key: cord-0034383-gb9727k9
authors: Zhang, Wenyi; Sato, Takeshi; Smith, Steven O.
title: NMR spectroscopy of basic/aromatic amino acid clusters in membrane proteins
date: 2006-07-30
journal: Prog Nucl Magn Reson Spectrosc
DOI: 10.1016/j.pnmrs.2006.04.002
sha: 393968f7d087a3bcaf78f58435fa1e99c6db2d1d
doc_id: 34383
cord_uid: gb9727k9

nan

Helical integral membrane proteins account for 20-30% of the proteins encoded in genomes ranging from archaea to eukaryotes [1] . When combined with proteins that associate with membranes, the proteins that interact with the internal and outer membranes in cells may well be close to half of all cellular proteins. Like integral membrane proteins, membraneassociated or peripheral membrane proteins have a broad range of cellular functions including catalyzing enzymatic reactions, mediating cell signaling and energy transduction, and transporting molecules to and from cell surfaces.

The membrane bilayer, with a hydrophobic interior and a polar interfacial region containing the lipid headgroups, creates a unique environment that dramatically influences the way membrane proteins fold and function. The architecture of most integral membrane proteins is a-helical due to the hydrophobic nature of the bilayer. Howe+ver, within this constraint the helices have evolved to often be flexible, associate in specific orientations, and to create internal hydrophilic pores and cavities. Membrane-associated proteins have fewer constraints on the type of structures they can adopt, and include proteins that are unstructured in their native state. The determination of the structures of simple transmembrane helix dimers by solid-state NMR [2] [3] [4] over the past few years has provided insights into how hydrophobic helices can specifically associate in membrane environments. The striking observation is that small and weakly polar amino acids (Gly, Ala, Ser and Thr) have high propensities for mediating helix-helix contacts [5] . When these amino acids line a helix interface, very short inter-helical distances result.

One common strategy for specific helix association appears to involve the formation of inter-helical hydrogen bonds upon the close approach of such helices.

The greater variety of structures that are possible for membrane-associated proteins has made it more challenging to discern the mechanisms these proteins have evolved for mediating specific protein-protein and protein-lipid interactions. Clusters of basic and aromatic amino acids are one common way that membrane associated proteins interact with cellular membranes. These clusters are enormously varied, ranging from highly basic sequences (as in the effector domain of the MARCKS protein) to highly aromatic sequences (as in the scaffolding domain of caveolin). However, it is striking how often aromatic residues, rather than large hydrophobic amino acids, seem to be critical for a specific mode of membrane interaction.

At first glance, the mechanism is simple for how membraneassociated proteins and peptides composed of basic and aromatic amino acids interact with membrane bilayers. The positively charged amino acids (Arg, Lys, His) bind to membrane surfaces through non-specific electrostatic interactions. Aromatic amino acids (Trp, Tyr, Phe) allow the membrane-associated peptides to penetrate the membrane surface. A closer look into how basic-aromatic membrane peptide sequences influence membrane structure suggests that peptide-lipid interactions can be both varied and complex. Besides simply binding to membrane surfaces, membraneassociated peptides can mediate membrane fusion, induce curvature, form transient pores and stable channels, cross membranes in a receptor-independent manner, or completely disrupt the bilayer structure.

NMR approaches are just beginning to answer several key questions involving the structure and function of basicaromatic clusters. For instance, are there common motifs that determine how strongly basic-aromatic clusters bind to cell membranes, how deeply they penetrate the membrane surface or whether they locally disrupt bilayer structure? What are the structural and functional differences between the aromatic amino acids? Similarly, what are the differences between the basic amino acids? What are the differences between the aromatic and large hydrophobic amino acids in mediating membrane interactions? How does the juxtaposition of residues in specific sequences confer specific structures and function?

In this review, we first describe the myriad of ways that basic-aromatic sequences can interact with membrane bilayers. NMR spectroscopy is one of the best approaches to study these sequences in their native membrane environment. We discuss the NMR methods that have been used to investigate the structures of membrane-associated proteins and the structure and dynamics of the membrane bilayers that these peptides interact with which they interact. An excellent review of NMR structural methods using magic angle spinning and specific applications to membrane-associated peptides has recently appeared in this series by Huster [6] . Fig. 1 presents in a cartoon form some of the diverse ways peptides with basic-aromatic clusters are thought to interact with membrane bilayers. These clusters can be contained within integral membrane proteins, membrane-associated proteins or short anti-microbial peptides. The aromatic amino acids in basic-aromatic clusters may (i) contribute hydrophobic energy to the binding of the cluster with membranes (ii) enhance the electrostatic potential of the basic amino acids, (iii) influence membrane curvature, and (iv) locally disrupt membrane bilayer structure. In this section, we present an overview of the major ways that basic-aromatic sequences have been found to interact with membranes. There are common themes that run throughout the examples presented (e.g. membrane penetration, non-specific electrostatic interactions), and often the challenge is to discern how subtle differences in protein structure may lead to very different membrane interactions.

Many cellular proteins have specialized modular membrane targeting domains that serve to recruit cytosolic proteins to the Fig. 1 . Cartoon illustrating possible modes of interaction of basic-aromatic amino acid clusters (represented by shaded ellipsoids) with membrane bilayers. Basicaromatic cluster can (a) bind to membrane surfaces by non-specific electrostatic interactions (b) insert into the membrane interface or penetrate into the acyl chain region of the bilayer, (c) locally disrupt the bilayer structure (d) form transient pores and stable channels, or (e) mediate membrane fusion. membrane surface [7, 8] . Some of the best characterized of these peripheral membrane proteins are involved in cell signaling. These include the pleckstrin homology (PH) domain [9, 10] , the Phox (PX) domain [11] , the protein kinase C conserved 1 (C1) and conserved 2 (C2) domains [12] , and the FYVE domain [13] . Basic and aromatic amino acids are common elements in these domains responsible for specific membrane binding and penetration. A general model for protein-lipid interactions in these systems is one where non-specific electrostatic interactions mediate the initial contact with the membrane, and hydrophobic and aromatic amino acids penetrate into the membrane interface. Many of the domains recognize specific lipids, which upon binding induce a conformational change in the domain that releases conserved aromatic residues buried within the protein for membrane penetration.

For example, membrane binding of the FYVE domain involves aromatic residues that allow penetration of the membrane surface and basic amino acids that mediate both non-specific electrostatic interactions with the membrane and specific lipid binding [14, 15] . NMR studies of the FYVE domain have shown that a critical loop containing hydrophobic and aromatic residues, undergoes a conformational change upon binding the phosphoinositide, PI(3)P [16] . Structural, mutational, and binding studies show that a conserved phenylalanine in this loop is buried within the protein and is forced out of its binding pocket to a position favorable for membrane insertion upon lipid binding.

A protein conformational change induced by membrane binding has also been proposed by Tuzi et al. [17] for the PLCd1 PH domain on the basis of solid-state NMR measurements. Conceptually, the model is similar to that of the FYVE domain. Even though PH domains do not have a deep ligandbinding pocket [9, 10] , a buried phenylalanine becomes exposed to the membrane bilayer upon binding of the phosphoinosotide PI(4,5)P 2 .

