The LDL Receptor Clustering Motif Interacts with the Clathrin Terminal Domain in a Reverse Turn Conformation

The peptides FDNPVYQKTT, FDNPVY, FDNPVSQKTT, FDNPVS, FDNPVA, FDNPVF, and FDNPVL were synthesized using standard FMOC chemistry with a free COOH terminus and an acylated NH₂ terminus. Crude peptides were purified by reverse-phase HPLC on a C18 column (Vydac, Hesperia, CA), equilibrated in H₂O with 0.1% trifluoroacetic acid, and eluted with a linear gradient of acetonitrile. The identities of the peptides were confirmed by mass spectrometry and amino acid analysis. Concentrated stock solutions of peptides were made by dissolving the peptide in D₂O and adjusting the pH to 6.0 with aliquots of concentrated NaOH.

Clathrin coat proteins were isolated from bovine brain by standard procedures (8). The coat proteins were then gel-filtered over a Superdex 200 column (Pharmacia Biotech, Inc., Piscataway, NJ) equilibrated in 0.6 M Tris at pH 9.0. Clathrin and AP fractions were pooled and loaded directly onto an anion exchange column (Source Q 15; Pharmacia Biotech, Inc.) equilibrated in 20 mM ethanolamine, pH 9.0, and eluted with a steep linear gradient of 1 M NaBr. Clathrin and AP fractions were pooled and concentrated using Centriprep 30 (Amicon Corp., Easton, TX). Concentrated proteins were polymerized (25) by extensive dialysis against NMR sample buffer (40 mM D₄-malonic acid [Cambridge Isotope Labs, Andover, MA], pH 6.0, 0.2 mM EDTA, 0.2 mM DTT, and 3 mM NaN₃). Protein concentration of the polymerized cage complex was determined using the Bradford reagent (Bio-Rad Laboratories, Hercules, CA).

For preparation of cages composed of clathrin or clathrin and AP2, the purification was slightly modified. An additional anion exchange chromatography step was performed to separate clathrin from AP2 before gel filtration. Each protein was then purified separately as before. AP-2 containing fractions from the second anion exchange step were then pooled on the basis of Coomassie blue–stained SDS-PAGE. The clathrin fraction was divided in half, and all of the AP-2 sample was added to one half of the clathrin. Both samples were concentrated in Centriprep 30 and dialyzed extensively as described above. Equal total protein amounts of clathrin or clathrin/AP-2 cages were added to each NMR sample, which contained clathrin (9.7 mg/ml) or clathrin (8.5 mg/ml) and AP-2 (1.2 mg/ ml). Sample purity was confirmed by Coomassie blue–stained SDS-PAGE gels and by Western blotting with an antibody against β-adaptin (100/1; Sigma Chemical Co., St. Louis, MO).

The clathrin terminal domain (residues 1–579) was expressed as a glutathione-S-transferase (GST) fusion protein in Escherichia coli BL21 cells, and was prepared as described (17). Glutathione agarose beads (Sigma Chemical Co.) with the bound GST construct were then washed with thrombin cleavage buffer (50 mM Tris, 2.5 mM CaCl₂, 150 mM NaCl, pH 7.4). The beads were incubated overnight with gentle agitation in the presence of thrombin (Sigma Chemcial Co.). DTT was added to 10 mM and allowed to incubate an additional 1 h before eluting the cleaved terminal domain construct from the beads. Samples were concentrated and dialyzed overnight against NMR sample buffer with 100 mM KCl.

The clathrin hub construct (residues 1074–1483) was expressed as a hexahistidine construct in E. coli BL21(DE) cells as described (29). Bacterial lysates were incubated with Ni-NTA agarose (QIAGEN Inc., Valencia, CA) overnight and eluted with 250 mM imidazole in 50 mM Tris, pH 7.8. The eluted polypeptides were then loaded directly onto an anion exchange column equilibrated in 30 mM Tris, 1 mM EDTA, pH 7.8, and eluted with a linear gradient of 0.5 M KCl. Fractions were pooled on the basis of Coomassie blue–stained SDS-PAGE gels. Because the hub polypeptide aggregated in the NMR sample buffer, both the terminal domain and hub polypeptides were dialyzed against 40 mM KPO₄, 150 mM KCl, 0.2 mM EDTA, 3 mM NaN₃, and 0.2 mM DTT pH 7.2.

