Structure-guided enzymology of the lipid Aacyltransferase LpxM reveals a dual activity mechanismDustin Dovalaa, Christopher M. Ratha, Qijun Hua, William S. Sawyera, Steven Shiab, Robert A. Ellingb, Mark S. Knappb,and Louis E. Metzger IVa,1

aInfectious Diseases, Novartis Institutes for BioMedical Research, Emeryville, CA 94608; and bGlobal Discovery Chemistry, Novartis Institutes for BioMedicalResearch, Emeryville, CA 94608

Edited by William Dowhan, McGovern Medical School, Houston, TX, and accepted by Editorial Board Member R. J. Collier August 23, 2016 (received for reviewJuly 1, 2016)

Gram-negative bacteria possess a characteristic outer membrane,of which the lipid A constituent elicits a strong host immune responsethrough the Toll-like receptor 4 complex, and acts as a component ofthe permeability barrier to prevent uptake of bactericidal compounds.Lipid A species comprise the bulk of the outer leaflet of the outermembrane and are produced through a multistep biosyntheticpathway conserved in most Gram-negative bacteria. The final steps inthis pathway involve the secondary acylation of lipid A precursors.These are catalyzed by members of a superfamily of enzymes knownas lysophospholipid acyltransferases (LPLATs), which are present inall domains of life and play important roles in diverse biologicalprocesses. To date, characterization of this clinically important class ofenzymes has been limited by a lack of structural information and theavailability of only low-throughput biochemical assays. In this work,we present the structure of the bacterial LPLAT protein LpxM, and wedescribe a high-throughput, label-free mass spectrometric assay tocharacterize acyltransferase enzymatic activity. Using our structureand assay, we identify an LPLAT thioesterase activity, and we provideexperimental evidence to support an ordered-binding and “reset”mechanistic model for LpxM function. This work enables the interro-gation of other bacterial acyltransferases’ structure–mechanism rela-tionships, and the assay described herein provides a foundation forquantitatively characterizing the enzymology of any number of clin-ically relevant LPLAT proteins.

LpxM | acyltransferase | lipid A | RapidFire mass spectrometry |phosphopantetheine ejection assay

Among the most intractable of the challenges in the develop-ment of new antimicrobial therapeutics is finding chemicalmatter that can permeate the Gram-negative cellular envelope.Such bacteria pose a unique challenge in this regard due to theircharacteristic outer membrane (OM)—an asymmetric barrier withan inner leaflet primarily composed of glycerophospholipids andan outer leaflet composed of lipopolysaccharide (LPS). In addi-tion to its structural roles in decreasing permeability and in-creasing the rigidity of the bacterial cell, LPS is a potent activatorof the innate immune response and is recognized at picomolarlevels by the Toll-like receptor 4 (TLR4) (1). For these reasons, itis critically important to develop a deep understanding of thepathways involved in LPS biogenesis and modification.The lipid anchor of LPS, lipid A, forms the outer leaflet of the

OM and is the epitope of LPS that is recognized by TLR4. Thiscomplex lipid is produced in a nine-step conserved pathway knownas the Raetz pathway (2), of which the first six steps are essential inmost Gram-negative bacteria, including in clinically relevant spe-cies such as Escherichia coli, Klebsiella pneumoniae, Pseudomonasaeruginosa, and most strains of Acinetobacter baumannii. The en-zymes in this pathway first convert UDP-GlcNAc, acyl-acyl carrierprotein (acyl-ACP), and ATP into lipid IVA, to which secondaryacyl chains and 3-deoxy-D-manno-oct-2-ulosonic acid (Kdo) sugarsare added to produce lipid A.The final steps in the Raetz pathway involve the secondary

acylation of the R-3-hydroxyacyl chains at positions 2′ and 3′ of

Kdo2-lipid IVA to produce lipid A, catalyzed by the LpxL andLpxM enzymes, respectively. These enzymes belong to a large anddiverse superfamily of proteins known as the lysophospholipidacyltransferases (LPLATs), which exist within all domains of life.Within this family, LpxL and LpxM belong to a group of proteinsknown as lipid A biosynthesis lauryl/myristyl acyltransferases(LABLATs). Other representative clades in the LPLAT familyinclude glycerol-3-phosphate acyltransferases (GPATs), mono-acylglycerol acyltransferases (MGATs), and acylglycerolphosphateacyltransferases (AGPATs), which are each necessary for the earlysteps in triglyceride biosynthesis (3, 4).To date, the structure of only one member of this superfamily

has been reported: a stromal GPAT protein from Cucurbitamoschata (the squash plant) (5). The squash protein is only dis-tantly related to the bacterial LPLAT family (with some similarityto acyltransferases in Chlamydiae species, but not to proteobacte-rial LPLATs). Furthermore, the C. moschata GPAT has not beencocrystallized with substrates or substrate analogs. Thus, despitecareful biochemical analyses of these proteins (6), there is currentlyno direct structural evidence published that identifies a mechanismof action, nor have binding sites for acyl-ACP and/or lipid substratebeen reported. Elucidation of bacterial LABLAT structures mayenable the discovery of compounds that perturb lipid A biosynthesis,

Significance

Lysophospholipid acyltransferase (LPLAT) proteins are requiredfor many essential biological activities involving the transfer ofacyl chains. One LPLAT, LpxM, is necessary for the biosynthesisof lipid A, which comprises the outer leaflet of the outer mem-brane in Gram-negative bacteria. Lipid A is important because itis a potent activator of the innate immune system and becauseof its role in preventing xenobiotics from permeating Gram-negative bacteria. In this work, we structurally and mechanisti-cally characterize LpxM, providing insights that may enable thetargeted discovery of inhibitors that prevent lipid A maturation;these might potentiate the uptake of extant antibiotics whoseclinical efficacy is hitherto limited by poor permeability. Our in-sights into the mechanism of LpxM may facilitate the study ofdiverse LPLATs.

