Small-Molecule Inhibitor Leads of Ribosome-Inactivating Proteins Developed Using the Doorstop Approach

Ribosome-inactivating proteins (RIPs) are toxic because they bind to 28S rRNA and depurinate a specific adenine residue from the α-sarcin/ricin loop (SRL), thereby inhibiting protein synthesis. Shiga-like toxins (Stx1 and Stx2), produced by Escherichia coli, are RIPs that cause outbreaks of foodborne diseases with significant morbidity and mortality. Ricin, produced by the castor bean plant, is another RIP lethal to mammals. Currently, no US Food and Drug Administration-approved vaccines nor therapeutics exist to protect against ricin, Shiga-like toxins, or other RIPs. Development of effective small-molecule RIP inhibitors as therapeutics is challenging because strong electrostatic interactions at the RIP•SRL interface make drug-like molecules ineffective in competing with the rRNA for binding to RIPs. Herein, we report small molecules that show up to 20% cell protection against ricin or Stx2 at a drug concentration of 300 nM. These molecules were discovered using the doorstop approach, a new approach to protein•polynucleotide inhibitors that identifies small molecules as doorstops to prevent an active-site residue of an RIP (e.g., Tyr80 of ricin or Tyr77 of Stx2) from adopting an active conformation thereby blocking the function of the protein rather than contenders in the competition for binding to the RIP. This work offers promising leads for developing RIP therapeutics. The results suggest that the doorstop approach might also be applicable in the development of other protein•polynucleotide inhibitors as antiviral agents such as inhibitors of the Z-DNA binding proteins in poxviruses. This work also calls for careful chemical and biological characterization of drug leads obtained from chemical screens to avoid the identification of irrelevant chemical structures and to avoid the interference caused by direct interactions between the chemicals being screened and the luciferase reporter used in screening assays.


Introduction
Shiga toxin (Stx) produced by the bacteria Shigella dysenteriae and Shiga-like toxins (Stx1 and Stx2) produced by certain strains of Escherichia coli are potent ribosome-inactivating proteins (RIPs) [1].Shiga-like-toxin-producing E. coli O157:H7 is an emerging bacterial pathogen responsible for outbreaks of foodborne disease with significant morbidity and mortality in the United States [2].E. coli O157:H7 is the most common cause of hemolytic uremic syndrome, causing more than 20,000 infections and as many as 250 deaths annually [3].Ricin is another potent RIP isolated from the seeds of the widely available castor plant, Ricinus communis [4], belonging to a family of dichain cytotoxins (type II RIPs) that includes abrin and several other plant toxins [5].While not frequently associated with disease, the toxicity of ricin has made it an attractive tool for both bioterrorism and the targeted killing of cancerous cells [4].
Type II RIPs have two subunits: subunit A, which binds to 28S ribosomal RNA (rRNA) and depurinates a specific adenine residue from the a-sarcin/ricin loop (SRL) thereby inhibiting protein synthesis [6,7], and subunit B, which recognizes specific receptors on the target cell and facilitates transfer of subunit A into the cell where the inhibition of ribosome activity occurs [8].According to site-directed mutagenesis and X-ray diffraction studies along with a transition-state analysis of the depurination caused by ricin subunit A (RTA) [9][10][11][12][13], the catalytic mechanism of depurination by RTA begins with sandwiching of the adenine ring of the substrate rRNA between Tyr80 and Tyr123 of RTA via pi-pi interactions [9] at the Michaelis-Menten state [14].These interactions enable the protonation of the adenine ring at N3 [9] by the cationic Arg180 of RTA that forms a hydrogen bond to the anionic Glu177 of RTA at the transition state [10].The protonation consequently cleaves the adenine in the zwitterionic form from the ribose by breaking the bond between N9 of the adenine and C1 of the ribose, thus leading to the formation of a cationic ribose intermediate stabilized by Glu177 at the transition state.A water molecule activated by the neutral Arg180 subsequently attacks the ribocation to form the ribose product and resume the cationic Arg180 [10][11][12][13].
Small-molecule inhibitors of ricin and Shiga/Shiga-like toxins are sought for potential pre-exposure or post-exposure treatment of RIP poisoning.Additionally, because ricin and abrin have potential medical use as immunotoxin components [15], small-molecule inhibitors can also serve as co-treatments to control immunotoxin toxicity.Some oligonucleotides developed through structure-based design-circular DNA and DNA/RNA hybrid molecules, for example-inhibit RTA at nanomolar concentrations [16][17][18][19]; the most potent has a K i value of 2.3 nM [19] and is potentially useful in the immunotoxin cancer therapy [15,19].These molecules are effective in neutralizing extracellular toxin to prevent further intoxication, but they generally cannot enter cells to neutralize intracellular toxin.Small-molecule inhibitors of RTA and Shiga/ Shiga-like toxins identified from high-throughput screens (HTSs) inhibit toxin transport at various stages at micromolar concentrations [20,21]; at a concentration of 200 mg/kg, one such inhibitor demonstrated full protection of mice against a dose of ricin that kills 90% of the unprotected control mouse population (1-4 in Figure 1) [21].Because of the lack of structural and regulatory information about components involved in toxin transport, optimization of these transport inhibitors can be difficult.
Both HTS and structure-based approaches have been pursued in the search for small molecules that can penetrate cells to neutralize intracellular toxins by inhibiting subunit A1 of Stx2 (Stx2A1) or RTA [22][23][24][25].The brute force approach has culminated in a small-molecule inhibitor of RTA with a half maximal inhibitory concentration (IC 50 ) value of 30 mM (5 in Figure 1) [22]; the rational approach has led to an inhibitor with an IC 50 value of 270 mM (6 in Figure 1) [25].Clearly, formidable challenges lie in the path of the structure-based design of molecules that can enter cells to inhibit RIPs directly.In our view, one key challenge is due to strong electrostatic interactions at the RIPNSRL interface [26][27][28][29] that make drug-like molecules ineffective competitors with polynucleotides for binding to RIPs because drug-like molecules are not highly charged.
In this article, we report the discovery of promising small molecules that demonstrate in vitro and ex vivo inhibition of Stx2 and ricin, using a novel approach to small-molecule inhibitors of proteinNpolynucleotide functions.This approach circumvents the challenge of the strong electrostatic interactions at the RIPNSRL interface.We discuss insights derived from these leads into structurebased design of improved RIP inhibitors, potential application of the new approach to other proteinNpolynucleotide-function inhibitors, and caveats for using chemical screens to uncover drug leads.

