PDZ domains are well known protein-protein interaction modules involved in various key signaling pathways that regulate essential functions, such as clustering of ion channels, protein targeting, expression of membrane receptors and cell-cell communications.
Inhibition of PDZ-peptide interactions can have important implications in the treatment of cancer, cystic fibrosis, Parkinson’s disease, Alzheimer’s disease, schizophrenia, cerebral ischaemia, pain and other disorders of the central nervous system.
Our study was focused on PDZ domains of PSD95 and MAGI1, as they were reported to play an important role in some types of cancer, neuropathic pain, reduction of consequences of stroke and human papilloma virus (HPV)-related diseases. Like many other PDZ domains, the structures of the PSD95 PDZ2 and MAGI1 PDZ1 domains were already described in literature. This allowed us to use structure-based virtual screening to computationally predict new PDZ ligands. The UNITY module of the SYBYL-X software package was used to screen a set of 50,000 diverse small-molecules obtained from The Life Chemicals Stock Collection pre-processed by means of special medicinal chemistry filters. These included PAINS and in-house developed structure filters aimed at discarding compounds with toxic and other unfavorable groups, Lipinski’s “Rule of Five” and Veber Rule.
1) First PDZ domain of Magi1(d1)
The Unity Query Model based on 2I04 PDB entry  was created as a superposition of features of the ligand structure (human papillomavirus (HPV) E6) and corresponding residues from the substrate binding groove of PDZ domain of MAGI1. Therefore, the final Unity model contained structural elements both from PDZ binding site residues and HPV E6 peptide pharmacophore (Fig. 1). The screening model included: nine hydrogen bond donor features (donor site), six hydrogen bond acceptor features (acceptor site) and two hydrophobic features (hydrophobic sites). Partial match constraints for hit compounds included at least one feature for donor and acceptor sites and two hydrophobic features (Fig. 2). After the screening, over 2,000 compounds have been selected for this Library.
|Fig. 1. UNITY search query for potential inhibitors of PDZ d1 MAGI1 with HPV E6 binding pose. The query was prepared by comparing pharmacophore model of HPV E6 and features created on the base of the binding pocket residues of MAGI1 PDZ1. Donor sites are colored in green, acceptor sites are colored in violet and hydrophobic feature is brown.||Fig. 2. Docking pose of a virtual hit molecule F2196-0085 that is characterized by three intermolecular hydrogen bonds with the substrate binding groove of PDZ (His512, Phe464 and Glu520). Hydrophobic interactions are observed in the hydrophobic pocket P0 and the hydrophobic pocket P1. QFit = 47.98.|
2) Second PDZ domain of PSD-95 (2 screening models)
When designing this library, two docking models were developed and used for screening. The first docking model was based on 2KA9 PDB entry (Fig. 3) . The query features were identified from the key residues of the PDZ domain responsible for ligand binding (Fig. 4).
UNITY query included the following (Fig. 5):
- 8 H - bond donor features, minimum 2 features match
- 4 H - bond acceptor features, minimum 2 match
- 1 hydrophobic feature match (tolerance is 2.2 Å)
Totally, at least 5 feature constraints were set to meet hit compound binding requirements.
MOLCAD Surface (solvent-accessible surface) was created using Fast Connolly function with VdW ratio 1Å.
As a result, 1,100 compounds were identified as hits.
|Fig. 3. Cypin peptide bound to PSD95 PDZ2 (2KA9). PDZ domain surface is colored by hydrophobicity from brown (hydrophobic) to blue (hydrophilic).||Fig. 4. Amino acid residues that are critical for the PDZ domain-ligand interaction.|
Fig. 5. Unity search query with surface volume of the PSD95 PDZ2. Donor sites are colored in green, acceptor sites are colored in violet and the hydrophobic feature is brown.
The second screening model was based on 1QLC PDB entry . It should be noted that there are some differences in the binding groove conformation of PDZ domain observed in 2KA9 and 1QLC (Fig. 6).
Fig. 6. Conformational differences of PSD95 PDZ d2 in PDB entries 1QLC and 2KA9.
The UNITY query contained:
- 8 H - bond donor features, minimum 1 feature must be matched
- 4 H - bond acceptor features, minimum 2 must be matched
- 1 hydrophobic feature which must be matched (tolerance is 2.5 Å)
At least 4 features must be matched by a hit compound.
MOLCAD Surface (solvent-accessible surface) was created by Fast Connolly type with VdW ratio of 1 Å.
This study enabled us to identify 1,800 hits (Fig. 7).
Fig. 7. Example of a virtual hit compound F3225-8235 binding mode. QFit = 65.03. The compound forms two hydrogen bonds with Ile114 and Asn120 of the PSD95 PDZ2 substrate binding site.
- Zhang Y., Dasgupta J., Ma R.Z., Banks L., Thomas M., Chen, X.S. Structures of a human papillomavirus (HPV) E6 polypeptide bound to MAGUK proteins: mechanisms of targeting tumor suppressors by a high-risk HPV oncoprotein. J. Virol. 2007, 81, pp. 3618–3626.
- Wang W.N., Weng J.W., Zhang X., Liu M.L., Zhang M.J. Creating conformational entropy by increasing interdomain mobility in ligand binding regulation: a revisit to N-terminal tandem PDZ domains of PSD-95. J. Am. Chem. Soc., 2009, 131, pp. 787–796.
- Tochio H., Hung F., Li M., Bredt D.S., Zhang M. Solution structure of the second pdz domain of postsynaptic density-95. J. Mol. Biol., 2000, 295 (2), pp. 225–237.