Mechanism of enzymatic reaction and protein-protein interactions of PLD from a 3D structural model.

The phospholipase D (PLD) superfamily catalyzes the hydrolysis of cell membrane phospholipids generating the key intracellular lipid second messenger phosphatidic acid. However, there is not yet any resolved structure either from a crystallized protein or from NMR of any mammalian PLDs. We propose here a 3D model of the PLD2 by combining homology and ab initio 3 dimensional structural modeling methods, and docking conformation. This model is in agreement with the biochemical and physiological behavior of PLD in cells. For the lipase activity, the N- and C-terminal histidines of the HKD motifs (His 442/His 756) form a catalytic pocket, which accommodates phosphatidylcholine head group (but not phosphatidylethanolamine or phosphatidyl serine). The model explains the mechanism of the reaction catalysis, with nucleophilic attacks of His 442 and water, the latter aided by His 756. Further, the secondary structure regions superimposed with bacterial PLD crystal structure, which indicated an agreement with the model. It also explains protein-protein interactions, such as PLD2-Rac2 transmodulation (with a 1:2 stoichiometry) and PLD2 GEF activity both relevant for cell migration, as well as the existence of binding sites for phosphoinositides such as PIP2. These consist of R236/W238 and R557/W563 and a novel PIP2 binding site in the PH domain of PLD2, specifically R210/R212/W233. In each of these, the polar inositol ring is oriented towards the basic amino acid Arginine. Since tumor-aggravating properties have been found in mice overexpressing PLD2 enzyme, the 3D model of PLD2 will be also useful, to a large extent, in developing pharmaceuticals to modulate its in vivo activity.


INTRODUCTION
PLD has been associated with a variety of physiological cellular functions, such as intracellular protein trafficking, cytoskeletal dynamics, membrane remodeling and cell proliferation (1,2) and with pathological activities such as angiogenesis and tumorigenesis (3,4). There are currently 6 mammalian isoforms of the gene (PLD1, 2, 3, 4, 5 and 6) (5)(6)(7)(8). Out of these, PLD1 has been extensively studied. The PLD1 gene has been localized to the long arm (q) of chromosome 3 (3q26) (9) and covers 210 kb of genomic DNA that is defined by 31 exons (10,11). The mammalian PLD2 gene is found on the short arm (p) of chromosome 17 (17p13) (12), is defined by 25 known exons of a genomic region spanning 16.3 kb and encodes for two splice variants (PLD2a and PLD2b) of 933 amino-acids in length each (13), which yields functionally indistinguishable proteins of 106 kDa MW (14).
All members of the PLD superfamily possess two highly conserved phosphatidyltransferase HKD catalytic domains (HKD1 and HKD2) that are defined by the consensus peptide sequence HxK(x)4D(x)6GSxN, which are vital to the lipase activity. PLD1 and PLD2 also bear the phox homology (PX) and pleckstrin homology (PH) domains, both at the Nterminal end and the phosphatidylinositol 4,5-bisphosphate [PIP 2 ] binding site (15). Two HKD motifs are requisite for catalytic activity (16). In case of salmonella endonuclease (nuc) the protein forms dimer to catalyze the reaction. In higher organisms the protein contains two HKD motifs suggesting a duplication and common origin. In these proteins, both the motifs tightly interact to form an active center (16,17). The HKD motifs are also present in other biologically diverse proteins such as bacterial phospholipid synthases, endonucleases, a pox envelope protein and a Yersinia toxin. PLD1 and PLD2 are classical mammalian PLD isoforms (15). The PX and PH domains of PLD1 and PLD2 function as strong modulators of the membrane recycling machinery. It regulates, for example, interaction with SH2/SH3-containing tyrosine kinases (18,19). PLD3, PLD4, PLD5 and PLD6 isoforms lack PX and PH domains (20) and, therefore, could be considered non-classical PLDs.
Apart from lipase activity, PLD2 possesses guanine nucleotide exchange activity (GEF) for the small GTPase Rac2 and Ras (21)(22)(23). The PX domain of PLD2 contains the site for GEF activity, whereas both the PLD2-PX and -PH domains are involved in interacting with the small GTPase substrate (24), which can be either Rac2 or Ras (22,25).
In spite of these accomplishments characterizing PLD, there are not yet any resolved structures of the protein either from a crystallized protein or from NMR. Therefore, we set out to model the isoform of PLD2, using computational methods and selecting from among the candidate structures on the basis of their consistency with the experimental results.
We report here a 3D model of PLD2, fully based on biochemical data, which will help explain the 2 intermolecular enzymatic activities, as well as to reasonably predict the physiological behavior of PLD in cells, protein-protein interactions and phospholipid binding sites.
Having a working modeled structure of PLD will tremendously help in the rational design of more specific inhibitors that are proving to become quite important for recovering tumorigenesis (26), as relates to the PLD/phospholipid fields.

