Variable Substrate Preference among Phospholipase D Toxins from Sicariid Spiders*

Background: Phospholipase D toxins from brown spider venoms can cause disease in humans. Results: Different toxin family members show specificity for lipid substrates with choline or ethanolamine headgroups or can be ambiguous. Conclusion: Spider phospholipase D toxins have evolved diverse substrate preferences. Significance: The diverse substrate preference may be significant for predation and the mammalian toxicity of venom. Venoms of the sicariid spiders contain phospholipase D enzyme toxins that can cause severe dermonecrosis and even death in humans. These enzymes convert sphingolipid and lysolipid substrates to cyclic phosphates by activating a hydroxyl nucleophile present in both classes of lipid. The most medically relevant substrates are thought to be sphingomyelin and/or lysophosphatidylcholine. To better understand the substrate preference of these toxins, we used 31P NMR to compare the activity of three related but phylogenetically diverse sicariid toxins against a diverse panel of sphingolipid and lysolipid substrates. Two of the three showed significantly faster turnover of sphingolipids over lysolipids, and all three showed a strong preference for positively charged (choline and/or ethanolamine) over neutral (glycerol and serine) headgroups. Strikingly, however, the enzymes vary widely in their preference for choline, the headgroup of both sphingomyelin and lysophosphatidylcholine, versus ethanolamine. An enzyme from Sicarius terrosus showed a strong preference for ethanolamine over choline, whereas two paralogous enzymes from Loxosceles arizonica either preferred choline or showed no significant preference. Intrigued by the novel substrate preference of the Sicarius enzyme, we solved its crystal structure at 2.1 Å resolution. The evolution of variable substrate specificity may help explain the reduced dermonecrotic potential of some natural toxin variants, because mammalian sphingolipids use primarily choline as a positively charged headgroup; it may also be relevant for sicariid predatory behavior, because ethanolamine-containing sphingolipids are common in insect prey.

clade members, generally belonging to the ␤I subclade, show little to no SMase D activity and have diminished toxic effects in animal models (7-10, 25, 26). SicTox proteins with confirmed low SMase D activity have also been termed "class IIb" PLDs (27,28); all these are members of the SicTox ␤ clade.
Sicarius SicTox genes are particularly interesting with respect to functional diversity within this gene family, because only ␤ clade paralogs have been found in Sicarius, and whole venoms of some species of New World Sicarius exhibit diminished SMase D activity compared with Loxosceles and Old World Sicarius venom. Even though there is variability in SMase D activity, all Sicarius species express diverse ␤ clade paralogs in their venom (24,29). ␤I clade (class IIb) PLDs could be generally less active or inactive SMase D enzymes, but it seems more likely that they are specific for non-SM substrates that have yet to be identified. In support of this hypothesis, they conserve all key residues thought to be important for enzyme activity (Fig. 2) and are present in diverse sicariid venoms. In addition, ␤I clade enzymes are present in multiple New World species of Loxosceles (see Figs. 1 and 2), although in lower abundance than ␣ clade enzymes. We propose that the ␤I clade (class IIb) proteins perform some important but undiscovered function in the venom mixture.
Identifying molecular target(s) for these toxins is important for understanding the evolution of functional specificity within the SicTox gene family, the biology of the sicariid spiders, and the molecular mechanism underlying the diminished toxicity of the ␤I clade toxins in mammalian models. Here, we utilize 31 P NMR-based substrate screening to identify substrates and substrate preferences for two diverse ␤I clade members as follows: St_␤IB1i from Sicarius terrosus, a New World species that lacks venom SMase D activity in vitro and has no known ␣ clade toxins (24); and La_␤ID1 from Loxosceles arizonica, a species that is known to cause loxoscelism and does produce ␣ clade toxins (30 -32). We include parallel characterization of La_␣IB2bi, a previously characterized ␣ clade toxin from L. arizonica (15,30), both as a positive control for activity against SM and LPC (Fig. 2) and to provide a direct comparison between substrate preferences of ␣ clade and ␤I clade members. Inclusion of La_␣IB2bi also provides an intraspecies comparison with La_␤ID1.
We report that St_␤IB1i, like other ␤I/class IIb toxins, has low activity against SM and LPC; however, it selectively turns over both ceramide and lysolipid substrates containing ethanolamine headgroups. La_␤ID1, by contrast, shows little preference between choline and ethanolamine, and La_␣IB2bi shows a preference for choline. Our results show for the first time that 1) SicTox enzymes vary widely in substrate selectivity and ambiguity and 2) that some ␤I/class IIb enzymes with little to no SMase D activity are actually active enzymes that prefer other substrates. To further investigate substrate preference by St_␤IB1i, we determined its crystal structure, the first of a class IIb PLD as well as the first structure of a SicTox PLD from a Sicarius species. Substrate docking studies using this structure did not reveal a clear basis for its different activity but did shed light on the enzyme mechanism.
Cloning of Coding Sequence of SicTox Isolates-Generation of cDNA libraries of SicTox homologs has been described previously (24). To compare the levels of enzyme ambiguity in the SicTox PLD family, and to investigate substrate specificity in the ␤ clade, we chose a set of three SicTox PLD from two different sicariid species. The three toxins in this study are as follows: La_␣IB2bi (gb:AY699703); La_␤ID1 from L. arizonica (gb:KM884812); St_␤IB1i from S. terrosus (gb:KM884813). La_␣IB2bi has been identified and studied previously (15,30) and is used here as a representative of the SicTox ␣ clade. St_␤IB1i from S. terrosus was chosen as a representative of the ␤ clade. St_␤IB1i is a close homolog to a previously reported (24) S. terrosus sequence, ␤IB1 (gb: FJ171474.1). The full-length clone was isolated from a previously obtained cDNA pool (24) using a 5Ј-rapid amplification of cDNA ends kit (Invitrogen). The GSP2 primer was designed to specifically amplify the missing 5Ј end of St_␤IB1 (5Ј-CAAATGTGCTCACGAATGC-3Ј), and St_␤IB1i was amplified using the using the manufacturer's SCHEME 1. Relevant substrates and products for our study of SicTox enzymes. The reaction at left shows conversion of sphingolipid substrates to cyclic ceramide phosphate; the reaction at right shows conversion of lysophospholipid substrates to cyclic phosphatidic acid. R 1 , R 2 , and R 3 represent alkyl chains of varying length. The four different headgroup alcohols (XOH) represented here are choline (SM and LPC), ethanolamine (CPE and LPE), glycerol (LPG), and serine (LPS).