All spectra were acquired at 500 MHz and 25°C in sample buffer containing 10% D₂O on a Varian Unity 500 spectrometer with a 5999.7-Hz spectral width using 3-(tetramethylsilyl) propionic acid as a chemical shift reference. 1D spectra were acquired with 128 transients of 64,000 points zero filled to 131,072 points (see Fig. 1) or 64 transients of 32,000 points zero filled to 65,536 points (see Figs. 4 and 5). Line-broadening was measured as the difference of peak widths at half height using the Felix 2.30 (Biosym Technologies) program. The broadening is reported as the line-width of the resonance peak in the presence of cages minus the line-width in the absence of cages. Crosspeak intensities were calculated in arbitrary units of volume using the Felix 2.30 program from spectra acquired at 60-, 100-, 140-, and 220-ms mixing times. The spectra were acquired in phase-sensitive mode (48) with 2 × 256 FIDs of 1024 complex points each. A spectral width of 5999.7 Hz was used in both dimensions, and 32 transients per FID were collected. The water signal was suppressed by presaturation.

The relative K_d for wild-type and position 807 mutant peptides from clathrin terminal domains was obtained from measurements of the peak intensities of the Phe (3, 5) aromatic protons for each of the peptides in the presence and absence of protein. The free and bound populations were calculated from the following equation:

where i is the intensity of the free (i_f), the bound (i_b), and observed (i_o) resonance, and p^f and p^b are the free and bound populations, respectively. Since peak intensities are inversely proportional to the transverse relaxation rate (1/T₂), the population of the free and bound peptide can be estimated from the change in intensity of a given peptide resonance upon addition of clathrin terminal domain if the T₂ of the resonance in the free and bound forms is known. The T₂ for a protein is approximately proportional to the molecular weight, and from the line-widths observed in 20-kD proteins (3), the expected T₂ for a 55-kD protein is ∼10 ms. The T₂ calculated from the line-widths of several resonances of the free peptides was ∼100 ms. The values of i_b were estimated to be i_f/10, assuming that the T₂s of the free and bound resonances are 100 and 10 ms, respectively. Therefore, the calculated populations of p^f and p^b were then used to obtain K_ds assuming standard protein–ligand equilibrium. Since the K_ds were based on T₂ estimates, their absolute values are likely to be inaccurate. However, the relative values of the K_ds can be compared because they were all obtained by the same method.

Distances were calculated only for interresidue NOEs seen in 60 ms mixing time data sets. Sequential Hα to NH NOEs were not used because they contained contributions from the free peptide. NOE distance restraints were calculated from both wild-type hexa- and decapeptide either by using bins of weak (w; 1.8–2.3 Å), medium (m; 1.8–3.5 Å), and strong (s; 1.8–5.0 Å) peaks, or by measured NOE volumes calibrated individually for each dimension with the average volume of the Phe, Tyr, and Pro (β/β) NOEs corresponding to 1.75 Å. Structures were calculated by simulated annealing performed with the programs INSIGHTII and NMRCHITECT (Biosym Technologies, San Diego, CA) incorporating either type of restraint. In short, peptides were heated to 1,000 K, and then the distance and chiral restraints were slowly increased over 30 ps. The peptides were then slowly cooled to 300 K while increasing covalent and nonbond forces over 32 ps. The final structures were then obtained by iteratively minimizing to a final derivative of 0.01 kcal/mol⁻¹/Å⁻² and simulating 30 ps of restrained dynamics. Similar sets of structures were obtained with both sets of restraints. Stereospecific assignments of Pro(β,d,γ) were based on interresidue distances and geometry. Removal of pseudoatom restraints for Tyr (δ, ε) and Phe(δ) was based on the geometry of an initial set of structures. The final set of 30 restraints was (* denotes pseudoatom and distance is denoted as s, m, or w): Ac(Me*)::F(Hδ₁), w; F(Hδ₁):: D(HN), w; F(Hβ₂)::D(HN), w; D(Hβ₂)::N(HN), w; N(Hβ₁)::P(Hδ_S), w; N(Hβ₂)::P(Hδ_S), m; N(Hβ₁)::P(Hδ_R), w; N(Hβ₂)::P(Hδ_R), w; N(HN):: P(Hδ_R), w; P(Hβ_R)::V(HN), m; P(Hβ_S)::V(HN), w; P(Hδ_S)::V(HN), w; V(HN)::Y(HN), s; V(HN)::Y(Hδ₁), w; V(Hγ₁*)::Y(Hδ₁), w; V(Hγ₁*):: Y(Hε₁), w; Ac(Me*)::D(HN), w; Ac(Me*)::N(Hδ₂₁), w; Ac(Me*):: N(Hδ₂₂), w; V(HN)::N(Hβ₁), w; V(HN)::N(Hβ₂), w; N(Hβ₁)::Y(Hδ₂), w; N(Hβ₂)::Y(Hδ₁), w; P(Hγ_s)::Y(Hδ₁), w; P(Hγ_s)::Y(Hε₁), w; P(Hγ_s):: Y(HN), m; P(Hβ_r)::Y(HN), w; P(Hβ_r)::Y(Hδ₁), w; P(Hβ_r)::Y(Hε₁), w; P(Hα)::Y(HN), w.

