The Adaptor Protein β-Arrestin2 Enhances Endocytosis of the Low Density Lipoprotein Receptor*

Endocytosis of the low density lipoprotein (LDL) receptor (LDLR) in coated pits employs the clathrin adaptor protein ARH. Similarly, agonist-dependent endocytosis of heptahelical receptors in coated pits employs the clathrin adaptor β-arrestin proteins. In mice fed a high fat diet, we found that homozygous deficiency of β-arrestin2 increased total and LDL plus intermediate-density lipoprotein cholesterol levels by 23 and 53%, respectively (p < 0.05), but had no effect on high density lipoprotein cholesterol levels. We therefore tested whether β-arrestins could affect the constitutive endocytosis of the LDLR. When overexpressed in cells, β-arrestin1 and β-arrestin2 each associated with the LDLR, as judged by co-immunoprecipitation, and augmented LDLR endocytosis by ∼70%, as judged by uptake of fluorescent LDL. However, physiologic expression levels of only β-arrestin2, and not β-arrestin1, enhanced endogenous LDLR endocytosis (by 65%) in stably transfected β-arrestin1/β-arrestin2 double-knockout mouse embryonic fibroblasts (MEFs). Concordantly, when RNA interference was used to suppress expression of β-arrestin2, but not β-arrestin1, LDLR endocytosis was reduced. Moreover, β-arrestin2–/– MEFs demonstrated LDLR endocytosis that was 50% less than cognate wild type MEFs. In fusion protein pull-down assays, β-arrestin2 bound to the LDLR cytoplasmic tail stoichiometrically, and binding was abolished by mutation of LDLR Tyr807 to Ala. Mutation of LDLR cytoplasmic tail Ser833 to Asp enhanced both the affinity of LDLR fusion protein binding to β-arrestin2, and the efficiency of LDLR endocytosis in cells expressing β-arrestin2 physiologically. We conclude that β-arrestin2 can bind to and enhance endocytosis of the LDLR, both in vitro and in vivo, and may thereby influence lipoprotein metabolism.

In human beings, supraphysiologic levels of plasma low density lipoprotein (LDL) 1 cholesterol are associated with virtually all cases of atherosclerosis (1). Approximately 75% of plasma LDL clearance occurs through endocytosis of the low density lipoprotein receptor (LDLR), predominately in the liver (2). The well characterized endocytosis of the LDLR through clathrincoated pits depends upon association of the 50-amino acid cytoplasmic tail of the LDLR with components of the cellular endocytic machinery. Although mutagenesis studies of the LDLR have delineated cytoplasmic tail domain residues necessary for LDLR endocytosis (3,4), proteins with which the LDLR cytoplasmic tail interacts have only recently been identified. The clathrin heavy chain terminal domain itself can interact with peptides from the LDLR cytoplasmic tail but with relatively low affinity (5). The N-terminal domain of the autosomal recessive hypercholesterolemia (ARH) protein also binds to the LDLR cytoplasmic domain, whereas the C-terminal domain of ARH binds to both clathrin and AP-2 (6). Thus, ARH appears to link the LDLR to the endocytic machinery, in a process that seems to be required for endocytosis of the LDLR in hepatocytes and lymphocytes (6,7). Whether other adaptor-type proteins may be involved in LDLR endocytosis remains to be determined.
Candidate clathrin adaptor proteins for the LDLR could include the ␤-arrestins, which play important roles in the endocytosis of heptahelical G protein-coupled receptors. ␤-Arrestin1 and ␤-arrestin2 were initially characterized as ubiquitously expressed proteins involved in heptahelical G protein-coupled receptor desensitization (8). With their N-terminal domains (9), the ␤-arrestins bind stoichiometrically to agonist-activated heptahelical receptors, and they do so with greater affinity after the receptors have been phosphorylated by G proteincoupled receptor kinases (10). The binding of a ␤-arrestin to the receptor inhibits receptor/heterotrimeric G protein interaction (8) but can also initiate signaling via c-Src and other kinases (11). To mediate G protein-coupled receptor internalization, ␤-arrestins link receptors to clathrin (12) and AP-2 (13), both of which are bound by the ␤-arrestin C-terminal domain. Recent data generated in ␤-arrestin-deficient cells suggest that particular receptors can interact preferentially with specific ␤-arrestin isoforms (14).
Endocytosis of G protein-coupled receptors in clathrin-coated pits is agonist-dependent (15), like the binding of ␤-arrestins to G protein-coupled receptors (16). In contrast, coated pit-mediated endocytosis of the LDLR is constitutive in fibroblasts (17). However, because ARH serves a clathrin adaptor function for the LDLR much like that served by the ␤-arrestins for heptahelical receptors (6), we tested the hypothesis that ␤-arrestins could act as LDLR adaptor proteins, and thereby enhance endocytosis of the LDLR.

