Acylation of Escherichia coli Hemolysin: A Unique Protein Lipidation Mechanism Underlying Toxin Function

E. coli HlyA and a Family of Bacterial Pore-Forming ToxinsHlyA is an important virulence factor in E. coliextraintestinal infections such as those of the upper urinary tract (81, 107, 124, 318). It is one of a close family of membrane-targeted toxins assumed or proven to be influential not only in urinary tract infections but also hemorrhagic intestinal disease, juvenile periodontitis, pneumonia, whooping cough, and wound infections; these toxins include enterohemorrhagic O157 E. coli hemolysin (257), the leukotoxin ofPasteurella haemolytica (291), the hemolysins and leukotoxins of Actinobacillus spp. (46, 52, 89,173), the bifunctional adenylate cyclase-hemolysin ofBordetella pertussis (100), and the hemolysins ofProteus vulgaris (167, 319), Morganella morganii (167), and Moraxella bovis(101) (Table 1). These toxins have 30 to 75% sequence identity to E. coli HlyA and share (i) posttranslational maturation, (ii) a C-terminal calcium-binding domain of acidic glycine-rich nonapeptide repeats that has led to the RTX (repeat toxin) family nomenclature, and (iii) export out of the cell by type I secretion systems (34, 59, 129, 171, 320). The posttranslational modification is unique to this toxin family, but Ca²⁺ binding and type I secretion are both common to other bacterial proteins (18, 34, 86, 221, 301). In HlyA there are between 11 and 17 glycine-rich repeats. When Ca²⁺ is bound (one calcium ion per repeat), these form short β-strands organized in an unusual “spring-like” structure called a parallel β-barrel or β-superhelix (18). Calcium binding is an absolute requirement for cytotoxic activity (36, 37, 187, 232) and occurs outside bacteria following export, the intracellular level of free calcium in E. coli being very tightly regulated to 0.1 μM (92), a level too low for HlyA activity.

Synthesis and Export of HlyA ToxinThe synthesis, maturation, and secretion of E. coliHlyA are determined by the hlyCABD operon (81, 132,171, 225) (Fig. 1). The membrane-located export proteins are synthesized at a lower level than the cytosolic HlyC and pro-HlyA, in part due to transcription termination within the hlyCABD operon (81). This termination is suppressed by the elongation protein RfaH and a short 5′ DNA sequence, ops (operon polarity suppressor) (7, 8,60, 225, 311), that act together to allow the transcription of long operons such as hly, rfa, and trathat encode the synthesis and export of extracellular components important to the virulence and fertility of gram-negative bacteria (7, 9).

• Open in new tab
• Download powerpoint

Fig. 1.

Hemolysin synthesis, maturation, and export by E. coli. (CM, cytoplasmic membrane; OM, outer membrane; ΔP, total PMF). Expression of the hemolysin operon is governed by components upstream of the hly genes. The hlyA gene encodes inactive prohemolysin, which is activated by HlyC. HlyA is secreted by a type I process. Ca²⁺ binds to the glycine-rich repeats of the toxin externally before interacting with the mammalian membrane.

The pro-HlyA protoxin is matured in the cytosol to the active form by HlyC-directed fatty acylation (see below). The maturation increases the hydrophobicity of the protein but is not required for export (186). E. coli HlyA and its toxin relatives are all secreted across both membranes by the type I export process employing an uncleaved C-terminal recognition signal (223,281) but no N-terminal leader peptide (82) or periplasmic intermediate (83, 168). The HlyA secretory apparatus comprises HlyB (an inner membrane traffic ATPase), HlyD (an inner membrane protein that is suggested but not shown to bridge to the outer membrane), and TolC (an outer membrane protein) (260, 313,315). In E. coli and most other pathogens, TolC is encoded by a gene separated from hlyCABD, but inB. pertussis the toxin locus includes tolC(cyaE). Type I secretion signal sequences have been located within the C-terminal 24 to 80 amino acids (137, 168, 293). The HlyA C terminus is predicted to contain an amphipathic helix (159, 281), and circular dichroism and nuclear magnetic resonance spectroscopy of the HlyA signal has shown that α-helices are formed in a membrane mimetic environment (330). Despite a lack of identity between their primary sequences, the interchangeability of the export genes suggests that higher-order structures in the signals are shared among the extended family of hemolysins, leukotoxins, and proteases (256, 262, 301, 329).

HlyB has an integral membrane domain fused to an ATP-binding cassette (traffic ATPase) cytoplasmic domain, which undergoes conformational change on ATP binding (170) and couples ATP hydrolysis to HlyA export (166). Topological models proposed from fusion data suggest that HlyB inserts in the cytoplasmic membrane via six helical transmembrane segments with both its N and C termini in the cytoplasm. The predicted HlyB cytoplasmic loops are large and positively charged, while the periplasmic loops are small (97). The function of HlyD is less clear, but it is proposed to have a single transmembrane segment with the N terminus in the cytoplasm and a large C-terminal region in the periplasm (260). TolC is a minor outer membrane protein believed to form ion-permeable channels. Electron microscopy of two-dimensional lattices of TolC in phospholipid bilayers has revealed it to be a trimeric porin-like structure with a C-terminal extramembrane domain believed to form a periplasmic bridge to the energized inner membrane components of the translocation complex (172). HlyBD-dependent type I secretion shares with the Sec secretion across the cytoplasmic membrane an early requirement for the total proton motive force (PMF) but also has a late stage that does not require PMF, membrane potential, or the proton gradient. A translocation intermediate identified in the PMF-independent late stage is closely associated with the inner membrane, possibly in a translocation complex spanning both membranes (169). The belief that HlyA interacts with HlyB is compatible with the finding that suppressor mutations in HlyB partially compensate for mutations in the HlyA secretion signal (273), and a view of the assembly of a transport-competent complex occurring in an ordered manner is now being substantiated by cross-linking experiments (34, 300). A consequence for HlyA secretion of the virtual absence of Ca²⁺ ions in the bacterial cytoplasm is that the glycine-repeat structures should be flexible, allowing translocation of the polypeptide across the bacterial membranes in an unfolded state.

Host Cell Specificities of the Toxin FamilyThe toxins of the HlyA (RTX) family have been grouped on the basis of their lytic and toxic effects on mammalian host cells: hemolysins exhibit little target cell specificity, while leukotoxins have pronounced species- or cell-specific effects (59, 320) (Tables 2 and3). E. coli HlyA has a wide spectrum of cytocidal activity, attacking erythrocytes, granulocytes (28, 91), monocytes (29), endothelial cells (294), and renal epithelial cells of mice, ruminants, and primates (155, 212). The 61% identical hemolysin (Ehx) from enterohemorrhagic E. coli O157 lyses sheep and human erythrocytes and kills bovine leukocytes (albeit with a specific activity more than 20-fold lower than that of HlyA), but, unlike HlyA, it has virtually no activity against human leukocytes (16). These contrast with the P. haemolytica andActinobacillus actinomycetemcomitans leukotoxins which have little erythrolytic activity but a potent cytotoxic activity toward phagocytic cells of particular animals: the P. haemolyticaleukotoxin (LktA) lyses leukocytes only from cattle, sheep, and other ruminants (274), while the A. actinomycetemcomitans leukotoxin (AaltA) specifically lyses human and primate polymorphonuclear lymphocytes (328). A. pleuropneumoniae produces hemolysins (ApxIA and ApxIIA) and a leukotoxin (ApxIIIA) (89) (Table 2).

How such reported host cell specificities arose during the evolution of the toxin family is not clear from phylogenetic trees derived from DNA sequences (320), and the physical basis for cell specificity is unresolved. Studies performed on hybrid toxins created by the exchange of putative domains between HlyA, LktA, ApxIIA, and AaltA have indicated that specificity is unlikely to lie in a single discrete feature. Activity against erythrocytes has been correlated with a central region of HlyA and an N-terminal region of ApxIIA (88). The ability to kill leukocytes has been mapped to C-terminal regions of LktA and ApxIIA but appears less sharply defined in HlyA (88, 202). The critical region of AaltA required for recognition of human target cells spans its glycine-rich repeats (175), whereas the specificity of LktA for ruminants has been linked to a feature at its N terminus (88).

