Phospholipase A activity of adenylate cyclase toxinmediates translocation of its adenylate cyclase domainDavid González-Bullóna,b, Kepa B. Uribea,b, César Martína,b, and Helena Ostolazaa,b,1

aBiofisika Institute (UPV/EHU, CSIC), University of Basque Country, 48080 Bilbao, Spain; and bDepartment of Biochemistry and Molecular Biology, Universityof Basque Country (UPV/EHU), 48080 Bilbao, Spain

Edited by Pascale Cossart, Institut Pasteur, Paris, France, and approved June 30, 2017 (received for review February 6, 2017)

Adenylate cyclase toxin (ACT or CyaA) plays a crucial role inrespiratory tract colonization and virulence of the whooping coughcausative bacterium Bordetella pertussis. Secreted as soluble pro-tein, it targets myeloid cells expressing the CD11b/CD18 integrinand on delivery of its N-terminal adenylate cyclase catalytic domain(AC domain) into the cytosol, generates uncontrolled toxic levels ofcAMP that ablates bactericidal capacities of phagocytes. Our studydeciphers the fundamentals of the heretofore poorly understoodmolecular mechanism by which the ACT enzyme domain directlycrosses the host cell membrane. By combining molecular biology,biochemistry, and biophysics techniques, we discover that ACT hasintrinsic phospholipase A (PLA) activity, and that such activity deter-mines AC translocation. Moreover, we show that elimination of theACT–PLA activity abrogates ACT toxicity in macrophages, particu-larly at toxin concentrations close to biological reality of bacterialinfection. Our data support a molecular mechanism in which in situgeneration of nonlamellar lysophospholipids by ACT–PLA activityinto the cell membrane would form, likely in combination withmembrane-interacting ACT segments, a proteolipidic toroidal porethrough which AC domain transfer could directly take place. Regu-lation of ACT–PLA activity thus emerges as novel target for thera-peutic control of the disease.

bacterial toxins | RTX toxin family | protein translocation | biologicalmembranes | membrane remodeling

The Gram-negative bacterium Bordetella pertussis causeswhooping cough (pertussis), a highly contagious respiratory

disease that is an important cause of childhood morbidity andmortality (1, 2). Despite widespread pertussis immunization inchildhood, there are an estimated 50 million cases and 300,000 deathscaused by pertussis globally each year. Infants who are too young tobe vaccinated, children who are partially vaccinated, and fully vacci-nated persons with waning immunity are especially vulnerable todisease (1).B. pertussis secretes several virulence factors, and among them

adenylate cyclase toxin (ACT or CyaA) is crucial in the early stepsof colonization of the respiratory tract by the bacterium (3). Uponbinding to the CD11b/CD18 integrin expressed on myeloidphagocytes (4), ACT invades these immune cells by delivering intotheir cytosol a calmodulin-activated adenylate cyclase (AC) do-main that catalyzes the uncontrolled conversion of cytosolic ATPinto cAMP (5), which causes inhibition of the oxidative burst andcomplement-mediated opsonophagocytic killing of bacteria (6, 7).ACT also has hemolytic properties conferred by its C-terminaldomain, which may form pores that permeabilize the cell mem-brane for monovalent cations, thus contributing to overall cytox-icity of ACT toward phagocytes (8–10). Bordetella ACT toxin,together with the anthrax edema factor, are the two prototypes ofbacterial protein toxin that attack the immune response by over-boosting the major signaling pathway in the immune response(cAMP, together with the MAPK pathway) (11).ACT is a single polypeptide of 1,706 amino acids, constituted

by an AC enzyme domain (≈residues 1–400) and a C-terminalhemolysin domain (residues ≈ 400–1,706) characteristic of theRTX family of toxins to which ACT belongs (12, 13). Both activities

can function independently as AC and hemolysin, respectively (14,15). The hemolysin domain contains, in turn, a hydrophobic regionwith amphipathic α-helices (∼500–750 residues) likely involved inpore-formation, two conserved Lys residues (Lys-863 and Lys-913)that are posttranslationally fatty acylated by a dedicated acyl-transferase (CyaC), and a C-terminal secretion signal recognized bya specific secretion machinery (CyaB, CyaD, and CyaE) (13, 16).ACT activity on cells requires binding of Ca2+ ions into the nu-merous (∼40) sites located in the C-terminal calcium-binding RTXrepeats that are formed by Gly- and Asp-rich nonapeptides har-boring a conserved X-(L/I/V)-X-G-G-X-G-X-D motif (17, 18). Thisrepeat domain has been reported to be involved in the toxin bindingto its specific cellular receptor, the CD11b/CD18 integrin (4, 19).To generate cAMP into the host cytosol, ACT has to trans-

locate its catalytic domain from the hydrophilic extracellularmedium into the hydrophobic environment of the membrane,and then to the cell cytoplasm. How this is accomplished remainspoorly understood. It is believed that after binding to the CD11b/CD18 receptor, ACT inserts its amphipathic helices into theplasma membrane and then delivers its catalytic domain directlyacross the lipid bilayer into the cytosol (20, 21) without receptor-mediated endocytosis (22). Temperatures above 15 °C (20),calcium at millimolar range, negative membrane potential, andunfolding of the AC domain (23) have been noted to be neces-sary for AC delivery.Several studies have highlighted the importance of various

amino acids and the contribution of distinct domains of the ACTpolypeptide in the translocation process (21, 24). However, a fulldescription at the molecular level of the individual steps that the

Significance

Numerous bacterial toxins can cross cell membranes, pene-trating the cytosol of their target cells, but to do so exploitscellular endocytosis or intracellular sorting machineries. Bor-detella pertussis adenylate cyclase toxin (ACT) delivers its cat-alytic domain directly across the cell membrane by an unknownmechanism, and generates cAMP, which subverts the cell sig-naling. Here, we decipher the fundamentals of the molecularmechanism of ACT transport. We find that AC translocationand, consequently cytotoxicity, are determined by an intrinsicACT–phospholipase A (PLA) activity, supporting a model inwhich in situ generation of nonlamellar lysophospholipids byACT–PLA activity remodels the cell membrane, forming pro-teolipidic toroidal “holes” through which AC domain transfermay directly take place. PLA-based specific protein transport incells is unprecedented.

Author contributions: D.G.-B., K.B.U., C.M., and H.O. designed research; D.G.-B. per-formed research; D.G.-B., K.B.U., C.M., and H.O. analyzed data; and H.O. wrote the paper.

The authors declare no conflict of interest.

This article is a PNAS Direct Submission.

Freely available online through the PNAS open access option.1To whom correspondence should be addressed. Email: elenaamaya.ostolaza@ehu.es.

This article contains supporting information online at www.pnas.org/lookup/suppl/doi:10.1073/pnas.1701783114/-/DCSupplemental.

E6784–E6793 | PNAS | Published online July 31, 2017 www.pnas.org/cgi/doi/10.1073/pnas.1701783114

polypeptide follows to cross the membrane is still missing. De-livery of the AC domain does not appear to proceed through thecation-selective pore formed by the hemolysin domain (25), but itrequires structural integrity of the hydrophobic segments (24), andit seems that the RTX hemolysin domain (residues 374–1,706)harbors all structural information required for AC translocation(26). The region encompassing the helix 454–485 of ACT has beenrecently noted as possibly playing an important role in promotingtranslocation of the AC domain across the plasma membrane oftarget cells (27, 28). A two-step model was proposed for ACtranslocation: in a first step, and upon ACT binding to its CD11b/CD18 integrin receptor, a translocation-competent ACT con-former would conduct extracellular Ca2+ ions across the plasmamembrane, which would activate a calpain-mediated cleavage oftalin, releasing the ACT–integrin complex from the cytoskeleton.In a second step, this ACT–integrin complex would move intolipid rafts, and there the cholesterol-rich lipid environment wouldpromote the translocation of the AC domain across the cellmembrane (29). A caveat of this model is the difficulty to explainhow ACT translocates its AC domain into cells that do not expressthe CD11b/CD18 integrin receptor (22, 30, 31).In recent years our laboratory has extensively explored ACT in-

teraction with lipid membranes, and we have discovered that toxininsertion into lipid bilayers induces transbilayer lipid motion (lipid“scrambling” or “flip-flop”) (32). Transbilayer lipid movement mayoccur as a result of the insertion of foreign molecules (detergents,lipids, or even proteins) in one of the membrane leaflets, or it mayresult from the enzymatic generation of lipids (e.g., lysophospholi-pids, diacylglycerol, or ceramide) at one side of the membrane (33),and it has been involved in enhancing the transmembrane perme-ability of the lipid membranes (33). Here we sought to investigatewhether ACT might possess intrinsic enzyme activity on lipids,which could participate in AC domain translocation. We discoveredthat ACT is a calcium-dependent phospholipase A (PLA) that, byremodeling the cell membrane, mediates translocation of its cata-lytic domain in target macrophages.

