Comparison and construction of membrane-bound fatty acid desaturase model
As the only active fatty acid desaturase complex crystal structure, the muSCD1_SA-CoA complex was parsed by X-ray diffraction. MpFADS6 is a membrane-bound desaturase that displays <20% identity with proteins in the PDB and only 14.9% identity with muSCD1 (Figure S1). Therefore, it is necessary to compare the models constructed using different modeling programs to obtain a highly accurate fatty acid desaturase model.
The SCD_SW homology model was constructed using the Swiss-model, which used the template manual screened in PDB. Models SCD_CI, SCD_CA, SCD_RB, SCD_CQ, and SCD_trR were auto-generated by C-I-TASSE, CATHER, Robetta, C-QUARK, and trRosetta, respectively. The global RMSD between the crystal and the five candidate models built by each program is shown in Fig. 1A. The model SCD_CI was closest to the crystal. However, this method relies on multiple templates or protein fragments to construct the final model and the 4YMK template was used to construct the SCD_CI. To fit FADS6, the following comparison excluded models that used 4YMK as a template. SCD_RB and SCD_trR were closer to muSCD1 in Fig. 1A. The TM-scores of both were >0.5, indicative of similarity with the crystal (Table 1). Alignment of muSCD1, SCD_RB, and SCD_trR (Fig. 1B) indicated that the RMSD of the transmembrane domains of these three models was <1 Å, indicating that they were structurally conserved regions. The conformational consistency of these domains was higher than that of the cytoplasmic region interacting with the substrate. Although the global RMSD of SCD_trR reached 3.04 Å, the TM score showed that SCD_trR was highly similar to muSCD1. RMSD is the average distance between atoms of the system, and it seems that the loose regions in the overall protein, especially the N- or C-terminus, increased the global RMSD value
To explore whether SCD_RB and SCD_trR can reflect the actual structure, SA-CoA was imported into the model and the distance between the interaction sites was analyzed. In Figure S2, We compared the changes in the spatial position of the residues that interact with the substrate SA-CoA which have been reported (Bai et al. 2015). It shows that the RMSD of the substrate tunnel between SCD_trR and SA-CoA was <1 Å, compared to 3.8 Å for the SCD_RB model (Table S2). In contrast, the distance between the residues of the SCD_trR model and the crystal was lower (Table S2). Although there were still some differences with muSCD1, the total deviation of the residues binding to the SA-CoA head group in SCD_trR was in the range of 2 Å (Table S2). Overall, the ab initio modeling method was more amenable for fatty acid desaturase than the other methods. Among the ab initio methods, the best model was constructed using trRosetta.
Therefore, the MpFADS6 model (Fig. 2) was constructed using trRosetta and the quality was evaluated using QMEAN and PROCKER. As shown in Table S3, 99.0% and 100% residues of the MpFADS6 model were in the most favored and additional allowed regions, respectively (Figure S3). QMEAN scores were accepted as -3.87. It indicated that the MpFADS6 model constructed by trRosetta, as well as SCD1_trR, were closest to the real confirmation and displayed satisfactory quality. Therefore, the MpFADS6 model was able to explore substrate binding sites.
Docking to explore potential binding sites of delta 6 fatty acid desaturase with the substrate
The MpFADS6 model in the present study had satisfactory stereochemical quality and was used for the docking of the substrate. However, the ligand head group of fatty acyl-CoA with higher flexibility caused the difficulty of docking to fatty acid desaturase. After evaluating the five mainstream academic docking procedures, it was found the best pose was obtained by dock6 in which the acyl chain extended well into the hydrophobic cavity in a reasonable range and formed a stable state with the key binding residues of fatty acid desaturase (Fig. 2 and Table 2). And LeDock and rDock showed better performance on ligand heme with few rotatable bonds (Fig. 4A and Table 2).
