QSAR for toxicities of polychlorodibenzofurans, polychlorodibenzo-1,4-dioxins, and polychlorobiphenyls

The toxic equivalency factors (TEF values) for 12 polychlorobiphenyls (PCBs), 14 polychlorodibenzo-1,4-dioxins (PCDDs), and 24 polychlorodibenzofurans (PCDFs) have been correlated with the extent and position of chloro-substitution using a variety of molecular descriptors., with Todeschini’s program Dragon 4.0. and the CODESSA program of Katritzky and coworkers.


Introduction
Among polychlorinated aromatic compounds, the most environmentally dangerous chemicals are polychlorodibenzofuran (PCDF) 1, polychlorodibenzo-1,4-dioxin (PCDD) 2, and.polychlorobiphenyl (PCB) derivatives 3. Like the persistent chlorofluorocarbons that were extensively used until the discovery of their deleterious effect on the ozone layer, these polychloroaromatic compounds are harmful to the environment.Similarly to other polychlorinated compounds (used as efficient pesticides such as DDT, Aldrin and Dieldrin, which led to the "Silent Spring" due to bioaccumulation in fatty tissues of higher organisms), the title compounds exert a powerful toxic effect on humans (endocrine disruptors, neurotoxic, carcinogens), 1,2 and one of the polychlorodibenzo-1,4-dioxin isomers is among the most toxic organic chemicals.2 ' 3 ' 4 ' 5 ' 6'

Number of possible isomers
The number of all possible isomers of the above three classes of polychlorinated aromatic compounds can be found easily by means of Polya's Theorem.First, one has to take into account the cycle index of the molecule, depending on the automorphisms (symmetry operations involving proper axes of rotation).For PCBs, one has to consider all possible rotamers of the substituted biphenyl molecule.The cycle index leads to the figure counting series, which is a polynomial in x whose degree equals the number of hydrogen or halogen atoms in the molecule.Each subscript a for x a indicates the order of the permutation, and each superscript b for x a b the number of permutations or the number of unchanged positions.Then, according to Polya's Theorem, by performing the substitution: x a b = (y a + 1) b one ends up with the generating function, which is a polynomial in y whose coefficients indicate the numbers of isomers.For PCBs, the possible permutations are indicated in Table 1.Therefore, there are 3 possible monochlorobiphenyls, 12 dichlorobiphenyls, 24 tri-, 42 tetra-, 46 penta-, 42 tetrachlorobiphenyls, etc.Many of the 209 possible polychlorobiphenyls have been synthesized and characterized chemically, physically, and toxicologically.Of these, fourteen have a pronounced toxic effect, similar to that of tetrachlorodibenzodioxins. 3,4 For PCDDs, the possible permutations are shown in Table 2. )/4.Generating function: y 8 + 2y 7 + 10y 6 + 14y 5 + 22y 4 + 14y 3 + 10y 2 + 2y + 1.Therefore, there are 2 possible monochlorodibenzodioxins, 10 dichloro-, 14 tri-, 22 tetra-, 14 pentachlorodibenzodioxins, etc.Many of the 75 possible polychlorodibenzodioxins have been synthesized and characterized chemically, physically, and toxicologically.5][6][7] This is the infamous and extremely toxic 2,3,7,8-tetrachlorodibenzo-1,4-dioxin or TCDD (LD 50 = 45 µg/kg in rats), formed as a low-yield byproduct during the syntheses of 2,4,5-trichlorophenoxyacetic acid (herbicide and defoliating agent) and hexachlorophene (germicide) from chloroacetic acid and sodium 2,4,5-trichlorophenoxide.In the Seveso accident in Italy, a large amount of soil contaminated with 2,3,7,8-tetrachlorodibenzo-1,4-dioxin had to be removed at high cost.During the Vietnam War, the defoliant (Agent Orange) used by the US Army contained small amounts of 2,3,7,8-tetrachlorodibenzo-1,4-dioxin, and US veterans were compensated financially for the resulting health problems.
Polychlorobiphenyls may have toxicities that are dioxin-like or non-dioxin-like.][36][37][38][39][40] When several polychlorinated aromatic compounds appear in mixtures 41 (as they occur in the environment), and when as usual they have the same mechanism of action, their cumulated toxic effect can be found by a relationship involving partial toxicities, reminiscent of Dalton's Law of partial pressures: the total equivalent toxicity is the sum of the products of their concentration with their TEF value. 31,42,43n the present paper we will use the same TEF data that have recently been used by Beger and Wilkes 43 as well as by Căprioară and Diudea. 44

Molecular Descriptors
The molecular descriptors that were tested are those included in Todeschini's Dragon 4.0 program. 46They may be grouped according to their dimensionality from zero to three.Only representative descriptors that have nonzero values have been selected.
After computing all descriptors, they were converted into common logarithms and correlations with TEF data were explored using the Origin 6.0 program, after eliminating zerovalue descriptors.In agreement with previous papers by Basak, 64,79 there are four levels of significance for molecular descriptors, conveying topostructural, topochemical, geometric, and quantum-chemical information.
1][82][83] Again, hundreds of molecular descriptors (structural, topological, quantum-chemical) are used and selected for the best correlation with a specified number of descriptors.