MATERIALS AND METHODS
Lipoprotein Metabolism in Mice-All animal care conformed to the "Guide for the Care and Use of Laboratory Animals" issued by the National Institutes of Health. The generation of ␤-arrestin2 Ϫ/Ϫ mice has been described previously (18). Derived from matings between ␤-arrestin2 ϩ/Ϫ founder mice, the ␤-arrestin2 Ϫ/Ϫ and ϩ/ϩ mice used in this study (n ϭ 30) were littermates and were hybrids of the C57Bl/6J and 129/SvJ genetic backgrounds. Serum cholesterol measurements were performed on ϳ200-l blood samples, obtained from tail cuts performed under methoxyflurane anesthesia. Blood samples were first obtained after 4 weeks on a Paigen diet (by weight: 15% fat, 1.25% cholesterol, and 0.5% cholic acid (19), Dyets, Inc., #615038). After initial blood sampling, mice were fed a normal, low fat diet (Purina Breeder Chow), and blood was sampled again after 8 weeks. Thus, our serum lipid studies constitute a dietary crossover study. The mice were supplied with only water for 12 h prior to blood sampling (19).
Blood was allowed to clot for 2 h at room temperature. The thrombus was pelleted at 4°C, and serum was removed. All mice were bled on the same day, and each batch of 30 samples (high and low fat diet) was assayed simultaneously. Lipid analyses were performed by the Duke University Medical Center Endomet Laboratory, which is standardized for human lipoprotein analyses by the Centers for Disease Control and Prevention. LDL cholesterol was measured with the Liquid Select TM LDL Direct Assay (Equal Diagnostics), and HDL cholesterol was measured with the Roche Applied Science direct HDL-cholesterol method. In pilot studies with mouse plasma divided into aliquots, we found that lipoprotein cholesterol results from these assays correlated very closely with those obtained by fast performance liquid chromatography (FPLC) fractionation using a Superose 6 HR10/30 column (Amersham Biosciences) and the Infinity TM Cholesterol Reagent kit (Sigma); however, LDL cholesterol determined by the LDL Direct assay corresponded to LDL plus IDL cholesterol determined by FPLC fractionation. Plasma lipoprotein fractionation by FPLC was performed on 100-l samples of mouse plasma as described previously (20).
Plasmid Constructs-The LDLR cDNA was obtained from the American Type Culture Collection (21). This construct was epitope-tagged at its N terminus with the FLAG TM octapeptide, as described previously (22). From the first amino acid of the mature LDLR (Ala) onward (21), the remainder of the LDLR sequence was left intact. The resulting construct was subcloned into pcDNA I (Invitrogen). This FLAG-LDLR construct behaved identically to the native LDLR in LDL uptake experiments (data not shown), and is referred to as "LDLR" throughout the text. To create S833D and S833A mutations in the full-length LDLR, we employed two sets of complementary 46-mer oligonucleotides, which included nucleotides 2545-2583 of the human LDLR sequence (containing a 5Ј-XhoI site) (21) as well as 3Ј EcoRI and XbaI sites. In these oligonucleotides, nucleotides 2560 -2562 were mutated to either GCC (Ala) or GAC (Asp). After annealing, these oligonucleotides were ligated into a XhoI/XbaI-cut LDLR/pcDNA I construct.
A GST fusion protein encompassing the 50 amino acids of the LDLR cytoplasmic tail (LDLRct) was created by cassette PCR of the human LDLR, and subcloning the PCR fragment into pGEX2T (Amersham Biosciences) with BamHI (5Ј) and EcoRI (3Ј). To create S833D, Y807A, and S833D/Y807A mutations in the GST/LDLRct construct, we used subcloning and cassette PCR similar to that for the wild type construct, but with oligodeoxynucleotides encoding the indicated mutations. Fidelity of PCR-amplified and synthetic oligonucleotide-generated DNA sequences was verified by dideoxy sequencing. Plasmids encoding native rat ␤-arrestin1 and rat ␤-arrestin2 (23), as well as a K44A ("dominant negative") rat dynamin I mutant (24) have been described previously. FLAG-tagged constructs encoding rat ␤-arrestin1 and ␤-arrestin2 (24) were subcloned into pcDNA3.1/Hygro (Invitrogen).
