User Guide to Thresholds and Classification - Environmental ...

342User Guide for Thresholds and Classificationsextent that toxicity is related to production of metabolites for a particular endpoint, as opposed to theproperties of the parent compound. Also, if toxicity is mediated by a specific receptor binding mechanism,dramatic effects may be observed with small changes in chemical structure.What ultimately governs the validity of such predictions is the degree to which the compounds used to derivethe QSAR for a specific biological endpoint, are acting by a common molecular mechanism. In many andperhaps most cases, a QSAR does not represent such a mechanistic model, but merely a correlative one. Atruly valid mechanistic model must be derived from a series of chemicals all acting by a common molecularmechanism, and fit to an equation using one or more parameters that relate directly to one or more steps ofthe mechanism in question. Such parameters or properties are more generally known as moleculardescriptors. It is also important to keep in mind that many such molecular descriptors in common use maynot have a direct physical interpretation. For a correlative model, the statistical fit of the data are likely to bepoorer than a mechanistic one given these limitations. Mechanisms are not necessarily completelyunderstood, but enough information may be known to provide confidence in this approach. For correlativemodels, the predictive reliability increases with the narrowness with which each is defined, for example,categories of electrophiles, such as acrylates, in which the degree of reactivity may be similar and toxicitycan be estimated for a ‗new‘ chemical using a model based solely on the log KOW parameter.As an example, primary and secondary alcohols containing a double or triple bond that is conjugated withthe hydroxyl function (that is, allylic or propargylic) are more toxic than would be predicted for a QSAR for thecorresponding saturated compounds. This behaviour has been ascribed to a proelectrophile mechanisminvolving metabolic activation by the ubiquitous enzyme alcohol dehydrogenase to the corresponding α,βunsaturatedaldehydes and ketones that can act as electrophiles via a Michael-type acceptor mechanism(Veith et al, 1989). In the presence of an alcohol dehydrogenase inhibitor, these compounds behave likeother alcohols and do not show excess toxicity, consistent with the mechanistic hypothesis.The situation quickly becomes more complex once one goes beyond such a homologous series ofcompounds. Consider, for example, simple benzene derivatives. A series of chlorobenzenes may be viewedas similar to a homologous series. Not much difference is likely in the toxicities of the three isomericdichlorobenzenes, so that a QSAR for chlorobenzenes based upon test data for one of these isomers islikely to be adequate. What about the substitution of other functional groups on benzene ring? Unlike analiphatic alcohol, the addition of a hydroxyl functionality to a benzene ring produces a phenol that is nolonger neutral, but an ionisable acidic compound, due to the resonance stabilisation of the resulting negativecharge. For this reason, phenol does not act as a true narcotic agent. With the addition of electronwithdrawing substituents to phenol (for example, chlorine atoms), there is a shift to these compounds actingas uncouplers of oxidative phosphorylation (for example, the herbicide dinoseb). Substitution of an aldehydegroup leads to increased toxicity via an electrophile mechanism for such compounds react with aminogroups, such as the lysine ε-amino group to produce a Schiff Base adduct. Similarly, a benzylic chloride actsas an electrophile to form covalent abducts with sulfhydryl groups. In tackling a prediction for an untestedcompound, the chemical reactivity of these and many other functional groups and their interaction with oneJanuary 2012 EPA0109