Binding of both the C1 and C2 domains of protein kinase C (PKC) are required for enzyme activity. The C1 domain follows a two-step membrane-binding mechanism in which the initial membrane adsorption by non-specific electrostatic interactions is followed by membrane penetration. The domain has a polar binding pocket for diacylglycerol that is surrounded by hydrophobic and aromatic residues, which itself is surrounded by a ring of basic amino acids. An NMR study of the PKCg C1B domain by Xu et al. [18] has shown that the hydrophobic and aromatic residues surrounding the DAGbinding pocket penetrate the membrane.

C2 domains are unique among membrane targeting domains in that they do not have a well-defined lipid-binding pocket and thus show relatively weak lipid specificity. Using EPR spectroscopy, Frazier et al. [19] have shown that the first and third Ca 2C binding loops of the C2 domain of cPLA2 penetrate as deeply as 15 Å below the lipid phosphates. These loops contain both hydrophobic (Met, Leu, Val) and aromatic (Phe, Tyr) residues. In this case, the membrane docking of the cPLA2 C2 domain is thought to be driven primarily by hydrophobic rather than electrostatic interactions. In contrast, the PKCR C2 domain interacts with the negative charges of the phospholipids. Kohout et al. [20] have shown that the Ca 2C binding loops insert directly into the membrane interface, and that an anion binding site on b-strands three and four lies near the phosphate head groups.

All of the domains mentioned above have a well-defined secondary structure. There are also many peripheral membrane proteins whose unstructured sequences bind to cell membranes through non-specific electrostatic interactions [21] . One representative example is the myristoylated alanine-rich C kinase substrate (MARCKS) effector domain. McLaughlin and co-workers [22, 23] have shown that the MARCKS effector domain (residues 151-175) binds strongly to negatively charged membranes and sequesters PI(4,5)P 2 . Phosphorylation of the MARCKS effector domain releases it from the membrane as part of a mechanism for regulating free PI(4,5)P 2 levels in the plasma membrane.

Basic-aromatic motifs are common in the juxtamembrane regions of integral membrane proteins as well. For example, McLaughlin and Murray [24] have postulated that unstructured regions of the NMDA receptor bind to membranes and laterally sequester PI(4,5)P 2 by the same non-specific electrostatic mechanism as found for the MARCKS effector domain. They propose that the basic-aromatic cluster may participate in closing the NMDA channel via a mechanism involving the binding of Ca 2C -calmodulin. As a result, it appears that many of the same functional roles that basic-aromatic clusters have in peripheral membrane proteins may exist in integral membrane proteins.

The peripheral membrane proteins described in Section 2.1 generally bind through non-specific electrostatic interactions and penetrate the membrane surface. The depth of penetration is often to the hydrocarbon core of the membrane. Membrane penetration is the hallmark of a series of small peptides that have been shown to cross through cell membranes without the aid of receptors [25, 26] and without forming stable pores. The general model for membrane penetration by these peptides involves non-specific binding followed by local disruption of the membrane. These peptides have gained considerable attention for their potential in drug delivery. They have been identified in normal cellular proteins, such as penetratin or the Antennapedia homeodomain [27] [28] [29] [30] [31] , and have been designed de novo [32] .

Many of the peptides that have been identified as cell permeant appear to cross cell membranes by endocytosis [33] . These peptides bind to membrane surfaces electrostatically and are rapidly internalized. Internalization by endocytosis allows these 'carrier' peptides to transport cargo ranging from peptides to DNA [34, 35] . While it remains controversial whether all of the putative cell permeant or penetrating peptides cross bilayers by endocytosis [33] , there does appear to be good evidence that some peptides are able to cross membranes by direct penetration of the hydrophobic core of the bilayer.

One of the first peptides shown to cross membranes in a receptor independent manner corresponds to amino acids 43-58 of the Antennapedia homeodomain protein Antp, a transcription factor found in Drosophila. In vivo and in vitro studies have shown that both basic and aromatic residues are crucial for the Antp peptide to be internalized within cells [27] [28] [29] [30] [31] . Importantly, when two tryptophan residues (Trp48 and Trp56) in the Antp (43) (44) (45) (46) (47) (48) (49) (50) (51) (52) (53) (54) (55) (56) (57) (58) sequence are substituted with phenylalanine, the mutated peptide is no longer cell permeant [25] . Such a subtle change in sequence would not be expected to influence membrane permeation by endocytosis.

A revealing example of a membrane peptide that is capable of crossing cell membranes is the MARCKS effector domain peptide. MARCKS(151-175) alone is not cell permeant. However, McLaughlin and co-workers found that labeling MARCKS(151-175) with Texas-Red, a large hydrophobic fluorescent label, allows the peptide to rapidly penetrate the outer membranes of giant vesicles [36] . As with the other cell permeant peptides, the detailed mechanism is not known [28, 37] . However, Cunningham et al. [38] observed that attaching a hydrophobic fluorescent probe, rhodamine B, to a basic peptide corresponding to a region of gelsolin allows the peptide to cross the cell membrane as well. In contrast, attaching a hydrophilic fluorescent label (Alexa-488) to the MARCKS(151-175) peptide does not confer cell permeation. It is important to note that many of the studies described in this review make use of large hydrophobic fluorescent labels to track the location of peptides with basic-aromatic clusters. These labels are generally aromatic systems, much like tryptophan, and the observations on their ability to facilitate membrane permeation emphasizes the importance of aromatic residues in basic-aromatic clusters, as well as raising a concern about how these fluorescent tags influence the behavior of the labeled peptide.

The ability of basic-aromatic sequences to mediate membrane binding and to penetrate into the hydrophobic membrane core appears to be widespread in normal cellular function. Many of the anti-microbial peptides represent the 'dark-side' of the basic-aromatic motif. There are literally hundreds of small anti-microbial peptides whose structure and activity have been characterized [39] . A variety of different mechanisms have been implicated in how these peptides cause cell death [40] [41] [42] [43] . While some interact with specific protein targets, most appear to directly interact with the cell membrane to form well-defined ion channels, transient pores, sinking rafts, or generally disrupt the integrity of the bilayer.

Gramicidin A is by far the most comprehensively studied anti-microbial peptide. The structure of gramicidin A in oriented lipid bilayers has been obtained by solid-state NMR [44] . The b-helical structure of alternating L-and D-amino acids orients the backbone carbonyls toward the center of an aqueous channel. It takes two monomers of gramicidin A to form a channel, each spanning half of the membrane and connected in a head-to-head geometry. There are no basic residues in gramicidin. However, four tryptophans in the C-terminal half of the peptide serve to orient the channel with respect to the lipid bilayer [45] , stabilize the channel conformation [46] and facilitate cation conductance [47] .

Many of the anti-microbial peptides are amphipathic helices [39, 48, 49] where the detailed mechanism of action is less certain. Basic residues are a common feature of these peptides, while aromatic residues are relatively rare [39] . Bechinger [41] has summarized in an elegant fashion many of the different models proposed for how amphipathic helices may function as anti-microbial agents.