RIP Inhibitors Identified Using a Virtual Screen
Site-directed mutagenesis and kinetic studies as well as the Xray crystallographic analysis [24,[30][31][32] show that, upon binding to the RTA active site, the adenine group of the SRL substrate markedly changes the side-chain conformation of Tyr80, a catalytically important active-site residue of RTA.As a result of this conformational change, the phenolic ring of Tyr80 can stack with the adenine group and catalysis proceeds (Figure 2a), whereas the side-chain conformation of another catalytically crucial activesite residue of RTA, Tyr123, remains unchanged upon the rRNA binding.
Informed by these seminal findings and the aforementioned challenge of obtaining proteinNpolynucleotide-interaction inhibitors, we decided to use a doorstop approach to identify smallmolecule inhibitors of RTA and Stx2.This new approach aims to identify small molecules that work as doorstops to prevent an active-site residue of an RIP (e.g., Tyr80 of ricin or Tyr77 of Stx2) from adopting the active conformation thereby blocking the function of the protein rather than work as contenders in the competition for binding to the RIP.
In this context, we performed the following virtual screen to identify small molecules that can bind at the active site of the RTA crystal structure of 1IFS [32] to stabilize conformation 2 thereby preventing its conversion to conformation 3. Our reason to stabilize conformation 2 instead of conformation 1 was that an adenine molecule can fit underneath the phenolic ring of Tyr80 in conformation 2 (Figure 2c) and permits the use of a clip-like molecule to stabilize conformation 2 with two functional groups simultaneously binding on both sides of the phenolic ring, whereas conformation 1 lacks space for an adenine-like molecule beneath the phenolic ring.
Using an automated computer docking program (EUDOC) [35][36][37], we screened 236,925 small molecules for molecules that bind favorably in the region enclosed by a docking box over the phenolic ring of Tyr80 in the 1IFS crystal structure (Figure 2c) [32].This screen identified 226 chemicals with EUDOC energies (intermolecular interaction energies) lower than 250 kcal/mol.All of the small molecules screened were selected from an in-house database of 2.5 million chemicals using the criterion that molecular weight must not be greater than 300 Da.Typically, we select  [32]) with the oligonucleotide-bound RTA at the Michaelis-Menten state (yellow; 3HIO [13]) showing that the adenine group markedly perturbs the conformation of Tyr80; b (top right): three distinct conformations of Tyr80: conformations 1, 2, and 3 represent the apo conformation in green (1IFT [32]), the less populated bound conformation in magenta (1IFS [32]), and the most populated bound conformation in yellow (1FMP [31]), respectively; c (bottom left): the phenolic ring with an adenine group underneath and a docking box atop in the less populated bound conformation (1IFS [32]); d (bottom right): overlay of the oligonucleotide-bound RTA at the Michaelis-Menten state (yellow; 3HIO [13]) with RTA in conformation 1 (green; 1IFT [32]), conformation 2 (magenta; 1IFS [32]), and conformation 3 (cyan; 1FMP [31]) showing the closeness of the Tyr80 conformations in 3HIO and 1FMP and the clash between the nucleotide and Tyr80 in 1IFS.doi:10.1371/journal.pone.0017883.g002chemicals using two energy criteria: (1) the EUDOC energy must be ,240 kcal/mol, and (2) the van der Waals component of the EUDOC energy must be ,225 kcal/mol.These criteria are derived from our observation that all experimentally confirmed micromolar inhibitors identified by EUDOC have EUDOC energies and the van der Waals components below these values [38][39][40].Because the proteinNpolynucleotide interaction in this study is mainly electrostatic [26][27][28][29], we lowered the EUDOC energy cutoff to 250 kcal/mol and removed the van der Waals component criterion.
We visually inspected the 226 selected molecules using the following criteria to identify those with the characteristics required to stabilize conformation 2: (1) a carboxylate group that mimics the phosphate group of the rRNA substrate to interact with Arg213 and Arg258 of RTA, (2) an aromatic ring that has the offcenter pi-pi interaction with Tyr80, and optionally (3) an alkyl group that forms the van der Waals interaction with the methylene group of Phe93 (Figure 3).We then weeded out those that were commercially unavailable or have multiple chiral centers, poor solubility, or poor cell permeability.Subsequently, we purchased 27 compounds for biological evaluation.