Clathrin triskelions and APs purified from bovine brain were assembled into membrane-free polyhedral cages using standard methods (25). We prepared synthetic hexa- and decapeptides corresponding to the wild-type LDL receptor internalization motif (FDNPVY and FDNPVYQKTT) and to an internalization defective motif (FDNPVS and FDNPVSQKTT). Aromatic (Fig. 1, A–F) and aliphatic (Fig. 1, G and H) side chain regions from one- dimensional ¹H NMR spectra are shown for peptide samples (2.5 mM) mixed with NMR sample buffer alone, or buffer containing clathrin cages (20 mg/ml). The wild-type and mutant peptides exhibited characteristic narrow ¹H line-widths in the absence of cages (Fig. 1, A, D, and G, and data not shown). Adding cages to wild-type hexa- and decapeptides caused a pronounced increase in the line widths, indicative of an interaction between the peptide and the cages and a consequent slower effective tumbling time for the peptide (Fig. 1, B, C, and H; 36). By contrast, the cage-induced line-broadening of mutant peptides was markedly less than that of wild-type for both the hexa- and decapeptides, indicating a weaker interaction (Fig. 1, E and F). As a control for nonspecific binding or viscosity-induced broadening, we examined the decapeptides in the presence of 20 mg/ml BSA. Neither peptide displayed appreciable line-broadening (data not shown).

NOE spectroscopy (NOESY) is another method of detecting interactions between peptides and large protein complexes (26). The hexa- and decapeptides have inefficient cross-relaxation due to their small size, and thus have NOE spectra relatively devoid of cross-peaks (data not shown). A peptide interacting with a large protein complex will develop stronger NOEs because cross-relaxation is more efficient in the bound state. The resulting spin populations are then maintained in the free state where they are observed as an increase in the number and intensity of NOE crosspeaks (10, 11). As shown in Fig. 2 (A and C), the NOESY spectra of both the deca- and the hexa- wild-type peptides in the presence of cages contain numerous cross-peaks, many of which are strong. In contrast, only a few cross-peaks are observed in the NOESY spectra of the mutant peptides in the presence of cages (Fig. 2, B and D). The large difference in number and intensity of NOEs at a short mixing time (60 ms) is consistent with the line-broadening results (Fig. 1), and indicates that the wild-type peptides have higher affinity for the cages. Transferred NOEs were not observed with either wild-type or mutant decapeptide in solutions with 20 mg/ml BSA, even at long mixing times (300 ms; data not shown).

When a peptide binds to a macromolecule, the effective tumbling rates are slower for protons in the region of the peptide involved in binding, since protons in other regions are likely to experience fast internal motions. Thus, cross-relaxation is more efficient in the regions immobilized by binding, resulting in faster NOE buildup rates for a given fixed distance. To analyze the mobility of residues of the wild-type decapeptide bound to the cages, we first compared the build-up rates of transferred NOEs corresponding to covalently fixed equivalent proton distances. The similarity in NOE buildup rates (measured with 60-, 100-, 140-, and 220-ms mixing times) for the geminal β protons from residues F, N, P, and Y indicates that these residues behave isotropically in the bound state (Fig. 3,A). This phenomenon was also apparent in the similar line-broadening of the Tyr (εH, 3.45 Hz) and Phe (εH, 3.60 Hz) ring protons (Fig. 1). By contrast, the buildup rates for Hβ/Hγ NOEs decrease towards the COOH terminus of the peptide. Val has the most rapid buildup rate, indicating that it is immobilized the most, then Thr₉, and then Thr₁₀ (Fig. 3,B). This trend was also observed in line-broadening of the Val (γH1, 5.20 Hz; γH2, 3.82 Hz), Thr₉ (γH, 3.18), and Thr₁₀ (γH, 2.95 Hz) γ protons (compare Fig. 1, G and H). These results indicate that the decapeptide residues outside the internalization motif have greater mobility than those within the motif. Therefore, the same residues known to be critical for internalization in vivo were immobilized in the presence of cages.

