The molecular dynamics simulations of fentanyl complexed with the μ-opioid receptor (μOR) were studied using both inactive 4DKL and active 5C1M opioid receptor crystal structures. Analogous simulations in morphine with or without a ligand were done for comparison. Simulations of the inactive states were carried out in the absence and presence of the Na⁺ ion. The obtained fentanyl’s binding mode agrees with some of the mutagenesis data, and it overlaps with that of morphine only to a minor extent. Notably, fentanyl stabilizes different rotameric states of Trp293^6.48 than observed for morphine or unliganded receptor. Another difference is tighter arrangement of the interaction between Asp147^3.32 and Tyr326^7.43 (a link between helices TM3 and TM7) in the presence of fentanyl. Principal component analysis reveals differences in the trajectories dependent on the ligand bound. The differences found could be linked to ligand-dependent efficacy with respect to receptor intracellular signaling events.

The NICS(1) magnetic aromaticity index is split into NICS(1) and NICS(−1) indices when the points 1 Å above and below the ring center are inequivalent by symmetry. The two indices characterize the aromaticity of the two ring faces rather than the ring itself.

Interactions of 21 fentanyl derivatives with μ-opioid receptor (μOR) were studied using experimental and theoretical methods. Their binding to μOR was assessed with radioligand competitive binding assay. A uniform set of binding affinity data contains values for two novel and one previously uncharacterized derivative. The data confirms trends known so far and thanks to their uniformity, they facilitate further comparisons. In order to provide structural hypotheses explaining the experimental affinities, the complexes of the studied derivatives with μOR were modeled and subject to molecular dynamics simulations. Five common General Features (GFs) of fentanyls’ binding modes stemmed from these simulations. They include: GF1) the ionic interaction between D147 and the ligands’ piperidine NH⁺ moiety; GF2) the N-chain orientation towards the μOR interior; GF3) the other pole of ligands is directed towards the receptor outlet; GF4) the aromatic anilide ring penetrates the subpocket formed by TM3, TM4, ECL1 and ECL2; GF5) the 4-axial substituent (if present) is directed towards W318. Except for the ionic interaction with D147, the majority of fentanyl-μOR contacts is hydrophobic. Interestingly, it was possible to find nonlinear relationships between the binding affinity and the volume of the N-chain and/or anilide’s aromatic ring. This kind of relationships is consistent with the apolar character of interactions involved in ligand–receptor binding. The affinity reaches the optimum for medium size while it decreases for both large and small substituents. Additionally, a linear correlation between the volumes and the average dihedral angles of W293 and W133 was revealed by the molecular dynamics study. This seems particularly important, as the W293 residue is involved in the activation processes. Further, the Y326 (OH) and D147 (Cγ) distance found in the simulations also depends on the ligands’ size. In contrast, neither RMSF measures nor D114/Y336 hydrations show significant structure-based correlations. They also do not differentiate studied fentanyl derivatives. Eventually, none of 14 popular scoring functions yielded a significant correlation between the predicted and observed affinity data (R < 0.30, n = 28).

A common practice in modern QSAR modelling is to derive models by variable selection methods working on large descriptor pools. As pointed out previously, this is intrinsically burdened with the risk of finding random correlations. Therefore it is desirable to perform tests showing the performance of models built on random data. In this contribution, we introduce a simple and freely available software tool SCRAMBLE’N’GAMBLE that is aimed at facilitating data preparation for y-randomization and pseudo-descriptors tests. Then, four close-to-real-world modelling situations are analysed. The tests indicate what the quality of obtained QSAR models is like in comparison to chance models derived from random data. The non-randomness is not the only requirement for a good QSAR model, however, it is a good practice to consider it together with internal statistical parameters and possible physical interpretations of a model.