Escherichia coli α-Hemolysin (HlyA) Is Heterogeneously Acylated in Vivo with 14-, 15-, and 17-Carbon Fatty Acids*

α-Hemolysin (HlyA) is a secreted protein virulence factor observed in certain uropathogenic strains ofEscherichia coli. The active, mature form of HlyA is produced by posttranslational modification of the protoxin that is mediated by acyl carrier protein and an acyltransferase, HlyC. We have now shown using mass spectrometry that these modifications, when observed in protein isolated in vivo, consist of acylation at the ε-amino groups of two internal lysine residues, at positions 564 and 690, with saturated 14- (68%), 15- (26%), and 17- (6%) carbon amide-linked side chains. Thus, HlyA activated in vivo consists of a heterogeneous family of up to nine different covalent structures, and the substrate specificity of the HlyC acyltransferase appears to differ from that of the closely related CyaC acyltransferase expressed by Bordetella pertussis.

Other RTX toxins that have been characterized include adenylate cyclase toxin (ACT, the CyaA protein from Bordetella pertussis) and leukotoxin (LktA) from Pasteurella hemolytica. This class of proteins is defined by a common feature, a repeat region rich in aspartic acid and glycine. They are secreted by pathogens during infection and are known to attack host immune system cells, such as macrophages and lymphocytes. At high concentrations they are lytic for their respective populations of target cells. At sublytic concentrations HlyA is known to disrupt host cell signal transduction and cytokine production (1,2).
An intracellular activation process is shared by the members of the RTX toxin family (3). In HlyA, for example, the activation process is dependent on the product of an accessory gene, hlyC, and is accomplished by acyl-acyl carrier protein (acyl-ACP) dependent fatty acylation (4,5). Studies of in vitro activated HlyA protein have revealed that two internal residues, 2 Lys-564 and Lys-690, are acylated (6). Analysis of the in vivo activated HlyA protein using two-dimensional gels has confirmed the sites of activation (7) but not the chemical structures of the modifications. The structural details of the fatty acylation of CyaA from B. pertussis have been studied. Active CyaA isolated from B. pertussis was observed to be palmitoylated at only one internal lysine residue, Lys-983, which corresponds to Lys-690 of HlyA (8) according to RTX toxin sequence alignments. Similarly, Lys-860 of CyaA corresponds to Lys-564 of HlyA. However, recombinant CyaA, isolated from E. coli containing cyaA and cyaC genes, is acylated at two sites. Lys-983 is about 87% palmitoylated and the remainder myristoylated, and Lys-860 is approximately two-thirds palmitoylated (9). More recently, it has become clear that acylations observed in CyaA show a high degree of variability, especially at Lys-860. We have now observed Lys-860 acylation, by an as yet uncharacterized side chain, in recent preparations of ACT expressed by B. pertussis (BP338), 3 the same strain we first analyzed in 1994 (8) and again in 1995 (9). Similarly, we have observed Lys-860 palmitoylation in a recombinant form of CyaA ex-  1 The abbreviations used are: Hly, hemolysin; ACP, acyl carrier protein; ACT, adenylate cyclase toxin; CAD, collision-activated dissociation; CyaA, synonymous with ACT; EI, electron impact; ESI, electrospray ionization; FAME, fatty acid methyl ester; GC/MS, gas chromatography/mass spectrometry; HPLC, high performance liquid chromatography; LB, Luria-Bertani; Lkt, leukotoxin; MALDI-TOF, matrix-assisted laser desorption/ionization time-of-flight; MS, mass spectrometry; RTX, repeat in toxin. 2 For the sake of clarity, we have consistently used the nomenclature of Stanley and coauthors (2,6) for the locations of the two acylation sites, which is based on a sequence 1024 amino acid residues in length. Certain strains of E. coli produce HlyA (e.g. HlyA J96 ) that contains 1023 amino acids and thus has modified residues at Lys-689 and Lys-563. The inferred amino acid sequences of J96 (GenBank accession no. P09983, 1023 amino acids) and pHly152 (GenBank TM accession number P08715, 1024 amino acids) differ at several positions with substitutions and deletions, yielding the net effect of a single amino acid difference in their length. 3 Chen, W., and Hackett, M., manuscript in preparation.
