Discussion

In our recent study, a library comprising 115 oximes was tested as reactivators of human BChE inhibited by nerve agents GA, GB, GF, and VX [9]. The determined kinetic parameters were used in the present work, in which PC analysis was applied to integrate them with the physicochemical properties of the oximes and pharmacological parameters relevant to drug design.

PC analysis was applied because of its primary advantage as a non-biased multiple variable method that preserves as much variability as possible while transforming it into new, mutually uncorrelated variables. Application of PC analysis to the reactivation efficacy of selected oximes resulted in three PCs with eigenvector values higher than 1 and a cumulative proportion of variance 72 %. PC1 has a relatively high proportion of variance, 42.4 %. Analysis of PC1 loadings across 27 variables revealed the variables most influential for reactivation efficacy. As expected, the highest positive PC1 loadings were associated with kinetic parameters of BChE reactivation. More interesting, however, was the distribution of loadings for pharmacological parameters (Table 1). Only the molecular fractional polar surface area exhibited positive PC1 loadings, while other parameters showed negative loadings. The ones with values below -0.9 include molecular weight, number of rotational bonds, molecular volume, and molecular surface area. All these parameters can be attributed to the molecule's size.

The correlation between PC1 and PC2 discriminates effective oxime reactivators from moderate or inactive ones. Effective oximes exhibited mainly negative PC1 values and positive PC2 values, while ineffective oximes were characterised by positive PC1 values (>5) and PC2 values (Figure 3). Components PC2 and PC3 jointly account for 29.6 % of the variance, and the PC2 vs. PC3 correlation makes the discrimination of oximes by reactivation efficacy more pronounced than the PC1 vs. PC2 correlation. Ineffective oximes have negative PC3 values, while effective oximes have positive values or close to zero. PC3 loadings for pharmacological parameters showed that the lowest values were for the number of rotational bonds (0.005), the number of H-bond acceptor atoms (0.088) and the molecular surface area (-0.031). Because the loadings essentially represent the Pearson correlation coefficient, providing a direct measure of their linear relationship, the results of PC3 indicate that three parameters (number of rotational bonds, number of H-bond acceptor atoms, and molecular surface area) have minimal effect on the distribution of transformed data, and consequently, oxime distribution by the reactivation efficacy.

For comparison, separate PCAs were performed for each OP compound. These individual analyses yielded similar proportions of variance for PC1, PC2, and PC3, as well as comparable transformed data distributions or loadings (Tables S2 to S5). Similarity is driven by a higher proportion of the same 14 pharmacological parameters than of the four OP kinetic parameters. The transformed data distribution showed that ineffective oximes were positioned far-right in the PC1 vs. PC2 correlation, and effective oximes in the far-left position, except for the GA reactivation dataset. GA reactivation is characterised by a high number of ineffective oximes, which are positioned in both the far-right and far-left. PC2 vs. PC3 correlation again more clearly discriminates between effective and ineffective oximes, with effective oximes exhibiting positive PC2 and PC3 values. For GA reactivation, ineffective oximes exhibited both positive and negative PC2 values, with predominantly negative PC3 values. Analysis of PC2 and PC3 loadings for the pharmacological parameters of an individual OP compound indicated that parameters below the 10 % threshold had a minor impact on the distribution of the transformed data. The corresponding variables were the number of aromatic rings and the number of rings for PC2 loadings of GB and GA data, while molecular weight and molecular volume were corresponding variables for GF and VX data.

For PC3 loadings, a larger number of variables (5 to 7) were below the 10 % threshold compared with PC2 loadings. All four OP compounds shared a low, variable molecular volume, while GB, GF and VX data shared the variables: number of rotational bonds and molecular surface area. Molecular volume was present in both PC2 and PC3 components as a low-value variable. When these findings were compared with the joint PC analysis for all four OP compounds, only one PC2 variable (dipole moment) fell below the 10 % threshold. In contrast, several PC3 variables, number of rotational bonds, molecular surface area, and molecular volume, were below this threshold. Interestingly, all low-threshold variables in individual PC analyses are repeated or preserved in the joint PC analysis, indicating overlap between the joint and individual analyses.

A previous study analysing a range of AChE inhibitors (68 compounds) aimed at correlating structural features of inhibitors with overall AChE inhibition potency, showed that inhibition potency was not highly correlated with topological polar surface area (TPSA) or the number of rotational bonds [25]. Yet, the structural features responsible for potent inhibition were molecular weight and the number of atoms. Structure-activity analysis provided insight into how the TPSA parameter, related to PSA 2D, can be linked to the aromatic nature of the AChE active-site gorge, where polar interactions are not required for high ligand affinity. Generally, potent inhibitors are expected to be larger ligands that induce a minor conformational change in the AChE active site residues by forming multiple hydrophobic interactions with aromatic residues of the AChE active site [25].

Variables that represent important pharmacological parameters significant for evaluating a compound’s BBB permeability, and consequently CNS activity, are Alog _P_98 and PSA 2D. According to the BBB plot, 87 % of the oximes in our library were predicted to penetrate the BBB and thus potentially reactivate ChEs in the CNS, accompanied by good human intestinal absorption [24] (cf. Figure 7). Loadings calculated from joint PC analysis revealed Alog _P_98 negative values with increasing magnitude: PC1-0.189, PC2-0.333 and PC3-0.870. In contrast, PSA 2D loadings changed from negative values to positive with a decline in the magnitude of PC1-0.814, PC2 0.491, and PC3 0.151. Since PC3 discriminated between ineffective and effective oximes in BChE reactivation, Alog _P_98 loading is important for transforming data distribution and plays a key role in oxime classification. The 2D PSA loading from PC1 influences the distribution of the transformed data. However, PC1 does not discriminate between oximes by reactivation efficacy as effectively as PC3 does.