Structural Insights into the Cytotoxic Mechanism of Vibrio parahaemolyticus PirAvp and PirBvp Toxins

In aquaculture, shrimp farming is a popular field. The benefits of shrimp farming include a relatively short grow-out time, high sale price, and good cost recovery. However, outbreaks of serious diseases inflict serious losses, and acute hepatopancreatic necrosis disease (AHPND) is an emerging challenge to this industry. In South American white shrimp (Penaeus vannamei) and grass shrimp (Penaeus monodon), this disease has a 70–100% mortality. The pathogenic agent of AHPND is a specific strain of Vibrio parahaemolyticus which contains PirAvp and PirBvp toxins encoded in the pVA1 plasmid. PirAvp and PirBvp have been shown to cause the typical histological symptoms of AHPND in infected shrimps, and in this review, we will focus on our structural understanding of these toxins. By analyzing their structures, a possible cytotoxic mechanism, as well as strategies for anti-AHPND drug design, is proposed.


Introduction
Acute hepatopancreatic necrosis disease (AHPND), which was originally known as early mortality syndrome (EMS), first broke out in China in 2009, then spreading to Vietnam, Malaysia and Thailand [1,2]. Because of this disease, shrimp production dropped to~60% compared with 2012, and total economic losses have been estimated at more than $1 billion per year, globally [3]. The causative agent of AHPND was soon found to be a specific strain of Vibrio parahaemolyticus. V. parahaemolyticus is a halophilic Gram-negative bacterium that is commonly found in estuarine, marine and coastal environments [4], and originally it was not known how this opportunistic bacterium had become virulent and capable of causing disease in shrimps.
In addition to the readily observable symptoms in infected P. monodon and P. vannamei-lethargy, an empty stomach and midgut, and a pale to white atrophied hepatopancreas [4]-histological examination of the diseased shrimp further showed that the HP tubule epithelial cells sloughed into the HP tubule lumens [4,5]. Meanwhile, in the initial, acute stage of AHPND, even when a large number of bacteria could be found in the stomach, there were still sometimes no obvious bacterial colonies in the hepatopancreas tube lumens [1,4,6]. This led Tran et al. to propose that the symptoms of AHPND were caused by a toxin secreted by the pathogen [4]. This proposal was further supported by reverse gavage experiments in which introduction of the bacteria-free supernatant of the bacterial culture into healthy shrimp induced typical AHPND symptoms [4,7]. Subsequent investigations focused on isolating AHPND variants [8] and on comparing the draft genome sequences of AHPND-causing versus non-AHPND-causing strains [1,[9][10][11][12]. By using a next-generation sequencing (NGS) platform to sequence and compare three virulent (3HP, 5HP and China) and one non-virulent (S02) V. parahaemolyticus strains [9], Yang et al. (2014) found that a 69-kb extrachromosomal plasmid was present in all AHPND-causing strains but not in the non-virulent strain. This plasmid was named pVA1, and sequence analysis showed that it contained homologs of the insecticidal Photorhabdus insect-related (Pir) binary toxin PirA/PirB [13]. The importance of these two toxins to AHPND was confirmed by subsequent studies [14][15][16], and they are now referred to as V. parahaemolyticus PirA/PirB (PirA vp /PirB vp ).

