Both acidic and basic amino acids in an amphitropic enzyme, CTP:phosphocholine cytidylyltransferase, dictate its selectivity for anionic membranes.

Amphitropic proteins are regulated by reversible membrane interaction. Anionic phospholipids generally promote membrane binding of such proteins via electrostatics between the negatively charged lipid headgroups and clusters of basic groups on the proteins. In this study of one amphitropic protein, a cytidylyltransferase (CT) that regulates phosphatidylcholine synthesis, we found that substitution of lysines to glutamine along both interfacial strips of the membrane-binding amphipathic helix eliminated electrostatic binding. Unexpectedly, three glutamates also participate in the selectivity for anionic membrane surfaces. These glutamates become protonated in the low pH milieu at the surface of anionic, but not zwitterionic membranes, increasing protein positive charge and hydrophobicity. The binding and insertion into lipid vesicles of a synthetic peptide containing the three glutamates was pH-dependent with an apparent pK(a) that varied with anionic lipid content. Glutamate to glutamine substitution eliminated the pH dependence of the membrane interaction, and reduced anionic membrane selectivity of both the peptide and the whole CT enzyme examined in cells. Thus anionic lipids, working via surface-localized pH effects, can promote membrane binding by modifying protein charge and hydrophobicity, and this novel mechanism contributes to the membrane selectivity of CT in vivo.

Proteins that interact reversibly with cell membrane lipids usually have selectivity for negatively charged phospholipids (1,2). Some of these proteins show specificity for a particular anionic phospholipid. For example, protein kinase C binds preferentially to phosphatidylserine (3); MARCKS (4) and proteins with PH domains (5) bind selectively to various phosphoinositides. However, many proteins exhibit non-selectivity with respect to the anionic phospholipid by a simple electrostatic interaction between clusters of basic residues on the protein and negatively charged lipid head groups (6). This binding affinity may be regulated by changes in membrane anionic lipid content, but more often by modification of the charge on the protein by a mechanism referred to as an "electrostatic switch" (7).
Phosphorylation of a basic patch on such proteins as MARCKS (7), Src (8), and ARNO (9) neutralizes positive charge. Alternatively, calcium binding to acidic residues in the C2 domain of protein kinase C and phospholipase A2 increases positive charge (10). Hisactophilin exhibits a variation on this theme, in which the protein charge may be modulated by cytosolic pH change (11). In this work, we provide an example in which increased anionic lipid composition modulates protein charge and hydrophobicity by influencing its protonation state and thereby increases membrane affinity.
CTP:phosphocholine cytidylyltransferase (CT) 1 catalyzes a key rate-limiting step in PC synthesis and contributes to maintenance of cell membrane PC homeostasis (12). When the relative membrane PC content is altered, CT could respond by recognition of ensuing changes in the physical features of PCdeficient or PC-overloaded membranes, including changes in the negative surface charge density (13). Thus, how CT responds to changes in surface charge has great bearing on the control of membrane phospholipid compositional homeostasis. There are three homologous mammalian CT isoforms, ␣, ␤1, and ␤2 (12). Most of the biochemical characterization, including the work in this study, has been done with CT␣. CT␣ activity is regulated by reversible membrane binding, which involves both electrostatic and hydrophobic interactions (12,13). It binds poorly to pure PC membranes in vitro, and its binding affinity increases in proportion to the negative surface charge of the membrane (14,15). Importantly, its translocation to membranes in living cells also increases as a function of anionic lipid content (16 -18). Membrane translocation is also accompanied by dephosphorylation of its C-terminal domain, but this event is subsequent to membrane binding (19), and the effects of phosphorylation status on membrane partitioning can be overcome by raising the anionic lipid content of the membrane (20). Based on in situ imaging with fluorescent antibodies or with GFP-tagged CT␣, the enzyme is predominantly nuclear in many cells (reviewed in Ref. 12) and translocates to the nuclear envelope upon stimulation with exogenous fatty acids (12,21). However, in other cells and contexts CT␣ appears to be cytoplasmic and ER-bound (12,(22)(23)(24). The reason for these cell-dependent differences in CT␣ localization is unresolved.
A well characterized membrane-binding domain present in all CT isoforms (domain M) consists of a long amphipathic ␣-helix ( Fig. 1A and Refs. [25][26][27]. Domain M can be subdivided further into: (i) an N-terminal polybasic region (subdomain N, residues 237-255), (ii) a central, net negative region containing three 11-mer repeats (the VEEKS subdomain, residues 256 -288), and (iii) a C-terminal aromatic-rich region terminating in a predicted bend (Fig. 1A). Peptides corresponding to either the entire domain M, subdomain N, or the VEEKS subdomain bind selectively to anionic lipids (27). The non-polar face of domain M helix creates a ϳ120 o wedge containing 18 aliphatic or aromatic side chains. The polar face is rich in acidic side chains. One of the interfacial strips is exclusively basic, and the other is a mixture of acidic and basic residues (Fig. 1, B and C).