Ambiguity in the experimental definition of the target cell range may contribute to the lack of understanding of targeting by these toxins. Targeting has been defined by the cytotoxic end result (death), but this is the result of a multistage mechanism that might involve receptor recognition, membrane insertion, and protein-protein interaction. There are therefore many reasons why a particular cell line might be insensitive to a particular toxin. The toxin may not interact with the membrane, and this may occur for AaltA with human erythrocytes (298) and for LktA with nonbovine erythrocytes (15, 42). The toxin may adsorb to the membrane but fail to insert into the lipid bilayer, and this may be the behavior of inactive, nonacylated protoxins. Even if the toxin does permeabilize the membrane, the cell line will appear insensitive if it possesses a lesion repair mechanism. It is also possible that the bacterial host will affect the final toxin activity. The B. pertussis CyaA hemolysin synthesized in E. coli has a fourfold-lower hemolytic activity than does the native CyaA (109), and while ApxIIA protein secreted from its native host A. pleuropneumoniae is both hemolytic and cytotoxic, expression of the apxIICA genes in E. coli generates a toxin that is cytotoxic but has little or no hemolytic activity (305). LktA from E. coli carrying the P. haemolytica lktCA genes caused weak lysis of erythrocytes despite reports that LktA has a target cell range limited to leukocytes and platelets (87, 125). Continued biochemical investigation of the action of these toxins should establish whether continued division into hemolysins and leukotoxins is appropriate.

An explanation for the apparent cell specificity of these toxins might be offered by their recognition of specific membrane receptors. The existence of receptors for many bacterial toxins, both enzymatic and pore forming, is well established. Diphtheria toxin enters mammalian cells by using the epidermal growth factor-like growth factor as a receptor (220), and the AB₅-type toxins such as cholera toxin, Shiga toxin, and verotoxin bind to cell surface ganglioside lipids (279). The binding ofStaphylococcus aureus pore-forming leukotoxins to human polymorphonuclear neutrophils is initiated via a calcium channel or a receptor linked to a calcium channel (280). Aerolysin ofAeromonas hydrophila binds to erythrocytes and T lymphocytes via high-affinity receptors that have been shown to be glycoproteins with common glycosylphosphatidylinositol (GPI) anchors (73);Clostridium perfringens enterotoxin also uses several structurally related proteins as receptors (152); and the receptor for C. difficile toxin A has been identified as sucrase-isomaltase (245). Whether there is a receptor for the E. coli hemolysin is uncertain. Dose-response binding assays performed by one group support an upper limit of 4,000 HlyA binding sites per erythrocyte, implying at least some degree of specificity (15). However, this contrasts with results from others, who estimated up to 50,000 toxin molecules bound per erythrocyte and, using radiolabelled hemolytically active HlyA, subsequently found no evidence for saturation in binding to erythrocytes and leukocytes, indicating that this process is not conditional on specific receptors (31, 76). The absence of a specific receptor could be the reason why HlyA can attack a wide spectrum of target cells; indeed, HlyA can even bind to synthetic planar lipid membranes, suggesting that binding is relatively nonspecific (203). Nonspecific adsorption to the cell surface, rather than high-affinity binding to a surface receptor, has also been proposed to precede membrane insertion of B. pertussis CyaA (133). Inactive (unacylated) precursor toxin does not inhibit cell binding by active, mature toxin for both HlyA and CyaA, compatible with there not being a saturable receptor (15, 133).

It is possible that receptor-ligand interactions do occur with some members of the toxin family, for instance LktA and AaltA, displaying a narrower range of target cell specificities. Competition between inactive LktA mutants and mature LktA has been demonstrated, suggesting that a receptor in the leukocyte membrane is involved in LktA target specificity (61), and LktA is unable to bind to bovine leukocytes pretreated with protease, suggesting the involvement of a proteinaceous component in toxin binding (42). AaltA has been suggested to use a β-integrin as a receptor, since it is able to bind both subunits of the lymphocyte function-associated antigen type 1. HlyA may use the same β₂-integrin for recognition of human leukocytes, but no direct binding has been shown (176). Integrins, a large family of αβ heterodimeric transmembrane receptors that are differentially expressed by a variety of cell types, have the potential variability to distinguish both cell type and species (except in erythrocytes) and are used as receptors during Shigella and Yersinia invasion of mammalian cells (84, 85, 130). Protein sequences potentially involved in putative receptor recognition would appear most likely to lie within the nonconserved domains of the toxin molecules (do the differences between E. coli HlyA and Ehx define regions of interaction with a human leukocyte receptor?). A model could be put forward in which toxin action against erythrocytes is an intrinsic property and does not use a receptor while action against leukocytes is defined by specific receptor interactions. The involvement of target cell components such as integrins would be consistent with the lack of receptor involvement in erythrocytes.

Subversion of Signal Transduction and Cytokine ProductionPathogenic bacteria frequently subvert host cell functions such as signal transduction pathways, cytoskeleton rearrangement, and vacuolar trafficking (reviewed in references 84 and85). For example, enteropathogenic E. coli (EPEC) secretes proteins such as EspA and EspB that stimulate host phospholipase C, inducing inositol triphosphate production, the mobilization of intracellular Ca²⁺ stores, tyrosine phosphorylation of host proteins, and cytoskeleton rearrangement (251). Cell entry by Yersinia,Listeria, Salmonella, and Shigellaalso involves the activation of host signalling and a more extensive cytoskeleton rearrangement, which is controlled by small GTP-binding proteins belonging to the Ras family, namely, Rac, Rho, and CDC42.Shigella, Listeria, and Yersiniaphosphorylate host proteins, including substrates for the nonreceptor tyrosine kinase Src. Salmonella, Mycobacterium,Chlamydia, and Legionella reside in intracellular membrane-bound vacuoles, and the stability of these vacuoles is believed to result at least in part from the bacteria interacting with vesicular trafficking that is normally mediated by a family of small GTP-binding proteins called Rabs. Proteins of the Ras, Src, and Rab families will be highlighted again in this review because of their possession of fatty acyl groups. There are also many examples of bacterial molecules inducing cytokine synthesis in eukaryotic cells, and although the most widely studied are lipopolysaccharide (LPS), capsular polysaccharides, and peptidoglycan, they also include porins, fimbrial proteins, heat shock proteins, and lipoproteins, as well as extracellular proteins such as proteases and toxins (reviewed in reference 120). Both gram-positive and gram-negative pathogens produce toxins that induce cytokine release; e.g., interleukin-1 (IL-1) and tumor necrosis factor (TNF) are induced in murine macrophages and human monocytes in response to listeriolysin O, pneumolysin, C. difficile toxin B, and Streptococcus pyogenes erythrogenic toxin A. Cholera toxin increases IL-6 synthesis and decreases TNF-α production by rat peritoneal mast cells, pertussis toxin enhances IL-4 production, and verotoxin induces the release of IL-1α, IL-6, and TNF-α in mouse macrophages.

Erythrocytes exposed to E. coli HlyA rapidly undergo cytoskeleton rearrangement, resulting in the formation of teardrop-shaped projections from the surface (147). HlyA also affects cytokine production. At very low, sublytic concentrations, HlyA is a potent trigger of G-protein-dependent generation of inositol triphosphate and diacylglycerol in granulocytes and endothelial cells, stimulating the respiratory burst and the secretion of vesicular constituents (30, 103). HlyA stimulates the release of IL-1β and TNF from human monocytes, the lipoxygenase products leukotriene B₄ and 5-hydroxyeicosatetraenoic acid and nitric oxide from endothelial cells, and IL-1β (but not TNF-α) from cultured monocytes, while it inhibits the release of IL-1β, IL-6, and TNF-α from human leukocytes (29, 102, 164, 295). When injected into mice, HlyA produces a cytokine response similar to those seen in vitro, elevating the levels of IL-1 and TNF in serum (200). This ability to affect cytokine release is shared by all the members of the HlyA toxin family; in vitro, LktA and AaltA reduce the lymphocyte response to various stimuli (195,197), at low concentrations, LktA stimulates the production of IL-1 and TNF by bovine mononuclear phagocytes (287, 326) and potentiates the production of histamine and certain eicosanoids in response to chemokines (1). LktA and ApxII activate bovine and porcine neutrophils, respectively (63, 121, 306). These effects may occur without necrotic cell death, since HlyA and AaltA cause apoptosis of human lymphocytes and LktA causes apoptosis of bovine mononuclear cells and granulocytes (142, 197). As discussed below, some inflammatory cytokines possess a comparable acyl modification to that of HlyA.