The Structural Similarity between V. parahaemolyticus PirA vp /PirB vp and Bacillus thuringiensis Cry Toxins
Photorhabdus PirA and PirB were first reported as potential toxins by genomic sequencing of the entomopathogenic bacterium Photorhabdus luminescens W14 [17], and in 2009, Waterfield et al. reported that both Photorhabdus PirA and PirB were necessary for the insecticidal activity against caterpillars of the moth Galleria mellonella [18]. Although sequence similarity had previously led to Photorhabdus PirB being initially identified as a juvenile hormone esterase-like (JHE-like) protein [19], Waterfield et al. found that Photorhabdus PirB did not have JHE activity [20], and another study further showed that it had sequence similarity to the pore-forming domain I of the B. thuringiensis Cry toxin [21]. However, although it was established that Photorhabdus PirA/PirB was an effective insecticidal binary toxin [18,[20][21][22], its cytotoxic mechanism remained unclear.
The first crystal structures to be reported for any PirA/PirB toxins were for V. parahaemolyticus PirA vp and PirB vp , and the accompanying structural analysis also suggested a relationship between B. thuringiensis Cry and PirA vp /PirB vp toxins [13]. Cry proteins are one of the B. thuringiensis insecticidal toxins, and they have an important potential use in agriculture [23,24]. Although Cry toxins can be divided into at least 75 primary subgroups, and can show differences in their amino acid sequences, the determined and predicted structures of almost all of the Cry toxins are similar [25]. Cry toxins have three functional domains: the pore-forming domain I, the receptor-binding domain II and the sugar-binding domain III [23][24][25][26][27][28][29]. The specificity and cytotoxic mechanisms of Cry toxins are mediated by these three domains, and they have been discussed in many review articles [23][24][25][28][29][30][31]. For example, B. thuringiensis Cry1A uses domains II and III to target receptors that are abundant in the midgut of insect larvae, such as alkaline phosphatase (ALP) or aminopeptidase N (APN). The concentrated Cry1A toxins then interact with another receptor, cadherin-like receptor (CAD), which facilitates the proteolytic cleavage of its domain I helix α1. This cleavage induces the formation of the Cry oligomer, which uses the activated domain I to form non-selective pores in the apical membrane. This causes colloidal osmotic lysis of the cells. Figure 1 shows the crystal structure of the PirA vp and PirB vp toxins. Figure 2 shows how PirB vp corresponds to Cry domains I and II, while PirA vp has similar topology to Cry domain III. These structural similarities suggest PirA vp /PirB vp binary toxin is a Cry-like, three-domain toxin, but with a dissociated domain III [13,27]. The following sections discuss this idea in more detail.

PirB vp Contains Both Cry-Like Pore-Forming and Receptor Domains
Both the N-terminal domain of PirB vp (PirB vp N) and Cry domain I contain a bundle of α-helices ( Figure 2A) [13]. Figure 2B further shows that there are abundant hydrophobic residues located in the center of the PirB vp N α-bundle, and that the hydrophobic α-helix 8 of PirB vp N is sheltered within a bundle of amphipathic α-helices. This "inside-out membrane fold" is consistent with other pore-forming toxins that can switch between soluble and transmembrane conformations [32]. After triggering conformational change, these hydrophobic residues become exposed on the surface of the protein, where they are able to interact with membrane lipids. It should also be noted that these helices are generally longer than 40 Å, which is sufficient to cross the cell membrane (the length of the lipid bilayer is ~40 Å). Similar features are seen in the pore-forming domain I of Cry toxin, as well as in other pore-forming toxins, like colicin [28,29,32,33]. All of this strongly suggests that PirB vp N has the ability to form a pore on the cell membrane that causes cell death. Meanwhile, the C-terminal domain of PirB vp (PirB vp C) has three antiparallel β-sheets arranged in a manner similar to that seen in Cry domain II ( Figure 2C) [13]. Since the Cry domain II contains an immunoglobulin-like folding that is involved in protein-protein or protein-ligand interactions [34], it seems likely that the PirB vp C domain plays a similar functional role. Further, since Cry domain II could interact with insect receptors [23][24][25][26][29][30][31], the structural similarity suggests PirB vp C is also a receptor binding domain.