Immunoprecipitations-The day after transfection, ldlA cells were seeded into 100-mm dishes in Ham's F12 supplemented with 10% lipoprotein-deficient newborn calf serum (LPDS medium). The following day, LPDS medium was replaced with fresh medium containing 5% LPDS lacking or containing LDL at 200 g of protein/ml, and incubated for 10 min at 37°C. Cells were then washed twice with PBS, and proteins were cross-linked with the reducible cross-linking agent dithiobis(succinimidylpropionate) (Pierce), as described (22). Cells were then scraped into ice-cold 500 mM NaCl/50 mM Tris-Cl, pH 7.4 (25°C) with protease inhibitors (buffer A) (23), and then disrupted by using a tightfitting Dounce homogenizer. Membranes from disrupted cells were pelleted (40,000 ϫ g, 30 min, 4°C), and washed in buffer A. Solubilized membranes were then processed for immunoprecipitation with M2 anti-FLAG TM IgG, as described (22). Immune complexes were heated in Laemmli buffer for 30 min at 37°C, to reduce intermolecular cross-links before SDS-PAGE and immunoblotting, which were performed as described previously (22).
Immunoblotting-To assay endogenous ␤-arrestin expression in MEFs, cells were lysed in 10 mM Tris-Cl/2 mM EDTA (pH 8.0) with protease inhibitors (buffer B), and the membrane fraction was pelleted at 20,000 ϫ g for 15 min at 4°C. Liver samples were obtained from C57Bl/6J mice sacrificed under pentobarbital anesthesia. Fresh liver was perfused with Ringer's Lactate, then minced and homogenized with a small Polytron TM (Brinkman) in buffer B. Insoluble debris was pelleted at 40,000 ϫ g for 30 min at 4°C. As throughout this report, protein concentrations of liver and cell supernatant fractions were assayed by a modified Lowry method, with IgG as the standard (23). Forty g of each specimen was subjected to SDS-PAGE and immunoblotting, as described (23).
LDL Uptake Assays-Either 16 h after transfection and 24 h before assay (ldlA cells, 293 cells), or 48 h before assay (MEFs and siRNAtreated 293 cells), cells were split and seeded at equal densities (ϳ50% confluence, 2.7-4.7 ϫ 10 4 cells/cm 2 ) into 6-well dishes, in growth medium containing 10% LPDS (in lieu of fetal bovine serum) (27). For LDLR-transfected 293 cells, 10% LPDS growth medium was supplemented with 10 g of cholesterol and 0.1 g of 25-hydroxycholesterol per milliliter, to down-regulate endogenous LDLRs (3). For siRNAtransfected 293 cells, 10% LPDS growth medium was supplemented with 1 M lovastatin. On the day of assay, separate aliquots of cells were processed for cell surface LDLR assay (see below) at 4°C, and LDL uptake (endocytosis) assays at 37°C. LPDS growth medium was replaced with cognate medium containing 5% LPDS and LDL labeled with the fluorescent dye 1,1Ј-dioctadecyl-3,3,3Ј,3Ј-tetramethylindocarbocyanine (DiI-LDL), at a concentration of 3-5 g of protein/ml ("labeling medium," for "total uptake"). Labeling medium containing unlabeled LDL at 500 g of protein/ml was used to assess nonspecific uptake of the DiI-LDL. Cells were incubated at 37°C in 5% CO 2 for 5 h (27). LDL uptake assays were then terminated by trypsinizing cells, washing twice with PBS, and fixing with 3.6% formaldehyde in PBS. Cellular uptake of LDL was assessed by flow cytometry for cellular DiI fluorescence (28). Specific ([total] Ϫ [nonspecific]) uptake of DiI-LDL was divided by the relative cell surface LDLR number (see below), to obtain the Internalization Index for LDLRs in each cell line (3). The values for internalization index were then normalized to that obtained for the control cells within each experiment, to facilitate averaging results across independent experiments. For DiI-LDL uptake, the intra-assay coefficient of variation was Ͻ5%, and mean nonspecific uptake constituted 25% of mean total uptake.
Cell Surface LDLR Assessment-For cells transfected with the Nterminal FLAG TM -tagged LDLR construct, cell surface LDLRs were quantitated by cell surface immunofluorescence and flow cytometry, as described (22). For MEFs, which expressed only endogenous LDLRs, cell surface receptor number was assessed at 4°C by ligand binding, with DiI-LDL (27). Cells were first detached in PBS/EDTA (Versene, Invitrogen), pelleted, and washed in ice-cold minimal essential medium. Next, cells were resuspended in 200 l of ice-cold DiI-LDL labeling medium lacking ("total binding") or containing unlabeled LDL at 500 g of protein/ml ("nonspecific binding"), and incubated at 4°C for 3 h.
Subsequently, cells were pelleted and washed with 1.5 ml of ice-cold HEPES-buffered saline (with 2.5 mM CaCl 2 ), then fixed in this solution with 3.6% formaldehyde. Specific cell surface DiI-LDL binding was then assessed as described above and was used to estimate the relative cell surface LDLR number within each experiment. The intra-assay coefficient of variation was Ͻ7%, and mean nonspecific binding constituted 35-60% of mean total binding. Within a given experiment, cells were used only if cell surface LDLR density was within 30% (transfected cells) to 50% (MEFs) of control cells.
