Cytotoxic activity of Kingella kingae RtxA toxin depends on post-translational acylation of lysine residues and cholesterol binding

Kingella kingae is a member of the commensal oropharyngeal flora of young children. Improvements in detection methods have led to the recognition of K. kingae as an emerging pathogen that frequently causes osteoarticular infections in children and a severe form of infective endocarditis in children and adults. Kingella kingae secretes a membrane-damaging RTX (Repeat in ToXin) toxin, RtxA, which is implicated in the development of clinical infections. However, the mechanism by which RtxA recognizes and kills host cells is largely unexplored. To facilitate structure-function studies of RtxA, we have developed a procedure for the overproduction and purification of milligram amounts of biologically active recombinant RtxA. Mass spectrometry analysis revealed the activation of RtxA by post-translational fatty acyl modification on the lysine residues 558 and/or 689 by the fatty-acyltransferase RtxC. Acylated RtxA was toxic to various human cells in a calcium-dependent manner and possessed pore-forming activity in planar lipid bilayers. Using various biochemical and biophysical approaches, we demonstrated that cholesterol facilitates the interaction of RtxA with artificial and cell membranes. The results of analyses using RtxA mutant variants suggested that the interaction between the toxin and cholesterol occurs via two cholesterol recognition/interaction amino acid consensus motifs located in the C-terminal portion of the pore-forming domain of the toxin. Based on our observations, we conclude that the cytotoxic activity of RtxA depends on post-translational acylation of the K558 and/or K689 residues and on the toxin binding to cholesterol in the membrane.


Introduction
Kingella kingae is a fastidious, facultative anaerobic, gram-negative coccobacillus of the Neisseriaceae family that was first isolated in 1960 by Elizabeth King [1][2][3] . Kingella kingae is a member of the commensal oropharyngeal flora of young children, and its transmission from child to child is believed to occur through close personal contact 1,4,5 . The process of colonization likely involves the adherence of K. kingae to respiratory epithelial cells through type IV pili 6,7 . The maximal colonization of children by K. kingae occurs between the ages of 6 and 36 months, peaking in the second year of life 3 . The carriage of K. kingae gradually decreases in older children and adults, indicating the acquisition of immunity that eradicates the bacterium from the pharynx 4,8 . Until recently, K. kingae was believed to be a rare cause of infection. However, improvements in culture techniques and molecular detection methods have led to the recognition of the bacterium as an important invasive pediatric pathogen 3,9 . In several reports, K. kingae has been Fig. 1 Schematic representation and purification process for RtxA and proRtxA. a Scheme of the K. kingae RtxA molecule, with several different areas predicted from homology with other RTX toxins. The arrowheads with a letter C indicate the predicted CRAC and CARC motifs. The RtxA (b) and proRtxA (c) proteins were produced in E. coli BL21/pMM100 cells and purified by a combination of affinity and hydrophobic chromatographies. Lanes: 1, crude extract from uninduced cells; 2, crude extract from cells induced with isopropyl β-D-1-thiogalactopyranoside (IPTG) to produce RtxA/proRtxA; 3, clarified crude urea extract from induced cells; 4, Ni-NTA agarose column flowthrough; 5, Ni-NTA agarose column wash; 6, fraction of eluted 105-kDa RtxA/proRtxA; 7, phenyl-sepharose column flowthrough; 8, phenyl-sepharose column wash; 9, fraction of eluted RtxA/proRtxA; and St, molecular-mass standards. The samples were analyzed on 7.5% polyacrylamide gels and stained with Coomassie blue