ResultsIn Vitro ACT–PLA Activity. To explore potential enzyme activity ofACT on lipids, liposomes (LUV) composed of the commonmembrane lipid phosphatidylcholine (PC) were incubated withACT (at four different lipid:protein molar ratios) for 30 min at37 °C, after which we performed lipid extraction of the samples.The extracted lipid fractions were spotted on silica plates togetherwith appropriate lipid standards, and after being chromato-graphed in one dimension (thin-layer chromatography, TLC) weperformed quantification of the spots by densitometry. The spotswhose intensity was proportional to the lipid:toxin ratio used ineach experiment were detected and found to migrate with anRf value similar to the lipid standard corresponding to free fattyacids, thus suggesting that ACT possess intrinsic PLA activity.Data from the densitometric analysis are depicted in Fig. S1. Acontrol value, which was arbitrarily set as the unity, corresponds tothe data obtained from liposomes treated without toxin. Becauseno spots corresponding to diacylglycerol, phosphatidic acid, or ceramidewere detected on the TLC gels analyzed, potential phospholipaseC (PLC), phospholipase D (PLD), and sphingomyelinase activitieswere in principle discarded.The ACT–PLA activity was further demonstrated by means of an

ultrahigh performance liquid-chromatography mass spectrometry(UHPLC-MS) analysis of the products of an in vitro ACT phos-pholipase reaction (Fig. 1). POPC vesicles (1-palmitoyl-2-oleil-phosphatidylcholine) with distinct fatty acids, palmitic acid (C16:0)and oleic acid (C18:1) on sn-1 and sn-2 positions, respectively, wereused as substrate. Superimposition of the chromatograms obtainedfor the POPC substrate incubated with ACT showed a considerableincrease in the relative abundance of a mass of 255.233 m/z (Fig.1A), which corresponds to the reference standard for free palmitic

acid (C16:0), and of a mass of 281.248 m/z (Fig. 1B), correspondingto free oleic acid (C18:1). Coincident with the appearance of freefatty acids, there was a quantitative reduction (35% reduction) in therelative abundance of the substrate with a mass of 760.586 m/z (Fig.1C) corresponding to intact POPC (C34:1). These data corroboratethat ACT cleaves the sn-1 and sn-2 ester bonds in POPC, and thuspossesses PLA activity.To identify more clearly whether the ACT–PLA activity

cleaves specifically the sn-1 or the sn-2 ester bond of the lipidsubstrate, we used highly sensitive fluorogenic phospholipidsubstrates called PED-A1 [N-((6-(2,4-DNP)Amino)Hexanoyl)-1-(BODIPY FL C5)-2-Hexyl-Sn-Glycero-3-Phosphoethanolamine](which measures PLA1 activity) and PED6 [N-((6-(2,4-Dinitrophenyl)amino)hexanoyl)-2-(4,4-Difluoro-5,7-Dimethyl-4-Bora-3a,4a-Diaza-s-Indacene-3-Pentanoyl)-1-Hexadecanoyl-sn-Glycero-3-Phosphoethanolamine] (which measures PLA2 activity)(34). Both substrates are modified glycerophosphoethanolamineswith BODIPY FL dye-labeled sn-1 or sn-2 acyl chains, respectively,and a dinitrophenyl group conjugated to the polar head group ofthe lipid to provide intramolecular quenching. Thus, sn-1 or sn-2acyl chain cleavage by PLA1 or PLA2 activity eliminates the dyequenching, which results in increased fluorescence of the fluo-rophore. As shown in Fig. 2, incubation of ACT (10 nM) withDOPC liposomes containing PED-A1 substrate (20% molar ratio)induced an increase of the fluorescence measured at 515 nm, in-dicating the release of free fatty acid from the sn-1 ester bond of thefluorescent substrate (Fig. 2A). Similarly, addition of ACT (10 nM)to DOPC liposomes containing PED6 (20%molar ratio) induced aswell fluorescence increase at 515 nm (Fig. 2A), indicating that thesn-2 ester bond was also being hydrolyzed by ACT and free fattyacid released at the same time. Comparison of EC50 valuesobtained from the fluorescence data for both fluorescent substratesat identical conditions (EC50 for PED-A1 = 2.42 nM; EC50 forPED6 = 12.85 nM) indicated, however, that ACT cleaves withnotably more efficiency the sn-1 ester bond of the fluorescentsubstrate than the sn-2 ester bond. For both fluorogenic substratesthe fluorescence increase was proportional to the toxin concentra-tion added (Fig. 2B), and interestingly, the data indicated that thetoxin PLA activity achieved saturation long before all of the sub-strate present was consumed (maximal relative fluorescence valuesfor the hydrolysis of PED-A1 and PED6 substrates by the respectivepositive controls, PLA1 and PLA2, at a concentration of 0.25 μg/mLupon incubation for 30 min at 37 °C, were 3.066 ± 0.046 and2.896 ± 0.078, respectively). In conclusion, this set of experimentsdemonstrates that ACT displays PLA activity, with preference forthe sn-1 ester bond, and hence, most likely that ACT acts as aphospholipase A1 (PLA1) on natural membranes. Importantly, italso indicates that a “product inhibition” mechanism might operatein ACT–PLA activity. Several proteins with PLA activities may alsodisplay lysoPLA activity (35); in contrast, ACT did not detectablyhydrolyze either LysoPC (C18:1) or DLPC:LysoPC (C18:1) lipidsubstrates, suggesting that it is not a lysophospholipase A (Fig. S2).To have an idea of the substrate specificity of the ACT–PLA

activity on several phospholipids, both neutral and anionicphospholipids were tested (Fig. S3). In this case, as a measure ofthe ACT–PLA activity, spots corresponding to free fatty acidswere resolved by TLC and quantified by densitometry after in-cubation of ACT with liposomes made of different lipid sub-strates (DOPC, DOPE, DMPG) at a lipid:protein ratio of1,000:1 for 30 min at 37 °C. Fig. S3 shows that ACT may hy-drolyze, in vitro, both neutral and anionic phospholipids withsimilar efficiency, indicating that ACT–PLA may not have a clearspecificity for any given phospholipid.