LDL and LPDS Preparation-LDL (1.019 Ͻ d Ͻ 1.055) from normal human plasma and newborn calf lipoprotein-deficient serum (d Ͼ 1.21 g/ml) were prepared by vertical spin density gradient ultracentrifugation, as described (27,29). LDL purity was assessed by Coomassie Blue staining of SDS 3-20% gradient gel electrophoresis. DiI-LDL was either purchased from Molecular Probes, Inc., or prepared as described (28).
Angiotensin II Receptor Internalization-Fibroblasts in 35-mm wells were incubated with serum-free medium (Dulbecco's modified Eagle's medium/1% bovine serum albumin/20 mM Hepes, pH 7.4) containing either vehicle ("naïve") or 100 nM angiotensin II for 30 min (37°C), and then washed with 40 mM sodium acetate/150 mM NaCl, pH 5.0 (10 min, 25°C) to remove cell surface-bound angiotensin II (30). After three washes with PBS, cells were incubated for 3 h at 4°C in serum-free medium containing 1 nM [ 125 I]sarile (an angiotensin II receptor antagonist, PerkinElmer Life Sciences), to assess cell surface angiotensin II receptor binding in the absence (total binding) or presence (nonspecific binding) of 500 nM angiotensin II (31). Cells were then washed thrice, solubilized in 0.1 N NaOH, and aliquoted to gamma counting and Lowry protein assay (31). Specific binding (total Ϫ nonspecific) was used to assess receptor internalization, as: 100 ϫ ([ 125 I]sarile bound to angiotensin II-challenged cells)/([ 125 I]sarile bound to naïve cells). Total binding in each well constituted Ͻ10% of the CPM used in each binding assay well, and nonspecific binding averaged 40 Ϯ 10% of total binding.
Fusion Protein Production-GST and the GST/LDLRct proteins were made in the Escherichia coli strain BL21 by standard methods (32). After elution from glutathione-agarose with reduced glutathione, fusion proteins were concentrated and dialyzed in Centriprep 10 units (Amicon). Purity of the preparations was ϳ90%, as determined by SDS-PAGE and Coomassie Blue staining.
In Vitro Binding of ␤-Arrestin2 to the LDLRct-[ 35 S]␤-arrestin2 was synthesized by in vitro translation, using the TNT® Quick Coupled Transcription/Translation System (Promega), [ 35 S]methionine/[ 35 S]cysteine (11 mCi/ml, PerkinElmer Life Sciences), and the rat ␤-arrestin2 expression plasmid (23), according to the manufacturer's instructions. Serially diluted reticulocyte lysate and purified ␤-arrestin2 were immunoblotted to quantitate [␤-arrestin2] in the lysate preparations (ϳ5.4 fmol/l). The indicated concentration of ␤-arrestin2 was incubated with 0.5 g of purified GST, GST/LDLRct, or GST/LDLRct-S833D and 10 l of glutathione-agarose (Sigma) in KOAc buffer (mM: K ϩ acetate 100, HEPES 50, MgSO 4 0.5, DTT 0.2, 0.2% bovine serum albumin, and protease inhibitors, pH 7.4), in a total volume of 240 l. After mixing for 2 h (4°C), beads were pelleted, washed thrice in KOAc buffer, and dried. Proteins were then desorbed by heating in Laemmli buffer at 65°C for 10 min and subjected to SDS-PAGE on 10% gels. Gels were stained with Coomassie Blue, and dried for autoradiography. Radioactivity in ␤-arrestin2 bands was measured with a PhosphorImager TM (Amersham Biosciences). For each [␤-arrestin2], nonspecific binding was defined as the ␤-arrestin2 cpm pulled down by GST alone and was subtracted from the ␤-arrestin2 cpm bound to the GST/LDLRct construct, to obtain "specific binding." Nonspecific binding constituted 40% of total binding for assays with the GST-LDLRct-S833D, and 60% of total binding for assays with the GST-LDLRct. Assays were performed in triplicate.
Statistical Analyses-For saturation binding experiments, nonlinear regression was performed with Prism 2 TM software (GraphPad, Inc.), using a variable Hill slope. LDL uptake in various cell lines and mouse serum lipid values obtained on low and high fat diets were compared using one-way analysis of variance, with Tukey's multiple comparison post test. All p values are two-tailed.