Cells and cell culture
COS-7 cells were obtained from the American Type Culture Collection. Cells were cultured in DMEM supplemented with 10% (v/v) fetal bovine serum (FBS).

Transfections
Transfections were performed using appropriate amounts of plasmid DNA, 5 μl Lipofectamine (Invitrogen), and 5 μl Plus reagent (Invitrogen) in Opti-MEM medium (Invitrogen), per the manufacturer's protocol. COS-7 cells or RAW264.7/LR5 macrophages were transfected for 3 h, washed, re-fed with prewarmed complete medium, and maintained for 48 h. After 48 h, cells were harvested for their respective experiment. When necessary, post-transfection before harvesting, cells were treated with 10 nM EGF or IL-8 depending upon the experiment and the cell line for indicated time lengths.

Purification of PLD2 and Rac2 using baculoviral expression system
To generate a purified, recombinant, and full-length PLD2 or Rac2, a baculovirus expression system was used for the overexpression and subsequent purification. Briefly, the PLD2-WT or Rac2-WT genes and the Bsu36I-digested BacPAK5 viral DNA were co-transfected into Sf21 insect cells in separate reactions, which rescued the very large viral DNA and effectively transferred the PLD2 or the Rac2 genes separately into the AcMNPV genome. PLD2 or Rac2 baculoviral stocks were selected that overexpressed PLD2-WT or Rac2, respectively, and were used to infect Sf21 insect cells for further overexpression of PLD2 or Rac2 and subsequent purifications of PLD2 or Rac2 using the TALON matrix, as previously described in (27).

Immunoprecipitation, SDS-PAGE and western blot analyses
After transfection, cells were harvested and lysed with special lysis buffer (5 mM HEPES, pH 7.8, 100 μM sodium orthovanadate, and 0.1% Triton X-100). The lysates were sonicated and the presence of overexpressed protein was confirmed by performing SDS-PAGE and Western blot analysis. For immunoprecipitation, the cell lysates were treated with 1 μl monoclonal antibody for the respective protein and 10 μl agarose beads (Millipore, Billerica, MA) and incubated at 4°C for 4 h. After incubation, the immunoprecipitates were washed with LiCl wash buffer (2.1% LiCl, 1.6% Tris-HCl, pH 7.4) and NaCl wash buffer (0.6% NaCl, 0.16% Tris-HCl, 0.03% EDTA, pH 7.4), respectively, and sedimented at 12,000 × g for 1 min. The resulting pellets were then analyzed using SDS-PAGE and Western blot (WB) analyses.

Prediction of full-length PLD2 structure: homology modeling
Based on the experimental data, a list of amino acid residues that is key for PLD2 activities as well as PLD2's interactions with other proteins was generated. This is shown in Table 1, and is used to assure the quality of the predicted structures.

I-TASSER
Human PLD2's amino acid sequence was obtained from NCBI, and submitted, to the protein structure prediction server, Iterative threading assembly refinement server (I-TASSER). I-TASSER threads the protein sequence through the PDB structure library and searches for the possible alignments. For unaligned regions, I-TASSER uses ab intio modeling. I-TASSER simulations can be run not only for the full chain, but also for the separate domains (28). Since I-TASSER-predicted full-length PLD2 is principally comprised of random secondary structures (not shown), we also obtained structures of domains separately. More than one solution was obtained for individual domain predictions. One out of the many solutions was chosen based on the criteria mentioned in Table 1. These were docked pairwise to obtain a group of candidate structures of the complete protein. From this group, one final model was chosen visually, based on the criteria mentioned in Table 1.

