The Biosynthesis of Gram-negative Endotoxin FORMATION OF LIPID A DISACCHARIDES FROM MONOSACCHARIDE PRECURSORS IN EXTRACTS OF ESCHERICHIA COLT*

We have discovered an enzyme in the cytosol of Escherichia coli that generates lipid A disaccharides from monosaccharide precursors by the following

saccharide lipid A precursor, 2,3-diacyl-GlcN-l-P,' and concurrent (unanticipated) revisions of the lipid A structure (4, discussed in the preceding article, have led us to hypothesize the two-stage scheme for the biosynthesis of lipid A (5) shown in Figs. 1 and 2. A central assumption is that 2,3-diacyl-GlcN-1-P and UDP-2,3-diacyl-GlcN condense to generate the lipid A disaccharide.
AS shown in the preceding article (6), we have documented the presence of the novel nucleotide, UDP-2,3-diacyl-GlcN, in living cells of Escherichia coli, consistent with our hypothesis (Figs. 1 and 2). We now demonstrate the existence of an enzyme in the cytosol of wild type E. coli catalyzing the reaction: 2,3-diacyl-GlcN-1-P + UDP-2,3-diacyl-GlcN + 2,3diacyl-GlcN ( p , 1 4 ) 2,3-diacyl-GlcN-l-P + UDP, as indicated in the lower half of Fig. 1. The disaccharide 1-phosphate product has been characterized by fast atom bombardment mass spectrometry and NMR spectroscopy, leading us to assign the structure shown at the bottom of Fig. 1. The striking deficiency of the disaccharide 1-phosphate synthase in extracts of mutants harboring the pgsBl lesion (14, 15) supports the view that this condensation is crucial for the formation of lipid A disaccharides. Furthermore, the deficiency of the disaccharide 1-phosphate synthase explains the concomitant accumulation of 2,3-diacyl-GlcN-1-P and UDP-2,3-diacyl-GlcN in mutants harboring the pgsBl mutation. The in uitro system described in the present article represents the first example of an enzymatic reaction for generating a lipid A disaccharide.

EXPERIMENTAL PROCEDURES
Materiak-'*Pi was a product of New England Nuclear. Glassbacked plates (5 X 20 cm), coated with a 250-micron layer of Silica Gel 60, were obtained from E. Merck, Darmstadt, Germany. Glassbacked plates (5 X 20 cm), coated with a 250-micron layer of Anasil H, were obtained from Analabs Inc., North Haven, CT. Silicic acid (200-400 mesh) was a product of Bio-Rad. Yeast extract, tryptone, and agar were products of Difco. Lipid X (2,3-diacyl-GlcN-l-P) was purified as described previously (3). All other chemicals were reagent grade and obtained from Sigma.
Preparation of Cell Extracts for Enzymatic Assays-Strains were grown on 50-100 ml of LB broth at 30 "C until A~M ) was 1.0. Cells were then harvested by centrifugation at 3,000 x gav, washed once, and suspended in 10 m M Hepes, pH 7.4. Next, the cells were disrupted using a French pressure cell at 18,000 psi. Protein concentration can cleave UDP-2,3-diacyl-GlcN to form lipid X.' Presumably, R-3hydroxymyristoyl acyl carrier protein (an intermediate in fatty acid biosynthesis (7, 8)) is the donor of the fatty acyl moieties to UDP-GlcN, or perhaps, UDP-GlcNac. UDP-2,3-diacyl-GlcN is thought to be a precursor of lipid X on the basis of pulse labeling experiments with 32Pi.3 In view of this, direct acylation of GlcN-1-P (5) seems less likely. In the above, R represents a 6-hydroxymyristoyl moiety.
Extraction of 32P-labeled 2,3-Diacyl-GlcN-l -P-The procedure used is a modification of the method of Takayama et al. (3). E. coli strain MN7 was grown on 10-20 ml of G56 medium at 30 "C until A m was approximately 0.5. Cells were sedimented in a clinical centrifuge and suspended in the above medium but without added phosphate. Next, %Pi was added (50 pCi/ml), and the culture was incubated at 42 "C for 3 h. The cells were harvested in a clinical centrifuge and extracted using the method of Bligh and Dyer (19) with phosphate-buffered saline, pH 7.4, as the aqueous component (14). Under these conditions lipid X is recovered in the aqueous methanol phase (14). Next, the aqueous methanol phase was acidified to pH 1 with HCl (14) and extracted with a fresh pre-equilibrated lower phase to obtain lipid X.