␤-Arrestin2
Deficiency Elevates Serum LDL-cholesterol Levels-To determine whether ␤-arrestin2 could affect LDL metabolism in vivo, we assayed lipoprotein levels in ␤-arrestin2deficient and wild type littermate control mice. On a low fat diet, ␤-arrestin2 Ϫ/Ϫ and wild type mice had indistinguishable serum lipoprotein cholesterol levels (Fig. 1). However, on a high fat, high cholesterol diet, ␤-arrestin2 Ϫ/Ϫ mice attained lipid values significantly higher than their wild type cohorts. Total cholesterol levels were 23% higher in ␤-arrestin2 Ϫ/Ϫ than in wild type mice (p Ͻ 0.05). Moreover, the sum of LDL and IDL cholesterol was 51% higher, and non-HDL cholesterol was 63% higher in ␤-arrestin2 Ϫ/Ϫ than in control mice (p Ͻ 0.01). Nonetheless, HDL-cholesterol levels were equivalent in both mouse cohorts. Thus, with a predominant effect on the LDLR ligands IDL and LDL (1), ␤-arrestin2 deficiency augmented diet-induced hypercholesterolemia. One possible explanation for this phenomenon was that ␤-arrestin2 enhanced endocytosis of the LDLR. To test this possibility, we undertook subsequent studies with purified proteins and with cells expressing both physiologic and supraphysiologic levels of the LDLR and ␤-arrestins.
Association of ␤-Arrestins with the LDLR-If ␤-arrestin2 augmented endocytosis of the LDLR in vivo, we would expect to observe an association of the LDLR with ␤-arrestin2. To determine whether the LDLR and ␤-arrestins could associate with each other in intact cells, we co-immunoprecipitated ␤-ar-

␤-Arrestin2 Enhances LDL Receptor Endocytosis
restins with the LDLR. As Fig. 2 demonstrates, ␤-arrestin2 and ␤-arrestin1 do indeed associate with the LDLR in intact cells, whether or not the receptor is bound to LDL (data not shown). Thus, the ␤-arrestin/LDLR interaction seems to differ from the ␤-arrestin/heptahelical receptor interaction: whereas association with the LDLR appears ligand-independent, association with heptahelical G protein-coupled receptors appears to be agonist-dependent (16). Interestingly, the ligand dependence of the ␤-arrestin/receptor interaction mirrors the ligand dependence of receptor endocytosis. LDLR endocytosis appears to be ligand-independent (17), whereas heptahelical receptor endocytosis appears to be agonist-dependent (15).
Functional Effects of ␤-Arrestins on LDLR Endocytosis-To test whether the observed association of ␤-arrestins with the LDLR in cells affected LDLR function, we assessed LDL uptake in the same transfected cell system used for co-immunoprecipitation. Overexpression of either ␤-arrestin1 or ␤-arrestin2 augmented LDLR endocytosis, by ϳ70% (p Ͻ 0.05), compared with cells expressing only endogenous levels of ␤-arrestins (Fig. 3). Thus, as with endocytosis of the ␤ 2 -adrenergic receptor (33), endocytosis of the LDLR, too, can be enhanced by ␤-arrestin overexpression.
If ␤-arrestins interacted with the LDLR as they do with heptahelical receptors, mutations in the N-terminal domain of the ␤-arrestins should impair the ␤-arrestin/LDLR interaction (9). To examine this possibility, we employed a V53D mutant of ␤-arrestin1. The binding of this ␤-arrestin1 mutant to heptahelical receptors is impaired (9), but its binding to clathrin and AP-2 is intact (13,34). For these reasons, this mutant has been used to inhibit heptahelical receptor endocytosis in HEK 293 cells, which express relatively high levels of endogenous ␤-arrestin isoforms (33). Just as it inhibits agonist-induced G protein-coupled receptor endocytosis, the ␤-arrestin1 V53D mutant inhibited constitutive endocytosis of the LDLR (Fig. 4). Moreover, the degree of inhibition seen with ␤-arrestin1 V53D was comparable to that observed with the K44A (dominantinhibitory) mutant of dynamin (Fig. 4). Thus, it appears that ␤-arrestins do bind the LDLR as they bind to heptahelical receptors via their N-terminal domain.