PirB vp Contains Both Cry-Like Pore-Forming and Receptor Domains
Both the N-terminal domain of PirB vp (PirB vp N) and Cry domain I contain a bundle of α-helices ( Figure 2A) [13]. Figure 2B further shows that there are abundant hydrophobic residues located in the center of the PirB vp N α-bundle, and that the hydrophobic α-helix 8 of PirB vp N is sheltered within a bundle of amphipathic α-helices. This "inside-out membrane fold" is consistent with other pore-forming toxins that can switch between soluble and transmembrane conformations [32]. After triggering conformational change, these hydrophobic residues become exposed on the surface of the protein, where they are able to interact with membrane lipids. It should also be noted that these helices are generally longer than 40 Å, which is sufficient to cross the cell membrane (the length of the lipid bilayer is~40 Å). Similar features are seen in the pore-forming domain I of Cry toxin, as well as in other pore-forming toxins, like colicin [28,29,32,33]. All of this strongly suggests that PirB vp N has the ability to form a pore on the cell membrane that causes cell death. Meanwhile, the C-terminal domain of PirB vp (PirB vp C) has three antiparallel β-sheets arranged in a manner similar to that seen in Cry domain II ( Figure 2C) [13]. Since the Cry domain II contains an immunoglobulin-like folding that is involved in protein-protein or protein-ligand interactions [34], it seems likely that the PirB vp C domain plays a similar functional role. Further, since Cry domain II could interact with insect receptors [23][24][25][26][29][30][31], the structural similarity suggests PirB vp C is also a receptor binding domain.  [35]. Briefly, each atom of the residues and the compound was first assigned an atom type (e.g., donor or acceptor) and formal charge based on their physiochemical properties. The scoring function of iGEMDOCK was then used to measure intermolecular interactions between PirA vp and GalNAc. In this docking model, the oxygen heteroatom of GalNAc forms hydrogen bonds with residue Lys29. Residue Glu36 yields a hydrogen bond with one of GalNAc's hydroxyl groups. Gly38 is a non-polar residue that is sandwiched in close proximity to two hydroxyl groups. Residues Val37 and Arg84 interact with the compound via van der Waals forces. The PDB code 1CIY was used to produce the figures for the Cry toxin. GalNAc is shown docked into the structure of PirA vp using the docking tool iGEMDOCK [35]. Briefly, each atom of the residues and the compound was first assigned an atom type (e.g., donor or acceptor) and formal charge based on their physiochemical properties. The scoring function of iGEMDOCK was then used to measure intermolecular interactions between PirA vp and GalNAc. In this docking model, the oxygen heteroatom of GalNAc forms hydrogen bonds with residue Lys29. Residue Glu36 yields a hydrogen bond with one of GalNAc's hydroxyl groups. Gly38 is a non-polar residue that is sandwiched in close proximity to two hydroxyl groups. Residues Val37 and Arg84 interact with the compound via van der Waals forces. The PDB code 1CIY was used to produce the figures for the Cry toxin.

PirA vp Contains a Possible Sugar-Binding Pocket
Like PirB vp , the biological functions of PirA vp may also be revealed by its structural features. PirA vp contains two antiparallel β-sheets that are packed together in a jelly-roll topology [13]. This folding is similar to domain III of the Cry toxin ( Figure 2D). Cry domain III contains a galactose-binding domain-like fold [36,37]; this is thought to be related to the toxin's specificity via its recognition of receptor-bound N-acetylgalactosamine (GalNAc) [23][24][25][26][29][30][31][36][37][38]. In the interaction between Cry1Ac and APN, Cry1Ac domain III first interacts with the GalNAc sugar on the APN receptor to facilitate the subsequent toxin-receptor binding [23]. PirA vp does indeed play a similar role to Cry domain III, then it should facilitate target-specific recognition by binding to certain ligands on the cell membrane/receptor. Interestingly, a potential sugar-binding cavity formed by three loops was found in PirA vp ( Figure 2E). The docking model shows that when the GalNAc molecule was fitted into this cavity, it could potentially interact with the PirA vp residues Lys29, Glu36, Val37, Gly38 and Arg84 ( Figure 2E). We further note that, since the potential binding cavity of PirA vp is deep and narrow ( Figure 2E), it may be possible that PirA vp not only targets the monosaccharides like GalNAc, but also oligosaccharides.

