Insertion and Organization within Membranes of the δ-Endotoxin Pore-forming Domain, Helix 4-Loop-Helix 5, and Inhibition of Its Activity by a Mutant Helix 4 Peptide*

The pore-forming domain of Bacillus thuringiensis Cry1Ac insecticidal protein comprises of a seven α-helix bundle (α1–α7). According to the “umbrella model,” α4 and α5 helices form a hairpin structure thought to be inserted into the membrane upon binding. Here, we have synthesized and characterized the hairpin domain, α4-loop-α5, its α4 and α5 helices, as well as mutant α4 peptides based on mutations that increased or decreased toxin toxicity. Membrane permeation studies revealed that the α4-loop-α5 hairpin is extremely active compared with the isolated helices or their mixtures, indicating the complementary role of the two helices and the need for the loop for efficient insertion into membranes. Together with spectrofluorometric studies, we provide direct evidence for the role of α4-loop-α5 as the membrane-inserted pore-forming hairpin in which α4 and α5 line the lumen of the channel and α5 also participates in the oligomerization of the toxin. Strikingly, the addition of the active α4 mutant peptide completely inhibits α4-loop-α5 pore formation, thus providing, to our knowledge, the first example that a mutated helix within a pore can function as an “immunity protein” by directly interacting with the segments that form the pore. This presents a potential means of interfering with the assembly and function of other membrane proteins as well.

The major defensive and offensive strategies chosen by bacteria are based on the interaction of toxins with cellular membranes (1)(2)(3). This interaction frequently results in membrane permeation or translocation into the cytoplasm (1, 4 -6). Consequently, toxins need to be soluble to reach their target cells. However, once the target is reached, the toxins are transformed into membrane proteins (highly insoluble by definition). Accumulating data reveal that the domain that interacts with the membrane uses one of two structural strategies (5,3). In the first and best known group of toxins, the membrane-interacting domain contains hydrophobic and amphipathic ␣-helices (1,7,8). In the soluble stage, the hydrophobic helices are sheltered from the solvent by a barrel of amphipathic helices. The second group of toxins is intrinsically soluble, and its structure is mainly composed of a ␤-barrel fold (9). These toxins depend on oligomerization to produce sufficient hydrophobicity for mem-brane insertion to occur.
Bacillus thuringiensis, a ubiquitous Gram-positive bacterium, produces parasporal crystals (10). Within these crystals lie the ␦-endotoxins, which are specific insect toxins. The ␦-endotoxins are digested by the insects in the midgut and are enzymatically modified into their active form (6), after which the toxins bind to target cells through a two-stage process to form channels/pores (11,12). The first step involves reversible binding to a receptor followed by an irreversible step in which membrane insertion occurs (3,6,13,14). The irreversible step is also implicated with toxicity; however, the mode of action of the toxin is not yet clearly understood. Biophysical and structural studies of the ␦-endotoxins as a paradigm for membrane toxins may help elucidate the specific interactions, mechanism of insertion, and organization within the membrane of toxins and channel proteins.
A major part of the current knowledge concerning the mechanism of ␦-endotoxin toxicity is through experience with three major proteins of the Cry family: Cry3A, Cry1Aa, and Cry1Ac. The Cry proteins are an extensive toxin family found within the ␦-endotoxin crystals. Their pore-forming domain comprises both amphipathic and hydrophobic ␣-helices. The Cry3A and Cry1Aa structures were resolved in aqueous solution; Cry1Ac protein is highly homologous to Cry1Aa (7,8). Furthermore, many mutations were done on the full-length proteins, including specific mutations in the ␣4 and ␣5 regions.