Author contributions: D.D., Q.H., and L.E.M. designed research; D.D., C.M.R., Q.H., W.S.S.,S.S., R.A.E., and L.E.M. performed research; C.M.R., W.S.S., and M.S.K. contributed newreagents/analytic tools; D.D., C.M.R., W.S.S., and L.E.M. analyzed data; S.S., R.A.E., andM.S.K. provided technical assistance with crystallography; and D.D. and L.E.M. wrotethe paper.

The authors declare no conflict of interest.

This article is a PNAS Direct Submission. W.D. is a Guest Editor invited by the EditorialBoard.

Freely available online through the PNAS open access option.

Data deposition: The atomic coordinates and structure factors have been deposited in theProtein Data Bank, www.pdb.org (PDB ID codes 5KN7 and 5KNK).1To whom correspondence should be addressed. Email: louis.metzger_iv@novartis.com.

This article contains supporting information online at www.pnas.org/lookup/suppl/doi:10.1073/pnas.1610746113/-/DCSupplemental.

E6064–E6071 | PNAS | Published online September 28, 2016 www.pnas.org/cgi/doi/10.1073/pnas.1610746113

Dow

nloa

ded

by g

uest

on

July

6, 2

021

http://crossmark.crossref.org/dialog/?doi=10.1073/pnas.1610746113&domain=pdfhttp://www.pdb.orghttp://www.rcsb.org/pdb/explore/explore.do?structureId=5KN7http://www.rcsb.org/pdb/explore/explore.do?structureId=5KNKmailto:louis.metzger_iv@novartis.comhttp://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1610746113/-/DCSupplementalhttp://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1610746113/-/DCSupplementalwww.pnas.org/cgi/doi/10.1073/pnas.1610746113

resulting in leaky membranes that may decrease virulence and po-tentiate the uptake of bactericidal compounds.Whereas the LABLAT proteins are not essential for bacterial

growth under typical laboratory conditions, their genetic deletionsresult in a wide variety of phenotypes among different bacterialspecies. In particular, disruption of lpxM can have a profound im-pact on the integrity of the OM, which in turn can have importanteffects on cell survival in certain environments. In K. pneumoniae,deletion of lpxM results in increased permeability in the OM andsubsequent sensitivity to cationic antimicrobial peptides (CAMPs)(7). A. baumannii (8), E. coli (9), and Salmonella typhimurium (9)also require functional LpxM for CAMP resistance, possibly due toincreased OM permeability in its absence. In Vibrio cholerae, theLpxM ortholog LpxN transfers a 3-hydroxylaurate to the 3′-linkedacyl chain on its lipid A precursor, which is then modified withglycine to confer resistance to CAMP (10, 11). Thus, LpxM, andother LpxM orthologs, may contribute both directly and indirectlyto CAMP resistance.Deletion of lpxM has also been shown to have important effects

on virulence in many human and animal pathogens. In the noso-comial pathogen A. baumannii, LpxM is required for virulence in theGalleria mellonella (greater wax moth) infection model, and forprotection from desiccation, which increases infection and trans-mission rates (8). Additionally, in E. coli clinical isolate H16, deletionof chromosomal lpxM results in a drastic decrease in virulence after i.p. injection into BALB/c mice (12), and deletion of the gene in theE058 strain of avian pathogenic E. coli (APEC) results in defects inavian macrophage invasion and decreased bacterial loads in severalorgans following inoculation into the left thoracic air sac in chickens(13). Salmonella enterica serovar Typhimurium infection of ligatedbovine ileal loops is also attenuated in lpxM deletion mutants,demonstrating reduced inflammatory response and correlating to adecrease in secretion of virulence factors through the type III se-cretion system (14). Interestingly, all documented pathogenic strainsof Shigella flexneri, another Gram-negative that causes disease inhumans, carry a virulence plasmid that includes a second copy oflpxM (15). This second lpxM (msbB2) is regulated in response tomagnesium by the two-component PhoP/PhoQ regulatory system,perhaps as a mechanism to decrease the permeability of the OM andto enhance S. flexneri survival in an intracellular environment (16).Deletion of either or both lpxM paralogs in S. flexneri results in de-creased host immune response, as well as defects in bacterial in-

vasion and replication in epithelial cells (15, 17). In virulent strains ofYersinia pestis, deletion of lpxM does not appear to affect the LD50after s.c. injection in mice (18). However, deletion of lpxM in thevaccine strain EV results in a significant increase in protective im-munity, while decreasing endotoxicity (18, 19). Thus, whereas notessential for viability of these pathogens, LpxM is still relevant withrespect to virulence and perhaps to the permeability of antibioticcompounds.In this work, we present the X-ray crystal structure of a bac-

terial LPLAT protein, the LABLAT LpxM from the pathogenicbacterium A. baumannii (AbLpxM), which possesses two knownacyltransferase activities (8) (Fig. 1). Further, we report a high-throughput label-free mass spectrometric assay for rapid andquantitative measurement of acyltransferase activity, and we usethis assay in conjunction with the structure to devise a model forthe mechanism of action of this important class of enzymes.

ResultsThe Structure of AbLpxM Reveals a Deep Hydrophobic Binding Pocket.To begin to build a structure–function relationship for LABLATproteins, we solved the structure of the A. baumannii LpxM ho-molog using single wavelength anomalous dispersion to 1.99-Åresolution (Fig. 2 and SI Appendix, Table S1). AbLpxM adopts aglobular structure bisected by a large seven-stranded β-sheet. Thepredicted transmembrane domain is composed of a single α-helixprotruding from the globular domain and forms a substantialcrystal contact with AbLpxM in a neighboring asymmetric unit.Electrostatic surface visualization of the AbLpxM structure

revealed a very large hydrophobic pocket, which bound a co-purifying n-dodecyl-β-D-maltoside (DDM) molecule (Fig. 2B andSI Appendix, Fig. S3C). Due to the hydrophobicity and size of thelipid substrate of AbLpxM, and given the lack of any other largehydrophobic surface, this pocket likely represents the bindingsite of the acyl chain acceptor. Intriguingly, this large pocket alsocontains several deep hydrophobic channels, which may serve ashydrocarbon “rulers,” and which may determine specificity foracyl chain length on either acyl-ACP or on the lipid A precursors(Fig. 2C, red arrows). Supporting a potential role as a hydro-carbon ruler, one of these pockets, in the structure of the cata-lytic mutant AbLpxME127A, contains electron density consistentwith the presence of an acyl chain (SI Appendix, Fig. S3D).