Our goal is to characterize the determinants in domain M responsible for CT selectivity toward anionic lipids. Using a VEEKS subdomain peptide, we previously showed that mutating to alanine the three serines interrupting the nonpolar face increases peptide hydrophobicity and reduces (but does not eliminate) the selectivity for anionic lipids (27). The N-terminal portion of domain M is the most highly conserved region of domain M (28), and has the highest concentration of positive charge. Here we have probed the contribution of these basic amino acids to the electrostatic interaction with anionic phospholipids by en bloc substitution of 5 or 8 interfacial Lys/Arg with glutamine (Fig. 1B). We found that the membrane affinity was progressively reduced upon progressive elimination of peptide positive charge. The electrostatic component of the membrane binding was virtually eliminated upon removal of positive charge from both interfaces flanking the hydrophobic face of the peptide.
CT binding to anionic lipid vesicles is enhanced as the pH is lowered from 7.4 to 6.3 (15). One hypothesis for this effect is that lowering the pH protonates the weakly acidic side chains in domain M, specifically the three interfacial glutamates, thereby neutralizing their negative charge and enhancing peptide hydrophobicity. We proposed (25) that the probability of protonation of these glutamates would be higher at the interface of an anionic membrane versus a zwitterionic membrane, because of the attraction of protons at the negative surface (Fig.  2). In this way, ironically, the positioning of three interfacial glutamic acid side chains could contribute to the selectivity for anionic phospholipids. To test this hypothesis we compared the binding characteristics of a wild-type VEEKS repeat peptide with a mutant analog, in which the interfacial glutamates were replaced with glutamine ( Fig. 1C). We found that the mutant peptide had reduced dependence on the anionic lipid content for binding, and that binding was pH-insensitive. These results imply that the interfacial glutamates contribute to the selectivity for anionic lipid. The same mutations were generated in the whole enzyme. The glutamine-substituted enzyme had higher affinity for membranes and, like the mutant peptide, was less dependent on high membrane surface charge for binding.

EXPERIMENTAL PROCEDURES
Materials-Cell culture materials, restriction enzymes, and PCR primers were from Invitrogen. All other chemicals were reagent grade or better.
Synthesis of Peptides-All peptides were acetylated and aminated on their N and C termini. The syntheses of the 33-mer peptides corresponding to the wild-type subdomain N sequence (residues 236 -268) and the VEEKS repeat sequence (residues 256 -288) of rat CT␣ (29) were described previously (27). All other peptides were synthesized by Dr. Krystyna Piotrowska at the University of British Columbia Peptide Service Laboratory on an ABI model 431A synthesizer using Fmoc chemistry. The peptides were Ͼ98% pure after HPLC purification using a VYDAC C 18 column. Their masses, determined by MALDI-mass spectroscopy, were within Ͻ2 daltons of their calculated mass, confirming the correct sequences. After lyophilization to remove the bound trifluoroacetic acid, the peptides were dissolved in water to a working stock concentration of 0.3-1 mM as confirmed by comparison to the fluorescence of a tryptophan standard.
Lipids-Egg PC and egg PG were from Northern Lipids (Vancouver,  (29). B, subdomain N peptide NMR structure (25). Only the helix-forming residues 242-268 are displayed. C, VEEKS-repeat peptide structure is a merge of the NMR structures of overlapping peptides, Pep-8K (providing residues 256 -267) and a 22-mer (providing residues 268 -288) (25). Structures are displayed using RasMol (58). Backbone and hydrophobic side chains are yellow, basic side chains are violet, acidic and polar side chains are red. Side chains mutated in the study are in ball-and-stick representation. Since the pH at the surface of anionic (but not zwitterionic) membranes is lower than the bulk pH due to the attraction of protons to the negative surface, the probability of glutamate protonation will be higher at the anionic surface. Protonation of interfacial glutamates would increase peptide hydrophobicity and eliminate charge repulsion, enhancing membrane affinity.
Preparation of Buffers with Varying Ionic Strength and pH-Ionic strength of samples was varied by addition of 1-100 mM phosphate buffer stocks at pH 7.0 and addition of NaCl. Ionic strength was calculated using Equation 1, where c and z are the concentration and charge of each contributing ion, respectively. For pH titrations, phosphate buffers were prepared from pH 4.25 to 7.4 (50 mM) and pH 7.6 to 8.5 (40 mM). The ionic strength was equalized to 131 mM with NaCl. We accounted for the slight decrease in pH that occurred upon addition of 3 mM SUVs to samples for CD analysis (Fig. 6).
Lipid Vesicle Filtration Binding Assay-SUVs of various compositions were incubated with peptide (10 M) in phosphate buffer, pH 7.0, at 20°C. Vesicle-bound peptide was trapped by centrifugation through Microcon-100 filters (Amicon; Beverly, MA) at 3000 ϫ g for 10 -30 min until one-half to three-fourths of the original volume had filtered (27). Samples of the filtrate were diluted 2-fold in phosphate and/or NaCl to equalize ionic strength, and adjusted to pH 10 with NaOH. The free peptide concentration was analyzed by derivatization with fluorescamine (0.4 mM, Sigma Chemical) (31). For binding assays with the VEEKSrepeat peptides, samples of the filtrate were diluted 1:1 with methanol, and the peptide concentration was measured via tryptophan fluorescence (excitation, 280 nm; emission, 345 nm) to increase sensitivity by a factor of ϳ5. For each analysis, standard curves were conducted with known concentrations of the appropriate peptide. A dimensionless partition coefficient (K x ) was calculated using samples where the percentbound peptide ranged between 16 and 70% unless otherwise stated, using Equation 2, Fluorescence Assay of Peptide Insertion-SUVs of various compositions were incubated with peptide (3 M) at 20°C in phosphate buffer as described for at least 5 min prior to spectral acquisition. Tryptophan fluorescence spectra were acquired as described previously (27). Tyrosine fluorescence of Pep-5KQ (15 M) was monitored at 304 nm (excitation, 280 nm). The increase in peptide fluorescence at 304 nm in the presence of vesicles was normalized to F 304 of peptide in buffer. All spectra were acquired at 20°C on an SLM 4800C or PTI model QM-1 spectrofluorometer. Spectra were smoothed, and the contribution of the lipid was subtracted. To calculate K x values from fluorescence measurements, the ratio of bound/total peptide was estimated from the fluorescence measurements by (F Ϫ F o )/(F max Ϫ F o ), where F ϭ fluorescence increase or blue shift of sample containing peptide ϩ lipid; F o ϭ fluorescence value of lipid-free peptide; F max ϭ fluorescence value at saturating lipid contents.