Syntheses of RIP Inhibitors
After an initial biological evaluation of the 27 purchased chemicals, we synthesized the four most promising inhibitors, R16, R19, R20, and R22 (Figure 3) for spectroscopic analyses that require relatively large quantities of materials or for making close analogues (R16b, R19b-d, and R20b; Figure 3).Although R16, R19, and R20 were commercially available from SPECS (www.specs.net),and R22 was available from ASINEX (www.asinex.com),synthesis or spectroscopic data for R16, R16b, R20, R20b, and R22 have not hitherto been reported.
R16 and R16b were made in good yields by reacting 3hydrazinobenzoic acid in acetic acid with benzaldehydes (Figure 4).R19 and its analogues were readily prepared according to a known procedure [41].R22 was obtained from a reaction of 5-chloroindoline-2,3-dione with 3-aminobenzoic acid in methanol (Figure 4).The proton NMR spectra of R16, R19, and R22 prepared in house are identical to those of the chemicals purchased from SPECS and ASINEX.The proton NMR and NOESY experiments show that R22 exists in a mixture of E and Z stereoisomers with the Z isomer being dominant, which is consistent with the Z stereochemistry of R22 used in our virtual screen.
R20 or R20b was prepared by coupling 4-formylbenzoic acid with a substituted pyrrole in the keto form for R20 or a mixture of keto and enol forms for R20b according to a reported process [42] (Figure 4).The substituted pyrrole was obtained via cyclization of 2-amino-2-(2-chloroacetyl)butenoate [43], which was prepared from 3-aminobutenoate using a literature procedure [44].R20 has the E stereochemistry according to the chemical structure specified by SPECS (catalog number AO-081/14455020).The proton NMR spectrum of R20 made in house matches that of R20 purchased from SPECS.Furthermore, the in vitro and ex vivo biological activities of the in-house and purchased R20 were the same.However, the NOESY spectrum shows that the in-house R20 exists in the Z stereochemistry because of our observed correlations of the nitrogen-attached proton with the methyl and phenyl protons in R20 (Figure 5).Consistent with the Z stereochemistry of R20, (Z)-ethyl 2-methyl-5-(4-nitrobenzylidene)-4-oxo-4,5-dihydro-1H-pyrrole-3-carboxylate-a close analog of R20b-has been reported to have the Z stereochemistry [45].Therefore, identification of R20 as an active RIP inhibitor resulted from sheer luck, because the E stereochemistry of R20 specified by the chemical vendor was used in our virtual screen.

