The cytosolic domain of T-cell receptor ζ associates with membranes in a dynamic equilibrium and deeply penetrates the bilayer

Interactions between lipid bilayers and the membrane-proximal regions of membrane-associated proteins play important roles in regulating membrane protein structure and function. The T-cell antigen receptor is an assembly of eight single-pass membrane-spanning subunits on the surface of T lymphocytes that initiates cytosolic signaling cascades upon binding antigens presented by MHC-family proteins on antigen-presenting cells. Its ζ-subunit contains multiple cytosolic immunoreceptor tyrosine-based activation motifs involved in signal transduction, and this subunit by itself is sufficient to couple extracellular stimuli to intracellular signaling events. Interactions of the cytosolic domain of ζ (ζcyt) with acidic lipids have been implicated in the initiation and regulation of transmembrane signaling. ζcyt is unstructured in solution. Interaction with acidic phospholipids induces structure, but its disposition when bound to lipid bilayers is controversial. Here, using surface plasmon resonance and neutron reflection, we characterized the interaction of ζcyt with planar lipid bilayers containing mixtures of acidic and neutral lipids. We observed two binding modes of ζcyt to the bilayers in dynamic equilibrium: one in which ζcyt is peripherally associated with lipid headgroups and one in which it penetrates deeply into the bilayer. Such an equilibrium between the peripherally bound and embedded forms of ζcyt apparently controls accessibility of the immunoreceptor tyrosine-based activation signal transduction pathway. Our results reconcile conflicting findings of the ζ structure reported in previous studies and provide a framework for understanding how lipid interactions regulate motifs to tyrosine kinases and may regulate the T-cell antigen receptor biological activities for this cell-surface receptor system.

The T-cell antigen receptor (TCR) 4 is a multisubunit cellsurface protein complex composed of eight single-pass transmembrane subunits arranged in pairs as ␣␤-␥⑀-␦⑀-2 (Fig. 1). The ␥⑀ and ␦⑀ subunits carry the CD3 cell-surface epitope and are referred to by this name. (The homodimeric 2 subunit carries the CD247 epitope but is not usually referred to by using this designation.) TCR signal transduction is triggered by interaction with MHC-peptide complexes on antigen-presenting cells. Variable and hypervariable regions on the extracellular antigen-binding domains of the TCR ␣␤ subunits contain the binding site for MHC-peptide. CD3␥, -␦, and -⑀ possess extracellular immunoglobulin-like domains that form heterodimers and associate with TCR␣␤. The 2 subunits have very small extracellular portions and associate with the receptor complex primarily via transmembrane interactions. The activation mechanism is incompletely understood at the molecular level, but it is well established that extracellular MHC-peptide binding triggers cytosolic signaling pathways (1), with the first step believed to be phosphorylation, by Src-family kinases, of tyrosine residues on intracellular TCR immunoreceptor tyrosinebased activation motifs (ITAMs) of the general sequence, YXX(L/I)X 6 -8 YXX(L/I) (2). The cytosolic domains of CD3␥, -␦, and -⑀ each contain one ITAM, and the cytosolic domain of (TCR cyt ) has three. cyt by itself can initiate signal transduction This work was supported by National Institutes of Health Grants AI-38996 (to L. J. S.) and GM-101647 (to M. L.) and by the Department of Commerce through its Measurement Science and Engineering program (Grant 70NANB13H009) and the IMS program "Precision Measurement for Integral Membrane Proteins." The authors declare that they have no conflicts of interest with the contents of this article. The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health. Certain commercial materials, equipment, and instruments are identified in this work to describe the experimental procedure as completely as possible. In no case does such an identification imply a recommendation or endorsement by National Institute of Standards and Technology, nor does it imply that the materials, equipment, or instrument identified are necessarily the best available for the purpose. 1 Present address: Harvard Business School, Boston, MA 02163. 2 Present address: Intel Corp., Beaverton, OR 97007. 3 To whom correspondence should be addressed: Dept. of Pathology, University of Massachusetts Medical School, 55 Lake Ave. N., Worcester, MA 01655. Tel.: 508-856-1831; Fax: 508-856-0019; E-mail: Lawrence.Stern@ umassmed.edu.
processes, as shown with chimeric proteins on which cyt was fused to an unrelated extracellular domain (3) and used as a basis for chimeric antigen receptor cancer immunotherapy (4). A schematic representation of the intact TCR(␣␤)(␥⑀)(␦⑀) ( 2 ) complex is depicted in Fig. 1, but its three-dimensional structure is unknown. However, structures have been determined for several sections (5)(6)(7)(8)(9)(10)(11)(12)(13), such as the isolated TCR␣␤ extracellular antigen-binding domains, both free (5) and MHCbound (6). Structures of the extracellular domains of CD3␥⑀ (7,8) and CD3␦⑀ (8,9) heterodimers were determined by NMR and X-ray crystallography, and recent small angle X-ray scattering, EM, and NMR studies suggested their orientations relative to each other in the TCR complex. These studies showed consistently that the extracellular CD3 domains are located underneath the TCR␣␤ chains (10 -12) but disagreed on proposed docking sites of the CD3 subunits to TCR␣␤ (11,12). Although the conformation of the small extracellular portion (9 residues) of is not known, the association of the helical transmembrane segments of 2 in micelles was determined by NMR (13) and refined in molecular dynamics simulations within a lipid bilayer (14,15). A model for the assembly of all eight transmembrane domains in the intact TCR complex has been proposed based on charge-pairing interactions within the bilayer (16). Finally, a recent study observed that cytosolic juxtamembrane regions of the two -subunits are divaricated in the TCR-CD3 complex but become juxtaposed upon TCR engagement (17).
In vivo, cyt binds to the inner plasma membrane as observed with FRET for chimeric protein in which the peptide was fused to the transmembrane and extracellular regions of CD2 or KIR2DL3 (irrelevant proteins in the context of MHC-TCR signaling) (25,28). The cytosolic domain of TCR by itself is thus sufficient to transduce activation signals across the cell membrane (3, 29 -31) and has been a focus of studies of the molecular basis of T-cell signaling (25,32). However, the disposition of cyt in the membrane and its implications for kinase accessibility remain unknown. In this study, we use a planar bilayer model, the sparsely tethered lipid bilayer membrane (stBLM) (33), to investigate the membrane association of isolated cyt in thermodynamic and structural terms by surface plasmon resonance (SPR) and neutron reflectometry (NR), respectively. Molecular models based on these results visualize the localization of cyt relative to the membrane. A phosphorylation assay based upon recombinant Lck tyrosine kinase was used to study the sequestering of ITAM tyrosines within the bilayer. Taken together, these results show conclusively that the interaction of the cyt with the bilayer controls accessibility of signaling components and thus regulates an intermediate stage of signal transduction.