E. coli HlyA and other RTX toxins alter the membrane permeability of host cells, causing lysis and death. Lysis of erythrocytes might provide bacteria with iron, and killing nucleated cells may prevent phagocytosis. The benefit to extracellular bacteria as a result of the induction of a host response is not so obvious. A consequence of the release of cytokines may be to induce inflammation, disrupt epithelial cell junctions, and favor translocation of bacteria through the intestinal barrier. The activation of host cell signal transduction pathways may also be advantageous to bacteria if the host responds by presenting a receptor used by them for binding. Such mechanisms are used by B. pertussis, where binding of filamentous hemagglutinin to a monocyte integrin causes the presentation of a second filamentous hemagglutinin-binding site (131), and by EPEC, where adherence to epithelial cells occurs after tyrosine phosphorylation by the host of an inserted bacterial protein, which then associates directly with intimin, an EPEC adhesin (160).

It is possible that some of the biological effects so far assigned to the HlyA toxin actually reflect cooperative responses to both toxin and LPS. In vivo and in vitro studies indicate that exposure to LPS, or mediators stimulated by LPS, increases the subsequent response to HlyA and LktA (261, 287). One response of host cells attacked by HlyA is the rapid and massive shedding of LPS receptors (CD14), which may then bind to neighboring cells and make them sensitive to LPS. Such a mechanism might underlie the long-range detrimental effects of pore-forming toxins in host organisms (312). The induction of a host response may not be an alternative to cell death but an alternative pathway to death, proceeding via programmed cell death or apoptosis instead of lysis. Many bacterial toxins as well as members of the HlyA family can induce apoptosis (126). Host cell death by apoptosis would occur without leakage of cellular components, without inflammation, and without damage to surrounding cells (4).

Pore Formation in Eukaryotic Membranes E. coli HlyA appears to have a two-stage interaction with eukaryotic membranes: a reversible adsorption sensitive to electrostatic forces and an irreversible insertion (10,233). Transition to the inserted form is associated with a change of conformation (211) and is favored by membranes with a fluid state and low cholesterol content, which may explain earlier findings that membranes made from asolectin, a crude lipid mixture from soybean, are very sensitive to HlyA, in contrast to membranes composed of pure lipids such as phosphatidylcholine or phosphatidylserine (21, 231). Once inserted into a membrane, HlyA behaves as an integral protein; it cannot be extracted without the use of a detergent (27). HlyA causes target cell lysis by forming pores which display cation selectivity and voltage and pH dependence, as shown in experiments with whole cells, planar lipid membranes, and liposomes (203, 204). HlyA pores in erythrocytes are thought to be asymmetric, since proteolytic enzymes digest the toxin only when applied to one side of the membrane. Similarly, when monoclonal antibodies raised against HlyA were applied from the same side as the toxin, they could open or close the pore, but they could not do so when applied from the opposite side (204, 205). The physical properties of the transmembrane pore formed by E. colihemolysin have been determined by using artificial lipid bilayers; it has a diameter of about 1.0 nm, a conductance of about 500 pS in 0.15 M KCl, and a mean lifetime of 2 s at low transmembrane voltages (21). HlyA pores in erythrocytes have been similarly sized, with a predicted physical cutoff of around 2 kDa (27). Patch-clamped human macrophages have been used as targets for HlyA, and this has confirmed that the leukotoxic action of HlyA is also due to the formation of pores with very different properties from the endogenous pores already present in the cell membrane (204). Similar physical properties have been assigned to the pores formed by other members of the HlyA toxin family (22, 55, 194, 203,258), suggesting that pore formation is a closely conserved step in the action of these toxins.

For many bacterial toxins unrelated to E. coli HlyA, pore formation is synonymous with oligomerization, with their structures varying from trimer to 100-mer. V. cholerae El Tor cytolysin may form 3- to 5-mer oligomeric structures (206). E. coli heat-labile enterotoxin and P. aeruginosacytotoxin form pentameric pores (228, 307), while aerolysin and S. aureus alpha-toxin oligomerize to produce heptameric pores (236, 278). The two gram-positive toxins streptolysin O and pneumolysin polymerize to form pores with a large and variable oligomeric superstructure of between 25 and 100 monomers (214,235).

Oligomeric forms of staphylococcal alpha-toxin and streptolysin O are unusually stable and can be extracted intact from membranes with detergent (26, 90). In contrast, HlyA recovered from deoxycholate-solubilized erythrocyte membranes is recovered in a monomeric form, indicating either that oligomerization is not required for pore formation or that oligomers are dissociated in the detergent (27). The lower stability, larger size, and lower solubility of HlyA-related toxins have made elucidation of the accurate subunit stoichiometry of the membrane-inserted pore even more complex and left it currently unresolved. As with streptolysin O and staphylococcal alpha-toxin, membrane insertion of HlyA and CyaA is believed to occur through a monomolecular mechanism (24, 286), with oligomerization, if any, occurring by the subsequent addition of monomers within the membrane. It has been estimated that only one to three HlyA molecules form the pore (21, 203), while CyaA pores in artificial lipid bilayers indicate a functional unit of a trimer or larger (297). The increases in membrane conductance on exposure to HlyA and Apx toxins have been explained by an association between nonconducting monomers and conducting oligomers (21, 194). The complementation of inactive deleted variants of HlyA to produce hemolytic activity and CyaA to produce cytotoxic activity has suggested that two or more toxin molecules aggregate before pore formation and substantiate the view that oligomerization is involved (133, 189). However, in the absence of a defined pore structure, it is worth bearing in mind that membrane disruption may occur through mechanisms other than formation of discrete pores. Examples are provided by the antibacterial peptide cecropin, where a “carpet” of peptide monomers disrupts phospholipid packing (93), and complement-mediated lysis by the formation of “leaky patches” in the lipid bilayer (327). Indeed, the exposure of membranes to toxin at high concentration or for long times may lead to similar lesions through a detergent-like mechanism (210, 231).

A highly conserved region (amino acids [aa] 238 to 410) of HlyA (20, 61) is essential for lysis and is believed to be involved in pore formation (Fig. 2). It spans the only pronounced hydrophobic sequences in the otherwise hydrophilic HlyA protein, and secondary-structure predictions suggest that it comprises four membrane-spanning α-helices, each of 21 aa. The sequences preceding and between these four putative HlyA α-helices have strong amphipathic α-helical properties and possess membrane-binding characteristics. It has been suggested that these repeated hydrophobic and amphipathic helices may insert into the cytoplasmic membrane and form a pore, allowing the influx of extracellular calcium and the escape of potassium (188,203). Mutations altering the hydrophobicity of this region reduce or abolish the pore-forming activity of the protein on erythrocytes and artificial membranes (186-188).

• Open in new tab
• Download powerpoint

Fig. 2.

Maturation of pro-HlyA to the active toxin by HlyC and acyl-ACP. The positively charged HlyC homodimer associates with the negatively charged ACP. Following binding to the HlyC recognition domains, FAI and FAII, the acyl chain is transferred to the corresponding acyl modification sites, K564 (KI) and K690 (KII). The acylated lysines KI and KII are shown relative to the hydrophobic pore-forming domain and the glycine-rich Ca²⁺-binding repeats of pro-HlyA.

The conserved glycine-rich repeat domain associated with HlyA Ca²⁺ binding is required for the hemolysis of erythocytes but not for pore formation in asolectin lipid bilayers, since its deletion leaves the pore-forming ability of HlyA unaffected (187) (Fig. 2). In addition, HlyA shows full channel-forming activity in artificial lipid bilayers even in the presence of 5 mM EDTA (74). This suggests that while Ca²⁺ binding is critical at some stage of the lytic process such as promoting the irreversible insertion of the toxin into the membrane (11), it does not directly contribute to the pore-forming structure (37,187). It also suggests that the asolectin lipid bilayer system is not a true reflection of binding and pore formation in real cell membranes, a conclusion supported by the finding that pore formation in asolectin bilayers is also independent of acylation (190). The conformation of the repeat region, which remains on the same side as that from which the toxin approached the membrane, may affect the state of the pore since the binding of a monoclonal antibody to this region led to pore closure (205).