Alamethicin is the prototypical amphipathic helix that functions by forming ion channels through the so-called barrel-stave mechanism [50] . This multistep mechanism involves (a) surface binding of the amphipathic helix, (b) oligomerization of peptide monomers, and (c) insertion of the oligomer into the membrane. Many other peptides have been found to form channels by this mechanism, such as magainin 2 [51] and paradaxin [52] . The N-terminus of paradaxin Pa4 is non-helical and has two consecutive phenylalanines that may initiate insertion by penetrating into the membrane core [52] .

The sinking raft mechanism has been proposed to explain how d-lysin, a 26-residue amphipathic helical peptide, is thought to function [53] . Pokorny and Almeida have observed that these peptides are excluded from liquid-ordered phases (rafts) in model membranes and are locally concentrated in liquid-disordered phases [53] . We discuss below in more detail how basic-aromatic clusters may be involved in forming microdomains in membranes. However, in the case of d-lysin the high local concentration of peptide that results from exclusion from liquid-ordered domains leads to peptide association and the formation of transient pores as the peptide 'rafts' sink into the membrane.

There is a large class of anti-microbial peptides with basicaromatic sequences that do not form amphipathic helices. For example, protegrin-1 (NH 2 -RGGRLCYCRRRFCVCVGR-NH 2 ) is an 18-residue anti-microbial peptide that forms an anti-parallel b-sheet in solution where the two strands are stabilized by two disulfide bonds [54, 55] . This disulfidestabilized b-hairpin motif is common to a number of other anti-microbial peptides, such as human defensins and tachyplesin [56] . 1 H spin diffusion experiments reveal close contacts of the CaH of Leu5 in protegrin-1 with both the terminal u-CH 3 group of the lipid acyl chains and lipid headgroup suggesting that the peptide locally disrupts the bilayer [57] .

Subtilosin A is an interesting anti-microbial peptide that has a net negative charge. The NMR solution structure of the peptide shows that a critical C-terminal tryptophan (Trp34) is in close proximity to the N-terminal lysine (Lys2) generating the minimal basic-aromatic cluster [58] . Fluorescence and NMR studies show that upon membrane binding, subtilosin A adopts an orientation in which the tryptophan-edge of the peptide is buried within the hydrophobic core of the membrane [59] . For subtilosin A, aggregation is thought to be involved in permeabilizing bacterial membranes. As a result, a cluster of basic-aromatic residues may be formed from the association of several peptides.

A final example involves two hexapeptide sequences, which were identified through combinatorial methods as having high anti-microbial activity [60, 61] . Interestingly, these peptides are not thought to disrupt membranes or form channels, but to target sites within bacteria. Vogel and co-workers found that when fluorescently tagged, the hexapeptides (Ac-FRWWHR-NH 2 and Ac-RRWWRF-NH 2 ) rapidly pass across the outer and inner membranes of E. coli and S. aureus. The remarkable feature of these peptides is the minimal sequence required for anti-microbial activity.

The wide array of mechanisms that have been proposed for anti-microbial peptides has generally highlighted how the specific amino acid sequence, the peptide concentration and the type of membrane can influence membrane interactions. For example, bacterial membranes have higher concentrations of anionic lipids than mammalian plasma membranes, and the ability of anti-microbial peptides to selectively disrupt membrane structure in many cases depends on this difference. Studies on magainin 2 show that the basic-aromatic clusters are only capable of inducing leakage across liposome membranes containing high concentrations of negatively charged phosphatidylglycerol [62] .

Section 2.3 discussed the ability of anti-microbial peptides to locally disrupt bacterial membranes. Membrane fusion is a normal process in mammalian cells where membrane associated peptides locally disrupt bilayers without causing cell death. Membrane fusion is involved in endo-and exocytosis, membrane recycling, fertilization, neuronal transmission and viral fusion [63] . The general model for membrane fusion involves the insertion of hydrophobic segments into the two membranes to be fused (e.g. vesicle membrane and plasma membrane). These segments often contain basic-aromatic clusters that are thought to facilitate fusion by locally disrupting the bilayer structure and countering the electrostatic interactions of negatively charged membranes.

Tryptophan and basic residues are abundant in the juxtamembrane regions of transmembrane SNARE proteins suggesting that these basic-aromatic clusters are involved in the general mechanism for membrane fusion. Using sitespecific spin labeling and EPR spectroscopy, Shin and co-workers have shown that a basic-aromatic cluster in synaptobrevin at the C-terminal end of its helical membrane spanning segment contains two tryptophan residues that are inserted deeply into the acyl chain region of the bilayer [64] .

Membrane fusion is also a necessary step (and studied extensively) in the entry of enveloped viruses. The viral envelope proteins that mediate fusion have two regions that interact with cell membranes, an N-terminal fusion peptide rich in glycines and aromatic amino acids, and a single transmembrane domain connected to a juxtamembrane region containing a basic-aromatic cluster. The transmembrane segment anchors the envelope protein to the viral membrane, while a fusion-induced conformational change allows the fusion peptide to insert into the host cell membrane [65] . Crystal structures of the soluble ectodomains of several fusion proteins [66] suggest that the long hydrophilic sequence intervening between these hydrophobic segments serves to bring the host and viral membranes into close proximity. In both the human immunodeficiency virus (HIV) type I [67] and the coronavirus [68] , the aromatic-rich juxtamembrane region has been implicated in the fusion of viral and host cell membranes. The predicted structure of these fusion proteins at the transmembrane-juxtamembrane boundary is remarkably similar to the mammalian SNARE proteins. Solid-state and solution NMR studies on fusion peptides derived from the HIV-1 gp41 protein [69] and the influenza hemaglutinin protein [70] have suggested how membrane binding may lead to local disruption of the host cell membrane.

One component of the fusion process is the ability of fusion proteins to induce curvature in the bilayer structure of membranes. There is a similar requirement for curvature in the formation of membrane vesicles. For example, endocytic clathrin-coated vesicles are one of the best characterized systems that have highly curved membranes [71] . There are several clathrin binding proteins (epsin, amphiphysin, and endophilin) that have the ability to induce curvature in membrane vesicles without clathrin [72] . Epsin induces curvature via an amphipathic a-helix that inserts into the cytosolic leaflet of the lipid bilayer via a PI(4,5)P 2 -dependent mechanism [73] . Amphiphysin and endophilin induce curvature primarily through electrostatic interactions between the positively charged concave surface of a Bin/amphiphysin/Rvs (BAR) domain and the negatively charged membrane. BAR domains form 'banana-shaped' dimers with clusters of basic residues on the concave surface [74, 75] . Interestingly, several BAR domains have an amphipathic helix at their N-terminus, which increases their ability to induce membrane curvature [76] . Lee et al. [77] have recently shown that the N-terminal amphipathic a-helix (MAGWDIFGWFRDVL) of Sar1p inserts into membranes and induces synthetic vesicles to form narrow tubules. Sar1p is a small GTPase and one of five core COPII proteins in COPII-coated transport vesicles. Substitution of the Trp or Phe residues with alanine within the amphipathic a-helix of Sar1p blocks its ability to generate highly curved membranes.