ACT–PLA Harbors Conserved Ser/Asp Catalytic Sites, Sharing Similaritywith Patatin-Like Phospholipases. To characterize the ACT–PLAactivity in greater detail, analysis of the ACT amino acid sequencewas carried out with the multiple sequence alignment program

González-Bullón et al. PNAS | Published online July 31, 2017 | E6785

BIOCH

EMISTR

YPN

ASPL

US

ClustalOmega (https://www.ebi.ac.uk), finding similarities withseveral patatin-like phospholipases, such as the Pseudomonasaeruginosa ExoU PLA2 (26.65% identity) or the Legionella pneu-mophila VipD PLA1 (21.60% identity) (Fig. S4). Patatin-likePLAs are most highly homologous to eukaryotic group IV cyto-solic PLA2α and group VI calcium-independent PLA2, which aredistinguished by a conserved Ser-Asp catalytic dyad within theactive site, with catalytic Ser located in a conserved G-X-S-X-Ghydrolase motif, and Asp in a D-X-G/A motif (boldface letterscorrespond to conserved catalytic residues involved in PLA ac-tivity) (36, 37). Remarkably, in its 1,706 amino acids long se-quence, rich in Gly and Asp residues, ACT harbors a single G-X-S-X-G motif, with a GASAG sequence, which is localized in thepredicted membrane-interacting α-helical region (residues 500–700) of ACT. On other side, the DXG/A motif that fits best in thealignment analysis against the sequences of several patatin-likephospholipases has a DGG sequence, and it is localized in thefirst one of the five Ca2+-binding β-rolls (block I, residues 1,014–1,087) that form the so-called C-terminal repeat domain. Addi-tionally, along its C-terminal sequence ACT contains, in addition tothe D1079GG motif, three more DXG motifs, but their fitting in thesequence alignment assay was worse.To ascertain whether ACT–PLA activity relies on the men-

tioned Ser and Asp residues, and thus to experimentally validateour prediction, we initially tested residue-specific agents, such asMAFP (methyl arachidonyl fluorophosphonate), for their abilityto inhibit ACT–PLA activity. MAFP is the MAFP analog ofarachidonic acid (AA), which is known to bind irreversibly to Serresidues and inhibit LysoPLA type I and PLA2 family membersthat exhibit LysoPLA activities (groups IV and VI) with an IC50of 0.5–5 μM and other hydrolases, such as fatty acid amide hy-drolase, with an IC50 of 2.5–20 nM (38, 39). MAFP is known to

also inhibit enzymes with PLA1 activity (38). At a concentrationof 10 nM, ACT was incubated with PED-A1–containing lipo-somes (lipid:protein molar ratio 1,000:1) in the presence of in-creasing amounts of MAFP (1–100 nM). Release of dye-labeledsn-1 fatty acid from DOPC–PED-A1 liposomes by ACT–PLAactivity was drastically halted by MAFP in a dose-dependentmanner (Fig. 3), which indicates that a catalytic Ser residue isinvolved in the ACT lipolytic activity.To further verify the involvement of the predicted catalytic

residues in ACT–PLA activity, we constructed two ACT mutants,one in which Ser-606 was substituted by Ala (S606A), and a sec-ond mutant in which Asp-1079 was substituted by Ala (D1079A).A secondary structure of WT ACT and of the two mutant proteinsat 1-μM concentration (buffer with 10 mM CaCl2) was checked byfar-UV circular dichroism (CD), and it was verified that thestructure of the three proteins was very similar (Figs. S5 and S6).PLA1 and PLA2 activities of the purified mutant proteins werechecked using the substrates PED-A1 and PED6. As shown in Fig.4, substitution of the Ser-606 by Ala had a dramatic effect on thetoxin in vitro both PLA activities, so that the mutant proteinS606A was almost inactive in cleaving the fluorogenic substrates,relative to the intact ACT toxin. Substitution of Ser-606 by Ala didnot affect any of the other ACT activities (production of cAMP insolution, hemolytic activity) in the mutant protein ACT-S606A(Table S1). Replacement of Asp-1079 by Ala showed a less del-eterious—although still substantial—decreasing effect on the mutantPLA1 activity, but its effect on the PLA2 activity was not significant(Fig. 4). Interestingly, this substitution also decreased the hemolyticpotency of the ACT-D1079A protein relative to WT ACT, whereasit did not alter at all of the cAMP-producing AC activity of themutant in solution (Table S1). Taken together, these results provedthat Ser-606 and Asp-1079 are directly involved in the ACT–PLA1

Fig. 1. Identification of the reaction products of ACT-PLA activity by UHPLC-MSE. Identification of the reaction products of an in vitro PLA reaction on POPClipid substrate were performed using UHPLC and simultaneous acquisition of exact mass at high and low collision energy, MSE. In the figure, representativeextracted ion UHPLC-ESI-MSE chromatograms are shown: electrospray ionization (ESI)−, m/z 255.233 of ion [Palmitic acid-H]− (A), m/z 281.248 of ion [Oleicacid-H]− (B), and ESI+, m/z 760.586 of ion [PC(16:0/18:1)+H]+ (C) obtained from samples containing POPC lipid substrate (green) or substrate exposed to ACT(red). On top of the figure is a schematic representation of PLA1 and PLA2 cleavage sites on phospholipids is shown.

E6786 | www.pnas.org/cgi/doi/10.1073/pnas.1701783114 González-Bullón et al.

catalytic reaction, corroborating our predictive data and suggestingthat the catalytic mechanism of ACT–PLA might proceed through aserine-acyl intermediate, in the likeness of the patatin-like PLAs.

In Vivo ACT–PLA Activity. To investigate whether ACT exerts lipidacyl hydrolase activity in vivo, enzyme assays were next con-ducted in J774A.1 macrophages exposed to the toxin. To this end,we first labeled the cell membranes with tritiated arachidonic acidby an overnight incubation of the phagocytes (1 × 105 cells) inmedium with 3H-arachidonic acid (1.5 μCi/106 cells) at 37 °C.

Upon extensive cell washing, cells were then incubated with ACT(20 nM) for 10 min at 37 °C. After that incubation time, a solutionof chloroform:methanol (1:1) was added to extract lipids, and thenseparation of free fatty acids from phospholipids (40) was carriedout by TLC. Radioactivity in the sample extracted was quantifiedby scintillation counting. About 2- to 2.5-fold more 3H-arachidonicacid was detected in the samples exposed to ACT compared withcontrol untreated cells (Fig. 5A), indicating that ACT exhibitsPLA activity in vivo as well. As expected, the release of tritiatedfree AA from the J774A.1 cell membrane was notably reducedupon overnight preincubation of ACT with MAFP (Fig. 5A), thuscorroborating the ACT–PLA activity features observed previouslyin the in vitro assays.Also as expected, the release of tritiated fatty acid was signifi-

cantly reduced in cells exposed to the inactive S606A mutant (Fig.5A). However, this inhibition by MAFP or mutation (S606A)observed in vivo was apparently less efficient than the observed inthe in vitro assays (Figs. 3 and 4). We hypothesized that becauseeukaryotic cells contain endogenous PLAs, some of which areCa2+-dependent (41), it might be that part of the tritiated AAdetermined in our in vivo assays could come from the activity ofsuch cellular PLAs. To settle the issue, we performed a controlassay in which cells were preincubated with known inhibitors ofboth cytosolic and secretory cellular PLAs, MAFP and LY311727,respectively, prior to their treatment with ACT, ACT+MAFP, orACT-S606A. Data from this experiment, depicted in Fig. 5B,confirmed our hypothesis and explained the apparent lower effi-cacy of inhibition. Fig. 5B shows, first, that as consequence of theinhibition of the cellular PLA2, the total amount of free arachi-donic acid quantified was inferior (≈1.7 times the control vs.2.3 times the control), and second, that the inhibitory effect ofMAFP or the point mutation ACT-S606A was now almost com-plete. In cells incubated with a lower ACT concentration (1.5 nM),the pattern observed was very similar, except that in this case thecontribution of the intracellular phospholipases was inferior (Fig.S7). We can thus state that ACT exhibits PLA activity in vivo,releasing free fatty acids into the membrane. A striking findinghere is that treatment of monocytes with ACT promotes acti-vation of cellular PLAs. We speculate that this activation mightFig. 2. ACT–PLA reaction determined by cleavage of PLA1- and PLA2-specific