The acyl tail of ALA-CoA was in a negatively charged cavity surrounded by three histidine conserved regions (Fig. 2B). The head group of CoA was attracted by the positive charge and anchored to the protein surface (Fig. 2C). As previously reported, heme bonded to the HPGG domain at the N-terminus of MpFADS6, in which a potential interaction site revealed that heme formed a Π-Π stack with residues H69 and H94 (Fig. 4A). Fig. 2B shows the potential sites where ALA-CoA interacted in the histidine conserved domains to introduce a double bond at the sixth position (highlighted in light red) from the methyl terminal. Residues H231, H358, K448, and N445 may promote substrate stability. In addition, as a preferred substrate for MpFADS6, ALA is formed by dehydrogenation of LA at the fifteenth carbon atom from the methyl terminal via omega-3 desaturase. This means that the LA and ALA structures only differ by one double bond (highlighted in magenta in Fig. 2B. Residues H228, F352, W224, M227, and F289 might strengthen the desaturation of the interaction between the double bond of ALA and these residues, resulting in substrate preference.
Effect of mutation site on substrate binding ability
To confirm the influence of the amino acid sites analyzed above on desaturation, the 10 residues were mutated to alanine, and the corresponding Saccharomyces cerevisiae recombinants were constructed because S. cerevisiae does not contain ALA and SDA, which was the substrate and product of FADS6, respectively. After induction by galactose for 24 h, ALA was added as a substrate for the enzymatic reaction, and the fatty acids were extracted for methyl esterification and analysis by GC–MS. The desaturase activity was shown in Fig. 3 and Table S4. The corresponding mutant models were built by Chimera and compared the structural difference with original model. And the affinities of the mutants to the substrate were analyzed by docking (Table 2). In addition to these 10 mutations by calculated analyses, two random mutations, S97A and M223A, were simultaneously analyzed for structural differences and enzyme activities.
As shown in Fig. 4A, H94 and H69 in the heme-binding region have the imidazole groups that attracted divalent iron at the heme center to form a stable binding. One of H94 and H69 was mutated to alanine, which suddenly lost the binding force and blocked desaturation (Fig. 3). The docking results in Table 2 show that H69A and H94A have low affinity for Heme, but the docking posts in Fig. 4B and C show that heme cannot form a stable binding with HPGG domain. However, the affinity of the random mutation S97A at the N-terminal reduced and Heme is right in the HPGG domain like Fig. 4A, so the deaturation activity retained 84.44% (Fig. 3). Therefore, H69A and H94A could not convert ALA into SDA, where electron transfer was most likely blocked because of the unstable reaction to heme.
In addition, one mutant among W290A, W224A, and F352A directly bound the substrate, which led to the inactivation of MpFADS6 (Fig. 3). According to the substrate mesh (Fig. 5), one of the mutants rendered the residues in the substrate tunnel incapable of contacting the substrate acyl chain, which failed to recognize the double bond of ALA-CoA and switch on desaturation. In contrast, the side chain group of F289A was distant from the substrate, which made the cavity space larger (Fig. 5E). This weakened the stability of the complex state of the protein and the substrate. The activity and affinity of MpFADS6 decreased, but inactivation was not observed after the mutation of F289, in which the deaturation activity retained 85.72% (Table 2 and Fig. 3). However, the random mutation M223A has 99.95% activity of MpFADS6 (Fig. 3). Molecular simulation analysis indicated that M223 was far from the cavity of the protein substrate that was not in contact with the substrate (Figure S4). The conformation of W224 and W290, which constituted the substrate channel, remained unchanged during the methionine to alanine mutation. Therefore, M223A was not the key binding residue of MpFADS6.
The head group of the ALA-CoA substrate was located on the surface of the protein and had pronounced flexibility. Many residues contacted the MpFADS6 surface, which made the analysis difficult. Therefore, H452, N445, K448, and H358 residues located at the entrance of the substrate cavity were selected for investigation. Alanine mutations were detected on these four amino acids (Fig. 6). The increased activity of K448A was speculated to reflect the enlarged entrance of the cavity after mutation (Fig. 6B), which increased the recognition and attraction of MpFADS6 to the substrate and blocked the electrostatic attraction of lysine to the CoA group. These changes increased the probability that the substrate would enter the cavity in the correct posture. Therefore, the desaturation activity of K448A increased by 1.08 times (Fig. 3). Due to the electrostatic interaction, N445A lost its covalent connection with the adenosine group (Fig. 6C). After mutations of H452 and H358, the positively charged domains were converted to neutral (Fig. 6D and E), which weakened the protein surface and attracted the adenosine group of ALA-CoA. Thereby reducing the recognition and attraction of MpFADS6 to the substrate, resulting in a significant decrease in the activity, and the retained activities are less than 50% of wild type (Fig. 3).