2 x Lauryl-ACP

2 x holo-ACP

AbLpxM

LpxMLpxM

O

ONH

O

O

OO

HO

O

HOO

O

NHO

O

OO

O

O

OO

O

PHO O

HO

OOH

HOOH

OHO

OOH

HOOH

O

HO

HO

O

POH

O

HO

12

12

12

12

12 14LpxL

KdtA

O

ONH

O

HO

OO

HO

HOO

O

NHO

O

OO

HO

O

OOP

HO O

HO

OOH

HOOH

OHO

OOH

HOOH

O

HO

HO

O

POH

O

HO

12

12

12

1414

4′

12

3

4 56

1′2′

3′

5′6′

14

4′

12

3

4 56

1′2′

3′

5′6′

Fig. 1. Reaction scheme for AbLpxM. The two acyl chains transferred by AbLpxM are highlighted in green. The Kdo sugars (added by KdtA) and the sec-ondary acylation on the R-3-hydroxylaurate at position 2′ (added by LpxL) may or may not be present in the AbLpxM lipid substrate and are highlighted inyellow. Acyl chain lengths shown are representative of A. baumannii strain ATCC 17978 (33). Note that the acyl chain lengths and compositions differ amongother bacterial species and strains.

Dovala et al. PNAS | Published online September 28, 2016 | E6065

BIOCH

EMISTR

YPN

ASPL

US

Dow

nloa

ded

by g

uest

on

July

6, 2

021

http://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1610746113/-/DCSupplemental/pnas.1610746113.sapp.pdfhttp://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1610746113/-/DCSupplemental/pnas.1610746113.sapp.pdfhttp://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1610746113/-/DCSupplemental/pnas.1610746113.sapp.pdf

90°

N

C

TM

A

B C

Fig. 2. The structure of AbLpxM. (A) Cartoon schematic of AbLpxM. From the N terminus to C terminus, the color shifts from blue to red. Two angles areshown. TM, predicted transmembrane helix. (B) Predicted surface electrostatics map generated by PyMol. The putative binding cleft is indicated with a dashedline. (C) Cutaway diagram showing deep pockets within the putative binding cleft (pockets indicated with red arrows; cutaway in black to show the depth ofthe binding cleft).

E6066 | www.pnas.org/cgi/doi/10.1073/pnas.1610746113 Dovala et al.

Dow

nloa

ded

by g

uest

on

July

6, 2

021

www.pnas.org/cgi/doi/10.1073/pnas.1610746113

As there is no discernible primary sequence identity betweenAbLpxM and the squash GPAT, or between AbLpxM and any otherstructures deposited in the Protein Data Bank (PDB), we lookedfor structural homologs using the programs PDBeFold (20) andDaliLite (21). Although there were no high-confidence hits, bothPDBeFold and DaliLite identified the squash GPAT as the closeststructural homolog to AbLpxM—albeit with less than half of thesecondary structural elements aligned. The residues with the mostsimilar secondary structure comprised the region surrounding theactive site, suggesting that the positioning of the catalytic residues inthe GPATs and LABLATs rely on conserved secondary structuralelements. Overall, however, the secondary structure differs greatlybetween LpxM and the squash GPAT (Q score 3 dB) and linearity down to∼100 nM of holo-ACP product (SI Appendix, Fig. S2 B and C)without any additional sample preparation.With this assay, we measured acyl-transferase activity for AbLpxM

using lauryl-ACP as an acyl chain donor and titrating lipid IVA as theacyl chain acceptor. As reported for E. coli LpxL (6), we found thatAbLpxM was capable of using lipid IVA as a substrate, even lackingthe Kdo moieties, with an apparent KM of 1.7 ± 0.6 μM (Fig. 3A).Indeed, AbLpxM produced both penta- and hexa-acylated lipid Aspecies when incubated with lauryl-ACP and lipid IVA as detected byquadrupole (Q)-TOF liquid chromatography (LC)-MS and TLC (SIAppendix, SI Results and Figs. S7C, S10, and S11).Interestingly, we found that AbLpxM could produce holo-ACP

in the absence of any acyl chain acceptor, suggesting a previouslyunreported acyl-protein thioesterase activity (Fig. 3A). Supportingthis hypothesis, we found that AbLpxM produced free lauric acidwhen incubated with lauryl-ACP, as determined by accurate-massQ-TOF LC-MS (Fig. 3C). Indeed, this enzyme-dependent hy-drolysis was ∼60-fold greater than the spontaneous rate of pro-duction without protein in our assay conditions. We note that thisactivity would be observed only by direct measurement of holo-ACP production, and not by tracking radiolabeled lipid A pre-cursors as has previously been done (6, 9).

The AbLpxM Active Site Resides Within the Deep Hydrophobic Cleft.Acyltransferase proteins typically contain an invariant histidineresidue in their active sites, followed by four nonconserved residues,and then a catalytic acidic residue (HX4D/E) (23). This catalyticdyad is thought to create a charge relay system, wherein the his-tidine abstracts the proton from the hydroxyl group and activatesthe oxygen for nucleophilic attack on the acyl thioester carbon (23,24). LpxM also contains a conserved HX4D/E motif (using glu-tamate in the A. baumannii ortholog), which is present within thedeep hydrophobic cleft (Fig. 3D). The positioning of these resi-dues within the cleft supports our hypothesis that this is the lo-cation of the active site. Aside from the invariant His122, analignment of 27 LpxM and LpxL orthologs revealed three othertotally conserved residues: Asn51, Arg159, and Asp192 (Fig. 3E).