pressed in B. pertussis strain 18323/pHSP9. 4 However, CyaA appears to require fatty acylation only at Lys-983 for cellinvasive activity. 5 In contrast to the somewhat more complex set of observations for CyaA, all preparations of native, active HlyA reported to date have probably been fully acylated at both lysines (2). Alterations in fatty acid composition are believed to result in functional changes, and the non-acylated protoxin is inactive in the case of both CyaA (9, 10) and HlyA (6). Interestingly, acylation is not a requirement for secretion of either CyaA or HlyA but is required at both sites in HlyA for full activity as a cytolysin or hemolysin (11). RTX toxin maturation and activation studies up to 1998 (2) have been reviewed in the context of other types of lipid protein modifications. In order to determine the structural details of fatty acylation present in in vivo activated HlyA, we analyzed the proteolytic fragments of HlyA using the techniques of peptide mapping, off-line HPLC, MALDI-TOF mass spectrometry, and electrospray ionization (ESI) mass spectrometry coupled with microcapillary HPLC, capillary gas chromatography coupled with electron impact mass spectrometry (GC/MS), and peptide synthesis. Two different sources of HlyA, one chromosomal (J96) and the other extrachromosomal (pHly152), and several different preparations from one source (J96) were used in order to control for potential variability between laboratories and individual preparations.

E. coli Strains Used for Toxin Production-Two forms of activated
␣-hemolysin were used, HlyA J96 and HlyA pHly152 . ProHlyA J96 (the inactive protoxin control) and HlyA J96 were prepared by either ammonium sulfate or polyethylene glycol precipitation of LB broth culture supernatants of E. coli strains WAM783 (12) and WAM1824 (13). Ammonium sulfate was preferred due to the tendency of polyethylene glycol residue to interfere with analysis by mass spectrometry. The hlyA gene in HlyA J96 is based on the hemolysin recombinant plasmid pSF4000 encoding the hly operon from the pathogenicity island (PAI IV) at 64 min in the chromosome of E. coli uropathogenic strain J96 (14,15). LB broth cultures (2 liters) were grown at 37°C with moderate aeration to an absorbance (A 600 nm ) of 0.8, or about 5 ϫ 10 8 cells/ml; the cells were removed by centrifugation, and ammonium sulfate was added to the supernatants to a final concentration of 60% by volume. The suspension was stirred and kept at 4°C for 1 h. The precipitate was collected by centrifugation at 16,000 rpm for 15 min; the pellet was resuspended in 8 M urea, 50 mM Tris, pH 8.0, and frozen at Ϫ70°C. ProHlyA and HlyA represented at least 90% of the protein present in the precipitates, based on SDS-polyacrylamide gel electrophoresis and Coomassie Brilliant Blue staining. The molecular genetics (16) and preparation (7) of activated HlyA pHly152 have been described elsewhere. A single preparation of roughly 300 pmol of HlyA pHly152 was used for this study.
Purification of Toxins-We purified approximately 500 pmol of protein per digest, using a Mini Prep tube gel apparatus (Bio-Rad). Approximately 140 fractions, 250 l each, were collected after eluting the purified protein at 200 V (constant voltage) and 100 l/min from the 1.5-cm 4% polyacrylamide stacking gel and the 5-cm 7.5% resolving gel. We combined all fractions containing hemolysin and concentrated them using a 30,000-Da cut-off Centricon concentrator (Amicon, Beverly, MA). The concentrated HlyA was washed three times with 2-ml aliquots of 50 mM Tris-HCl, pH 8.5, 4 M urea, and 15% acetonitrile to ensure removal of SDS. The acetonitrile was necessary to minimize losses of the hydrophobic acylated toxin and to maintain solubility during proteolytic cleavage. HlyA pHly152 was purified using SDS-polyacrylamide gel electrophoresis as described (7) and digested in the wash buffer under the same conditions as HlyA J96 .
Enzymatic Digestion with Lys-C-Achromobacter proteinase I (EC 3.4.21.50, from Dr. T. Masaki, Ibaragi University) was used for all initial digestions of HlyA. The toxin was digested at a ratio of 0.2 g of enzyme per 1 nmol of substrate for 20 -24 h at 37°C. The digested products (ϳ500 l) were frozen at Ϫ40°C until fractionation by HPLC.