Physiologic Levels of ␤-Arrestin2
Enhance LDLR Endocytosis-If ␤-arrestins play a role in LDLR endocytosis, then cells lacking ␤-arrestin isoforms would be expected to demonstrate impaired LDL uptake. We tested this possibility with fibroblasts from mouse embryos deficient in both ␤-arrestin1 and ␤-arrestin2. To minimize the effects of genetic variability on comparisons between ␤-arrestin-expressing and ␤-arrestin1/2 double-knockout cells, we stably transfected the double-knockout cells to express either ␤-arrestin1 alone or ␤-arrestin2 alone, each at levels equivalent to those prevailing in wild type fibroblasts (Fig. 5A). Either in the absence of both ␤-arrestins or in the presence of physiologic levels of ␤-arrestin1, LDL uptake via endogenous LDLRs was equivalent in these fibroblasts. In contrast, physiologic ␤-arrestin2 levels enhanced LDL uptake via endogenous LDLRs by 65% (p Ͻ 0.05). Although physiologic levels of ␤-arrestin1 failed to augment LDLR internalization, they succeeded in augmenting angiotensin II AT 1 receptor (35) internalization (Fig. 5B), as we have shown previously (14). In these fibroblasts, therefore, physiologic levels of ␤-arrestin2, but not ␤-arrestin1, enhanced LDLR endocytosis, and thereby recapitulated results obtained with   Fig. 2, a V53D ␤-arrestin1 mutant (V53D ␤arr1), or a K44A dynamin I mutant (K44A Dyn I) construct. DiI-LDL uptake was assayed as described under "Materials and Methods." The Internalization Index was used as in Fig. 2 and was normalized to the value obtained for the control (LDLRϩempty vector (Ϫ)) cells. Shown are the means Ϯ S.E. of four experiments performed in duplicate. Compared with control cells: *, p Ͻ 0.05. the ␤ 2 -adrenergic receptor (14). This selectivity of the LDLR for ␤-arrestin2 expressed at physiologic levels, in fibroblasts, contrasted with the lack of ␤-arrestin selectivity observed when we overexpressed ␤-arrestins and LDLRs in CHO cells (Fig. 3).
␤-Arrestin2 Binds the LDLR Stoichiometrically-Functionally significant interaction between ␤-arrestins and phosphorylated heptahelical receptors occurs with a receptor:␤-arrestin stoichiometry of ϳ1 (10). We therefore sought to determine whether the augmentation of LDLR endocytosis by ␤-arrestin2 could correspond to stoichiometric binding of the LDLR with ␤-arrestin2. For this purpose, we employed a GST fusion protein encompassing the 50 amino acids of the LDLR cytoplasmic tail domain (LDLRct). In addition, we constructed a S833D mutant LDLRct fusion protein, intended to mimic phosphorylation of Ser 833 by the LDLR kinase, a process believed to occur in adrenocortical cells (36,37). We used these fusion proteins in glutathione-agarose pull-down assays with in vitro translated, [ 35 S]␤-arrestin2. Although the native LDLRct did bind ␤-arres-tin2, the S833D LDLRct mutant did so with ϳ2-fold greater FIG. 5. LDLR endocytosis is augmented by physiologic levels of ␤-arrestin2, but not ␤-arrestin1. A, ␤-arrestin expression: C57Bl/6J mouse embryo fibroblasts (MEFs) were either wild type (WT) or ␤-arrestin1/2-double knockout (␤arr1 0 /2 0 ). ␤arr1 0 /2 0 fibroblasts were stably transfected with either empty vector (None), or a vector expressing a FLAG TM -tagged construct of either ␤-arrestin1 (␤arr1) or ␤-arrestin2 (␤arr2). Forty g of protein from the indicated MEF line or from mouse liver were subjected to SDS-PAGE and immunoblotting, with an antiserum that recognizes ␤-arrestin1 5-fold more avidly than ␤-arrestin2 (14). Arrows indicate the migration positions of (untagged) ␤-arrestin1 (M r ϳ 50,000) and ␤-arrestin2 (M r ϳ 45,000) (8). Shown is a single immunoblot, representative of Ն3 performed. WT fibroblast [␤-arrestin2] was 2.0 Ϯ 0.2 times that observed in liver (n ϭ 4). B, receptor internalization: DiI-LDL uptake assays (left panel) and angiotensin II receptor internalization assays (right panel) were performed with these cells, as described under "Materials and Methods." LDLR internalization index and angiotensin II receptor internalization for each cell line was normalized to that obtained for ␤-arrestin1 0 /2 0 ("control") MEFs, to obtain "% of control." Cell surface LDL binding in ␤arr1 and ␤arr2 cells was 111 Ϯ 18% and 93 Ϯ 8% of control cells, respectively. With 1 nM [ 125 I]sarile, there were 12.0 Ϯ 0.6 and 14 Ϯ 2 fmol of angiotensin II (AT 1 ϩAT 2 ) receptors per milligram of protein in control and ␤arr1 cells, respectively; of these, 14 Ϯ 10% and 53 Ϯ 9% were internalized in control and ␤arr1 cells, respectively. C, DiI-LDL uptake experiments were performed with MEFs from WT ("control") and ␤-ar-  6. LDLR endocytosis is impaired by siRNA-mediated suppression of ␤-arrestin2 but not ␤-arrestin1. Efficacy of RNAi was assessed by immunoblotting 25 g of cell protein for ␤-arrestins 1 and 2 (␤arr1 and ␤arr2) with the same antiserum used in Fig. 5 (top panel). Subsequently, film exposure time for chemiluminescence blots was optimized for visualizing either ␤arr1 or ␤arr2, and the blot was stripped and re-probed for actin, to control for protein loading. Displayed (top panel) is an immunoblot (IB) from a single experiment performed in duplicate, representative of two performed. These siRNAtreated cells were used in DiI-LDL uptake assays, as in Fig. 5. Displayed (bottom panel) are the mean Ϯ S.E. of two experiments performed in duplicate. Cell surface LDL binding in cells treated with siRNA for ␤arr1 and ␤arr2, respectively, was 68 Ϯ 20% and 155 Ϯ 15% of that measured in control cells. Compared with control cells: *, p Ͻ 0.05. affinity (Fig. 7), much as phosphorylated heptahelical receptors bind ␤-arrestins with higher affinity than their non-phosphorylated counterparts (10). Saturation binding experiments with these fusion proteins demonstrated that, at saturation, ϳ1.5 mol of ␤-arrestin2 was bound per mole of either the wild type or the S833D mutant LDLRct (Fig. 7B). This stoichiometric binding of ␤-arrestin2 to the LDLRct further substantiated the potential physiologic importance of their interaction.