To obtain a divergent ␤ clade paralog from L. arizonica, allowing for comparison of two intraspecific paralogs, an Illumina cDNA library of L. arizonica cephalothorax (including venom glands) was searched for the motifs DNPW, HMVN, and HXXPCDCXRXC. Transcript 1239 was isolated and phylogenetically resolved in the ␤ clade. This transcript is confirmed to be venom expressed based on hits to venom proteins using proteomic analysis (MudPIT, data not shown). The fulllength clone was then amplified from the venom gland cDNA pool using two primers designed from the initial transcript sequence, TOPO cloned (Invitrogen) and sequenced. The predicted amino acid sequence of locus 1239 shares 85% sequence identity with LiRecDT5 (LiSicTox-␤ID1) from L. intermedia (9), another SicTox PLD reported to have diminished SMase D activity compared with ␣ clade isoforms from the same species. Because of this relatedness (Fig. 1), we named the 1239 clone "La_␤ID1" as outlined in the SicTox nomenclature (24).
Signal sequences and mature protein sequences were predicted by analysis of the primary sequence with SignalP (33). Cloning of DNA sequences, encoding the mature PLD toxins St_␤IB1i and La_␤ID1 obtained from cDNA libraries, into pHis8 expression vectors (34) was performed as described previously for La_␣IB2bi (15,30). Sequence variants of St_ ␤IB1i were generated using QuikChange site-directed mutagenesis according to the manufacturer's protocol.
Heterologous Expression and Purification of SicTox PLD Toxins-N-terminally His 8 -tagged recombinant proteins were expressed from pHis8-toxin constructs in Escherichia coli strain BL21(DE3), closely following published methods (15,35). Freshly transformed cells were grown with shaking at 250 rpm at 37°C in 1 liter of 2ϫ YT media (20 mg/ml tryptone, 10 mg/ml Bacto yeast extract, 5 mg/ml NaCl (pH 7)) containing 30 g/ml kanamycin. When the cultures reached an A 600 of ϳ0.6, FIGURE 1. Simplified cladogram of SicTox gene products that have been biochemically characterized. The tree, which is based on a previously reported phylogeny (24), includes all SicTox gene products that have been tested for PLD activity following isolation from whole venom or purification from recombinant sources. The names of toxins investigated in this study are highlighted in purple. The ingroup includes two major clades (␣ and ␤). IsSMase from Ixodes scapularis (74) is included as an outgroup. Homologs with Ͼ95% amino acid sequence identity are considered isoforms and grouped under the same name. Names follow the SicTox nomenclature (24), but alternative names used in the literature are also given, and PDB codes are given for known protein structures. Branch colors correspond to species groups as previously presented (24): red, L. reclusa; green, Loxosceles spadicea; purple, Loxosceles gaucho; blue, L. laeta, and pink, South America Sicarius species. Branches lacking terminal names represent clades for which no member has been characterized biochemically. Except for toxins characterized in this study, high or low SMase D activity is based on comparisons with the activity of whole venom (at the same protein amount) from the toxin source species. In the case of ␤1A1 (SMase II) from L. laeta, SMase D activity has been observed but with a 3-fold higher K m value and a reported 2-fold lower reaction rate at saturating substrate concentrations compared with ␣III1 (SMase I) from the same species (26). References for structure and activity data are as follows: IsSMase (74); LiRecDT6 (11); LiRecDT7 (14); SMase I (8,18,27,35,59,64,75); Ll2 (8); LiRecDT4 (9); LgRec1 (76); Lr1/Lb1/Lb2 (8,25); L. arizonica ␣IB2bi (15,30); Lr2 (19,25); P1/P2 (77-79); LiRecDT1 (6,7,12,13,20,60); 3RLG (61); 3RLH (28); LiRecDT2 (7); LiRecDT5 (9); Lb3 (8,25); LiRecDT3 (7, 10); and SMase II (26). overexpression was induced by addition of isopropyl ␤-D-1thiogalactopyranoside to a concentration of 0.1 mg/ml. After 2-3 h, cells were harvested by centrifugation at 5,000 ϫ g for 10 min. To extract soluble proteins, pelleted cells were lysed with BugBuster Protein Extraction Reagent (Novagen. Madison, WI) according to the manufacturer's protocol. Cleared lysate was brought to ϳ20 mM imidazole by addition of NE250 buffer (0.1 M Tris-HCl (pH 7.4), 0.2 M NaCl, and 0.25 M imidazole) and loaded onto a gravity flow column containing 3 ml of nickelnitrilotriacetic acid-agarose resin (Qiagen) equilibrated in NW20 buffer (0.1 M Tris-HCl (pH 7.4), 0.2 M NaCl, and 20 mM imidazole). Resin was washed twice with 30 ml of NW20 buffer following collection of flow through. His 8 -tagged PLD toxin was eluted with 2ϫ 10-ml aliquots of NE250 buffer. Eluate fractions containing Ͼ80% pure protein, as judged by inspection of Coomassie-stained SDS-polyacrylamide gels run under nonreducing conditions, were pooled, centrifugally concentrated to a volume of 2 ml, and loaded at a flow rate of 0.5 ml/min onto a HiPrep 16/60 Sephacryl S-100 HR column (GE Healthcare) equilibrated in TBS buffer (0.1 M Tris-HCl (pH 7.4), 0.2 M NaCl). Fractions containing monomeric, purified protein were identified by SDS-PAGE analysis, pooled, and centrifugally concentrated (molecular mass cutoff ϳ10 kDa) to an appropriate volume and exchanged into a low ionic strength buffer (25 mM Tris-HCl (pH 8), 50 mM NaCl). Protein concentration was determined by ultraviolet absorption at 280 nm using appropriate estimated extinction coefficients (36). For 31 P NMR assays, concentrated protein was stored in 50% glycerol at Ϫ20°C until use. For crystallization trials, concentrated protein was either used immediately, or flash-frozen and stored at Ϫ80°C. All enzymes could be heterologously expressed in Escherichia coli systems and Ͼ95% enzyme purity was achieved by nickel affinity purification and size exclusion chromatography. 