TCR cyt membrane association
representation of the inner cell membrane leaflet. Time-dependent SPR responses to cyt incubation of such stBLMs were recorded for protein concentrations between 0.05 and ϳ23 M. SPR measurements were performed at 25.0°C. While the bilayer membranes are in the fluid phase at 25.0°C, as they would be at physiological temperature, the entropic barrier to binding is slightly higher at physiological temperatures. Therefore, apparent binding constants observed under our experimental conditions would be expected to be somewhat reduced, but the general behavior of the protein is not expected to be qualitatively different. Fig. 3 shows representative SPR data. Binding of cyt to pure POPC stBLMs was insignificant, even at 25 mM NaCl, where electrostatic interactions are minimally shielded (Fig. 3A). Binding of cyt to POPG-containing membranes (POPC/ POPG, 60:40) leads to a fast increase of the SPR signal after each injection, followed by an approximately exponential decay of the signal to a new baseline above that of the previous, lower concentration (Fig. 3B). This indicates that protein binds to the membrane in two distinct modes: an initial fast association followed by a slow process that does not reach equilibration during the time course of the experiment. Extrapolated SPR response values, R ∞ , at the end of each incubation are well described by a Langmuir binding model with dissociation con-  (Fig. 3D), which is comparable with those measured for POPC/POPG and DOPC/ DOPS membranes. At 150 mM NaCl, binding of cyt to POPScontaining stBLMs is strongly reduced (Fig. 3E). Here, binding kinetics were considerably slowed down, low concentrations of cyt caused small decreases of the SPR response, and increases in SPR signal were only observed at high cyt concentrations (Ͼ6.7 M, n ϭ 3), yielding an estimated dissociation constant (K d ) of Ͼ25 M.

Bilayer integrity
In light of a previous report that cyt may disrupt vesicles upon incubation (23), we monitored the integrity of bilayers in contact with the protein. Negative-stain transmission EM showed that POPC/POPG (60:40; total lipid 3 mM) vesicles were not affected in their morphology by 10 M cyt (data not shown). In addition, the hydrodynamic radii of these vesicles without and with 5 M protein were indistinguishable, as reported by dynamic light scattering. As a more sensitive probe of membrane integrity, we used electrical impedance spectroscopy (EIS), which measures dielectric (insulating) properties and is a sensitive readout of bilayer quality in stBLMs (33). The EIS response has been modeled using an equivalent circuit yielding the capacitance of the intact membrane and the conductance and capacitance of parallel conductive pathways induced by protein interaction (33). Using this approach, we monitored the EIS signatures of POPC/POPS (70:30) bilayers before and after completion of an SPR experiment at which maximum surface coverage has been achieved ( Fig. 4). We typically observed an increase in membrane conductance that was Ͻ5 S cm Ϫ2 . For comparison, the introduction of a pore-forming toxin like ␣-hemolysin would lead to an increase of the membrane conductance of Ͼ100 S cm Ϫ2 , well below saturation surface coverage (35). We therefore conclude that our studies of cyt on stBLMs report protein interaction with intact membranes.