Production and purification of recombinant RtxA and proRtxA
Biologically active RtxA is normally present in the cell culture supernatants of K. kingae isolates in low quantities, and its isolation and purification from bacterial cultures is a relatively complicated process. Therefore, we developed a procedure to overproduce and isolate a highly purified recombinant form of RtxA. The rtxA gene, encoding the inactive protoxin (proRtxA), and the rtxC gene, encoding a predicted fatty-acyltransferase that activates proRtxA, were PCR amplified from chromosomal DNA of the K. kingae septic arthritis isolate PYKK081 (ref 20 .) and cloned in the expression vector pT7-7 (ref 21 .). To express unmodified proRtxA, only the rtxA gene was inserted into pT7-7. Next, the rtxA gene was fused in frame at its 3′-terminus to a sequence encoding a double-hexahistidine purification tag. The RtxA and proRtxA proteins were expressed in E. coli BL21/pMM100 cells and purified from urea-solubilized inclusion bodies by a combination of affinity chromatography using Ni-NTA agarose and hydrophobic chromatography using phenyl-sepharose (Figs. 1b, c). The identities of the purified RtxA (Supplementary Table 1) and proRtxA (Supplementary Table 2) proteins were confirmed by peptide mass fingerprinting. This procedure yielded 10 to 15 mg of highly purified RtxA and proRtxA proteins per 1 L of bacterial culture.
RtxA but not proRtxA is covalently acylated at lysine residues 558 and 689 Previous studies of other RTX toxins, namely the B. pertussis toxin CyaA [22][23][24] , the E. coli α-hemolysin HlyA 25,26 and the A. actinomycetemcomitans leukotoxin LtxA 27 , demonstrated that they are post-translationally activated through amide-linked fatty acylation of the εamino groups of two conserved internal lysine residues. Sequence alignment analysis of these three RTX toxins with RtxA revealed lysine residues 558 and 689 of pro-RtxA to be putative acylation sites (Fig. 2a). To analyze the acylation of the purified proRtxA and RtxA proteins, they were separated by SDS-PAGE and digested in-gel by trypsin, after which the resulting peptides were separated by reversed-phase microcapillary high-performance liquid chromatography and analyzed by mass spectrometry (MS). The tryptic peptides with m/z signals that were exclusively present in the trypsin-digested RtxA sample (Supplementary Fig. 1) were further analyzed using a tandem MS/MS sequencing approach. As shown by the spectra in Fig. 2b and Supplementary Fig. 2a Table 3).

RtxA but not proRtxA exhibits cytotoxicity against human cells
The cytotoxic activities of the acylated RtxA and the unacylated proRtxA were tested on different human cell types that may be attacked by the toxin during K. kingae infection, including laryngeal HLaC-78 squamous cells, hypopharyngeal FaDu epithelial cells, bone osteosarcoma U-2 OS epithelial cells, synovial SW 982 cells and monocytic THP-1 cells. The cells were exposed to different concentrations of the highly purified RtxA and proRtxA proteins, and the susceptibility of cells to RtxA/ proRtxA-mediated killing was determined by a cell viability assay using the nucleic acid stain Hoechst 33258 followed by flow cytometry. As shown in Fig. 3a, the viability of all tested cells was reduced in a dosedependent manner upon treatment with RtxA, while the viability of cells incubated with unacylated proRtxA remained unaffected over the 1-h testing period. The viability of the most susceptible cells, HLaC-78 and FaDu, was reduced by~70 and~90% upon incubation with 0.5 and 1 µg/ml of RtxA, respectively (Fig. 3a). The remaining three cell types, U-2 OS, SW 982 and THP-1, were approximately two-fold less susceptible to RtxA activity than the HLaC-78 and FaDu cells (Fig. 3a). The cells had to be incubated with RtxA in the presence of 2 mM calcium ions, since the viability of cells treated with the toxin in the absence of calcium ions remained unaffected ( Supplementary Fig. 3). This requirement indicates that as with other RTX toxins, the cytotoxic activity of RtxA is strictly dependent on the presence of free calcium ions 17 . The results of a time course analysis of the cytotoxic effect of RtxA on HLaC-78 cells demonstrated that cell viability decreased rapidly, even at the lowest concentration of the toxin tested. As shown in Fig. 3b, the cell viability was reduced to 50% after HLaC-78 cells were incubated with 0.5 or 1 µg/ml of RtxA for 5 and 2.5 min, respectively. A complete loss of cell viability was observed after HLaC-78 cells were incubated with 1 µg/ml of RtxA for 15 min (Fig. 3b).
We further tested the cytotoxic effect of the recombinant RtxA toxin on sheep erythrocytes, which are commonly used as a simple model to analyze the hemolytic (cytolytic) activity of some RTX toxins 28,29 . Sheep erythrocytes were incubated with different concentrations of purified RtxA, and the hemolytic activity was measured as the amount of hemoglobin released over time by photometric determination. As shown in Fig. 3c,~50% of erythrocytes lysed after exposure to 0.2 µg/ml of RtxA for 15 min, and the complete lysis of erythrocytes was observed when the cells were incubated with 1 µg/ml of the toxin for 10 min. No lysis over the 45-min testing period was observed when erythrocytes were incubated with 1 µg/ml of RtxA in the absence of calcium ions or when incubated with 1 µg/ml of proRtxA (Fig. 3c).
These results demonstrate that the purified recombinant RtxA toxin exerts calcium-dependent cytotoxicity toward various cell types and that this cytotoxicity is dependent on acylation of the K558 and/or K689 residues.