In Section 2.1, we described how the binding of peptides containing clusters of basic and aromatic residues are able to sequester highly negatively charged lipids like PI(4,5)P 2 . The ability of these clusters to locally sequester poly-phosphoinosotides is important in cell membranes since these lipids are involved in many signal transduction pathways [78] [79] [80] . For example, the formation of small microdomains enriched in PI(4,5)P 2 by binding of the MARCKS protein provides a mechanism for regulating the free PI(4,5)P 2 levels in membranes.

The concept of cell membranes composed of microdomains enriched in different lipid and protein components has slowly replaced the fluid-mosiac model of membrane bilayers proposed by Singer and Nicolson [81] . For example, over the past 10 years membrane domains enriched in cholesterol have been implicated in cellular processes ranging from signal transduction to generating cell surface polarity [82] . These cholesterol-rich lipid rafts appear to segregate and concentrate membrane proteins [83] [84] [85] . Proteins with basic-aromatic clusters have been found to bind strongly to cholesterolcontaining rafts raising the question of whether these proteins serve to nucleate microdomain formation in cells or whether existing lipid rafts tend to recruit proteins with basic-aromatic clusters [86] . Anderson has suggested that cholesterol-rich domains can form in membranes as a consequence of preferential sequestration of particular lipids as a shell surrounding membrane bound proteins [87] . For example, he suggests that when the MARCKS effector domain binds to membranes and recruits negatively charge lipids, these lipids associate with cholesterol to create a lipid shell.

Caveolae are the prototypical membrane microdomain enriched in cholesterol. These invaginations in cellular membranes require caveolin, a 178 residue membrane protein, for formation. Caveolin has a short hydrophobic segment, which is thought to form a helical hairpin in membranes, and a basic-aromatic cluster termed the scaffolding domain (residues 82-101). The scaffolding domain is essential for both caveolin oligomerization and the interaction of caveolin with other proteins. The domain contains an unusual six residue basic-aromatic sequence, KYWFYK. While the mechanism of cholesterol sequestration by caveolin is not understood, this basic-aromatic sequence containing tyrosine is part of a general cholesterol recognition consensus (CRAC) motif identified by Li and Papadopoulos [88] . Epand and co-workers [89] found from magic angle spinning (MAS) NMR that peptides corresponding to the caveolin scaffolding domain insert into the interfacial region of membrane bilayers and that the insertion is deeper when cholesterol is present.

Other proteins containing the CRAC motif include, apolipoprotein [88] and the HIV-1 gp41 fusion protein [90, 91] . Epand and co-workers [86] have also studied a peptide containing a CRAC-like motif from NAP22 using 1 H NOESY-MAS-NMR. They found that this motif induces formation of a cholesterol-depleted domain in membranes and that this domain is abolished when the sole aromatic amino acid, Tyr11, is replaced with leucine. These results suggest that there is a direct interaction between the aromatic amino acid tyrosine and cholesterol.

The gp41 protein from HIV-1 also contains a CRAC motif (centered on the sequence LWYIK). The basic-aromatic cluster in gp41 exhibits many of the properties discussed in the subsections above. The gp41 cluster promotes membrane binding and interfacial penetration, membrane fusion, sequestration of cholesterol and negatively charged lipids, and induces membrane curvature. An important question is whether these properties are simply a manifestation of the same property inherent in any basic-aromatic cluster or whether they are independent. Nieva and co-workers [92] addressed this question by comparing the juxtamembrane sequence of HIV-1 gp41 (DKWASLWNWFNITNWKWYIK) with a sequence where the first three tryptophans are replaced by alanines. They found that both peptide sequences bind to and penetrate the membrane surface (via the WKWYIK sequence), but that the Trp-to-Ala mutations abolish the amphipathic helical secondary structure of the peptide, along with its ability to oligomerize and mediate membrane fusion.

The tyrosine residue in the CRAC motif of other viral fusion proteins is not conserved [90] suggesting that this motif itself is not strictly required for cholesterol sequestration. In fact, simple insertion of aromatic residues may be a general mechanism of sequestering cholesterol. This can be done for unstructured peptides binding to membrane surfaces or in the context of transmembrane helices. For example, Killian and co-workers [93] have found that tryptophan at the end of transmembrane helices can interact with cholesterol. They observed that the formation of an isotropic lipid phase resulted from a synergistic effect between a tryptophan-containing transmembrane peptide and cholesterol.

Taken together, the studies described above illustrate that the behavior of basic-aromatic clusters depends on a number of factors including the secondary structure, the local concentration of the cluster, formation of oligomers, and the nature of the membrane.

Solid-state NMR spectroscopy is well suited for high resolution structural studies of membrane proteins attached to or embedded in membrane bilayers [94] [95] [96] . In order to address the structure and function of basic-aromatic clusters and how they influence cell membranes, these studies most often rely on the use of membrane bilayers or bicelles rather than detergent micelles. The first part of this review described different membrane protein-lipid interactions, emphasizing the often subtle relationships between membrane protein structure (conformation and oligomerization) and lipid bilayer structure. In this section, we summarize the use of 1 H high-resolution magic angle spinning (MAS) NMR for studying membraneassociated proteins and 13 C and 15 N MAS NMR for studying integral membrane proteins and peptides. 31 P and 2 H NMR methods and applications are described for characterizing the structure and dynamics of the membrane bilayer.

Magic angle spinning (MAS) is a solid-state NMR technique used for obtaining high-resolution NMR spectra of membrane proteins and peptides [94, 97, 98] . Protons in solid samples give broad and featureless lines because of strong 1 H-1 H dipolar couplings, which cannot be averaged to zero by MAS at usual spinning speeds. However, membrane lipids exhibit a high degree of rotational and segmental motion when in the liquid crystalline phase, which partially averages the proton dipolar couplings, which can be reduced further by MAS. Oldfield and co-workers were the first to fully recognize that high resolution 1 H spectra of lipid membranes could be obtained using MAS [99] , while Cafiso and co-workers pioneered the use of NOESY methods combined with MAS to investigate the structure of lipid membranes [100, 101] . The high resolution 1 H NOESY-MAS approach has subsequently been extended to characterize the location of a wide range of small hydrophobic molecules in membrane bilayers [102, 103] and to investigate the structure and location of both membraneassociated and transmembrane peptides [104] [105] [106] [107] [108] [109] .

The 2D 1 H MAS-NOESY spectrum of a simple basicaromatic model compound, phenylalanine methyl ester, bound to the surface of a lipid bilayer comprised of a 3:1 mixture of DMPC and DMPG is presented in Fig. 2a [107] . This small molecule has a single aromatic ring and a single positive charge on its unprotected amino terminus (Fig. 2c) . The hydrophobic phenylalanine side chain has a favorable free energy for partitioning into the bilayer head group region, while the N-terminal positive charge should be attracted to, but not penetrate, the negatively charged lipid bilayer surface [110] . The spectrum is dominated by the proton resonances from the DMPC and DMPG lipids. These lipids at a molar ratio of 100:30 provide a net bilayer surface charge comparable to that of the cytoplasmic leaflet of typical mammalian plasma membranes.

