An Experimentally Based Computer Search Identifies Unstructured Membrane-binding Sites in Proteins

Abstract

Programs exist for searching protein sequences for potential membrane-penetrating segments (hydrophobic regions) and for lipid-binding sites with highly defined tertiary structures, such as PH, FERM, C2, ENTH, and other domains. However, a rapidly growing number of membrane-associated proteins (including cytoskeletal proteins, kinases, GTP-binding proteins, and their effectors) bind lipids through less structured regions. Here, we describe the development and testing of a simple computer search program that identifies unstructured potential membrane-binding sites. Initially, we found that both basic and hydrophobic amino acids, irrespective of sequence, contribute to the binding to acidic phospholipid vesicles of synthetic peptides that correspond to the putative membrane-binding domains of Acanthamoeba class I myosins. Based on these results, we modified a hydrophobicity scale giving Arg- and Lys-positive, rather than negative, values. Using this basic and hydrophobic scale with a standard search algorithm, we successfully identified previously determined unstructured membrane-binding sites in all 16 proteins tested. Importantly, basic and hydrophobic searches identified previously unknown potential membrane-binding sites in class I myosins, PAKs and CARMIL (capping protein, Arp2/3, myosin I linker; a membrane-associated cytoskeletal scaffold protein), and synthetic peptides and protein domains containing these newly identified sites bound to acidic phospholipids in vitro.

Introduction

Recently, there has been considerable interest in characterizing protein domains, such as PH, FERM, C2, and ENTH, that are responsible for specific binding to membrane lipids (for review, see Ref. 1). These domains have highly defined tertiary structures comprising α-helices, β-sheets and loops, and there are multiple programs for recognizing them in protein sequences. However, a growing number of membrane-binding proteins, including cytoskeletal proteins (2), GTP-binding proteins (e.g. Rit), and GTP-binding protein effectors (e.g. some PAKs) (see references below), do not fit within these categories and bind membrane lipids through much less structured regions. Here, we describe the development and testing of a simple computer search program that identifies such potential membrane-binding sites. This novel search program evolved from our previous effort (3) to identify the membrane-binding site in the heavy chain of Acanthamoeba myosin IC (AMIC)² (4), which was known to bind to acidic phospholipids (3, 5, 6) and cell membranes (7–10) through its 220-residue basic region (5).

We had found (3) that AMIC binds nonspecifically to acidic phospholipid vesicles in proportion to their negative charge. Prompted by the report (11) that several proteins bind to acidic phospholipids through a basic-hydrophobic-basic (BHB) region consisting of two small clusters of basic amino acids separated by hydrophobic residues, we identified, by visual inspection, a 13-residue BHB sequence, KVKPFLYVLKRR, within the basic region of the AMIC heavy chain (3). A synthetic peptide with this sequence inhibited the binding of AMIC to acidic phospholipid vesicles with an IC₅₀ of 16 μm (3). The relative IC₅₀ values of synthetic peptides with the sequences of BHB segments in similar positions in the heavy chains of the two other Acanthamoeba class I myosins correlated with the localization of the myosins in either membranes (low IC₅₀ values) or cytoplasm (high IC₅₀ values) in the amoebae (3).

In the experiments reported in this paper, we first confirmed and extended the previous results by measuring directly the binding affinities for acidic phospholipids of the three Acanthamoeba myosin I peptides and one Dictyostelium myosin I peptide previously studied, peptides from two additional Dictyostelium class I myosins, and mutations of these peptides. We found that the basic and hydrophobic amino acid composition of the peptides, but not the sequence of the residues, was the determining factor for binding affinity. We then developed a computer search program for identifying protein regions enriched in basic and hydrophobic amino acids by changing the values for arginine and lysine from negative to positive in the Wimley and White (12) hydrophobicity scale. This new basic and hydrophobic (BH) scale, which successfully identified the known membrane-binding sites in 16 proteins, provides a useful method for rapidly identifying potential membrane-binding sites.

EXPERIMENTAL PROCEDURES

Phospholipid Vesicles

PS and PC were purchased from Avanti Polar Lipids (Alabaster, AL). Large unilamellar lipid vesicles containing either 50% PS and 50% PC, or 5% PIP₂, 25% PS, and 70% PC were prepared by repeated passages through an extruder fitted with two 100-mm membrane filters and repeated freezing and thawing cycles, as described previously (3).

Binding of Synthetic Peptides to Phospholipid Vesicles

Peptides with an N-terminal dansyl group were synthesized by EZBiolab (Carmel, IN). Peptide purity was higher than 95%, and molecular mass was confirmed by mass spectroscopy. Peptides, final concentration 1 μm, were mixed with varied concentrations of phospholipid vesicles in 20 mm Hepes buffer, pH 7.0, 100 mm NaCl, 1 mm EGTA, and 0.5 mg/ml bovine serum albumin. The amount of bound peptide was determined by the increase in the fluorescence intensity of the dansyl group (λ_ex, 335 nm; λ_em, 515 nm). K₅₀ for total lipid is the concentration of total phospholipid that resulted in 50% of the maximum increase in fluorescence. K₅₀ for accessible PS is the concentration of PS on the outer bilayer required for a 50% increase in fluorescence.