Phyre-2
We also used the server, Phyre 2 (protein homology/analogy recognition engine), which is an improved version of Phyre. After creating the profile of non-redundant sequence, the secondary structure was predicted. Using the profile and model of secondary structure a fulllength 3D models were generated. The missing parts that arise during the queues are determined by the loop library method (26). Finally, molecular modeling and visualization was performed using the Swiss PDB viewer (SPDBV). More than one solution was observed before finalizing the final structure. By visual inspection we concluded the PLD2 model predicted by Phyre2 server meets most of the criteria listed in Table 1. However, this model was not entirely energetically favorable as certain amino acid conformations are not viable. These issues were fixed by manually perturbing the structure using SPDBV and minimizing the result using steepest decent and conjugate gradient methods. Finally, the complete 3D model was energy minimized which resulted in a modeled structure, which was energetically favorable and consistent with experimental criteria listed in Table 1.

Molecular docking
The computer-simulated docking studies were performed using both HEX 6.1 and AutoDock Vina. In order to model the full-length 3D PLD2 structure from the different domains of PLD2 (obtained from I-TASSER) as well as for docking two (PLD2/Rac2) the HEX (29) was used. The structure of Rac2 was obtained from PDB database (PDB IDs 2W2T). HEX is a macromolecular docking program that takes into consideration both shape and electrostatic charge. In predicting molecular docking juxtapositions, it uses spherical polar fourier (SPF) correlations to accelerate the docking calculations. In many respects, this approach is similar to conventional fast Fourier transform (FFT) docking methods, which use Cartesian grid representations of the molecular shape and electrostatic properties and translational FFTs to perform docking correlations. The default docking control parameters of the HEX program were used to arrive at 100 docked conformations (30). In all the HEX dockings, PLD2 was treated as receptor while the other protein partner was treated as ligand.

Enzyme-substrate docking
For docking PLD2 with its substrate 1, 2 di-octanoyl phosphatidyl choline (PC8), PC8 structure was retrieved from the PDB database, ID: 2LYA. PLD2 and the substrate 1,2dioctanoyl-sn-glycero-3-phosphocholine (PC8) were submitted to Autodock Vina (31). Autodock Vina allows the ligand to have flexible/rotatable bonds. Docking with AutoDock Vina starts by defining a search space or binding site in a restricted region of the protein.
The coordinates of Lys 444 zeta nitrogen lie between the two catalytic Histidines (442 and 756) in our 3D model. Therefore, in the present study, the receptor grid was generated using the coordinates of Lysine 444. The resulting docking conformation was further visualized using the PyMOL.
Using the same conditions, docking experiments were performed with phosphatidyl ethanolamine (PE) and di-oleoyl phosphatidyl serine (PS), in order to explain the ability and specificity of PLD2 towards PC. Structures of PE and PS were retrieved from the PDB IDs 3B74 (32) and 2LYB (33) respectively. To determine PIP 2 binding sites on PLD2, docking experiments were performed with PLD2 and phosphatidyl inositol 4, 5-bisphosphate (PIP 2 ) whose structure was retrieved from the PDB ID, 3W68 (34).

Coordinates
Access to the full-length modeled structure created in this paper can be found in the Supplementary data file "PLD2 coordinates of 3D structure model" with PDB that contains PDB coordinates in text format. These coordinates can be used by any investigator to reconstruct the three-dimensional model we are presenting in this study.

Statistical Analysis
Data are presented as mean + SEM. The difference between means was assessed by the Single Factor Analysis of Variance (ANOVA) test. Probability of p<0.05 indicated a significant difference.

