Bacterial endotoxins: extraordinary lipids that activate eucaryotic signal transduction

Over 100 years ago, Pfeiffer and Centanni described heat-stable, nonsecreted toxin(s) in gram-negative bacteria that caused fever and pathology when injected into animals (53). Unlike heat-labile protein toxins, such as diptheria and tetanus toxins, the cell-associated toxins of gram-negative bacteria (endotoxins) consisted of carbohydrates and lipids (36, 53) and are also termed lipopolysaccharides (LPS) (see Fig. 1). Uncertainties about the chemical nature of endotoxins lingered until 1983, when the structure of the lipid A anchor of LPS (see Fig. 1 and 2) was elucidated (36, 46, 53), and chemically synthesized lipid A was shown to possess the biological activities of endotoxins (36, 46, 53). The long delay in establishing the identity of the active component of endotoxins occurred because the structure of lipid A is complex (29, 36, 46) and differs from those of all other lipids (45). The approximate composition of lipid A was recognized 40 years ago (36, 53), but its biosynthesis and pharmacology could not be studied without a complete understanding of its covalent chemistry (46). In 1983, just as several laboratories were close to defining the structure of lipid A (54), my group characterized a novel glucosamine-derived phospholipid (58), previously observed in certain phosphatidylglycerol-deficient mutants of Escherichia coli (40). Together with Laurens Anderson and Kuni Takayama at the University of Wisconsin (58), we showed that this material, termed lipid X, was a 2,3-diacylglucosamine 1-phosphate (see Fig. 2). The discovery of a monosaccharide substructure of lipid A provided an important clue to elucidating the sites of acylation on lipid A (40, 54, 58) and the enzymatic pathway for lipid A biosynthesis (1, 14, 46, 50). The functional connection between lipid X accumulation and phosphatidylglycerol deficiency remains unknown (39, 40). In this minireview, I summarize current knowledge of the enzymatic synthesis of lipid A, most of which is derived from studies with E. coli extracts (46, 65). I also provide a brief overview of the mechanisms by which lipid A activates signal transduction in animal cells.


INTRODUCTION
Over 100 years ago, Pfeiffer and Centanni described heat-stable, nonsecreted toxin(s) in gram-negative bacteria that caused fever and pathology when injected into animals (53). Unlike heat-labile protein toxins, such as diptheria and tetanus toxins, the cell-associated toxins of gram-negative bacteria (endotoxins) consisted of carbohydrates and lipids (36,53) and are also termed lipopolysaccharides (LPS) (see Fig. 1). Uncertainties about the chemical nature of endotoxins lingered until 1983, when the structure of the lipid A anchor of LPS (see Fig. 1 and 2) was elucidated (36,46,53), and chemically synthesized lipid A was shown to possess the biological activities of endotoxins (36,46,53).
The long delay in establishing the identity of the active component of endotoxins occurred because the structure of lipid A is complex (29,36,46) and differs from those of all other lipids (45). The approximate composition of lipid A was recognized 40 years ago (36,53), but its biosynthesis and pharmacology could not be studied without a complete understanding of its covalent chemistry (46).
In 1983, just as several laboratories were close to defining the structure of lipid A (54), my group characterized a novel glucosamine-derived phospholipid (58), previously observed in certain phosphatidylglycerol-deficient mutants of Escherichia coli (40). Together with Laurens Anderson and Kuni Takayama at the University of Wisconsin (58), we showed that this material, termed lipid X, was a 2,3-diacylglucosamine 1-phosphate (see Fig. 2). The discovery of a monosaccharide substructure of lipid A provided an important clue to elucidating the sites of acylation on lipid A (40,54,58) and the enzymatic pathway for lipid A biosynthesis (1,14,46,50). The functional connection between lipid X accumulation and phosphatidylglycerol deficiency remains unknown (39,40).
In this minireview, I summarize current knowledge of the enzymatic synthesis of lipid A, most of which is derived from studies with E. coli extracts (46,65). I also provide a brief overview of the mechanisms by which lipid A activates signal transduction in animal cells.