Removal of minor contaminants was achieved by chromatography on a column of BioSil A (3). The purified, 32P-labeled lipid X was stored at -80 "C in CHC13/methanol (21) at a specific radioactivity of 5 X 1 0 ' cpm/nmol.
Conditions for Fast Atom Bombardment Mass Spectrometry and NMR Spectroscopy-Fast atom bombardment mass spectrometry was B. L. Ray and C. R. H. Raetz, unpublished observations. C. E. Bulawa 2. Hypothetical pathway for the biosynthesis of lipid A, the "disaccharide stage." In vitro phosphorylation of the disaccharide 1-phosphate can he detected? S. typhimurium mutants defective in the biosynthesis of 3-deoxy-D-manno-octulosonate (9-11) accumulate a family of disaccharides of this kind under nonpermissive conditions, including a predominant one with the structure shown in the figure.4 Furthermore, the lipid A precursors that accumulate in 3-deoxy-D-manno-octulosonate deficient mutants are acceptors for the 3-deoxy-~-manno-octulosonate moiety of CMP-3-deoxy-~manno-octulosonate (12). Later stages of lipid A acylation occur after KDO addition (9, 12) but have not yet been studied in vitro. The biosynthesis of core and outer sugars has been reviewed elsewhere (13).
carried out as described previously for the analysis of lipid X (3). In general, mass assignments were made with an accuracy of f l atomic mass unit, when the MS-50 spectrometer (AEI/Kratos, Manchester, England) was interfaced with a DS-55 data system. 'H NMR spectra were recorded either at 200 MHz on a Nicolet NT-200 Fourier transform, superconducting spectrometer interfaced with a Nicolet 1280 computer or at 600 MHz at the National Institutes of Health National Facility for Biomedical Research at Carnegie-Mellon University. The samples were derivatized with diazomethane to methylate the phosphomonoester function increasing the solubility in CDCb, as reported previously (3,21). In all cases spectra were recorded at 25 "C with a sample concentration of 5-10 mg/ml in CDC13/dimethyl sulfoxide-ds (91, v/v). A complete 'H decoupling analysis was obtained from a two-dimensional 600 MHz proton spin echo J correlated (SECSY) NMR spectrum. For the SECSY experiment, 256 1K data blocks (100 transients each) were collected using a 90"-1/2 t1-90"-1/2 t,-FID (t2) pulse sequence where tl was varied in 358-ps increments and the relaxation delay was 0.75 s. The twodimensional data matrix was processed using zero-filling and sine bell apodization in both domains to give a final 512 X 2K time domain data matrix. 13C NMR spectroscopy was carried out on the NT-200 instrument operating at 50.31 MHz with bilevel proton decoupling to minimize sample heating. For analysis, 20 mg of underivatized disaccharide 1-phosphate (free-acid form) was dissolved in 0.5 ml of CDC13/methanol (91, v/v). When 32P-labeled lipid X was incubated by itself with a cellfree extract of E. coli, no appreciable change was detected over the course of 60 min under the conditions of Fig. 3, as judged by thin layer chromatography and autoradiography ( Fig. 3, lanes I, 3, and 5). Inclusion of 1 mM UDP-2,3-diacyl-GlcN during the incubation, however, caused a time-dependent transformation of the 32P-labeled lipid X to a new, more rapidly migrating product (Fig. 3, lanes 2,4, and 6 as indicated by the arrows). This observation was consistent with the formation of a tetraacyldisaccharide 1-phosphate, since the latter substance is less polar than X and has solubility properties that lie between those of lipid X and 2,3-diacyl-GlcN (dephospho X) (3). The inclusion in the reaction mixture of UDP-2,3-diacyl-GlcN also caused the formation of small amounts of a slower moving radioactive material (coincident with UDP-2,3-diacyl-GlcN) (Fig. 3), but this was not further characterized.
The enzyme catalyzing the reaction between 2,3-diacyl-GlcN-1-P and UDP-2,3-diacyl-GlcN has an optimal pH of about 8. The reaction is not dependent upon the addition of divalent cations or Triton X-100, while octyl-j3-D-glucoside is inhibitory (data not shown). Salmonella typhimurium extracts possess a similar enzyme activity. Approximately 70% of the E. coli enzyme is recovered in the supernatant fraction after a 90-min centrifugation at 150,000 x gaV, and it is inactivated by heating for 30 min a t 60 "C.