fluorogenic substrates. Two different specific fluorogenic phospholipid sub-strates, PED-A1 and PED6, were used for measurement of PLA1 and PLA2 acti-vities, respectively. These substrates incorporate a BODIPY FL-dye–labeled sn-1 ora BODIPY FL-dye–labeled sn-2 acyl chain, and a dinitrophenyl quencher group.Cleavage of the corresponding labeled-acyl chain by PLA1 or PLA2 activitiesresults in the formation of nonfluorescent lysophospholipid and fluorescent fattyacid tagged to the fluorophore-BODIPY FL. The release of the correspondingfatty acid is determined as an increase of the fluorescence intensity at 515 nm(a.u.). The respective substrate hydrolysis was determined 30 min after in-cubation of the substrate with the toxin (10 nM) at 37 °C. PLA1 and PLA2 ac-tivities are expressed as an increase of the fluorescence intensity at 515 nm (a.u.)relative to the corresponding control (lipid substrate in buffer, without toxin).The phospholipid substrates in the assay were liposomes (LUV) prepared withmixtures of DOPC:PED-A1 (4:1 mol:mol) or DOPC:PED6 (4:1 mol:mol) as describedin SI Materials andMethods (A). PLA1 and PLA2 activities expressed as increase ofthe fluorescence intensity at 515 nm (a.u.) as a function of ACT toxin concen-tration (0.5–40 nM) determined 30 min after incubation with the toxin (B). Dataare expressed as average values ± SD from at least three independent experi-ments (n = 3).

Fig. 3. Dose-dependent inhibition of the ACT–PLA activity by MAFP, aresidue specific irreversible PLA inhibitor. Release of dye-labeled sn-1 fattyacid from DOPC:PED-A1 (4:1 mol:mol) liposomes by ACT–PLA1 activity wasdetermined by fluorescence spectroscopy, in the presence of different con-centrations of the PLA inhibitor MAFP, in the range 1–100 nM. PLA1 activityof ACT (10 nM) on DOPC:PED-A1 (4:1 mol:mol) liposomes after 30-min in-cubation, normalized with respect to the control (substrate in buffer,without toxin) was taken as 100% activity. Average values ± the SEM (n = 3)are shown. ***P < 0.001 with respect to intact ACT.

González-Bullón et al. PNAS | Published online July 31, 2017 | E6787

BIOCH

EMISTR

YPN

ASPL

US

be mediated through ACT capacity to induce Ca2+ elevations intotarget cells, which is greater the higher the toxin concentration is(42). Elucidation of the potential biological consequence of suchactivation in target cells needs further investigation.

ACT–PLA Activity Is Involved in AC Domain Translocation.To understandthe biological consequence of the novel ACT phospholipase activ-ity, we hypothesized that it might modulate cellular processes re-lated to membrane remodeling. Lysophospholipids, one of the endproducts of the PLA activity, are nonlamellar lipids with intrinsicpositive curvature (43), and formation of nonlamellar structures inthe membrane is thought to be one way for “lipid pores” to occur init (44). Moreover, insertion of amphipathic regions of many cellularproteins is now recognized as possibly resulting in the formation ofhybrid proteolipidic “toroidal pores” by contributing to the fusion ofinner and outer bilayer leaflets, in a way that is analogous to theeffect on a pure lipid system of an external charge or the presenceof detergents (45–47).With this basis, we explored the possibility that ACT–PLA

activity could contribute in vivo to AC domain translocation byinducing membrane-remodeling effects through the generationof lysophospholipids into the cell membrane. As a measure ofthe AC domain translocation efficiency, we quantified the in-tracellular cAMP accumulation induced by intact ACT (5 nM),by ACT preincubated with MAFP, and by the mutant proteinsS606A (5 nM) and D1079A (5 nM) in J774A.1 macrophages. As

shown in Fig. 6A, preincubation of ACT with MAFP abrogatedalmost completely the accumulation of cAMP into the macro-phage cytosol. Possible side effects of the drug on ACT ACenzyme activity were ruled out, because cAMP production insolution by the MAFP-treated ACT was almost identical to in-tact ACT (Fig. 6B). Similarly, cAMP accumulation was impor-tantly halted in macrophages incubated with the S606A andD1079A mutants, whereas the point substitutions of the catalyticSer and Asp did not have any effect on the capacity of the re-spective ACT mutants to generate cAMP in solution (Fig. 6 Aand B). Therefore, these results proved that ACT is not capableof translocating efficiently its AC domain into the host cytosol ifthe toxin intrinsic PLA activity results abrogated, both chemi-cally (MAFP) or by mutagenesis. Thus, this set of experimentsdemonstrated a direct causal relationship between ACT–PLAactivity and AC domain translocation.One uncontestable feature of the AC domain translocation

process is the requirement of calcium in the millimolar range. To

Fig. 4. ACT–PLA has Gly-X-Ser-X-Gly hydrolase motif. PLA1 (A) and PLA2

(B) activities of intact ACT, and of the ACT mutants S606A and D1079A, as afunction of the corresponding protein concentration (0–40 nM), were determinedusing the method of the PLA1- and PLA2-specific fluorogenic substrates PED-A1 and PED6. The S606A mutant has a point mutation in the amino acid Ser-606 of the conserved Gly-X-Ser-X-Gly hydrolase motif, which was substituted byan Ala, and the D1079Amutant has a mutation in the Asp-1079 of the Asp-X-Gly/Ala motif. Cleavage of the sn-1 and sn-2 ester bonds was followed as an increaseof the fluorescence intensity at 515 nm determined 30 min after incubation withthe corresponding protein toxin. Data are expressed asmeans± SD (n= 3) from atleast three independent experiments.

Fig. 5. In vivo ACT–PLA activity. PLA activity of ACT was determined inmacrophages by quantification of free [3H]-arachidonic acid released fromJ774A.1 cells upon treatment with ACT for 10 min, as described in SI Mate-rials and Methods. (A) PLA activity of intact ACT (20 nM), of ACT (20 nM)preincubated with the PLA inhibitor MAFP (25 μM), and of the ACT mutantS606A (20 nM) was determined by the same procedure and graphicallydepicted as the ratio of scintillation counting relative to control untreatedcells. (B) Cells were first preincubated with a combination of inhibitors ofcytosolic and secreted PLA (MAFP + LY311727; each inhibitor at 25 μM) andthen cells were treated with ACT, or ACT preincubated with MAFP (25 μM),or S606A proteins (20 nM), and upon 10-min incubation free [3H]-arachidonicacid released was quantified. Average values ± the SEM (n = 3) are shown.***P < 0.001 with respect to intact ACT.

E6788 | www.pnas.org/cgi/doi/10.1073/pnas.1701783114 González-Bullón et al.

determine whether the ACT–PLA activity was calcium-sensitive,the hydrolysis of the fluorogenic substrates was assayed in bufferswith different CaCl2 concentrations, We found that the acylhydrolase activity of ACT (PLA1 and PLA2) required Ca2+ in themillimolar range (Fig. 6C), thus indicating that ACT is a calcium-dependent PLA. This was a key finding, because this dependenceof the ACT–PLA activity for calcium may rationally explain atthe molecular level the strict requirement shown by the ACtranslocation process for this divalent cation.Finally, we wanted to corroborate whether the mechanism by

which the ACT–PLA activity dictates AC translocation is based onthe membrane remodeling capacity of the lipid hydrolase action.With this aim, we determined the capacity of ACT to inducetransbilayer lipid motion in DOPC vesicles, using for that a flip-flopassay previously adapted for this purpose in our laboratory (32). Thisassay is based on the phenomenon of FRET (32). In the presentcase, the donor and acceptor are, respectively, 7-nitrobenz-2-oxa-1,3-diazol-4-yl (NBD) and rhodamine. NBD is bound to a phospholipidinitially located only in the inner monolayer of the membrane, andrhodamine is bound to a large protein (antibody) located outside thevesicle, so that at time 0 no energy transfer can occur. However, iftransbilayer lipid motion occurs, some of the NBD lipid is trans-ferred to the outer monolayer and can interact with the external,protein-bound rhodamine, thus decreasing the fluorescence signal at545 nm, the maximum of NBD emission. As expected, addition ofACT (10 nM) to DOPC liposomes induced a rapid reduction of thefluorescence signal at 545 nm, in a time scale of minutes (Fig. 7A),consistent with ACT capacity to promote lipid flip-flop, as previouslynoted by our group (32). In contrast, no changes in the fluorescencesignal were detected upon addition of S606A or D1079A mutantproteins to liposomes (Fig. 7A), reflecting that the membranerestructuring capacity of ACT is directly linked to its intrinsic PLAactivity. We provide a control of binding (Fig. S8) for the threeproteins used in the flip-flop assay, from which we ruled out thatmutations on Ser-606 or Asp-1079 had gross effects on the ability ofthose proteins to bind to the liposome membranes.The concentration-dependent inhibitory effect of MAFP on the