Interestingly, Arg159 also resides within the cleft and is positionedvery close to the putative catalytic histidine (His122) and glutamate(Glu127). Analysis of the crystal structure suggests that this residueinteracts with the C6-hydroxyl on the bound DDM molecule. Ifthe diglucosamine head group of lipid IVA binds similarly to themaltose head group of DDM, this interaction would correspond tothe C1 phosphate on lipid IVA, suggesting that Arg159 may play arole in substrate binding.We wished to examine the possible role(s) of Arg159 for both

acyltransferase and acyl-protein thioesterase activities. We there-fore expressed and purified AbLpxMR159A and measured its ac-tivity with increasing concentrations of lipid IVA (Fig. 3A). We alsoproduced and assayed AbLpxME127A, which was expected toeliminate activity through substitution of the putative catalyticglutamate. We found that both AbLpxMR159A and AbLpxME127Asubstitutions greatly abrogated activity of the enzyme (Fig. 3A).However, a closer examination of the kinetics of both variantsrevealed residual activity that differed greatly between the two (Fig.3B). Whereas the AbLpxME127A variant strongly responded to in-creasing lipid IVA substrate, the AbLpxMR159A variant displayed amuch more attenuated response (Fig. 3B). These data are consis-tent with the hypothesis that the conserved Arg159 residue con-tributes to either substrate recognition and binding or to the properpositioning of substrates before the acyl chain transfer.To confirm proper folding of the active site in the catalytic var-

iant, we solved the structure of AbLpxME127A to 1.9-Å resolution(SI Appendix, Fig. S3A and Table S1). AbLpxME127A adopted analmost identical conformation as the wild-type protein, except fortwo notable differences in the rotomeric orientation of His98 andTrp126 (which defines one end of the active site; SI Appendix, Fig.S3B). Like the wild-type protein, the catalytic variant also copurifiedand cocrystallized with DDM (SI Appendix, Fig. S3C).

AbLpxM Demonstrates Acyl Chain Length Selectivity and SubstrateInhibition. We hypothesized that the deep channels in the putativeactive site (Fig. 2C) may act as rulers to gauge acyl chain lengthof either acyl chains on the acceptor lipid IVA or the donor acyl-ACP. To test whether AbLpxM displays a preference for acylchain length on acyl-ACP, we measured AbLpxM activity usingeither capryl-, lauryl-, myristyl-, or palmityl-ACP. We found thatAbLpxM displayed a strong preference for lauryl-ACP over theother acyl chain lengths tested (Fig. 4 A and B and Table 1), witha measured apparent KM of 1.8 ± 0.2 μM. As the acyl chainlength deviated from 12 carbons in either direction, the KM in-creased and the specific activity decreased. A strong preferenceby AbLpxM for a lauryl-ACP donor is also consistent with carefulmeasurements of the mass of lipid A produced by A. baumanniiwith and without LpxM (8).We also observed that AbLpxM displays substrate inhibition

with respect to lauryl-ACP, with a steady decrease in activity atlauryl-ACP concentrations above 7 μM. This substrate inhibitionwas observed in both the presence and absence of lipid IVA (SIAppendix, Fig. S4A), and is consistent with an ordered bindingmechanism. We hypothesize that premature binding of acyl-ACPto AbLpxM may sterically occlude the binding site and preventproper binding of lipid IVA. Indeed, our observed biochemicalinhibition may be physiologically relevant as reported cellularACP concentrations in E. coli are on the order of 40–100 μM(25). Whereas the exact distribution of the bound acyl species isunknown, it is reasonable to propose that the concentration oflauryl-ACP falls at or near the inhibited regime observed.

Kinetic Analyses Suggest AbLpxM Possesses Multiple ACP-BindingSites. Contrary to our initial hypothesis that AbLpxM has oneacyl-ACP binding site that is substrate inhibited (SI Appendix, Fig.S4C, orange curve), we observed that as lauryl-ACP concentrationwas increased, the substrate inhibition decreased AbLpxM activ-ity only by approximately twofold. Classical substrate inhibition

Dovala et al. PNAS | Published online September 28, 2016 | E6067

BIOCH

EMISTR

YPN

ASPL

US

Dow

nloa

ded

by g

uest

on

July

6, 2

021

http://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1610746113/-/DCSupplemental/pnas.1610746113.sapp.pdfhttp://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1610746113/-/DCSupplemental/pnas.1610746113.sapp.pdfhttp://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1610746113/-/DCSupplemental/pnas.1610746113.sapp.pdfhttp://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1610746113/-/DCSupplemental/pnas.1610746113.sapp.pdfhttp://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1610746113/-/DCSupplemental/pnas.1610746113.sapp.pdfhttp://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1610746113/-/DCSupplemental/pnas.1610746113.sapp.pdfhttp://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1610746113/-/DCSupplemental/pnas.1610746113.sapp.pdfhttp://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1610746113/-/DCSupplemental/pnas.1610746113.sapp.pdfhttp://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1610746113/-/DCSupplemental/pnas.1610746113.sapp.pdfhttp://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1610746113/-/DCSupplemental/pnas.1610746113.sapp.pdfhttp://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1610746113/-/DCSupplemental/pnas.1610746113.sapp.pdfhttp://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1610746113/-/DCSupplemental/pnas.1610746113.sapp.pdfhttp://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1610746113/-/DCSupplemental/pnas.1610746113.sapp.pdfhttp://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1610746113/-/DCSupplemental/pnas.1610746113.sapp.pdfhttp://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1610746113/-/DCSupplemental/pnas.1610746113.sapp.pdf

kinetics, assuming a single ACP-binding site, would predict that asthe concentration of substrate approaches infinity, the velocityshould asymptotically approach zero (SI Appendix, Fig. S4C, or-ange curve, and SI Results for equations).This prediction was not observed in the case of AbLpxM sub-

strate inhibition with lauryl-ACP. We therefore attempted tomodel our kinetic data with the assumption that AbLpxM containsat least two ACP-binding sites, one of which follows standardnonsubstrate-inhibited kinetics. We found that assuming thepresence of two ACP-binding sites, we could accurately model

the behavior of the enzyme (Fig. 4D and SI Appendix, Fig.S4C, purple curve, and SI Appendix, SI Results for equations),supporting the hypothesis that AbLpxM indeed possessesmultiple ACP-binding sites. Because AbLpxM acylates lipid IVAat two distinct sites, the hypothesis of multiple ACP-binding sitesis reasonable.We hypothesized that the observed substrate inhibition resulted

from premature ACP binding sterically occluding the active sitecleft and preventing the second substrate (lipid IVA or H2O in thecase of the thioesterase activity) from binding and/or properly