Evaluation of RIP Inhibitors Using in Vitro and ex Vivo Methods
Firefly-luciferase-based cell-free translation assays with rabbit reticulocyte lysate (RRL) [46] confirmed that 22 of the 27 compounds identified in our virtual screen showed some degrees of RTA inhibition at an inhibitor concentration of 50 nM.Of the 22 active compounds, R16, R19, R20, and R22 were the most promising.Further studies of these inhibitors and their analogs (R16b, R19b, R19c, R19d, and R20b) showed a 1.1-to 1.7-fold increase in luciferase activity resulting from the translation in the RRL after treatment with 1 nM RTA and 1 nM inhibitor, relative to the activity after treatment with 1 nM RTA only (Table 1).R19b and R16b showed 1.7-and 1.6-fold increases in luciferase activity, respectively.Interestingly, the luciferase activity in the RRL treated with R16b alone increased as the concentration of R16b increased, whereas that of the RRL treated with RTA and R16b decreased as the R16b concentration increased (Figure 6).Other inhibitors showed similar concentration effects on luciferase activity.These concentration effects made the determination of IC 50 values difficult and suggested that these inhibitors might interact with both RTA and firefly luciferase owing to the structural similarity of the inhibitors such as R16b to D-luciferin that is the substrate of firefly luciferase [47] and to 3-(5-(4-(trifluoromethyl)phenyl)-1,2,4-oxadiazol-3-yl)benzoic acid that is a known inhibitor of firefly luciferase [48] (Figure 7).
Twenty 10-ns (1.0-fs time step) molecular dynamics simulations of the R16b-bound firefly luciferase showed that R16b binds at the luciferase active site in almost the same way as D-luciferin does (Figure 7 and Datasets S1 and S2).The average intermolecular interaction energy calculated from 20,000 conformations of the R16b-bound luciferase obtained from the simulations is of 2134.2 kcal/mol, whereas the corresponding energy computed from 7,000 conformations of D-luciferin-bound enzyme obtained from seven 10-ns (1.0-fs time step) molecular dynamics simulations is 2102.0kcal/mol.These results suggested that R16b could bind to the luciferase active site.Indeed, subsequent experimental studies showed that R16b had two direct effects on the firefly luciferase activity in a dose-dependent manner.As apparent from Figure 8, R16b first increased and then decreased the luciferase activity as the concentration of R16b gradually changed from 0 to 10 mM, and this bell-shaped dose response is most noticeable when the luciferase concentration is low (0.74-0.19 ng of firefly luciferase).These results demonstrate that R16b has direct interactions with firefly luciferase.
To confirm that our compounds directly inhibit RTA as well, we performed cell titer 96 AQuaeous non-radioactive cell proliferation assays (Promega, Madison, Wisconsin) using Sp2 mouse myeloma cells [49] to test the ability of these compounds to protect cells against ricin.This assay determines cell viability by measuring the absorbance of the formazan product produced by viable cells rather than detecting cellular adenosine 59-triphosphate (ATP) levels through the use of firefly luciferase.At a concentration of 300 nM, these inhibitors showed 0.7-15.7%cell protection against ricin (Table 2), confirming that these compounds are capable of inhibiting RTA in the absence of firefly luciferase.Most of these inhibitors showed similar cell protections at 3 mM, 30 mM, and 300 nM (Table 2), suggesting possible interactions of these inhibitors at high concentrations with other off-targets in the Sp2 mouse myeloma cells.Superimposition of the crystal structures of RTA (PDB ID: 1IFS [32]) and Stx2 (PDB ID: 1R4P [50]) showed that the conformations of active-site residues of Stx2A1 are similar to those of RTA (Figure 9), although the sequence identity of the two proteins is only 18%.This similarity prompted us to test our inhibitors against Stx2 both in vitro and ex vivo.Indeed, R16b and R22 demonstrated 1.4-and 1.3-fold increases in luciferase activity resulting from the translation in the RRL after treatment with 0.5 nM Stx2 (activated by reacting with trypsin and dithiothreitol and 1 nM inhibitor) relative to the activity after treatment with 0.5 nM activated Stx2 only (Table 1).Furthermore, R22 showed 21% cell protection against Stx2 at a drug concentration of 300 nM.Of mechanistic importance, among the nine inhibitors tested, R22 is the most potent in inhibiting Stx2, whereas R19b is the most potent in inhibiting ricin at the same drug concentration, which demonstrates the preferential interactions of the tested inhibitors with RTA and Stx2A1.