Binding of the AMIC-truncated Mutants to Phospholipid Vesicles

The T1, T2, and T4 fragments of AMIC heavy chain (AAC98089) with N-terminal FLAG were expressed in SF9 insect cells, with or without coexpression of light chain, and purified by anti-FLAG-affinity chromatography as described previously (13). Binding of the expressed proteins to 50% PS vesicles was performed as described (3). Briefly, the expressed proteins were mixed with various concentrations of vesicles in 20 mm imidazole, pH 7.0, 100 mm NaCl, 1 mm EGTA, and 0.5 mg/ml bovine serum albumin and centrifuged at 200,000 × g for 40 min at 20 °C. Aliquots were removed before and after centrifugation, and the myosin concentrations determined by measuring K⁺-ATPase activity.

Binding of CARMIL Fragment to Phospholipid Vesicles

The CAH3 fragment of mouse CARMIL (Q6EDY6) corresponding to residues 961–1085 of the full-length protein, with N-terminal GFP and C-terminal His tags, was expressed in Escherichia coli and purified by absorption on nickel resin as described by Fujiwara et al. (14). In brief, bacteria from a 150-ml expression culture were lysed by two rounds in a French press at 1,300 pounds per square inch in 10 ml of buffer A (phosphate-buffered saline buffer supplemented with complete EDTA-free protease inhibitor mixture (Roche Diagnostics). Triton X-100 was added to a final concentration of 0.4%, and the lysate was clarified by centrifugation at 12,000 × g for 15 min. The supernatant was mixed for 2 h with 1 ml of nickel-nitrilotriacetic acid agarose (Qiagen, Valencia, CA); the resin was washed with buffer A containing 20 mm imidazole, pH 7.5, and 300 mm NaCl; and the green GFP-mouse CAH3 (CARMIL homology 3 domain) fusion protein was eluted with 3 ml of buffer A containing 250 mm imidazole, pH 7.5. The concentration of monomeric GFP-mouse CAH3 was determined by Bradford assay. The purified monomeric GFP-mouse CAH3 fragment (0.6 μm) was then mixed with varied concentrations of 50% PS vesicles in the same buffer used for binding synthetic peptides and centrifuged as described above for the peptide binding assay. Aliquots were removed before and after centrifugation at 200,000 × g for 40 min at 20 °C, and the amount of unbound CAH3 fragment was determined by measuring GFP fluorescence (λ_ex, 490 nm; λ_em, 510 nm).

Computer Program for BH Search

As discussed briefly in “Results,” analyses were initially performed on the EMBOSS Pepinfo Web site using one of the value scales shown in Table 2. Beginning with the first residue in the protein sequence, the program averages the values in Table 2 for each amino acid in a segment of the selected length (the window size) and gives that value as the score for the middle residue in the segment (see Ref. 15). Thus, for a window of 19, the first score, which is the average of the values for residues 1–19, is given to residue 10. The program then averages the values for residues 2–20 and gives that score to residue 11, continuing in this manner until the end of the sequence with the last residue scored being 10 residues from the C terminus. Smaller windows must be used to score residues closer to the N and C termini. After BH search was developed, the EMBOSS program was modified to be more convenient for this application (http://helixweb.nih.gov/bhsearch).

TABLE 2.

Scales used for computer searches

Amino acid	K&Da	OMH	Consensus	W&W	BH
Ala	1.8	−0.4	0.62	−0.17	−0.17
Cys	2.5	0.17	0.29	0.24	0.24
Asp	−3.5	−1.31	−0.9	−1.23	−1.23
Glu	−3.5	−1.22	−0.74	−2.02	−2.02
Phe	2.8	1.92	1.19	1.13	1.13
Gal	−0.4	−0.67	0.48	−0.01	−0.01
His	−3.2	−0.64	−0.4	−0.17	−0.17
Ile	4.5	1.25	1.38	0.31	0.31
Lys	−3.9	−0.67	−1.5	−0.99	2.00
Leu	3.8	1.22	1.06	0.56	0.56
Met	1.9	1.02	0.64	0.23	0.23
Asn	−3.5	−0.92	−0.78	−0.42	−0.42
Pro	−1.6	−0.49	0.12	−0.45	−0.45
Gln	−3.5	−0.91	−0.85	−0.58	−0.58
Arg	−4.5	−0.59	−2.53	−0.81	2.00
Ser	−0.8	−0.55	−0.18	−0.13	−0.13
Thr	−0.7	−0.28	−0.05	−0.14	−0.14
Val	4.2	0.91	1.08	−0.07	−0.07
Trp	−0.9	0.5	0.81	1.85	1.85
Tyr	−1.3	1.67	0.26	0.94	0.94

^a References are as follows: K&D, Kyte and Doolittle (15); OMH, Sweet and Eisenberg (19); Consensus (19); W&W, Wimley and White (12); and BH (this paper; W&W scale with Lys and Arg values changed to 2; see “Results”).

RESULTS

Binding of Fluorescently Labeled Peptides to Acidic Phospholipid Vesicles

The fluorescence of an N-dansyl-peptide with the sequence of the previously identified BHB segment within the basic region of the AMIC heavy chain tail increased ∼10-fold when exposed to unilamellar vesicles composed of 50% PS and 50% PC (Fig. 1A). There was no significant increase in fluorescence when the peptide was exposed to vesicles composed of 100% PC (Fig. 1A) or when dansyl-l-glutamine was exposed to 50% PS or 100% PC vesicles (data not shown). These results are consistent with the increase in fluorescence of the dansyl-AMIC(tail) peptide being due to binding of the peptide to the acidic phospholipid vesicles. The K₅₀ for binding of the dansyl-AMIC(tail) peptide to 50% PS vesicles was 13 μm for total lipid or 3.3 μm for accessible PS (Fig. 1B and Table 1A). These values are similar to the K₅₀ of 4 μm for total lipid and 1.3 μm for accessible PS obtained previously (3) for binding of full-length AMIC to 50% PS vesicles. In the peptide-binding studies described in this paper, we used only vesicles consisting of 50% PS and 50% PC and will hereafter cite only the K₅₀ values for total lipid; the K₅₀ value for accessible PS is always 25% of the K₅₀ value for total lipid.