Microbore HPLC and Screening by MALDI-TOF Mass Spectrometry-A PLRP-S 100-Å, 5-m, 250 ϫ 2.1-mm HPLC column (Polymer Laboratories, Amherst, MA) was used to separate the Lys-C digest. All digests were fractionated with a binary gradient HPLC system (Shimadzu model LC-10AD pumps, Kyoto, Japan) modified in-house with a Michrom mixer (Auburn, CA) for use with 2-mm inner diameter columns (17), 200 l/min, and detection at 214 nm (Shimadzu Model SPD-10A) with a 2.5 l flow cell. Gradient conditions were 5% acetonitrile (0.085% trifluoroacetic acid) for 5 min to 40% acetonitrile over 50 min, where the majority of fragments eluted, followed by a 10-min linear gradient to 95% acetonitrile with a 5-min hold. Fractions were collected at a rate of 1/min. The acylated peptides eluted during the second gradient. Each fraction was screened for molecular weight using MALDI-TOF mass spectrometry (Perspective Biosystems Model Voyager-Elite, Foster City, CA). Predicted mass values were calculated with the aid of Sherpa, a Macintosh-based protein and peptide analysis program, from the hlyA gene sequences for J96 (18) and pHly152 (16).
Microcapillary HPLC Electrospray Ionization Tandem Mass Spectrometry-Results from the initial screening indicated which HPLC fractions contained putative modification sites. Those fractions were analyzed by microcapillary HPLC coupled to a TSQ 7000 electrospray tandem quadrupole mass spectrometer (Finnigan, San Jose, CA). One l of each Lys-C fraction was eluted from a 50-m inner diameter ϫ 12-cm capillary column packed with Monitor 5-m 100-Å C18-modified silica (Column Engineering, Ontario, CA), at a flow rate of 150 nl/min, 5-95% acetonitrile (1% acetic acid) over 15 min. Initial main beam results (a single stage of mass spectrometry) indicated that charge states of the major peaks of interest were ϩ3 or higher. Tryptic subdigestion was used to create fragments with abundant ϩ2 charge states, which served as better precursor ions for the second stage in a sequential product ion experiment, with respect to ease of data interpretation. Peptides were subdigested using sequencing grade modified trypsin (Promega, Madison, WI). A small portion of each fraction was diluted with an equal volume of 50 mM Tris, pH 8.5, and the sample pH was verified. Trypsin, 1 g, was added to roughly 10 pmol of substrate at 37°C for 8 h. For the hydrophobic and acylated tryptic peptides, we used a less retentive PLRP-S 5-m 4000-Å packing (Polymer Labs) in a 12-cm microcapillary column. All product ion collision-activated dissociation (CAD) experiments were conducted with 3 millitorrs of argon in the collision cell (q2) at collision offsets between Ϫ20 and Ϫ40 V. The details of our approach to capillary HPLC coupled with electrospray tandem mass spectrometry, including plumbing of the Shimadzu HPLC and modifications to the Finnigan electrospray interface, have been described (17). Our approach is based on previous work in Hunt's laboratory (19) using column and sample loading technology originally developed by Jorgenson and co-workers (20,21).
Modification of Native Peptides and Synthesis of Acylpeptide Standards-We treated the Lys-C and tryptic subdigest fractions containing acylated peptides with methanolic HCl to convert any carboxylate side chains, as well as C termini, to their methyl esters (22). This allowed unambiguous identification of Asp and Glu. Acetylation was employed to derivatize unmodified lysine and N termini, allowing easy differentiation between Lys and Gln, and y series and b series ions (23) in the CAD spectra. Synthetic standards, prepared using Fmoc (N ␣ -9-fluorenylmethoxycarbonyl) chemistry for both acylated and non-acylated peptides, were used to confirm the structures assigned by mass spectrometry. The details of the derivatization reactions and the synthetic chemistry have been described (24).
Fatty Acid Analysis-Both peptide fragments and intact protein were analyzed for fatty acid content measured as fatty acid methyl esters (FAMEs) using the direct transesterification method (25). In addition to C12:0 as a positive control and internal standard, additional positive control and blank experiments were performed concurrently to ensure the validity of the data. GC/MS conditions were as follows: 30 m ϫ 0.32 mm inner diameter BPX5 column (SGE, Austin, TX) with 0.25-m film thickness, 100 -250°C 15-min temperature gradient on a Hewlett-Packard (Palo Alto, CA) 5890 GC and 7673 autosampler, interfaced with a Trio 2000 quadrupole MS (Micromass, Manchester, UK) 70-eV electron impact ionization. For quantitation, the GC/MS total ion currents for each fatty acid were measured relative to the C12:0 internal standard and corrected for any background level contamination.

RESULTS AND DISCUSSION
We had tried unsuccessfully to digest HlyA with trypsin, possibly due to incomplete denaturation of HlyA in 1 M urea. Lys-C, which can tolerate a higher urea concentration (4 M) was a better choice. This was consistent with the difficulties ob-served by Ludwig et al. (7) when they attempted to digest their HlyA with trypsin or V8 protease.