Structural characterization of membrane-bound cyt
NR is a well-established technique to determine interfacial structure in the normal direction of supported planar bilayers as time and ensemble averages of molecular conformations within the plane of the sample (36). Changes in bilayer structure upon protein incubation and the distribution of proteinacious material along the bilayer normal are thus obtained. Using NR, we probed the structure of cyt bound to stBLMs (POPC/ POPG ϭ 60:40 and POPC/POPS ϭ 70:30) in buffers that contained 25 mM NaCl. Fig. 5 shows an exemplary set of reflectivity curves for a POPC/POPS stBLM before and after the addition of 15.8 M cyt in isotopically distinct buffers. Modeling established the corresponding neutron scattering length density (nSLD) distributions shown in the inset. Tables 1-3 contain median fit values and 68% confidence limits of the parameterized one-dimensional structural profiles that give rise to these nSLD distributions for three distinct NR experiments.
In 25 mM NaCl, the distribution of cyt , as measured with protein in the buffer, occupies the hydrocarbon region and headgroups of the substrate-distal bilayer leaflet (Fig. 6, A and C). Outside the bilayer, the peptide extends away from the membrane surface up to ϳ60 Å from the bilayer center. After rinsing with protein-free buffer, we observed significantly less bilayer-associated cyt following bimodal distributions of proteinaceous material (Fig. 6, B and D) with one peak within the hydrocarbon chains and the other ϳ40 -50 Å from the membrane center.
In 150 mM NaCl, where electrostatic interactions are highly screened, the localization of membrane-bound cyt depended on sample history. Protein incubation of the DOPC/DOPS Synthetic lipids tethered to a 10-nm-thick gold film contain an oligo(ethylene oxide) segment that separates a lipid bilayer from the atomically flat gold surface, which is passivated by co-adsorption of ␤ME. The hydrated space between the ␤ME film and the bilayer avoids direct contacts between the membrane and the solid surface. This configuration is sufficiently resilient to withstand buffer exchanges by flowing solutions across the bilayer.

TCR cyt membrane association
bilayer that was previously exposed to cyt in low-salt buffer and then rinsed (Fig. 6, C and D), resulted in a distribution in which the protein on the membrane surface was essentially removed but membrane-embedded protein was unaffected (Fig. 7A). Subsequent rinsing with protein-free high-salt buffer removed some of the membrane-embedded cyt (Fig. 7B). On the other hand, incubation of a pristine POPC/POPS bilayer with 15.8 M cyt resulted in a layer of protein peripherally associated with the membrane (Fig. 7C), and rinsing with protein-free buffer removed most of this protein (Fig. 7D).
The protein distributions measured at low salt concentrations ( Fig. 6) can be interpreted as a single homogeneous protein configuration in which one region in the protein intercalates the membrane and another region extends substantially

TCR cyt membrane association
into the adjacent buffer phase. Alternatively, the same distributions would arise from two distinct protein populations of which one is membrane-inserted and a second one is peripherally associated with the membrane. The observation that rinsing with high-salt buffer removes the membrane-peripheral protein density but leaves the membrane-inserted density intact (Fig. 7, A and B) suggests that latter model is correct. This conclusion is supported by Fig. 8, in which we quantitatively compare the protein distribution profiles on POPS-containing membranes at low and high ionic strength. The difference (red trace) between the profiles in 25 mM NaCl (black) and after subsequent incubation with protein in 150 mM NaCl (green) is identical to the protein distribution profile after incubation of a pristine membrane with cyt at 150 mM NaCl (blue).