RtxA loses its cytotoxic activity under native conditions
All the experiments described above were performed with RtxA purified from inclusion bodies under denaturing conditions in 8 M urea, since we observed that the toxin had low cytotoxic activity toward target cells when it was purified from bacterial cytosol under native conditions (Figs. 4a, b). Moreover, RtxA that was purified in 8 M urea had to be diluted (100×) into native buffer immediately before being added to cells, as a longer preincubation of the toxin in native buffer resulted in the rapid loss of its cytotoxic activity. As shown in Fig. 4c, HLaC-78 cell viability was reduced by~90% after incubating for 5 min with 2 µg/ml of RtxA that was directly 100-fold diluted from the RtxA stock in 8 M urea. In contrast, when the toxin was first prediluted in native buffer and preincubated for 1, 5 or 10 min before being added to HLaC-78 cells, cell viability was only reduced bỹ 70, 30, and 20%, respectively. Similarly, erythrocytes incubated for 10 min with 0.5 µg/ml of RtxA that was directly diluted from the RtxA stock in 8 M urea were lysed to 90%, while erythrocytes treated with RtxA that was preincubated in native buffer were lysed less efficiently (Fig. 4d).
Recombinant RtxA forms pores in artificial membranes with substantially higher overall membrane activity than proRtxA To characterize the membrane pores formed by the purified recombinant RtxA and proRtxA proteins at the molecular level, we analyzed the single-pore conductance and lifetime of pores formed upon incubation of artificial planar lipid bilayers made of 3% asolectin with RtxA and proRtxA. As shown in Fig. 5 and summarized in Table 1, RtxA generated pores ( Fig. 5a) with three welldistinguishable conductance states, with the most frequently observed values being 38.4, 210.7, and 419.5 pS ( Fig. 5b and Table 1). The majority of pores formed by RtxA exhibited short lifetimes, with 0.24 s being the most frequently observed value, while the remaining toxingenerated pores exhibited longer lifetimes, with 2.12 s being the most frequently observed value ( Fig. 5c and Table 1). A minor proportion of RtxA pores were opened for tens of seconds (Fig. 5a). The proRtxA protoxin formed pores with similar single-pore conductance states  Table 1). The most frequent longer lifetime of proRtxA was 4.11 s, two times higher than that of RtxA ( Fig. 5c and Table 1). However, the acylated RtxA and unacylated proRtxA importantly differed in their overall membrane activity (i.e., the combined contributions of the number of poreforming molecules in the membrane, conductance and lifetime of single pores, and the frequency of pore formation), which was 8.3 pA/s for RtxA and only 1.1 pA/s for proRtxA ( Fig. 5d and Table 1). These results indicate that proRtxA was impaired in its ability to correctly insert into the lipid bilayer and/or in its propensity to form oligomeric pores with the same frequency as the acylated toxin. a Different cell types (1 × 10 6 /ml) were incubated with increasing concentrations (0.5, 1, 2, and 4 µg/ml) of purified RtxA or proRtxA in the presence of 2 mM calcium ions for 1 h at 37°C. b HLaC-78 cells (1 × 10 6 /ml) were incubated with the indicated concentrations of purified RtxA for different times at 37°C. a, b Cell viability was determined by a cell viability staining assay using 1 µg/ml of Hoechst 33258 followed by flow cytometry. The viability of cells incubated without RtxA was reported as 100%. Each point represents the mean value ± SD of four independent experiments. c Erythrocytes (5 × 10 8 / ml) were incubated at 37°C in the presence of increasing concentrations (0.1, 0.2, and 1 µg/ml) of purified RtxA or 1 µg/ml of proRtxA. Erythrocytes incubated with 1 µg/ml of RtxA in the absence of calcium ions were used as negative control. Cytolytic (hemolytic) activity was measured at different times as the amount of released hemoglobin by photometric determination (A 541nm ). Complete lysis of erythrocytes was reported as 100%. Each point represents the mean value ± SD of three independent determinations performed in duplicate