C14:0 Fatty Acylation-HPLC fraction 39 from HlyA J96 (Fig. 1A) contained two peptides that can be related to the predicted Lys-C fragments, with C14:0 acylation at Lys-690 (VLQEVVKEQEVS-VGK C14:0 RTEK and EQEVSVGK C14:0 RTEK), based on their molecular weights. Subdigestion of fraction 39 with trypsin and the MALDI-TOF MS results showed a peak at m/z 1243 which corresponded to the product, EQEVSVGK C14:0 R. CAD experiments showed a fragmentation spectrum ( Fig. 2A) which yielded the expected y and b series ions. Virtually identical CAD fragmentation patterns were observed when a separate CAD experiment was performed using a C14:0 acylated synthetic peptide, thus confirming the C14:0 acylation at Lys-690 on HlyA J96 . This work was repeated, with the same results, for HlyA pHly152 . HPLC fraction 44 contained a Lys-C fragment with putative C14:0 acylation at Lys-564, FVTPLLTPGEEIRERRQSGK C14:0 YEYITELLVK (m/z 3777). This peptide was subdigested; CAD spectra were acquired, and the C14:0 modification was confirmed by synthesizing the tryptic frag-ment and analyzing the fatty acyl group itself by GC/MS. Again, the work was repeated for HlyA pHly152 , with identical results. No modification at Lys-690 or Lys-564 was observed with the biologically inactive ProHlyA J96 control.
C15:0 and C17:0 Fatty Acylation-Our initial goal was to identify the expected C14:0 acylation sites on HlyA, based on our interpretation of work by the Cambridge group (6) and Ludwig et al. (7). However, we found Lys-C HlyA J96 fragments that eluted later than fractions 39 and 44 and that yielded m/z values not predicted by theoretical peptide maps, allowing for acylation by common fatty acids such as C12:0, C14:0, C16:0, C18:0 or their unsaturated forms. Examples include peptides at m/z 1615 and 2411 in fraction 42 (Fig. 1B) and m/z 1643 and 2439 in fraction 41 (Fig. 1C), which are 14 and 42 mass units larger than the two C14:0-acylated Lys-690 peptides found in fraction 39 (Fig. 1A). Similarly, in fraction 45, two Lys-C fragments 14 and 42 mass units larger than the C14:0 acylated Lys-564 peptide in fraction 44 were also observed. We observed similar results for HlyA pHly152 but not for the inactive ProHlyA J96 control. A number of potential modifications could give rise to these observations, based on molecular weight information alone, but subsequent analysis of these proteolytic fragments by CAD and electron impact GC/MS of the FAMEs clearly identified the fragments as C15:0 and C17:0 modifications of Lys-690 and Lys-564. Fig. 2B shows the CAD spectra of the C15:0-modified Lys-690 peptide (HPLC fraction 42) acquired following a tryptic subdigestion. The fragmentation pattern was identical to its C14:0 counterpart (Fig. 1A), except that starting at the y 2 and b 8 ions, an increase of 14 mass units was noted. Similar CAD results were observed for the Lys-C fragments of HlyA pHly152 . We also converted the putative C15: 0-acylated peptides in HPLC fraction 42 to methyl esters. Fig.  3 shows the expected mass increases from the original peptides (Fig. 1B) due to the addition of methyl ester groups to the C terminus and the aspartic acid residues. Tryptic subdigestion of these methyl ester derivatives also yielded a peptide with the expected CAD spectrum, suggesting a C15:0 acylation at Lys-690. We repeated these procedures for the remaining HPLC fractions, 41 and 45, which contained the putative C17:0 acylation at Lys-690 and both C15:0 and C17:0 at Lys-564. The MALDI-TOF and electrospray CAD data can successfully pinpoint the locations of these fatty acylations, but studies of the modifying groups themselves were necessary to confirm the conclusions suggested by the peptide CAD data. FAME analy-sis by capillary GC/MS provided both retention time data and the EI fragmentation patterns that taken together were unequivocal for the assignments of C15:0 (Fig. 4A) and C17:0 ( Fig.  4B) as modifying groups.
Quantitative Analysis-Based on the capillary GC retention times for well characterized FAME standards, we identified the peaks corresponding to the C14:0, C15:0, and C17:0 modifications of HlyA. The amounts of each present, as calculated from the peak areas with background corrections, were 68% C14:0, 26% C15:0, and 6% C17:0, with a relative standard deviation of about 8%. This was in good agreement with our expectations based on the peptide electrospray data.