Currently, the umbrella model best explains the Cry mechanism of toxicity. Briefly, the protein goes through a metamorphosis upon membrane binding in which structural rearrangement occurs (7,15,16). Presumably, this is followed by a hydrophobic hairpin insertion into the phospholipid bilayer, whereas the amphipathic helices are spread on the surface. Later, oligomerization occurs, and a channel or pore is formed. Two highly hydrophobic helices of the Cry3A domain I, namely ␣4 and ␣5, were each shown to form homo-and hetero-oligomers in the membrane milieu (17)(18)(19). Helix ␣5 of Cry3A (17) and Cry1Ac (20) were shown also to have membrane permeation capabilities. These data suggest that ␣4 and ␣5 might be the region that is inserted into the membrane and actually participates in pore formation. This model is supported by mutagenesis analysis done on full-length proteins, where loss of toxicity is a common characteristic of most ␣4 and ␣5 mutations (16,(21)(22)(23). Previous studies indicate that although mutations in ␣4 usually cause loss of toxicity, they do not change the aggregation capabilities of the toxin (23). On the other hand, a strong correlation was found between toxicity and ␣5 oligomerization. Specifically, for mutations that did not affect toxicity, the protein retained its oligomerization ability, excluding one known exception H168R (23). Recent results based on site-directed mutagenesis place ␣4 in the aqueous interface of the pore, whereas the more hydrophobic ␣5 is located at the back of the oligomer and interacts mainly with ␣4 and the lipid bilayer (16). Furthermore, it was shown that mutations I132S and Q133R, located on the polar surface of ␣4, substantially decrease the pore-forming activity, whereas F134L results in a 3-fold increase in activity (23). However, there is still no direct evidence that ␣4 inserts together with ␣5 as a hairpin into the membrane to form a pore.
To directly support this hypothesis, we have synthesized and characterized the corresponding peptides derived from Cry1Ac, namely, ␣5, ␣4, and its active and non-active mutants and the full-length membrane-inserted domain ␣4-loop-␣5 (Table I).
Note that the benefits of studying the membrane-inserted segment of a protein were clearly demonstrated by revealing the crystal structure of the KcsA potassium channel, which was done using only the membrane-inserted segments, helix M1pore region-helix M2, of the channel (24). Importantly, the structure and the organization of this region were in agreement with studies carried out with the isolated helix M1, the pore region H5, and the helix M2 (25,26).
Membrane permeation studies, binding experiments, and competition assays between the ␣4-loop-␣5 and the ␣4 and ␣5 peptides provide direct evidence of the role of ␣4-loop-␣5 as the membrane-inserted pore-forming hairpin in which ␣4 and ␣5 lines the lumen of the channel, and ␣5 participates also in the oligomerization of the toxin. Furthermore, the striking finding that an ␣4 mutant can inhibit ␣4-loop-␣5 pore formation provides, to our knowledge, the first example that a mutated helix of a pore-forming toxin can function as an immunity protein by directly interacting with the segments that form the pore, similar to what has been proposed for the immunity proteins of colicin E1 (27). This presents a potential means of interfering with the assembly and function of other membrane proteins as well. These results are discussed also with regard to the importance of understanding protein-protein interaction within the membrane milieu (28,29).
Peptide Synthesis, Fluorescent Labeling, and Purification-The peptides were synthesized by a standard solid phase method on peptidylglycine ␣-amidating monooxygenase resin as described (30). NBD and Rho labeling of the N terminus of the resin-bound peptides was achieved as described previously (31). The peptides were cleaved from the resin by HF treatment and purified by reverse transcription-HPLC. Purity (ϳ99%) was confirmed by analytical HPLC. The peptides were subjected to amino acid analysis and to Platform LCZ micromass electrospray to confirm their composition.
Preparation of Lipid Vesicles-Small unilamellar vesicles (SUV) were prepared from PC/PS (1:1 w/w) or from PC by sonication as described previously in detail (32).