Time (min)
Cholesterol is important for the binding and cytotoxic activity of RtxA The presence of cholesterol in the cell membrane was previously observed to be important for the cytotoxic activities of some RTX toxins, two of which, LtxA and HlyA, have been shown to specifically bind cholesterol [30][31][32] . To examine if cholesterol has a role in the membrane binding of RtxA and proRtxA, we visualized the affinities of both proteins to 1-palmitoyl-2-oleoyl-sn-glycero-3-phosphocholine (POPC) and POPC/cholesterol (75/25) membranes using a fluorescent giant unilamellar vesicle (GUV) assay 30 . RtxA was labeled with AF555 (red) and GUVs were labeled with NBD-PE (green). As shown in Fig. 6a, little RtxA was observed to be associated with the 100% POPC membrane, but the toxin bound readily to GUVs composed of 75% POPC and 25% cholesterol. The experiment was repeated for proRtxA, where proRtxA was labeled with Dy495 (green) and GUVs were labeled with Marina Blue (blue). As with RtxA, lower binding of proRtxA was observed to 100% POPC GUVs than to cholesterol-containing GUVs (Fig. 6b).
The affinities of RtxA and proRtxA for membranes with and without cholesterol were quantified using surface plasmon resonance (SPR). As shown in Fig. 6c and summarized in Supplementary Table 4, the equilibrium dissociation constant (K D ) of RtxA for POPC membranes was approximately 1.5 × 10 -9 M, while that of proRtxA was slightly lower, although this difference was not significant. Consistent with other RTX toxins, including LtxA 30 , the K D of RtxA for cholesterolcontaining membranes was lower than its K D for POPC membranes, indicating a stronger affinity for the POPC/cholesterol membrane ( Fig. 6c and Supplementary Table 4). Similarly, the K D of proRtxA for POPC/ cholesterol membranes was lower than its K D for POPC membranes, indicating a stronger affinity for the POPC/ cholesterol membrane ( Fig. 6c and Supplementary   Table 1, and the numbers in each KDE represent the most frequent values of three well-distinguishable singlepore conductance states. c For lifetime determination, approximately 400 individual pore openings were recorded on several different asolectin membranes with 0.25, 0.5, and 1 nM RtxA or proRtxA (all other parameters were the same as in (a)), and the KDE of dwell times was fitted with a double-exponential function. The error estimates of lifetimes were obtained by bootstrap analysis. The results are summarized in Table 1, and the numbers in each graph represent the most frequent values of two well-distinguishable lifetimes. d Overall membrane activity of the purified RtxA and proRtxA proteins on asolectin/decane:butanol (9:1) membranes using the same conditions as in (a) also similar, with the exception of RtxA binding to POPC/ cholesterol, which was one order of magnitude slower than the other dissociation rates. This slower dissociation rate drives the increase in the overall affinity (K D ) of this toxin for cholesterol-containing membranes relative to POPC membranes. To analyze whether RtxA is able to interact with cholesterol in the absence of lipid membranes, the wells of a 96-well plate were coated with cholesterol conjugated to BSA or free BSA, with the latter used as a negative control. The coated wells were incubated with different concentrations of purified RtxA that was subsequently detected by the antibody 9D4, which recognizes the RTX domain of RtxA 18 . As shown in Fig. 6d, the RtxA toxin exhibited a 2-3-fold greater binding capacity to cholesterol-BSA than to free BSA at all tested concentrations.
We further examined the role played by cholesterol in the membrane binding and cytotoxic activity of RtxA toward erythrocytes. First, erythrocytes were preincubated in the absence or in the presence of 1 mM methyl-βcyclodextrin (MβCD), which is commonly used to deplete membrane cholesterol 33,34 . Next, the cells were incubated with different concentrations of fluorescently labeled RtxA, and binding of the toxin to erythrocytes was analyzed by flow cytometry. As shown in Fig. 7a, RtxA bound with an approximately 30 to 40% lower efficiency to MβCD-treated cells compared with untreated cells, indicating that membrane cholesterol facilitates binding of the toxin to the cell surface.
To analyze whether free cholesterol decreases the cytotoxic activity of RtxA, the toxin was preincubated with 5 or 50 nM of free cholesterol and was subsequently incubated in the presence of free cholesterol with erythrocytes for different periods of time. As shown in Fig. 7b, the exposure of RtxA to free cholesterol significantly reduced its hemolytic activity over the 1-h testing period.
Taken together, these results demonstrate that the interaction of RtxA with the target membrane is enhanced by membrane cholesterol, which directly interacts with the toxin.
Phenylalanine substitutions of the Y343 and Y352 residues in the putative cholesterol binding motifs reduce the cytolytic activity of RtxA A primary sequence analysis of RtxA revealed five potential cholesterol binding sites located within or adjacent to the predicted pore-forming domain (residues 140 to 410) of the toxin (Fig. 1a and Supplementary Table 5). Two of these sites are so-called recognition/ interaction amino acid consensus (CRAC) motifs, consisting of the pattern L/V-(X)(1-5)-Y-(X)(1-5)-R/K (where (X) (1)(2)(3)(4)(5) represents between one and five residues of any amino acid) 35 , and three are CARC motifs, which are inverted CRACs with the consensus sequence R/K-(X) (1-5)-Y/F-(X)(1-5)-L/V 36 . The presence of these motifs suggests that RtxA may directly interact with host cells through these predicted CRAC/CARC motifs and membrane cholesterol, as has been demonstrated for the RTX toxin LtxA 30,31 . To test which of the predicted CRAC/ CARC motifs in RtxA mediate membrane cholesterol binding, we replaced the key tyrosine residues of these motifs (Y54, Y285, Y343, Y352, and Y448) with phenylalanine residues. As shown in Fig. 7c, the Y54F, Y285F, and Y448F substitutions in the putative CRAC/CARC motifs did not significantly affect the lytic activity of the RtxA mutants toward erythrocytes. In contrast, the Y343F and Y352F substitutions in the CARC 340-348 and CRAC 349-354 motifs, respectively, reduced the lytic activity of the mutant RtxA proteins to~50% that of wild-type RtxA after a 20 min incubation of the toxins with erythrocytes (the half time of cell lysis with RtxA; Fig. 7c). Thus, the tandem CARC-CRAC motifs located between residues 340-354 of RtxA may be involved in the interaction between the toxin and membrane cholesterol.