A 3D modeled structure of PLD2
The absence of a crystal structure for mammalian PLD2 hinders advances in the field of PLD biology. To remedy this situation, we combined modeling tools with biochemical evidence, such as substrate accessible active site(s) and solvent accessible phosphorylation sites to obtain a structural model. A compilation of criteria including considerable experimental evidence is shown in Table 1 (10,21,24,(35)(36)(37)(38)(39)(40)(41)(42)(43)(44)(45)(46)(47). These criteria are a summary of data from both our lab and from other investigators, as listed in the references of the table. They present the starting experimental evidence used to test the model that, for example, certain amino acids that interact with other signaling proteins with known motifs (such as SH2 or SH3) need to be on the surface of PLD2 in order to bind to said proteins; or that certain alcohol-amino acids (Ser, Thr, Tyr) need to be also exposed to the surface in order to be phosphorylated; or that the histidines on the two catalytic motifs should be in close spatial proximity. Since the large majority of the criteria were fulfilled with the initial model, we used it to further refine a predicted PLD2 structure.
The strategy we followed to obtain the full-length 3D model of PLD2 is pictorially shown in Figure 1A, which used two different protein prediction programs algorythms, implemented by either I-TASSER or Phyre and compared the resulting structures (28,48). In either case, the scoring that the structure solving programs provide had to be orthogonal with the biological and biochemical data on Table 1. For example, energy minimization was run only after a candidate structure has checked out all requisites of Table 1.

PLD2 structure prediction -Phyre versus I-TASSER
When a full sequence was submitted to I-TASSER, the HKD motifs were comparable to the crystallized bacterial HKD members. However, a major part of the resulting structure was found to be random without the expected PX and PH domains. To remedy this, domain-wise structures were obtained from I-TASSER (28). More than one solution was obtained for each domain. Individual domains were chosen based on the set criteria in Table 1. Next, using the program HEX, we docked the globular domains of the possible solutions. The lowest energy docking orientation was selected and the primary structures of the fragments were linked to obtain the full-length modeled structure (predicted at pH 7). In a complementary approach, the full-length PLD2 sequence was submitted to the Phyre 2 server for "Intensive'' modeling (48). Multiple solutions were obtained out of which, the 3D model structure that satisfied the set criteria in Table 1 and that with the lowest energy score was selected. The confidence of the resulting model estimated by Phyre 2 is summarized in Figure 1B, which represents the overall confidence as a function of sequence. It is worth mentioning that in the Phyre-predicted structure, 71% of the residues of PLD2 were modeled at greater than 90% confidence. Analysis of I-TASSER-and Phyre-predicted PLD2 structures revealed that the criteria listed in Table 1 are convincingly demonstrated to a larger extent in Phyre-PLD2 and therefore we pursued the Phyre predicted model for PLD2 3D structure.

Refining Phyre PLD2
The initial structural model obtained (that fulfilled criteria in Table 1) was not energetically minimized. In several instances, sidechains were in untenable conformations such as steric hindrances between some of the amino acid sidechains. Therefore, the model was globally adjusted manually and through a combination of steepest decent (1000 step) and conjugate gradient (300 step) energy minimizations implemented by SPDBV (49). Bulk dielectric constants were used for these calculations instead of explicit solvent molecules. All these steps provided gave us a biochemically agreeable, energetically favorable, full-length PLD2 3D model. This model is shown in Figure 2, where active sites, protein binding sites and phosphorylation sites are labeled in different colors (Figure 2). Access to the full-length modeled structure created in this paper can be found in the Supplementary data file "PLD2 coordinates of 3D structure model" with PDB that contains PDB coordinates in text format.

