Fatty acid synthase (FASN, EC 2.3.1.85), is a multi-enzyme dimer complex that plays a critical role in lipogenesis. This lipogenic enzyme has gained importance beyond its physiological role due to its implications in several clinical conditions-cancers, obesity, and diabetes. This has made FASN an attractive pharmacological target. Here, we have attempted to predict the theoretical models for the human enoyl reductase (ER) and β-ketoacyl reductase (KR) domains based on the porcine FASN crystal structure, which was the structurally closest template available at the time of this study. Comparative modeling methods were used for studying the structure-function relationships. Different validation studies revealed the predicted structures to be highly plausible. The respective substrates of ER and KR domains-namely, trans-butenoyl and β-ketobutyryl-were computationally docked into active sites using Glide in order to understand the probable binding mode. The molecular dynamics simulations of the apo and holo states of ER and KR showed stable backbone root mean square deviation trajectories with minimal deviation. Ramachandran plot analysis showed 96.0% of residues in the most favorable region for ER and 90.3% for the KR domain, respectively. Thus, the predicted models yielded significant insights into the substrate binding modes of the ER and KR catalytic domains and will aid in identifying novel chemical inhibitors of human FASN that target these domains.
Fatty acid synthase (FASN, E.C. 2.3.1.85) is a multi-enzyme complex that synthesizes endogenous fatty acids. Seven cycles of FASN-catalyzed reactions result in the conversion of acetyl-CoA and malonyl-CoA into 16-carbon palmitate [
Two major forms of FASN are known: FASN type I (FASNI) and type II. FASNI is a multimeric multi-enzyme complex involved in the synthesis of palmitate in an integrative manner on a single polypeptide chain (
Besides the important biochemical role that FASN plays in lipid biogenesis, this complex enzyme has been implicated in several pathological conditions. In comparison with normal cells, the overexpression of FASN has been reported in several types of cancers, including prostrate, breast, ovarian, and colon cancers [
In this study, we have attempted to predict the 3D structure of the enoyl reductase (ER) and KR domains of human FASN in unbound form by comparative modeling. At the time of this model development, the structurally closest template was the reported crystal structure of porcine FASN (PDB ID: 2VZ8, 3.2 Å). This mammalian FASN template revealed a complex architecture, covering five catalytic domains and also the inter-connecting linkers [
In order to validate the biological closeness of the predicted
The amino acid sequences of the ER and KR domains of human FASN, were retrieved from Uniprot Kb/Swissprot server (uniprot ID: P49327). The sequences of the ER and KR domains were 229 and 255 amino acids in length, respectively. The retrieved sequences were BLAST-analyzed against PDB towards identifying suitable templates for homology modeling. From the available crystal structures deposited in PDB during the time of this study (early 2014), the crystal structure of mammalian (porcine,
The pairwise sequence alignments between the template and the targets (human ER and KR) were built using MODELLER 9v7 [
The validations of structural geometric properties, like backbone conformation, and of the compatibility of residue interactions were performed using the Structural Analysis and Verification Server (SAVES;
MD simulations for the modeled proteins were carried out using the Desmond program, an explicit solvent MD package (version 3.1, Desmond Molecular Dynamics System; D. E. Shaw Research, New York, NY, USA and version 3.1, Maestro-Desmond Interoperability Tools; Schrödinger) with inbuilt optimized potentials for liquid simulation (OPLS 2005) force field [
Site Map 2.6 (Schrödinger, LLC) was used to identify the active sites on the predicted models. Various countermaps were also generated to distinguish the hydrophobic and hydrophilic regions on the active site regions. Finally, the best binding pockets were ranked based on the Sitemap score.
The substrates for the ER and KR domains-namely, the
Receptor grid files for glide docking were generated to cover the volume of the predicted active sites. Here, van der Waals radius is scaled to 1.0 with a partial cut-off of 0.25 to soften the potential for non-polar parts of a receptor, where other atoms are free of scaling. The receptor grid file and the prepared substrate were docked using Glide standard precision, where the ligand sampling was set to be flexible, ensuring the sample ring conformation and nitrogen inversions. Ligands were set to select only less than 300 atoms and less than 50 rotatable bonds with a van der Waals scaling factor of 0.8 with a partial cut-off of 0.15. Out of the 1,000 poses generated per docking run, 10 energetically favorable poses per ligand were selected. Glide score, an empirical docking scoring function that implements the OPLS 2005 force field, was used to infer the affinity and binding mode of the substrate. The best docked substrate conformation for the ER and KR (holo) domains was subjected to MD simulation for 5 ns, similar to that of the methods discussed for MD simulations of the apo form, except that maximum iterations of 2,000 steps were applied on solute heavy atoms alone with a convergence threshold as 1.0 kcal/mol/Å.