Unanswered Questions Relating to the Cytotoxic Mechanism of PirA vp /PirB vp
We have shown that the PirA vp /PirB vp toxin has structures that are similar to the functional domains of Cry. This further suggests that PirA vp /PirB vp might also induce cell death via the respective Cry-like steps of receptor binding, oligomerization and pore forming. To explore this model, identification of the cell receptors that might interact with PirA vp /PirB vp is a logical place to start. We note that although the main folding of Cry domain II and PirB vp C is similar, the loop regions between these two domains are quite different ( Figure 2C). Since the loop á-8, loop 2 and loop 3 of Cry Domain II are very important to aminopeptidase N (APN)-, alkaline phosphatase (ALP)and cadherin (CAD)-receptor binding [26,[39][40][41], these divergent loop regions suggest either that the toxin-interacting regions on shrimp's APN, ALP and CAD receptors are different to those found in insects, or else that PirB vp targets different receptors on the shrimp cell's membrane. In either case, given that PirA vp /PirB vp toxin induces cell death in the shrimp's hepatopancreas, but not in the stomach or other organs, it seems very likely that these putative PirA vp /PirB vp receptors will be found exclusively in the hepatopancreas membrane. However, we caution that there is as yet no experimental evidence in support of this; at present, the structure of these shrimp receptors remains unknown. We also note that several other critical processes still need to be investigated experimentally. For example, we do not yet know whether the cleavage of N-terminal á-helices on PirB vp is important for toxin activation, or whether PirA vp /PirB vp forms an oligomer in order to make a pore in the membrane.
Determination of the binding ligand of PirA vp is also worth investigating. Although the binding model between PirA vp and GalNAc seems reasonable, this interaction still needs to be confirmed by experiments such as surface plasmon resonance. To explore more possibilities, a high-throughput screening of PirA vp bound ligands would be useful, and we note that a feasible chip platform designed for carbohydrate-protein interactions has recently been developed [42][43][44].
To become a true three-domain toxin, PirA vp and PirB vp must first form a complex. Although the complex formation of PirA vp /PirB vp was confirmed using gel filtration [13], the resulting structure is still unknown, so how these two toxins bind to each other is still unclear. Based on the locations of the corresponding domains in the Cry toxin, a possible binding model of PirA vp and PirB vp was proposed (Figure 3; [13]). In this model, á-helices 1, 2, 12 and 13, and loops 12 and 13 of PirB vp create a potential binding cavity for PirA vp , while the â-sheets 1, 3 and 9 of PirA vp interact with PirB vp (Figure 3A). Figure 3B shows how the surface charges on the PirA vp /PirB vp interface are complementary to each other, further suggesting that this model is reasonable. However, as noted above, this PirA vp /PirB vp binding model still needs to be verified experimentally.
Furthermore, although there are many structural similarities, some physiological characteristics between Cry and PirA vp /PirB vp toxins may be different. For example, the Cry protoxins generally form crystals in the mother cell compartment [45,46]. Since the crystals have to be solubilized in the gut of insect larvae to become biologically active, this ability of the protoxins to crystallize may decrease their susceptibility to premature proteolytic degradation [45]. Previous reports have shown that the solubility of these Cry crystals is dependent on pH [45,47,48]; the crystals that form in the neutral pH of the mother cells subsequently dissolve in the acidic environment (<pH 4) of the insect gut. However, unlike Cry toxins, there are no reports of in vivo crystal formation for PirA vp /PirB vp , and although in vivo crystallization of Cry toxins is an important control step of their toxicities, it seems unlikely that PirA vp /PirB vp would use a similar control mechanism. Nevertheless this has not yet been demonstrated experimentally.
A more complete understanding of the cytotoxic mechanisms of PirA vp /PirB vp toxins is likely to be important for AHPND research, but could also be important for agricultural applications. Although there is genetic distance between PirA vp /PirB vp and the PirA/PirB homologs that are found in other bacteria such as Photorhabdus asymbiotica (WP_015835800/WP_015835799) [18], Photorhabdus luminescens (ABE68878/ABE68879) [19], Xenorhabdus doucetiae (CDG18638/CDG18639), Xenorhabdus cabanillasii (CDL79383/CDL79384), Xenorhabdus nematophila (WP013183676/WP010845483) and Alcaligenes faecalis (WP003801867/WP003801865), these insecticidal PirA and PirB toxins have allowed Photorhabdus and Xenorhabdus to be used in biological pest control [18,21,22]. The study of PirA vp /PirB vp should also therefore provide useful information for insecticidal applications. be solubilized in the gut of insect larvae to become biologically active, this ability of the protoxins to crystallize may decrease their susceptibility to premature proteolytic degradation [45]. Previous reports have shown that the solubility of these Cry crystals is dependent on pH [45,47,48]; the crystals that form in the neutral pH of the mother cells subsequently dissolve in the acidic environment (<pH4) of the insect gut. However, unlike Cry toxins, there are no reports of in vivo crystal formation for PirA vp /PirB vp , and although in vivo crystallization of Cry toxins is an important control step of their toxicities, it seems unlikely that PirA vp /PirB vp would use a similar control mechanism. Nevertheless this has not yet been demonstrated experimentally. A more complete understanding of the cytotoxic mechanisms of PirA vp /PirB vp toxins is likely to be important for AHPND research, but could also be important for agricultural applications. Although there is genetic distance between PirA vp /PirB vp and the PirA/PirB homologs that are found in other bacteria such as Photorhabdus asymbiotica (WP_015835800/WP_015835799) [18], Photorhabdus luminescens (ABE68878/ABE68879) [19], Xenorhabdus doucetiae (CDG18638/CDG18639), Xenorhabdus cabanillasii (CDL79383/CDL79384), Xenorhabdus nematophila (WP013183676/WP010845483) and Alcaligenes faecalis (WP003801867/WP003801865), these insecticidal PirA and PirB toxins have allowed Photorhabdus and Xenorhabdus to be used in biological pest control [18,21,22]. The study of PirA vp /PirB vp should also therefore provide useful information for insecticidal applications.