Mutations in the LPS biosynthetic genes affect the expression and/or activity of E. coli HlyA (17, 286, 314). A transposon insertion in rfaP (required for attachment of phosphate-containing substituents to the LPS inner core) reduces the extracellular activity but not the export of HlyA as a consequence of the formation of large extracellular toxin aggregates; biological activity is restored with chaotropic agents (286). Why this occurs is not clear, but LPS possesses calcium-binding sites that may act as outer membrane reservoirs of calcium, and alterations in LPS structure may reduce the availability of calcium, resulting in the secretion of an inactive, unstable HlyA prone to aggregation. Alternatively, direct contact between wild-type LPS and HlyA may prevent toxin aggregation.

None of the dramatic effects of HlyA on eukaryotic host cells outlined above occur in vivo if the toxin is produced in the absence of the cotranslated protein, HlyC. The elucidation of the reason for this strict requirement has revealed a mechanism of bacterial toxin maturation unknown outside the HlyA family.

Bacterial Mechanism for Modifying Lysine ResiduesHlyA toxin is synthesized as an inactive 1,024-residue protoxin, pro-HlyA, which is activated intracellularly to the mature toxin by the action of the cosynthesized HlyC. It was shown by using an in vitro system that this maturation is a fatty acylation and that HlyC is a homodimeric putative acyltransferase, using acyl-acyl carrier protein (acyl-ACP) as the fatty acid donor (112, 128, 129, 132, 282,283) (Fig. 2). In in vitro reaction mixtures containing only purified acyl-ACP, HlyC and pro-HlyA, the acquisition of hemolytic activity is directly related to the binding of fatty acid by pro-HlyA. Mass spectrometry and Edman degradation of proteolytic products from mature HlyA toxin activated in vitro by HlyC and [³H]acyl-ACP revealed two fatty-acylated internal lysine residues, K564 (KI) and K690 (KII), and resistance of the acylation to hydroxylamine suggested that the fatty acid is amide linked. Substitution of the two lysines confirmed that they are the only sites of acylation and showed that although each is acylated in the absence of the other, both sites are required for in vivo toxin hemolytic activity (282). The K564 and K690 residues were subsequently confirmed to be modified in in vivo-synthesized and secreted HlyA by peptide mapping and two-dimensional electrophoresis (190). The internal pro-HlyA modification explains the earlier isolation of a monoclonal antibody that recognized only the active form of HlyA, specifically epitope aa627 to 726 (238).

The other HlyA-related toxins secreted by pathogenic gram-negative bacteria all require HlyC-type protoxin activation, and indeed many (PvxA, MmxA, ApxIA, and LktA) have been activated by E. coliHlyC (87, 106, 167). In addition, CyaC, ApxIIC, and AaltC are able to activate pro-LktA (175, 202, 322). CyaA, the adenylate cyclase hemolysin of B. pertussis, requires posttranslational activation both to deliver its catalytic domain into the mammalian target cell (cell-invasive activity) and to form transmembrane channels (hemolytic activity). It has been demonstrated by mass spectrometry that CyaA toxin secreted by B. pertussis is modified by amide-linked palmitoylation on the ɛ-amino group of K983 (analogous to HlyA KII, K690) (108). Only modification sites of HlyA and CyaA have been demonstrated to be acylated, although the loss of activity caused by deletion of the region between aa 358 to 548 of P. haemolytica LktA is compatible with the KI residue, K554, as a potential site of modification (61), and activation of a CyaA-LktA hybrid toxin supports the notion that the region aa 379 to 616 of LktA is important for LktC recognition (322). Substitution of LktA K554 (to T and C) reduced the lytic activity of LktA against bovine lymphocytes by only ca. 40%, but the retention of lytic activity does not rule out K554 as an LktA acylation site, since HlyA mutants with either KI or KII deleted also retained about 50% activity against BL-3 cells despite being inactive to erythrocytes (239, 282). Bovine lymphocytes may be less sensitive to the state of acylation of these toxins than are either erythrocytes or human lymphocytes, but it remains a possibility that LktA contains an acylation site different from the KI and KII sites identified for HlyA and CyaA. No data for the other toxins have been presented.

HlyC, a Novel AcyltransferaseThe ability to transfer an acyl group to an internal lysine residue of a target protein distinguishes HlyC from all other bacterial acyltransferases, and hlyA⁺ hlyC strains are nonhemolytic, indicating that the various constitutive acyltransferases of E. coli are unable to substitute for HlyC to even a small degree. HlyC has no significant sequence homology to known acyltransferases such as the lipid A acyltransferases (3, 54, 56,158), the Rhizobium Nod factor acyltransferases NodL and NodA (35, 69), or the well characterized acyltransferases such as glycerol-3-phosphate acyltransferases (182) or eukaryotic N-myristoyltransferases (140, 332). HlyC may therefore be an acyltransferase that is structurally and functionally distinct from all other acyltransferases.

Little is known about the biochemical properties of HlyC, the enzyme responsible for protoxin acylation. Bacterial and eukaryote acyltransferases generally accept either acyl coenzyme A (acyl-CoA) or acyl-ACP as an acyl donor, but ACP is a strict requirement for HlyC-directed pro-HlyA acylation (acyl-CoA cannot be used). Myristoyl-ACP gave the highest hemolytic activity of HlyA acylated in vitro with a range of fatty acids (C₁₂ to C_18:1) (132), which is consistent with an apparent selection of myristoylation at both the KI and KII sites in vitro. The affinity of KI for myristoyl-ACP (a saturated acyl group of 14 carbons) in vitro is approximately twice that for palmitoyl-ACP (a saturated acyl group of 16 carbons), while the affinity of KII for myristoyl-ACP is approximately five times that for palmitoyl-ACP (285). In vivo, both K564 and K690 appear fully acylated, suggesting that unlike CyaA, HlyA is predominantly myristoylated (190). Not only is HlyC able to discriminate between acyl-ACPs carrying fatty acids of different lengths but also it is presumably able to do this at low substrate concentrations, sinceE. coli does not have significant pools of acyl-ACPs with fatty acids longer than 3 carbons (249). In vitro measurements of K_m for both ACP and protoxin substrates suggest that HlyC does indeed recognize its substrates at extremely low concentrations (∼100 and ∼10 nM, respectively). HlyC is able to bind tightly but, it seems, noncovalently both the acyl chain from acyl-ACP and phosphopantetheine, but the locations of the active site of HlyC and regions that bind its acyl-ACP and pro-HlyA substrates are not known (284). The specificity for acyl-ACP as the acyl donor is shared in E. coli by the acyltransferases involved in lipid A biosynthesis, LpxA, LpxD, HtrB, and MsbB, but the use of a protein as a substrate is unique to HlyC (43, 158, 246). The cytosolic enzyme, LpxA, catalyzes the first step of lipid A biosynthesis, transferring (R)-3-hydroxymyristate from ACP to UDP-N-acetylglucosamine (3). LpxD is also specific for (R)-3-hydroxymyristoyl-ACP (158). The so-called “late” acyltransferases, HtrB and MsbB, can both use lauroyl-ACP and myristoyl-ACP as substrates for generating acyloxyacyl residues in lipid A, although HtrB shows a fivefold preference for lauroyl-ACP (54).