Promising RIP Inhibitor Leads
We have two lines of evidence for the significant and direct inhibition of ricin and Stx2 by our inhibitors at an inhibitor concentration of 1 or 300 nM, despite the unexpected interactions between our inhibitors and firefly luciferase.First, these inhibitors showed two different structure-activity relationships for RTA and activated Stx2 assayed under the same conditions involving firefly luciferase and substrate D-luciferin.If the observed enhancement of the firefly luciferase activity by our inhibitors were solely due to the interaction with the luciferase, it would be highly improbable to have two different structure-activity relationships (Table 1).Second, our inhibitors exhibited up to 20% cell protection against ricin or activated Stx2 under the assay conditions that are devoid of the luciferase.
Inhibitors of firefly luciferase are known to have the ability to increase the luciferase activity through a modest increase of the enzyme half-life by reversible binding to the enzyme [48,51,52].In this study we found that R16b first increases and then decreases the low-concentration luciferase activity as the concentration of R16b changes from 0 to 10 mM (lower panel of Figure 8).Although further experimental verification is needed, these results may explain why the luciferase activity in the RRL treated with R16b alone increases as the concentration of R16b increases, whereas the luciferase activity in the RRL treated with RTA and R16b decreases as the R16b concentration increases (Figure 6).When the RRL was treated with 1-100 nM R16b alone, R16b was presumably in shortage relative to an ample amount of firefly luciferase resulting from the translation in the RRL.In this case, a relatively small amount of R16b could modestly enhance the halflife of the enzyme but would not be enough to inhibit the binding of D-luciferin substrate to the enzyme; therefore, R16b could increase the luciferase activity.When the RRL was treated with 1 nM RTA and 1-100 nM R16b that inhibits RTA, R16b was presumably in excess relative to a residual amount of the luciferase resulting from the translation in the RRL that was incompletely inhibited by RTA in the presence of R16b in low concentrations.In this case, a relatively large amount of R16b decreased the enzymatic activity because the inhibition of D-luciferin binding to the enzyme by R16b outweighs the increase of the luciferase half-life by R16b.Taken together, our cell-free and cell-based studies as well as the hypothetical mechanisms of the inhibitor concentration effects on the luciferase activity suggest that R16, R16b, R19, R19b, R19c, R19d, R20, R20b, and R22 are promising inhibitor leads of ricin and Stx2.

The Doorstop Approach to Inhibiting ProteinNPolynucleotide Functions
In this work, we used the doorstop approach to identify small molecules that work as doorstops to prevent the active-site residue Tyr80 in RTA from adopting its active conformation, thereby blocking the function of the protein rather than work as a contender in the binding competition.Further studies are needed to validate the approach and determine the scope of its application.At this stage, informed by the fortuitous finding of R20 as an RIP inhibitor and the complication of unexpected interactions with firefly luciferase, it is reasonable to question whether the identification of R16, R19, and R22 as RIP inhibitor leads resulted from sheer luck or the use of the doorstop approach.
For the following reasons, we attribute the finding of these leads to the doorstop approach.
Our screening work hinges on the assumption that conformation 2 is a bound conformation that inhibits catalysis.If this assumption were incorrect, our RIP inhibitor discovery would be serendipitous.As apparent from the superimposition of the crystal structure of RTA liganded with a cyclic tetranucleotide (PDB ID: 3HIO [13]) over the crystal structures of RTAs in complex with small-molecule inhibitors (PDB IDs: 1IFS [32], 1IFT [32], and 1FMP [31]) shown in Figure 2d, Tyr80 in conformation 2 (found in the 1IFS crystal structure) clashes with the nucleotide of the 3HIO crystal structure at the Michaelis-Menten state, whereas Tyr80 in conformation 3 (found in the 1FMP crystal structure) overlays well with that of the 3HIO crystal structure.The clash of Tyr80 in conformation 2 with the nucleotide substrate supports the assumption that conformation 2 inhibits catalysis.This suggests that the discovery of the RIP inhibitors was likely conferred by the doorstop approach and that the doorstop approach might be useful for developing inhibitors of other proteinNpolynucleotide functions.