Strategies for Designing Drugs to Block the Cytotoxic Effects of V. parahaemolyticus PirA vp and PirB vp Toxins
Although AHPND-detection methods that can monitor the shrimps and the environment during cultivation have already been developed [7,49,50], there are still no available drugs that can be used in the treatment of AHPND. It has already been clearly established that PirA vp and PirB vp toxins are the main cytotoxic source of AHPND; for example, the deletion/mutation of their pirA vp and pirB vp genes from pVA1 can decrease AHPND severity and reduce the mortality of the shrimps [12,13,15,16]. Additionally, PirA vp and PirB vp are both secreted proteins [13,15], which means that they could be easily targeted by drugs/inhibitors. Neutralization of PirA vp and/or PirB vp toxicity is

Strategies for Designing Drugs to Block the Cytotoxic Effects of V. parahaemolyticus PirA vp and PirB vp Toxins
Although AHPND-detection methods that can monitor the shrimps and the environment during cultivation have already been developed [7,49,50], there are still no available drugs that can be used in the treatment of AHPND. It has already been clearly established that PirA vp and PirB vp toxins are the main cytotoxic source of AHPND; for example, the deletion/mutation of their pirA vp and pirB vp genes from pVA1 can decrease AHPND severity and reduce the mortality of the shrimps [12,13,15,16]. Additionally, PirA vp and PirB vp are both secreted proteins [13,15], which means that they could be easily targeted by drugs/inhibitors. Neutralization of PirA vp and/or PirB vp toxicity is therefore a rational direction for AHPND drug design. Further, since the structures of PirA vp and PirB vp are both available, a structure-based drug design can be used to achieve this goal more efficiently.
Structure-based drug design has been successfully used before. For example, in the well-studied pore-forming toxins, such as colicin and hemolysin, structural biology provided a wealth of useful knowledge regarding conformation rearrangement, receptor/ligand binding regions and oligomerization [32]. Structural insights into toxins also enables the development of novel therapeutic strategies [32]. For example, small molecules or engineered antibodies can be designed to interact with specific sites on the toxins. In the case of Aeromonas hydrophila aerolysin, which targets glycosylphosphatidylinisotol (GPI)-anchored proteins, synthetic GPI molecules and GPI analogues have been proposed as inhibitors [51]. It has also recently been shown that Staphylococcus aureus hemolysin can be neutralized by an antibody that targets the receptor binding site of this toxin [52]. Similarly, with other pore-forming toxins, receptors that bind these toxins, such as CCR5 and ADAM10, can also be considered in a reverse strategy for drug design [53,54]. For example, Leukotoxin ED pore-forming toxin targets human CCR5 receptor, and CCR5 receptor antagonists such as maraviroc were shown to block Leukotoxin ED-induced cell death [53]. The structural characteristics of PirA vp and PirB vp suggests three regions that are potentially suitable for structure-based drug design: (1) the potential receptor-binding region of PirB vp ; (2) ligand-binding region of PirA vp and (3) the interacting region between PirA vp and PirB vp (Figure 4, Table 1). Interface information such as amino acid sequences and structural motifs can be used for antibody engineering, as well as for in silico compound screening.
