ABCC5 belongs to the ABC transporters, which are structurally related membrane proteins featuring intracellular motifs that exhibit ATPase activity [1] . This nucleotide binding domain (NBD), which contains the Walker A and Walker B motifs, cleaves ATP’s terminal phosphate to energize the transport of substrate molecules against a concentration gradient. ABC genes are highly conserved within species, indicating that these genes may have been present since the beginning of eukaryotic evolution [1] [2] . The putative structure of ABCC5 consists of 12 transmembrane helices (TMH) and 2 NBDs (Figure 1).

Clinical studies have shown that extracellular cGMP levels are elevated in various types of cancer [3] - [6] . The human ABC transporter ABCC5 transports cGMP out of cells, and the increased cGMP efflux from cancer cells may be caused by an up regulation of ABCC5. The ability of ABCC5 to transport cGMP suggests that it may be playing a biological role in cellular signaling by eliminating the nucleotides from the cells [7] . Increased cGMP efflux by ABCC5 may be one mechanism whereby cancer cells can develop resistance against endogenous growth control, and also against antineoplastic drugs which are substrates for ABCC5. cGMP regulates several apoptosis-associated genes, and therefore intracellular elevation of cGMP concentration by inhibition of ABCC5 efflux may give cytotoxic effects [8] .

No X-ray crystal structure of ABCC5 has been reported, but molecular models of ABCC5 may be constructed by homology using a known 3D crystal structure of an evolutionary related protein as a template. Different models of ABCC5 [9] [10] are previously presented in outward-facing conformations and inward-facing conformations based on the Staphyllococcus aureus ABC transporter Sav1866 [11] , which has been crystallized in an outward-facing ATP-bound state, and on the Escherichia coli MsbA, which has been crystallized in a wide open inward-facing conformation [12] . In this study, multiple homology models of ABCC5 are constructed on the X-ray crystal structure of five different ABC transporters (PDB IDs: 4F4C, 4Q9H, 4M1M, 4M2T and 4KSD) and all models are ranked as per their structural quality. The template resolution of different crystal structures ranged from 3.8 - 4.1 Å and the sequence identity between templates and human ABCC5 are in range 20.7% - 22.8%, clearly implying that the ABCC5 homology models has elements of uncertainty.

The best homology model of ABCC5 was model number 22 (Table 1 lists all models that were built on different templates) and this model was based on crystal structure of P-glycoprotein (PDB ID: 4Q9H) of 3.4 Å (Figure 2). The model was in an inward facing conformation with the NBD’s separated by approximately 26 Å. A large internal cavity open to the cytoplasm was formed by 2 TMD. TMD 1 comprise of TMHs 1, 2, 3, 6, 10, 11 and TMD 2 comprise of TMHs 4, 5, 7, 8, 9, 12 as shown in Figure 3. The secondary structure of the model is depicted in Figure 4.

The Walker A motifs consisted of a coiled loop and a short α-helix (P-loop), and the Walker B motifs were in β-sheet conformation and localized in the NBD’s hydrophobic cores, which consisted of 5 parallel β-sheets and 1 anti-parallel β-sheet. In NBD1 the Walker A has residues from Gly595-Thr602, Q-loop from Gln630-Asn636, C motif from Leu685-Arg692, Walker B from Ile705-Asp709, D-loop from Ser713-Asp716, H-loop from Thr741-Gln743 and in NBD2, the Walker A starts from Gly1227-Ser1234, Q loop from Gln1275-Gly1282,

C motif from Phe1330-Gln1336, Walker B from Ile1350-Asp1354, D-loop from Ala1358-Asp1361 and H-loop from Ala1385-Arg1387. The connecting extracellular loop between TMH7 and TMH8 in the model was about 39 amino acids long (Ser872-His910) and this was consistent with the PROSITE [14] predictions of TMH from 848 - 868 and 917 - 937 in ABCC5. NBD1 and TMD2 were connected with 2 α-helices (Thr800-Lys821, Gln832-Gly837) and three loops (Asn783-Glu799, Ala822-Val831 and Gln838-Trp843). The fact that needs to be considered here that this long loop was not present in the template, and that modeling loops of this length is relatively inaccurate and consequently the modeled loop structures must be regarded as ambiguous. The loop is approximately 42 Å from the binding pocket, and accordingly, the inclusion of this loop may not be necessary for the purpose of any future studies like docking on this model.