TCR cyt membrane association Accessibility of ITAMs in membrane-bound cyt to phosphorylation
To investigate the effects of membrane interaction on phosphorylation of ITAM tyrosine residues, we used recombinant Lck-G2⌬ tyrosine kinase in a phosphorylation assay in which cyt was exposed to large unilamellar POPC/POPG (60:40) vesicles (LUVs). These experiments included 6.5 mM Mg 2ϩ in the assay buffer, as described previously (19), to allow MgATP-dependent tyrosine kinase activity (37); the increase in ionic strength relative to 150 mM NaCl alone (12%) might weaken but is not likely to substantially alter the nature of the cytmembrane interaction. (Ca 2ϩ at high local concentration has been reported to promote release from the membrane of ITAM-containing cytosolic domains (37), but Mg 2ϩ interacts with acidic membrane more weakly than Ca 2ϩ (38). As shown in Fig. 9, Lck phosphorylation of cyt ITAM tyrosines is reduced as the lipid concentration is increased. This decrease cannot be attributed to a lipid-dependent reduction of Lck catalytic activity, because a control experiment with an unrelated Fyn-substrate peptide does not depend on lipid concentration. In contrast to a previous study, which showed that ITAMs on cyt bound to lysomyristoylphosphatidylglycerol (LMPG) micelles were completely protected from phosphorylation (19), we observed only a partial reduction that leveled off near 50% of the phosphorylation in the absence of lipid. Therefore, a fraction of tyrosine residues is still exposed to buffer, or there is a dynamic equilibrium between the membrane-embedded and exposed populations. No significant differences in the time courses of cyt phosphorylation were observed in the absence or presence of zwitterionic POPC LUVs (Fig. 10A), but substantial differences appeared in the presence of POPC/POPG LUVs, where phosphorylation kinetics slowed significantly at 0.5 mM lipid compared with lipid-free phosphorylation, and only very slow phosphorylation was observed for 7 mM lipid (Fig. 10B). A similar delay in phosphorylation was observed in the presence of pure POPG LUVs, but equivalent levels of phosphorylation were reached at all LUV concentrations after long exposure times to the Lck kinase (Fig. 10C).

Discussion
TCR cyt (25) and CD3⑀ cyt (21) interact with the inner leaflet of the plasma membrane. Such interactions inhibit ITAM phosphorylation by Src-family kinases (19,21) and may control the accessibility of ITAM tyrosines in the cellular context (19,21,39), providing a regulatory mechanism for signal transmission. However, this hypothesis has been controversial, and mutations designed to disrupt membrane association can in some cases inhibit rather than potentiate tyrosine phosphorylation (25-27, 40, 41). This study was designed to shed light on this long-standing problem by investigating the membrane association of cyt with well-defined membrane models and to correlate structural and functional aspects of protein-membrane association under well-controlled experimental conditions.

cyt adopts distinct conformations at the membrane
The cyt -membrane interaction has often been viewed as a simple equilibrium involving a single bound species (19,20). cyt is unstructured in solution but may adopt a helical structure in association with detergent micelles or acidic lipid vesicles, depending on detergent or lipid composition (19,20,22,23). The SPR results in Fig. 3 confirm that cyt membrane binding is contingent on acidic lipids. Not surprisingly, binding to charged membranes depends on ionic strength of the aqueous buffer because of electrostatic shielding by counterions (42). Accordingly, the dissociation constants (K d ) of ϳ4 -6 M, measured in 25 mM NaCl increase to Ͼ25 M in 150 mM NaCl, which is a lower limit because full binding isotherms could not be determined due to limited amounts of protein. This is consistent with a sucrose-loaded vesicle pull-down assay at 100 mM KCl that determined a molar partition coefficient (K) of 2,200 M Ϫ1 for an equimolar mixture of POPC and POPG and K ϭ 84,000 M Ϫ1 for 100% POPG (20). Although the interaction of isolated cyt is relatively weak, because the full-length domain is membrane-anchored and the local concentration near the membrane is therefore high, these results indicate a substantial and strong membrane interaction of the soluble domain by itself.
Examination of individual SPR traces in 25 mM NaCl after protein additions shows a slow decay that follows an initial spike (Fig. 3, B and C). Such a decay indicates a reduction of the refractive index of the interfacial layer, which may result from a degradation of the lipid bilayer (i.e. formation of water-filled defects) or bilayer thinning (i.e. rearrangement of the aliphatic chains within the bilayer upon protein adsorption), or the replacement of lipid in the bilayer by protein, which can outweigh the increase in refractive index due to peripherally bound protein. EIS results show that bilayer integrity is maintained (Fig. 4), and therefore membrane degradation is unlikely to contribute to the observed changes in SPR. We observe that bilayer thinning is minimal upon protein incubation, with an upper limit of 5% of the initial hydrocarbon thickness according

TCR cyt membrane association
to NR (Tables 1-3). This effect corresponds to an estimated reduction of 3-5 SPR response units at a protein concentration of ϳ16 M, where we observe SPR signals of Ͼ30 units (Fig. 3). Membrane thinning is therefore a component of the observed reduction in SPR response, but too small to account for much more than 10% of the observed effects. The time courses of the SPR signals therefore most likely result from slow conforma-tional changes of cyt following its adsorption to the membrane surface. We propose a process in which cyt initially binds peripherally to the membrane and converts to a second, membrane-embedded state thereafter, leading to a reduction in SPR response. NR corroborates this interpretation, as it shows proteinaceous material in two distinct distributions (Fig. 6, B and  D): one within the membrane and another that extends from  showing the same POPC/POPS ϭ 70:30 stBLM that was initially exposed to cyt at 25 mM NaCl (Fig. 6). C and D, CVO profiles from a pristine POPC/POPS ϭ 70:30 stBLM during protein exposure (C) and after rinse with protein-free buffer (D).