Biological Significance-Our results demonstrate that E. coli ␣-hemolysin from two sources with different genetic histories consists of a heterogeneous group of acylated proteins, as many as nine (two acylation sites and three possible modifying groups at each site). In contrast to the situation with CyaA described earlier, the data acquired over a 3-year period, involving many preparations of HlyA J96 , were remarkably consistent. The results for a single preparation of HlyA pHly152 were identical to those from HlyA J96 , within the expected limits of experimental error. No evidence of palmitoylation was observed in HlyA from either source.
Our data support the hypothesis that the substrate specificities of HlyC and CyaC (the acyltransferase from B. pertussis FIG. 3. MALDI-TOF mass spectrum of the HPLC fraction containing C15:0-acylated Lys-690, as shown in Fig. 1B, but after chemical conversion to methyl ester (*) derivatives. The conversion took place at the C terminus and the aspartic acid residues. The first peptide increased its mass-to-charge ratio from its original m/z 1615 (see Fig. 1B) to m/z 1671, which corresponded to an addition of four methyl ester units. Similarly, the increase in mass of the second peptide, from m/z 2411 to 2481, corresponded to the addition of five methyl ester groups. Similar results were observed for the C17:0-containing fragments and the acylated peptides at Lys-564 (data not shown).
FIG. 4. GC/MS EI FAME mass spectra of the odd-carbon fatty acyl groups cleaved from HlyA. The identities of the C15:0 (A) and C17:0 (B) side chains were confirmed by comparing their fragmentation patterns and GC retention times to those of known standards. We were able to rule out methyl-branched or cyclopropane-containing fatty acids due to retention time differences and (or) known mass spectral fragmentation properties. that activates CyaA) or their respective complexes with acyl-ACPs are different. CyaC-mediated acylation of CyaA involves primarily palmitoylation (8,9,26) or unsaturated C-16 side chains. HlyC-mediated acylation of HlyA in a cell-free system was most efficient when C14:0 acyl-ACP was used as a substrate (2). Similarly, C14:0-modified HlyA produced in vitro was the most active form in the sheep red blood cell assay for hemolytic activity (5). Stanley et al. (2) and observations from recent biophysical studies with model membranes (27) have suggested that C14:0 side chains are not hydrophobic enough in themselves to allow HlyA to penetrate target cell membranes in a nonspecific manner, but they are sufficient to promote surface binding to target cell membranes. These observations may shed some light on the functional significance of the approximately 68% C14:0 in vivo acylation that we report here. However, the observations of C15:0 and C17:0 were unexpected and difficult to rationalize because these fatty acids are present in E. coli at concentrations too small to normally be considered significant. However, the cellular machinery responsible for HlyA activation is known to function with very low concentrations of acyl-ACP substrate (2). Odd-carbon fatty acids are not mentioned in authoritative reviews of chemical composition (28) and endogenous fatty acid metabolism (29) in E. coli, although they are known to occur in the membranes of Gram-negative bacteria generally (30) and in other taxonomic groups. However, C17:0 and C15:0 fatty acyl groups have not, to our knowledge, been reported as protein modifications, much less as substrates for the known internal lysine acylations (2). Therefore, the suggestion that their presence is simply a byproduct of enzymes and (or) carrier proteins with loose specificity, whose primary end products are C14:0-modified lysine at Lys-564 and Lys-690, is probably not tenable. On the other hand, if HlyA were to achieve a better "fit" with C15:0 or C17:0 during a specific host cell protein interaction, e.g. a receptor, or between monomers in the putative oligomerization step some believe may be involved in pore formation, their presence could be rationalized. It will be interesting to examine the modification status of previously described HlyA mutants and HlyA/ LktA hybrid proteins with altered target cell specificities (11,31). Lally et al. (32) have suggested the ␤ 2 integrins as possible receptors for HlyA, but recent attempts to demonstrate ␤ 2 integrin-specific HlyA-mediated cytotoxicity have not been successful. 6 Another possibility is that these side chains may promote interference with host cell signaling pathways by mimicking the action(s) of a lipid-containing host cell messenger. The fact that CyaA is acylated primarily with 16 carbon side chains whereas HlyA is acylated with 14, 15, and 17 carbon groups suggests that the two toxins differ significantly in the physical details of their interactions with target cell membranes.