Potential Applications of the Doorstop Approach
Without knowing whether a molecule can grip conformation 2 by binding atop the phenolic ring of Tyr80, we initially wanted to design a clip-like molecule to stabilize conformation 2 using two functional groups simultaneously binding on both sides of the phenolic ring.For this reason, we performed virtual screening using the 1IFS crystal structure with conformation 2 [32], which has space to accommodate an adenine-like molecule underneath the phenolic ring of Tyr80, rather than using the 1IFT crystal structure with conformation 1 [32], which lacks space beneath the ring.However, conformation 2 may not be readily available for other proteinNpolynucleotide complexes.Its rarity may raise a concern about the generality of the doorstop approach.Interestingly, our in vitro and ex vivo testing results show that molecules such as R16b, R19b, and R22 with functional groups binding on one side of the phenolic ring appear to be effective doorstops, suggesting that effective doorstops might be obtainable from screening chemicals for their binding atop the Tyr80 ring in conformation 1, which is a common apo-state conformation.In this context, it is worth noting the following potential application of the doorstop approach.
The N-terminal domain of the E3L protein of vaccinia virus binds Z-DNA and causes pathogenicity in mice [53].In addition, the side-chain conformation of Tyr48 at the Z-DNA binding site of E3L is in an equilibrium of UP (c = x+0u) and DOWN (c = x+180u) states; the E3L protein with Tyr48 in the UP state cannot bind Z-DNA and does not cause pathogenicity whereas the one in the DOWN state binds Z-DNA and causes pathogenicity [54].Furthermore, the amino acid sequence of vaccinia virus E3L is almost identical to those of poxviruses [55].These results suggest that an active-site Tyr residue may serve as a common switch for functions of proteinNpolynucleotide complexes.The doorstop approach might be applicable for the development of other proteinNpolynucleotide inhibitors-such as small-molecule inhibitors of Z-DNA binding proteins in poxviruses-as antiviral agents.

Caveats for the Use of Chemical Screens for Potential Drug Leads
In our previous virtual screen for farnesyltransferase inhibitors, we found that six out of 27 compounds purchased from chemical vendors had serious chemical identity or purity issues [38].In this study, of the two chemicals (R20 and R22) that required spectroscopic analyses to confirm their stereochemistry, one (R20) turned out to have incorrect stereochemistry.These results call for careful chemical characterization of leads identified from chemical screens using spectroscopic analyses to avoid issues of chemical identity and purity.
Having long been used in HTSs for drug leads, luciferase is a reporter that detects the transcriptional activity in translation assays and measures cellular ATP levels in cell viability assays or kinase activity assays [56].However, only recently has attention been directed to interference caused by direct interactions between luciferase and chemicals being screened [48,51,52,57].The experimentally confirmed direct interaction of R16b with firefly luciferase reported herein reinforces the reported advocation that appropriate control studies must be performed before interpreting HTS results [48].

Conclusion
Using the doorstop approach that aims to identify molecules that can prevent an RIP active-site residue from adopting an active conformation, we identified chemicals with significant in vitro inhibitory potency at nanomolar concentrations and up to 20% cell protection against ricin and Stx2 at an inhibitor concentration of 300 nM.This work offers promising leads for structural optimization to achieve better cell protection.The results suggest that the doorstop approach might be applicable to the development of small-molecule inhibitors of poxvirus Z-DNA binding proteins as anti-poxvirus agents.This work also calls for careful chemical and biological characterization of leads obtained from HTSs to avoid identification of irrelevant chemical structures and interference caused by unintended interactions with luciferase.