Membrane Permeation Induced by the Peptides-Membrane permeation was assessed utilizing the calcein release assay (33) as described previously (34). Briefly, SUV (7 mg/ml) were prepared in a 60 mM calcein and a 10 mM HEPES solution, pH 7.4. Liposomes were then passed through a Sephadex G-50 column (Amersham Pharmacia Biotech) in order to separate the free calcein. A final concentration of 2.5 M SUV was used. Trapped inside vesicles, the calcein dye is selfquenched. Thus, membrane permeation can be detected by an increase in fluorescence as peptides are added. The percentage of fluorescence recovery, F t , was defined as F t ϭ (f t Ϫ I o ) /(I max Ϫ I o ) ϫ 100, where I 0 ϭ initial fluorescence, I max ϭ the total fluorescence observed after the addition of paradaxin, and f t ϭ the fluorescence observed after adding the peptide, at time t. Paradaxin was used at a concentration of 0.6 M, which caused 100% calcein leakage from the vesicles.
Rhodamine Fluorescence Dequenching Measurements-Rhodamine fluorescence is highly sensitive to self-quenching but poorly affected by the dielectric constant of its environment. Therefore, the tendency of the peptides to oligomerize in aqueous solution was tested using Rholabeled peptides. Changes in the fluorescence were measured after adding proteinase K (10 mg/ml final concentration) to Rho-labeled peptides (0.1 M) previously dissolved in PBS. More proteinase K was added until there was no further change in the fluorescence emission. Excitation was set at 530 nm, and emission was set at 580 nm. All the fluorescence measurements in the present study were done on an Aminco Bowman Series 2 SLM spectrofluorometer (SLM-Aminco Spectronic Instruments).
Binding Experiments-The degree of peptide association with PC or PC/PS SUV was measured by adding increasing amounts of vesicles to 0.1 M NBD-labeled peptides dissolved in PBS. The fluorescence intensity was measured as a function of the lipid/peptide molar ratio, with excitation set at 467 nm, and emission set at 530 nm. The fluorescence values were corrected by subtracting the corresponding blank (PBS with the same amount of vesicles). The binding isotherms were analyzed as partition equilibria (31,(35)(36)(37)(38) using the formula X b * ϭ K p * C f , where X b * is defined as the molar ratio of bound peptide per 60% of the total lipid, assuming that the peptides were initially partitioned only over the outer leaflet of the SUV, as has been previously suggested (38); K p * corresponds to the partition coefficient; and C f represents the equilibrium concentration of the free peptide in the solution. The value X b was calculated by extrapolating F ϱ (the fluorescence signal obtained when all the peptide is bound to lipid) from a double-reciprocal plot of F (total peptide fluorescence) versus C L (total concentration of lipids) (35). Knowing the fluorescence intensities of unbound peptide, F 0 , as well as the bound peptide, F, the fraction of membrane-bound peptide, f b , could be calculated using the formula Having calculated the value of f b , it is then possible to calculate C f as well as the extent of peptide binding, X b *. The curves that result from plotting X b * versus free peptide, C f , are referred to as the conventional binding isotherms.
Fluorescence Resonance Energy Transfer (FRET) Measurements-FRET experiments were performed by using NBD-labeled peptides serving as energy donors and Rho-labeled peptides serving as energy  (39). Fluorescence spectra were obtained at room temperature, with excitation set at 467 nm (8-mm slit). In a typical experiment, a donor peptide (final concentration, 0.05 M) was added to a dispersion of SUV (350 M) in PBS followed by the addition of an acceptor peptide in several sequential doses. Fluorescence spectra were obtained before and after adding the acceptor. The efficiency of energy transfer (E) was determined by measuring the decrease in the quantum yield of the donor as a result of the presence of the acceptor. The value E was determined experimentally from the ratio of the fluorescence intensities of the donor in the presence (I da ) and in the absence (I d ) of the acceptor at the wavelengths of the maximal donor emission. The percentage of transfer efficiency (E) is given by %E ϭ (1 Ϫ I da / I d ) ϫ 100. Subtracting the signal produced by the acceptor-labeled analogue alone made correction for the contribution of acceptor emission as a result of direct excitation. The contribution of buffer and vesicles was subtracted from all measurements.