transbilayer lipid motion induced by ACT (Fig. 7B) was a furthercorroboration of the relation between ACT–PLA activity andmembrane remodeling ability of ACT. This direct interrelationbetween both processes was furtherly proved into target cells, inwhich we could detect, through the widely used method of annexinV staining, that a few minutes (0–10 min) after the treatment ofJ774A.1 monocytes with ACT (5 nM), phosphatydylserine (PS; aphospholipid normally located in the cytoplasmic monolayer), wassignificantly located in the outer monolayer of the plasma mem-brane (Fig. 7C), and was reaffirmed by the experiments using thePLA inactive mutants ACT-S606A and ACT-D1079A, which werealmost unable to promote such transmembrane lipid movement inthe treated macrophages (Fig. 7C). Taken together, these resultsconstitute a very solid indication that AC translocation relies onthe generation of nonlamellar intermediates in the cell membraneby the ACT–PLA activity.

ACT–PLA Activity Is Instrumental to Induce Cell Cytotoxicity. Toxicaction of ACT on phagocytes results primarily from the cytotoxicsignaling of toxin-produced cAMP (12). Because production ofcAMP into the host cell cytosol requires AC domain trans-location, we hypothesized that ablation of the ACT–PLA activitywould redound in a lower ACT cytotoxicity on macrophages.Therefore, we examined the effect of eliminating the ACT–PLAactivity on the ACT-induced cytotoxicity, measured as lactatedehydrogenase (LDH) release. If PLA activity is necessary for AC

Fig. 6. ACT–PLA activity is involved in AC domain translocation. The cata-lytic activities of intact ACT, ACT preincubated with MAFP, and of the ACTmutants S606A and D1079A, were measured in J774A.1 cells (A) or in solu-tion (B) at a toxin concentrations of 5 nM and 1 nM, respectively, andexpressed as picomole cAMP generated per microgram of the correspondingprotein. The control sample corresponds to the cAMP amount measured inuntreated cells. The cyclase activity was determined as described in SI Ma-terials and Methods. Calcium-dependence of ACT–PLA1 activity was mea-sured for the three proteins, ACT, ACT-S606A, and ACT-D1079A, by themethod of the fluorogenic substrate PED-A1, as a function of the calciumconcentration, and expressed as relative fluorescence values (515 nm) in a.u.

(C). Average values ± the SEM (n = 3) are shown. ***P < 0.001 with respectto intact ACT.

González-Bullón et al. PNAS | Published online July 31, 2017 | E6789

BIOCH

EMISTR

YPN

ASPL

US

translocation, treatment of J774A.1 cells with the PLA-deficientS606A or D1079A ACT mutants would be expected to be lessdeleterious to cells. As shown in Fig. 8, intact ACT induced aprogressive, concentration-dependent increase in cell cytotoxicity,inducing the release of LDH in about 70% of the cells after 2 h ata toxin concentration of 2 nM. In contrast, the PLA-deficientmutant ACT-S606A did not result in cytotoxicity at this samerange of toxin concentrations (0–2 nM) (Fig. 8). Only at toxinconcentrations above 2–4 nM did the cytoxicity induced by thePLA-deficient mutant ACT-S606A become substantial, very likelybecause of its intact pore-forming activity, which has been repor-ted that at this higher toxin concentrations synergizes with cAMPgeneration and ATP depletion in inducing cell death (9). Thecytotoxicity induced by the ACT-D1079A mutant was almostnegligable at both toxin concentration ranges (0–2 nM and ≥2 nM),very likely because besides the PLA deficiency, this point mutationdecreases the toxin hemolytic potency as well (Table S1), whichcontributes more importantly to general toxicity at the higher toxinconcentrations (9). According to a recent study performed in ababoon model of infection (48), at the bacterium–target cell in-terface during infection of the respiratory tract, the concentrationof ACT can be around 100 ng/mL (≈0.5–1 nM); thus, our resultsindicated that PLA activity is instrumental for ACT to be cytotoxic,particularly at toxin concentrations close to biological reality ofbacterial infection.

DiscussionIn this study, we decipher a fundamental event underlying thehitherto elusive and unique mechanism of translocation of theB. pertussis ACT.We discover that ACT bears intrinsic PLA activity, both in

vitro and in vivo. Our in vitro results obtained by means of twodifferent biophysical techniques, highly sensitive fluorescent lipidsubstrates and MS, concur and clearly demonstrate that ACThydrolyzes both the sn-1 and the sn-2 ester bonds of glycer-ophospholipid substrates, thus releasing into the membrane freefatty acids and lysophospholipids (Figs. 1 and 2). The compari-son of the efficiencies of the respective catalytic activities clearlyindicates, however, that ACT cleaves with notably more effi-ciency the sn-1 ester bond of glycerophospholipid substrates,suggesting that ACT most likely acts as a PLA1 on naturalmembranes. Although many of the known PLA1 also displayLysoPLA activities, we could not detect lysophospholipase ac-tivity in ACT (Fig. S2). Interestingly, our data also suggest that a“product inhibition” or “feedback-like” mechanism might oper-ate in the ACT–PLA activity (Fig. 2). This may be a relevantcharacteristic for the ACT–PLA mechanism, as we explain later,as it suggests that the ACT–PLA activity is not to massively“destroy” the target cell outer membrane lipids, but quite theopposite: that this ACT lipid hydrolytic activity has likely evolvedas a “security mechanism” by inducing a local, subtle and “sur-gical” effect on the cell membrane of the target cells. The in vivodata obtained upon treatment of target macrophages with ACT(Fig. 5) are fully consistent with the in vitro results, and provethat ACT releases free fatty acids from the macrophage cellmembrane in the same toxin concentration range in which itbecomes cytotoxic for cells. Interestingly, we note here that ACT

Fig. 7. ACT induced membrane remodeling (lipid flip-flop) in liposomes andin macrophages. Time course of changes in fluorescence intensity of NBDrepresented as the ratio of the fluorescence intensities of the probe at 545/575 nm, in control liposomes or upon incubation of LUVs with WT ACT orwith the ACT–PLA inactive mutants S606A and D1079A. Lipid concentrationwas 0.1 mM. ACT was added at 100 nM and liposomes were composed ofDOPC and NBD-PE (DOPC:NBD-PE, 1,000:6 molar ratio) (A). Effect of two

different concentrations (50 nM and 1 μM) of the irreversible inhibitor MAFPon the ACT flip-flop activity is shown. Lipid and protein concentrations wereas above (B). Externalization of the lipid PS to the outer monolayer of thecell membrane of macrophages, determined by flow cytometry as annexin Vstaining, by ACT, ACT-S606A, and ACT-D1079A. Macrophages were in-cubated at 37 °C with the respective toxin at a concentration of 5 nM, for10 min. The control represents cells that were treated with medium withouttoxin (C). Average values ± the SEM (n = 3) are shown.