Because the S833D mutant LDLRct binds ␤-arrestin2 with higher affinity than the wild type LDLRct, and because phys-iologically expressed ␤-arrestin2 appears to enhance LDLR endocytosis, we asked whether the S833D mutation of the full-length LDLR would enhance LDLR endocytosis in intact cells (Fig. 7C). We found that LDLR endocytosis was augmented 36% when Ser 833 was mutated to Asp. By contrast, as previously reported (3), LDLR endocytosis was not affected when Ser 833 was mutated to Ala. Thus, a mutation that increases the affinity of LDLRct/␤-arrestin2 binding also enhances LDLR endocytosis in cells expressing physiologic levels of ␤-arrestin2.
A Role for LDLR Tyr 807 in the ␤-Arrestin2/LDLRct Association-Endocytosis of the LDLR depends critically upon the reverse-turn conformation of the receptor's cytoplasmic tail, conferred by the tetrapeptide sequence 804 NPVY 807 (38). Mutation of Tyr 807 to any but aromatic residues reduced LDLR internalization by ϳ80% in transfected CHO cells (3). Because the Y807A LDLR mutation reduced LDLR endocytosis in fibroblasts (3), and because ␤-arrestin2 appears to enhance LDLR endocytosis in fibroblasts (Fig. 5), we asked whether the LDLR Y807A mutation would affect the association of ␤-arrestin2 with the LDLR. To address this question, we employed glutathione-agarose pull-down assays with the wild type and S833D LDLRct constructs used in Fig. 7. Mutation of Tyr 807 to Ala abrogated binding of ␤-arrestin2 to the LDLRct constructs (Fig.  8). Thus, the association of the LDLR with ␤-arrestin2, as with the ARH protein (6) and clathrin itself (5), appears to require Tyr 807 . The LDLR Y807A mutation could reduce LDLR endocytosis in fibroblasts, at least in part, by eliminating the association of ␤-arrestin2 with the LDLR.

DISCUSSION
The LDLR represents a previously unrecognized, functionally significant binding partner for ␤-arrestin2. By showing that physiologic levels of ␤-arrestin2 can augment LDLR endocytosis in cultured fibroblasts and 293 cells, these studies demonstrate a novel mechanism by which LDLR endocytosis may be facilitated. Although ␤-arrestin2 is clearly not necessary for LDLR internalization, ␤-arrestin2 can clearly enhance LDLR internalization. Indeed, ␤-arrestin2-mediated enhancement of hepatic LDLR endocytosis could very plausibly explain the results of our in vivo lipoprotein metabolism studies. Alto- FIG. 7. Stoichiometric binding of ␤-arrestin2 to the LDLR cytoplasmic tail: increased binding affinity with mutation of LDLR Ser 833 to Asp. Glutathione-agarose pull-down assays were performed with in vitro translated [ 35 S]␤-arrestin2 and either GST itself, a GST fusion protein encompassing the 50 amino acids of the LDLR cytoplasmic domain (GST-LDLRct, "WT"), or the GST-LDLRct with a S883D mutation (S833D). After glutathione-agarose pull-down, specimens were subjected to SDS-PAGE and autoradiography. A, a sample autoradiogram is shown. B, glutathione-agarose pull-downs were performed with the indicated concentration of [ 35 S]␤-arrestin2 and 0.5 g of either the GST-LDLRct WT, GST-LDLRct S833D ("total binding"), or GST alone ("nonspecific binding"). Specific binding (total Ϫ nonspecific) was calculated as under "Materials and Methods" and normalized to the value obtained with binding of 107 nM ␤-arrestin2 to the LDLRct S833D construct, to obtain percentage of control maximum. Shown are the means Ϯ S.E. from Ն3 experiments performed in triplicate. Data were fit to a sigmoidal dose-response curve with variable Hill slope, yielding R 2 values of 0.94 -0.96. Saturation was reached for both LDLRct constructs at 97 nM ␤-arrestin2. C, 293 cells were transfected with plasmids encoding either the WT LDLR or the LDLR in which Ser 833 was mutated to either Asp or Ala, as indicated. DiI-LDL uptake assays were performed as described for gether, our in vivo and in vitro studies demonstrate for the first time that ␤-arrestin2 can participate in receptor endocytosis that is constitutive, and not just agonist-promoted.