31 P NMR Enzymatic Assays-All one-dimensional 31 P NMR spectra were recorded at 310 K on a Bruker DRX-500 spectrometer equipped with a BBO-500 MHz S2 5-mm probe with a Z gradient. Spectra were acquired at a spectrometer frequency of 202.11 MHz with 1 H decoupling. Data points (32 K) were acquired per spectrum at a spectral width of 80 ppm, with signal averaging over 99 scans. Spectra were obtained with a 60°pulse width and a 7-s relaxation delay for quantitative signal integration (15). Samples for initial assays on short chain lipid substrates contained either 2 mM 6:0 SM solubilized with 2 mM Triton X-100 or 80 mM 8:0 LPC, in 100 mM Tris-HCl (pH 7.4) containing 10 mM MgCl 2 , 1 mM trimethyl phosphate as an internal chemical shift and concentration standard, and 10% D 2 O (15). Samples for assays on long-chain lipids were similar except that they contained 2 mM phospholipid substrate solubilized in 50 mg/ml CHAPS detergent. Each sample contained a volume of 0.6 ml in a 5-mm thin wall NMR sample tube. Following an initial spectrum obtained in the absence of enzyme, 10 g of pure recombinant enzyme was added to the sample. After a delay of 5 min to allow for magnet shimming, five consecutive spectra were obtained at 15-min intervals. Samples were then stored at 37°C for 24 h, and an NMR spectrum was obtained at that time. Data were processed and analyzed with MestReNova 7.1.1 (Mestrelab Research, Santiago de Compostela, Spain). A baseline correction was applied to the spectra, and quantitative peak integration was performed with the autodetect function in MestReNova. All integrated peaks were normalized to the 1 mM trimethyl phosphate standard. For hexanoyl SM (6:0 SM) and octanoyl LPC (8:0 LPC), percent loss of substrate at each time point was calculated by dividing the normalized integrated peak area of substrate resonance by the initial substrate signal. Where no product signal was observed, a zero was assigned to indicate that substrate to product turnover was not detected in our assay. For the CHAPS assay, product yield for each time point was calculated by dividing the normalized integrated peak area of product resonance by the initial substrate signal. Substrate and product resonances were identified based on analogous results in a previous study (15).
Crystallization Trials of St_␤IB1i-Initial crystallization screens were carried out with a Phoenix robotic liquid dispenser (Art Robbins Instruments, Sunnyvale, CA) in a 96-well Intelli-Plate using the Index Crystallization Screen (Hampton Research, Aliso Viejo, CA). Each sitting drop in the screen contained a mixture of purified St_␤IB1i and precipitant solution at a ratio of 1:2 or 2:1 in a total drop volume of 200 nl. Tetragonal rods appeared within 24 -48 h in 0.2 M magnesium chloride hexahydrate, 0.1 M BisTris-HCl (pH 5.5), 25% w/v PEG-3350. Crystal growth was optimized in 24-well plates using the hanging drop method with 2 l of 3 mg/ml purified St_␤IB1i to 2 l of precipitant containing between 10 and 25% PEG 3350 buffered with 0.1 M BisTris (pH 5.5) and 0.2 M magnesium chloride hexahydrate. Thin rectangular crystals grew in 6 -8 days from precipitant solution containing 14% PEG 3350. Crystals were crushed and added to decreasing concentrations of PEG 3350 for seeding trials. Diffraction quality crystals of St_␤IB1i grew in a precipitant solution containing 12% PEG 3350, 0.1 M Bis-Tris (pH 5.5), 0.2 M magnesium chloride hexahydrate. Large single crystals (0.2 ϫ 0.3 ϫ 0.4 mm) were flash frozen using 40% PEG 3350 as a cryoprotectant.
Diffraction and Model Building of St_␤IB1i-Diffraction data for crystals of St_␤IB1i were collected remotely on beam line 7-1 at the Standard Synchrotron Radiation Laboratory using Blu-Ice software (37). The phase problem was solved with molecular replacement using the programs Chainsaw and MR phaser of the CCP4 suite (38). The crystal structure of a homologous toxin from L. intermedia (PDB code 3RLH, chain A) was used as a starting model. Manual model building and modification were performed using COOT (39). Bulk solvent correction and standard refinement with Refmac5 (40) yielded an R crys of 18% and an R free of 24%. Optimum TLS groups were obtained from the TLS motion determination server (41,42). Refinement including TLS groups yielded a new R crys of 16% and a new R free of 20%. Coordinates have been deposited in the RCSB Protein Data Bank under PDB code 4Q6X.