Discussion
The K. kingae RtxA toxin belongs to a broad family of the secreted RTX cytotoxins, which are gaining attention as important virulence factors of various gram-negative bacterial pathogens 17 . While some RTX toxins, such as the E. coli α-hemolysin HlyA and the B. pertussis adenylate cyclase toxin CyaA, have been intensively studied over the last decades 17,37-39 , the cytotoxin RtxA is one of the least characterized members of the RTX toxin family. To perform biological, biochemical, biophysical and structural studies of RtxA, we developed a simple procedure for the overproduction and purification of milligram amounts of biologically active recombinant RtxA. Using different approaches, we have demonstrated that the cytotoxic activity of RtxA depends on post-translational acylation of its lysine residues 558 and/or 689 and that cholesterol facilitates the interaction of the toxin with artificial and cell membranes.
Although it is generally accepted that the RTX toxins are post-translationally acylated on the ε-amino groups of two conserved lysine residues, the type of acyl chains has only been experimentally identified for three RTX toxins, namely the B. pertussis CyaA 22-24 , the E. coli α-hemolysin HlyA 25,26 and the A. actinomycetemcomitans leukotoxin LtxA 27 . Analysis of the predicted acylated segment of RtxA and its alignment with the corresponding regions of CyaA, HlyA and LtxA revealed that the RtxA protein contains two conserved lysine residues, K558 and K689 (Fig. 2a), that could be post-translationally modified. This modification was experimentally confirmed by MS analyses of the RtxA protein coexpressed with a putative acyltransferase (RtxC), which revealed that the K558 residue was modified by C14:0 and C14:0-OH fatty acyl groups ( Fig. 2b and Supplementary Fig. 2a), whereas the K689 residue was modified by several fatty acyl groups, including C12:0, C14:0, C14:0-OH and C16:1 (Fig. 2c and Supplementary Fig. 2b-d). As demonstrated recently, C12:0, C14:0 and C16:1 are the major fatty acids present in K. kingae ATCC 23330 T and K. negevensis Sch538 T (ref 40 .). The saturated C14:0 acyl group was observed to be the most dominant acyl form, accounting for~18 and 71% of K558 and K689 acylation of RtxA, respectively (Supplementary Table 3  We demonstrated that similar to RtxA secreted by K. kingae 11 , the highly purified recombinant RtxA is a potent cytotoxin capable of rapid and efficient killing of various human cell types (Figs. 3a, b). Moreover, RtxA was able to efficiently lyse sheep erythrocytes (Fig. 3c) that are commonly used as a model to test the cytolytic activity of some other RTX toxins 28,41 . In contrast, the viability of cells incubated with unacylated proRtxA remained unaffected (Fig. 3a), and the protoxin was unable to lyse erythrocytes (Fig. 3c). Our data are in agreement with many previous studies showing that the acylation of RTX toxins is crucial for their cytotoxic activities 25,42,43 . Although the exact molecular mechanism by which the acylation contributes to the toxicity of the RTX toxins is unknown, the acyl chains have been demonstrated to have a structural role in supporting the folding of CyaA into biologically active states 44 , are responsible for the irreversible binding of HlyA to target membranes 45 , and/or promote proteinprotein interactions between HlyA monomers to enable toxin oligomerization in the membrane during pore formation 46 . Therefore, we hypothesize that the acyl chains of RtxA may have a direct impact on these steps, which are also believed to be crucial for the cytotoxic activity of RtxA. In agreement with the inability of proRtxA to reduce cell viability (Fig. 3a) or to lyse erythrocytes (Fig. 3c), the unacylated protein displayed a substantially Fig. 7 Cholesterol is important for the binding and cytotoxic activity of RtxA toward erythrocytes. a Erythrocytes (5 × 10 8 /ml) were pretreated with 1 mM MβCD at 37°C for 30 min, diluted to 1 × 10 7 /ml and placed on ice before RtxA labeled with Dy495 was added at the indicated concentrations. The cells were incubated with the toxin for 15 min at 4°C, and the amount of fluorescently labeled RtxA bound was determined by flow cytometry. The results are expressed as the relative binding of RtxA according to the following formula: relative binding = (sample binding)/(maximum binding) × 100. The data shown are the mean values ± SD of four independent experiments. b Purified RtxA (1 µg/ml) was preincubated for 10 min at room temperature in the presence (5 and 50 nM) or absence of cholesterol and then was added to erythrocytes (5 × 10 8 /ml). The cells were incubated for different times at 37°C and analyzed as described in the Fig. 3 legend. Each point represents the mean value ± SD of three independent determinations performed in duplicate. c Substitutions of the tyrosine residues Y343 and Y352 by phenylalanine residues in the putative CARC/CRAC motifs reduce the cytolytic activity of RtxA. Erythrocytes (5 × 10 8 /ml) were incubated with 100 ng/ ml of intact RtxA or its mutant variants for different times at 37°C and analyzed as described in the c reduced overall membrane activity in planar lipid bilayers compared to RtxA (Fig. 5d). The unacylated protein was most likely impaired in its ability to bind/insert into the membrane and/or in its propensity to form oligomeric pores, as discussed above, since once inserted to the membrane, proRtxA formed pores with similar singlepore conductance states and lifetimes as RtxA (Fig. 5 and Table 1). This result is in agreement with previous studies showing that despite highly decreased overall membrane activity, the unacylated proHlyA and proCyaA proteins formed pores with similar properties as the acylated toxins 42,47,48 . The inefficient binding of proHlyA to the lipid bilayers 47 and the highly reduced propensity of proCyaA to form membrane pores 42,48 was suggested to be responsible for the reduced membrane activity.