The Errat option of the Structural Analysis and Verification Server (SAVES)

http://nihserver.mbi.ucla.edu/SAVES/ reported that the overall quality factor of the ABCC5 model was 98.022,

and a value above 90 indicates a good model. According to the Ramachandran plot (Figure 5) provided by the Procheck option, 91.2% of the ABCC5 residues were in the most favored regions, 7.6% were in additional allowed regions, 0.6% were in generously allowed regions, and 0.6% were in disallowed regions. The overall G-factor value of the model was −0.17 indicating that geometry of model corresponds to the probability conformation with 91.2% residue in the core region showing high accuracy of model prediction.

The summary of the what check option reported that the ABCC5 model was found to be satisfactory. ProSA was used to check the three dimensional model of ABCC5 for potential errors [15] . The program displays 2 characteristics of the input structure: its Z-score and a plot of its residue energies. The Z-score of −11.61 indi

cates the overall model quality of ABCC5 (Figure 6) and it also measures the deviation of total energy of the structure with respect to an energy distribution derived from random conformations. The score is indicating a highly reliable structure that is well within the range of scores typically found for proteins of similar size. The energy plot shows the local model quality by plotting knowledge-based energies as a function of amino acid sequence position.

QMEAN analysis was also used to evaluate and validate the model [16] . The QMEAN score of the model obtained was 0.54 and the Z-score was −2.44. The QMEAN analysis reflects the good quality of the model because the estimated reliability of the model was expected to be in between 0 and 1 and this can be inferred from the density plot for QMEAN scores of the reference set (Figure 7). Comparison with non-redundant set of PDB structures in the plot between normalized QMEAN score and protein size revealed different set of Z-values for different parameters such as C-beta interactions (−2.37), interactions between all atoms (−1.40), solvation (−2.25), torsion (−3.69), SSE agreement (0.37) and ACC agreement (−1.55) which can be clearly observed (Figure 8). Some local error was also obtained for the ABCC5 model which was higher somewhere in between the residue from 1250 and 1300. The PROCHECK and QMEAN analysis of other best ranked ABCC5 models are summarized in Table 2.

All 26 models were validated using DOPE [17] which computes the energy of the tertiary structure of a protein as the sum of the energy of pairs of atoms in the protein. The detailed structural analysis of all the models is provided in Table 1. The top ranked ABCC5 model (Model no. 22) has normalized dope score of −0.43, suggesting that 60% of its C-α atoms are within 3.5 Å of their correct positions implying a reliable model. The overall Z-score of this model as determined by structural validation tools of YASARA was −1.433 and this model was ranked 1 in terms of stability and stereo chemical aspects.

The presence of a well-defined binding site, shown in Figure 9, was probed through the MOE Site Finde program. Table 3 reports the binding site score for the ABCC5 model and lists the residues composing the site. Predicted active site residues like ASN 441, GLN 1138 were consistent with ABCC5 model [9] .

Other ABCC5 models that were ranked best among the five models built on each template are shown in Figure 10. The PDB files of the ABCC5 models are available upon request.