The amino acid sequences of the human FASN ER and KR domains were retrieved from Uniprot (P49327). Comparative modeling method was implemented to predict the structures of the ER and KR domains. The structural templates for modeling the query sequences were searched against PDB using BLASTP. Many of the hits obtained were similar to the query sequences, mainly representing the reductase families. The best templates were selected in accordance with the optimal pairwise alignment, sequence coverage, and sequence conservation. Accordingly, mammalian (porcine,
We had attempted to model the protein structure of human FASN ER domain earlier for the purpose of understanding its bonding interactions with the known inhibitor triclosan [
The predicted structure of the human ER domain showed a Rossmann fold at region 1651 to 1794, which harbors a nucleotide binding cavity (1671 to 1688) and also overlapping substrate binding cavities at 1650-1653 and 1795-1863, as proposed by Maier et al. [
PDBsum server, which uses Gail Hutchinson's PROMOTIF program to compute the secondary structure motif information of the models generated (
The initial models were subjected to different structure evaluation tools for understanding the geometries, backbone configuration, dihedrals, and residue-residue interactions. The Ramachandran plot for the ER domain showed 0.5% residues in a disallowed region. In the case of the KR domain, 0.0% of the residues were in disallowed regions. Hence, these models were further refined using the WHAT IF program to remove the atomic bumps and subsequently loop-refined using MODELLER 9v7 scripts and the MODREFINER algorithm. The final refined structures of both domains showed no residues in the disallowed region of the Ramachandran plot (
The overall stereochemical parameters for the modeled proteins were measured using G-factor and ERRAT calculations by PROCHECK. G-factor is a measure of the proper dihedrals and covalent bond orders, and it is expressed overall as log-odd score. G-factor scores for optimal structures range from 0.0 to 0.1 with increasing order of confidence. A negative G-factor score indicates improper conformation of the residues, while higher positive scores indicate proper conformation. In the case of the predicted model of the ER domain, the overall log-odd score was found to be 0.30, which indicates a high plausibility of the structure with negligible improper conformations. Similarly, the KR model also showed an overall G-factor score of 0.22, suggestive of a high plausibility of the structure. Moreover, the non-bonded atomic interactions of the models were analyzed using the ERRAT tool, wherein the overall quality score for the predicted structures of the ER and KR domains were found to be 95.92 and 80.85, respectively. The ERRAT scores for both models were above 50 and are considered a standard for good models. To validate this further, the modeled structures were also assessed for their quality using the ProQ server. ProQ assessment for the ER model predicted the LGscore to be 4.362 (>4.0, extremely good model) and the MaxSub score to be 0.416 (>0.1, fairly good model). Similarly, for the KR domain, the LGscore was 3.509 (>2.5, very good model) and the MaxSub score was 0.246 (>0.1, fairly good model), suggesting higher plausibility of the models.
MD simulation was carried out for the predicted models to understand the stability and conformational changes of the modeled proteins in holo and apo forms. The simulation was carried out in a water (SPC-molecule)-solvated system with optimal physiological conditions, wherein the temperature and pressure were maintained at 300 K and 1 atm for both the ER and KR domains. The computed total energy and potential energy remained stable from start of the dynamics simulation until the course of the the 5-ns simulation in the case of both proteins. The ER domain had a total energy of -82,181.468 kJ/mol and a potential energy of -101,788.978 kJ/mol. Similarly, KR also had stable distribution with a total energy of -141,285.655 kJ/mol and potential energy of -174,168.487 kJ/mol. The event trajectory was observed to remain stable throughout the simulation process. Further, the stability of the modeled proteins was verified by plotting the root mean square deviation (RMSD) graph for backbone atoms during the production run. The RMSD for the ER domain remained stable for the 5-ns time frame with minimal deviation until the end of 5 ns (
Site Map 2.6 (Schrödinger, LLC) was used to predict the active sites in the modeled proteins. The presently modeled human FASN ER domain comprises two subdomains forming the Rossmann fold and substrate binding fold. According to Maier et al. [
Normally, in FASN-catalyzed lipid biosynthesis, seven cycles of two carbon additions are essential for synthesizing the 16-carbon palmitate. The respective initial physiological substrates for the KR and ER domains-namely, β-ketobutyryl and
Finally, both docked complexes were validated for the stability of complex formation, implementing a MD simulation using DESMOND. The simulations for both holoenzyme complexes were carried out for 5 ns. The RMSD plot for all atoms of each holoenzyme during the production run was analyzed. The plots showed no significant increase in deviation until the completion of 5 ns (
Hence, all of these findings strongly suggest the highly stable complex formation of the ER and KR domains with their respective substrates (
Based on the
The work described in this article was supported by the Department of Biotechnology, Ministry of Science & Technology, New Delhi for program support on retinoblastoma (Project sanction order no: BT/01/CEIB/11/V/16, dated 08/05/2012).