We observed that when expressed as recombinant protein in E. coli and purified from urea-solubilized inclusion bodies under denaturing conditions in 8 M urea, RtxA was much more active against target cells than when the toxin was purified according the same scheme from the bacterial cytosol under native conditions (Figs. 4a, b). Moreover, to attain maximal biological activity, denatured and stabilized RtxA stored in 8 M urea had to be diluted into urea-free buffer just prior to being added to cells, since preincubation of RtxA in the native buffer resulted in a rapid reduction in its cytotoxic activity (Figs. 4c, d). This result indicates that RtxA has a tendency to selfaggregate and lose cytotoxic activity in solutions that do not contain denaturing concentrations of chaotropic agents. A similar loss of biological activity was described for other members of the RTX toxin family, which also tend to form biologically inactive aggregates in buffers that do not contain high concentrations of chaotropic agents 44,49-53 . Self-aggregation and inactivation of the RTX toxins has been proposed to result from the exposure and interaction of hydrophobic segments that are predominantly present in the membrane-binding poreforming domains of the toxin molecules 44,51,53 . Therefore, RtxA and other RTX toxins have to be diluted into ureafree buffers just prior to addition to target cells to minimize the time for formation of toxin self-aggregates. In the presence of target cells, the hydrophobic regions of the toxins can directly interact with and insert into cell membranes, minimizing the rapid inactivation of the toxin molecules and resulting in highly specific cytotoxic activity of the RTX toxins 44,[49][50][51][52][53] .
The RTX toxins HlyA and LtxA have been shown to specifically interact with lipid monolayers, liposomes and/ or target cells through membrane cholesterol [30][31][32] , which is an essential structural component of all animal cell membranes. In this study, we used several different approaches to demonstrate that RtxA also has an affinity for cholesterol. We clearly visualized binding of RtxA and proRtxA to GUVs composed of 75% POPC and 25% cholesterol, whereas only weak binding of both proteins to GUVs made of 100% POPC was observed (Figs. 6a, b). In agreement with this result, stronger affinities of RtxA and proRtxA for the cholesterol-containing POPC membranes than for pure POPC membranes were observed via SPR ( Fig. 6c and Supplementary Table 4). No significant difference in binding of either the acylated RtxA or the unacylated proRtxA to the cholesterol-containing POPC membranes was observed (Fig. 6c), indicating that the acyl chains of RtxA are not involved in its ability to bind cholesterol. Using an ELISA assay, we demonstrated that the toxin exhibits a 2-3-fold greater ability to bind cholesterol-BSA than free BSA (Fig. 6d), indicating that RtxA can bind cholesterol independently of the presence of other biological membrane components. In line with this result, preincubation of RtxA with free cholesterol significantly reduced its cytolytic activity toward erythrocytes (Fig. 7b). In addition, decreasing the cholesterol content of erythrocyte membranes by extraction with MβCD resulted in a 30-40% reduction in RtxA binding to MβCD-treated cells (Fig. 7a), indicating that membrane cholesterol facilitates binding of RtxA to the cell surface.
Many proteins that interact with membrane cholesterol have been shown to possess specific cholesterol binding motifs, of which the linear cholesterol binding motif CRAC and its reversed form CARC are well-known and widespread among cholesterol-binding proteins of diverse functions [54][55][56] . Two CRAC motifs were identified in the RTX leukotoxin LtxA, but only the CRAC 333-339 motif, highly conserved among RTX toxins, in the putative membrane-interacting domain (residues 1-420) of LtxA was shown to be required for its toxicity 30 . Seven CRAC and thirteen CARC motifs were also predicted within the RTX hemolysin HlyA sequence, but their involvement in the interaction of HlyA with membrane cholesterol has not been experimentally demonstrated 32 . In this study, we identified five putative CRAC and CARC motifs located within or adjacent to the predicted pore-forming domain of RtxA (residues 140 to 410) ( Fig. 1a and Supplementary Table 5). Assays using a panel of RtxA CRAC/CARC mutants demonstrated that the CARC 340-348 and CRAC 349-354 motifs located between residues 340-354 are required for the full cytolytic activity of RtxA (Fig. 7c). Since both of these motifs are adjacent to each other in the predicted membrane-interacting and pore-forming domain of RtxA, it is tempting to speculate that this CARC-CRAC tandem may be involved in the interaction of the toxin with membrane cholesterol. Indeed, the presence of both CARC and CRAC within the same transmembrane segment was shown to allow for its interaction with two cholesterol molecules 55 .
In summary, we developed an E. coli expression system that enables the production of milligram amounts of highly purified and biologically active RtxA. We expect that this system will greatly facilitate structural and functional studies of this key virulence factor, eliminating the need for time-consuming genetic manipulations and RtxA production in K. kingae. The production of recombinant RtxA allowed us to demonstrate that the cytotoxic activity of this protein depends on posttranslational acylation of the K558 and/or K689 residues and on its ability to bind to membrane cholesterol.

