Read-across is an expert-based in silico approach, whereby experimental data from one or more chemical analogs (termed source compounds) are used to support the prediction of toxicity for a chemical with no data (termed target compound) for a toxicological endpoint of interest.94-97 A one-to-one prediction is termed an analoge approach and a many-to-one prediction is termed a category approach. Qualitative predictions like hazard or quantitative predictions such as NOAELs are possible.
Any read-across approach and workflow is usually handled on a case- by-case basis. As it is time consuming and requires expert knowledge, read- across is most often applied to predict complex toxicological endpoints such as the expected toxicity after repeated exposure of low doses or reproductive toxicity, as well as endpoints for which validated QSAR models or in vitro testing batteries are not able to replace in vivo testing.
In Europe under REACH, read-across was the most frequently used alternative method to predict repeated-dose toxicity.98 It was used in 3220 (33%) out of 9786 dossiers entries to predict repeated-dose toxicity (all routes, all study durations) of phase-in-substances with production volumes of 100-1000 tonnes per year.
The challenge for read-across is to identify “similar” source compounds. In principle, convincing evidence of “similarity” of source and target compounds have to be provided with regard to chemical as well as biological similarity. In cases where metabolism is an important aspect, chemical and biological similarity must also be evaluated for critical metabolites.
Chemical similarity is defined by shared structural features, e.g. functional groups, but also by physico-chemical properties. These features and/or properties should be carefully evaluated for their relevance to the predicted toxicological endpoint. Endpoints related to reactivity may focus on the presence or absence of reactive functional groups contained within the source compounds (or their metabolites). In other cases a consistent trend, e.g. the effect of different aliphatic side chain lengths, may also be appropriate. Physicochemical properties are often used as a first indication of absorption, distribution, and elimination in vivo, and might accordingly be similar or follow a consistent trend. Next, the biological similarity has to be addressed. The question is “do the source compounds have a similar mode of action?” and “how can this be proven?” Based on the analysis of appropriate in vivo data, e.g. from animal or human investigations, a read-across hypothesis has to be defined. For repeated-dose toxicity, biological similarity may include the evaluation of similar critical findings, e.g. neoplastic or non-neoplastic lesions in identical target organs. As an example, organo(thio)phosphates typically induce a decrease in acetylcholinesterase (AChE) which leads to neurological symptoms. Beside shared toxicodynamics, toxicokinetic information is crucial to conclude that the behavior of the source compounds in vivo is similar and to make a prediction such as a noael value with confidence. when looking for analogs with relevant toxicological data, toxicity databases, as discussed in section 9.2 provide a critical resource for this process.
In traditional risk assessment, which is mainly based on in vivo animal studies, often the underlying mechanisms of actions or adverse outcome pathways are not known. However, this mechanistic information might be a key element for further supporting the hypothesis of biological similarity. Recently, a framework has been developed in the sEuRAT-1 project with the aim to predict systemic toxicity after repeated exposure for chemicals and cosmetics through the integration of mechanistic data from in vitro investigations.99 Four different read-across scenarios were investigated, to demonstrate how evidence from in vitro molecular screening, '-omics' assays, and computational models can support a traditional read-across based on structural similarities between source and target substance.100 these four read- across scenarios are as follows.
- (1) Chemically similar compounds that do not require/undergo metabolism to exert a potential adverse human health effect; for example perfluoroalkyl acids101
- (2) Chemically similar compounds involving metabolism (resulting in exposure to the same/similar toxicant); for example p-unsaturated alcohols
- (3) Chemicals with general low or no toxicity; for example primary alcohols
- (4) Chemicals in a structurally similar category with variable toxicities based on a mode of action hypothesis; for example alkylphenols.
Read-across is a prediction containing a certain uncertainty, which needs to be assessed. the selection and accuracy of the analoges, the quality of the experimental data, the relevance of data and accuracy with regard to the toxicological endpoint, as well as data gaps need to be evaluated and documented.102 ECHA has recently drafted a read-Across Assessment Framework for human endpoints to support a transparent and systematic read- across assessment and workflow.103