FIGURE 1.

Binding of dansyl-labeled synthetic AMIC(tail) peptide to acidic phospholipid vesicles. A, fluorescence spectra of dansyl-labeled synthetic peptide with the sequence of the BH region in the tail of AMIC (see Table 1) in the presence of buffer alone, or 500 μm phospholipid vesicles contain 100% PC, 100% PS or 50% PS and 50% PC. B, increase in fluorescence intensity of dansyl-AMIC(tail) peptide as a function of concentration of 50% PS vesicles. The open and closed symbols are two independent experiments. Excitation, 335 nm; emission, 515 nm.

TABLE 1.

Affinities of binding of dansyl-peptides to acidic phospholipid vesicles

Total lipid refers to total concentration of combined PS and PC. Accessible PS concentration is one-fourth of above because only half of the total lipid is PS and only approximately half of PS is on the outer side of lipid bilayer vesicles and accessible for binding of peptide. For A, these peptides were selected by visual identification of basic/hydrophobic/basic sequences within the basic regions of the tails of the myosin I heavy chains (1). For B, these peptides were selected from BH program searches of the proteins (e.g. Figs. 2 and 6).

Protein source	Peptide sequence	K50
		Total lipid	Accessible PS
		μm
A
AMIA(tail)	840KKKVATHVLDKK851	≥209	≥52
AMIB(tail)	785RKKKSGQVVYNLKRR799	20	5
AMIC(tail)	802KKVKPFLYVLKRR814	13	3.3
AMIC(scramble)	VKPKFRLKYRVKL	19	4.8
DMIB(tail)	801KKKVLVHTLIRR812	33	8.3
DMIB(hydro/Ala)	KKKAAAHTAARR	≥200	≥50
DMIB(basic/Ala)	AAAVLVHTLIAA	>500	>125
DMIC(tail)	849KKNLATYLLDRR860	≥113	≥28
DMID(tail)	802RKKRPWIYVQKRR814	8	2
B
AMIC(head)	616YRQVYDKFFYRYR628	22	5.5
AMIC(neck)	699RKTAMRKYYYEVKK712	43	10.8
CARMIL	22RKIKISVKKKVK33	14	3.5
CARMIL	1068RRSGFLNLIKSR1079	18	4.5

The K₅₀ values for binding of the dansyl-peptides of the corresponding regions of the heavy chains of AMIA, AMIB, and DMID to 50% PS vesicles (Table 1A) were proportional to the previously determined (3) abilities of the unlabeled peptides to inhibit binding of AMIC to phospholipid vesicles. For example, AMIA(tail) peptide, which was a poor inhibitor of AMIC binding to PS vesicles (3), bound only weakly in the direct binding assay (Table 1). The K₅₀ values of the dansyl-peptides of the BHB regions of the three Acanthamoeba myosins paralleled their preferred membrane (AMIB, AMIC) or cytoplasmic (AMIA) localization in cells (7–9). The cellular localizations of DMIB, DMIC, and DMID have not been clearly defined (16–18), but the data in Table 1 for the corresponding dansyl-peptides suggest that DMID and DMIB may have higher affinities for membranes than DMIC.

To test the relative importance of the basic and hydrophobic amino acids for binding of the dansyl-peptides to acidic phospholipids, we measured the binding of the DMIB(tail) peptide with Ala substituted for either all of the hydrophobic or all of the basic amino acids. Both substitutions substantially weakened binding. Basic amino acids seemed more important than the hydrophobic residues; compare the K₅₀ values of DMIB(hydro/Ala), DMIB(basic/Ala) and DMIB(tail) in Table 1A. These results are similar to those obtained previously for the ability of AMIC(tail) peptide to inhibit binding of AMIC (3). Although these peptides were originally tested as potential membrane-binding sites because of the basic-hydrophobic-basic patterns of their sequences, the affinity of the AMIC(tail) peptide was almost unchanged when the sequence was rearranged so that the hydrophobic and basic residues were in alternating positions, AMIC(scramble) (Table 1A). Thus, the composition, and not the sequence, of the residues is the principal determinant for peptide binding. These results also argue that the hydrophobic residues do not form a loop that inserts deeply into the membrane bilayer, but rather that the side chains of the hydrophobic residues interact individually at the interface between the polar head groups and the acyl chains of the phospholipids.

Designing the BH Search Program

More than 20 different amino acid hydrophobicity scales have been developed for searching protein sequences for patches of hydrophobic amino acids; see, for example, the ExPASy ProtScale Web site and the EMBOSS Pepinfo Web site. The hydrophobicity scales are useful for identifying protein regions that may penetrate deeply into the membrane but have limited application for identification of potential membrane surface-binding sites because the hydrophobicity scales do not consider the contribution of basic residues, which, as discussed in the preceding section, are important for protein binding to negatively charged regions in phospholipid vesicles and cell membranes.

