Incorporation of the concept of tipping points into adverse outcome pathways (AOPs)

SUMMARY The concept of the adverse outcome pathway (AOP) has evolved since inception. Recent advances include a variety of methods for developing quantitative AOPs (qAOPs). The goal is prediction of adverse outcomes from a range of stressors that activate molecular initiating events (MIEs) that in turn trigger a sequence of key events (KEs) leading ultimately to the adverse outcome (AO). Inherent in qAOPs is the implicit assumption of probabilities of transition along this sequence of key event relationships (KERs) based on the degree of activation of the MIE and each subsequent KE. These probabilities are the basis of the KER tipping points and are a function of not only the dysregulation induced by the stressor but also adaptation or resilience of the organism to counter this dysregulation. Two groups of qAOPs are considered here. The first group relies on data and systems biology modeling, an effort dependent on data at all levels of organization including molecules, cells, tissue, organs, organisms, and populations. The second group use methods that are much less data-dependent and are based on certain types of Bayesian networks (BNs). The difficulty with reconciling the biology inherent in the AOP concept with a purely statistical technique such as BNs is the presence of feedback control loops in biological systems that are inconsistent with the one-way information flow permitted in BNs. However, recent developments in BNs may permit such feedback loops. Here we envision a prediction model for qAOPs that depends on newer types of BNs along with a time-dynamic model to estimate the probabilities of exceeding tipping points. Such a model has not yet been developed, but several qAOPs developed with BNs seem ripe for testing this model.