A 3D comparison of PLD isoforms
Next, we sought to compare sequence as well as the modeled structure of PLD2 with other mammalian isoforms. First, we obtained the predicted structures of PLD1 and PLD3-6 via Phyre 2 server (energy minimizations were not performed for these structures). A multiple sequence alignment of PLD1 through 6 was performed using Clustal W, following which we compared predicted 3D structures of PLD1, 3, 4 and 6 isoforms with our PLD2 structural model. It is known that PLD1 and 2 are the only mammalian isoforms that contain the regulatory PX and PH domains. Figures 3A and B shows the structural alignment of the PX and PH domains of PLD1 and 2, respectively. The two HKD motifs are spatially or threedimensionally conserved among the PLD1 through 6 isoforms ( Figure 3C). Sequence alignments of the HKD motifs of PLD isoforms is shown in Figure 3D. Amongst the mammalian PLD isoforms, only PLD1, 2 and 6 possess lipase activity. The lack of lipase activity in PLD3 and 4 cannot be attributed to the absence of PX and PH domains, since PLD6 also lacks these domains and is still active upon dimerization.
Next, we wanted to compare the catalytic site of our PLD2 structure model, with that of streptomyces PLD that was crystallized along with its substrate dibutyrylphosphatidyl choline (PDB ID: 1VOV) (50). Figure 3E and F shows the superimposed streptomyces (white) and human PLD (green) active site residues, while Figure 3G shows sequence alignment of the same. These results indicate that the active site residues of our predicted model for the PLD2 structure fits with the existing bacterial PLD crystal structure to a larger extent. Specifically the orientation of the two key amino acids, Histidines 442 and 756, those that play an important role in the phosphatidyl choline hydrolysis are aligned.

Modeling of PLD2 complexed with phosphatidyl choline
Once we found that the active site residues superimpose, we decided to test it by performing theoretical docking studies with its substrate phosphatidylcholine (PC) and with known protein interacting partners. Examining these binding/interactions with the PLD2 model of Figure 3 allowed for us to test our model in silico. Based on biochemical experimental data, the two histidines in the HKD motifs play an important role in the catalytic conversion of PC to PA and choline (15,50,51). We first performed lipase assays using variety of phosphatidylcholine species ( Figure 4A) as well as other phospholipids ( Figure 4B). Results shown in Figure 4A indicate that PLD2-WT, but not PLD2-K758R can breakdown a variety of PC species, including 1,2-dioctanoyl-sn-glycero-3-phosphocholine (PC8), 1,2diarachidonoyl-sn-glycero-3-phosphate (AraPC) and 1,2-dioleoyl-sn-glycero-3phosphocholine (DOPC). Figure 4B indicates that C8-PC followed by DOPC, but neither PE nor PS serves as substrates for this PLD2.
Next, to demonstrate phospholipid accommodation in PLD2's presumptive active site pocket of the 3D structural model of PLD2, initial docking was performed with PLD2 and PC8 (from PDB ID: 2LYA) (33) by employing autodock vena (31). PLD2 was treated as receptor, whereas PC8 as a small molecule ligand. The search space was centered at the location of lysine 444 that lies between the two Histidine residues. Of the solutions derived by autodock, the best scoring conformation was presented in Figure 4C and D, which shows PLD2's active site ( Figure 4C) with the substrate PC8. The orientation of the key amino acids in the active site is more clearly shown in Figure 4D. Figure 4B suggests that PLD2 while hydrolyzes PC, it is not able to hydrolyze other phospholipids, PE and PS. Therefore, we next wanted to determine the specificity of PLD2's putative lipase active site pocket in silico. Using the same conditions that we used to dock PC8, we performed docking with the phospholipids phosphatidylethanolamine (PE) (PDB ID: 3B74) (32) and phosphatidylserine (PS) (PDB ID: 2LYB) (33). Panels in Figures 4E, F and G show PLD2's presumtive active site docked with PC8, PE and PS. A closer observation of these results led us speculate that the choline group on PC is oriented outwards making the choline release easy. In case of both PE and PS, the head groups are oriented inward into the presumptive active site pocket. We propose that this makes the release of head group difficult when compared to that of PC, further affecting the hydrolysis. All this explains the PLD2's ability to hydrolyze PC but not PE and PS ( Figure 4B). However whether PS or PE can act as inhibitors of PLD2 is not clear, as these experiments were performed with one substrate at a time (PS or PC or PE). In reality, other factors such as, affinity of the enzyme (PLD2) to the substrate and the amount of the substrate available and accessible to the enzyme determine the formation of enzyme substrate complex resulting in the product. Also additional in silico experiments with PLD2 3D structure model with other possible phospholipid substrates such as lysoPC, will give insight on PLD2's catalysis.

