in supplemental Fig. 3B). MFR max gives the size of the total cycling pool, and [i] ϭ MFR max Ϫ MFR 10 min . In our experiments, basal P ϭ 0.12/0.88, and insulinstimulated P ϭ 0.3/0.7. Using these values and our calculated k ex , insulin has no effect on k en (0.088 min Ϫ1 basal versus 0.084 min Ϫ1 insulin; k en ϭ k ex /P). It is important to note that the partitioning coefficient must be calculated as a percentage of the total cycling pool of Glut4 under each condition, not the total amount of cellular Glut4. Inclusion of noncycling Glut4 (i.e. Glut4 in noncycling GSVs or in the biosynthetic or degradative pathways) will lead to an overestimate of [i], an underestimate of P, and an overestimate of k en . In our experiments, only 2.6% of the total cycling Glut4 (measured as MFR max at saturating concentrations of insulin) is at the plasma membrane in basal cells. However, [p] ϭ 12% in basal cells, because the basal cycling pool is only 22% of the total insulin-responsive cycling Glut4. After maximal insulin stimulation, 30% of the total cycling pool is at the plasma membrane. The net effect is that insulin increases cell surface Glut4 ϳ12-fold, through a 3-fold increase in k ex and a 4-fold increase in pool size.
In the work of both Martin et al. (3) and Govers et al. (4), the observed relaxation rate constants calculated from the single exponential fits were assumed to be k ex . This was based on the explicit assumption that binding at the cell surface was instantaneous and irreversible. If these assumptions are correct, then antibody labeling would be independent of antibody concentration [Ab], since the rate of binding increases linearly with increasing [Ab]. To test this, we measured uptake at [Ab], ranging from 1.25 to 80 g/ml (Fig. 4). Although we found that 50 g/ml is nearly saturating with respect to binding (it is 17 times the EC 50 of binding), it is not saturating with respect to uptake. Under these conditions, the rate of antibody uptake will be a function of both binding and traffic, and k obs is not simply k ex .
Both Martin et al. (3) and Govers et al. (4) interpret their data based on a "two-pool" model. In this model, the cycling Glut4 exchanges between the plasma membrane and intracellular compartments via exocytosis and endocytosis, with the rate constants k ex and k en . Intracellular Glut4 is treated as a single kinetically homogeneous pool (in the work of Govers et al. (4), there is also a separate static pool). Others have used similar models (6,7). To better understand the relationship between k obs and k ex , we modified the model to include antibody binding (supplemental Fig. 3). This assay is a relaxation experiment. At t ϭ 0, no antibody is bound. When the system reaches equilibrium, antibody is bound to 95% of the HA epitopes both at the surface and inside of the cell at 50 g/ml antibody. There are three reversible processes occurring: 1) intracellular Glut4 (i) is exchanging with plasma membrane Glut4 (p), with the rate constants k ex and k en ; 2) antibody is binding to and being released from HA-Glut4 at the plasma membrane, with the rate constants k on and k off ; 3) the labeled HA-Glut4 at the plasma membrane (p*) is exchanging with the labeled intracellular pool (i*), with the rate constants k ex and k en (supplemental Fig. 3A).
Antibody uptake in this model is described by a series of ordinary differential equations (supplemental Fig. 3B). Numerical methods were used to simulate uptake over a range of antibody concentrations, varying k ex , k on , and the partition coefficient, P (k en ϭ k ex /P; k off ϭ k on ϫ K d ). We fit a single exponential relaxation to the simulated data and assessed k obs from this analysis. We found that k obs was a function of k ex , k en , k on , and [Ab]. In contrast, varying k off had little effect on k obs but affected the maximal binding (through effects on K d ). When k ex is Ͼ Ͼk en , P is high, and most of the Glut4 is at the cell surface. Under these conditions, the rate of antibody accumulation is simply the rate of binding (k obs ϭ k on [Ab] ϩ k off ). When k en is Ͼ Ͼk ex , P is low, and most of the Glut4 is intracellular. Under these conditions, k obs is approximately k ex at saturating concentrations of antibody. When P is set to the values calculated from our data (basal P ϭ 0.12/0.88; insulin P ϭ 0.3/0.7), k obs is a function of both binding and trafficking and does not simply equal k ex (supplemental Fig. 3D). However, the k obs versus [Ab] curve flattens out once k obs is Նk ex (supplemental Fig. 3F). At [Ab] close to this break point (40 -80 g/ml with this antibody), the rate of binding is close to the rate of exocytosis, and k obs is a good approximation for k ex (supplemental Fig. 3E).