To design a new search program that would account for the contribution of both basic and hydrophobic amino acids to membrane binding, we chose to start with the Wimley and White (W&W) hydrophobicity scale (12) because it was developed specifically for application to proteins at membrane interfaces by determination of the free energy of transfer of model peptides from a lipid bilayer to water. However, because neutral lipid bilayers were used in those experiments (appropriately for the determination of hydrophobicity uninfluenced by electrostatic interactions), the hydrophilic positive charges of Lys and Arg hindered, rather than contributed to, lipid binding. Therefore, based on our data that Lys and Arg can be at least as important as hydrophobic amino acids for binding of peptides to negatively charged lipid vesicles (Table 1A), we modified the W&W scale by giving Lys and Arg values of +2, which is approximately equal to the value for the most hydrophobic amino acid, Trp, and equal but opposite to the value for the least hydrophobic amino acid, Glu, in the W&W scale.

Application of BH Search Program to Acanthamoeba Class I Myosins

We first compared the results of the EMBOSS Pepinfo analysis of AMIC heavy chain using our BH scale to the results using the four hydrophobicity scales in Table 2, with a window of 19 (see “Experimental Procedures”). The BH search showed three distinct peaks rising above a BH score of 0.6 (Fig. 2); such clearly defined peaks were not identified by the searches based on the four hydrophobicity scales in Table 2 (Fig. 2). The most C-terminal of these peaks, with peak residue at Leu⁸¹¹, was expected because we had shown that the synthetic peptide corresponding to residues Lys⁸⁰²-Arg⁸¹⁴ binds to acidic phospholipid vesicles (Table 1A and Ref. 3). The two other peak regions in Fig. 2, with peak residues Val⁶¹⁹ and Arg⁶⁹⁹, are in the head and neck domains of the AMIC heavy chain (see Fig. 4A), respectively, neither of which was known to bind to membranes. Interestingly, the potential membrane-binding site in the neck, residues Lys⁶⁹⁰-Leu⁷¹⁸, overlaps the IQ domain, residues Ile⁶⁹³-Lys⁷⁰⁷, to which the AMIC calmodulin-like light chain binds (20).

FIGURE 2.

Comparison of hydrophobicity plots and BH plot of the Acanthamoeba myosin IC heavy chain. The amino acid values for the five different scales are given in Table 2. A window size of 19 was used for all plots.

FIGURE 4.

Binding of truncated Acanthamoeba myosin IC constructs to acidic phospholipid vesicles. A, schematic representation of the structure of the AMC heavy chain and the truncated constructs T4, T2, and T1. The head domain contains the actin-activated ATPase site. The neck domain contains the binding site for a single light chain. The tail domain contains an N-terminal basic region, two Gly/Pro/Ala-rich regions (G), and an SH3 domain (S). The BH region (bar) is located within the basic region. T4 comprises the head, neck and basic regions; T2 comprises the head and neck regions, and T1 is the head region only. B and C, binding of AMIC constructs to 50% PS vesicles (B) and 5% PIP₂ and 50% PS vesicles (C). Binding was assayed by pelleting the vesicles assay and determining the percentage of unbound ATPase activity in the supernatant (see “Experimental Procedures”). T4 (filled circles) was coexpressed and purified with light chain, T2 both with (filled triangles) with and without (open triangles) light chain, and T1 (open circles) without light chain.

A BH search of the AMIB heavy chain (Fig. 3) gave very similar results, with peaks residues at Glu⁶¹⁰/Phe⁶¹¹, Arg⁶⁹², and Gal⁷⁹⁰. In contrast, the BH plot for AMIA (Fig. 3) showed only a few minor peaks at Pro⁶¹9, Arg⁷¹⁴/Leu⁷¹⁷, and Ala⁹⁰⁶/Glu⁹⁰⁷. The results of the BH plots for the three Acanthamoeba myosin I heavy chains are consistent with the predominantly membrane localizations of AMIB and AMIC and the predominantly cytoplasmic localization of AMIA (7–9).

FIGURE 3.

Comparison of BH plots of heavy chains of Acanthamoeba myosins IA, IB, and IC.

To assess the validity of the identification of the segments in the head and neck regions of AMIC as potential membrane-binding sites, we first determined the affinity for acidic phospholipid vesicles of synthetic dansyl-peptides corresponding to sequences from those regions. Both the AMIC(head) and AMIC(neck) peptides had relatively low K₅₀ values (Table 1B), as expected given their high content of basic and hydrophobic amino acids, although higher than the K₅₀ of the AMIC(tail) peptide (Table 1A). To extend the analysis to the protein level, we compared the binding to acidic phospholipid vesicles of three constructs of AMIC with truncated heavy chains (Fig. 4A) (13, 21): T4, which includes the head, neck, and basic region of the heavy chain (but not the C-terminal Gly/Pro/Ala and SH3 domains); T2, which includes only the head and neck domains; and T1, which includes only the head domain. T4 was expressed with the myosin light chain (20) that binds to the neck domain, and T2 was expressed both with and without the myosin light chain.