Reagents and Toxins
Hexanes (Hex), ethyl acetate (EtOAc), and trifluoroacetic acid (TFA) were purchased from Fisher Scientific (Pittsburgh, PA).All commercially available reagents such as dimethyl sulfoxide (DMSO) were used as received.Rabbit reticulocyte cell-free lysate (2 parts water and 1 part lysate) was obtained from Green Hectares (Oregon, WI).Ricin and RTA were purchased from Vector Laboratories (Burlingame, CA).Stx2 holotoxin was provided by the Phoenix Laboratory (Tufts-NEMC Microbial Products & Services Facility).Hybridoma Serum Free Medium, Glutamax, and a mixture of penicillin and streptomycin were obtained from Invitrogen (Carlsbad, CA).The Cell Titer 96 AQuaeous Non-Radioactive Cell Proliferation Assay reagents were purchased from Promega (Madison, WI).

in Silico Screening
The two-stage docking of 236,925 small molecules into the active site of RTA was carried out by using the EUDOC program [35][36][37] according to a published protocol [35].The translational and rotational increments at the first stage were 1.0 A ˚and 10 degrees of arc, respectively, and default increments were used at the second stage.A cutoff of 250 kcal/mol for intermolecular interaction energies was used.The 236,925 small molecules were selected from an in-house database of 2.5 million small molecules using the criterion that each selected molecule has a molecular weight less than 301.All small molecules to be screened were protonated or deprotonated according to physiological pH of 7.4 and their three-dimensional structures and atomic charges were obtained from AM1 semi-empirical calculations.Conformations of RTA and small molecules were not allowed to change during the docking.A docking box (6.063.566.0A ˚3) was defined to confine the translation of the mass centre of each molecule within the active site of RTA crystal structure (PDB ID: 1IFS [32]).The box was surrounded by Asp100, Ile-104, Asp75, Asn78, Tyr80, Val82, Phe93, Gly120, Gly121, Asn122, His94, Pro95, and Asp96 whose conformations were defined in the 1IFS structure (see Figure 2c).All water molecules and the bound adenine were removed from the 1IFS structure.The docking studies were performed on a dedicated cluster of 470 Xeon Processors (2.2/2.4GHz).