TCR cyt membrane association
the membrane surface into the buffer, where it is selectively removed by rinsing (Fig. 7, A and B). We conclude that bound cyt coexists in a membrane-embedded and in a membraneperipheral state at low ionic strengths.
At physiological ionic strength (150 mM NaCl), cyt interaction with the membrane is different; individual SPR traces did not show bimodal time courses (Fig. 3E), and membrane-bound protein does not penetrate the membrane, as shown by NR (Fig. 7,  C and D). We conclude that soluble cyt interacts with the membrane too weakly or resides at the membrane too briefly to form a membrane-embedded state. However, this is only true for the truncated cytosolic domain of the protein; hence, the situation may differ for full-length , whose transmembrane segment ties the soluble protein region more tightly to the membrane.
Whereas cyt inserts deeply into the bilayer under conditions where it interacts strongly with the membrane, EIS shows that bilayer integrity is not compromised (Fig. 4). In addition to functional relevance of the membrane-inserted state as a regulator of signal strength (see below), membrane insertion may also play a role in sensing or regulating local lipid curvature (43) or in intracellular sorting or trafficking (44,45).

cyt conformations are in a kinetic equilibrium
Our studies of cyt phosphorylation (Figs. 9 and 10) correlate protein structure on the membrane with function and provide insight into the kinetics of the distinct protein conformations. We considered distinct kinetic models to interpret the data. First, we assumed that ITAM tyrosines of both peripherally bound and membrane-embedded cyt populations were accessible to phosphorylation at different rate constants, such that the time course for p cyt formation is a sum of two exponentials. Alternatively, only the peripherally bound cyt population might be accessible to phosphorylation, such that p cyt follows a single exponential with a rate constant similar to that in the absence of lipids. Here, p cyt is limited by the number of exposed ITAM tyrosines, and due to inaccessibility of membrane-embedded protein, the final phosphorylation level will be lower than in the absence of lipids. Finally, we considered a dynamic equilibrium between membrane-embedded and peripherally bound cyt where only the latter is accessible to phosphorylation. However, because phosphorylation of sur-

TCR cyt membrane association
face-bound cyt reduces the interaction with acidic lipids (19,20), this step is effectively irreversible.
cyt membr -| 0 If the sum of rate constants for interconversion of the surface-associated and membrane-inserted cyt conformations (k 1 ϩ k Ϫ1 ) is similar to the phosphorylation rate (k 2 ), a time lag in the formation of p cyt is expected. Whereas our experimental phosphorylation data are at odds with the first two models, the observed delay in cyt phosphorylation in the presence of acidic vesicles (Fig. 10, B and C) supports the latter, as it is reasonable to assume that the membrane-embedded conformation of cyt is less susceptible to tyrosine phosphorylation. Therefore, it is likely that the two conformations are in a dynamic equilibrium, such that interaction of cyt with the membrane does not fully abolish phosphorylation but rather controls the reaction rate. The idea of a dynamic equilibrium between a membrane-associated and free state has also been suggested for the cytoplasmic domains of full-length CD3⑀ (41).

A molecular model of the membrane-bound conformations of cyt
Molecular models for membrane-associated cyt consistent with the NR results took three considerations into account. First, cyt is an unstructured protein in aqueous solution, but interaction with lipid vesicles can induce helicity (19,20,23); in micellar solutions, ϳ35% of the protein was ␣-helical as observed by CD spectroscopy (22). Second, regions of cyt that interact with acidic lipid headgroups have been mapped using single amino acid substitutions in the full-length (25), its cytoplasmic domain (27), and peptide fragments of cyt (23,27). These studies revealed basic-rich sequences (Fig. 11A) (19), suggesting sequestration of these residues into the hydrophobic cores. Accordingly, we designed molecular models with different levels of helicity in which tyrosine hydroxyls were sequestered by the membrane and most of the Lys and Arg side chains were in contact with phosphates or carboxylates of lipid headgroups. However, we were unable to generate membrane-embedded models consistent with the scattering profiles in which all Lys and Arg side chains of cyt were outside the bilayer. Therefore, the final models contain 4 -7 buried Arg and 4 -6 buried Lys residues. Because of the high energetic cost to immerse an arginine residue into a lipid bilayer, estimated at ϳ60 kJ/mol (46), such models might seem unfavorable. However, TCR transmembrane domains in close proximity within the intact receptor are replete with buried acid and basic residues, including Lys-Asp, Lys-Glu, and Arg-Asp salt bridges, with an excess of acidic residues (16). Moreover, computer simulations suggest that transferring multiple Arg residues into lipid bilayers is energetically nonadditive (47), so that the energetic penalty of burying multiple Arg sites might be lower than anticipated from single amino acid transfer studies and could be partially compensated by negative transfer energies for neighboring Leu, Ile, and Val residues. We did not explicitly account for any potential temperature-or salt-dependent alterations in bilayer structure or protein stability, but these are not expected to be significant for the conditions investigated.
Potential helical peptide regions were assigned, as shown in Fig. 11, based upon hydropathy scores of the Kyte/Doolittle (48) and Wimley/White (49) hydrophobicity scales as well as secondary structure analysis (50) and guided the design of three models with different helical contents (Fig. 11, C-E) with 0, 25, and 38% helicity. Scaled three-dimensional models of cyt with these properties were placed into a mixed lipid bilayer with their N-terminal residues near the membrane surface and basic-rich sequence residues preferentially in contact with lipid headgroups. Depending on whether the models were designed