E6790 | www.pnas.org/cgi/doi/10.1073/pnas.1701783114 González-Bullón et al.

binding to target cells activates endogenous cellular Ca2+-sensi-tive PLAs (Fig. 5), most likely through its capacity to induceCa2+ influx. At present, the biological consequences of suchactivation are unknown.From the analysis of the toxin sequence, we conclude that

ACT shares sequence similarity with members of the patatin-likephospholipases and mammalian cPLA2 (Fig. S4), suggesting thatACT may belong to the esterase/lipase hydrolase superfamily(37). Consistent with this, we identify in the ACT toxin sequencetwo conserved active-site catalytic motifs, namely, the GAS606AGsequence (consensus motif GXSXG) and D1079GG sequence(consensus motif DXG/A), and by means of pharmacological in-hibition and site-directed mutagenesis, we provide evidence of thedirect involvement of both Ser-606 and Asp-1079 as essential cat-alytic residues in the ACT–PLA1 activity (Figs. 3–5), and thus wevalidate that ACT is a serine acylhydrolase.A first characterization of the substrate specificity of the

ACT–PLA activity shows that ACT may hydrolyze in vitro bothneutral and anionic phospholipids (Fig. S3) with similar effi-ciency, suggesting that ACT–PLA may not have a clear substratespecificity for any given phospholipid in the likeness of thepatatin PLA enzyme (49). Recently, it has been shown that theso-called ABH effector domain of the Vibrio cholerae MARTXtoxin, another member of the RTX family of proteins to whichACT toxin belongs, possess PLA1 activity (50). Relative to theACT–PLA activity, ABH appears to be highly specific for itsphospholipid substrates, cleaving preferentially phosphatidyli-nositols, in particular PI3P (50). Compared with ACT however,ABH takes several hours to hydrolyze PI3P (50), which mightperhaps be because of an in vivo ABH requirement of some yetunknown intracellular cofactor. As shown here however, ACT–PLA does not require any extracellular or intracellular cofactorother than calcium to rapidly hydrolyze common membranephospholipids; and given that ACT acts on the cell membraneouter monolayer where PC—in particular POPC and DOPC—are the most abundant glycerophospholipids, we hypothesizethat PC could be an in vivo likely ACT–PLA lipid substrate.

Of great significance in the ACT context is the elucidation hereof the biological consequence of this novel ACT–PLA activity,that ACT lipid hydrolytic activity is directly linked to the ACdomain transport across the plasma membrane of target phago-cytes. Two types of experiments allow us to conclude this. First,the sole inhibition of the ACT–PLA activity by MAFP preventstotally the accumulation of cAMP into the cytosol of the targetcells, whereas it does not affect at all of the ACT production ofcAMP in solution, thus proving that MAFP does not interfere withthis second enzymatic activity of ACT (Fig. 6). Second, eradicationof the ACT–PLA1 activity by introduction of a point substitutionin the consensus catalytic center (S606, D1079) equally halts theaccumulation of cAMP into the target cytosol (Fig. 6), whereas themutated proteins remain fully functional in generation of cAMP insolution (Table S1). Hence, all of the results concur that ACtransport across the membrane relies and is directly coupled to thehydrolysis of membrane phospholipids by ACT itself. Consistentwith this essential role of the ACT–PLA activity in AC trans-location, we show that toxin-induced cell toxicity dramaticallydecreases in the macrophages treated with PLA-deficient ACTmutant proteins (Fig. 8), particularly at the toxin concentration-range close to biological reality of bacterial infection, indicatingthat PLA activity is instrumental for ACT cytotoxicity.A key question here is how the ACT–PLA activity may mecha-

nistically allow the transference of the AC domain across theplasma membrane. The reaction products of ACT–PLA activity onplasma membrane phospholipids are lysophospholipids and freefatty acids. For decades it has been widely recognized that themodification of the cell-membrane composition through the directaction of hydrolytic enzymes, such as phospholipases on membranelipids, may regulate cell-membrane physical properties, such as themembrane mechanical stretching and bending constants (51),membrane curvature, or membrane lateral organization, favoringthe formation of transient nonlamellar structures (52–56). Lyso-phospholipids, nonlamellar lipids with positive spontaneous curva-ture, have been proven to induce formation of hydrophilic permeableproteolipidic “toroidal pores” in membranes in which amphipathicpeptides and proteins are inserted, via a decrease in line tensionwithin the lipid bilayer because of relief of curvature stress along themembrane edge of the pore wall (44, 57). By definition, in a toroidalpore its constituent monolayers become continuous via the pore-lining lipids (44, 57), and thus this structure allows the movementof lipid molecules from one monolayer of the bilayer to the other.Here we provide evidence that ACT–PLA activity is directly involvedin promoting transbilayer lipid motion both in pure phosphatidyl-choline liposomes and, moreover, also in macrophages, and provethat this is the direct inhibition of such ACT-induced lipid scrambling(flip-flop) by MAFP and point mutations in the PLA catalytic sites(Fig. 7), thus indicating that ACT–PLA activity could form in thelipid bilayer toroidal pores. Therefore, we provide herein an expla-nation to the question of how ACT–PLA activity may dictate ACdomain transport through the plasma membrane. We postulate thatin situ generation of nonlamellar lysophospholipids by ACT–PLAactivity on plasma membrane phospholipids would form a hydro-philic lipid pore of toroidal architecture through which AC domaintransfer could directly take place. The location of the catalytic Ser-606 within one of the predicted α-helices (helix IV) of the so-calledhydrophobic pore-forming domain of ACT, leads us to plausiblyhypothesize, that such a hydrophilic lipidic pore might be of pro-teolipidic nature, constituted likely by both lipids and one or morehelical segments of the pore-forming region. Consistent with thisidea, others have previously noted that mutations in specific aminoacids of several α-helices from the hydrophobic domain impair notonly the hemolytic activity, but also the translocation capacity ofACT (21).Our translocation model is also consistent with the extraor-

dinary versatility of the AC domain to transport into the cytosolof CD11b+ target cells large heterologous cargo polypeptides of

Fig. 8. Cell toxicity induced by low ACT concentrations requires toxin in-trinsic PLA activity. Cytotoxicity induced by WT ACT or by the ACT mutantsS606A and D1079A, in J774A.1 phagocytes after 2-h incubation at 37 °C, as afunction of toxin concentration is depicted. Cell cytotoxicity was determinedthrough release of LDH into the culture medium, and is expressed as percentLDH released into the medium relative to the total LDH content, after 2-hincubation at 37 °C. Average values ± the SEM (n = 3) are shown.

González-Bullón et al. PNAS | Published online July 31, 2017 | E6791

BIOCH

EMISTR

YPN

ASPL

US

up to 200 residues in length within the AC domain (23, 26), asamong the unique characteristics of the proteolipidic toroidalpores are structural flexibility, low selectivity, and variable size(44, 57). Direct participation in AC domain translocation of anACT–PLA-generated hydrophilic toroidal tunnel, in which thepore-walls may in part be lined by phospholipid head groups,may also be consistent with previous observations showing a netpositive charge requirement for AC domain to be translocated(58). Moreover, the possible formation of ion couples amongthose positively charged amino acid lateral chains with negativelycharged phospholipid and fatty acid headgroups exposed at thepore walls may represent an energy source that might contributeto lowering the energy barrier for the AC domain insertionwithin the lipid bilayer (59, 60).Remarkably, our results may substantiate—at the molecular

level—previous data on calcium dependence of the ACT cata-lytic domain translocation in eukaryotic cells (20), and this is acritical aspect that previous models on AC transport have failedto explain convincingly (20, 29). We show indeed that to be fullyoperative, the ACT–PLA activity requires calcium in the milli-molar range, thus providing a straightforward explanation to thestrict calcium dependence of the AC transport across the cellmembrane (20). It is tempting to speculate that the requirementof a cofactor, such as calcium ions for the ACT–PLA activity,might be linked to the absence of toxicity of ACT whenexpressed in bacteria, acting somehow as a security mechanismfor the prokaryotic cell.A model cited in the ACT field to explain AC translocation is the