In the 2D spectrum shown in Fig. 2 , the 1 H resonances are observed as contours. The resonances on the diagonal correspond to the isotropic 1 H resonances that would appear in a conventional 1D NMR spectrum. The most intense diagonal peak at w1.3 ppm corresponds to the methylene protons of the lipid acyl chains. Using a mixing time of 300 ms, cross peaks are observed between most of the proton resonances that lie on the diagonal, and their intensities in a NOESY spectrum depend on through-space dipolar couplings. The dipolar couplings and corresponding cross peak intensities are strong for protons that are in close proximity to one another. In Fig. 2 , the proton resonances are dominated by DMPC, the most abundant lipid.

One-dimensional rows taken from the 2D spectrum provide a qualitative ruler for establishing the depth of peptides in the lipid bilayer [107] . If the peptides are buried in the hydrocarbon interior of the bilayer, the peptide protons exhibit cross peaks predominantly to the protons of the lipid acyl chains, whereas if the peptides are bound in the polar head group region of the bilayer, they exhibit cross peaks predominantly to the protons of the glycerol backbone and choline head group. Peptides that do not insert into the interfacial region of the bilayer, e.g. poly-lysine, exhibit cross peaks to only the phosphocholine head group. This approach assumes that the mixing times are sufficiently short to limit spin diffusion, and that molecular motions of the peptide and lipid allow efficient cross relaxation.

In order to illustrate how the cross peak intensities change with mixing time, Fig. 3 shows rows from the 2D MAS-1 H NOESY spectra of phenylalanine methyl ester obtained with mixing times of 50 (a) and 300 ms (b). The rows were taken through the diagonal resonance of the aromatic phenylalanine protons at 7.25 ppm. Based on equilibrium dialysis of model tripeptides with a charged N-terminus and protected C-terminus, White and Wimley [110] estimated that the free energy for partitioning of the aromatic side chain of phenylalanine into the head group region of 1-palmitoyl-2-oleoyl-snglycero-3-phosphocholine (POPC) is K4.73 kJ/mol. In these model tripeptides, positively charged amino acids, e.g. lysine and arginine, have free energies of w4.19 kJ/mol for partitioning into the head group region of POPC. Energetically these results suggest that the positive charge on the phenylalanine methyl ester model compound should not penetrate the phosphocholine region of the bilayer.

In the 50 ms spectrum (Fig. 3a) , the most intense cross peaks correspond to the a-protons (3.7 ppm), b-protons Fig. 2b for resonance assignments). The resonance at 3.4 ppm marked with an asterisk results from the b-CH 2 group of the phenylalanine side chain. The short (50 ms) mixing time limits spin diffusion and the spectrum indicates that the aromatic side chain is located in the region of the choline head group, as expected. When the mixing time is increased to 300 ms (Fig. 3b) , additional intense resonances occur at 4.7 and 1.3 ppm, which are assigned to water and the lipid acyl chain CH 2 resonances, respectively. The observation of cross peaks between the aromatic protons of the phenylalanine methyl ester and protons on the lipid acyl chains suggests that spin diffusion occurs at longer mixing times. In order to illustrate how aromatic amino acids partition into membrane interfaces in the context of a peptide sequence, Fig. 4 shows rows taken from the 2D MAS-1 H NOESY spectra of a model basic-aromatic peptide, Ac-KKFSFKK-NH 2 , containing two phenylalanines and four lysines. As in Fig. 3 , the rows correspond to the position of the aromatic proton resonances of the phenylalanines and were obtained with mixing times of 50 ms (a) and 300 ms (b). The N-and C-termini are protected in order to address how the peptide sequence might behave in the context of a membrane protein domain, and the spectra were obtained of the peptide bound to DMPC:DMPG model membranes. With a mixing time of 50 ms, cross peaks are observed between the aromatic protons of the peptide and the lipid acyl chain protons. These data indicate that the aromatic rings of phenylalanine in the seven-residue peptide are more deeply inserted into the membrane bilayer than those of the phenylalanine methyl ester model compound.

The next level in complexity is presented by the effector domain of the MARCKS protein discussed in Section 1 [107] . The peptide sequence of the MARCKS effector domain (151-KKKKKRFSFKKSFKLSGFSFKKNKK-175) is unstructured and contains five aromatic and 13 basic residues. Fig. 5 presents rows at the position of the aromatic proton resonances from 2D MAS-1 H NOESY spectra of MARCKS(151-175) bound to DMPC:DMPG model membranes. As above, the NOESY spectra were obtained with mixing times of 50 ms (a) and 300 ms (b). Remarkably, the 50 ms spectrum exhibits an intense cross peak between the aromatic protons and the lipid acyl chain protons. These data indicate that the aromatic rings of the phenylalanines penetrate the model membrane to below the level of the acyl chain carbonyls. The location of the MARCKS effector domain deep in the membrane is supported by EPR measurements of Cafiso and co-workers [111] . Importantly, replacing the five phenylalanine residues in the MARCKS(151-175) peptide with alanines shifts the equilibrium binding position of the peptide more than 10 Å so that it resides in the aqueous phase [112] . It is estimated that the deep penetration of the aromatic rings of the five phenylalanines increases the affinity of MARCKS(151-175) with bilayer membranes by a factor of 10-100 [23, 111, 113] .

The differences in the NOESY spectra between these three different basic-aromatic model systems demonstrate the ability of this approach to establish the depth of the peptides in membrane bilayers. The differences are most dramatic when comparing the spectra obtained with a 50 ms mixing time; the 50 ms spectrum of phenylalanine methyl ester exhibits no intensity in the cross peak region corresponding to the lipid acyl chain protons, whereas the 50 ms spectrum of the MARCKS(151-175) peptide exhibits almost no intensity in the cross peak region corresponding to the protons of the phospholipids head group. The influence of spin diffusion can be quantified by measuring the buildup of cross peak intensity as a function of the mixing time [107] . Fig. 6 presents a plot of the intensities of the acyl chain CH 2 resonance at 1.3 ppm as a function of the NOESY mixing time for phenylalanine methyl ester and MARCKS(151-175). The slight negative curvature for the buildup curve of the phenylalanine methyl ester suggests that magnetization transfer between the aromatic protons and the lipid acyl chains results from spin diffusion [107] . In contrast, the rapid rise of the NOE buildup curve for MARCKS indicates the Phe rings are in direct contact with the lipid acyl chains [107] . One of the striking observations that emerges from the NMR studies on the MARCKS peptide is that the aromatic rings of the phenylalanine side chains are located within the hydrophobic core of the bilayer. This contrasts with both the small molecule studies of White and Gawrisch related to Trp and the preference for Trp and Tyr at the ends of transmembrane helical segments in membrane proteins of known structure [103] . Phenylalanine has a much higher occurrence in the transmembrane segments of membrane proteins than either tyrosine or tryptophan [5] . In an elegant study on the preferred position of Trp and Phe in membrane bilayers, Braun and von Heijne [114] found that Phe, but not Trp, is accommodated in the hydrophobic core of the bilayer when inserted into a poly-Leu transmembrane helix. Moreover, measurements on the partitioning of benzene into bilayers shows that there is no preference for benzene partitioning into either the interfacial or hydrophobic core regions of the bilayer [115, 116] .