QSAR results
The best results with the lowest numbers of molecular descriptors are presented separately for the three classes of polychloro aromatic compounds, in each case with three QSAR equations: one (A), starting with the Dragon program, and two other ones (B and C) continuing with the CODESSA program.4 and Figure 1) Two PCDFs from the original set were left out.C. CODESSA program with three descriptors (Table 9 and Figure 6)

Discussion of Results
In Tables 4-12 one can see for each of the polychloroaromatic compound the parameters selected for the QSAR, the experimental and calculated logTEF values, and the difference (residual) between these two values.Diagrams presented in Figures 1-9 indicate the linear dependence between experimental and calculated TEF values, and the equation for the linear correlation that may be used for prediction within each class with the corresponding molecular descriptors.
The correlation coefficients r are satisfactory in all cases.Only two molecular descriptors appear twice: DP05 (a Randić-type molecular profile, which appears in two equations, namely eq.1A and 3A with similar negative coefficients) and MNRIO (the nucleophilic reactivity of the oxygen heteroatom, which appears in eq.2B and 2C).Attempts to find a unique set of up to four molecular descriptors for the combined set of all 50 or 52 polychloroaromatic compounds did not yield any correlation with r >0.90.
On comparing the r values for the QSAR obtained with the Dragon and CODESSA programs, one may see that for a comparable number of descriptors the latter program yields higher r and lower s values.Of course, these two programs are based on different molecular descriptors, so that for predictions of unknown representatives from each class, one should compare the results for each of the three equations A, B, or C in the corresponding class.
Out of 209 possible PCBs, the present correlation covers only 12 (about 6%); out of the 75 possible PCDDs, equations 2A-2C include only 14 structures (about 19%); and out of the 135 possible PCDFs, we have toxicity data only for 26 compounds (also about 19%).
One can argue that the range of toxicities (3.5 for PCBs or 4 orders of magnitude for PCDDs and PCDFs) may not be large enough for all structural effects to become manifest.This is probably true for the PCBs, where out of 10 possible substitution sites only 3-6 correspond to experimentally known toxicity data.In the case of PCDD there are toxicity data for compounds from mono-to all octa-substitution sites, whereas for PCDFs out of the 8 substitution sites there are toxicity data from mono-to hexa-substituted compounds.However, many possible substitution patterns are absent from the experimental data.Therefore, till more experimental data will become available, it is still difficult to argue whether the present correlations may serve for predicting accurately toxicity values for the remaining polychloroaromatics from the three classes of compounds examined here.What can be undeniably inferred from examining the toxicity data is that no single factor or parameter is responsible for the global toxicity of these compounds.Rather, a subtle combination of electronic, steric, geometric (i.e. dihedral angle for PCBs), and hydrophobic effects are at play in determining interactions with cellular receptors that become responsible for the high toxicity of some of the polychloroaromatics.
We believe, however, that the present QSAR study can provide an informed guess about the toxicity of the many still unknown compounds from the three classes of polychloroaromatics examined here.

Conclusions
Di-, tri-, or tetra-parametric linear correlations between toxicity values (logTEF) of three classes of polychloroaromatics, namely polychlorobiphenyls (PCBs), polychlorodibenzo-1,4-dioxins (PCDDs), and polychlorodibenzofurans (PCDFs) have been presented.This QSAR study has used the Dragon program of Todeschini and the CODESSA program of Katritzky and their coworkers.The latter program was slightly superior.
Only about 6% of all possible PCBs, and 19% of all possible PCDDs and PCDFs were available with such toxicity data, but since 3.5-4 orders of magnitude of toxicities were involved, one may confidently assume that most major structural factors could manifest themselves in these data.However, many substitution patterns are not present in the experimentally available toxicity data, so that some of these patterns may escape detection.

Table 1 .
Permutations of the substituents in polychlorobiphenyls

Table 3 .
Permutations of the substituents in polychlorodibenzofurans

Table 4 .
Descriptors and TEF data for polychlorodibenzofurans (Dragon)