RESULTS
Peptide-induced Pore/Channel Formation-Previous studies have indicated that the ␣4 and ␣5 helices of domain I are important for toxicity and pore formation (15, 17, 20 -23, 40, 41). We have investigated whether helices ␣4, ␣5, and ␣4loop-␣5 can cause membrane leakage of encapsulated calcein from vesicles. Fig. 1 shows the dose response of peptide-induced calcein release from PS/PC vesicles. Clearly the results show that ␣4 and its mutants are incapable of causing any appreciable calcein release on PS/PC vesicles. Compared with ␣4 peptides, ␣5 had low activity (Fig. 1). These results are in accord with published data on a homologue Cry3A ␣5 peptide (18) and single channel experiments with an elongated version of Cry1Ac ␣5. Among the mutated ␣4 peptides, F9L showed activity, albeit low, on PC vesicles, which is more relevant to the midgut membranes. This result is in agreement with a mutation done on the full-length protein, where the F9L mutation was found to exhibit hyperactivity compared with the WT protein (23). Interestingly, we found ␣4-loop-␣5 to be much more active than ␣4 and ␣5 alone or together. For example, the ␣4-loop-␣5 reached 70% activity on PS/PC vesicles at a peptide/ lipid molar ratio of 0.04, whereas ␣4 and ␣5 were practically inactive at the same ratio (Fig. 1). A mixture of ␣4 or its mutants with ␣5 did not exhibit any synergism, implying that a linkage of the two helices with the loop is necessary for activity (Fig. 1). All other peptides showed similar activities on PC vesicles, and therefore, only ␣5 is shown as an example. These results support the umbrella model, where ␣4-loop-␣5 hairpin is inserted into the membrane in an anti-parallel manner to form the active pore/channel (7,41,15). To test whether the inability of ␣4 peptides to cause calcein release is a result of their inability to bind phospholipid membranes, we performed additional binding experiments.
Peptides Binding to Membranes-Characterization of vesicle binding by WT, I7S, Q8R, and F9L ␣4 was facilitated through the attachment of an NBD moiety to the peptides. NBD is very sensitive to the dielectric constant of the environment. Therefore, membrane binding can be detected as a substantial increase in NBD fluorescence. The attachment of the NBD did not alter the membrane-permeating activity of the peptides, which has also been shown in other studies (19). A constant concentration of peptide (0.1 M) was titrated with the desired vesicles (e.g. PC, PS/PC). Increases in the fluorescence intensities of NBD-labeled peptides as a function of the lipid/peptide molar ratio yielded the conventional binding curves (see Fig. 2, panels A, B, C, and D for ␣4, Q8R, I7S, and F9L, respectively, for PC/PS vesicles). Similar curves were obtained with PC vesicles and, therefore, are not shown. The light scattering of the vesicles was subtracted by repeating the experiments with unlabeled peptides. Since all peptides apart from ␣5 are monomeric in aqueous solution, as revealed by rhodamine dequenching experiments (see next paragraph), we were able to analyze their binding isotherms as partition equilibria. The surface partition coefficients were estimated from the initial slopes of the curves shown in the insets of all panels in Fig. 2 (for PC/PS) (38). Table II shows the calculated partition coefficients (and the free energy of binding, Ϫ⌬G) for both PC and PC/PS phospholipid membranes. F9L shows positive cooperativity as the binding isotherm curves upward. The wild type and the rest of the ␣4 mutants give linear binding isotherm curves. WT-␣5 binding to vesicles was also facilitated through NBD-labeled peptides (Fig. 3). The binding isotherm was not extrapolated for ␣5, since it is an oligomer in solution (see next paragraph and Fig. 4). However, the saturation level of the ␣5 binding isotherm is similar to that of the ␣4 peptides, and therefore it is reasonable to assume a partition coefficient of the same scale (10 Ϫ4 M Ϫ1 ). Qualitatively, since the fluorescence of NBD-labeled ␣5 did not dequench upon binding, we qualitatively plotted the binding isotherm, which displayed strong positive cooperativity (Fig. 3, inset), similar to that previously shown for Cry3A ␣5 (15).