Mar. Drugs 2017, 15, 373 7 of 12 therefore a rational direction for AHPND drug design. Further, since the structures of PirA vp and PirB vp are both available, a structure-based drug design can be used to achieve this goal more efficiently. Structure-based drug design has been successfully used before. For example, in the well-studied pore-forming toxins, such as colicin and hemolysin, structural biology provided a wealth of useful knowledge regarding conformation rearrangement, receptor/ligand binding regions and oligomerization [32]. Structural insights into toxins also enables the development of novel therapeutic strategies [32]. For example, small molecules or engineered antibodies can be designed to interact with specific sites on the toxins. In the case of Aeromonas hydrophila aerolysin, which targets glycosylphosphatidylinisotol (GPI)-anchored proteins, synthetic GPI molecules and GPI analogues have been proposed as inhibitors [51]. It has also recently been shown that Staphylococcus aureus hemolysin can be neutralized by an antibody that targets the receptor binding site of this toxin [52]. Similarly, with other pore-forming toxins, receptors that bind these toxins, such as CCR5 and ADAM10, can also be considered in a reverse strategy for drug design [53,54]. For example, Leukotoxin ED pore-forming toxin targets human CCR5 receptor, and CCR5 receptor antagonists such as maraviroc were shown to block Leukotoxin ED-induced cell death [53]. The structural characteristics of PirA vp and PirB vp suggests three regions that are potentially suitable for structure-based drug design: (1) the potential receptor-binding region of PirB vp ; (2) ligand-binding region of PirA vp and (3) the interacting region between PirA vp and PirB vp (Figure 4, Table 1). Interface information such as amino acid sequences and structural motifs can be used for antibody engineering, as well as for in silico compound screening. Although engineered antibodies can be used to investigate the importance of these various PirA vp and PirB vp regions in the laboratory, they are probably too expensive and difficult to use in the field. By contrast, small compounds may be more suitable for AHPND treatment in aquaculture. Although engineered antibodies can be used to investigate the importance of these various PirA vp and PirB vp regions in the laboratory, they are probably too expensive and difficult to use in the field. By contrast, small compounds may be more suitable for AHPND treatment in aquaculture. Recently, in silico screening approaches have been used to identify small molecules that disrupt protein-protein interactions [55,56], and currently, data is available for over 35 million compounds on databases such as ZINC (http://zinc.docking.org/; [57]). By using in silico screening approaches, these compounds can be virtually docked into specific sites on PirA vp or PirB vp . Furthermore, the binding affinities between PirA vp /PirB vp and compounds can be predicted using molecular docking tools, such as iGEMDOCK [35] and AutoDock Vina [58]. Ligand-based screening is another approach that can be used to identify inhibitors [59]. On the assumption that similar compounds can mimic physicochemical properties of the interacting regions and occupy the interface of the target protein, this approach uses online chemistry tools (e.g., Open Babel; [60]) to search for compounds that are similar to interacting peptides (e.g., a loop) of partner proteins (e.g., the PirA vp binding interface on PirB vp ). Ultimately, compounds with high docking scores that predict greater affinity can be considered as potential inhibitors, and these can then be validated through bioassays, as well as shrimp challenge assays. Using these approaches, we are hopeful that a potential PirA vp /PirB vp drug/inhibitor can be discovered in the near future. Table 1. Potential interacting regions on PirA vp and PirB vp that may be suitable targets for structure-based drug design.

Potential Function Regions Involved in Possible Interactions Amino Acid Sequences
Receptor binding

Conclusions
In this review, we have presented structural views of the major pathogenic factors of AHPND: V. parahaemolyticus PirA vp and PirB vp . Based on the structural similarity to B. thuringiensis Cry pore-forming toxin, we hypothesized that PirA vp and PirB vp may use similar mechanisms to cause cell death in shrimps. Furthermore, strategies for drug/inhibitor design against these two toxins were proposed. As more details are discovered, we anticipate that the future safety and usefulness of the insecticidal applications of this toxin family will also be improved.