In Silico Docking Studies-In silico docking of substrates to StSicTox-␤IB1i was performed using Autodock Vina (43). PDBQT files for StSicTox-␤IB1i receptor and substrate ligands were prepared using UCSF Chimera 1.8.1 (44) and/or AutoDockTools-1.5.6 in the MGLTools package. St_␤IB1i con-tains two active site histidine residues, His-12 and His-47. In one set of runs, His-12 was deprotonated, whereas His-47 was protonated; this is the starting configuration in the mechanism proposed by Shi et al. (45) for GDPD enzymes. In a second set of runs, the opposite configuration was used. For docking of LPE and LPC substrates, the dihedral angles for the phosphoglycerol moiety of the ligand were restricted to a conformation in which the 2-hydroxyl group was poised to attack the phosphorus atom, leading to CPA formation. This conformation was based on the observed geometry of sn-glycerol-3-phosphate found in the active site of a glycerol phosphodiester phosphodiesterase from Oleispira antarctica (PDB code 3QVQ). Substrate molecules were docked into a 30 ϫ 30 ϫ 30 Å box centered on the magnesium atom in the active site of St_␤IB1i. The charge on magnesium was manually set at ϩ2. The binding mode search was conducted at maximal exhaustiveness setting. For each ligand in each protonation state of the enzyme, 30 -60 binding modes were calculated, depending upon the strength of structural consensus observed among the lowest five modes in a smaller initial test calculation (strong for LPE; less strong for LPC). In the docking of LPE, the lowest overall energy binding modes calculated for the two histidine protonation states were practically identical. For LPC, however, docking to the His-47protonated version of St_␤IB1i yielded a highly improbable lowest energy mode in which the phosphate group was ϳ10 Å from the magnesium center. Results were visualized using UCSF Chimera.

Enzymatic Activity and Substrate Preference of Three Diverse
SicTox Proteins-We initially tested the three selected SicTox homologs for PLD activity on hexanoyl SM/Triton X-100 and octanoyl-LPC micelles using previously described 31 P NMR assays (15). These short chain substrates give small micelles and thus minimize resonance linewidths (46). Both enzymes from L. arizonica (La_␣IB2bi and La_␤ID1) showed high activity against both SM and LPC, whereas St_␤IB1i showed much lower activity (Fig. 3A). The L. arizonica enzymes consumed all the SM within the dead time of the experiment (ϳ10 min), whereas more than 50% of this substrate remained after ϳ80 min of incubation with the S. terrosus enzyme. Similar differential activity was also observed with 8:0 LPC as substrate (Fig.  3A). Both L. arizonica enzymes also consumed substantial amounts of 8:0 LPC after 80 min (initial rate of ϳ0.24 mM/min for La_␣IB2bi and 0.8 mM/min for La_␤ID1), whereas St_␤IB1i showed no activity against 8:0 LPC even after 24 h. These initial experiments demonstrate that the homologs have variable PLD activity against SM and LPC and are consistent with other studies showing reduced SMase D activity among some ␤ clade members (Fig. 1).
We then tested the three enzymes against a wider panel of commercially available long-chain lysolipid and phosphosphingolipid substrates with various headgroups. In addition to LPC, the panel included the lysolipids LPE, LPG, and LPS, which are analogous to LPC but contain ethanolamine, glycerol, and serine headgroups, respectively, in place of choline (see Scheme 1). In addition to SM, the panel included the phosphosphingolipid ceramide phosphoethanolamine (CPE), which differs from SM only in replacement of the choline headgroup with ethanolamine (see Scheme 1). The ethanolamine-containing substrates LPE and CPE were particularly interesting to us as possible alternative substrates to LPC and SM, because ethanolamine is a common headgroup in insect lipids (47)(48)(49)(50). We included LPG and LPS as examples of lipids with neutral/zwitterionic headgroups. These lipids have been previously reported as substrates for Loxosceles reclusa protein LrSicTox-␣IB1 (19).
For our panel comparison (Fig. 3, B-D), we utilized 31 P NMR in mixed micelles with the nondenaturing detergent CHAPS (51). CHAPS effectively solubilizes the long-chain lipids and has precedent in phospholipase activity studies (52). The CHAPS NMR assay allows comparison of diverse long-chain lipid substrates at a common concentration of ϳ2 mM and allows direct detection of sharp, well resolved substrate and product resonances in all cases. It is less well suited to extraction of detailed primary kinetic parameters, partly because of the low sensitivity of NMR (53) and because such measurements are frequently problematic for interfacial enzymes in substrate/detergent mixtures (54). We regard these strengths and limitations as an acceptable compromise in an initial panel study of substrate preference across multiple SicTox family members. We thus characterize apparent substrate preferences (Table 1) by comparing initial reaction rates at a single initial concentration of substrate, both in single-substrate assays (Fig.  3) and in competition assays of selected substrate pairs (Fig. 4).
The substrate preferences of the three enzymes measured from single-substrate assays have two common features (Fig. 3 and Table 1). First, all three enzymes show activity against lysophospholipids with positively charged headgroups (LPC and/or LPE) but show no detectable activity against neutral/zwitterionic headgroups (LPG and LPS). Second, the enzymes show more rapid turnover of ceramide-based substrates (SM or CPE) versus the corresponding lysophospholipids (LPC or LPE, respectively). This latter preference is marginal in the case of La_␤ID1 (ϳ2-fold difference in measured rate) but increases to an order of magnitude or more for St_␤IB1i and La_␣IB2bi.
The enzymes have strikingly different substrate preferences for choline and ethanolamine, the two positively charged head-FIGURE 3. 31 P NMR assay results of three different SicTox PLD toxins with different phospholipid substrates. A, preliminary assessment of activity against SM and LPC using 10 g of pure enzyme with either 2 mM hexanoyl sphingomyelin (6:0 SM) and 2 mM Triton X-100, or 80 mM octanoyl lysophosphatidylcholine (8:0 LPC). St_␤IB1i shows much lower activity toward the two substrates than La_␣IB2bi and La_␤ID1, (B-D). Panel assays using 10 g of pure enzyme with ϳ2 mM each of six different phospholipid substrates solubilized in CHAPS detergent. Fitted slopes of product accumulation during the linear portion of the reaction were used to extract initial reaction rates. Error bars have been added to points used for slope determination; these are based on statistical analysis of instrument precision and represent standard deviations. In some cases, no product was observed after 75 min but did appear after 24 h. Substrate conversion and product yield were derived from integration of NMR signals relative to a trimethyl phosphate internal standard, as described under "Experimental Procedures."