A B C

D E

Fig. 3. Kinetic analysis of AbLpxM activity. (A) Specific activity of AbLpxM as a function of lipid IVA concentration for wild-type AbLpxM, AbLpxME127A, andAbLpxMR159A. (B) Close-up of the kinetics of AbLpxME127A and AbLpxMR159A. (C) Production of lauric acid by AbLpxM in the presence of lauryl-ACP after 2 h. (D)Close-up of the AbLpxM putative active site with conserved residues and their interaction distances shown. DDM is shown in purple. (E) Logo showing conservationof residues shown in D. All error bars represent the SEM. All experiments were repeated in triplicate. Specific activity represents the production of holo-ACP.

E6068 | www.pnas.org/cgi/doi/10.1073/pnas.1610746113 Dovala et al.

Dow

nloa

ded

by g

uest

on

July

6, 2

021

http://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1610746113/-/DCSupplemental/pnas.1610746113.sapp.pdfhttp://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1610746113/-/DCSupplemental/pnas.1610746113.sapp.pdfhttp://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1610746113/-/DCSupplemental/pnas.1610746113.sapp.pdfhttp://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1610746113/-/DCSupplemental/pnas.1610746113.sapp.pdfwww.pnas.org/cgi/doi/10.1073/pnas.1610746113

positioning itself with respect to the AbLpxM catalytic residues.These observations led us to reexamine the AbLpxM structures forputative ACP-binding sites within or near the active site cleft. Bylooking for large patches of predicted electropositive surfacecharge, we identified a putative ACP-binding site (Fig. 4C and SIAppendix, Fig. S4D), which would position ACP directly over the

entrance to the active site. To test our hypothesis, we substitutedtwo residues (K282E/K285E) within the putative binding site toinvert the surface charge and abrogate ACP binding (Fig. 4C, Inset,and SI Appendix, Fig. S4D). That ACP interaction with proteins isprimarily driven by electrostatics is well established (26, 27). Indeed,we found that by substituting these residues, the observed substrateinhibition was attenuated, and the remaining AbLpxM activityexhibited a Vmax of ∼0.6·min−1, directly supporting our model (SIAppendix, Fig. S4C, blue curve, and Fig. 4D, red curve). These datasuggest that: (i) the electropositive patch near the active site rep-resents an ACP-binding site; (ii) ACP binding to this site is re-sponsible for the substrate inhibition; and (iii) there is at least oneother ACP-binding site located elsewhere on the protein that isresponsible for the residual, noninhibited activity.

DiscussionA Proposed Model for LpxM Activity. Based on the available data,we propose the following model for an ordered-binding and“reset”mechanism for LpxM (Fig. 5). We hypothesize that underideal conditions, LpxM first binds to either lipid IVA, Kdo2-lipidIVA, or Kdo2-(lauryl)-lipid IVA (the preferred substrate is notclear), followed by lauryl-ACP. Proper positioning of the ac-ceptor lipid substrate is facilitated by interactions with Arg159,possibly at either the 1- or 4′-phosphate on the diglucosaminehead group, whereas the acyl chain lengths on either (or both)substrates are regulated by deep hydrophobic channels within theputative binding cleft. Upon proper positioning of the substrate,coordinating the hydroxyl of the R-3-hydroxyacyl chain at position3′ (and position 2, for the A. baumannii ortholog) near the cata-lytic dyad, the acyl chain transfer is catalyzed and holo-ACP andlipid A products are generated and released. However, if lauryl-ACP binds first, then the lipid acceptor substrate is sterically oc-cluded from binding to the cleft, and the enzyme is trapped in anonproductive state. In this case, the LpxM thioesterase activityensures that the lauryl-ACP is eventually hydrolyzed to holo-ACPand laurate, thereby regenerating active LpxM enzyme at an en-ergy cost of just one ATP (one ATP is required to regenerate thelauryl-ACP), as opposed to, in the extreme example, synthesizinganother LpxM molecule at a much higher energetic cost.The reset mechanism may also prevent acyl-ACP from being

sequestered by LpxM during conditions when its lipid acceptorsubstrates are scarce (such as when cells slow their biosynthesisof lipid A precursors and of other OM substituents). Rather thanbeing an aberrant or promiscuous activity, we hypothesize thatthis release allows bacteria to efficiently conserve resources asthe supply of lipid A precursors fluctuates, which is in turn re-flective of the cells’ energy status.In addition to this reset mechanism, we propose that the

A. baumannii LpxM ortholog binds to two acyl-ACP molecules onseparate and distinct binding surfaces, one of which is partiallyregulatory.