LIPID A COMPARED WITH GLYCEROPHOSPHOLIPIDS
The minimal LPS required for growth of gram-negative bacteria, termed Re LPS (36,38,46), occurs in mutants lacking heptose (see Fig. 1). Re LPS consists of lipid A and two 3-deoxy-D-manno-octulosonic acid (KDO) moieties (see Fig. 2) (36,46). Re LPS possesses 24 chiral centers, whereas classical glycerophospholipids contain one or two. Lipid A is therefore an "information-rich" molecule, with many possibilities for specific recognition by procaryotic and eucaryotic proteins. The acyl moieties of glycerophospholipids are often 2 to 6 carbons longer than those attached to lipid A (45,46), but they are not hydroxylated (45,47). The acyl moieties of lipid A generally do not contain double bonds (36,46).
There are _106 lipid A molecules and i107 glycerophospholipids in a cell (24,45), arranged as shown in Fig. 1. The reasons why lipid A and KDO are required for growth (46) are not known.
The LPS molecules of wild-type cells are further glycosylated with 6 to 8 additional sugars that constitute the nonrepeating core domain and -1 to 50 0-antigen repeats ( Fig. 1) (36,46). The latter are often tetrasaccharides (36,46) characterized by the presence of deoxyand dideoxyhexoses.