Mechanism of PLD2 lipase reaction, based on structural details
Based on our model of PLD2 3D structure and its docking with the substrate PC8 ( Figure  4C-D), we propose the working model for human PLD2 lipase activity ( Figure 5). The reaction could take place in three steps that are explained as follows:

3.
Histidine 756, on the other hand will participate in the reaction possibly by donating a proton (Nd) leading to the release of choline. Subsequently, with the help of Tyrosine 754, Histidine 756 activates water. A second nucleophilic attack will takes place where water will nucleophilically attack the phosphate moiety. This will result in the formation of phosphatidic acid (PA).
Also the superimposition of bacterial PLD and human PLD2 ( Figure 3E) indicates the conserved orientation of the key aminoacids as well as the secondary structures. All this is in agreement with the mechanism of lipase action of bacterial PLD that was based on crystal structure (50,51). These facts suggest that the presented model for PLD2 3D structure is comparable to the bacterial PLD in terms of lipase reaction mechanism. In addition, we are able to explain other existing biochemical findings with the proposed 3D structure model.

Modeling of PLD2 structure model with PIP2
It is well established that phosphatidyl inositol 4,5-bisphosphate (PIP 2 ) is one of the positive regulators of PLD2 (53). Therefore we sought to determine the possible putative binding site for PIP 2 on PLD2. Biochemical data from the literature suggest that PIP 2 might have two binding sites on PLD2, one at PH domain and the other towards the C-terminus (54). In the same study they showed that Arginine 237 and Tryptophan 238 are essential for PIP 2 mediated localization of PLD2. In the same study they also suggested the possible PIP2 binding regions binding in the C-terminal region (54). PIP2 binding regions in various proteins are not well defined, but generally is comprised of clusters of basic and aromatic aminoacid residues. In a study by Sciorra et al (55), it was speculated that the region between 554-575 aminoacids in mouse PLD2 might be a putative PIP 2 binding region. This region corresponds to 531-564 in human PLD2 (hPLD2). By mutational analysis they showed that Arginines 554 (R553 in hPLD2) and 558 (R557 in hPLD2) contribute in responding to PIP 2 (55). However many other possible basic amino acids in that region, which might play a role in binding to PIP 2 binding cannot be ruled out. In order to determine if our model can predict the new binding sites or if it can dock PIP2 to the proposed binding sites, we performed docking experiments.
Docking was performed with PIP 2 (PDB ID: 3W68) and our PLD2 structure model with the search site focused at PH domain and between 532 and 564 aminoacids. Out of the solutions obtained, we chose three solutions ( Figure 6). In each of these, polar inositol ring is oriented towards the basic aminoacid residue and the hydrophobic tail is oriented towards aromatic aminoacids, mostly tryptophan. The first two solutions are R236 and W238 ( Figure 6A), R557 and W563 ( Figure 6B) that were suggested to be involved in PIP 2 interaction. This information is in consistent with our model for PLD2 3D structure. Figure 6C on the other hand, shows a novel putative PIP2 binding region in PH domain, which involves R210, R212 and W233. Nevertheless, this requires experimental confirmation.
These results indicate that the polar region of PIP 2 might interact with the basic aminoacid residues that are clustered and are surface exposed. R210, 212, and 236 as well as K235 are all present as a cluster in the near vicinity. The possibility of more than one PIP 2 molecule interacting with multiple basic aminoacid clusters at the same time is quite possible. In Figures 6A, B and C, the docking conformation of PLD2 and PIP 2 were shown in ribbon and surface modes. Along with the basic aminoacids, aromatic aminoacids such as W might also play a role by interacting with hydrophobic regions of PIP 2 .
Next, we wanted to explore the effect of the novel putative PIP2 binding mutants (R210 and R212 on lipase activity in the presence of PIP2. In order to do that a PLD2 double mutant was made where R210 and R212 were mutated to Alanine. Intact cells or cell lysates overexpressing either PLD2-WT or PLD2-R210/212A, were treated with increasing concentrations of PIP2 ranging from 0 to 10 mM and lipase activity was determined. The results presented in Figure 6G and H suggests that PIP2 exerted a positive effect on PLD-WT expressing cells in a dose-dependent fashion, as expected. Interestingly mutating the putative PIP2 sites (R210 and R212), made PLD2 irresponsive to PIP2.