Circular Dichroism-CD spectra were acquired at 20°C as described previously (27). Peptides (30 M) were mixed with lipid vesicles or trifluoroethanol for at least 5 min prior to spectral acquisition. Spectra were smoothed and the contribution of the buffer and/or lipid was subtracted. The CD values were converted to mean residue molar ellipticity (; deg cm 2 dmol Ϫ1 ), and the percent helix was estimated from 222 nm as described (27).

Construction and Expression of Wild-type and Mutant
CTs-To generate a mutant CT␣ containing glutamines at codons 257, 268, and 279, we designed the following mutagenic primers: CL1, 5Ј-cGGATCCaaA-TCGATAGATCTcatccagaagtggcaggagaagtCCCGGGagttcattggaagt-3Ј; CL2; 5Ј-cGGATCCAGATCTATCGATttctcctgcactttctgcacaaattctttcgacttttcctgcacatctttcacttt-3Ј. The mutagenic nucleotides are in bold and the engineered restriction sites, BamHI, BglII, ClaI, and SmaI (uppercase) created silent mutations. Two PCR reactions were performed using Pfu polymerase with wild-type rat CT␣ cDNA (32) inserted into the SalI site of pBSKS (ϩ) as a template. The CL1 mutagenic primer was paired with the vector reverse T7 primer, and the CL2 mutagenic primer was paired with the vector forward T3 primer to generate PCR products of 910 and 430 bp, respectively. These PCR products were cut with ClaI and SalI and inserted into the appropriate sites of pBSKS (ϩ). The accuracy of the resulting constructs was confirmed by sequencing. The two PCR fragments were joined at the ClaI site to generate CT-3EQ. CT-3EQ was moved to the expression vector pAX142 (33) using SalI.
Expression and Membrane Partitioning of CT-WT and CT-3EQ in COS Cells-COS-1 cells were cultured and transfected with pAX-142 constructs as described (34) except that the seeding density was 1 ϫ 10 6 cells/10-cm dish, and the cells were glycerol-shocked for 2.5 min following rather than preceding treatment with 100 M chloroquine for 3 h at 37°C. Cells were transfected with 3 g of pAX142-CT␣-WT per 10-cm dish for 20 h and 10 g of pAX142-CT␣-3EQ for 60 h to achieve equivalent expression levels. To enrich cells with oleic acid they were incubated for 1 h at 37°C with media containing 1 mM sodium oleate (Sigma Chemical) and 0.5-10 mg ml Ϫ1 BSA (fatty acid free; Calbiochem) to achieve OA/BSA molar ratios of 133:6.6. The oleate was prepared as a 10ϫ sonicated stock in phosphate-buffered saline and was co-sonicated with the BSA prior to addition to cells. At the highest OA:BSA ratio, the viability of the cells after 1 h was Ͼ90%.
The transfected cells were harvested with phosphate-buffered saline containing 2.5 mM EDTA, and homogenized by sonication for 2 ϫ 15 s at 4°C in 0.4 ml 10 mM Tris, pH 7.4, 1 mM EDTA, 3 mM MgCl 2 , 0.5 mM phenylmethylsulfonyl fluoride, 2 mM dithiothreitol. The protein concentration of the homogenates was ϳ5 mg ml Ϫ1 (35). Electrophoresis and Western blotting using an antibody against the N-terminal 17 amino acids of CT␣ was as described (36). After adding K 2 HPO 4 to a final concentration of 0.2 M, the homogenates were centrifuged at 100,000 ϫ g for 1 h at 4°C to generate a particulate and soluble fraction. The particulate fraction was resonicated in homogenization buffer/0.2 M K 2 HPO 4 as above. Aliquots of both fractions were assayed for CT activity for 15 min under optimum conditions of substrates (15) and lipid activator (200 M egg PC/oleic acid (1:1)).
Determination of the Oleic Acid Content of COS Cell Particulate Fraction-The lipids in the particulate fractions were extracted (37) and analyzed for phospholipid phosphorus content (30). Fatty acid, cholesterol, and triacylglycerol were separated by TLC on precharred silica H plates (Analtech) using hexane/diethyl ether/acetic acid (60:40: 1). The TLC plates were scanned and densitometry was performed with Scion-Image software by reference to standards of oleic acid, cholesterol, and triolein, which were spotted on each TLC plate. Linear regression analysis of the plots of density versus nmol of standard lipid between 0 -20 nmol gave r ϭ 0.93-0.995. The mol fraction oleic acid was calculated as mol fatty acid/(mol PL ϩ mol cholesterol ϩ mol triacylglycerol ϩ mol fatty acid).