Peptide Oligomerization in Solution-To test whether helices ␣5 and ␣4 and its mutants are monomers or oligomers in aqueous solution, we used Rho-labeled peptides. The fluorescence of rhodamine is quenched when several Rho-labeled molecules are in proximity, and the increase in fluorescence after enzymatic cleavage is a result of dissociation of these aggregates. A final concentration of 0.1-1 M Rho-labeled peptide was added to PBS, and fluorescence levels were measured at 580 nm. After equilibrium was reached, proteinase K was added, and the change in the fluorescence was recorded. Peptide ␣5 exhibited a strong decrease in fluorescence in PBS at all peptide concentrations, after which the addition of proteinase K caused an increase of almost 3-fold in fluorescence level. The difference in fluorescence before and after proteinase K was added indicates that ␣5 oligomerizes in aqueous solution. Rholabeled ␣4 and its mutants did not show a significant change in fluorescence levels before or after adding proteinase K. Thus, ␣4 and its mutants seem to be monomers in solution. Fig. 4 shows, for example, the results obtained at a peptide concentration of 1 M.
Inhibition of the ␣4-Loop-␣5 Hairpin Activity-We further tested whether ␣4 and its mutants or ␣5 can interfere with the functional assembly of the ␣4-loop-␣5. In these experiments, the ␣4-loop-␣5 was used at a concentration in which it has ϳ50% calcein release activity (0.015 M), and ␣5 has ϳ10% activity. The peptides ␣4 and all mutants were used at the maximal concentration tested in the calcein release assay. Fig.  5 shows the results for PC vesicles. Similar results were obtained with PC/PS vesicles and, therefore, are not shown. The results indicate that F9L ␣4 has a very strong antagonism on the activity of ␣4-loop-␣5 (Fig. 5). Whereas peptides ␣5, ␣4, and mutants I7S and Q8R displayed only weak inhibition abilities, the F9L ␣4 inhibition of the hairpin was marked; specifically, the activity of ␣4-loop-␣5 was completely abolished when mixed with F9L ␣4 (Fig. 5). This surprising result occurred even when F9L ␣4 was incubated with the vesicles before adding the hairpin.
Peptide-Peptide Interaction within the Membrane Milieu-The ability of the peptides to oligomerize in their membranebound state was investigated using FRET. The experiments were performed both with PC and PS/PC vesicles at a molar ratio where 100% binding was expected (Fig. 2) using NBDlabeled peptides as energy donors and Rho-labeled peptides as energy acceptors. Fig. 6 depicts the curves of the experimentally derived percentage of energy transfer versus the bound acceptor/lipid molar ratio. When Rho-labeled Cry1Ac ␣5 was sequentially added to its NBD derivative, a decrease in the NBD fluorescence and an increase in the rhodamine fluorescence were revealed, suggesting that Cry1Ac ␣5 self-associates in its membrane-bound state (Fig. 6). Similarly, it was found that the Cry3A counterpart of ␣5 could also form oligomers when bound to the membrane (15,17,19).