Table 1. Analysis of acyl-ACP acyl chain length specificity onAbLpxM thioesterase activity as determined by high-throughputmass spectrometry

Acyl chaindonor KM

App, μMSpecific

activity, min−1 Hill coefficient

Capryl-ACP 9.887 ± 0.481 0.238 ± 0.004 1.134 ± 0.041Lauryl-ACP* 1.845 ± 0.238 1.483 ± 0.099 1.392 ± 0.134Myristyl-ACP 10.35 ± 0.943 0.346 ± 0.013 1.328 ± 0.115Palmityl-ACP 30.47 ± 10.3 0.057 ± 0.009 1.128 ± 0.196

Values were calculated from data shown in Fig. 4A using Matlab; errorshown as 95% confidence interval. All experiments were repeated in triplicate.*Parameters for lauryl-ACP were generated using data in the noninhibitedregime only.

k cat, m

in-1

0

0.2

0.4

0.6

0.8

1.0

1.2

1.4

1.6

Capry

l-AbA

CP

Laury

l-AbA

CP

Myris

tyl-AbA

CP

Palm

ityl-AbA

CP

1.6

1.4

1.2

1.0

0.8

0.6

0.4

0.2

0

0 10 20 30 40 50 60[Acyl-AbACP], μM

Spe

cific

Act

ivity

, min

-1

Capryl-AbACPLauryl-AbACP

Myristyl-AbACPPalmityl-AbACP

[AbLpxM] = 50 nM

Lys282

Lys285

1.4

1.2

1.0

0.8

0.6

0.4

0.2

0

0 10 20 30 40 50[Lauryl-AbACP], μM

Spe

cific

Act

ivity

, min

-1

AbLpxMAbLpxMK282E,K285E

A B

C

D

Fig. 4. Analysis of AbLpxM activity with acyl-ACP. (A) Specific activity ofAbLpxM with increasing concentrations of acyl-ACP with carbon lengths 10(capryl-), 12 (lauryl-), 14 (myristyl-), and 16 (palmityl-) as determined by high-throughput mass spectrometry. Inhibited region for lauryl-ACP shown in red.Error bars represent the SEM. (B) Calculated apparent kcat for AbLpxM witheach acyl-ACP. Error bars represent 95% confidence interval. (C) Identifica-tion of possible ACP-binding site near the putative active site. Inset showsresidues that were predicted to drive the interaction and were substituted.(D) Data shown in A for lauryl-ACP fit to new kinetic model (black), as well asthe kinetics of AbLpxMK282E,K285E with increasing lauryl-ACP (red). All ex-periments were repeated in triplicate. Specific activity represents the pro-duction of holo-ACP.

Dovala et al. PNAS | Published online September 28, 2016 | E6069

BIOCH

EMISTR

YPN

ASPL

US

Dow

nloa

ded

by g

uest

on

July

6, 2

021

http://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1610746113/-/DCSupplemental/pnas.1610746113.sapp.pdfhttp://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1610746113/-/DCSupplemental/pnas.1610746113.sapp.pdfhttp://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1610746113/-/DCSupplemental/pnas.1610746113.sapp.pdfhttp://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1610746113/-/DCSupplemental/pnas.1610746113.sapp.pdfhttp://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1610746113/-/DCSupplemental/pnas.1610746113.sapp.pdf

Substrate Specificity of LpxM. Whereas LpxM is typically depictedas accepting Kdo2-lauryl-lipid IVA as its substrate, making it theterminal enzyme in the Raetz pathway, many LpxM orthologsexhibit substantial substrate promiscuity. For example, Neisseriameningitides produces fully acylated lipid A even when the Kdotransferase, kdtA, is knocked out, suggesting LpxL and LpxM canuse lipid IVA in this species (28). Similarly, P. aeruginosa can pro-duce fully acylated lipid A even when CMP-Kdo synthase (KdsB) isinhibited or when acting upon purified lipid IVA (29, 30), suggestingthat the presence of the Kdo moieties on the lipid acceptor is dis-pensable for some LpxL/LpxM orthologs. In A. baumannii, LpxMadds lauryl groups to the R-3-hydroxyacyl chains at both the 3′- and2-positions of lipid A precursors, producing hepta-acylated lipid A(8). This secondary activity may necessitate flexibility in the activesite to accommodate multiple conformations of lipid A, so as toposition the substrate properly. Our data, taken together with pre-vious studies (8), suggest that the A. baumannii LABLAT proteinslikely lack any defined order of operations, as both AbLpxL andAbLpxM produce lipid products in the absence of the other. In-terestingly, when an alignment of LpxM orthologs is examined,those enzymes with predicted substrate promiscuity (Neisseria,Pseudomonas, and Acinetobacter) align more closely together,compared with orthologs that are reported to be relatively specific

in their acyl transfer activities (SI Appendix, Fig. S5). This alignmentsuggests that there may exist a common set of mutations that relaxesligand specificity. Close analysis of the sequence differences be-tween these and other LpxM orthologs may reveal more informa-tion about substrate specificity.

Concluding Remarks. LPLAT proteins are key enzymes in manyimportant biological processes in humans and may represent usefultargets for pharmacological intervention—especially in the treat-ment of diabetes and obesity (31). To date, several MGAT inhib-itors have been developed and patented for this purpose (32).Whereas there is still no published structure of a mammalianLPLAT protein, let alone representative structures from eachfamily within the LPLAT superfamily, comparisons betweenthe structures of A. baumannii LpxM and C. moschata GPATmay enable researchers to make inferences about which structuralelements are broadly conserved and likely found among mam-malian LPLAT homologs and which may be more specific tocertain subfamilies.Our detailed characterization of A. baumannii LpxM may be

relevant to the discovery of inhibitors of lipid A maturation inA. baumannii and in other Gram-negative pathogens, as our workmay enable both high-throughput biochemical screening andstructure-aided inhibitor discovery. We hypothesize that it may beadvantageous to select for inhibitors of the LpxM thioester hy-drolase activity or of its acyl-ACP binding, rather than for inhib-itors that compete with the lipid IVA acceptor substrate (as suchlipid IVA-competitive inhibitors might tend to possess unfavorableproperties such as high molecular mass and lipophilicity). Ourhigh-throughput MS assay system may allow for the discovery ofsuch inhibitors in the absence of lipid IVA. We hypothesize thatmolecules that stabilize the LpxM–acyl-ACP complex (Fig. 5, step5) might have the dual benefit of preventing lipid A maturationand of sequestering acyl-ACP, thus depleting ACP pools andpossibly causing broad metabolic disruption in the bacteria.Our findings may facilitate several interesting areas of inquiry in

the acyltransferase field. We have discovered that at least oneLPLAT protein contains a thioesterase activity in addition to itsknown acyltransferase activity. It is likely that other LPLAT pro-teins, even ones that have been studied extensively using radio-labeled lipid analysis, also contain secondary activities thatmay be regulatory in nature. Further examination using ourassay platform may identify opportunities for both direct andallosteric inhibition of targeted enzymes relevant to human disease.Overall, our structural and mechanistic characterizations, coupledwith a robust and target-adaptable biochemical MS assay platform,set the stage for a broad characterization of LPLAT structure–function relationships.