A quantitative assay for the enzyme is presented in Fig. 4, illustrating linearity with respect to protein concentration and time. The specific activity of the enzyme in crude E. coli extracts is about 1 nmol/min/mg under the conditions employed. This is comparable to most enzymes of glycerophospholipid metabolism (7). Despite the crude nature of the enzyme preparation, 70% (or more) of the 32P-labeled lipid X is converted to the new product after a prolonged incubation (Fig. 4B).
Rapid Purification of the Newly Synthesized Product-Since the enzyme responsible for the UDP-2,3-diacyl-GlcN-dependent metabolism of lipid X is not membrane bound, one can easily remove endogenous phospholipids and lipopolysaccha-FIG. 3. Conversion of "P-labeled lipid X to a more rapidly migrating compound dependent upon the addition of UDP-2,3-diacyl GlcN and an E. coli cell extract. The above is an overnight autoradiogram of a thin layer plate (Merck, Silica Gel 60) developed in the solvent chloroform/pyridine/formic acid (20307, v/ v). In each case a 4-pl sample of a 100-pl reaction mixture consisting of 20 mM Hepes, pH 8, 16 p~ 32P-labeled lipid X (5 X lo' cpm/nmol), and 1.7 mg/ml of crude extract of strain R477 was spotted on each lane. The numbers at the bottom indicate the minutes of incubation at 30 "C before the sample was spotted on the plate. The + sign indicates that the reaction mixture also contained 1 mM UDP-2,3diacyl-GlcN. The major, centrally located spot in each lane is 32Plabeled lipid X. The more rapidly migrating component, the formation of which is dependent upon UDP-2,S-diacyl GlcN, is the putative tetraacyldisaccharide 1-phosphate and is marked with an arrow. No product is generated in the absence of enzyme (not shown). A small amount of a slowly moving radiolabeled compound is also found when the nucleotide is present. It migrates with the same RF as the nucleotide and may be generated by an exchange reaction between 32Plabeled lipid X and the nonradioactive nucleotide. A portion was incubated for 5 h a t 30 "C without added enzyme, while the rest was incubated with a 150,000 X g supernatant of a wild type cell extract a t a final protein concentration of 2 mg/ ml. Next, 3 pl of each reaction mixture was spotted on a thin layer plate (250-micron Anasil H, Analabs, North Haven, CT) and developed in the solvent chloroform/pyridine/formic acid (2030:7, v/v). The solvent was removed by heating the plate. After cooling, the plate was sprayed with 10% H2S04 in ethanol, and spots were visualized by charring. Phte A represents the sample incubated without enzyme. The slower moving compound is UDP-2.3-diacyl-GlcN. while the more rapidly moving material is 2,3-diacyl-GlcN-l-P (lipid X). Plate B shows the results if the incubation is carried out in the presence of enzyme. After 5 h, both substrates have been depleted, but a new, very rapidly migrating component (the putative disaccharide 1-phosphate) is observed. It is important to note that the crude enzyme preparation employed is essentially free of contaminating phospholipids or lipopolysaccharide, since these would be detected by charring. The enzyme preparation by itself contains no disaccharide product (not shown). For purposes of large scale purification of the disaccharide product, we prepared a reaction mixture having a final volume of 18 ml and 1 mg/ml of supernatant protein prepared from a strain that carries a pgsE clone (15). The latter overproduces the synthase? In this case (data not shown) the reaction went to completion after 3 h. The disaccharide product was extracted from the reaction mixture without the addition of carrier by the Bligh-Dyer method with phosphate-buffered saline as the aqueous component. After conversion of the product to the free acid form by washing the lower layer with an acidic upper phase (14), the lower CHCl3 phase was dried under N2. We recovered 22 mg of product that was greater than 95% pure, as judged by thin layer chromatography. This material was used for subsequent physical studies. ride by ultracentrifugation (see above). If a reaction mixture containing 1 mM lipid X and 1 mM UDP-2,3-diacyl-GlcN is incubated in the presence of supernatant protein (1-2 mg/ ml), the conversion of lipid X and UDP-2,3-diacyl-GlcN to the new product can be monitored directly by thin layer chromatography (Fig. 5). At various times, a 3-pl sample of the reaction mixture is spotted on a Silica Gel H plate, subjected to chromatography in the solvent chloroform/pyridine/formic acid (20307, v/v), and visualized by charring. In the absence of enzyme (Fig. 5A), lipid X and UDP-2,3-diacyl-GlcN do not react with each other, while in the presence of D. N. Crowell and C. R. H. Raetz, unpublished observations. enzyme (Fig. 5B) both substrates disappear, and there is net synthesis of a more rapidly migrating product that corresponds to the fast moving 32P-labeled material shown in Fig.   3. Since lipid X and UDP-2,3-diacyl-GlcN are recovered in the aqueous methanol phase of a Bligh-Dyer partitioning carried out at pH 7, while the desired product is recovered in the lower chloroform phase, one can easily purify it from a reaction mixture such as that shown in Fig. 5. Using this approach, we isolated about 22 mg (Fig. 5, legend) and subjected it to various physical studies.