Bacterial strains
The K. kingae septic arthritis isolate PYKK081 (ref 20 .) was grown on Columbia agar with 5% sheep blood under a 5% CO 2 atmosphere at 37°C for 24 h. The E. coli strain XL1-Blue (Stratagene, La Jolla, CA) was used throughout this study for DNA manipulations and was grown in Luria-Bertani medium at 37°C. The E. coli strain BL21 (Novagen, Madison, WI) carrying the plasmid pMM100 (encoding LacI and tetracycline resistance) was used to express the RtxA proteins.

Plasmid construction
To construct a plasmid to overproduce proRtxA, the rtxA gene was amplified from K. kingae isolate PYKK081 (ref 20 .) genomic DNA and was cloned into the vector pT7-7 (ref 21 .), yielding pT7rtxA. To express RtxCactivated RtxA, the plasmid pT7rtxC-rtxA was constructed by amplifying the rtxC gene from PYKK081 genomic DNA and cloning it into pT7rtxA. Both the rtxA and rtxC genes were placed under the control of the strong transcriptional and translational initiation signals of gene 10 from bacteriophage T7 and of the lactose promoter-operator region. A sequence encoding a double-hexahistidine tag 58 was fused in frame to the 3'end of the RtxA-encoding gene. Oligonucleotide-directed PCR mutagenesis was performed to construct pT7rtxC-rtxA-derived plasmids for the expression of RtxA mutant variants.

Protein production, purification, and labeling
The recombinant proRtxA, RtxA and RtxA mutant variants were produced in E. coli BL21/pMM100 cells and purified by a combination of affinity and hydrophobic chromatography. Dy495-NHS ester was used to label the proRtxA and RtxA proteins. The details of this procedure are provided in the Supplementary Information.

MS analysis
The purified proRtxA and RtxA proteins were separated by SDS-PAGE, digested in-gel with trypsin, and the resulting peptides were analyzed in parallel via liquid chromatography-mass spectrometry (LC-MS, ESI-qTOF) or off-line peptide separation followed by matrix-assisted laser desorption/ionization time-of-flight MS analysis. All details for these procedures are provided in the Supplementary Information.

Planar lipid bilayers
Measurements on planar lipid bilayers 59 were performed in Teflon cells separated by a diaphragm with a circular hole bearing the membrane, as described in detail in the Supplementary Information.