Materials and MethodsNative and SeMet-derivatized N-terminally polyhistidine-taggedA. baumanniiLpxM (AbLpxM) were heterologously overexpressed in E. coli membranes.These membranes were detergent solubilized, and AbLpxM was purified byimmobilized metal affinity and ion exchange chromatographic methods.Crystals of AbLpxM were grown by sitting-drop vapor diffusion at roomtemperature in 200 mM NaBr, 2.2 M (NH4)2SO4, at a protein:mother liquorratio of 70:30 and an AbLpxM concentration of 21 mg/mL (SI Appendix, Fig.S2). The crystals were cryoprotected with the addition of 20% glycerol tothe mother liquor solution. Data were collected for native and Se-MetAbLpxM crystals at the Advanced Light Source beamline 5.0.2 at wave-lengths of 1.000 Å and 0.9797 Å, respectively; these data were used to solvethe structure using single anomalous dispersion methods. Enzyme assay datawere collected by high-throughput automated SPE-MS using an AgilentRapidFire 300 front-end and an AB SCIEX 5500 mass spectrometer. Thephosphopantetheine ejection assay, which monitors a diagnostic fragmention formed from the preferential release of the serine-bound prostheticgroup, was used to monitor the depletion of the acyl-ACP substrates and thegeneration of holo-ACP products of the reaction (SI Appendix, Fig. S2). Fullmethods are available in the SI Appendix, SI Materials and Methods.

Lipid IVA binds Lauryl-ACP bindsand blocks binding cleft

Lipid IVA cannot fitinto the active site

Lauryl-ACP bindsat two sites

Laurate transferis catalyzed

Thioester linkages onlauryl-ACP are cleaved

Substrates are released,regenerating the enzyme

holo-ACP

holo-ACP

Lauryl-ACP

dilauryl-Lipid IVA

Lauric Acid

AMP + PPi

ATP

AasS

1

2

3

4

5

6

7

8

9

Fig. 5. Model for AbLpxM mechanism. (Step 1) AbLpxM binds to its lipidsubstrate (shown here as lipid IVA). (Step 2) Following binding of the lipidsubstrate, lauryl-ACP binds at two sites, one of which is over the active sitecleft. (Step 3) The acyl chain transfer is catalyzed by the HX4D/E catalyticdyad. (Step 4) The holo-ACP and acylated lipid products dissociate fromAbLpxM, regenerating the enzyme. (Step 5) If lauryl-ACP concentrations aretoo high, or lipid substrate concentrations are too low, then lauryl-ACP canbind AbLpxM first. (Step 6) When this happens, the lipid substrate cannot fitinto the active site. (Step 7) AbLpxM then cleaves the thioester linkage be-tween the lauric acid and the phosphopantetheine on the lauryl-ACP. (Step8) This regenerates the enzyme, which may otherwise be trapped in anonproductive state. (Step 9) Each lauryl-ACP can be regenerated with oneATP by acyl-ACP synthase (AasS).

E6070 | www.pnas.org/cgi/doi/10.1073/pnas.1610746113 Dovala et al.

Dow

nloa

ded

by g

uest

on

July

6, 2

021

http://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1610746113/-/DCSupplemental/pnas.1610746113.sapp.pdfhttp://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1610746113/-/DCSupplemental/pnas.1610746113.sapp.pdfhttp://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1610746113/-/DCSupplemental/pnas.1610746113.sapp.pdfhttp://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1610746113/-/DCSupplemental/pnas.1610746113.sapp.pdfhttp://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1610746113/-/DCSupplemental/pnas.1610746113.sapp.pdfwww.pnas.org/cgi/doi/10.1073/pnas.1610746113

ACKNOWLEDGMENTS. We thank David A. Six for helpful discussions,Jennifer Leeds and David Carpenter for critical reading of the manuscript,and members of the Emeryville Protein Sciences, Structural Chemistry,Bacteriology New Technologies, Global Discovery Chemistry, and legal

groups at Novartis Institutes for BioMedical Research for help andsupport. We acknowledge the legacy of Christian R. H. Raetz (1946–2011), discoverer of the Raetz pathway of lipid A biosynthesis and mentorto L.E.M.

1. Zughaier SM, Zimmer SM, Datta A, Carlson RW, Stephens DS (2005) Differential in-duction of the toll-like receptor 4-MyD88-dependent and -independent signalingpathways by endotoxins. Infect Immun 73(5):2940–2950.

2. Whitfield C, Trent MS (2014) Biosynthesis and export of bacterial lipopolysaccharides.Annu Rev Biochem 83(1):99–128.

3. Shi Y, Cheng D (2009) Beyond triglyceride synthesis: The dynamic functional roles ofMGAT and DGAT enzymes in energy metabolism. Am J Physiol Endocrinol Metab297(1):E10–E18.

4. Takeuchi K, Reue K (2009) Biochemistry, physiology, and genetics of GPAT, AGPAT,and lipin enzymes in triglyceride synthesis. Am J Physiol Endocrinol Metab 296(6):E1195–E1209.

5. Turnbull AP, et al. (2001) Analysis of the structure, substrate specificity, and mecha-nism of squash glycerol-3-phosphate (1)-acyltransferase. Structure 9(5):347–353.

6. Six DA, Carty SM, Guan Z, Raetz CRH (2008) Purification and mutagenesis of LpxL,the lauroyltransferase of Escherichia coli lipid A biosynthesis. Biochemistry 47(33):8623–8637.

7. Clements A, et al. (2007) Secondary acylation of Klebsiella pneumoniae lipopoly-saccharide contributes to sensitivity to antibacterial peptides. J Biol Chem 282(21):15569–15577.