As shown previously (3), T4 bound to 50% PS vesicles with a K₅₀ of ≤5 μm (Fig. 4B), similar to the K₅₀ for the dansyl- AMIC(tail) peptide (Table 1A). T2, with or without light chain, also bound to the vesicles, but significantly less well, with a K₅₀ of ∼110 μm (Fig. 4B), similar to the affinity of the dansyl-AMIC(neck) peptide (Table 1B). Thus, the light chain does not block the phospholipid-binding site. Although the neck region bound only weakly to the PS vesicles, this weak binding might be sufficient to contribute to the binding of full-length AMIC by acting cooperatively with the stronger binding site in the tail. T1 bound with very much lower affinity, K₅₀ > 500 μm (Fig. 4B), and much more weakly than the dansyl-AMIC(head) peptide (Table 1B), most likely because the affinity of the potential phospholipid-binding site in the head was affected by the secondary and tertiary structure of the protein. Very similar results were obtained for binding of T4, T2, and T1 to 5% PIP₂ and 25% PS vesicles (which have the same net negative charge as 50% PS vesicles); the K₅₀ values were ≤5 μm for T4, ∼94 μm for T2, with and without light chain, and >500 μm for T1 (Fig. 4C). Thus, and consistent with previous results (3), there appears to be no phospholipid specificity for binding. It is noteworthy that this is the first piece of evidence that a region in the neck domain of an amoeba class I myosin may contribute to its association with membranes.

Application of BH Search to Proteins with Known Membrane-binding Sites

Although the data for the Acanthamoeba class I myosins were indicative of the potential usefulness of BH search for other proteins, it was necessary to test the ability of BH search to identify known unstructured binding sites in membrane-associated proteins. We tested 16 proteins (Table 3) in which one or more such membrane-binding sites have been defined. We used a window of 19 (see “Experimental Procedures”), except for two proteins for which a window of 11 was used: cortexillin, where the lipid-binding site is extremely C-terminal, and gelsolin, where the lipid-binding site(s) consists of multiple regions.

TABLE 3.

Published lipid-binding regions found by BH search

Protein, source (accession no.)	Published binding sites (in boldface)a	Refs.	Found by BH searchb	Peak residue(s) of regions found by BH search		Window size	Comments
				Published regions	Additional regionsc
Actophorin (cofilin), Acanthamoeba (AAA02909)	93KSKMMYTSTKDSIKK107	22	+	Thr101		19	Overlaps with actin-binding site
CapZα, mouse (P47753)	255FKALRRQLPVTRTKIDW271	23	++	Val264	Arg172	19
Cortexillin I, Dictyostelium (AAB62275)	422MKLLNQKEDDLKAQKLKSSKSKK444	24	+	Lys438	Lys53, Tyr111	11	Published site is C-terminal and can be detected only with small window
		−			Ile49, Thr107	19
EGFR, bovine (XP_592211)	646RRRHIVRKRTLRRLLQ661	25	+++	Val651		19	Overlapping lipid and calmodulin binding sites
GAP43 (neuromodulin), bovine (AAO60065)	30KAHKAATKIQASFRGHITRKKLKGEKK56	26	+++	Ile46		19	IQ site binds calmodulin and lipid
Gelsolin, human (BAH14236)	135KSGLKYKKGGVASGF149	27	++	Lys141, Gal137		19	Gelsolin binds phospholipids through three or more cooperating sites
	150KHVVPNEVVQRLFQVKGRRVVR172	28	+	Lys166	Lys272, Phe316, Phe365	11
	620GKAAYRTSPRLKDKK634	29	+	Arg629	Lys272, Phe316, Phe365	11
Gic1, Baker's yeast (P38785)	89SNHKSLTNKKKNFLGMFKKKDLLSRRHGSA117	30	+++	Lys106		19
Gic2, Baker's yeast (Q06648)	96SSSSSSSANKTNHKKVFLKLNLLKKKLLGAQPD128	30	+++	Lys114		19
MARCKS, bovine (P12624)	151KKKKKRFSFKKSFKLSGFSFKKNKK175	31, 32	+++	Lys160		19	Overlapping lipid and calmodulin binding sites
Myosin VI, chicken (Q9I8D1)	1107RRLKVYHAWKSKNKKR1122	33	++	Ala1114	Ser802, Gal839	19	Additional regions are in myosin neck
NMDA NR1, human (NP_015566)	837YKRHKDARRKQMQLAFAAVNVWRKNLQ863	26	+/−	Tyr837		19
	875KKKATFRAITSTLASSFKRRRSSK898		++	Ser885		19	Binds calmodulin
N-WASP, bovine (BAA11082)	186KEKKKGKAKK195	34	+++	Gal191	Asp136	19	Very basic, no hydrophobic residues
PLCζ1, mouse (Q8K4D7)	374KKRKKMKIAMA385	35	++	Ala383		19	Very basic, only one hydrophobic residue
Rit, human (Q92963)	195KKSKPKNSVWKRLKSPFRKKKDSVT219	11	+++	Lys205		19	BH search also finds lipid-binding sites identified in the same ref. for GEM, Rem, Rin, and Rad
STE5, Baker's yeast (P32917)	42LSPLSRGKKWTEKLARFQRSSAKKKRFSP70	36	++	Phe58		19
STE20, Baker's yeast (Q03497)	277KSYYSSSSKKRKSGSNSGTLRMKDVFTSFVQNIKR311	30	+	Ser281	Arg599	19	BH search finds low, broad peak that includes published site

^a When the amino acid numbers cited in paper and GenBank are different, numbers are from GenBank sequence.

^b The + and − symbols refer to the ability of BH search to find the published binding site. Multiple plus signs mean that site is easily identified in the BH plot by a combination of peak height and width compared to other regions in the plot.

^c Additional sites found by BH search are listed only if the size of the peak is at least equal to peaks corresponding to published sites found by BH search of the same protein.