Cytotoxicity assay
Cultured cells were harvested, washed once in a HEPESbuffered salt solution (HBSS buffer; 10 mM HEPES (pH 7.4), 140 mM NaCl, and 5 mM KCl) supplemented with 2 mM CaCl 2 and 2 mM MgCl 2 (HBSS-Ca/Mg buffer) and diluted with the same buffer to 1 × 10 6 cells/ml. For experiments in the absence of calcium ions, the cells were prepared in HBSS buffer supplemented with 2 mM EDTA (HBSS-EDTA buffer). The cells were subsequently incubated with the indicated concentrations of purified pro-RtxA or RtxA for specified times at 37°C. Cell viability was determined by a cell viability staining assay using 1 µg/ml of Hoechst 33258 followed by flow cytometry on a FACS LSR II instrument (BD Biosciences, San Jose, CA) 60 . Data were analyzed using the FlowJo software (Tree Star, Ashland, OR), and appropriate gatings were used to exclude cell aggregates and dying/dead cells (Hoechst 33258-positive staining). Cell viability in each sample was calculated according to the following formula: (count of viable cells in a sample with proRtxA or RtxA/count of viable cells in a sample without proRtxA or RtxA) × 100.

Hemoglobin release assay
Sheep erythrocytes (LabMediaServis, Jaromer, Czech Republic) stored in Alsever's Solution (Sigma-Aldrich) were repeatedly washed with TNC buffer (50 mM Tris-HCl (pH 7.4), 150 mM NaCl and 2 mM CaCl 2 ). The washed erythrocytes (5 × 10 8 /ml) were then incubated with the indicated concentrations of proRtxA, RtxA or the RtxA mutant variants in 1 ml of TNC buffer, and hemolytic activity was measured over time by photometric determination (A 541 ) of the hemoglobin release.

GUV preparation
GUVs with compositions of POPC (99% POPC, 1% N-(7-nitrobenz-2-oxa-1,3-diazol-4-yl)−1,2-dihexadecanoylsn-glycero-3-phosphoethanolamine (NBD-PE) or Marina Blue) or POPC/cholesterol (74% POPC, 25% cholesterol, and 1% NBD-PE or Marina Blue) were created by mixing the required amounts of lipids dissolved in chloroform to a final lipid concentration of 4 mg/ml. The lipid/chloroform mixture was spin-coated onto a glass slide coated with indium tin oxide (SPI, West Chester, PA). Next, the slides were dried under vacuum for 30 min to remove any residual solvent. A polydimethylsiloxane spacer was used to separate two slides, creating a compartment that was filled with ultrapure water and sealed. Subsequently, an electric field was applied to the setup for 3 h at room temperature to form GUVs, which were used the same day they were prepared.

Confocal microscopy
The proRtxA and RtxA proteins were labeled with Dy495-NHS ester (Dyomics, Jena, Germany) as described in the Supplementary Information or with Alexa Fluor (AF) 555 NHS Ester (Molecular Probes, Eugene, OR) according to the manufacturer's instructions. GUVs were incubated with 30 ng of the fluorescently labeled proteins in ibiTreat µ-dishes (Ibidi, Martinsried, Germany) for 30 min. Imaging was conducted using a Nikon C2si + confocal microscope equipped with an LU-N4S laser unit and a 60 × oil objective (NA = 1.4). The images were processed using Nikon's imaging software suite, Elements v.4.3.
Liposome preparation for surface plasmon resonance (SPR) and SPR of proRtxA and RtxA SPR was performed to quantify the affinities of RtxA and proRtxA for membranes with and without cholesterol as described in the Supplementary Information.

Enzyme-linked immunosorbent assay (ELISA)
Cholesterol-captured ELISA was performed to analyze whether RtxA interacts with cholesterol in the absence of lipid membranes. The details of this assay are provided in the Supplementary Information.

RtxA binding to cholesterol-depleted erythrocytes
Washed sheep erythrocytes (5 × 10 8 /ml) in TNC buffer were preincubated at 37°C for 30 min in the absence or presence of 1 mM MβCD. Next, the cells were washed with TNC buffer, diluted to 1 × 10 7 /ml with TNC buffer supplemented with 75 mM sucrose (TNCS buffer), placed on ice and incubated with different concentrations of purified and Dy495-labeled RtxA for 15 min at 4°C. Finally, the cells were washed with and resuspended in cold TNCS buffer, and the amounts of fluorescently labeled RtxA bound to the surface of erythrocytes were determined by flow cytometry.

Statistical analysis
The results were expressed as the arithmetic means ± the standard deviation (SD) of the mean. Statistical analysis was performed by one-way ANOVA followed by Dunnett's post-test using GraphPad Prism 6.0 (GraphPad Software, La Jolla, CA).