Protein interactions of PLD2: PLD2 and Rac2 molecular modeling
We next wanted to determine if our model could explain the protein-protein interactions of PLD2. From previous studies in our lab, we know that PLD2 and Rac2 interact with each other and in addition, Rac2 influences PLD2 activity (38,56). While this suggests that PLD2 is an effector of Rac2, PLD2 acts as GEF for Rac2 and acts upstream (24). Representative data showing the same is presented in Figure 7A-B. Figure 7C shows our modeled PX domain (from Figure 2) has PLD2-GEF catalytic residues highlighted. From the threedimensional model, it is obvious that residues F107, F129, L166, and L173 might form a hydrophobic pocket, which we speculate might be involved in promoting nucleotide exchange on Rac2.
The fact that Rac2 is both upstream affecting PLD2 and downstream activated by PLD2 suggests the possibility of more than two binding sites for Rac2 on PLD2. Taking into consideration these sets of facts, we wanted to see if our PLD2 3D modeled structure was in agreement with these experimental findings. We used the PLD2 modeled structure and performed HEX docking (57) with a first Rac2 molecule (PDB id: 2W2T) (58). From the several docking results that were generated, an energetically favorable complex was chosen and docked with another molecule of Rac2 ( Figure 7D, E). In agreement with the experimental data, two Rac2 molecules are in close proximity to the PLD2-PX and -PH domains.
We believe that the upstream Rac2 (shown in navy blue) that regulates PLD2 via binding is in closer proximity to the CRIB motifs (gray and purple) of the PLD2-PH domain ( Figure  7D, E). On the other hand, the PLD2-GEF catalytic site (green) is in close proximity to the downstream Rac2 (magenta) switch I (red) ( Figure 7D, E), which in conventional GEF catalysis is usually perturbed by upstream GEFs. All this correlates with the results of mutational analysis, which identified residues in the PLD2-PX as GEF residues. Overall, PLD2-GEF activates Rac2, while at the same time PLD2 is an effector protein for active Rac2-GTP.

Why is our model for PLD2 3D structure significant?
Modeling experiments were performed using the ITASSER server (28) and the Phyre 2 (48) server. In both cases multiple models are predicted and a "best" model is suggested. Groups that created each of these servers used them in the recent Critical Assessment of Protein Structure Prediction experiment (CASP 10). This experiment assesses the ability of protein prediction algorithms and servers to predict protein structures for proteins prior to their determination by x-ray crystallography. From about 70 protein prediction servers created by structural biomedical/bio computational scientists across the protein structure prediction community, I-TASSER performed best and Phyre 2 was in the upper 30% for prediction of three dimensional protein structure from primary sequences. An interesting finding in this experiment was that these servers found the correct conformation much more frequently among the top scoring 10 or so candidate models, but were unable to correctly choose which of these was best.
This argues first, that the prediction methods used in this paper were among the very best at the current state of the art and second, that the chances of selecting the best structural model are greatly improved by supplementing largely energy-based scoring methods with alternative information about a protein's structure. This information includes biological activity and information, for example molecule binding properties. Here we use just such biological information in conjuction with sound structure prediction methodologies to identify models that are consistent with known Phospholipase biology and are heuristic, suggesting both features of these enzymes and how these features can be tested.