Analysis of the Reaction Product by Fast Atom Bombardment Mass Spectrometry-Fast atom bombardment mnss spectrometry was carried out both in the negative and in the positive mode. In the negative mode the molecular ion (M -H)-was observed at m/z 1323, as shown in Fig. 6. The small peak a t 1332, as well as the prominent peaks a t 148 and 297 were from the matrix used to apply the sample to the probe. There were no mass peaks above 1332. Based on this analysis the molecular weight of the reaction product is 1324 f 1, consistent with the structure shown at the bottom of Fig. 1.
Other peaks in the spectrum may be interpreted as follows: Taken together, the mass spectrometry strongly supports the proposed disaccharide structure (Fig. 1). The molecular formula of this material is C68H129N2020P, and the predicted molecular weight corrected for atomic abundance is 1325.74.

Characterization of the Disaccharide Product by 'H NMR
Spectroscopy-The purpose of the NMR analysis of the disaccharide product was 4-fold: (a) to verify the presence of two glucosamine residues in the product; ( b ) to confirm the location of the phosphate residue a t position 1 and the aanomeric configuration at the reducing end; (c) to verify the presence of ester-linked hydroxymyristoyl moieties at positions 3 and 3'; and (d) to show the presence of an axially disposed H-1' and a &1+6 linkage between the sugars.
As in previous 'H NMR spectroscopy of the monosaccharide precursors (3, 21), we methylated the phosphate residue of the disaccharide product by treatment with diazomethane. This enhanced its solubility in organic solvents. Initial spectra were recorded at 200 MHz at room temperature. The sample (5-10 mg/ml) was dissolved in CDC13/Me2SO-d6 (9:1, v/v). Fig. 7 shows the spectrum of the dimethyl disaccharide 1phosphate (designated DSMP in A) in comparison to that of dimethyl lipid X ( B ) . Both spectra are similar in the regions attributed to the terminal methyls (6, 0.9 ppm) and to the bulk of the methylene protons (6, 1.3) of the fatty acid moieties. There is a large envelope of unresolved resonances centered a t 6 3.9 ppm due to protons attached to hydroxylated carbon atoms (Fig. 7A). Superimposed on this are four sharp resonances due to the methyls attached to the phosphate. It is significant that upon integration (data not shown) the ratio of fatty acid methyl protons (6, 0.9) to phosphorus-linked OCH3 protons (6, 3.8-3.9) is exactly two times greater in the product generated in uitro (A), than in lipid X (B).
Resonances of particular interest are further downfield (6, 4.2-7.5 ppm), and these are especially well resolved. As expected for the structure under consideration (Fig. 7A), there are two distinct NH resonances (6 7.3 and 7.45 ppm), one Assignments were verified by systematic decoupling of all resolved resonances. Initial decoupling experiments were performed at 200 MHz, as described previously for lipids X and Y (3,21), but in addition the material was analyzed by twodimensional proton spin echo J correlated (SECSY) NMR spectroscopy at 600 MHz (25). The results for the region of interest (3-6 ppm) are shown in Fig. 8; they establish unequivocally the downfield shifting of H-3 and H-3' due to acylation at positions 3 and 3'. At 600 MHz, H-2 is completely resolved from overlapping resonances and shows a 1.8 Hz coupling to the phosphate group at C-1 (Fig. 8). Coupling constants for the spin-spin interactions of the various protons in the two glucosamine residues of the disaccharide product are assembled in Table I. Previously obtained values for dimethyl lipid X and dephospho X are provided for comparison (3). Importantly, the Jl,z of the reducing end unit (residue I) is 3.36, diagnostic of an a-anomeric configuration as in lipid X (23,24). On the other hand, the J1,,z,, representing the nonreducing unit (residue I1 in Table I (-) while those of residue I1 are indicated by a doshed line (---). The asterisk (*) indicates couplings between H-2 and NH, and H-2' and NH', but the NH and NH' amide protons resonate downfield of the 3-6 ppm window at 7. 33 and 7.46 ppm, respectively, as described in the legend to Fig. 7.   Analysis of pertinent resonances in the 13C spectrum of the disaccharide 1-phosphate confirms the presence of a / 3 linkage between the two sugars, since C-1 is observed at 94.8 ppm (typical of a GlcN residue with an a-anomeric configuration), while C-1' is shifted downfield to 101.2 ppm, typical of the / 3 configuration (Table 11) (4, 26, 27). In lipid X, the C-6 resonance is observed at 61.8 ppm under these conditions (data not shown), but in the disaccharide 1-phosphate it is shifted downfield to 67.9 ppm while C-6' is observed at 61.7 (Table 11). The magnitude and direction of this downfield shift, also observed in lipid A (27, 28), demonstrate the presence of a glycosidic linkage at C-6 rather than (2-4. Furthermore, the NMR evidence for a & l 4 linkage is consistent with model building studies in which steric problems are encountered when models are constructed with a ,8,14 linkage, related to the presence of the fatty acid substituents at 2, 3, 2', and 3'.