Design of Mutant Peptides-
The contribution of the interfacial lysines and glutamates to membrane binding were analyzed using peptides in which the motifs are best represented. The role of the interfacial basic strip was investigated using peptides corresponding to amino acids 236 -268 of rat CT␣ (29). The membrane interactions of the wild-type version of this peptide have been previously characterized (27), and its structure in complex with SDS micelles has been solved (25). It consists of a continuous ␣-helix between residues 242-268, linked via a ϳ50 o bend to a loosely coiled N terminus (25). Fig.  1B shows the positioning of amino acids that were targeted for mutation. In the mutant peptide 5KQ, five basic amino acids on the right-hand interfacial zone were changed to glutamines: Arg-245, Lys-248, Lys-252, Lys-259, and Lys-266. These residues are generally conserved in animal CTs. Assuming protonation of His-241, the substitutions changed the net charge from ϩ4 to Ϫ1. In the mutant peptide 8KQ, an additional three lysines on the opposite face, Lys-250, -254, and -261, were changed to glutamine, to generate a peptide with a net charge of Ϫ4. In the wild-type and 8KQ peptide F263 was substituted with tryptophan to facilitate monitoring of fluorescence.
The three interfacial glutamates (Glu-257, Glu-268, Glu-279) were mutated to glutamine using the VEEKS subdomain peptide (residues 256 -288; Fig. 1C). The membrane binding behavior and secondary structure of the wild-type version of this peptide have been extensively studied (26,27). The NMRderived structure of a 22-mer peptide containing the second and third VEEKS repeats in complex with SDS micelles is a continuous ␣-helix (25). The 3E to 3Q mutation changes the net charge from Ϫ2 to ϩ1. This peptide also contains 6 interfacial basic residues that likely contribute to electrostatic binding.
The substitutions did not alter the propensity of the peptides for helix formation, as determined by CD in 50% trifluoroethanol, a solvent that promotes internal H-bonding (data not shown). The helical contents estimated from the molar ellipticity at 222 nm were similar: 60% for Pep-8K, 73% for Pep-5KQ and 56% for Pep-8KQ; 69% for Pep-3E and 62% for Pep-3EQ. All peptides were predominantly random coil in water, except for Pep-8KQ, which adopted mostly ␤-structure in water (data not shown).

Loss of Interfacial Basic Residues Diminishes Peptide Affinity for Anionic Membranes by Inhibiting the Electrostatic Attraction-The interactions of subdomain N peptides with SUVs
were monitored by a direct vesicle-filtration binding assay, by CD, and by fluorescence changes. Thus, the binding event can be correlated with conformational changes and bilayer insertion processes. The progressively weakened response to PG/PC (1:1) lipid vesicles at 22 mM ionic strength upon removal of 5 or 8 interfacial lysines is evidenced by all three methods in the sets of parallel binding curves shown in Fig. 3. Weakened membrane binding (Fig. 3A) is accompanied by reduced ␣-helix formation (Fig. 3B) and by weaker insertion (Fig. 3C). The bilayer insertion of wild-type peptide and Pep-8KQ was monitored via the blue shift in the fluorescence (330:350 nm) of a tryptophan located in the nonpolar face of the amphipathic helix. This change reached a plateau at a lipid/peptide ratio between 30 -50 for the wild-type peptide. By contrast, no change in the fluorescence of Pep-8KQ was observed below a lipid/peptide ratio of 100. For Pep-5KQ, which was not engineered with a tryptophan, we monitored an increase in fluo-rescence at 304 nm, indicative of the movement of Tyr-240, the lone fluorophore in this peptide, into a more hydrophobic environment. A lipid/peptide ratio of Ն100 was required for maximum change of the F 304 for Pep-5KQ. These data suggest that both strips of interfacial lysines contribute to peptide binding. In addition they provide the first evidence for membrane insertion of the lone aromatic residue in the bend at the N terminus of domain M.
We next probed the electrostatic nature of the interaction of WT, 5KQ, and 8KQ peptides by measuring their partitioning between aqueous and lipid phases as a function of mole percent anionic lipid and medium ionic strength. The effect of vesicle anionic lipid content is shown in Fig. 4A. The binding of the peptides to pure PC vesicles was too weak to obtain a reliable estimate of their molar partition coefficients, K x (sensitivity limit of the assay was K x Ϸ 1 ϫ 10 4 ). The K x of the wild-type peptide increased at least 2 orders of magnitude when the mol% PG was increased from 10 to 100% (Fig. 4A). The response to increasing anionic lipid content was progressively muted upon substitution of the 5 or 8 lysines. This is in keeping with an elimination of the electrostatic component of the binding. The effect of ionic strength is shown in Fig. 4B. Using 50% PG vesicles, the K x for the wild-type peptide was reduced ϳ50fold as the ionic strength was raised from 11 mM to 0.75 M. At the lowest ionic strength, substitution of 5 or 8 lysines reduced the K x values 10-and 15-fold, respectively. The affinity of Pep-5KQ and 8KQ was lowered an additional ϳ3-fold by raising the ionic strength to 44 mM. These data show that the ionic strength required to eliminate the electrostatic component of the binding reaction is greatly reduced for the peptides missing 5 or 8 of the interfacial lysines.