Homology modelling of ABCC5 was performed using YASARA Structure (V. 16.3.28) that fully automatically takes all the steps from an amino acid sequence to a refined high-resolution model using a CASP approved protocol [18] . The target sequence (Uniprot ID: O15440) was PSI-BLASTed [19] against Uniprot to build a

position-specific scoring matrix (PSSM) from related sequences, then this profile was used to search the PDB for potential modelling templates. Common protein purification tags were excluded to avoid false positives. The templates were then ranked based on the alignment score and the structural quality according to WHATCHECK [20] obtained from the PDB Finder 2 database [21] . Usually models built using high-resolution X-ray templates are more accurate than those created from lower resolution X-ray or NMR templates, even if the latter share a higher percentage sequence identity. Models were finally built for the top scoring five templates in this study (Table 4).

To aid alignment correction and loop modelling, a secondary structure prediction for the target sequence was obtained by running PSI-BLAST to create a target sequence profile and feeding it to the PSI-Pred secondary structure prediction algorithm [25] . For each available template, the alignment with the target sequence was performed after taking into account additional information like sequence-based profiles of target and template calculated from related Uniprot sequences, optionally augmented with structure-based profiles from related template structures. The alignment of the ABCC5 sequence with all five templates is depicted in Figure 11.

The alignment also considered the structural information contained in the template (avoiding gaps in second- dary structure elements, keeping polar residues exposed etc.), as well as the predicted target secondary structure [27] . This structure-based alignment correction is partly based on SSALN scoring matrices [28] . If the alignment is not certain, alternative high-scoring alignments were created using a stochastic approach [29] , and models were built for all of them. If templates exist in oligomeric states (according to the PQS database), models may

be built in the same state, so that interactions between side-chains across the interface can be considered. This

includes all kinds of hetero-oligomers, e.g. a homo-dimer of two hetero-dimers. In case of insertions and deletions, an indexed version of the PDB is used to determine the optimal loop anchor points and collect possible loop conformations. If templates contain ligands, these molecules were parameterized and fully considered in the homology modelling procedure, including hydrogen bonding and other interactions with the peptide chain. After initial model building the loops were optimized by trying hundreds of different conformations and re-optimizing the side-chains for all of them. Side-chain rotamers were fine-tuned considering electrostatic and knowledge-based packing interactions as well as solvation effects. After the side-chains optimization the newly modelled parts were subjected to a combined steepest descent and simulated annealing minimization (i.e. the backbone atoms of aligned residues were kept fixed to avoid potential damage). The model’s hydrogen bonding network was optimized, including pH-dependence and ligands. An unrestrained high-resolution refinement with explicit solvent molecules was run, using the latest knowledge-based force fields. The result was validated to ensure that the refinement did not move the model in the wrong direction. The tasks above were performed for all combinations of templates and alignments, per-residue quality indicators for the resulting models were determined. A hybrid model (model no. 26) was also built in which bad regions in the top scoring model were iteratively replaced with corresponding fragments from the other models. The binding site of the model was identified through the MOE Site finder program, which uses a geometric approach to calculate putative binding sites in a protein, starting from its tridimensional structure. This method is not based on energy models, but only on alpha spheres, which are a generalization of convex hulls [30] .

In silico studies in general and molecular modeling techniques have been of great help in understanding the structure, function and mechanism of the action of proteins, particularly the membrane proteins. The present investigation was carried out with major objective to model the ABCC5 protein using different templates and subjecting models so obtained for structural validation using different analysis tools like PROCHECK, ProSA and QMEAN. ABCC5 model based on 4Q9H was found to be reliable in terms of stereochemistry with G-factor value of −0.17 indicating that the geometry of model correlated to the probability conformation with 91.2% residue in the core region of Ramachandran plot showing very high accuracy of model prediction. Z score of −11.61 predicted by ProSA represented the good quality of the model. The predicted binding site had the residues as enumerated in previously reported ABCC5 model. The predicted 3D structure will provide more insight in understanding the structure and function of the protein. Moreover, this structure can be used for drug designing or understanding the interactions of ligand with the active site residues.

Natesh Singh, (2016) Molecular Modelling of Human Multidrug Resistance Protein 5 (ABCC5). Journal of Biophysical Chemistry,07,61-73. doi: 10.4236/jbpc.2016.73006