It should be noted that all resonances in the 13C spectrum can be assigned, assuming the structure shown at the bottom of Fig. 1.
Rebase of UDP from UDP-2,3-diacyl-GlcN in the Presence of Lipid X-Using the high performance liquid chromatography system described in the preceding article (6), we attempted to demonstrate UDP release during the formation of the disaccharide 1-phosphate under conditions otherwise identical with those described in Fig. 5. Unfortunately, the conversion of UDP to UMP and Pi was too rapid. However,  we have recently subcloned the pgsB gene from pLC26-43 of the Clarke and Carbon collection. In such plasmid bearing strains the specific activity of the disaccharide 1-phosphate synthase is 10-20 times higher than In extracts from these strains it is possible to detect UDP release from UDP-2,3-diacyl-GlcN dependent upon the addition of lipid X (data not shown). It will be necessary to utilize a highly purified enzyme preparation to demonstrate stoichiometry of disaccharide 1-phosphate synthesis and UDP formation.
Deficiency of the Disaccharide 1-Phosphate Synthase in Mutants Defective in the pgsB Gene-As shown in the preceding article (6), both 2,3-diacyl-GlcN-1-P (lipid X) and UDP-2,3-diacyl-GlcN accumulate in strains like MN7 (pgsA444 p g s B l ) or MN7B (pgsA'pgsB1) that harbor the pgsBl lesion. A plausible explanation for this biochemical alteration would be a deficiency in the disaccharide 1-phosphate synthase characterized above.
We prepared cell-free extracts of R477 (pgsA+pgsB+), MN1 (pgsA444 pgsB+), and MN7 (pgsA444 pgsBl ) and compared the rates of formation of the disaccharide 1-phosphate from lipid X and UDP-2,3-diacyl-GlcN. As shown in Fig. 9, there is a gross deficiency of the synthase in MN7 that can be attributed to the pgsBl lesion, since MN1 and MN7 are isogenic transductants (15). The specific activity of the enzyme is reduced by at least 30-fold.
The fact that the synthase is deficient in a mutant that is temperature-sensitive for growth and accumulates monosaccharide lipid A precursors (3,14,15,21,29) provides strong evidence that the enzyme represents a major (and perhaps the only) route for the biosynthesis of lipid A disaccharides in E. coli.

DISCUSSION
Considering the relative simplicity of the lipid A structure (Fig. 1 of the preceding article), it is surprising that little is known about its enzymatic synthesis. An important clue was provided by the discovery of incomplete lipid A disaccharide precursors in 3-deoxy-D-manno-octulosonate biosynthesis mutants (9)(10)(11), but the incorrect assumption that the 3deoxy-D-manno-octulosonate substituent was attached a t po-sition 3' of the disaccharide (30, 31) hindered elucidation of the relevant enzymology. The recent suggestion by Strain et al. (4) that 3-deoxy-D-manno-octulosonate is actually at position 6' and our concurrent discovery of the monosaccharide lipid A precursor, 2,3-diacyl-GlcN-1-P ( E . coli lipid X), makes it possible, for the first time, to postulate plausible biosynthetic routes (Figs. 1 and 2, and Ref. 5).