TCR cyt membrane association
to conform to the NR-derived protein distribution profiles at 150 mM NaCl within or outside of the bilayer (Fig. 7, A and B), ITAMs and their neighboring residues were either membraneimmersed or membrane-peripheral, as depicted in Figs. 12 and 13, respectively. For the membrane-embedded cyt model, this implies that the central stretch of the peptide chain is located near the membrane surface, whereas overall, the protein may be anchored to the bilayer by its hydrophobic flanks (N terminus to residue 30 and residue 90 to C terminus), which may penetrate the bilayer deeply. Fig. 12 (A-C) shows representative projections of the protein models with different helical contents. NR protein distribution profiles were calculated for these hypothetical models, scaled, and compared with the experimental data. Fig. 12D shows that models of the membrane-embedded cyt population agree well with the experimental data, independent of their helical content. In the surface-associated cyt configurations (Fig. 13), the protein is also ligated to the bilayer by its hydrophobic flanks, but its hydrophilic central stretch forms loops that stretch out some 40 Å from the membrane surface, thus exposing the ITAM tyrosines to the aqueous compartment, where they are easily accessible to phosphorylation by Src family kinases. Again, reconstructed nSLD profiles of proteins with distinct secondary structures, although indistinguishable, all capture the experimental result quite well. Although these simple models are not unique, and thus not conclusive, they illustrate that peptide configurations that sat-isfy known constraints in conformational space can rationalize the NR results (Fig. 13D) and phosphorylation data. It will require more elaborate molecular dynamics simulations to quantify the underlying structures of the cyt protein near membranes more stringently, explore the energetic landscapes that govern the distinct association modes, and thus define the conditions that control interconversion between coexisting cyt conformations.

Conclusions
In this work, we presented an unconventional approach to quantify the role of protein-membrane interactions in regulating the accessibility of ITAM tyrosines, and thereby their phosphorylation, on the cytoplasmic domain of the T-cell receptor, a process that is at the root of signal transduction in T-cell signaling. Although the impact of cyt phosphorylation and its variation though membrane interactions have been extensively studied, mostly in the cellular context (25,27), consensus on these processes has not been established due to contradicting results from distinct experiments (19,26,51,52). In vitro experiments that reconstruct significant aspects of the T-cell receptor signaling cascade may help to elucidate this problem (51,52). Here we took the radical approach to study the soluble TCR cyt domain in the extremely simple but well-controlled experimental environment of tethered synthetic bilayers probed by SPR and NR. This reductionist approach necessarily engenders certain compromises: we used soluble cyt , whereas in the native protein, this domain is covalently attached to a transmembrane span; we modulated protein-lipid interaction by variation of solution ionic strength and lipid bilayer composi- Figure 12. A-C, cyt models with distinct helical contents (0 (A), 25 (B), and 38% (C)) embedded in a lipid bilayer. ␣-Helical segments are shown in red, key ITAM residues (Tyr and Leu/Ile) in magenta, and Arg/Lys side chains in blue. D, NR protein distribution profiles calculated for these models are compared with the experimental data (black) in 25 mM NaCl.

Figure 13. cyt models with distinct helical contents (0 (A), 25 (B), and 38% (C)) peripherally bound to a lipid bilayer.
Sections in the cyt -sequences arranged into ␣-helices are highlighted in red. D, NR protein distribution profiles calculated for these models are compared with the experimental data (black) in 25 mM NaCl. The color code is as in Fig. 12.