DISCOVERY OF THE LIPID A PATHWAY
Recognition that lipid X was equivalent to the reducing end glucosamine of lipid A (Fig. 2) (58) suggested that it must be a precursor or breakdown product (50,58). The discovery in 1983 that KDO was attached at position 6' on lipid A ( Fig.  2) (54) and that position 3' (Fig. 2) was acylated with R-3-hydroxymyristate revealed that the nonreducing end glucosamine of lipid A was acylated with R-3-hydroxymyristate ( Fig. 2) in the same places as the reducing end unit. Accordingly, we decided to search for a pathway by which the 2,3-diacylglucosamine moiety of lipid X would be incorporated into both sugars of the lipid A disaccharide (14,50).
As an initial approach, Ray et al. (50) synthesized UDP-2,3-diacylglucosamine (Fig. 3), a putative nucleotide derivative of lipid X. When UDP-2,3-diacylglucosamine and lipid X were incubated with a crude extract of wild-type E. coli, efficient formation of a new, more hydrophobic product was observed (50). Isolation and physical analysis demonstrated that this material consisted of two glucosamines, four R-3hydroxymyristates, and one anomeric phosphate (Fig. 3) (50). The presence of a novel disaccharide synthase in cell extracts, capable of generating the 1B, 1'-6 linkage of lipid A, demonstrated that lipid X and its UDP derivative were precursors (50). Neither compound had been discovered earlier, because each represents less than 0.01% of the total lipid of wild-type E. coli (14,45). Ovals and rectangles represent sugar residues, whereas circles depict the polar head groups of glycerophospholipids (phosphatidylethanolamine in red and phosphatidylglycerol in yellow). MDO represents membrane-derived oligosaccharides (45). The core region shown is that of E. coli K-12 (36,46), a strain that does not normally contain an 0-antigen repeat unless transformed with an appropriate plasmid (36,46 (2), suggesting that R-3-hydroxymyristoyl-ACP (a thioester) may be more stable than UDP-3-0-(R-3-hydroxymyristoyl)-GlcNAc (an oxygen ester). Whatever the chemical explanation, the significance of this observation is that deacetylation of UDP-3-0-(R-3hydroxymyristoyl)-GlcNAc ( Fig. 4) (4) is the first irreversible step of lipid A biosynthesis. Exposure of the glucosamine nitrogen by deacetylase permits incorporation of the N-linked R-3-hydroxymyristate to form UDP-2,3-diacylglucosamine ( Fig. 4) (1,3,30). As with the O-acyltransferase, coenzyme A thioesters and myristoyl-ACP are not substrates (3,30).
The key role of UDP-GlcNAc in lipid A biosynthesis shows that this nucleotide is situated at an important branch point in E. coli, as it is also a precursor of peptidoglycan (2). Similarly, R-3-hydroxymyristoyl-ACP can be used either in the biosynthesis of lipid A or in the generation of palmitate (33).
The demonstration of UDP-GlcNAc acyltransferases in all gram-negative bacteria so far examined has established the general importance of lipid X and UDP-2,3-diacylglucosamine as endotoxin precursors (65).
32P-labeled lipid IVA, prepared with this kinase (28), can be used for studying endotoxin binding to animal cells (27) or for detecting the distal enzymes of lipid A metabolism (12,13) (Fig. 5). Unlike disaccharide synthase, this kinase is unstable and has not been cloned or purified (51). Since phosphorylation of the 4' position is thermodynamically favorable, access to pure 4' kinase would be useful for the synthesis of 1,4' disaccharide bis-phosphate analogs of lipid A. These analogs are known to include potent endotoxin antagonists (26,31,57) with potential clinical utility.
The acceptor for the innermlost KDO residue is the 6' OH of lipid IVA (Fig. 5 (Fig.  5), at the 4 OH of the innermost KDO. KDO1-lipid IVA accumulates to a small extent in enzymatic incubations at low KDO concentrations (8). Given the significant difference in the structures of the acceptors for the first and second KDO residues, it is remarkable that both KDO residues are incorporated by a bifunctional enzyme consisting of a single polypeptide (8) (Fig. 5). This enzyme is encoded by the kdtA gene (15) near min 81 on the E. coli chromosome, just clockwise of the rfa cluster. Overproduction of the kdtA gene product increases the specific activities of both the first and second KDO transferases in extracts (8,15). The KDO transferase of Chlamydia trachomatis, which displays some sequence similarity to the E. coli enzyme, is a single polypeptide capable of incorporating at least three KDO residues (7).
Purified KDO transferase of E. coli (8,12) recognizes lipid A disaccharide bis-phosphates as acceptors, but the extent of their acylation is not crucial. The presence of a 4' phosphate residue is an absolute requirement for substrate recognition (8,12). This enzyme may share some common structural features with the serum LPS-binding protein (LBP) (59,66), with lipid A recognizing components of the limulus clotting cascade (34,37) or with putative endotoxin receptor(s) on animal cells (36,46,49). No sequence homologies between KDO transferase and eucaryotic lipid A-binding proteins have been discovered as yet.
Following attachment of KDO, additional acyltransferases complete the formation of lipid A by transfeffing laurate and myristate residues to KD02-lipid IVA (Fig. 5) Fig. 4 and 5 are very low (less than 1,000 molecules per cell) (14). ferases of E. coli utilize thioesters of ACP as substrates (13). The late acyltransferases display an amazing dependence upon the KDO domain (13), accounting for the fact that KDO-deficent mutants accumulate lipid IVA (25,48,52). All of the lipid A synthetic enzymes are cytosolic or associated with the inner membrane (46). Novel transport functions needed for LPS export probably also exist but have not been defined.