The disaccharide 1-phosphate synthase characterized above represents the first example of an enzyme capable of generating a lipid A disaccharide in vitro. The fact that the synthase makes the proper P , l 4 linkage found in mature lipid A suggests that the enzyme is a significant (and perhaps the only) source of this material in vivo. The existence of the disaccharide 1-phosphate synthase, the presence in vivo of its substrates 2,3-diacyl-GlcN-1-P and UDP-2,3-diacyl-GlcN (6), and the accumulation of both substrates in mutants deficient in the synthase (6) provide compelling evidence for the scheme shown in Figs. 1 and 2. Obviously, it will be necessary to search for additional enzymes and to isolate more mutants in order to verify the overall hypothesis. The characterization of the substrate specificity of the homogeneous synthase will also be very helpful.
The observation that the disaccharide 1-phosphate product synthesized in vitro has its four 8-hydroxymyristoyl substituents attached a t 2, 3, 2', and 3' (Fig. 8) has important implications for the conformation of lipid A. Presumably, the 8-hydroxymyristoyl residues are not relocated during subsequent lipid A maturation. An attractive feature of disaccharides consisting of 2 glucosamine residues, linked @,I-& and acylated at positions 2,3, 2', and 3', is their ability to assume tightly packed, planar configurations, compatible with insertion into phospholipid bilayers. A space filling model of this kind is shown in Fig. 10. It is important to note that C-6' (the site of polysaccharide attachment indicated by the uppermost arrow in A ) projects away from the hydrophobic domain. The phosphate moieties a t positions 1 (arrow on the right) and 4' (arrow on the left) are likewise removed from the hydrophobic region.
Additional constraints on the structure of lipid A disaccharides deserve mention: 1) the GlcN residues are locked into a chair conformation, as indicated by the NMR analysis; 2) based on x-ray studies of acylated sugars (32), it is likely that the carbonyl oxygens of the amide-and ester-linked hydroxymyristoyl functions are in a cis-like configuration relative to their respective H-2, H-3, etc.; and 3) the hydrophobic chains are likely to be extended and hexagonally packed, as suggested by the preliminary x-ray studies of Wawra et al. (33). In this context, one can speculate that the lipid A molecule has two nonidentical surfaces. We designate these as the "amide face" and the "ester face." The former may be defined by the presence of the @-OH moieties of the amide-linked /3-hydroxymyristoyl residues (visible and cross-hatched in Fig. lOA), while the latter is defined by the presence of the &OH moieties of the ester-linked @-hydroxymyristoyl residues. In a side view (reducing end in front), the amide face is on the left, while the ester face is on the right (Fig. 10B). Mature lipid A of E. coli and S. typhimurium bears two additional esterified fatty acyl residues (not shown in Fig. lo), a lauroyl and a myristoyl group (21, 34), both on the nonreducing residue (35). Based on the work of Wollenweber et al. (36), the lauroyl residue must be on the N-linked hydroxymyristoyl function, restricting the myristoyl residue to the 0-linked hydroxymyristoyl function at 3'. In the molecular analysis of Fig. 10, the lauroyl moiety would reside on the amide face (left side), while the myristoyl moiety would reside on the ester face (right side). In principle, the two surfaces of lipid A could interact with each other, and perhaps, this is important for outer membrane biogenesis. It will be of great interest to subject the disaccharide 1-phosphate (and other defined lipid A preparations) to further biophysical studies, such as NMR spectroscopy and x-ray diffraction. A conformational analysis of these substances before and after insertion into model lipid bilayers would be especially relevant.
The tentative identification of the pgsB locus as the structural gene for the tetraacyldisaccharide 1-phosphate synthase may facilitate the elucidation of the mechanism(s) by which pgsB interacts with pgsA (14,15), and with the metabolism of phosphatidylglycerol (14,15). We currently favor the view that UDP-2,3-diacyl-GlcN (a CDP-diglyceride-like molecule) or metabolite Z (6) inhibits the phosphatidylglycerophosphate synthase specified by the pgsA444 allele. Such an inhibition could even occur at the level of enzyme synthesis (14). In any event, the molecular cloning of the pgsB gene causes massive synthase overproduction,5 simplifying purification. In view of the high efficiency with which the enzyme converts our monosaccharide precursors to lipid A disaccharides in uitro, its potential for bio-organic synthesis in conjunction with chemical methods (37-39) deserves consideration. Using recombinant DNA techniques and colony autoradiography (40), it may be possible to broaden the substrate specificity of the disaccharide 1-phosphate synthase, facilitating the preparation of new lipid A analogs and probes for endotoxin function.