TCR cyt membrane association
tion, whereas in the physiological cellular environment, the composition of aqueous and membrane phases are complex and dynamic; and we performed SPR and NR at subphysiological temperature because of instrumental constraints. Nonetheless, despite these caveats, we were able to gain insight into basic features of the cyt -membrane interaction.
The results presented here lead to a more complex concept of cyt -membrane interactions than discussed previously. We determined the affinities of (soluble) cyt to acidic membranes to be rather low, in the micromolar range, at low ionic strength and to drop further at physiological salt concentration. Despite this relatively weak interaction, in its native environment, we expect that the cyt component of the polytopic membranespanning TCR complex will be membrane-associated, as attested to by the observation of cyt -membrane interaction in living cells (25,28). The time courses of SPR adsorption traces indicated that the protein undergoes large-scale reconfiguration following adsorption to acidic membranes. The observation that the SPR signal drops slowly after a fast initial rise suggests, after exclusion of more trivial processes, that the protein inserts into the membrane. Indeed, corresponding phosphorylation assays with Lck showed that the ITAM accessibility of soluble cyt is severely compromised in the presence of membranes. Two distinct membrane association states of cyt , one deeply inserted and one peripherally bound, were also detected and characterized with NR, and it is tempting to assume that they are in a dynamic equilibrium at the membrane, as suggested by the phosphorylation assay. Both structures of the cytmembrane complexes clearly bear the signatures of molecular dimensions: one in which the central part of the protein that holds the ITAMs is membrane-inserted and one in which it is solvent-exposed. Even if we could not discriminate between distinct models with widely varying helicities, these results show clearly that the two association states must be grossly distinct in their functional roles; whereas the ITAM tyrosines are deeply buried within the membrane in the inserted state, they are exposed and thereby accessible to kinase activity in the membrane-peripheral state.

Vesicle preparation
Definite amounts of lipids dissolved in chloroform were well mixed to obtain homogeneity of chosen lipid compositions. Chloroform was evaporated under a continuous N 2 stream and completely removed by lyophilization. Lipids were hydrated in 20 mM HEPES, pH 7.4, 150 mM NaCl and subjected to five freeze-thaw cycles. LUVs were prepared by extrusion (11 times) through a stack of two polycarbonate filters (pore size of 100 nm) using an Avanti Mini-Extruder (Avanti Polar Lipids). Vesicles were characterized by negative-stain electron microscopy and dynamic light scattering. Dynamic light scattering experiments were performed with a DynaPro-MS800 (Protein Solutions) instrument at 20°C. Samples of LUVs (2 mM) were incubated at 20°C with or without cyt (45 M) in 20 mM HEPES, pH 7.5, 25 mM NaCl; diluted (1:10,000); and filtered (Millex filter unit 0.22 M) before measurements. Scattering data were analyzed using DYNAMICS autocorrelation analysis software (version 5.25.44).
Samples analyzed by negative-stain electron microscopy were prepared by incubating LUVs (3 mM) with or without cyt (10 M) in 20 mM HEPES, pH 7.7, 150 mM NaCl, 6.5 mM MgCl 2 and applied to carbon-coated Formvar 300-mesh gold grids, stained with 1% (w/v) uranyl acetate solution. Micrographs were taken with a Philips CM10 transmission electron microscope equipped with a Gatan CCD camera system.

Surface plasmon resonance
SPR measurements were conducted at 25 Ϯ 0.01°C in a single-batch set-up using a custom-built instrument (SPR Biosystems, Germantown, MD). SAM-covered, gold-coated glass slides were assembled in the Kretschmann configuration by index-matching to a prism. stBLMs were completed by vesicle fusion in situ, as described above. The setup allows for simultaneous SPR and electrochemical impedance spectroscopy (EIS) measurements, which are used to assess the quality of the stBLMs before and after protein is introduced. In the singlebatch experiments, the untreated bilayer is measured first to determine a baseline before adding protein in increasing concentrations, c P , and measuring the equilibrium SPR response, R eq . R eq (c P ) was fitted to the Langmuir isotherm to determine the protein affinity in terms of the dissociation constant (K d ) and the surface density (R ∞ ) of bound protein extrapolated to infinite concentration c P .

Electrochemical impedance spectroscopy
EIS data were taken using a Solartron (Farnborough, UK) system (model 1287A potentiostat and model 1260 frequency response analyzer). Gold-coated glass slides served as the working electrode in a setup that allowed simultaneous SPR and EIS measurements. The cell has a volume (V) of ϳ250 -300 l and a surface area (A el ) of ϳ0.33 cm 2 confined by a Viton O-ring. Copper contrast was used to measure the geometric electrode surface area (53), and raw EIS data were normalized to A el and a roughness factor (␤) of 1.4, estimated from the gold surface oxidation/oxide stripping charge (54). A saturated silver-silver chloride (Ag͉AgCl͉NaCl(aq,sat)) microelectrode (Microelectrodes, model M-401F) was used as reference. The auxiliary electrode was a 0.25-mm diameter platinum wire (99.99% purity; Sigma-Aldrich) coiled around the barrel of the reference electrode. The distance between the tip of the reference and working gold electrode surface was set to 2-3 mm. Measurements were carried out in aerated solutions with 10-mV alternating current at 0-V bias versus reference electrode. Data analysis was carried out by fitting to an equivalent circuit using ZView (Scribner Associates), as described (33).