HlyC Recognition Domains on the ProtoxinBy using deleted protoxin variants and protoxin peptides as substrates in an in vitro maturation reaction dependent on only HlyC and acyl-ACP, two independent HlyC recognition domains (FAI and FAII) have been identified on the HlyA protoxin, each of which spans one of the target lysine residues, KI or KII, respectively (283) (Fig. 2). Each domain requires 15 to 30 aa for basal (ca. 10%) recognition and 50 to 80 aa for full wild-type acylation. The peptide recognition sequences for HlyC appear to be larger than those of other acyltransferases, since acylation of mutagenized target substrate proteins, chimeric proteins, and substrate peptides has indicated that peptide sequences of 4 to 15 aa are recognized byN-myristoyltransferases and prenyltransferases (198,240, 302, 303, 331). However, these modification sites are generally at the N and C termini of the substrate proteins, where they may be relatively open and accessible. In contrast, the internal acylation sites of HlyA may require HlyC to recognize a larger topology rather than a linear sequence. While the HlyA FAI domain is symmetrically centered on the modified lysine residue, the FAII domain appears to be asymmetrical, not requiring amino acids N-terminal to the modified residue (283). The two domains are nevertheless functionally indistinguishable and compete with each other for HlyC both in cis and in trans. No other HlyA sequences are required for toxin maturation, including the immediately C-terminal Ca²⁺-binding repeats (aa 722 to 849). Indeed, in vitro, Ca²⁺ ions prevent acylation at both the KI and KII sites (283). The extreme sensitivity of the pro-HlyA activation reaction to free Ca²⁺ supports the view that intracellular Ca²⁺ levels in E. coli are too low to affect toxin activity (92) and that Ca²⁺ binding does not occur until the toxin is outside the cell.

The HlyC recognition sites of E. coli HlyA have been compared with each other and with the corresponding sequences of other members of the toxin family. Although FAI and FAII of HlyA have the same function, there is little primary sequence identity (only 21%), and this lack of homology is also evident within each site among the toxins; only 20% of the FAI residues and 17% of the FAII residues are identical in the five toxins shown (Fig.3). This divergence may reflect differences in the HlyC-type proteins, the pattern of acylation at the two sites (where present, some toxins of the Pasteurellaceaefamily lack the KII target lysine), and the preferred target cell range of each toxin. Consensus sequences constructed for each of the two acylation sites have little in common apart from the central “GK” motif, suggesting that similarity between FAI and FAII may be of a higher structural order. Secondary-structure predictions of FAI and FAII indicate that both regions are rich in β-turns, quite regularly spaced approximately every 10 aa, especially in FAI. Other predicted features are not shared by FAI and FAII, except perhaps for a helical region immediately N-terminal to both KI and KII. However, helical-wheel projections do not show amphipathic distributions of charged or hydrophobic residues as related to membrane-associated structures. The lack of identity between FAI and FAII domains in pro-HlyA and corresponding sites on related protoxins currently deters an explanation for the basis of HlyA recognition by HlyC.

• Open in new tab
• Download powerpoint

Fig. 3.

Alignment of the E. coli HlyA FAI and FAII HlyC recognition domains with corresponding sequences of the related toxins A. actinomycetemcomitans AaltA, A. pleuropneumoniae ApxIA, P. haemolytica LktA, andB. pertussis CyaA. Identical residues are in boldface type; amino acids belonging to the same class are in capitals and non-homologous amino acids are in lowercase. Hyphens represent breaks introduced to maximize identity. Consensus sequences for FAI and FAII indicate residues shared by four toxins (boldface) and by all five toxins (underlined). A cumulative histogram representation of secondary structure predictions by the Chou and Fasman and the Robson and Garnier methods are shown under the primary sequences.

A lack of cross-complementation between protoxin activator C proteins supports the view that C-proteins may be unable to modify a KI or KII site even when it is present. E. coli HlyC is able to activate P. haemolytica LktA, whereas LktC is unable to activate HlyA and CyaA (87, 322). One explanation would be that LktC can acylate only KI even when KII is present. However, this explanation could not extend to A. pleuropneumoniae ApxIIC, which can activate P. haemolytica LktA, whereas LktC is able to activate ApxIIA only marginally, despite both toxins apparently possessing only KI (202) (Table 3). It seems possible that there are additional factors influencing the recognition between protoxin and HlyC-like activator proteins. One such additional factor appears to be the host cell in which the toxin operon is expressed. In contrast to the native CyaA protein from B. pertussis, which is palmitoylated exclusively at K983 (KII), recombinant pro-CyaA protoxin modified by CyaC in E. coli is acylated at K983 and contains an additional acylation site at K860 (analogous to KI) (109). It is perhaps this anomalous acylation in E. coli that allows CyaC to activate LktA (322). The specificity of CyaC peptide recognition may be affected by the host cell ACP, but this presumes that it is a C-ACP complex which recognizes the toxin. Such a mechanism is supported for HlyC by in vitro kinetic data that also indicate the involvement of a ternary HlyA-HlyC-ACP complex during the transfer of fatty acid from acyl-ACP to pro-HlyA (284). An ordered Bi-Bi mechanism would be analogous to that of N-myristoyltransferases, for which it is the binding of acyl-CoA which determines subsequent peptide binding (252,332). This might explain the earlier failures of LktC to activate the proforms of HlyA, ApxIIA, and CyaA (87, 202, 322) in experiments which were performed in or with extracts from an E. coli background. Alternatively, E. coli may possess an escort protein, similar to that observed with prenylated proteins, which enhances the acylation at KI but which is absent in B. pertussis (269). However, one might expect such an escort protein to have been identified by mutagenesis studies carried out in many laboratories that have successfully identified other accessory proteins, such as TolC and RfaH, required for toxin activity. Although the data available at present are incomplete and address only HlyA and CyaA, it appears that there is selection for the fatty acid attached to the toxin and that E. coli and B. pertussis have evolved hemolysin acyltransferases with different affinities for fatty acid. The E. coli HlyC protein may preferentially acylate HlyA KII with myristic acid, whereas theB. pertussis CyaC protein preferentially acylates CyaA KII with palmitic acid. Similar variations in substrate specificity between related proteins of different species appear between E. coli, Neisseria meningitidis, and P. aeruginosa LpxA, which are specific for (R)-3-hydroxymyristoyl-ACP, 3-hydroxylauroyl-ACP, and 3-hydroxydecanoyl-ACP, respectively (75, 227), and E. coli and Haemophilus influenzae HtrB, which are specific for lauroyl-ACP and myristoyl-ACP, respectively (54,224). Not only do differential fatty acid affinities require an explanation, but so do cross-species complementation tests with C-proteins and host-specific acylation patterns. Presumably, these explanations will lie in the differences between the structures of the family toxins, activator C-protein acyltransferases, and ACPs.

An insight into the modification of HlyA might be expected from an examination of other examples of protein lipidation. We present an extensive review of these and show that while the different types of modification vary considerably in terms of the location and structure of the lipid moiety, it is possible to define universal themes in terms of their function. Despite the absence of data for acylation events exactly equivalent to that of the prokaryotic toxins, it is still possible to draw conclusions from the more extensively studied examples that are relevant to the roles that toxin acylation may play in determining biological function. Lipidated proteins found in and on eukaryotic membranes are included in this comparison since the eukaryotic cell membrane is the site of action of HlyA and its related toxins. It is also becoming apparent that these acylated proteins are the sites of action of many bacterial effector proteins either directly by modification (e.g., ADP-ribosylation, glucosylation, and deamidation of small G-proteins and heterotrimeric G-protein subunits) or indirectly by interfering with their enzymatic pathways (e.g., protein phosphatases and kinases) (84, 85).

Increasing Affinity of Proteins for Biological MembranesThe increase in hydrophobicity resulting from the attachment of a fatty acid chain increases the association of many proteins with membranes. In gram-negative bacteria, N-terminal acylation localizes lipoproteins to the outer membrane, e.g., FlgH involved in membrane anchoring of the flagellar basal body, and PulS, promoting the correct localization of other membrane proteins involved in type III secretion of pullulanase (65, 259). In gram-positive organisms, many surface proteins are anchored by their C termini, but N-terminal lipidation may offer an additional membrane anchor generating a topology similar to that for Braun’s lipoprotein (292). A similar function can be attributed to the acyl group of N-myristoylated proteins, many of which are cytosolic when unmyristoylated but acquire biological activity when bound to membranes through simple insertion of their myristate chain into the hydrophobic interior of the lipid bilayer (40, 140, 163, 265, 277, 296). Many thiopalmitoylated proteins are either integral membrane proteins or additionally N-myristoylated and do not require palmitoyl groups for membrane association. An exception appears to be the Gα-interacting protein (72). Membrane interaction is also promoted by prenylation, e.g., a-factor (53), G-protein γ subunits (217), and RTKs (64, 234, 290). Many eukaryotic cell surface proteins are anchored to the membrane by GPI, which in terms of stability is comparable to a hydrophobic peptide domain. If the GPI anchor is not attached, the protein is not localized on the cell surface but is instead retained in the cell (98).