The interaction of helices ␣4 with ␣5 in the membrane was studied as well. Peptide ␣4 and each one of its three mutants exhibited resonance energy transfer with ␣5, indicating that ␣4 and its mutants hetero-oligomerize with ␣5 (Fig. 6). In Fig. 6, the curve corresponding to a random distribution of monomers All experiments were performed at room temperature in PBS. Excitation was set to 467 nm, emission was recorded at 530 nm, and 4-nm slits were used. Insets, binding isotherms, derived from the respective binding curve by plotting X b *, the molar ratio between bound peptide and lipids in the outer leaflet, versus C f , the equilibrium concentration of free peptide in solution. FIG. 3. Binding curve of ␣5 peptide. Increase in fluorescence of NBD-labeled WT ␣5 upon titration with SUV (PS/PC 1:10). The experiment was performed at room temperature in PBS. Excitation was set to 467 nm, emission was recorded at 530 nm, and 4-nm slits were used. Inset, binding isotherm of WT ␣5 derived as described in the legend of Fig. 2. (42), assuming an R 0 (distance at which 50% FRET occurs) of 51 Å (17), is depicted as a dashed line. The interaction between Rho-␣5 and NBD-N36 (43), a peptide corresponding to a segment from human immunodeficiency virus-1 gp41 that was shown to bind membranes, was investigated as a negative control. Only random energy transfer was observed for the negative control, suggesting no specific interaction with ␣5.

DISCUSSION
In the current study we show that the Cry1Ac segment ␣4-loop-␣5 is much more active in membrane permeation than either of the hydrophobic helices by themselves (Fig. 4). For example, helix ␣4 shows no pore-forming activity at all, whereas ␣5 has low activity (Fig. 4). This provides strong evidence that both helices and the loop connecting them are needed to form the channel and subsequently cause toxicity. According to the crystal structure of the homologue proteins Cry1Aa and Cry3A, the ␣4-loop-␣5 hairpin is hidden within the hydrophobic core of the protein (7,8). Upon membrane binding, a metamorphosis occurs, and the hairpin is inserted into the membrane in a manner consistent with the umbrella model (7,15). Furthermore spectrofluorometric analysis of the shorter helical segments, such as pore-forming, membrane binding, FRET, or competition assays, suggests that ␣4 and ␣5 of Cry1Ac have distinct roles within the hairpin.
Recently, Schwartz and co-workers (16) used site-directed mutagenesis studies and suggested a more detailed organiza-tion for the inserted hairpin. According to their findings, domain I drifts away from domain II and III after initial binding. A hairpin inserts into the membrane with ␣4 facing the lumen of the channel and ␣5 facing the lipid interface, whereas the other helices are spread on the surface of the membrane. Using cysteine labeling, Schwartz and co-workers showed that ␣4 is accessible to the lumen of the channel. Their conclusion was that the ␣4-loop-␣5 is inserted into the membrane and that ␣4 faces the lumen of the channel, whereas ␣5 is at an outer circle, facing the hydrophobic environment of the membrane milieu. There is concrete evidence to support this model. For example, Aronson and co-workers (23) show that ␣5 is highly involved in the oligomerization of the channels and that its ability to oligomerize directly affects the protein toxicity. Furthermore, ␣4 is also highly involved in toxicity, although the oligomerization of ␣4 is not necessary for toxicity (23). Thus, with the help of ␣5 oligomerization, the more polar ␣4 inserts into the membrane and lines the lumen of the channel.
Our results show that ␣5 can create pores (Fig. 4) and homooligomers (Fig. 6) and that it exhibits a cooperative binding isotherm curve (Fig. 3). In addition, ␣5 alone can release calcein molecules trapped inside vesicles (Fig. 4), causing perforation of the membrane. However, pore formation activity is extremely high when it is connected to ␣4. Furthermore, loss of toxicity in Cry1Ac with mutations in ␣5 was directly correlated with the loss of oligomerization, suggesting that ␣5 is crucial for membrane oligomerization (44,23,45). In short, ␣5 can play an important role both in membrane insertion and in the oligomerization of several proteins to form a channel. Helix ␣4 can also bind vesicles, although no cooperativity was observed ( Fig. 2A). This can be explained by the more polar nature of ␣4 compared with ␣5, which makes it harder for ␣4 to insert into the membrane. Nevertheless, helix ␣4 has the ability to oligomerize specifically within the membrane environment, both with ␣5 and with itself (Fig. 6). In contrast, ␣4 and its mutants were incapable of membrane permeation, except for the slight ability of F9L ␣4 (Fig. 4). This may indicate that ␣4 plays a role in oligomerization within the phospholipid bilayers but is incapable of sufficient membrane insertion without the help of ␣5.