The internalization motif peptide could be interacting with multiple proteins in the lattice. To distinguish between the contributions of AP2 and clathrin, we prepared clathrin cages composed of either clathrin alone or clathrin plus purified AP2. AP2 alone could not be tested because it aggregated under our experimental conditions. Fig. 4 shows that the wild-type hexapeptide in the presence of clathrin cages displayed distinct line-broadening (compare Fig. 4,A with Fig. 4,B). The mutant peptide, by contrast, showed much less line-broadening (compare Fig. 4,D with Fig. 4,E). Some additional line-broadening was seen when the wild-type peptide was mixed with cages composed of both clathrin and AP2 (compare Fig. 4,B with Fig. 4,C). However, it is likely that the broadening arose primarily from clathrin (see Discussion). The mutant peptide (Y₈₀₇S), by contrast, showed much less line-broadening (compare Fig. 4,B with Fig 4,E, and Fig. 4,C with Fig. 4 F). These results suggest that the LDL receptor interacts with clathrin through the FXNPXY motif during receptor clustering in coated pits.

We examined further the interaction between the hexapeptides and clathrin using recombinant portions of the clathrin molecule. The terminal domain of clathrin (residues 1–579) expressed as a GST fusion protein was purified, and thrombin was cleaved from GST. Clathrin hubs (residues 1074–1483) were expressed as a hexaHis-tagged polypeptide, and were purified by a Ni resin and anion exchange. Line-broadening measurements were first made using the terminal domain at the acidic pH used in the cage assays. The wild-type peptide (Fig. 5) displayed the same degree of line-broadening (compare Fig. 5,A with Fig. 5,B) seen when peptide is incubated in the presence of clathrin cages. The mutant peptide, by contrast, displayed much less broadening (compare Fig. 5,C with Fig. 5,D). Equimolar amounts of either BSA or GST under the same conditions had no effect (data not shown). We saw a similar degree of line-broadening at neutral pH (compare Fig. 5,E with Fig. 5,F), but no appreciable broadening was detected when the hub region was substituted for the terminal domain at neutral pH (compare Fig. 5,E with Fig. 5 G). These results show that the LDL receptor tail selectively interacts with the terminal domain of clathrin.

The amino acid requirements at position 807 for LDL receptor internalization (14) have been found to correlate with the propensity of peptides from that region to form a reverse turn (2). We examined position 807 mutants for their ability to interact with clathrin terminal domains by measuring the change in peak intensity of 1D proton resonances for each peptide. Because peak heights are inversely proportional to the effective molecular weight, they can be used to calculate the relative binding constant for each peptide as it interacts with the terminal domain (see Materials and Methods). Table I shows that the relative affinity of each peptide for the clathrin terminal domain strongly correlated with the predicted turn propensity (2) as well as the ability of receptors bearing these mutations to internalize LDL (14). Thus, peptides with Tyr and Phe at position 807 had the highest affinity, the greatest turn propensity, and the highest internalization rate, while a Leu at this position had an intermediate affinity and both Ser and Ala had the lowest affinities.

Transferred NOEs have been used previously to determine the conformation of peptides bound to proteins (26, 27, 49). These results have later been confirmed from crystal structures of the complexes (40, 57, 62). We used data from transferred NOE spectra to determine the structure of the FDNPVY motif bound to clathrin cages. Similar NOE results were obtained when cages were replaced with recombinant terminal domains (data not shown). Structures were calculated by simulated annealing (37) using 30 nontrivial transferred NOE restraints. Among these, 23 NOEs correspond to the residues NPVY, including the strong NH/NH NOE between Val and Tyr characteristic of a turn conformation (Fig. 6,A), and other short and intermediate range, structurally diagnostic NOEs. A superposition of the peptide backbones for the 21 structures with distance violations of <0.1 Å obtained among 30 calculated structures is shown in Fig. 6,B. All resulting structures had an extended NH₂ terminus followed by a Type 1 β turn around residues NPVY. The two NH₂-terminal amino acids did not align as well in the superposition. This result is more likely to arise from a lack of proton density in the extended structure of this region than from increased mobility, since line-broadening and transferred NOE build-up rates indicate that Phe is immobilized (see above). The 21 calculated structures had a backbone root mean square deviation of 0.61 Å, indicating that a single conformational form of the peptide was bound by the cages. Based on this information, a representative structure is shown in Fig. 6 C.