The Apparent pK a for Membrane Binding of the Wild-type VEEKS-repeat Peptide Is a Function of the Anionic Lipid Content-CT binding to PG/PC (2:3) vesicles is induced over a pH range from ϳ7.1 to 6.5 (15), suggesting that a protonation event in domain M drives binding. We explored the effect of vesicle charge density on the pH-dependent binding of the wild-type peptide, Pep-3E (Fig. 1C), using tryptophan fluorescence blue shifts as a measure of membrane binding. Binding to vesicles containing 5-35 mol% PG was enhanced as the pH was lowered from 7.5 to 4.25 (Fig. 5A). The protonation state of the PG is not affected over this pH range (38) thus the effect is due to a change in the protonation state of the peptide. Increasing the anionic lipid content of the vesicles increases the differential between the bulk pH and membrane surface pH, as protons are attracted by the negative membrane surface. This should result in an increase in the apparent pK a for binding shown in Equation 3, where F ϭ Faraday's constant, R ϭ gas constant, T ϭ temperature, and o ϭ the membrane surface potential, calculated using Equation 4, where kT/e ϭ 25.26 mV (59), Z, the charge of Na ϩ counterion ϭ ϩ1, A ϭ 136 Å 2 (59), c ϭ ionic strength (131 mM), and the variable , the surface charge density, was calculated for each PG content using values of 68 Å 2 for both egg PC and PG (40).
In agreement with this prediction, the pK a increased from 5.37 to 5.9 as the PG content increased from 5 to 35% (Fig. 5, A and  B). Although the measured pK a values are uniformly higher than predicted based on the calculated membrane surface potential (Fig. 5B), the measured trend agrees with the calculated plot, both showing an increase in apparent pK a with increasing negative charge density. The higher than predicted pK a val- ues may reflect not only the lower surface pH, but also the trapping of the protonated peptide upon insertion into the membrane. These data support the notion that anionic lipids can, via modulation of the surface pH, affect the protonation state of the wild-type peptide.

Substitution of Three Interfacial Glutamates with Glutamine Eliminates the VEEKS Peptide Binding Dependence on pH-
The interfacial glutamates of the VEEKS peptide provide likely sites for protonation, as this would neutralize their negative charge and thus promote pH-dependent lipid interaction. A prediction of this model is that Pep-3EQ, as a mimic of the protonated form of the wild-type peptide (Pep-3E), should not require a reduction in pH for membrane binding to acidic vesicles. The data shown in Fig. 6 validate this prediction. Fluorescence and CD spectra of Pep-3E in the presence of 30 mol% PG vesicles were pH-dependent with apparent pK a values of 5.7 (fluorescence) or 5.9 (CD), whereas the spectra of Pep-3EQ were nearly equivalent over a pH range of 7.5-4.9. The binding of Pep-3EQ to 10 and 20% PG vesicles, which was less than 100% under these conditions, was also pH independent (data not shown). Since there was no evidence of protonation using conditions where membrane binding is incomplete, the pH independence of Pep-3EQ is not merely because enhanced hydrophobicity increases the affinity so as to mask the effects on binding of protonation at another (non-interfacial) site. Rather, these data indicate unequivocally that the interfacial glutamates are the sites of protonation.

Substitution of Three Interfacial Glutamates to Glutamine Lowers the Anionic Lipid Requirement for Peptide Binding-
The elimination of negative charge upon substitution of glutamate with glutamine should increase peptide affinity by creating more favorable electrostatic interactions and by enhancing peptide hydrophobicity. Both should be reflected by decreased dependence on membrane negative charge density and decreased sensitivity to medium ionic strength. In Fig. 7 the mol% PG was varied, and vesicle interactions were probed via tryptophan fluorescence blue shifts. In 110 mM ionic strength medium the wild-type peptide (Pep-3E) did not interact with vesicles containing Ͻ35% PG, whereas Pep-3EQ interacted with vesicles containing only 10 mol% PG. In fact, the small blue shift for Pep-3EQ in the presence of 100% PC vesicles was significant. CD analysis of the peptides using the same conditions revealed parallel changes in helical content; a lower membrane negative charge density was required to induce ␣-helical content in the 3EQ peptide as compared with the wild-type Pep-3E (Fig. 7). Fluorescence blue shift analyses were used to obtain partition coefficients for binding of Pep-3E and 3EQ to PC vesicles containing 0, 20, 30, and 50% PG at 22 mM ionic strength (Fig. 10B). The results clearly show reduced anionic lipid selectivity and increased hydrophobicity for the 3EQ peptide.
The effect of the E to Q substitutions on peptide hydrophobicity was also investigated by monitoring the fluorescence FIG. 4. Effect of anionic lipid content and ionic strength on wild-type, 5KQ, and 8KQ peptide membrane affinity. Binding was measured by the direct vesicle-filtration assay method and partition coefficients were calculated from data obtained at two lipid/protein ratios using Equation 2 under "Experimental Procedures." The data are averages of 2-8 independent determinations. Wild-type Pep-8K (q), Pep-5KQ (OE), Pep-8KQ (f). A, dependence on the mol% PG (balance egg PC) was determined using an ionic strength of 22 mM, pH 7.0. B, ionic strength dependence was determined using PC/PG (1:1) vesicles, pH 7.0.