It is easy to assume that the addition of a fatty acid with a long hydrophobic chain would be sufficient to convert a protein into a membrane-associated form, but the thermodynamics are not so unequivocal. It has been calculated from experiments with acylated peptides mixed with phospholipid vesicles that binding energy increases by 0.8 kcal/mol per -CH₂ group along an acyl chain (237, 266). The hydrophobic energy supplied by myristate (C₁₄) may not therefore be sufficient to attach a protein firmly to a membrane. This has been confirmed experimentally, since myristoylated proteins are anchored at a membrane surface but not embedded within it (161), and even a palmitoyl (C₁₆) chain cannot force a hydrophilic peptide sequence into a hydrophobic environment. Therefore, while palmitoylation next to a transmembrane segment may increase the partitioning of this hydrophobic segment into the membrane, the sole presence of a fatty acid moiety may not be sufficient for membrane binding, especially in hydrophilic proteins (148). However, the effort invested in the incorporation of fatty acids suggests that there is a benefit, especially where the lipid donor is a minor component of the substrate pool. This advantage may be that the low hydrophobicity of single lipid groups allows a reversible association with membranes (302).

Where single lipophilic groups determine low membrane binding, a simple means of increasing the membrane affinity of a protein is to incorporate a second fatty acid. The incorporation of fluorescently labelled peptides bearing double-lipid modifications (C₂₀ + C₂₀, C₁₄ + C₁₆, C₁₆ + C₁₅, and C₁₆ + C₁₆) into phospholipid vesicles has been used to monitor the effect of double acylation on membrane affinity (266). Lipopeptides with “double-anchor” motifs do indeed have a greater membrane affinity than that of monoacylated or monoprenylated peptides (78). Surprisingly, peptides possessing palmitoyl (C₁₆) groups associate with lipid bilayers even more than do geranygeranylated (C₂₀) peptides, and double palmitoylation may be the most hydrophobic lipid modification currently known in intracellular proteins. This view is supported by the finding that a MARCKS derivative bearing two N-terminal palmitic acids in place of the wild-type single myristoyl moiety is no longer released from the membrane (265).

Many membrane-associated proteins use a double anchor. Among the intracellular proteins that are covalently modified with acyl or isoprenyl groups, a significant number carry more than one hydrophobic group. In some N-myristoylated proteins such as those of the Src family (e.g., Lck and Fyn but not Src and Blk) and α-subunits of heterotrimeric G proteins, membrane binding is reinforced with additional palmitoylation of nearby cysteine residues (209, 247,272, 308, 317, 334). The isoprenoid group binds to membranes with about the same affinity as myristate (266), and so membrane-bound prenylated proteins are also commonly doubly acylated. For instance, H-Ras and N-Ras which are farnesylated on their C-terminal cysteine residue are also reversibly S-palmitoylated on one or more nearby cysteines (198) (Fig.6). The combined hydrophobic interactions of the two lipid moieties are necessary to anchor these proteins to the plasma membrane, as shown by the inability of prenylated, nonpalmitoylated H-Ras mutants to bind to plasma membranes (323). Other double-lipid substitutions, including double geranylgeranyl and geranylgeranyl-farnesyl modifications, have been suggested to occur on mammalian Rab proteins and yeast Ypt (71,99, 269).

• Open in new tab
• Download powerpoint

Fig. 6.

Mechanisms to enhance the effects of single lipid groups. Double anchors increase hydrophobicity over single lipid groups. Members of the Src family of nonreceptor tyrosine kinases such as Lck or Fyn possess both an N-myristoylated glycine residue and a thioester-linked palmitoyl group. Small G-proteins such as H-Ras contain a C-terminal farnesyl group and a thioester-linked palmitoyl group. Electrostatic switches modulate the penetration of acyl chains into the membrane through the interaction of positively charged amino acids with the polar head groups of lipids. The myristoyl group of Src inserts into the bilayer when a basic domain interacts with acidic lipids, but phosphorylation of serine residues within this domain reduces the interaction. The hydrophobic interaction of the C-terminal farnesyl moiety of K-Ras with the bilayer is enhanced by an adjacent polybasic domain. The calcium-myristoyl switch of recoverin determines the accessibility of its lipid group. In the Ca²⁺-free state, the myristoyl group occupies a hydrophobic cluster (shaded oval), while in the Ca²⁺-bound state, the myristoyl group leaves the hydrophobic cluster and is exposed to interact with membranes or other proteins.

The membrane affinity of single acyl chains is not only enhanced by additional fatty acids. Both myristoylated and prenylated proteins use electrostatic interaction with the negative phospholipid head groups in the lipid bilayer to increase and decrease association with the membrane (218). The myristoyl groups of human immunodeficiency virus type 1 matrix protein cooperate with basic residues to interact with the acidic phospholipids in the membrane (333), and similar cooperative interactions between a C-terminal farnesyl moiety and an adjacent hexalysine domain have been observed for K-Ras (198, 222) (Fig. 6). Further subtlety arises from this additional electrostatic interaction, since in some proteins the phosphorylation of serine residues within basic sequences reduces the attraction for acidic lipids, providing an “electrostatic switch” for reversible binding, e.g., MARCKS and Src (161, 247,296). N-myristoylated hisactophilins possess an electrostatic binding site containing both basic and acidic residues, so that below pH 7 electrostatic attraction increases interaction with a lipid bilayer while above pH 7 it determines electrostatic repulsion (111). Other myristoyl switch mechanisms are responsible for the reversible membrane-binding of ADP-ribosylation factor and recoverin. ADP-ribosylation factor requires the binding of GTP first to expose an otherwise hidden myristoyl group and a patch of basic residues that then promote membrane insertion, while in the case of recoverin it is the binding of Ca²⁺ that exposes a myristoyl moiety masked within a hydrophobic groove, allowing it to interact with a lipid bilayer (2, 5, 38) (Fig. 6).

Enhancement of Protein-Protein InteractionsLipid attached to a protein acts not only as a mediator of membrane association but also as a determinant of specific protein-protein interactions and the assembly of membrane protein complexes. One obvious reason why membrane binding aids the interaction between proteins is that it reduces diffusion from three to two dimensions and increases their effective concentration approximately 1,000-fold. It may also expedite the lateral movement of membrane proteins, which may be required for rapid cross-linking and oligomerization. Nevertheless, very little is known about how acyl groups alter protein structure, especially during the change from the monomeric non-membrane-associated protein to the multimeric membrane-associated protein. Signal transduction is one area in which the contact of proteins with other proteins is crucial, and acylation plays a central role in these processes. The palmitoylation of GPCRs, the myristoylation, palmitoylation and prenylation of heterotrimeric G-proteins, the prenylation of Ras-related small G-proteins, the myristoylation and palmitoylation of Src family tyrosine kinases, and the glypiation of GPI-linked membrane proteins are all critical at different stages of signal transduction for reasons other than influencing membrane affinity (110, 199, 222, 247, 317, 331) (Fig. 5).

Approximately 1,000 different GPCRs exist in mammals, each member interacting with a limited number of closely related G-protein heterotrimers, and palmitoylation may play a role in this interaction since it is necessary for coupling to β₂-adrenoceptors although it is not essential for other receptors (115, 139,153). The proximity of sites for palmitoylation, G-protein binding, and RTK-mediated phosphorylation on GPCRs (321) allows an interplay such that the state of palmitoylation influences RTK recognition of the agonist-occupied receptor and plays a role in signal desensitization through the removal of the GPCR from the cell surface (213, 234). As well as associating with GPCRs, G-protein α-subunits interact with βγ-subunit complexes and with effector molecules such as adenylyl cyclase. The N-terminal myristoyl group on Gα subunits specifically contributes to Gβγ binding (19, 183, 317). There may not be direct contact between the fatty acid on α and the protein in βγ (32), but the isoprenoid group attached to the γ polypeptide does contact Gα in the αβγ trimer (217, 317, 325, 331). In addition, nearly all Gα subunits are reversibly palmitoylated, especially when the α subunit is not complexed with βγ (145, 317).