two-step model proposed by Bumba et al. (29), in which in a firststep, and upon ACT binding to its CD11b/CD18 integrin receptor, atranslocation-competent ACT conformer would conduct extracel-lular Ca2+ ions across the plasma membrane, which it would acti-vate a calpain-mediated cleavage of talin, releasing the ACT–integrin complex from the cytoskeleton. In a second step, this ACT–integrin complex would move into lipid rafts, and there thecholesterol-rich lipid environment would promote the translocationof the AC domain across the cell membrane (29). However, thismodel cannot explain convincingly the uncontestable dependenceshown by the AC domain transport for millimolar calcium con-

centrations, because at very low ACT concentrations (<100 ng/mL)at which AC translocation is supposed to be physiologically morerelevant for cytotoxicity, the intracellular calcium elevation inducedby ACT is nearly null. Another feature also difficult to explain byBumba et al.’s (29) model is how ACT can successfully transport,upon fusion to the C-terminal ACT-hemolysin domain, such a va-riety of N-terminal polypeptides carrying such different sequencesand lengths. A path that has to assist indistinctly the translocation ofa 400 amino acids-long chain or of other different polypeptides hasto be expectedly dynamic, adaptable, and versatile, and these aretypically features shared by lipid/protolipid toroidal pores, such asthe ones that may be generated by a local PLA activity, as the ACT–PLA activity revealed here.In conclusion, this study reveals intrinsic PLA activity of ACT

that explains a fundamental molecular event underlying ACTprotein translocation. The coupling in a single molecule ofmembrane remodeling capacity—given by an enzymatic activity—with the transport of proteins across membranes is unprecedented.Because delivery of the AC domain is essential for ACT-inducedcell toxicity, regulation of the ACT–PLA activity emerges as anovel drug target for the therapeutic control of whooping cough, aninfectious disease that is the fifth largest cause of vaccine-preventable death in infants.

Materials and MethodsExpanded methods are available in SI Materials and Methods.

In vitro PLA activity was assayed using TLC and ulterior densitometricanalysis, UHPLC-MS analysis, and selective fluorogenic phospholipid substrates:namely, PED-A1 and PED6. For the in vivo PLA activity determination, cellularlipids were labeled with [3H]-arachidonic acid. Lipid flip-flop was assessed byFRET using NBD-PE and rhodamine as donor and acceptor, respectively.

ACKNOWLEDGMENTS. We thank the technical and staff support providedby the Servicio de Lipidómica of the SGIker (Universidad del País Vasco/Euskal Herriko Unibertsitatea, Ministeria de Ciencia e Innovación, GV/EJ,European Social Fund) and Rocío Alonso for excellent technical assistance.This study was supported by Basque Government Grants Grupos Consolida-dos IT849-13 and ETORTEK Program KK-2015/0000089. D.G.-B. and K.B.U.were recipients of a fellowship from the Bizkaia Biophysics Foundation.

1. Carbonetti NH (2010) Pertussis toxin and adenylate cyclase toxin: Key virulence factorsof Bordetella pertussis and cell biology tools. Future Microbiol 5:455–469.

2. Mattoo S, Cherry JD (2005) Molecular pathogenesis, epidemiology, and clinicalmanifestations of respiratory infections due to Bordetella pertussis and other Bor-detella subspecies. Clin Microbiol Rev 18:326–382.

3. Goodwin MSM, Weiss AA (1990) Adenylate cyclase toxin is critical for colonizationand pertussis toxin is critical for lethal infection by Bordetella pertussis in infant mice.Infect Immun 58:3445–3447.

4. Guermonprez P, et al. (2001) The adenylate cyclase toxin of Bordetella pertussis binds totarget cells via the alpha(M)beta(2) integrin (CD11b/CD18). J Exp Med 193:1035–1044.

5. Berkowitz SA, Goldhammer AR, Hewlett EL, Wolff J (1980) Activation of prokaryoticadenylate cyclase by calmodulin. Ann N Y Acad Sci 356:360.

6. Confer DL, Eaton JW (1982) Phagocyte impotence caused by an invasive bacterialadenylate cyclase. Science 217:948–950.

7. Weingart CL, Weiss AA (2000) Bordetella pertussis virulence factors affect phagocy-tosis by human neutrophils. Infect Immun 68:1735–1739.

8. Ehrmann IE, Gray MC, Gordon VM, Gray LS, Hewlett EL (1991) Hemolytic activity ofadenylate cyclase toxin from Bordetella pertussis. FEBS Lett 278:79–83.

9. Hewlett EL, Donato GM, Gray MC (2006) Macrophage cytotoxicity produced by ad-enylate cyclase toxin from Bordetella pertussis: More than just making cyclic AMP!Mol Microbiol 59:447–459.

10. Benz R, Maier E, Ladant D, Ullmann A, Sebo P (1994) Adenylate cyclase toxin (CyaA) ofBordetella pertussis. Evidence for the formation of small ion-permeable channels andcomparison with HlyA of Escherichia coli. J Biol Chem 269:27231–27239.

11. Dal Molin F, et al. (2006) Cell entry and cAMP imaging of anthrax edema toxin. EMBOJ 25:5405–5413.

12. Ladant D, Ullmann A (1999) Bordatella pertussis adenylate cyclase: A toxin withmultiple talents. Trends Microbiol 7:172–176.

13. Glaser P, Sakamoto H, Bellalou J, Ullmann A, Danchin A (1988) Secretion of cyclolysin,the calmodulin-sensitive adenylate cyclase-haemolysin bifunctional protein of Bor-detella pertussis. EMBO J 7:3997–4004.

14. Bellalou J, Sakamoto H, Ladant D, Geoffroy C, Ullmann A (1990) Deletions affectinghemolytic and toxin activities of Bordetella pertussis adenylate cyclase. Infect Immun58:3242–3247.

15. Sakamoto H, Bellalou J, Sebo P, Ladant D (1992) Bordetella pertussis adenylate cyclasetoxin. Structural and functional independence of the catalytic and hemolytic activi-ties. J Biol Chem 267:13598–13602.

16. Hackett M, Guo L, Shabanowitz J, Hunt DF, Hewlett EL (1994) Internal lysine palmi-toylation in adenylate cyclase toxin from Bordetella pertussis. Science 266:433–435.

17. Bauche C, et al. (2006) Structural and functional characterization of an essential RTXsubdomain of Bordetella pertussis adenylate cyclase toxin. J Biol Chem 281:16914–16926.

18. Rose T, Sebo P, Bellalou J, Ladant D (1995) Interaction of calcium with Bordetellapertussis adenylate cyclase toxin. Characterization of multiple calcium-binding sitesand calcium-induced conformational changes. J Biol Chem 270:26370–26376.

19. El-Azami-El-Idrissi M, et al. (2003) Interaction of Bordetella pertussis adenylate cyclasewith CD11b/CD18: Role of toxin acylation and identification of the main integrininteraction domain. J Biol Chem 278:38514–38521.

20. Rogel A, Hanski E (1992) Distinct steps in the penetration of adenylate cyclase toxin ofBordetella pertussis into sheep erythrocytes. Translocation of the toxin across themembrane. J Biol Chem 267:22599–22605.

21. Basler M, et al. (2007) Segments crucial for membrane translocation and pore-forming activity of Bordetella adenylate cyclase toxin. J Biol Chem 282:12419–12429.

22. Gordon VM, et al. (1989) Adenylate cyclase toxins from Bacillus anthracis and Bor-detella pertussis. Different processes for interaction with and entry into target cells.J Biol Chem 264:14792–14796.

23. Gmira S, Karimova G, Ladant D (2001) Characterization of recombinant Bordetellapertussis adenylate cyclase toxins carrying passenger proteins. Res Microbiol 152:889–900.

24. Osicková A, Osicka R, Maier E, Benz R, �Sebo P (1999) An amphipathic alpha-helixincluding glutamates 509 and 516 is crucial for membrane translocation ofadenylate cyclase toxin and modulates formation and cation selectivity of its mem-brane channels. J Biol Chem 274:37644–37650.