The 16 proteins in Table 3 fall into diverse categories: structural and regulatory cytoskeletal proteins (actophorin, CapZα, cortexillin, gelsolin, MARCKS, myosin VI, N-WASP); GTP-binding proteins and their effectors (Rit, Gic1, Gic2, and STE20); kinases (STE20, EGFR); a scaffolding protein (Ste5); calmodulin-binding proteins (GAP43, MARCKS, NMDA-NR1, EGFR); and a phospholipase (PLCζ1). BH search detected all of the previously identified sites, and identified additional potential membrane-binding sites in 6 of the 16 proteins (Table 3). These latter sites need to be confirmed by appropriate experiments as both false positives and false negatives are possible. Representative BH plots are shown in Fig. 5.

FIGURE 5.

BH plots of proteins with known membrane-binding sites. N-WASP, EGFR, PLCζ1, and myosin VI were run with window size 19, and gelsolin was run with with window size 11.

The identified binding sites can comprise basic residues only (N-WASP), or almost only (PLCζ1), or a mixture of basic and hydrophobic residues (EGFR, myosin VI, Rit). BH search provides a means for identifying short basic-hydrophobic regions, which are difficult to identify by previously available methods, and which may be common in cortical proteins (30). The lipid-binding sites identified in GAP43 (Table 3) and the myosin I neck domain (Fig. 3) overlap with calmodulin- and light chain-binding sites, respectively. Similarly, there is independent experimental evidence that myristoylated alanine-rich C kinase substrate and possibly EGFR, GAP3, and NMDA-NR1 bind calmodulin and PIP₂ through the same basic-hydrophobic region (37). In addition, the membrane-binding sites of actophorin (22) and CapZα (23) overlap with actin-binding sites.

Application of BH Search to Proteins with Undetermined Membrane-binding Sites

In addition to the three Acanthamoeba myosins shown in Fig. 3, BH searches were performed on 24 other class I myosins with the following results. BH regions, with varied scores, were identified in the tails of Dictyostelium myosins IA–IF (P22467, P34092, P42522, P34109, Q03479, and P54695, respectively), and Schizosaccharomyces pombe Myo1 (Q9Y7Z8), Saccharomyces cerevisiae MYO3 (P36006), Candida albicans Myo5 (Q59MQ0), Aspergillus myosin IA (1C4A5), chicken brush border myosin IB (Q90748), mouse myosin IF (Q811E7), rat Myo1b (Q05096) and class I myosins in Penicillium (B6H033), Caenorhabditis elegans (Q19901), Entamoeba (B0EJD8), and Xenopus (Q6GME8). BH regions were not found in the tail regions of class I myosins from bovine (P10568), Neurospora (Q7SDM3) or fungi (Coprinopsis (A8N2Y6), Cryptococcus (Q5K8T7), Malassezia (A8PWF6), Lacarria (B0CRJ3)), but potential membrane-binding sites were found in the neck (IQ) domains of these myosins. A BH region was not found in the tail of human Myo1c (O00159), which agrees with the published data that the probable membrane-binding site in the basic region of the Myo1c tail is a structured putative PH domain (38), but BH search did identify a potential membrane-binding site in the IQ region of human Myo1c. BH searches did not find any potential membrane-binding sites in the tail of striated muscle myosin II.

We also performed BH searches on three PAK kinases: Acanthamoeba PAK (AAD11799) and Dictyostelium PAK (AAC71063), both of which are myosin I heavy chain kinases (39, 40) and mammalian PAK1 (P35465). All three PAKs bind acidic phospholipids (39–44), and Acanthamoeba PAK binds to plasma membranes in vivo and in vitro (45). BH search identified strong potential lipid-binding regions located N-terminal to the CRIB domains in all three PAKs (Fig. 6), consistent with localizations previously suggested (40, 41, 43). Interestingly, BH search also identified strong potential lipid-binding sites N-terminal to CRIB domains in Drosophila PAK A (NP_731073) and C. elegans PAK A (AAA68805). Thus several PAKs, in addition to STE20 (Table 3 and Ref. 30), have the potential to bind membranes through a BH region. There are exceptions, however, e.g. yeast Cla4, which binds lipids through a PH domain (46).

FIGURE 6.

BH plots of Acanthamoeba, Dictyostelium, and mammalian PAKs. The positions of the CRIB domains (*) are Acanthamoeba PAK-(93–106), Dictyostelium PAK-(356–369), and mammalian PAK-(75–88).

CARMIL is an interesting membrane-associated cytoskeletal scaffold protein with binding sites for actin capping protein, Arp2/3 and myosin I (47). BH search revealed potential membrane-binding regions at the N-terminal and C-terminal ends of mouse CARMIL (Q6EDY6) with peaks at Lys²⁵, Arg⁹³¹, Leu⁹⁹², and Leu¹⁰⁷³ (Fig. 7A). Synthetic dansyl-peptides with sequences from N-terminal (Arg²²-Lys³³) and C-terminal (Arg¹⁰⁶⁸-Arg¹⁰⁷⁹) regions bound to acidic phospholipid vesicles with high affinity (Table 1B). The N-terminal region is strongly basic and readily identifiable as a potential membrane-binding site by visually scanning the protein sequence. The C-terminal region is less obvious, consisting of both basic and hydrophobic residues, and is located within the CARMIL homology 3 domain (CAH3) that is a binding site for actin capping protein (48). For these reasons we focused on the lipid-binding properties of the C-terminal site, and expressed a GFP fusion protein of the entire 125-residue CAH3 domain (Pro⁹⁶¹-Pro¹⁰⁸⁵), which accounts for 9% of the total CARMIL sequence, including the last two of the three C-terminal potential membrane-binding peaks identified by the BH search. GFP-CAH3 bound to acidic phospholipid vesicles with a K₅₀ of ∼10 μm (Fig. 7B). It is likely, therefore, that one or more of the C-terminal regions identified by the BH search is involved in the association of CARMIL with the plasma membrane.