The role of the farnesylation of the Saccharomyces pheromonea-factor is not restricted to increasing hydrophobicity but may be important in binding the a-factor receptor (68). Lipidation also affects the interaction of many small G proteins such as Rho and Rab with their regulatory proteins, e.g., GDP-dissociation inhibitors (GDIs) and GDI displacement factor (198). The prenyl groups of small G proteins may bind to pockets in GDIs to form soluble complexes (150, 157, 324).

GPI-anchored proteins are cell surface proteins that are involved in signal transduction; e.g., GPI-anchored glycoproteins ofTrypanosoma cruzi induce proinflammatory cytokine synthesis by macrophages (49). To generate intracellular signals, GPI-anchored proteins such as Thy-1 that do not extend into the cytoplasm must interact with proteins on the inner face of the membrane such as Gα subunits and the Src family of nonreceptor tyrosine kinases. These interactions are influenced by the presence of saturated acyl chains; e.g., Src family members need to be N myristoylated and palmitoylated to interact with GPI-linked proteins (271), but this influence may be nonspecific and caused by concentration of lipidated proteins within detergent-resistant membrane domains of eukaryotic cells such as caveolae (41, 276). Dual palmitoylation may have additional effects more subtle than an increase in membrane affinity such as preferential targeting to the plasma membrane or Golgi apparatus (33, 149). Recognition between Src kinases and their substrates is also affected by their state of palmitoylation (255).

Eukaryotic Internal Amide-Linked AcylationLipopeptides with long-chain fatty acids attached to the side group of lysine residues have been used to investigate the biological effect of the unusual lipidation involving internal amide-linked acylation. A 14-aa hydrophilic peptide with a palmitoylated lysine at either the N- or C-terminal end is able to enter intact human cells, whereas the unmodified peptide is not (185). This system used synthetically modified peptides, but biological peptides do contain internal N ^ɛ-palmitoyl-lysine residues. An isoform of the 35-residue surfactant protein SP-C possesses two thioester-linked palmitoylated cysteine residues as well as a palmitoyl group linked to the ɛ-amino group of Lys11. The origin of this third modification appears to be nonenzymatic, and its biological significance is unclear (105). The acylation of phospholipase A₂ at the ɛ-amino groups of lysine residues converts the soluble enzyme into a membrane-penetrating form with an increased interaction with phospholipid monolayers (270). While this modification followed chemical treatment with short (C₈) acyl chains and may not be physiologically relevant, the findings nevertheless appear relevant to the other eukaryotic proteins (e.g., TNF-α, IL-1, and cytochrome coxidase) known or suggested to carry internal amide-linked fatty acids (Table 4).

The acylated lysine residues of TNF-α are located immediately downstream from a hydrophobic, membrane-spanning segment of the propiece, and so acylation may facilitate the membrane localization or insertion of the precursor molecule by enhancing overall hydrophobicity. Such a role for the acylation of cytokine propieces is supported by the fact that both the TNF-α and IL-α precursor proteins are active as plasma membrane-associated proteins whereas the poorly myristoylated IL-1β precursor is active only as the processed secreted form (25, 70). Dansylated myristoyl acceptor peptides derived from IL-α have demonstrated the specific binding of the myristoylated form to the inner leaflet of erythrocyte ghosts, suggesting that acylation does indeed facilitate membrane binding (289). The unmodified subunit 1 of cytochrome coxidase is already extremely hydrophobic, with several predicted transmembrane α-helices, and so it seems unlikely that acylation is required for integration of the protein into the lipid bilayer of the inner mitochondrial membrane. It is possible that myristoylation of subunit 1 encourages protein-protein interactions, contributing to assembly with subunit 2 and perhaps other subunits of this enzyme complex (309). The role of covalently bound fatty acids in the insulin receptor is unclear, but they may have important implications in terms of receptor stability, life cycle, or function (119). Immunoglobulin heavy chain (μ_m) is initially synthesized as a relatively hydrophilic protein, which is nonetheless stably anchored in the endoplasmic reticulum membrane. Myristoylation of μ_m correlates with its transport to the cell surface and its conversion to a relatively hydrophobic form. However, myristoylation does not result in detergent solubility, suggesting that its effect on μ_m may be an indirect one through mediating intermolecular interactions during its assembly and transport to the cell surface (242). Similarly, acylation of the nicotinic acetylcholine receptor, an integral membrane protein, appears to have a greater influence on protein-protein interaction rather than membrane affinity. These interactions may not be restricted to the receptor subunits but may include molecular chaperones such as calnexin, which forms complexes with nascent α subunits (96). Subunit assembly of nicotinic acetylcholine receptor is relatively slow and begins with the interaction of α and δ subunits in the endoplasmic reticulum to form a heterodimer. This process requires the tethering of the α subunit to the membrane, but whether this is provided by a lipid group is not known (316). Nevertheless, in the absence of acylation, subunit assembly is impaired, possibly because nonacylated subunits are degraded before they can be transported to the plasma membrane (207).

The roles of acylation at internal lysine residues thus appear to be the same as those of other lipid modifications. Eukaryotic proteins with internal amide-linked acyl groups form active membrane-associated complexes. Can this be linked to HlyA toxin function in eukaryotic membranes? Is it a coincidence that six of the seven eukaryotic proteins with internal amide-linked acyl groups are involved in signalling events either as cytokines or as receptors?

Host Cell Targeting by the HlyA Toxin FamilyWe have set out how fatty acylation of proteins can be central to anchoring to membranes, signal transduction, and oligomerization. The functions of the HlyA toxin suggest that its acylation could contribute in all three of these areas. It is tempting to try and explain the differences in the cell type and host species attacked by the toxins of the HlyA family in terms of what might be differences in their posttranslational modification. One interpretation of domain exchanges between hemolysins and leukotoxins and between leukotoxins that attack cells from different species could be that separate sequences account for erythrocyte specificity (∼aa 450 to 700 on HlyA) and leukocyte specificity (∼aa 700 to 850) (88, 175, 202). Although these regions approximate to the acylation sites of HlyA, examination of KI and KII within the toxin family does not allow a simple view that their acylation might individually determine erythrocyte and leukocyte specificity (Table 3). For instance, ApxIIA and LktA are cytolytic but do not possess KII, while AaltA and LktA are not hemolytic but do possess KI. The number of acylation sites also appears not to be decisive, since ApxIIA, a hemolysin, and LktA, a leukotoxin, both have one site whereas HlyA, a hemolysin, and AaltA, a leukotoxin, both have two sites. More subtle models are also not substantiated experimentally. Of the two (KI and KII) acylated lysines responsible for the hemolytic activity of HlyA, only KI is conserved in the strictly leukotoxic LktA (282). A simple hypothesis would be that the acylation of KI determines targeting to leukocytes while acylation at both KI and KII is required for erythrolysis. Although simple, this hypothesis has proved wrong, since HlyA with a substituted KII lost not only hemolytic activity but also its leukotoxic activity against human polymorphonuclear leukocytes (31, 239), and an N684K substitution, placing a lysine residue at the prospective KII site of LktA, failed to confer activity against erythrocytes (239). As previously indicated by studies with hybrid gene constructions (88, 175, 202), the structural basis of target cell specificity is unlikely to lie in single features of the toxins. In addition, the contribution made by putative receptors is unknown. Identification of all the acylation sites of the family of toxins will not be sufficient to explain cell targeting but will provide a further understanding of the mode of action of individual toxins at the molecular level.