25. Osickova A, et al. (2010) Adenylate cyclase toxin translocates across target cellmembrane without forming a pore. Mol Microbiol 75:1550–1562.

26. Holubova J, et al. (2012) Delivery of large heterologous polypeptides across the cy-toplasmic membrane of antigen-presenting cells by the Bordetella RTX hemolysinmoiety lacking the adenylyl cyclase domain. Infect Immun 80:1181–1192.

E6792 | www.pnas.org/cgi/doi/10.1073/pnas.1701783114 González-Bullón et al.

27. Karst JC, et al. (2012) Identification of a region that assists membrane insertion andtranslocation of the catalytic domain of Bordetella pertussis CyaA toxin. J Biol Chem287:9200–9212.

28. Subrini O, et al. (2013) Characterization of a membrane-active peptide from theBordetella pertussis CyaA toxin. J Biol Chem 288:32585–32598.

29. Bumba L, Masin J, Fiser R, Sebo P (2010) Bordetella adenylate cyclase toxin mobilizesits beta2 integrin receptor into lipid rafts to accomplish translocation across targetcell membrane in two steps. PLoS Pathog 6:e1000901.

30. Rogel A, et al. (1989) Bordetella pertussis adenylate cyclase: Purification and char-acterization of the toxic form of the enzyme. EMBO J 8:2755–2760.

31. Eby JC, Gray MC, Mangan AR, Donato GM, Hewlett EL (2012) Role of CD11b/CD18 inthe process of intoxication by the adenylate cyclase toxin of Bordetella pertussis.Infect Immun 80:850–859.

32. Martín C, et al. (2004) Membrane restructuring by Bordetella pertussis adenylatecyclase toxin, a member of the RTX toxin family. J Bacteriol 186:3760–3765.

33. Contreras FX, Sánchez-Magraner L, Alonso A, Goñi FM (2010) Transbilayer (flip-flop)lipid motion and lipid scrambling in membranes. FEBS Lett 584:1779–1786.

34. Gaspar AH, Machner MP (2014) VipD is a Rab5-activated phospholipase A1 thatprotects Legionella pneumophila from endosomal fusion. Proc Natl Acad Sci USA 111:4560–4565.

35. Tamura M, et al. (2004) Lysophospholipase A activity of Pseudomonas aeruginosatype III secretory toxin ExoU. Biochem Biophys Res Commun 316:323–331.

36. Hirschberg HJHB, Simons JW, Dekker N, Egmond MR (2001) Cloning, expression,purification and characterization of patatin, a novel phospholipase A. Eur J Biochem268:5037–5044.

37. Rydel TJ, et al. (2003) The crystal structure, mutagenesis, and activity studies revealthat patatin is a lipid acyl hydrolase with a Ser-Asp catalytic dyad. Biochemistry 42:6696–6708.

38. Wang A, Yang HC, Friedman P, Johnson CA, Dennis EA (1999) A specific human ly-sophospholipase: cDNA cloning, tissue distribution and kinetic characterization.Biochim Biophys Acta 1437:157–169.

39. Lio YC, Reynolds LJ, Balsinde J, Dennis EA (1996) Irreversible inhibition of Ca(2+)-independent phospholipase A2 bymethylarachidonylfluorophosphate. Biochim BiophysActa 1302:55–60.

40. Dole VP, Meinertz H (1960) Microdetermination of long-chain fatty acids in plasmaand tissues. J Biol Chem 235:2595–2599.

41. Schaloske RH, Dennis EA (2006) The phospholipase A2 superfamily and its groupnumbering system. Biochim Biophys Acta 1761:1246–1259.

42. Martín C, Gómez-Bilbao G, Ostolaza H (2010) Bordetella adenylate cyclase toxinpromotes calcium entry into both CD11b+ and CD11b- cells through cAMP-dependent L-type-like calcium channels. J Biol Chem 285:357–364.

43. Yoo J, Cui Q (2009) Curvature generation and pressure profile modulation in mem-brane by lysolipids: Insights from coarse-grained simulations. Biophys J 97:2267–2276.

44. Basañez G, et al. (2002) Bax-type apoptotic proteins porate pure lipid bilayersthrough a mechanism sensitive to intrinsic monolayer curvature. J Biol Chem 277:49360–49365.

45. Gilbert RJC (2016) Protein-lipid interactions and non-lamellar lipidic structures inmembrane pore formation and membrane fusion. Biochim Biophys Acta 1858:487–499.

46. Allende D, Simon SA, McIntosh TJ (2005) Melittin-induced bilayer leakage depends onlipid material properties: Evidence for toroidal pores. Biophys J 88:1828–1837.

47. Karatekin E, et al. (2003) Cascades of transient pores in giant vesicles: Line tension andtransport. Biophys J 84:1734–1749.

48. Eby JC, et al. (2013) Quantification of the adenylate cyclase toxin of Bordetella per-tussis in vitro and during respiratory infection. Infect Immun 81:1390–1398.

49. Strickland JA, Orr GL, Walsh TA (1995) Inhibition of diabrotica larval growth by pa-tatin, the lipid acyl hydrolase from potato tubers. Plant Physiol 109:667–674.

50. Agarwal S, et al. (2015) Autophagy and endosomal trafficking inhibition by Vibriocholerae MARTX toxin phosphatidylinositol-3-phosphate-specific phospholipaseA1 activity. Nat Commun 6:8745; erratum in Nat Commun (2015) 6:10135.

51. Henriksen J, et al. (2006) Universal behavior of membranes with sterols. Biophys J 90:1639–1649.

52. Bagatolli LA, Gratton E (2000) A correlation between lipid domain shape and binaryphospholipid mixture composition in free standing bilayers: A two-photon fluores-cence microscopy study. Biophys J 79:434–447.

53. Leidy C, Wolkers WF, Jørgensen K, Mouritsen OG, Crowe JH (2001) Lateral organi-zation and domain formation in a two-component lipid membrane system. Biophys J80:1819–1828.

54. van den Brink-van der Laan E, Killian JA, de Kruijff B (2004) Nonbilayer lipids affectperipheral and integral membrane proteins via changes in the lateral pressure profile.Biochim Biophys Acta 1666:275–288.

55. Epand RF, Martinou JC, Fornallaz-Mulhauser M, Hughes DW, Epand RM (2002) Theapoptotic protein tBid promotes leakage by altering membrane curvature. J BiolChem 277:32632–32639.

56. Epand RM (1996) Functional roles of non-lamellar forming lipids. Chem Phys Lipids 81:101–104.

57. Terrones O, et al. (2004) Lipidic pore formation by the concerted action of proa-poptotic BAX and tBID. J Biol Chem 279:30081–30091.

58. Karimova G, Pidoux J, Ullmann A, Ladant D (1998) A bacterial two-hybrid systembased on a reconstituted signal transduction pathway. Proc Natl Acad Sci USA 95:5752–5756.

59. Bychkova VE, Pain RH, Ptitsyn OB (1988) The ‘molten globule’ state is involved in thetranslocation of proteins across membranes? FEBS Lett 238:231–234.

60. Ptitsyn OB, Pain RH, Semisotnov GV, Zerovnik E, Razgulyaev OI (1990) Evidence for amolten globule state as a general intermediate in protein folding. FEBS Lett 262:20–24.

61. Karst JC, et al. (2014) Calcium, acylation, and molecular confinement favor folding ofBordetella pertussis adenylate cyclase CyaA toxin into a monomeric and cytotoxicform. J Biol Chem 289:30702–30716.

62. Gray M, Szabo G, Otero AS, Gray L, Hewlett E (1998) Distinct mechanisms for K+efflux, intoxication, and hemolysis by Bordetella pertussis AC toxin. J Biol Chem 273:18260–18267.

González-Bullón et al. PNAS | Published online July 31, 2017 | E6793

BIOCH

EMISTR

YPN

ASPL

US