The surprising conclusion of this study is that clathrin may be the molecule in coated pits responsible for capturing the LDL receptor during receptor-mediated endocytosis. Four stringent criteria were used to distinguish specific from nonspecific interactions between the clustering motif peptide and clathrin. First, changing tyrosine 807 to serine, which matches the mutation that prevents LDL receptor clustering in coated pits (13), substantially reduced interaction of the peptide with the cages. Second, the wild-type peptide bound to the cages in a single conformation, and had the predicted (2, 12) reverse turn conformation. Third, binding was limited to those residues that lie within the FXNPXY motif. Fourth, once clathrin was identified as a candidate molecule, the wild-type peptide was found to interact with the terminal domain, but not with the hub portion of recombinant clathrin. Furthermore, the relative affinity of terminal domains for peptides with various position 807 substitutions correlated strongly with both the LDL receptor internalization index (14) and the predicted turn propensity (2). These results suggest that the receptor internalization rates depend on the affinities of the hexapeptide motif for clathrin, and that the affinity depends on a reverse turn conformation.

The apparent affinity we measured between the hexapeptides and clathrin terminal domains is in the same range as that reported for the interaction between YXXΦ peptides and the μ subunit of AP (4, 44). It is hard to judge the functional significance of these affinities. Within the cell, LDL receptors probably exist as multimers interacting with the rigid clathrin lattice in a two-dimensional plane (52). In this environment, the receptors will have a restricted rotational and translational mobility as they interact with a multivalent, nearly crystalline set of terminal domains. Assuming that a 76-nm–diameter clathrin-coated vesicle contains 60 clathrin triskelions (21), we calculated the concentration of terminal domains to be at minimum 1.5 mM. This result suggests the effective concentration of terminal domains in the region where receptor cluster is quite high, so the relatively low-affinity interactions we have measured should be sufficient to mediate clustering.

Clathrin terminal domains have been implicated before in receptor clustering (17, 18). The binding of β-arrestin to the phosphorylated cytoplasmic tail of the β₂-adrenergic receptor appears to be required for receptor migration to coated pits. The binding of β-arrestins to clathrin is stoichiometric, clustering in coated pits is coincident with receptor internalization, and expression of β-arrestin mediates internalization (18). The β-arrestin binding site has been localized to the first 100 amino acids of the clathrin heavy chain (17). Structural studies show that the terminal domain of clathrin is optimally positioned in an assembled lattice to interact with the cytoplasmic tails of receptors (53). However, none of these studies have sufficient resolution to determine the location of the terminal domains in an assembled clathrin coat.

Although clathrin terminal domains clearly interact with the NPXY clustering motif, we have not ruled out the possibility that APs can bind to the cytoplasmic portion of the LDL receptor, or that they can modulate the interaction of the NXPY motif with clathrin. The presence of AP2 in our clathrin cage preparations moderately enhanced line-broadening of the NMR spectra compared with clathrin alone, suggesting that the wild-type peptides may have a higher affinity for cages containing AP2 (Fig. 4). The bound peptide displayed a well-defined conformation in the presence or absence of APs, so broadening must be due to a single type of binding site for the hexapeptide. Therefore, most likely the enhanced broadening is not due to introduction of additional hexapeptide binding sites present in the APs.

Our results stand in contrast to previous studies showing that purified AP2 binds substoichiometrically to a fusion protein containing the cytoplasmic tail of the LDL receptor (42). The fusion protein used in this study contained 31 residues from the λcII protein, a thrombin cleavage site, 112 residues of myosin light chain, a second thrombin cleavage site, and the 50-residue LDL receptor tail (42). Interestingly, the cII fragment contains a dileucine motif that can directly interact with APs (20, 43). Since a fusion protein containing an internalization-defective LDL receptor tail (e.g., Y807C) was not tested, these studies did not rule out the possibility that the interaction with AP2 was mediated by the dileucine motif. Moreover, there is compelling in vivo evidence that AP2 does not cluster LDL receptors. Depletion of intracellular K⁺ and hypertonic treatment of cells both cause reversible loss of clathrin lattices from the cell surface (22, 28), which is accompanied by LDL receptor unclustering (22, 28). By contrast, these conditions do not uncluster AP2 (19). If receptor clustering were primarily dependent on AP2, then LDL receptors should have remained clustered under these conditions.