The focus above has been on locating the aromatic residues in basic-aromatic motifs. 1 H NOESY experiments can also be used to identify the position of the basic amino acid side chains. In both lysine and arginine, the basic amine or guanidium functional group is attached at the end of a relatively long hydrophobic chain. This structure allows the basic residues to insert into the membrane interface with the charged end of the side chain 'snorkeling' into the membrane surface. Killian and co-workers have convincingly demonstrated that basic amino acids in transmembrane peptides can snorkel through the membrane interface [117] .

For 1 H MAS-NMR studies, the side chain of arginine offers an advantage over lysine since the protons associated with the guanidium group have chemical shifts in the region of 7-8 ppm and are consequently resolved from the lipid resonances that dominate the spectrum. Cafiso and co-workers found that the arginine side chains of an 11-residue peptide derived from a secretory carrier membrane protein (SCAMP) lie below the level of the lipid phosphates [118] . SCAMPs are membrane proteins with four transmembrane segments that function in membrane fusion during exocytosis.

The position of the charged residues relative to the low dielectric interior of the bilayer raises an important issue concerning electrostatics. The aromatic amino acids in basicaromatic clusters may influence the electrostatic interactions with the membrane surface [110] by dragging the positive charge on the flanking lysine and arginine side chains deeper into the polar head group so they can interact more effectively with the negative lipid charges (e.g. on the PI(4,5)P 2 or phosphotidylglycerol (PG) head groups). The electrostatic potential extending from the positively charged residues would be expected to increase substantially as the charge approaches the low dielectric membrane surface. Electrostatic calculations on a simple ion of charge q at an idealized membrane-water interface show that if ions are excluded from the membrane phase, then the potential at the membrane interface extends much further than it would for the same charge in bulk water [119] . In fact, the electrostatic potential at a distance r from an ion at a low dielectric interface is twice that predicted by Debye-Hückel theory for an identical ion in the bulk aqueous phase. This increase in electrostatic potential enhances the ability of the basic-aromatic peptides to sequester PI(4,5)P 2 . For example, MARCKS(151-175) is able to electrostatically sequester PI(4,5)P 2 with high selectivity (w1000 fold) over monovalent acidic lipids, such as phosphatidylserine [120] . The high selectivity is predicted from the high valence (K4) of PI(4,5)P 2 and the Boltzmann factor associated with the MARCKS(151-175)-PI(4,5)P 2 interaction [80] .

Solid-state NMR spectroscopy of biological systems has typically focused on the 13 C, 15 N and 19 F nuclei for structural studies. Structural studies of membrane bound peptides have exploited the orientational dependence of 15 N chemical shifts and 15 N-1 H dipolar couplings. Several reviews are available on these methods [41, 121] . The structure determination of gramicidin A in membrane bilayers was a tour de force of solid-state NMR spectroscopy using this approach [44] . This approach has also been applied to the M2 peptide studies by Cross and co-workers [122] . M2 has a hydrophobic transmembrane helix and N-terminal amphipathic helix containing a basic-aromatic cluster. The amphipathic helix lies approximately parallel to the surface of the membrane with several phenylalanines penetrating into the hydrophobic core of the membrane. Hong and co-workers have used 13 C and 15 N chemical shifts to determine the orientation of protegrin 1 in lipid bilayers [123] . The orientation favors interaction of the hydrophobic backbone of the peptide with the hydrophobic core of the bilayer and positions the cationic arginine side chains to interact with the anionic phosphate groups.

Structural studies based on the measurement of 13 C-13 C homonuclear dipolar couplings and 13 C-15 N heteronuclear dipolar couplings have also been extensively used to establish the structures of membrane bound proteins and peptides. These methods have been reviewed [124] [125] [126] . For example, Ramamoorthy and co-workers [52] have described 13 C-15 N rotational echo double-resonance experiments in multilamellar vesicles to establish the helical conformation of the C-terminal segment of paradaxin Pa4 in lipid membranes.

Hong and co-workers have developed 15 N-detected, 1 H spin diffusion experiments and 13 C-detected, 1 H spin diffusion experiments to measure how deeply specific protein residues are located in membrane bilayers [127] . They applied the approach to gramicidin A, where they were able to resolve 4-5 Å differences in depth between three 15 N-labeled backbone sites (Gly2, Val7, Trp13). They have also described 2D 13 C and 1 H spin diffusion experiments showing that ordered aggregates of the anti-microbial peptide protegrin are oriented parallel to each other [128] .

Recently, Hong and co-workers [129] have shown that 19 F spin diffusion measurements can be made in MAS experiments to determine the oligomeric state of peptides bound to lipid bilayers. They observed that magnetization transfer between chemically equivalent, but orientationally different, 19 F spins on different molecules reduces the 19 F magnetization in an exchange experiment and applied the method to show that the anti-microbial peptide protegrin-1 is almost completely dimerized in POPC bilayers at a concentration of 7.4 mol%. Decreasing the peptide concentration reduced the dimer fraction.

The sections above investigate the interaction of basicaromatic clusters with membrane bilayers from the perspective of the protein. In the next two sections, we discuss solid-state 31 P and 2 H NMR studies used to determine how these clusters influence the bilayer structure. Both 31 P and 2 H NMR have been standard approaches for characterizing membrane structure and dynamics for over 30 years. Reviews have focused on both lipid systems (e.g. [130] ) and lipid-protein interactions [131] [132] [133] [134] As a simple introduction to 31 P NMR, Fig. 7 shows static and MAS 31 P spectra of DMPC:DMPG bilayers containing the bound basic-aromatic model peptide, Ac-KKFSFKK-NH 2 , described above. The static 31 P spectrum exhibits an axially symmetric lineshape resulting from axial diffusion of the lipids (Fig. 7a) . The DMPC and DMPG lipids have slightly different lineshapes giving rise to the observed splitting in the static spectrum. In the 31 P spectrum obtained with MAS, the anisotropic chemical shift interactions are averaged resulting in sharp features at the isotropic chemical shifts for DMPC and DMPG (the dashed line in Fig. 7b shows the spectra of DMPC:DMPG bilayers without bound peptide). Binding of the Ac-KKFSFKK-NH 2 peptide produces a distribution of chemical shifts for both DMPC and DMPG (solid line, Fig. 7b ). 31 P measurements can also be made on oriented bicelles. These spectra provide information on both the orientation and the environment of the lipid head groups. Bicelles are typically composed of a mixture of long chain (DMPC and DMPG) and short chain (DHPC) lipids. The molar ratio (or q-value) of the long to short chain lipids determines the morphology of the bicelle [135] . Isotropic bicelles are formed with a ratio of 1:1 (or lower), while oriented bicelles are formed with a molar ratio of w3:1 (or higher). 31 P spectra of oriented membrane bicelles titrated with the MARCKS effector domain peptide are shown in Fig. 8 . The bicelles have a q value of 4. The ratio of DMPC and DMPG to DHPC is 4:1, while the ratio of neutral DMPC to DMPG is 3:1. The 31 P spectrum is sensitive to the orientation of the phosphate head group of the three phospholipids; three distinct resonances are observed corresponding to the 31 P group of DMPC (K5 ppm), DMPG (K2 ppm) and DHPC (2-3 ppm).