CONCLUSIONS
We present here a 3D modeled structure of PLD2 fully based on biochemical and biological data and, in turn, explain several models of how this enzyme interact with other cell signaling proteins. Since a 3D structure for mammalian PLD has not previously been determined (crystallography, NMR or otherwise) or predicted by any laboratory, this contribution is first as far as we can understand.
The model provides a structural basis that explains previously published biological and biochemical findings by our and other labs. The predicted modeled structure explains the active site, several motif homologies with other mammalian PLD isoforms and interaction with known small molecule inhibitors. This is important since PLD2 has been recently described as a tumorigenic and metastasis inducing protein. Finding good docking inhibitors will be important for understanding and modulating pathology. The model also predicts what other proteins should bind to PLD (we are giving the example of Rac2). Further, it can be used to study the other enzymatic activity in PLD that is, a GEF activity.
This study established a firm foundation for future bioinformatics-based studies using fulllength PLD2 as a model. While PLD2 activity has been shown to be necessary for cellular processes, like leukocyte chemotaxis and phagocytosis, and general cell migration. Deregulated PLD2 levels were reported in several cancers such as breast, colorectal and renal (59)(60)(61)(62)(63). A recent publication from our own lab reported a study in which the presence of PLD2 promoted the formation of aggressive tumors in mice, while silencing PLD2 reduced tumor size (3). All this suggests an increasing demand for the availability of PLD2's 3D structure, which can then be used to design specific and efficient small molecule inhibitors or modulators that will regulate PLD2 activities/protein interactions.
The predicted 3D model of PLD2's structure reported here is in accordance with the experimental evidence to a very large extent. From the docking results with the known PLD2 binding partners presented in the current study, we believe that exceptional modulators of PLD2 can be developed more efficiently. The current model has also paved a way to determine how PLD2 associates with the membrane when stimulated and the regulatory structural dynamics. The model should also constitute the basis for the discovery of new enzymatic properties on PLDs and new protein-protein associations with signaling proteins or lipids.

Supplementary Material
Refer to Web version on PubMed Central for supplementary material.

Highlights
• The first structure model for mammalian PLD, in the absence of any resolved structure, crystallized or NMR.
• A model based on biochemical data that predicts the active site accommodation of substrate, mechanism of reaction and interactions with other proteins during cell signaling • The model for the PLD2 structure is consistent with experimental observations and predicts new findings, such as the discovery of a phosphoinositide-binding site previously unknown.
• As PLD2 is tumorigenic, the 3D structure is crucial for developing new inhibitors of in vivo activity. (A) Self-explanatory schematic representing the strategy applied in this study to obtain a PLD2 three-dimensional structure. Briefly, PLD2 primary sequence was submitted to I-TASSER and Phyre 2 Extensive protein prediction servers. Both the servers provided with more than one solution, which were evaluated based on set criteria (the laboratory bench marks) listed in Table 1  (Ci-Civ) Different snapshots (snapshot 2 is turned 50° X, snapshot 3 is turned 60° X, 125° Y, snapshot 4 is turned 70° anticlockwise Y, 30° Z from snapshot 1 as indicated in the Figure) of PLD2 3D structure model in surface mode indicating that the important amino acids that interact with the binding partners or subject to post translational modifications are exposed to the solvent. Color code is as follows: Yellow -PX domain, Silver-PH domain, Cyan -Rest of the PLD2 bearing HKD motifs, Magenta -Lipase sites, Red-GEF sites, Orange -putative Phosphorylated sites, Brown -Putative lipid binding sites, Dark blue PKC binding sites, Skyblue -PLC binding sites, Pink -CRIB1, Purple -CRIB2. Full access of this model can be found along with the paper in the website "PLD2 3D structure model" with PDB coordinates in text format.    upstream and downstream Rac2 molecules, respectively. In both Rac2 molecules, red and blue colors represent switch I and II, respectively. Labels are represented in same colors as that of each domain. Table 1 Criteria based on experimental evidence that need to be applied to any predicted structure for validation. List of important amino acid residues in PLD2 and their functional role that are used in this paper as criteria for the PLD2 molecular modeling. These criteria are a summary of data from both our lab and from other investigators, as listed in the references of the table. They present the starting experimental evidence used to test the model that, for example, certain amino acids that interact with other signaling proteins with known motifs (such as SH2 or SH3) need to be on the surface of PLD2 in order to bind to said proteins; or that certain alcohol-amino acids (Ser, Thr, Tyr) need to be also exposed to the surface in order to be phosphorylated; or that the histidines on the two catalytic motifs should be in close spatial proximity.