FIG. 5. Binding of Pep-3E to anionic vesicles is pH-dependent.
A, insertion of Pep-3E (3 M) into vesicles (0.3 mM) composed of 5 (⌬), 20 (OE), 30 (q), and 35 (E) mol% PG (balance PC) in phosphate buffers of various pH was monitored via tryptophan fluorescence at 330 nm/350 nm. Ionic strength was 131 mM. In the absence of lipid, there was a small blue shift in the tryptophan fluorescence spectra of the peptide at pH Ͻ6 and this was subtracted from values obtained in the presence of lipid vesicles. The data were fit to the equation for a single group titration using GraFit ® (solid line). Representative titration curves are shown, which were repeated with similar results. B, calculated and experimental apparent pK a values are dependent on anionic lipid content. The experimental apparent pK a values (E) were obtained from 2-3 separate titration curves using GraFit ® as described above. The calculated apparent pK a values (q) were obtained using Equations 3 and 4. blue shift as a function of medium ionic strength. Fig. 8 shows that the binding of Pep-3EQ to PG/PC (3:7) vesicles is less sensitive to ionic strength than the wild-type 3E peptide. Increasing ionic strength decreases the negative surface potential at the membrane, reducing both electrostatic interactions with peptide and interfacial changes in apparent pK a (see Equations 3 and 4 in the legend to Fig. 5). Whereas at Ն200 mM salt the wild-type peptide fluorescence ratio is the same as the lipidfree peptide, Pep-3EQ shows significant ionic strength-independent binding. Thus, a key effect of the substitution of the 3E to 3Q is an increase in peptide hydrophobicity. Interestingly, the binding of the wild-type Pep-3E increased at high ionic strength (Ͼ500 mM). This may be due to charge neutralization of the interfacial glutamates by the sodium counterion.
Substitution of three interfacial glutamic acid residues with glutamine reduces the selectivity for anionic lipid and en-hances partitioning into cell membranes. To test whether the three interfacial glutamates contribute to anionic lipid selectivity of CT in cells, we engineered the 3Glu 3 Gln substitution in the CT cDNA and expressed wild-type and mutant protein in COS cells. The wild-type and mutant cDNAs were expressed at very high (100ϫ) levels, effectively swamping out the endogenous activity. We adapted transfection conditions to achieve equivalent expression of the mutant and wild-type CTs (see "Experimental Procedures" and Fig. 9, inset). The specific activity of the mutant CT was equivalent to wild-type enzyme when assayed in the cell homogenate (Fig. 9, inset), suggesting that the folding of the enzyme was unperturbed by the mutation. To examine the effects of the mutation on membrane partitioning and selectivity for anionic membranes we incubated cells with various ratios of BSA/OA to elevate the anionic lipid content of cellular membranes (Fig. 9). This is a standard protocol for promoting reversible membrane binding of CT, as assessed by sedimentation, digitonin permeabilization, or in situ immunofluorescence (16,17,21,34,39). There is evidence that fatty acids can serve as regulators of CT activity and membrane association in vivo. CT activity correlates positively with elevated fatty acid content in lung microsomes during developmental maturation (18) and the fatty acid content in CT-associated lipidic particles in lung cytosol following glucocorticoid administration (41). Moreover, up-regulation of fatty acid synthesis in CHO cells overexpressing a regulator of fatty acid synthesis genes enhanced CT activity, and inhibition of fatty acid synthesis in these cells with cerulenin reversed the effect on CT (21).
The oleic acid content of the cell particulate fraction from the BSA/OA-treated cells ranged from 0 to 28 mol% of total lipid (Fig. 9). The distribution of CT-3EQ between soluble and particulate fractions was altered in a manner reminiscent of the membrane affinity changes of the mutant peptide. In control cells CT-WT was distributed ϳ13% in the membrane fraction. Membrane-associated CT increased gradually with increasing oleic acid content, and then jumped to ϳ65% of total as the mol% oleic acid increased between 8 and 16 mol%. In contrast, CT-3EQ partitioned ϳ50% in the membrane fraction of control cells, and increased to 80% in cells containing 17 mol% oleic acid. The data in Fig. 9 are similar to those obtained with wild-type and mutant peptides in Fig. 7. The higher affinity of CT-3EQ for membranes from control cells in Fig. 9 is expected, as cellular membranes contain acidic lipid (typically 20 mol%), unlike the pure PC lipid vesicles that served as the reference lipid for the experiment in Fig. 7.
Several additional experiments confirmed that the increased partitioning of CT-3EQ into the particulate fraction was the result of enhanced membrane affinity and not aggregation into an insoluble form. The partitioning of both CT-WT and CT-3EQ into the soluble fraction was dependent on the homogenate volume and the medium ionic strength (data not shown). Particulate CT-3EQ, like CT-WT, was released into the soluble fraction upon solubilization of the membrane phospholipid with Triton X-100 (data not shown), implying a membrane association. In all experiments, the particulate/soluble ratio of CT-3EQ was greater than that of CT-WT. Together these data suggest that the 3EQ mutation enhanced membrane partitioning.