MOLECULAR GENETICS OF LIPID A BIOSYNTHESIS
We have obtained genetic evidence that the lipid A biosynthesis pathway (Fig. 6) is the major one for lipid A generation in E. coli and is essential for cell viability (24). We first found that silent mutations inpgsA (encoding phosphatidylglycerophosphate synthase near min 42) (39, 40) were rendered conditionally lethal by otherwise silent, second site mutations in lpxB, the structural gene for the disaccharide synthase near min 4 (18,39,40). Cloning and sequencing of the disaccharide synthase (18,19) revealed the presence of the lpxA gene, encoding UDP-GlcNAc O-acyltransferase, just upstream (Fig. 6) (17, 18). Characterization of temperature-sensitive mutations mapping near min 4 resulted in the isolation of SM101 (lpxA2), an organism with a defective UDP-GlcNAc O-acyltransferase and a 10-fold-reduced rate of lipid A synthesis in cells at 42°C (24). The properties of SM101 demonstrate the quantitative importance of this pathway (24). The rapid loss of viability of SM101 at 42°C suggests that inhibitors of these enzymes may be novel antibiotics (24).
The lipid A content of SM101 is 30% less than normal at the permissive temperature (30°C) (24). At 30°C, SM101 is hypersensitive to antibiotics that normally do not permeate the outer membrane, like rifampin and erythromycin (61).
In searching for other structural genes of lipid A synthesis, we examined all available rifampin-hypersensitive mutants. We found that deacetylase (Fig. 4) is encoded by the envA gene, located near min 2 on the E. coli chromosome (67). envA was first isolated and mapped in 1969 (41) as a mutation that causes antibiotic hypersensitivity and a delay in cell separation. Sequencing of envA (now designated lpxC) failed to reveal any similarity to genes of known function (6).
The specific activity of deacetylase is elevated 5to 10-fold when lipid A synthesis is inhibited (2). How the deacetylase is up-regulated is unknown, but this observation is consistent with its function as the first committed step.
The firA gene of E. coli (now designated lpxD) (Fig. 6) was first defined as a second site mutation that caused reversal of rifampin resistance associated with certain mutations in the 1 subunit of RNA polymerase (21). Recently, the product of the firA gene (30) was identified as the N-acyltransferase of the lipid A pathway. The effects of firA mutations on rifampin resistance are presumably due to enhanced outer membrane permeability. The firA gene maps near lpx4 and lpxB at min 4 (21). firA is identical to the ssc gene of Salmonella typhimurium (60). There is homology between the sequences of firA (IpxD) and lpx4, consistent with their functions (Fig. 4) (30, 60). The role of the operon(s) that includes lpxA, lpxB, and lpxD (17,19,30) requires further evaluation.

Synthetic endotoxin antagonist (Eisai)
Rhizobium leguminosarum (Possible short-chain species) FIG. 7. Structures of lipid A from R. sphaeroides and R. leguminosarum, contrasted with a synthetic endotoxin antagonist. Lipid A of R. sphaeroides was one of the first endotoxin antagonists to be identified (43,57). The -100to 1,000-fold-more-potent synthetic antagonist of the Eisai Co. was modeled on Rhodobacter lipid A. The Eisai compound is -10-fold more active as an antagonist than E. coli LPS is as an agonist. The structure of lipid A from R. leguminosarum is not fully established (10,11), and its biological activities are not yet known. Recent work by Carlson and Bhat (14a) indicates the presence of a 2-deoxy-2-aminogluconate residue in place of the reducing end glucosamine, as well as a 4' galacturonic acid in place of phosphate. In addition to R-3-hydroxymyristate (as shown), R. leguminosanum lipid A contains 16-, 18-, and 28-carbon acyl chains (10,11). R may be either an acyl residue or H.

LIPID A IN PHOTOSYNTHETIC BACTERIA AND RHIZOBIA
Examination of lipid A biosynthesis in Rhodobacter sphaeroides (43) and Rhizobium leguminosarum (10, 11) (Fig. 7) may provide some novel insights. Extracts of R. sphaeroides possess the UDP-GlcNAc acyltransferases (Fig. 4) (29a), but the origin of the unusual 3-keto-myristate residue (Fig. 7) (43) is uncertain. Lipid A from R. sphaeroides is interesting, in that it is an antagonist of the action of E. coli lipid A on animal cells (26,57). More-potent, synthetic endotoxin antagonists (Fig. 7) modeled on Rhodobacter lipid A have recently been reported (14b).
Lipid A from R. leguminosarum lacks phosphate altogether and does not contain a glucosamine disaccharide (10,11). Carlson and Bhat (14a) have proposed a structure like that in Fig. 7, in which carboxyl groups seem to function as surrogates of the phosphates that are usually attached at positions 1 and 4'. R leguminosarum lipid A contains a 2-deoxy-2-aminogluconate residue in place of the glucosamine 1-phosphate at the reducing end, and it bears a galacturonic acid moiety instead of a monophosphate at position 4'. We have detected the UDP-GlcNAc acyltransferases (Fig. 4) (29a) in extracts of R. leguminosarum. It will be interesting to determine whether R. leguminosarum also contains a disaccharide synthase, a 4' kinase, etc., and how the 2-deoxy-2-aminogluconate is made. Another feature of R leguminosarum is the presence of very long (28-carbon) acyl chains (11). The unique structure of lipid A in R. leguminosarum may reflect functional roles in symbiosis.