TABLE 1 Initial reaction rates (M/min) of SicTox PLD toxins with various phospholipid substrates
Reported rates are slopes of product accumulation in single substrate assays (see Fig.  3). Rates in parentheses derive from direct competition assays (SM/CPE or LPC/ LPE) based on disappearance of the substrate signal (see Fig. 4). The two bottom rows represent rate ratios for SM and CPE or LPC and LPE, based either on comparison of the rates in single substrate assays or relative rates in a competition assay (values in parentheses). Note that values derived from single substrate and competition assays are similar. All rates were directly measured at 10 g of enzyme and ϳ2 mM initial substrate concentration, with the exception of St_␤IB1i with CPE. In this case, the rate was too fast to be measured, so a rate was measured at 1 g of enzyme and multiplied by 10. In cases where the reported rate is Ͻ1 M/min, no substrate decay was evident after ϳ75 min (and in the case of the single substrate assays, no product formation was evident either). For these cases, the rates were estimated from the 24-h time points, unless no product was seen even at 24 h, in which case the rate was reported as Ͻ0.1 M/min (see Fig. 3). groups. La_␤ID1 shows very little discrimination between choline and ethanolamine. Meanwhile, La_␣IB2bi is specific for choline and St_␤IB1i is specific for ethanolamine. La_␣IB2bi turns over SM faster than CPE by an estimated 1 to 2 orders of magnitude at ϳ2 mM initial substrate concentration (Table 1). La_␣IB2bi also prefers LPC over LPE, although neither substrate is turned over rapidly in this assay. St_␤IB1i, however, turns over CPE and LPE at least 3 and 2 orders of magnitude faster than SM and LPC, respectively. The CPE/SM comparison is particularly striking; all CPE substrate is gone within the first 10 min, while no turnover of SM is evident up to 80 min (Fig. 3D).

Phospholipid
Direct competition experiments between substrates with choline and ethanolamine headgroups show similar substrate preferences to those seen in single-substrate assays (Fig. 4 and Table 1). For each enzyme, we set up two competition experiments using either phosphosphingolipids (CPE versus SM) or lysolipids (LPC versus LPE). The most notable result is that we continue to observe rapid turnover of CPE, and extremely slow turnover of SM, by St_␤IB1i in direct competition experiments. In sum, our experiments show for the first time that these venom toxins have variable substrate specificity and reveal that toxins with low sphingomyelinase activity, such as St_␤IB1i, may be highly specific for ethanolaminecontaining phospholipids.
Crystal Structure of St_␤IB1i-To begin exploring the basis of differences in substrate specificity, we solved the structure of SicTox protein ␤IB1i from S. terrosus using crystallographic methods ( Table 2). Refinement of the St_␤IB1i structure converged to a crystallographic residual of 17% (R free ϭ 22%) for all FIGURE 4. Stacked 31 P NMR spectra from competition assays of three SicTox PLD enzymes with choline-and ethanolamine-containing substrates. Left panels show SM/CPE competition experiments, and right panels show LPC/LPE competition experiments. The spectra illustrate the preference, or lack thereof, of the different enzymes for choline (SM and LPC) or ethanolamine (CPE and LPE) headgroups. Assays contained 10 g of each SicTox PLD and ϳ2 mM of each substrate. In the SM/CPE assay, cleavage of either substrate yields cyclic ceramide phosphate (CCP), a six-membered ring phosphate with an upfield 31 P resonance. In the LPC/LPE assay, cleavage of either substrate yields cyclic phosphatidic acid (CPA), a five-membered ring phosphate with a far downfield 31 P resonance. data between 38.5 and 2.14 Å. The refined model of St_␤IB1i contains 278 amino acid residues, 122 solvent water molecules, and 1 Mg 2ϩ ion in the active site. Model quality was assessed with SFCHECK (55), and all stereochemical parameters are within acceptable ranges. A Ramachandran diagram of the model generated with RAMPAGE (56) indicates that all mainchain dihedral angles are within the most favored and additionally allowed regions.
St_␤IB1i contains two disulfide bonds as follows: one between Cys-51 and Cys-57, part of a hairpin turn in the catalytic loop, and one between Cys-53 and Cys-197, linking the catalytic loop to a second flexible loop region (27). This disulfide bonding pattern, along with low SMase D activity, places St_␤IB1i in the class IIb group (27,28). The flexible, catalytic, and variable loop regions of St_␤IB1i all exhibit large temperature factors, as also seen in both LiRecDT1 and SMase I. These apparently dynamic regions are likely to be involved in substrate recognition and formation of the membrane interface (2, 27, 28, 61). Modeling of these highly flexible regions in St_␤IB1i required TLS refinement (40, 41, 62).
The active site of St_␤IB1i contains all the residues previously identified as essential for catalytic activity (Figs. 2 and 5B) (2,27,28,59). With the exception of Asp-52, which is in the highly flexible loop region of the molecule (Fig. 5B), the catalytic side chains of St_␤IB1i also superimpose closely with those of LiRecDT1 and SMase I. The Mg 2ϩ cofactor is coordinated by six groups: the OD1 atom of Asp-33 (Asp-34 in SMase I numbering); the OE2 atom of Glu-31 (Glu-32 in SMase I numbering); the OD2 atom of Asp-91, and three water molecules. Thus, the inner hydration shell of the divalent magnesium is observed in the St_␤IB1i structure (63). The active site of St_␤IB1i does not show clear electron density for any ligand other than Mg 2ϩ ; all residual electron density in the active site was attributed to water molecules.