Simulations of Firefly Luciferase Liganded with R16b or D-luciferin
Model Preparation.The atomic charges of R16b and Dluciferin were generated according to the RESP procedure [61] with ab initio calculations at the HF/6-31G*//HF/6-31G* level using the Gaussian 98 program [62].The force field parameters including the charges of R16b and D-luciferin are included in Datasets S1 and S2, respectively.The starting structure of firefly luciferase in complex with R16b was generated by manually docking R16b into the active site of the luciferase structure that was taken from the crystal structure of luciferase bound with 59-O-[N-(dehydroluciferyl)-sulfamoyl]adenosine (PDB ID: 2D1S [47]).In the manual docking, the carbonyl carbon atom of R16b was placed near the carbonyl carbon atom of D-luciferin in the 2D1S crystal structure, and the alkyl-substituted aromatic carbon atom was placed near the hydroxy-substituted aromatic carbon of the D-luciferin structure.The starting structure of luciferase in complex with D-luciferin was extracted from the 2D1S crystal structure.For the luciferase structure, all histidine residues were treated as HID, and crystallographically determined water molecules were removed.The topology and coordinate files of luciferase in complex with R16b or D-luciferin were generated by the PREP, LINK, EDIT, and PARM modules of the AMBER 5.0 program [63].The complex was refined by energy minimization using a dielectric constant of 1.0 and 100 cycles of steepest-descent minimization followed by 100 cycles of conjugate-gradient minimization.The refined complex was solvated with 13,617 and 13,540 TIP3P water molecules (named WAT) [64] for R16b and D-luciferin, leading to a system of 12,802 and 12,842 atoms, respectively.The WAT molecules were obtained from solvating the complex using a pre-equilibrated box of 216,000 TIP3P molecules, whose hydrogen atom charge was set to 0.4170, where any water molecule was removed if it had an oxygen atom closer than 2.2 A ˚to any solute atom or a hydrogen atom closer than 2.0 A ˚to any solute atom, or if it was located further than 8.2 A ålong the x-, y-, or z-axis from any solute atom.Multiple Molecular Dynamics Simulations.The solvated complex system was energy-minimized for 100 cycles of steepestdescent minimization followed by 100 cycles of conjugate-gradient minimization to remove close van der Waals contacts in the system, then heated from 0 to 300 K at a rate of 10 K/ps under constant temperature and volume, and finally simulated independently with a unique seed number for initial velocities at 300 K under constant temperature and pressure using the PMEMD module of the AMBER 8.0 program [65] with the AMBER force field (ff99SB) [66,67].All simulations used (1) a dielectric constant of 1.0, (2) the Berendsen coupling algorithm [68], (3) a periodic boundary condition at a constant temperature of 300 K and a constant pressure of 1 atm with isotropic moleculebased scaling, (4) the Particle Mesh Ewald method to calculate long-range electrostatic interactions [69], (5) a time step of 1.0 fs, (6) the SHAKE-bond-length constraints applied to all the bonds involving the H atom, (7) saving the image closest to the middle of the ''primary box'' to the restart and trajectory files, (8) formatted restart file, and ( 9) default values of all other inputs of the PMEMD module.Twenty different molecular dynamics simulations (each lasted 10 ns) were carried out for luciferase in complex with R16b, and seven difference simulations were performed for luciferase bound with D-luciferase.
Structure Analysis.The average structure of the simulations for each luciferase complex was obtained by using the PTRAJ module of the AMBER 11 program [65].The average intermolecular interaction energy for each complex was obtained using an in-house program (ISE, wrote by Yuan-Ping Pang) that computes the average of the intermolecular interaction energies of all trajectories saved at the 1-ps intervals during the last 1-ns period of the 20 (for the R16b complex) or 7 (for the D-luciferin complex) different simulations.

Cell-Free Assay for Inhibition of RTA and Stx2
N-Glycosidase Preparation.A stock solution of RTA (7.5 mM) was prepared using phosphate buffered saline solution (PBS).A stock solution of activated Stx2 was prepared according to published reports [70][71][72] with slight modifications.Briefly, lyophilized Stx2 powder was re-suspended in PBS to prepare a 2-mM solution.From this solution, 2 mg of Stx2 was incubated with trypsin (prepared in 1 mM HCl, pH 3.0 and used at a final concentration of 2 ng/mL) and 30 mM dithiothreitol for 2 hours at 37uC.The treatment with trypsin was stopped by adding phenylmethylsulfonyl fluoride at a final concentration of 0.04 mg/ mL.The concentrated stock solutions of RTA and activated Stx2 were maintained at 4uC.The diluted working solutions of the two N-glycosidases were disposed after each experiment.
Inhibitor Preparation.R16, R19, and R20 in their acid form were dissolved in neat DMSO to a final concentration of 10 mM.Stock solutions of these compounds (1 mM) were prepared in 10% DMSO.The final DMSO concentration for the in vitro translation reaction assay was set at 0.67% DMSO.R16b, R19b, R19c, R19d, R20b, and R22 in their sodium salt form were dissolved in double distilled water at 5.06 mM, 60 mM, 1.41 M, 37 mM, 0.31 M, and 46 mM, respectively.Stock solutions of the water-soluble compounds were prepared at 1 mM.
Cell-Free Translation Assay.The final protein concentration of the diluted RRL was adjusted to approximately 80 mg/mL as measured using a spectrophotometer (OD 280 ).The

Figure 3 .
Figure 3.Chemical structures and binding modes of smallmolecule inhibitors of ricin subunit A. The complexes were generated using the EUDOC program with the 1IFS crystal structure [32].The residues from left to right are Phe93, Tyr80, Arg258, and Arg213.doi:10.1371/journal.pone.0017883.g003

Table 1 .
Increase of rabbit reticulocyte lysate in vitro translation caused by exposure to 1 nM ribosome-inactivating protein inhibitor and 1 nM ricin subunit A (RTA) or 0.5 nM Shiga-like toxin 2 (Stx2) relative to the exposure to 1 nM RTA or 0.5 nM Stx2.