OVERVIEW OF ENDOTOXIN PHARMACOLOGY
Considerable progress has been made recently in defining the mechanisms by which endotoxins (lipid A) interact with animal cells (Fig. 8). During infections, LPS can dissociate from bacteria (or membrane fragments) and bind to LBP (59). LBP may function as a lipid transfer protein that delivers LPS to CD14 (59, 66). CD14 is a surface protein of macrophages and other LPS-responsive cells (55).
Discovery of the scheme in Fig. 8 was based on the effects of antibodies directed against macrophage surface proteins on the LPS response and on the LBP requirement for optimal LPS stimulation (59,66). Certain cells, such as 70Z/3 pre-B lymphocytes, lack CD14 (32). Although these cells can respond to high levels (nanomolar) of lipid A (32,49), they respond to low levels (picomolar) when transfected with CD14 (32).
CD14 is bound to the cell surface by a phosphatidylinositol anchor (55). CD14 may direct LPS to another, as yet unknown, transmembrane protein that generates intracellular signal(s). In macrophages, the signals resulting from this association stimulate transcription of mRNA encoding cytokines, like tumor necrosis factor and interleukin-1 (9,22). Excessive production of these proteins (not lipid A itself) is responsible for the symptoms of endotoxin-induced shock (9,22,35).
The model shown in Fig. 8 bears some resemblance to what is known about receptors for interleukin-6 (56). The relevant intracellular signals resulting from LPS stimulation remain to be defined, but rapid activation of tyrosine phosphorylation and MAP kinase(s) is known to occur (63).
Lipid A and LPS are also bound and internalized by cells that express the macrophage scavenger receptor (27; not shown in Fig. 8). Inhibition of uptake mediated by the scavenger receptor does not prevent stimulation of cytokine synthesis (27), indicating that the scavenger receptor is not involved in signal transduction.  (32,49,59,66). The identity of the putative transmembrane protein downstream of CD14 is still unknown, but binding of LPS to CD14 has recently been demonstrated (59a). The functioning of LBP as a transfer protein for lipid A is suggested by its sequence similarity to cholesteryl ester transfer protein (23). TNF, tumor necrosis factor; IL-1, interleukin-1.

APPROACHES TO THE TREATMENT OF
ENDOTOXIN-INDUCED SHOCK Complications of gram-negative infections are a common cause of death in debilitated patients (42). On the basis of current knowledge of the pathophysiology of endotoxemia (Fig. 8), three general approaches are being explored (20). The first involves sequestration of lipid A with antibodies (62) or with a bacteriocidal protein of neutrophils (64). Clinical trials with the former have been disappointing (20,62), and results with bacteriocidal protein are not yet available. The second approach involves blocking of the action of key cytokines, as with antibodies to tumor necrosis factor (16) or with a protein antagonist that binds to the interleukin-1 receptor (5,20). Despite initial promise, phase III trials in patients have failed to confirm good efficacy (5,20). The newest approach (not yet tested in patients) involves lipid A antagonists, such as analogs of R. sphaeroides lipid A (26, 57) (Fig. 7), to block LPS activation of macrophages (Fig. 8).
Blocking the initial event(s) should prevent the production of all mediators.
The difficulties encountered so far (20,62) in developing a therapy for endotoxin-induced shock will eventually be overcome. The enormous progress that has been made since 1983 in unravelling the chemistry and biology of bacterial endotoxins (36,46,53) has yielded a wealth of new approaches that remain to be tested.