Docking and Mutational Studies-To examine possible substrate binding modes, we docked LPE and LPC ligands into the S. terrosus ␤IB1i active site (Fig. 6) using Autodock Vina (see under "Experimental Procedures"). We focused on lysophospholipid substrates rather than sphingolipids because of their simpler structure, and because other structural data yielded clues to the binding mode for lysophospholipids. Specifically, the structure of a bacterial GDPD from O. antarctica (PDB code 3QVQ) contains a glycerol 3-phosphate bound in a mode where the 2-hydroxyl is poised to attack the phosphorus center and make a five-membered ring (see Fig. 8, below). GDPDs are distantly related to the SicTox proteins (27,64,65), and their mechanism has been proposed to proceed through a five-membered cyclic phosphate intermediate (45) analogous to the product of the toxin action on lysophospholipids (15). We felt that the conformation of the glycerol phosphate in the GDPD active site was a reasonable model for the phosphoglycerol portion of LPC or LPE bound to the toxin in a catalytically competent mode. Consequently, we simplified our docking calculations by restricting the dihedral angles of the phosphoglycerol moiety of LPE and LPC to mimic the glycerol phosphate structure. The lowest energy binding modes for LPE and LPC are highly similar (Fig. 6A) and suggest that the headgroup is accommodated within a pocket bordered by Val-89, Asp-91, Asn-95, Ser-132, Glu-134, Asp-165, Ser-167, Gly-195, Tyr-228, Trp-230, and Met-251 (Fig. 6B). Part of this region (residues 89, 132, 134, 165, 167 and 195) has previously been named the "distal binding pocket" and is suggested to be important for recognition of different substrates (24). This pocket is quite well conserved between St_␤IB1i and the ␣ SMase D toxins of known structure. Only residues 134 (Glu versus Pro) and 195 (Gly versus Ser) differ (Figs. 6B and 2). In the lowest energy binding modes for LPE and LPC, neither position directly contacts the headgroup. Thus, the docking studies do not generate a strong sense that the active sites of these proteins should preferentially bind more strongly to choline than to ethanolamine headgroups or vice versa.
Nonetheless, we tested a variant of St_␤IB1i containing the substitutions E134P and G195S. In a competition experiment between SM and CPE, the variant does have measurable activity against SM during the 1st h, unlike the wild type, but it still turns over CPE much faster (Fig. 7). We also see no evidence that the variant favors LPC over LPE (data not shown). Interestingly, the rate of SM consumption in the SM/CPE competi- where ͗I(hkl)͘ is the mean intensity of all symmetry-related reflections I i (hkl). c R cryst ϭ (⌺͉F obs Ϫ F calc ͉)/⌺F obs . R free as for R cryst , using a random subset of the data (5%) not included in the refinement. d r.m.s.d., is root mean square deviation. tion slows markedly after the 1st h (44% SM consumed after 24 h) and is also very slow in an experiment with pure SM (data not shown). This suggests not only that activity of the variant against SM is limited, but also that it may depend on the presence of CPE.
Asn-95, a residue adjacent to Glu-134 (Fig. 6B), has also been proposed (along with Glu 134) to help exclude choline-containing substrates from the active site of class IIb enzymes (27). This residue is conserved as Gly in the ␣ clade toxins (see Fig. 2). To test the importance of these residues, we engineered a E134P/ N95G variant of St_␤IB1i. The variant showed rapid consumption of CPE and no measurable activity against SM in a competition experiment (Fig. 7). Because the rates are outside the dynamic range of the experiment for both the variant and the wild type, we cannot exclude some mutational effect on the relative rates of turnover. We conservatively concluded that site-directed mutations in or near the putative headgroup bind-ing region may have a limited impact on substrate specificity, but they do not alter it qualitatively.
Mechanism of SicTox PLD Activity-Although the above analysis of St_␤IB1i does not explain its substrate specificity, it does provide an opportunity to reconsider the general mechanism of the SicTox enzymes. The currently accepted mechanism (2,27,59) is based on the mechanisms of two other phosphodiesterases, a bacterial SMase C (66) and hydrolytic PLD enzymes with a conserved HKD motif (67). The first step is proposed to involve nucleophilic attack by a conserved histidine at the scissile phosphodiester bond, followed by displacement of the headgroup, leading to formation of a pentacoordinated covalent intermediate. In the second step, deprotonation of water forms a nucleophile, which attacks the covalent adduct to release a monoester product. This mechanism must be reexamined in light of our recent discovery that SicTox enzymes catalyze cyclic phosphate formation rather than hydrolysis (15), combined with the lack of an evolutionary relationship between SicTox and HKD PLD enzymes (67).
One possibility is that the SicTox mechanism is homologous to the first step of the mechanism of GDPD enzymes, which are distantly related (27,64,65). GDPDs hydrolyze phosphoglycerol substrates and have been proposed to form a cyclic intermediate analogous to the lipid product of the action of SicTox enzymes on lysophospholipids (45). In the proposed cyclization step, one histidine (His-17) acts as a general base, accepting a proton from the 2Ј-hydroxyl of the glycerol moiety, generating a nucleophile; a second histidine (His-59) acts as a general acid, donating a proton to the leaving group. This mechanism is supported by both the conformation and positioning of glycerol 3-phosphate product bound to O. antarctica GDPD (PDB code 3QVQ).