Toxin Association with Eukaryotic Cell MembranesA discussion of cell targeting in terms of toxin acylation may be more revealing if one considers the separate processes of membrane insertion, host cell signalling, and pore formation. The influence of fatty acylation on HlyA membrane binding has been investigated, but the results are contradictory. Several research groups have concluded that unmodified, inactive pro-HlyA is unable to bind to erythrocyte membranes (37, 187) while the acylation of CyaA has been proposed to be essential for tight association with target cells (109). However, other work has suggested that acylation does not increase the affinity of toxin either to artificial liposomes or to real mammalian target membranes (15, 76, 210, 275). An explanation of these apparently opposing conclusions may lie in small differences in the preparation of the toxin samples. Fully modified toxin preparations lose their activity after dilution, possibly through self-aggregation and an inability to bind, and it appears that this tendency may be greater for unacylated than for acylated toxin and for calcium-bound rather than calcium-free toxin (11, 285). In addition, toxin at least partially denatured by the action of chaotropic agents appears to acquire the potential to bind lipid bilayers in an acylation-dependent manner (285). Under certain conditions, HlyA binding to erythrocytes is effective in the absence of acylation, Ca²⁺ binding, and the N-terminal hydrophobic region (15); there may therefore be no single factor that is decisive for in vivo membrane insertion. Whether the primary interaction and binding of HlyA to target membranes is dependent on fatty acylation is therefore unresolved and will probably require a more sophisticated approach than that allowed by in vivo systems.

Theoretically, by increasing the hydrophobicity of HlyA, the influence of fatty acylation would be to increase membrane binding. However, an alternative view is that the N-terminal hydrophobic domain of HlyA, possessing eight transmembrane helices, is sufficient for at least initial membrane binding and that before insertion is complete, the acyl chains are separated from the aqueous environment, perhaps hidden within hydrophobic tunnels. The transition from a membrane-bound to a membrane-inserted state could involve a conformational change that may at least partially release the acyl chains to the lipid bilayer. This transition could lie behind the temperature sensitivity of the lytic process; binding of HlyA to a membrane is slower at 4°C than at 37°C, but the amount eventually bound is constant. However, at the lower temperature, no cell lysis results despite the binding (15, 76, 210).

An essential step in toxin action, in addition to acylation, is the binding of calcium at the glycine-rich repeats of the toxin. Here the “Ca²⁺-myristoyl switch” of recoverin may be useful as a functional model (Fig. 6) (2), with the binding of Ca²⁺ being required to expose the acyl groups and make them available to the lipid bilayer. Certainly, Ca²⁺ binding by HlyA does appear to increase its overall surface hydrophobicity (11). It is possible that the binding of calcium affects the conformation of the toxin not only at KII adjacent to the glycine-rich repeats but also at KI more than 170 aa away, since in the presence of Ca²⁺ both KI and KII are no longer recognized by HlyC in vitro (283). While a large conformational change has not been detected by circular dichroism for HlyA on calcium binding, it has been shown for CyaA (11, 250). Why should the doubly acylated HlyA require an additional mechanism to increase membrane affinity when double acylations are among the strongest modifications known? The answer may lie in the HlyA acylation sites being 126 aa apart whereas in doubly modified N-myristoylated and prenylated proteins the two lipid groups are close together. It is more correct to view HlyA not as a doubly anchored protein but as being monoacylated in two places, a view corroborated by the existence of a sole palmitoyl group on native CyaA. The highly charged nature of the residues either side of KI and KII (30% of the flanking 20 residues are DEKR, with no overall positive charge) further increases the instability of any potential membrane association. The acylated lysines of HlyA do not have neighboring clusters of hydrophobic or positively charged residues as for K-Ras whether considered as a linear sequence, a helix, or a β-sheet (Fig. 3). Without these, single myristic or even palmitic side chains are unlikely to provide sufficient hydrophobic interaction to stabilize a transmembrane region even if there are additional contributions by the lysine side groups. Rather, the acyl groups on HlyA may act at most as anchorage points to the surface of the lipid bilayer.

Effects of HlyA Toxin on Host Cell Signalling and Cytokine ProductionThe hydrophobicity of fatty acylation does not appear to be a simple key to HlyA targeting to mammalian cell membranes, but it may be significant in mediating specific protein-protein and/or protein-lipid interactions. Investigations into the effect of acylation on the ability of sublytic concentrations of HlyA to stimulate a host cell response have been complicated, perhaps significantly, by the presence of LPS in toxin preparations. At least the major component of HlyA binding to granulocytes appears to be non-receptor mediated, with subsequent triggering of the respiratory burst presumably not through the occupancy of a receptor but through a short-circuiting of the classical signal transduction pathway (103, 165). Acylation may therefore enhance protein-protein interactions between HlyA and components of a signal transduction pathway, perhaps by targeting to detergent-resistant domains (Fig. 5). With acylation sites C-terminal to a multiple transmembrane helical domain, HlyA is at least superficially similar to the GPCRs, but unlike the labile thioester-linked palmitoylation of GPCRs, the amide-linked modification of HlyA would not be reversible. Effects of acyl groups on cytokine induction, however, are not restricted to a model incorporating complex protein-protein interactions. A synthetic lipopeptide of the N terminus of a bacterial lipoprotein stimulates the synthesis of IL-1, IL-6, and TNF-α by murine macrophages (114). Lipoteichoic acids fromStreptococcus faecalis stimulate murine mononuclear cells to release both IL-1 and TNF, but deacylation stops the cytokine-stimulating activity (304), and deacylation of lipoarabinomannan of Mycobacterium spp. almost totally inhibits its capacity to induce cytokine synthesis (14). Intriguingly, another response of monocytes exposed to lipoteichoic acid is to increase the expression of CD11a/CD18, a putative receptor that may be recognized by RTX toxins (51, 176).

HlyA Pore Formation in Host Cell MembranesAs well as being a pseudo-chemokine, HlyA is cytotoxic as a result of its ability to form pores. If monomers of the toxin must combine within the membrane to form pores, perhaps the acyl groups are involved in protein-protein interactions to bring about oligomerization. By providing anchorage points in the membrane, acyl groups may prevent essential domains from looping away from the membrane surface and may enhance lateral diffusion to accelerate the contact between them. The assembly of the acylated subunit 1 into the cytochrome oxidase complex and the interaction of the acyl groups involved in contact between the α and γ subunits of heterotrimeric G-proteins may provide useful analogies to this process (Fig. 5; Table 4). The most dramatic effect of the absence of fatty acylation on the action of HlyA indeed appears to be at the pore-forming step. Both acyl modifications must be present for HlyA to form channels. Loss of either acyl group results in the virtual abolition (>99.5%) of cytolytic activity (282). This is at least true for erythrocytes and human polymorphonuclear leukocytes. Monoacylated HlyA derivatives lacking either KI or KII do retain a reduced ability to form pores in bovine lymphocytes (BL-3 cells) (239). Why pore formation in BL-3 cells should be less dependent on toxin acylation is unknown, but the impression of systems involving bovine lymphocytes and the specific P. haemolytica leukotoxin LktA is that they represent a special case.

An influence of acylation on pore formation has also been demonstrated for CyaA. After expression in E. coli, CyaA carries an acyl group on K860 in addition to the K983 acylated in the native host. This overacylated toxin is impaired in the formation of the CyaA channels, being about 20% as hemolytically active as the wild-type CyaA (109). The hemolytic activity of HlyA absolutely requires double acylation, while monoacylation is sufficient for the hemolytic activity of CyaA. However, while both HlyA and CyaA are described as hemolysins, there is a dramatic difference in their erythrolytic activity, such that CyaA hemolytic assays are performed at nanomolar toxin concentrations but HlyA assays require only picomolar toxin concentrations. Monoacylated HlyA derivatives retaining only <0.5% of the wild-type activity are correctly defined as nonhemolytic but are still more hemolytic than the wild-type CyaA toxin. Nevertheless, whereas acylation at KI is essential for pore formation by HlyA, acylation at KI actually seems to impede the hemolytic activity of CyaA. The primary sequences 50 aa either side of KI are only 30% identical between HlyA and CyaA (283), so perhaps a reason for this difference will be confined to individual structures. It might be significant that KI is the acylation site closest to the toxin hydrophobic regions, with the hydrophobic region of CyaA possibly having an intrinsically stable topology while that of HlyA requires its C-terminal side to be stabilized by an anchorage to the membrane (Fig.2). Evidence that the acyl chains attached to HlyA may provide membrane anchors and that their stability in the membrane is influential in pore formation is provided by the observation that doubly palmitoylated HlyA has a specific hemolytic activity at least fivefold higher than that of doubly myristoylated HlyA (285).