DISCUSSION
In this study we have characterized two strategies used by CT as the membrane negative charge sensors. The first, a strategy commonly used by amphitropic proteins, consists of strips of lysines flanking both sides of the nonpolar face of the amphipathic helix, which provide the electrostatic drive. The second, a novel strategy, is the selective protonation of three interfacial glutamates at the acidic lipid surface, which eliminates charge repulsion and elevates hydrophobicity.
Contribution of the Interfacial Lysines to the Binding-The data in Fig. 4, A and B indicate that successive elimination of the interfacial lysines in the subdomain N peptide results in elimination of the electrostatic component of binding. That there is a small effect of ionic strength and PG content on the binding of Pep-8KQ suggests that the two remaining lysines and/or histidine at the N terminus that were not substituted with glutamine participate in the charge interaction with membranes. The decrease in the binding of the Pep-5KQ and 8KQ is the result of the reduction in peptide positive charge and not an altered binding mechanism, since the binding of the mutant peptides is also accompanied by ␣-helix formation and insertion into the bilayer. The Lys 3 Gln substitutions did not increase the hydrophobicity of this peptide (Fig. 4B), in contrast to the Glu 3 Gln substitutions in the VEEKS peptide.
How much binding energy does each lysine contribute to the electrostatic interaction? The ion pairing of lysines with negatively charged lipid head groups in 0.1 M salt has been estimated to contribute 1 kcal/mol based on analysis of the adsorption of simple unfolded polybasic peptides of variable positive charge (42). The free energies associated with the binding of the 3 subdomain N peptides to PC/PG (1:1) vesicles at 22 mM ionic strength were calculated from K x values obtained from fluorescence measurements shown in Fig. 3C and from the filtration assay shown in Fig. 4A. The relationship between the number of peptide lysines and the ⌬G is linear for both assays (Fig. 10A), and indicates a ⌬⌬G for removal of 8 lysines of 2 kcal/mol (data derived from the filtration assay) and 2.65 kcal/ mol (data derived from the fluorescence assay). Thus the electrostatic contribution per lysine at a low ionic strength of 22 mM is only 0.25 to 0.32 kcal/mol. The ⌬⌬G associated with charge neutralization of Pep-8K by raising the ionic strength  ). B, partitioning of Pep-3E (q) and 3EQ (E) between aqueous and membrane phases as a function of PG content was measured via fluorescence. For A and B the K x values were computed using Equation 2 in "Experimental Procedures" for data obtained at lipid/peptide ratios in which the peptide was 16 -70% bound, except for Pep-3E at 10% PG where the peptide was only 10% bound. ⌬G ϭ ϪRTln(K x ). Data represent means Ϯ errors of 2-6 determinations. from 0.1 to 1 M is only 1.1 kcal/mol (values derived from the data in Fig. 4B) or 1.65 kcal/mol (values derived from analogous fluorescence binding analyses, data not shown). Thus, the contribution of each lysine at 100 mM ionic strength is computed to be only 0.14 -0.2 kcal/mol. All these analyses reveal that the contribution per lysine is much less than the 1 kcal/ mol associated with a purely electrostatic attraction. Because CT peptides are not only adsorbed but inserted into the bilayer, an energetic cost of dehydrating the charged lysines decreases the net electrostatic component (43).
Pep-5KQ, 8KQ, and the wild-type VEEKS peptide (Pep-3E) have net charges of Ϫ1, Ϫ4, and Ϫ2 respectively, yet they demonstrate electrostatic responses to acidic membranes (Figs. 4,7,and 8). This could be explained if peptide secondary structures segregate the acidic from the basic residues. Moreover, the electrostatic interactions of wild-type and mutant peptides were maximal at ionic strengths less than 20 mM. This observation suggests that the acidic residues in the polar face of the amphipathic helix (see Fig. 1, B and C) do not come into close contact with the negatively charged bilayer. Otherwise higher ionic strength would serve to mask their negative, repulsive charge, and the binding would show an optimum at medium ionic strength (e.g. 100 mM). This is a feature of the binding of some other membrane surface interacting peptides containing mixed charges (11,44). If peptide folding into an ␣-helix precedes surface attraction and insertion, this would place strips of basic charge flanking the hydrophobic wedge, and would position the negatively charged polar face away from the membrane surface (Fig. 1B). In this model the ␣-helical conformer of the peptide would form in solution, and the membrane would serve as a trap for this conformer. The alternative model, that the peptide is attracted electrostatically to the surface in an open, unfolded form and folds into a helix on the surface, seems less likely.
Contributions of Glutamates 257, 268, and 279 to Membrane Binding-The Glu 3 Gln substitution in the VEEKS peptide increased the hydrophobicity of the peptide, as evidenced by increased affinity for vesicles with low or zero negative charge (Figs. 7 and 10B) and increased ionic strength resistance (Fig.  8). The increased hydrophobicity of the glutamine-substituted domain M was also apparent when we examined the distribution of CT-WT and CT-3EQ between membrane and soluble fractions of transfected cells. The increased partitioning of CT-3EQ into the cellular particulate fraction was a result of its increased membrane affinity rather than its aggregation into an insoluble form since CT-3EQ had the same specific activity as CT-WT, suggesting a native fold. Moreover, the partitioning of CT-3EQ, like CT-WT, was influenced by the anionic lipid content of the cell membranes. This would be unlikely if it were an insoluble aggregate.