FIGURE 7.

BH plot and phospholipid-binding of CARMIL. A, schematic representation of domain structure of mouse CARMIL 1: leucine-rich repeat (LRR); CARMIL homology domain 3, which binds actin capping protein, (CAH3); proline-rich region (P). B, BH plot with window size 19. C, binding of GFP-CAH3 to 50% PS vesicles. CAH3 was expressed with N-terminal GFP. Binding was assayed by pelleting the vesicles and measuring GFP fluorescence of the supernatant (see “Experimental Procedures”).

DISCUSSION

Based on our experimental data showing that both basic and hydrophobic amino acids contribute to the affinity of synthetic peptides for acidic phospholipid vesicles, as quantified by the increase in fluorescence of an N-terminal dansyl-tag, we designed a computer search program (BH search) for potential membrane-binding sites by replacing the negative values for Arg and Lys in the Wimley and White (12) hydrophobicity scale with positive values. The BH search data in this paper were obtained by substituting this new BH scale for one of the hydrophobicity scales on the EMBOSS Pepinfo Web site, but BH search is now available on its own user-friendly Web site (http://helixweb.nih.gov/bhsearch), which links to EMBOSS Pepinfo.

A window size of 19 and a value of 2 for Arg and Lys, with a threshold BH value of 0.6 were the optimal parameters for identifying the maximum number of known lipid-binding sites for the 16 proteins in Table 3, representative of different protein families. As discussed by the authors of the original hydropathy plot program (15), small window sizes give noisier plots and large window sizes may miss sites near the N or C terminus (with a window size of 19, the first and last amino acids scored are 9 residues from the N and C termini) and short/weak internal sites. For example, with a window of 19, the published lipid-binding site of Rit1 (11) was the only major peak (Table 3 and supplemental Fig. S1). Lowering the window size to 7 increased the height of several other peaks so that the peak corresponding to the known site no longer dominated (supplemental Fig. S1), while raising the window to 29 missed the C-terminal 14 residues that contribute to the peak at the lipid-binding site. Correspondingly, a smaller window (size 11) was necessary to detect known lipid-binding sites that are close to the N or C terminus of other proteins, for example cortexillin 1 (Table 3) and relatively weak sites, for example gelsolin (Table 3).

As expected, values for Arg and Lys lower than 2 disfavored identification of lipid-binding sites with a strong basic component, so that known lipid-binding sites were either not detected or became much less distinct, for example PLCζ1 (supplemental Fig. S2). Increasing the values for Arg and Lys to 4 did not affect identification of lipid-binding sites for the proteins in Table 3, but it became much more difficult to establish a common threshold above which a peak could be considered to be a potential lipid-binding site, e.g. compare the plots with a value of 4 for actophorin and Rit1 (supplemental Fig. S3). In both cases, the peak corresponding to the single known lipid-binding site can be detected, but to do so unambiguously requires a threshold for Rit1 of 1.5 that is higher than the principal peak for actophorin.

With a value of 2 for Arg and Lys and a window size of 19 (with the exceptions of cortexillin and gelsolin previously noted) a single threshold of 0.6 identified all of the known lipid-binding sites for the 16 proteins in Table 3, and potential lipid-binding sites in the class I myosins, PAKs and CARMIL. An equally useful single set of parameters was not identified when we substituted the Kyte and Doolittle (15) or Sweet and Eisenberg (19) hydrophobicity scales (Table 2) for the Wimley and White (12) hydrophobicity scale. However, users of BH search should recognize that different parameter values may be useful for analysis of other proteins.

One should also keep in mind that BH search identifies only potential lipid-binding sites; both false positives and false negatives are to be expected. Therefore, predicted lipid-binding sites should be confirmed by experimental data. We recommend that BH search be used for proteins that are known to bind to membranes but do not have a recognizable, specialized lipid-binding domain. Regions consisting of mixtures of basic and hydrophobic residues often form inter- or intramolecular protein-protein interaction sites, which are likely be the main source of false positives, especially if the sites are buried within the protein molecule. For example, the site identified in the head domain of AMIC (Fig. 3), which did not bind to lipid vesicles (Fig. 4, B and C), might be such a site because it lies within the converter domain, which has been shown to be involved in dynamic intramolecular interactions in other myosins (49).

The 16 proteins analyzed in Table 3 contain a total of 19 known lipid-binding sites. With a window of 19, BH search identified 16 of the known sites (albeit one of the NMDA NR1 sites was borderline positive) and 6 previously unidentified sites. Thus, there were 3 false negatives and a maximum of 6 possible false positives. The false negatives were 2 of the 3 known sites in gelsolin and the 1 known site in cortexillin, all of which were identified when a window of 11 was substituted for a window of 19. There were no false negatives with a window of 11, but there were a maximum of 5 possible false positives, 3 for gelsolin and 2 in cortexillin that were also identified with a window of 19. Of course, some or all of the 9 possible false positives in Table 3 might be previously unidentified lipid-binding sites and not false positives. From these data, we would suggest using a window of 19 for the initial BH search, and then a window of 11 if no potential lipid-binding sites are found or to search for sites close to the N and C termini.

