3 Methods
3.1 Virtual Screening 1. The database including 0.5 million compounds from Specs
Ltd. is first converted into 3D structures using the CON-
CORD program [27] and filtered using drug-like property
criteria [28] by the FILTER 2.0.1 software [29].
- The protein structure is added with hydrogen atoms. All atoms
are assigned Kollman-all charges by the SYBYL 7.1 program
(SYBYL 7.1, 2005 Tripos Inc., St. Louis, MS). - The 3D structures of the small molecules are added with
hydrogen and assigned AM1-BCC partial charges. - The active site for the virtual screening is in the radius of 6 A ̊
around the center of AI-2, includes residues Pro74, Gln77,
Ser79, Asp80, Tyr81, Trp 82, Pro109, Asn159, His180,
Phe206, Ile211, Arg215, Cys264, Ser265, Thr266, Asp267,
Trp289, Gly290, Gly291, Glu295, and Arg310. In the LuxPQ
holo-form crystal structure, the borate moiety of AI-2 generates
several hydrogen bonds with the positively charged residues
Arg215 and Arg310 for stabilizing the negative charge. Mean-
while, this borate group can also form hydrogen bonding with
Ser79 and Thr266. Further, residues Trp82, Ser79, and Gln77
provide multiple hydrogen bonds with the two hydroxyl
groups of AI-2 also have, and Asn159 and Arg215 has hydro-
gen bonds with the oxygen atom of furanoyl ring (Fig.3). - The position and conformation of each compound are opti-
mized first by the anchor fragment orientation and then by the
torsion minimization method implemented in the DOCK6
program [30, 31]. - Each compound is generated with 50 conformations and a
maximum of 100 anchor orientations. By the way, all of
the docked conformations are energy minimized by 100 itera-
tions [31]. - The docked molecules are ranked based on the sum of thevan
der Waalsand electrostatic energies to get the top 1000 hits. - In order to identify a specific association between the ligands
and LuxP, the resulting structures are analyzed using PyMOL
0.99, HBPLUS 3.06, and Ligplotþ1.4.5.
3.2 Compounds
Selection
- Carry out a consensus scoring evaluation of the top 1000 hits
from virtual screening results and select the top 10% com-
pounds (seeNote 2). Our consensus scoring efforts consisted
of ChemScore, OEChemscore, PLP, ScreenScore, Chem-
Gauss3, CGO, and ShapeGauss implemented in the FRED
2.2.3 software (FRED 2.2.3, 2007 OpenEye Scientific Soft-
ware, Inc., Santa Fe, NM).
In silicoIdentification of AI-2 Inhibitors 357