How much binding energy is associated with the Glu 3 Gln substitution? The substitution of the three glutamates with glutamine should theoretically increase the interfacial hydrophobicity by 4.3 kcal/mol (45). Protonation of the three glutamates should increase the interfacial hydrophobicity by 6 kcal/ mol (45). The partition coefficients and derived ⌬G values for the interaction of peptides 3E and 3EQ with 0 -50% PG vesicles clearly show the increase in the hydrophobic component upon charge neutralization of the three glutamates (Fig. 10B). Although the binding of Pep-3EQ to 100% PC vesicles yielded a reliable K x of (1.6 Ϯ 0.5) ϫ 10 4 by the filtration binding assay and (1.9 Ϯ 0.4) ϫ 10 4 by the fluorescence binding assay, the binding of Pep-3E to PC vesicles was below the detection limit of both assays. Thus we cannot directly obtain the increase in the hydrophobic component of binding. Extrapolation to 0% PG of the data describing Pep-3E binding in Fig. 10B might predict a ⌬⌬G associated with the Glu 3 Gln mutation of 4 -6 kcal/ mol. However, we feel the extrapolation is not reliable since the shape of this line at low mol% anionic lipid is unknown. The anionic lipid dependence is influenced not only by the electrostatic attraction of the interfacial lysines, but also by protonation of the glutamates.
These considerations provide an explanation for why glutamates have evolved at interfacial positions on the CT amphipathic helical membrane binding domain. If polar uncharged amino acids such as glutamine occupied these positions, the CT-membrane interaction would be too strongly hydrophobic to allow regulation by small changes in anionic lipid content. If lysines occupied these positions the electrostatic interaction might be too strong at low anionic lipid contents. By contrast, the interfacial glutamates ensure that CT is not bound to cell membranes until the anionic lipid content increases well above the homeostatic norm.
The binding scenario we envision is that domain M, in a transient ␣-helical conformation, is electrostatically attracted to enriched anionic lipid surfaces via the basic strip on one interface. The glutamates on the opposite interface become protonated in the low pH milieu at the anionic membrane surface (Fig. 2). Protonation eliminates the charge repulsion and increases the hydrophobicity of the domain, thus overcoming the barrier preventing insertion of the hydrophobic face of the amphipathic helix. Once inserted, the glutamic acids remain protonated during the protein residence time. Thus, the anionic membrane traps the protonated form. Therefore, there are two factors that increase the apparent pK a of the glutamates. The first is the negative surface potential that lowers the surface pH. The second is the lower dipole moment of the interfacial zone of the membrane. These two factors may account for the observed apparent pK a values in Fig. 5B.
Lipid interactions driven by protonation of protein acidic side chains have emerged as a regulatory principle in in vitro studies with other membrane proteins and peptides, where the lipid interaction is promoted by lowering the pH. Protonation drives a conformational change to a molten globule form prior to lipid insertion of such proteins as colicin A (46), cytochrome c (47), a trichosanthin toxin (48), and apolipophorin III (49). Low pH at the surface of anionic membranes was proposed to promote this protonation event. Other studies with peptides analogs of hemagglutinin fusion proteins (50,51) or bacteriorhodopsin membrane-spanning peptides (52) have identified cases where protonation of glutamates at low bulk pH promotes amphipathic ␣-helix formation and membrane insertion into zwitterionic PC vesicles. A correlation between anionic lipid content and the activity of several amino acid transporters in Escherichia coli (53) and yeast (54) was postulated to involve surface potential modulation of the pK a values of acidic residues on the transporters involved in substrate transport.
A role for anionic lipid in modulating the protonation state of a model membrane-interacting peptide emerged from studies with myristoylated model hexapeptides (55). A tryptophan residue adjacent to the C-terminal residue was inserted into PG/PC (1/1) vesicles at a higher bulk pH than observed for insertion into PC vesicles. This was attributed to protonation of acidic residues in the low interfacial pH at the surface of anionic lipid membranes. Our study has gone a step further in establishing this feature of anionic lipids in vivo by demonstrating that the engineering of mutations at the candidate glutamic acid sites that mimic their protonated form can greatly alter the affinity of an amphitropic protein for membranes in a living cell. Acidic residues are found in proximity to the anionic membrane-binding domains of such amphitropic proteins as Src (8) and MARCKS (4), which interact via a polybasic motif, and Dna A (56) and vinculin (57), which interact via an amphipathic helix. Whether protonation of these residues provides a mechanism for anionic membrane selectivity in these or other proteins remains to be established.
In summary we have identified two membrane negative charge sensors featured in domain M of CT; the interfacial basic strips contribute an electrostatic driving force for negatively charged surfaces, whereas the interfacial glutamates, like the three serines interrupting the non-polar face (27), reduce the binding affinity for zwitterionic membranes. The surface of anionic, but not zwitterionic membranes induces a modification of domain M (protonation of the glutamates), thereby increasing its net positive charge and hydrophobicity. We demonstrated that the interfacial glutamates serve as functional determinants of the CT membrane affinity in living cells, and this suggests that we have unraveled a bona fide mechanism for regulating the lipid selectivity of CT and for maintaining the anionic phospholipid/PC ratio at its functional optimum in cells.