8. Boll JM, et al. (2015) Reinforcing lipid A acylation on the cell surface of Acinetobacterbaumannii promotes cationic antimicrobial peptide resistance and desiccation sur-vival. MBio 6(3):e00478-15.

9. Tran AX, et al. (2005) Resistance to the antimicrobial peptide polymyxin requiresmyristoylation of Escherichia coli and Salmonella typhimurium lipid A. J Biol Chem280(31):28186–28194.

10. Hankins JV, et al. (2011) Elucidation of a novel Vibrio cholerae lipid A secondaryhydroxy-acyltransferase and its role in innate immune recognition. Mol Microbiol81(5):1313–1329.

11. Hankins JV, Madsen JA, Giles DK, Brodbelt JS, Trent MS (2012) Amino acid addition toVibrio cholerae LPS establishes a link between surface remodeling in gram-positiveand gram-negative bacteria. Proc Natl Acad Sci USA 109(22):8722–8727.

12. Somerville JE, Jr, Cassiano L, Darveau RP (1999) Escherichia coli msbB gene as a vir-ulence factor and a therapeutic target. Infect Immun 67(12):6583–6590.

13. Xu H, et al. (2013) Role of the lpxM lipid A biosynthesis pathway gene in pathoge-nicity of avian pathogenic Escherichia coli strain E058 in a chicken infection model.Vet Microbiol 166(3–4):516–526.

14. Watson PR, et al. (2000) Mutation of waaN reduces Salmonella enterica serovarTyphimurium-induced enteritis and net secretion of type III secretion system1-dependent proteins. Infect Immun 68(6):3768–3771.

15. D’Hauteville H, et al. (2002) Two msbB genes encoding maximal acylation of lipid Aare required for invasive Shigella flexneri to mediate inflammatory rupture and de-struction of the intestinal epithelium. J Immunol 168(10):5240–5251.

16. Goldman SR, Tu Y, Goldberg MB (2008) Differential regulation by magnesium of thetwo MsbB paralogs of Shigella flexneri. J Bacteriol 190(10):3526–3537.

17. Ranallo RT, et al. (2010) Virulence, inflammatory potential, and adaptive immunityinduced by Shigella flexneri msbB mutants. Infect Immun 78(1):400–412.

18. Anisimov AP, et al. (2007) Effect of deletion of the lpxM gene on virulence andvaccine potential of Yersinia pestis in mice. J Med Microbiol 56(Pt 4):443–453.

19. Feodorova VA, et al. (2009) Pleiotropic effects of the lpxM mutation in Yersinia pestisresulting in modification of the biosynthesis of major immunoreactive antigens.

Vaccine 27(16):2240–2250.20. Krissinel E, Henrick K (2004) Secondary-structure matching (SSM), a new tool for fast

protein structure alignment in three dimensions. Acta Crystallogr D Biol Crystallogr

60(Pt 12 Pt 1):2256–2268.21. Holm L, Rosenstrom P (2010) Dali server: Conservation mapping in 3D. Nucleic Acids

Res 38(Web Server issue):W545–W549.22. Meluzzi D, Zheng WH, Hensler M, Nizet V, Dorrestein PC (2008) Top-down mass

spectrometry on low-resolution instruments: Characterization of phosphopantethei-

nylated carrier domains in polyketide and non-ribosomal biosynthetic pathways.Bioorg Med Chem Lett 18(10):3107–3111.

23. Heath RJ, Rock CO (1998) A conserved histidine is essential for glycerolipid acyl-

transferase catalysis. J Bacteriol 180(6):1425–1430.24. Röttig A, Steinbüchel A (2013) Acyltransferases in bacteria. Microbiol Mol Biol Rev

77(2):277–321.25. Lu P, Vogel C, Wang R, Yao X, Marcotte EM (2007) Absolute protein expression

profiling estimates the relative contributions of transcriptional and translational

regulation. Nat Biotechnol 25(1):117–124.26. Masoudi A, Raetz CRH, Zhou P, Pemble CW, 4th (2014) Chasing acyl carrier protein

through a catalytic cycle of lipid A production. Nature 505(7483):422–426.27. Zhang Y-M, Wu B, Zheng J, Rock CO (2003) Key residues responsible for acyl carrier

protein and beta-ketoacyl-acyl carrier protein reductase (FabG) interaction. J Biol

Chem 278(52):52935–52943.28. Tzeng YL, Datta A, Kolli VK, Carlson RW, Stephens DS (2002) Endotoxin of

Neisseria meningitidis composed only of intact lipid A: Inactivation of the me-

ningococcal 3-deoxy-D-manno-octulosonic acid transferase. J Bacteriol 184(9):2379–2388.

29. Goldman RC, Doran CC, Kadam SK, Capobianco JO (1988) Lipid A precursor fromPseudomonas aeruginosa is completely acylated prior to addition of 3-deoxy-D-

manno-octulosonate. J Biol Chem 263(11):5217–5223.30. Mohan S, Raetz CR (1994) Endotoxin biosynthesis in Pseudomonas aeruginosa: En-

zymatic incorporation of laurate before 3-deoxy-D-manno-octulosonate. J Bacteriol

176(22):6944–6951.31. Nelson DW, Gao Y, Yen M-I, Yen C-LE (2014) Intestine-specific deletion of acyl-CoA:

monoacylglycerol acyltransferase (MGAT) 2 protects mice from diet-induced obesity

and glucose intolerance. J Biol Chem 289(25):17338–17349.32. Ohshiro T, Tomoda H (2015) Acyltransferase inhibitors: A patent review (2010-pre-

sent). Expert Opin Ther Pat 25(2):145–158.33. Pelletier MR, et al. (2013) Unique structural modifications are present in the lipo-

polysaccharide from colistin-resistant strains of Acinetobacter baumannii. Antimicrob

Agents Chemother 57(10):4831–4840.

Dovala et al. PNAS | Published online September 28, 2016 | E6071

BIOCH

EMISTR

YPN

ASPL

US

Dow

nloa

ded

by g

uest

on

July

6, 2

021