What is "k ex "? It is the apparent rate constant for delivery of HA-Glut4/GFP to a compartment where it can be labeled (at the plasma membrane, unoccluded). This is not necessarily the rate constant for fusion of GSVs to the plasma membrane, however. Although the cycling pool of Glut4 behaves as a kinetically homogenous population (uptake can be fit by a single exponential; two-pool model), this does not mean that there is only a single intracellular compartment through which Glut4 traffics. In an elegant modeling paper, Holman et al. (22) (also see Ref. 5) showed that a simple two-pool model in which endocytosis is essentially the reverse of exocytosis is insufficient to describe Glut4 trafficking. In the two-pool model, the rate constant for the transition from the basal to the insulin-stimulated state will be less than or equal to the rate constant for steady state uptake after insulin stimulation. It is not. The t1 ⁄ 2 for the basal to insulin-stimulated transition is 5-6 min (Fig. 8), whereas the t1 ⁄ 2 for uptake in insulin-stimulated cells is 17-19 min (Fig. 5). Holman et al. (22) showed that a "three-pool" model, in which Glut4 traffics through two sequential intracellular pools, the endosomes and a "sequestered" pool, could account for the discrep-ancy between the t1 ⁄ 2 values (Fig. 9). In a cycle of nonreversible steps (i.e. clathrin-coated pit/vesicle to transport vesicle to early endosome to sorting endosome to GSV to PM), flux through the pathway will be dominated by a single step if that step is significantly slower than all of the others. Under these conditions, k ex is approximately the rate constant for the slow step, and the population will appear to behave as a single homogenous pool. The population will also follow an apparent single exponential if the rate constants for the sequential steps are similar. However, in this case, k ex is a function of all of the rate constants. In Holman's three-pool model, Glut4 accumulates in GSVs in basal cells, because the rate constant for fusion of GSVs to the plasma membrane (k ex ) is much slower than the rate constant for transfer of Glut4 from endosomes to GSVs (k seq ). The basal-to-insulin transition is fast due to a burst of Glut4 release caused by a large (Ͼ100-fold) increase in k ex . In contrast, steady state uptake is relatively slow, because it is a function of both k seq and k ex .
Although the three-pool model describes the behavior of the static pool of Glut4, this model does not fully account for the observed kinetic data. It predicts a very slow k obs under basal conditions (0.001 min Ϫ1 ), which increases Ͼ80-fold after insulin stimulation. However, in all of the kinetic studies reported, there is a pool of Glut4 in basal cells that is cycling much faster than this, with a k obs ϭ 0.007-0.024 min Ϫ1 . This k obs increases only 2-7-fold with insulin. The three-pool model also predicts a slow labeling of the Glut4 in the GSV pool in basal cells, which we do not observe; it does not predict the fast filling of a small pool of Glut4 (20 -25%) that is observed (Figs. 5 and 9B). Finally, the threepool model cannot account for the quantal release of Glut4 at submaximal insulin concentrations (Fig. 6) (4). In contrast, static retention can account for the rapid transition kinetics with a k obs that increases only 3-fold (from 0.012 to 0.036 min Ϫ1 ; Fig. 9C). It also accounts for the rapid labeling of a small pool in basal cells and the quantal release.
Static retention is a modification of the three-pool model (Fig. 9E). Rather than a sequential pathway where cycling Glut4 must traverse the sequestered (GSV) pool before exit, there is a bifurcation in the pathway. Cycling Glut4 in the endosomal pathway can either traffic to the plasma membrane (with rate constant k ex ) or to the GSV pool (with rate constant k seq ). At steady state, however, Glut4 does not traffic into or out of the static GSV pool, under either basal or insulin-stimulated conditions. Therefore, in the static retention model, there are two distinct pools of Glut4 in both basal and insulin-stimulated cells: one that is rapidly cycling and one that is static. Insulin controls the total amount of Glut4 in the actively cycling pool through , k ex set to 1 ϫ 10 Ϫ4 s Ϫ1 ( 1 ⁄2 basal k ex ), and k en set to 14.7 ϫ 10 Ϫ4 s Ϫ1 (basal k en ). E, comparison of the three-pool model versus static retention. In both models, Glut4 traverses three pools: the PM, endosomes (end), and GSVs. Glut4 is trafficked to the PM with a rate constant k ex , Glut4 is retrieved from the PM with the rate constant k en , and Glut4 is recycled from the endosomes to GSVs by sequestration (k seq ). However, in the three-pool model, k ex is the rate constant for the fusion of GSVs to the PM, Glut4 traffics to the cell surface only through GSVs, and k seq is fast and does not change. In static retention, Glut4 traffics to the PM through the endosomal cycling pool with the rate constant k ex , insulin transiently stimulates quantal release of Glut4 from GSVs into the cycling pool, and sequestration is transiently elevated in response to insulin withdrawal (k seq is very slow in both basal and insulin-stimulated cells).