The large shift difference between DHPC and DMPC, both of which have PC head groups, is due to the morphology of the qZ4 bicelles. The DMPC lipids are in the bilayer portion of the bicelle, and the lipid axis is approximately perpendicular to the direction of the applied magnetic field of the spectrometer, whereas the DHPC lipids are on the edges of the bicelle approximately perpendicular to the DMPC lipids. DMPG is in the bilayer portion of the bicelle, but has a slightly different 31 A dramatically different picture for how basic-aromatic clusters can interact with membrane bilayers is obtained with the cell permeant peptide, Antp(43-58). Fig. 9 presents 31 P spectra of the Antp(43-58) peptide bound to model membrane bicelles having the same composition as in Fig. 8 [136] . As Antp (43) (44) (45) (46) (47) (48) (49) (50) (51) (52) (53) (54) (55) (56) (57) (58) is titrated into this system, the chemical shift of the DMPC resonance does not change. In contrast, the DMPG resonance changes from K1.74 to K0.27 ppm, moving closer to the resonance from DHPC. The dramatic shift of the 31 P resonance of DMPG can be caused by a change in the orientation of the DMPG head group or by a change in the local electrostatic environment. The 31 P MAS spectrum of unoriented membrane bicelles is very similar to that shown for DMPC:DMPG (dashed line) in Fig. 7 , which means that the 31 P chemical shift is not due to local electrostatic interactions. Instead, this argues that the observed DMPG shifts are due to a change in the headgroup orientation.

The Ab42 peptide associated with Alzheimer's disease is yet another example of how peptides with basic and aromatic amino acids can interact with biological membranes. Ab42 is derived from amino acids 672 to 713 of the amyloid precursor protein (APP). Most gene mutations that are associated with the inherited forms of Alzheimer's disease cause an increase in the ratio of Ab42 over shorter Ab peptides. APP is cleaved at Asp672 in the extracellular domain by b-secretase and at Ala713 within the transmembrane domain by g-secretase [137, 138] . The mechanism of toxicity has been variously attributed to the formation of ion pores [139] [140] [141] [142] , direct destabilization of membranes [143] [144] [145] [146] or to permeation through neuronal cells and inhibition of cellular enzymes such as alcohol dehydrogenase [147] . There have consequently been several studies using solid-state NMR to establish how this peptide interacts with membranes [148] [149] [150] [151] . 31 P spectra of the Ab42 peptide bound to membrane bicelles having the same composition as in Figs. 8 and 9 are shown in Fig. 10 . Remarkably, the influence of Ab42 binding on the 31 P spectrum is different than for both the MARCKS and Antp (43) (44) (45) (46) (47) (48) (49) (50) (51) (52) (53) (54) (55) (56) (57) (58) peptides. Binding of Ab42 results in a broadening of the 31 P linewidths with the most dramatic change occurring for DMPG. However, the chemical shifts of the 31 P resonances do not change dramatically indicating that the bilayer structure of the bicelles remains intact. Consistent with these results, Seelig and co-workers [148] used 31 P NMR of membrane vesicles to show that the bilayer structure is not disrupted by the Ab40 peptide. Moreover, using 2 H NMR they found no change in the conformation of the choline headgroup or in the ordering of the hydrocarbon chains. As a result, they concluded that Ab40 binds electrostatically to the outer envelope of the polar headgroup region without penetrating the membrane interface.

Bonev et al. [152] compared the 31 P spectra of DMPC:DMPG membranes containing bound Ab40 peptide and pentalysine. Pentalysine does not penetrate the membrane interface and should only interact electrostatically with the negatively charged DMPG lipids. They found that the intensity of the 31 P resonance of DMPG is reduced relative to DMPC when Ab40 is added to multilamellar membrane vesicles, whereas no change was observed in the relative intensities of the 31 P resonances with pentalysine. These results suggest that the aromatic (and other hydrophobic) amino acids contribute to membrane binding.

There are many additional ways that 31 P NMR can be exploited to characterize lipid structure and dynamics. For instance, Auger and co-workers [153] have used 2D 31 P solidstate NMR spectroscopy [154] to determine how b-purothionin affects the slow motions (w10 3 Hz) of the lipids and more specifically, the lateral diffusion of DMPG. In 2D 31 P-EXSY solid-state NMR spectra, cross-peak intensity appears when there is a change of orientation during the mixing time. The b-purothionin protein is found to decrease the lateral diffusion of the DMPG and may act as an obstacle to the lipid diffusion. Auger and co-workers also suggest that the presence of b-purothionin significantly modifies the lipid packing at the surface of the bilayer, which in turn increases the accessibility of water molecules in the interfacial region.

Finally, 31 P experiments are well suited for addressing curvature of the membranes due to the binding and penetration of basic-aromatic motifs. The peptide strongly perturbs the polar head group region in order to penetrate into the acyl chain region of the membrane. As the aromatic rings become buried in the hydrophobic core of the membrane, the lipid head groups are pushed aside by the backbone and the side chains of the peptide. Monolayer studies demonstrate that the increase in the surface pressure is small (0.3 mN/m) for w1 MARCKS(151-175) peptide bound per 100 lipids, but increases markedly as the number of peptides bound per unit area increases. Hence binding would be expected to result in significant membrane curvature in regions of the cell membrane where MARCKS is concentrated. Membrane curvature is critical for endocytotic vesicle formation, a process that is often mediated by proteins that specifically interact with PI(4,5)P 2 [155] . MARCKS is concentrated in regions of cells associated with inward curvature, such as the nascent phagosomes of macrophages Fig. 10 . 31 P spectra of membrane bicelles containing the Ab42 peptide associated with Alzheimer's disease. The molar ratio of DMPC to DMPG to DHPC is 10:3:3.25, and the bicelles were titrated with Ab42 peptide in molar ratios of peptide to lipid from 0:400 to 4:400. The temperature was maintained at 27 8C. 31 P chemical shifts are referenced to external 85% H 3 PO 4 . Spectra were obtained at a 31 P frequency of 283.4 MHz. [156, 157] . The combination of 31 P with 1 H and 2 H NMR (below) may be sufficient to distinguish between membrane curvature and local membrane disruption.

Deuterium NMR spectroscopy provides a complementary method for investigating the structure and dynamics of membrane bilayers. Like 31 P NMR methods, it has been used extensively for the past 30 years. NMR studies have focused on both the lipid head group and lipid acyl chains. Seelig and colleagues developed the idea of using deuterium NMR of the lipid headgroup as a voltage sensor [158] . They found that the phosphocholine dipole is sensitive to the electric surface charge and that a change in the orientation of the PC headgroup results in a linear change, at low surface charge density, in the quadrupole splittings of the aand b-CD 2 deuterons.

Chain 2 H-labeled lipids provide a good probe for observing the chain mobility of the membrane. The distribution of 2 H quadrupolar doublet splittings are dependent on the order parameters of the deuterons on the lipid acyl chain [159, 160] . The order parameter reflects the average orientation and mobility of the lipid acyl chains [161] . Measurements can be successfully made on both static multilamellar systems and on membrane bicelles. There are several advantages of using bicelles, including improved spectral resolution and the ability to titrate membrane associated peptides into the bicelle solution.