As expected, BH search did not recognize highly structured lipid-binding sites such as PH, FERM, C2, and ENTH domains, and we ran at least two representatives of each family; however, BH search recognized a short basic-hydrophobic region within some PH domains (data not shown). Somewhat surprisingly, BH search recognized lipid-binding sites in α-helical structures (for example, IQ domains within the neck domains of class I myosins) and identified a lipid-binding site consisting of several regions (see gelsolin in Table 3).

BH search successfully identified the known membrane-binding sites of 16 proteins, which serves to document the program's usefulness. Also, BH search identified potential membrane-binding sites in the tails of Acanthamoeba myosin IB and IC, which are known to binding to cell membranes in situ, but not in Acanthamoeba myosin IA, which is principally, if not exclusively cytoplasmic. In addition, previously unknown possible membrane-binding sites were identified in the neck domains of Acanthamoeba myosins IB and IC, and truncated myosin IC heavy chain and a synthetic peptide with the corresponding sequence were shown to bind with relatively high affinity to acidic phospholipid vesicles. Similarly, the BH search identification of a previously unknown potential membrane-binding site in the C-terminal half of mouse CARMIL was supported by phospholipid-vesicle binding data for both a short synthetic peptide and a 125-residue protein fragment of CARMIL.

The results of multiple additional searches support several interesting generalizations. Calmodulin-binding IQ domains had consistently high BH scores, i.e. were identified as potential membrane-binding sites. This may just be coincidence resulting from an affinity of calmodulin for sites composed of basic and hydrophobic amino acids, but it suggests the possibility that IQ domains may also function as membrane-binding sites, possibly regulated by competition with calmodulin or calmodulin-like proteins. Calmodulin-binding and lipid-binding sites are known to overlap in several proteins (37) including class I myosins (50) and PAKs (40, 42), and in some cases, binding of lipid and calmodulin to overlapping sites is mutually exclusive, for example, myristoylated alanine-rich C kinase substrate (37), Acanthamoeba PAK (42) and Dictyostelium PAK (40). IQ domains and other calmodulin-binding sites are usually α-helical, and binding of lipids to α-helices has been predicted by others (51).

In fact, lipids have been shown to bind to the neck (IQ) regions of some vertebrate class I myosins (38, 50, 52–55). Although lipid binding was enhanced by Ca²⁺-induced dissociation of bound calmodulins, even in the absence of Ca²⁺ (i.e. when all the calmodulin was present), Myo1c bound to 60% PS vesicles with an effective dissociation constant of 37 μm (38), not very different from the K₅₀ of ∼110 μm that we found for the binding of the T2 fragment (head and neck) of AMIC with or without bound light chain to 50% PS vesicles (Fig. 4B). Our observation that bound light chain had no effect on lipid-binding indicates that the single bound light chain of AMIC and lipid-binding are not mutually exclusive. These data suggest that, although the tails of at least some vertebrate class I myosins and AMIC contain a strong lipid-binding site, lipid-binding regions in their necks may play a supportive role in the binding of these myosins to cell membranes (38, 54).

In addition, different isoforms within the same protein family often have either a BH region or a more specialized lipid/membrane-binding domain in homologous positions, presumably fulfilling similar roles. For example, many (but not all) of the multiple class I myosins, from yeast to mammals, have a BH region within the basic region of the heavy chain tail (this paper; ref. 3) where mammalian Myo1c has a putative PH domain (38, 56); yeast Cla4 (a PAK) has a PH domain N-terminal to its CRIB domain (46) where Acanthamoeba PAK and yeast STE20 (also a PAK) have BH regions (Ref. 30 and this paper); and PLCδ has a PH domain but PLCζ1 has a BH region (35, this paper).

Membrane localization of a protein in vivo may depend on more than one binding site. A protein can have more than one membrane lipid-binding site (for example Acanthamoeba myosin IB and IC and gelsolin), or both a membrane lipid-binding site and a membrane protein-binding site. Examples of the latter are the yeast scaffold protein Ste5, which binds to membrane lipids and to receptor-activated Gβγ (36), and yeast PAK Ste20, which binds to membrane lipids and, through a CRIB domain, to membrane-associated Cdc42 (30). CARMIL might also have dual membrane-binding capability. As shown in this paper, CARMIL has potential membrane-lipid binding sites, but CARMIL also binds to the SH3 domain of long-tailed amoebae class I myosins (47), which are associated with the plasma membrane (7–10). Multiple membrane-binding sites could both amplify the affinity and allow fine-tuning of membrane binding.

Interestingly, slight modifications of a membrane-binding site can “direct” the protein to different membranes. For example, hydrophobic residues within a polybasic amino acid cluster direct Rit and Rin to the plasma membrane of mammalian cells but a single point mutation of a hydrophobic residue within the BHB region, Trp/Ala, causes strong nuclear localization (11).

In summary, we have quantified the affinity to acidic phospholipid vesicles of synthetic peptides containing basic and hydrophobic amino acids (by measuring the increase in fluorescence of an N-terminal dansyl group attached to the peptide) and found that the affinity depends on the peptide content of both basic and hydrophobic amino acids, but not on the amino acid sequence. We then modified the hydrophobicity scale of Wimley and White (12), giving Arg and Lys values of +2, to create a BH scale that serves as the basis for a computer search for unstructured potential membrane-binding sites (http://helixweb.nih.gov/bhsearch). We demonstrated that BH search correctly identifies known membrane-binding sites in 16 diverse proteins and previously unknown membrane-binding sites and thus provides a useful experimental tool for identifying membrane-binding sites that do not fall into any category of highly structured membrane-lipid binding domains.