The apparent selectivity of the LDLR for ␤-arrestin2 may result from greater binding affinity to ␤-arrestin2 than ␤-ar-restin1. If true, such a mechanism would replicate that invoked to explain the selectivity of the ␤ 2 -adrenergic receptor for ␤-ar-restin2 (39), as well. Alternatively, however, the apparent ␤-ar-restin2 selectivity of the LDLR could result from ␤-arrestin isoform-specific mechanisms for linking receptors to the endocytic machinery. ␤-Arrestin2 has a ϳ6-fold greater affinity for clathrin than ␤-arrestin1 (12), and ␤-arrestin2 interacts with AP-2 more robustly than ␤-arrestin1 in yeast two-hybrid assays (40). Although both ␤-arrestins are phosphorylated in their C-terminal domains, this phosphorylation impairs endocytosis mediated by ␤-arrestin1 more so than that mediated by ␤-arrestin2 (41,42). Dephosphorylation of both ␤-arrestins at the plasma membrane has been shown to be agonist-dependent, by mechanisms that remain unknown. If ␤-arrestin binding to the LDLR fails to promote ␤-arrestin dephosphorylation, then ␤-arrestin2 should be expected to mediate receptor internalization more readily than ␤-arrestin1.
For the LDLR, identification of ␤-arrestin2 as an adaptor protein may help to elucidate previously enigmatic aspects of receptor regulation. Phosphorylation of the LDLR on Ser 833 , a residue conserved from frog to human (43), occurs in adrenocortical cells but not in fibroblasts (36,37). Over more than a decade, potential physiologic purposes for this phosphorylation event have remained obscure. Mutation of Ser 833 to Ala failed to affect either LDLR endocytosis in CHO fibroblasts (3), the basolateral targeting of LDLRs in transgenic mouse livers, or the extent to which hepatocyte LDLRs reside outside of coated pits (44). However, modeling Ser 833 phosphorylation with an LDLRct S833D mutation increased both the affinity of ␤-arres-tin2/LDLRct binding and the efficiency of LDLR endocytosis in cells expressing physiologic levels of ␤-arrestin2 (Fig. 7). In adrenocortical cells, Ser 833 phosphorylation conceivably could, by enhancing LDLR/␤-arrestin2 interaction, enhance LDLR endocytosis. However, even in the absence of Ser 833 phosphorylation, ␤-arrestin2 may still enhance LDLR endocytosis, as suggested by our data from ␤-arrestin1/2 double-knockout and ␤-arrestin2 knockout MEFs (Fig. 5) and siRNA-treated 293 cells (Fig. 6).
Another potential role for ␤-arrestin2 in LDLR endocytosis is suggested by the human disease autosomal recessive hypercholesterolemia (ARH). The ARH protein, like ␤-arrestin2, can link the LDLR to clathrin and AP-2 (6). LDLR endocytosis is dramatically reduced in lymphocytes and hepatocytes of ARH subjects. However, there is little, if any detectable abnormality of LDLR endocytosis in ARH fibroblasts (6,7). Could ␤-arres-tin2 compensate for ARH deficiency in fibroblasts but not in hepatocytes or lymphocytes? This possibility is suggested by our immunoblot analysis of mouse liver and fibroblasts, in Fig.  5, and LDL uptake studies. In mice, fibroblasts appear to have ϳ2-fold more ␤-arrestin2 than liver. In LDL uptake studies with ␤-arrestin1/2 double-knockout fibroblasts stably transfected with ␤-arrestin2, expression of ␤-arrestin2 failed to augment LDLR endocytosis when ␤-arrestin2 levels were Յ0.5 times those obtaining in wild type fibroblasts (data not shown).
Our in vivo studies demonstrate that, unlike ARH in humans (7), ␤-arrestin2 is not necessary for normal lipoprotein metabolism in the low fat-fed mouse. However, on a high fat diet that down-regulates LDLRs (45), ␤-arrestin2 function does appear to contribute to steady-state levels of IDL and LDL cholesterol in the mouse. Increases in IDL and LDL associated with ␤-ar-restin2 deficiency are consonant with a role for ␤-arrestin2 in hepatic LDLR endocytosis, but they do not exclude an alternative or concomitant role for ␤-arrestin2 in the endocytosis of other receptors, including the hepatic LDLR-related protein (46). Whether ␤-arrestins may affect endocytosis of the LDLRrelated protein or other members of the LDLR family remains to be explored.