Figure 4. Plot of the relation between p-phenols toxicity, log(1/IC 50), and interaction energies,E int, of data sets A (black squares), B (red circles) and C (blue triangles). Dashed lines correspond to linear parameters in Table 2.
Finally, the cytotoxicity correlation with the interaction energy data, Eint, (Table S1 in Supplementary information) as a measure of the phenol reactivity[11] according to Eq. (2) exhibit the best statistical parameters (Table 2, Fig. 4) for the dataset A.
There are significant differences between the regression parameters of the Eq. (2) for both A and B datasets. Except for Eint, the B dataset ones are nearly doubled. In agreement with worse statistical parameters for the B dataset, a similar relation holds for their standard deviations as well. With few exceptions, the B dataset cytotoxicity values are higher than the A ones for similar predictor values whereas the C ones are even lower. The rightness of inclusion of the iodine substituent into the A dataset is confirmed by comparing the corresponding data evaluated by the Eq. (2) by using its regression parameters both for A and B datasets (Table S3 of Supplementary information).
We performed regression analysis for the predictor data of the Brown-Hammett parameters (σ +), the HOMO–LUMO energy difference (L–H gap), the logarithmic n-octanol - water partition coefficients (log P ) and homolytic bond dissociation energies (BDE ) used by Selassie et al. [2] as well. The obtained results for the regression parameters of Eq. (2) and related statistical data are presented in Table 2 and Figs. S4 – S7 of Supplementary information. In all cases these variables offer much worse statistical parameters than our predictors related to a Cu(II) probe. The distribution of A and B dataset data is more chaotic despite the b parameter of the Eq. (2) for both datasets is nearly identical.
Finally it can be concluded that the QSAR model based on quantum-chemical calculations of the Cu(II) coordination ability of p-substituted phenols offers much better results than classical treatments.[2] The predictors based on the O → Cu electron density transfer, such as copper charges and BCP electron density Laplacian of the Cu-O bond in the2[Ph Cu]2+complexes, as well as the corresponding interaction energy are able of a very good toxicity prediction. The p-substituted phenols with electron withdrawing and electron donating substituents must be treated separately due to different linear dependencies of both groups. Nevertheless, there may be more than one mechanism for various types of toxicity as indicated by the phenols of the C dataset. Further experimental as well as theoretical studies in this field are desirable.