A reasonable GDPD-like mechanism for cyclization of LPC by the SicTox proteins is shown in Fig. 8C. To visualize the probable GDPD cyclization step in the context of a SicTox protein, we superimposed the O. antarctica and S. terrosus structures to position the phosphoglycerol moiety in the active site of St_␤IB1i (Fig. 8, A and B). The superposition yields a reasonable binding mode for the phosphoglycerol moiety of LPC or LPE, with the two  Fig. 2). The essential magnesium ion cofactor from the structure of St_␤1B1i is shown as a small red sphere, along with a sulfate ion from the L. laeta SMase I structure. With the exception of Asp-52, in the flexible/variable catalytic loop, all the amino acid residues have a similar location and conformation in St_␤1B1i, a ␤ clade member, as in the two ␣ clade representatives. Residue numbering is referenced to Ll_␣III1/SMase I. The magnesium cofactor is shown as a small green sphere interacting with the phosphate group of each substrate. For both substrates, the headgroup is situated in a binding pocket that has been previously identified as potentially important for substrate specificity (24). B, side chains in the putative headgroup binding pocket of St_␤IB1i (tan; PDB code 4Q6X) compared with those of LiRecDT1, an ␣ clade protein from L. intermedia (green; PDB code 3RLH). The two pockets are quite similar but contain amino acid sequence differences at positions 95, 134, and 195 (see also amino acid residues in blue boxes in Fig. 2). Residue numbering is referenced to Ll_␣III1/SMase I. active site histidines (His-12 and His-47) nicely positioned for general base and acid catalysis, respectively.
Our in silico docking of full-length lipid substrates suggests a related but different mechanism in which the orientation of the phosphoglycerol moiety is reversed (Fig. 8, D-F). In the GDPDlike mechanism, the orientation of the phosphoglycerol moiety implies that the lipid chain of LPC or LPE would face deep into the binding pocket, although the headgroup would protrude from the active site (Fig. 8, A-C). The docking studies imply essentially the opposite positioning of lipid chain and headgroup. One advantage of this configuration is that the lipid chain may not easily fit into the deep part of the binding pocket in the GDPD-like mechanism. The reversed orientation also leads to a swapping of the roles of the two histidines as follows: His-47 is now in a position to activate the nucleophile, whereas His-12 can protonate the leaving group. His-47 could also be assisted by the adjacent Asp-52 residue (Fig. 8E). A final difference is that in the GDPD-like mechanism, the nucleophilic hydroxyl occupies one coordination site of magnesium; in the docking-based mechanism, the oxygen of the leaving group occupies this position. In either case, the magnesium ion could accelerate the reaction by stabilizing a developing negative charge.

DISCUSSION
We have biochemically characterized three diverse SicTox PLD toxins to compare substrate preference and ambiguity among venom-expressed enzymes in the SicTox gene family. The results demonstrate for the first time that a wide range of preference exists for choline versus ethanolamine headgroups. Most strikingly, we found that St_␤IB1i, a class IIb toxin with low PLD activity against SM and LPC, is actually a highly active enzyme with strong specificity for ethanolamine-containing substrates. We determined the structure of St_␤IB1i, the first of a class IIb enzyme as well as the first structure of a SicTox PLD from a Sicarius species. The structure led to new insights into substrate binding mode and mechanism, but it did not immediately reveal the basis for differences in specificity. Nevertheless, our findings show that SicTox proteins are functionally diverse, which may in turn help to explain their variable mammalian toxicity (7-9, 25, 26).
The specificity of St_␤IB1i for ethanolamine headgroups may explain why other class IIb enzymes tend to show diminished sphingomyelinase activity compared with whole venoms. St_␤IB1i shares over 60% sequence identity with all previously reported class IIb enzymes (Figs. 1 and 2), and phylogenetic analysis groups these toxins within the ␤I clade of the SicTox gene family (24). It is therefore plausible that many ␤I SicTox PLD toxins have evolved specificity for ethanolamine-containing targets such as CPE.
Our test set also included two paralogs, La_␣IB2bi (15,30) and La_␤ID1, from the venom of L. arizonica. La_␣IB2bi, an ␣ clade member, strongly preferred choline-containing over ethanolamine-containing substrates. The preference of La_␣IB2bi for LPC over LPE resembles the specificity previously observed for another ␣ clade toxin, ␣IA1 from L. reclusa (19). Surprisingly, La_␤ID1 also slightly preferred choline-containing substrates, despite belonging to the ␤I clade like St_␤IB1i. The observation of robust SMase D activity in La_␤ID1 is intriguing given that it shares 84% sequence identity with LiRecDT5 (␤ID1) from L. intermedia, an enzyme with very limited SMase D activity compared with ␣ clade toxins from the same species (9). Direct experimental comparison of La_␤ID1 and LiRecDT5 could further illuminate whether these two enzymes differ in SMase D activity and/or substrate specificity. Functional differences between these two proteins would suggest that the ␤ID clade is functionally diverse and represents a transition toward ethanolamine specificity in the ␤I-ABC clades, beginning from either low specificity or a preference for SM in the ancestral sequences (see Fig. 1). They would also potentially lead to identification of limited sequence changes that modulate function.
At present, the determinants of substrate specificity remain a mystery. The structure of St_␤IB1i, including the active site, is quite similar to the previously determined structures of a class I enzyme (PDB code 1XX1) and a class IIa enzyme (PDB code 3RLH). As a class IIb enzyme, St_␤IB1i is most similar to the latter, as all class II enzymes contain two disulfide bonds. The largest structural variations were found in the flexible loop regions, as observed previously in comparisons of class I and class IIa structures (2,28). Docking experiments with LPC and LPE substrates and St_␤IB1i suggested a location for the headgroup recognition pocket deep in the active site, away from variable loops. The sequence and structure of this pocket are largely conserved between the different enzymes, and both substrates docked in a similar mode. Mutation of variable residues in this region of St_␤IB1 led to at most minor alterations in its specificity. If these enzymes use active site interactions to discriminate between substrates, it is not obvious how they do so.