Neutron reflectometry
NR measurements were performed at 22 Ϯ 2°C on the NGD-MAGIK reflectometer at the NIST Center for Neutron Research (55). Reflectivity curves were recorded for momentum transfer values 0.008 Å Ϫ1 Յ q z Յ 0.25 Å Ϫ1 . stBLMs were prepared on 3-inch diameter silicon wafers assembled in a flow cell (56). Samples were measured at room temperature using three isotopically different bulk solvents of distinct H/D contrasts (D 2 O, H 2 O, and a 2:1 (v/v) mixture of the two), with adequate counting statistics typically obtained after ϳ6 h. The flow cell allows for in situ buffer exchange; therefore, subsequent measurements were performed on the same sample area. NR data were first collected from an as-prepared stBLM. Thereafter, protein was added to the stBLM, and NR data were collected either during protein incubation or after rinsing the bilayer with buffer.
To characterize the structures of proteins associated with the membrane, one-dimensional nSLD profiles along the lipid bilayer normal were determined by fitting to the experimental NR data. These one-dimensional structural profiles were composed using a hybrid of a stratified slab model for the solid substrate (57), a continuous distribution of lipid components for the stBLM (58), and a monotonic Hermite spline for the model-free envelope of the protein (36). In this compound model, bulk silicon, silicon oxide, a chromium bonding layer (ϳ50 Å), and the substrate-terminal gold layer (ϳ150 Å) were parameterized as subsequent slabs in terms of thickness and nSLD for each layer, except for the bulk silicon, for which the nSLD is known. One global roughness parameter was applied to all substrate interfaces. Individual submolecular groups implemented in the continuous distribution model of the stBLM are as follows: ␤ME, tether PEG chains, tether glycerol groups, substrate-proximal and substrate-distal PC and PS headgroups, and substrate-proximal and substrate-distal methylene and methyl chains of lipid and tether molecules. Fit parameters are the bilayer hydrocarbon thickness for each bilayer leaflet, bilayer completeness, tether surface density, tether thickness, ␤ME surface density, and one roughness parameter applied to all of these distributions.
The Hermite spline that encodes the protein distribution profile is defined by control points that are on average 20 Å apart. The number of control points depends on the spatial extension of the protein along the bilayer normal and is determined iteratively during model optimization. Fit parameters for each control point are the volume occupancy of protein and the deviation from a position defined by equidistant control points throughout the spline. A constant nSLD is associated with the spline functions that thus describe the contributions to the scattering of either completely protiated or partially deuterated, intrinsically homogeneous protein material.

TCR cyt membrane association
Data modeling was performed with the ga_refl and Refl1D software packages (56) using a differential evolution Markov chain global optimizer. Reflectivity curves recorded from one sample under different conditions were fitted simultaneously to the same model by sharing fit parameters, for example, for the solid substrate. The recorded Monte Carlo Markov chain was used to determine the fit parameter confidence limits using a statistical analysis. This procedure yields a bias-free and objective estimate of the uncertainties of the resulting nSLD profiles while avoiding overparameterization (36).

Molecular modeling
Two amino acid residues (Gly 1 and Ser 2 ) were N-terminally appended to the primary sequence of cyt to account for the thrombin recognition site in the recombinant protein. Residue numbering used here refers to the cytosolic domain of , such that residue Leu 51 in the native full-length protein (with signal sequence) is considered number 3. The hydropathy score and the secondary structure prediction were determined using ExPASy ProtScale (59) and PSIPRED version 3.3 (50), respectively. Molecular models were derived from an initial ideal helix model, which was interspersed with perturbations to maintain helical segments at the desired positions. Three-dimensional conformations of these initial models were adjusted using Lego tools in the O software environment (60) to reflect local conformations found in a library of high resolution protein structures. Final protein models were used to form a protein/membrane system by using the CHARMM-GUI Input Generator (61). Protein structures were first preoriented relative to the membrane normal and then surrounded by lipid-like spheres whose positions were subsequently replaced by lipid molecules (replacement method). A rectangular protein/membrane system was formed choosing DOPC/DOPS molecules in a molar ratio of 70:30.
Author contributions-K. Z., R. E., F. H., S. R., B. J., and P. S. conducted experiments and analyzed results. K. Z. generated three-dimensional protein models. K. Z., F. H., M. L., and L. J. S. wrote the paper. M. L. and L. J. S. conceived the idea.