The deuterium spectrum of DMPC:DMPG bicelles containing chain deuterated DMPC is shown in Fig. 11a . The terminal methyl groups give rise to the sharp resonances at G1000 Hz. Their position near the center of the spectrum reflects the high mobility and disorder at the ends of the acyl chains. A decrease in mobility and/or an increase in order are observed as one moves toward the glycerol backbone. The largest quadrupolar splittings (G11 kHz) are associated with the CD 2 groups adjacent to the acyl chain carbonyls.

When the MARCKS effector domain is titrated into the bicelle solution at a peptide to lipid molar ratio of 1:50 (Fig. 11b) , there is a subtle, but characteristic change in the deuterium quadrupolar splittings. Comparison with the spectrum of lipid alone in Fig. 11a clearly shows that binding of the MARCKS effector domain results in an increase of the quadrupolar splitting of the acyl chain deuterons, which is consistent with a decrease in the mobility of the acyl chains. This result suggests that the deep insertion of the phenylalanines of the MARCKS effector domain reduces chain mobility.

There has been considerable interest in how binding of peptides with basic-aromatic clusters sequesters cholesterol. The large change of the quadrupolar splittings of chain deuterated DMPC with addition of cholesterol suggests a way to monitor cholesterol sequestration. If cholesterol is sequestered by binding of a basic-aromatic cluster, then the quadrupolar splittings should decrease with an increase in bound peptide. The simple idea is that sequestration removes cholesterol from the bulk DMPC, which then should exhibit greater flexibility and less order.

Cholesterol packs against the acyl chains of DMPC and DMPG, thus restricting the motion of the acyl chains [162] . Figs. 11a and 12a , respectively, reveals a dramatic increase in the quadrupolar splittings when cholesterol is in the membrane. This effect is very similar to the insertion of the aromatic rings of the phenylalanine into the acyl chain region of the bilayer when the MARCKS(151-175) peptide binds.

When the MARCKS(151-175) peptide is titrated into DMPC:DMPG:DHPC bicelles containing cholesterol, there is a small, but measurable, decrease in the quadrupolar splittings of chain deuterated DMPC when the peptide binds (Fig. 12b) DMPC:DMPG membranes containing cholesterol as it does into membranes without cholesterol (unpublished results). We present this final example of the MARCKS(151-175) peptide binding to cholesterol-containing membranes to emphasize again the interplay between the binding of basicaromatic clusters and changes in membrane structure. The deep penetration of the MARCKS peptide due to the aromatic groups may be responsible for cholesterol sequestration since parallel experiments with pentalysine show no change in the deuterium quadrupolar splittings (unpublished results). The sequestration may be a response to the local disruption of the bilayer (where the cholesterol is somehow filling in the gaps in the membrane created by peptide binding) or may be due to an interaction with the negatively charged PG lipids that are attracted to the positively charged MARCKS (as suggested by Anderson [87] in his lipid-shell mechanism for membrane domain formation).

The diversity of ways that basic-aromatic clusters can interact with membranes is remarkable. This review only scratches the surface of what may be one of the least understood areas of cell biology. Besides simply binding to membranes and inducing a change in structure, the interactions of basic-aromatic clusters with membranes are often regulated. For instance, membrane interactions of many proteins, including MARCKS, are regulated by phosphorylation. In MARCKS, phosphorylation of serine residues within its basic-aromatic cluster introduces a negatively charged phosphate that lowers the binding affinity of the cluster for negatively charged membranes. The binding of other basicaromatic clusters is regulated by protein-protein interactions, such as interaction with Ca 2C -calmodulin [24] . As a result, the cell biology that relies on the specific ways that proteins influence membrane structure is both complex and dynamic. NMR methods that provide high-resolution information on both structure and dynamics are well suited for these types of systems. The membrane environment is an integral part of the system and consequently these problems are not amenable to protein crystallography.

There are three separate areas where NMR can have an impact on understanding the structure and function of basicaromatic clusters: (i) protein or peptide structure, (ii) membrane structure and (iii) lipid-protein interactions.

(i) Membrane protein structure. One of the themes that emerges from this review is that the structure of the basic-aromatic cluster often determines the mechanism of membrane binding and interaction. Basic-aromatic clusters can be contained in amphipathic helices, b-sheets, and structure loops, as well as being part of natively unstructured proteins. Moreover, the secondary structure of the cluster can be induced upon membrane binding. It is important to note that besides secondary structure, the function of basicaromatic clusters often depends on the tertiary and quaternary structure (or oligomeric state) of the peptides or proteins involved. In addition, local cation-p interactions between the positively charged side chains of basic amino acids and the aromatic rings of Trp, Phe and Tyr may play a role in the ability of basic residues to partition into the membrane interface. NMR has a tremendous advantage over CD and IR spectroscopy in terms of establishing to high resolution the local structure and the interactions of basicaromatic clusters in lipid bilayers. (ii) Membrane structure. There is a vast body of literature on using 31 P and 2 H NMR to characterize the structure and dynamics of lipid membranes. The challenges posed by the complex and dynamic nature of membranes and how they respond to membrane-associated proteins is illustrated in the three examples presented in the 31 P section: MARCKS, AntP, and Ab42. These proteins all contain basic and aromatic residues, yet binding to membrane bilayers results in each case in unique changes in the 31 P spectrum. These changes likely originate from electrostatic interactions between the basic amino acids and the negatively charged DMPG lipids. However, the nature of the interactions must be different for the different basicaromatic clusters. (iii) Lipid-protein interactions. There is also a substantial literature on using NMR to characterize lipid-protein interactions. Over the past few years there has been a resurgence of interest in how membrane proteins influence membrane structure, and a recognition that the interactions can be subtle and specific. Biophysics has much to offer in way of understanding the cell biology that relies on these interactions. Moreover, cell membranes are increasingly being viewed as proteinrich environments able to form distinct microdomains that direct cellular processes. The recent proposal by Anderson [87] that lipid shells are involved in the formation of microdomains in membranes is an excellent example. The challenge for the future is to develop NMR approaches that move beyond model systems and capture the complexity and dynamics of cellular membranes.

Proc. Natl. Acad. Sci

Proc. Natl. Acad. Sci. USA

Proc. Natl. Acad. Sci

Proc. Natl. Acad. Sci. USA 90

Glossary Antp: Antennapedia APP: amyloid precursor protein BAR: Bin/amphiphysin/Rvs CRAC: cholesterol recognition/interaction amino acid consensus DAG: diacylglycerol DHPC: 1,2-dihexoyl-sn-glycero-3-phosphocholine DMPC: 1,2-dimyristoyl-sn-glycero-3-phosphocholine DMPG: 1,2-dimyristoyl-sn-glycero-3-phospho-rac-(1-glycerol)] HIV: human immunodeficiency virus MARCKS: myristoylated alanine-rich C kinase substrate MAS: magic angle spinning NMDA: N-methyl-D-asparate NOESY: nuclear Overhauser enhancement spectroscopy PH

This work was supported by NIH-NSF instrumentation grants (S10 RR13889 and DBI-9977553), and grants from the National Institutes of Health to S.O. Smith (GM 46732 and GM 69651). We gratefully acknowledge the W.M. Keck Foundation for support of the NMR facilities in the Center of Structural Biology at Stony Brook, and Martine Ziliox and Markus Eilers for critically reading the manuscript.