Because PLDs are interfacial catalysts (67), an alternative source of specificity might involve interactions between the i-face of the enzyme (68) and a lipid interface. The i-face of the PLD toxins has not been identified experimentally but has been proposed to consist of the catalytic loop (visible in the lower left corner of Fig. 5B) and the loops following the fifth and sixth FIGURE 7. 31 P NMR competition assays of St_␤IB1i variants against SM and CPE. The data for wild-type St_␤IB1i (black) derive from the spectra shown in Fig. 4 and the N95G/E134P (blue) and E134P/G195S (red) variants that were assayed in the same manner. All three variants turn over CPE considerably faster than SM, although E134P/G195S may exhibit a decreased preference for CPE. See "Experimental Procedures" for details of assay conditions. Error bars are based on statistical analysis of instrument precision and represent standard deviations.
strands of the barrel (27). These regions could be targets for future mutational studies. Ethanolamine and choline could present very different interaction surfaces to the i-face of an interfacial catalyst. Although both are positively charged headgroups, they differ in hydrogen bonding capability, hydrophobicity, and size. For example, unlike their choline counterparts, ethanolamine-containing phospholipids can form a hydrogenbonded ring structure between the positively charged amine group and a negatively charged phosphate oxygen (69).
Substrate preferences could result from changes in kinetics and thermodynamics of the enzyme-lipid interactions at the i-face, including tighter binding and/or longer residence times for processive catalysis. They may also result from allosteric coupling between the i-face and the active site, which can affect substrate binding as well as catalytic steps (70). Because of these complexities, apparent substrate preferences may themselves depend somewhat on the form of the lipid interface, e.g. micelles versus vesicles, present in the assay. This is a caveat in our study, which uses detergent-mixed micelles; we note, however, that the trends we observed were similar in most single substrate and mixed substrate assays and that the enzymes showed similar trends in SMase/LPCase activity in different micelle mixtures (Figs. 3 and 4 and Table 1).
This study suggests generally stronger substrate specificity for the SicTox enzymes in comparison with previous studies. Specifically, the three PLD toxins showed moderate to strong FIGURE 8. Two plausible mechanisms of LPC cyclization by SicTox enzymes, based on two possible substrate binding modes. A, active site pocket of St_␤IB1i with glycerol 3-phosphate bound as seen in a GDPD enzyme from O. antarctica (PDB code 3QVQ). B, this binding mode places the nucleophilic 2Ј-OH of the glycerol moiety adjacent to His-12. The putative leaving group position is not evident, as glycerol 3-phosphate is the product of removal of the leaving group by a GDPD enzyme. However, the phosphate oxygen opposite the nucleophile (see arrow) is likely to represent the leaving group oxygen of substrate. C, mechanism for LPC/LPE cyclization derived from the glycerol 3-phosphate binding mode. His-12 deprotonates the 2Ј-hydroxyl generating the nucleophile for attack on the phosphate. His-47 acts as a general acid to protonate the leaving group. D, active site pocket of St_␤IB1i with LPE docked as in Fig. 6. E, this binding mode inverts the orientation of the glycerol phosphate moiety relative to that shown for glycerol 3-phosphate in A-C, and places the 2Ј-OH of glycerol adjacent to His-47. F, mechanism for LPC/LPE cyclization based on the LPC/LPE docking. His-47 activates the 2Ј-OH for in-line attack on the phosphodiester. His-12 protonates the leaving group.
preference for sphingolipid over lysolipid substrates, as well as a very strong preference for positively charged over neutral headgroups. By contrast, for an ␣ clade enzyme from L. reclusa, Lee and Lynch (19) reported very similar kinetics for sphingolipid (SM) and lysolipid (LPC) substrates, and observed strong activity against LPG and LPS. van Meeteren et al. (18) also reported similar levels of turnover of SM and LPC for an ␣ clade enzyme from Loxosceles. laeta. These different findings may reflect variable substrate preference within the SicTox family. Indeed, one of our three proteins, La_␤ID1, shows considerably lower preference forsphingolipids,andlessdiscriminationbetweencholineandethanolamine, than the other two. The general preference for sphingolipids found in our study, however, may have biological significance; sphingolipid substrates exist at higher concentrations than lysolipids in the lipidome of insect prey (49,50).
Discrimination between choline and ethanolamine headgroups may also be biologically significant for both mammalian toxicity and predation. CPE is a major phosphosphingolipid in insects (47), the main prey item of the sicariid spiders (30); SM, by contrast, dominates in mammals (71,72). As noted above, some ␤I clade members show diminished toxicity in mammalian models (7-10, 25, 26). Our work suggests that this reduced toxicity may relate to specificity for ethanolamine-containing lipids like CPE. Given the specificity of St_␤IB1i and the paucity of CPE in humans, we also propose that St_␤IB1i, or similar toxins, have potential as safe insecticides if utilized in a transgenic system (73).
As for predation, lipidome and genetic analyses suggest that most arthropods can produce both CPE and SM, but some insects (e.g. the Brachycera) may produce only CPE (47). The diversification of headgroup specificity in the SicTox gene family may allow some spider species to tailor venom activity specifically toward prey species in which either SM or CPE is the predominant phosphosphingolipid. A question for future research is whether variation in whole venom specificity for SM or CPE correlates with the levels of each lipid in typical prey.
Purified La_␣IB2bi acts as a paralytic neurotoxin in crickets (30), and this may be one general function of the SicTox proteins, whether they are specific for SM or CPE or act on both. CPE is concentrated in neural tissue of certain insects, comprising about 4% of the total lipid content of brain tissue (49) in Drosophila and about 6 -7% of total lipid content of the nervous system in Musca domestica (50). In Drosophila, CPE is required by glial cells to properly insulate axons, and lack of CPE leads to neuropathy (48). Another future research question is whether the neurotoxic effects of isolated SicTox proteins on different prey will depend on a match between enzyme specificity and the content of SM and/or CPE in neurological tissue.