An important prediction from these results is that LDL receptors should be sorted and concentrated during clathrin-coated vesicle formation regardless of whether budding occurs from Golgi, endosome, or surface membranes. Coated vesicles purified from adrenal glands are enriched in masked LDL receptors four to fivefold (33). These isolated vesicles contain coated vesicle derived from both the plasma membrane and the Golgi apparatus that can be separated on the basis of size (5). Ligand blotting indicates that both populations of vesicles are equally enriched in LDL receptors relative to the plasma membrane (5). Therefore, even though there are no quantitative EM studies showing LDL receptors concentrated in Golgi-coated vesicles, these biochemical experiments suggest that sorting does occur.

AP2 does appear to be responsible for capturing receptors that contain YXXΦ (39) and dileucine internalization motifs (20, 43). Two-hybrid screen (38, 39) and surface plasmon resonance analysis (4, 39) both show that YXXΦ interacts with the μ chain of AP2. The properties of YXXΦ motif are distinct from those for FXNPXY since FXNPXY motifs did not bind the μ subunit, and a combinatorial library showed no preference for this motif (4). Dileucine motifs also bind clathrin AP2 and AP1, yet receptors with this motif do not compete with receptors containing YXXΦ during internalization (31). Receptor internalization by coated pits can therefore be mediated by at least three distinct motifs (two tyrosine-based and one dileucine-based), each of which is captured by a different binding site in the lattice.

Tyrosine phosphorylation alters the binding specificity of both YXXΦ and NPXY, although phosphorylation is not essential for binding to all phosphotyrosine binding (PTB) domains (6, 59). pYXXΦ binds SH₂ domains (46) while NPXpY binds proteins that contain a PTB domain (51). The structures of both pYXXΦ/SH₂ (16, 54, 55) and NPXpY/PTB domain (15, 60, 61) complexes have been solved. The former shows that pYXXΦ binds as an extended two-pronged plug. The structure of the latter confirms transferred NOE data (61) showing that the tyrosine in the bound NPXpY is in a tight turn preceded by an extended region. Since we found that bound FXNPXY is in a tight turn (Fig. 6 C), it is possible the simple addition of phosphate or specific expression of a PTB domain blocks coated pit clustering. The conformation of YXXΦ bound to the μ chain of AP has not been determined, but if the phosphate is also what changes binding specificity, then the bound conformation of YXXΦ should be an extended, two-pronged plug rather than a tight turn.

This study demonstrates the power of NMR spectroscopy to identify the molecules in coat protein complexes responsible for sorting membrane proteins during vesicular traffic. At the same time, this method provides dynamic structural information about the targeting sequence when it is bound. Multiple targeting motifs regulate receptor trafficking throughout the cell. Internalization motifs can function at multiple steps along an internalization pathway, differentially using clathrin and AP subunits to generate different migratory patterns. The specificity of clathrin for FXNPXY and APs for YXXΦ appears to be as distinct as PTB and SH2 domains are for their phosphorylated counterparts. YXXΦ and dileucine motifs may target receptors to plasma membrane, late endosome, lysosome, and TGN structures via combinatorial interactions with AP complexes and possibly other membrane coats. By contrast, the FXNPXY motif may be needed for efficient clathrin-mediated recycling of LDL receptors between the plasma membrane and endosomes. Based only on the primary sequence of these motifs, it has not been possible to predict the cellular distribution of receptors. NMR spectroscopy should be useful to characterize further the conformational determinants of sorting motifs that ultimately determine receptor sorting patterns.

We would like to thank Dr. Jim Keen for the gift of the GST-TD(1–579) construct and Dr. Frances Brodsky for the gift of the hexaHis Hub(1074– 1483) construct. We would also like to thank William Donzell and Sarah J. Stradley for valuable technical assistance and Stephanie Baldock for administrative assistance.

This work was supported by grants from the National Institutes of Health (HL20948 and GM27616) the Welch Foundation I-1304, and the Perot Family Foundation.

GST

glutathione-S-transferase

m

medium

NMR

nuclear magnetic resonance

NOE

nuclear Overhauser effect

PTB

phosphotyrosine binding

s

strong

w

weak