F9L ␣4 is a homologue peptide to the F134L mutation on the full-length protein, which was previously shown to have a 3-fold increase in toxicity (23). Importantly, the F9L ␣4 mutant is able to form oligomers within the membrane. It also has cooperative binding and slight membrane permeation capabilities. These facts may imply better insertion into the membrane milieu than the wild type (Fig. 2, A and D, insets). Substitution of an aromatic amino acid, which prefers the phospholipid interface, with leucine, which prefers the acyl chain environment, can explain why F9L ␣4 would be more capable of membrane insertion. Thus, we suggest that the cooperative binding observed for F9L ␣4 is due to the depleting surface-bound F9L ␣4 as it inserts into the membrane (46,47). The pore-forming activity of F9L ␣4, albeit low, supports the hypothesis that ␣4 has a face exposed to the lumen of the channel.
Surprisingly, we found F9L ␣4 to exhibit a strong inhibition on the highly active ␣4-loop-␣5 (Fig. 5). This inhibition occurs within the membrane environment, since F9L ␣4 was also pre-incubated with the vesicles before the addition of ␣4-loop-␣5. All the other short peptides (including ␣5) did not exhibit any significant inhibition capability. Why is the inhibition so dramatically increased with a more active mutation? This interference is most likely due to F9L ␣4 interacting with ␣5 within the membrane, which was previously shown to be crucial for oligomerization of the pore. This increased interaction between F9L ␣4 and the ␣5 segment within the hairpin is enough to disrupt the natural ␣4to-␣5 interaction between two hairpins, thus interfering with channel formation. Wild type ␣4 is probably unable to have significant inhibition, because it cannot insert deep enough into the membrane or in the proper orientation. This leads us to the conclusion that the interaction most important for oligomerization of the channel is that of ␣5 with ␣4. Furthermore, this can explain why all the ␣4 mutants showed an energy transfer with ␣5. Note that the ␣4 mutations affected the insertion ability or channel properties of ␣4 but not its interaction with ␣5. Thus, these mutations did not have a significant effect on oligomerization as in the full-length protein while changing the activity of the channel (23). The interaction of ␣5 with ␣5 may be less dominant for oligomerization, since we would then expect wild type ␣5 to have a greater inhibitory effect on the hairpin activity. The striking finding that an ␣4 mutant can inhibit ␣4-loop-␣5 pore formation demonstrates the importance of the specific interaction between ␣4 and ␣5 within the membrane. In addition, it provides the first example that a mutated helix of a poreforming toxin can function as an immunity protein by directly interacting with the segments that form the pores. This mode of action has been proposed for the immunity proteins of colicin E1 (27).
In conclusion, our results give direct support to the umbrella model (7,15,16). After the initial binding to the receptor, the protein undergoes a transformation. Domain I binds the membrane, and as its helices are spread around on the surface, the ␣4-loop-␣5 hairpin is inserted into the membrane. Oligomerization with other hairpins produces a channel in which the lumen of the channel is constructed mainly by ␣4 and partly by ␣5. ␣5 and the loop play a major role in the insertion of the hairpin into the membrane and the oligomerization of the channel.
Further investigation of specific molecular recognition within the membrane is of great importance, since this seems to be a very general phenomenon. For instance, the ability of the T cell receptor to assemble is encoded by their transmembrane domains (48). Likewise, the dimerization of glycophorin A (49 -51) is through its transmembrane domains. Thus, understanding the molecular organization of the ␦-endotoxin can shed light on the general principles that govern protein-protein interactions within the membrane. In addition, inhibition of ␣4-loop-␣5 pore formation by an ␣4 mutant presents a potential strategy for interfering with the assembly and function of other membrane proteins as well.