Toxicity assessment is a key part of the drug discovery and development process. Many of the tests that use laboratory animals to investigate drug safety are time-consuming, expensive and complicated by ethical concerns. Investigative toxicology strategies have therefore adopted a tiered approach employing in silico and in vitro methods in order to ideally eliminate, or at least reduce, in vivo experimentation.
In silico testing enlists the computer to model the safety of compounds based on their physicochemical properties and predicted chemical reactivity. Such assessments are used prior to, or alongside testing in vitro or in animals, and can lead to the future termination of development of a drug candidate.
Here, we attempt to give a simplified overview of the complex and multifaceted topic that is in silico toxicology assessment. Strap in and hold on tight…
Why should I consider in silico toxicity assessments?
As we imply above, in silico methods can offer a fast and cost-efficient alternative to other strategies. They can even be performed before a compound is synthesised, leading to a more streamlined development process. Differences between the physiology of humans and other species suggest that the use of predictive human tools (such as in silico assessments) might one day be more effective than laboratory animal models for drug safety testing [1].
Despite their advantages, in silico approaches to toxicity assessment can be, no let’s be honest, are, often complex and usually require expert knowledge and review. Fear not, team bibra has a lot of experience with in silico toxicity models – we’ve got your back! Where possible, we strive to help clients fill data gaps without a potentially costly trip to the testing laboratory.
First, let’s consider the different types of in silico assessments…
Types of in silico toxicity assessments
In silico toxicity models are based on experimental data, structure-activity relationships and scientific knowledge, such as structural alerts reported in the literature.
Expert rule-based (or expert/structural alerts)
These methods use the presence or absence of specific and established molecular substructures (structural alerts) to predict the toxicity of compounds. They are based on the concept that if a particular structural alert is present in a higher percentage of toxic than non-toxic molecules, the structure might be responsible for the toxicity. The presence of potential deactivating structural features (i.e., those that might counteract the effects of other structures) in the test compound can be a confounding factor in structural alert-based analyses. Since such analyses often do not consider the whole molecule, they should then only be used as a heath precautionary start to an analysis.
Statistical-based
Also known as (quantitative) structure-activity relationship, or (Q)SAR, modelling, this method involves the use of data from a training set of example chemicals that were found to be positive or negative in a given toxicological study (e.g., the Ames mutation assay), or to induce a continuous toxicological response. The expected toxicity of a test compound can then be predicted based on the similarity of its physicochemical properties and chemical structure to those of the compounds in the training set. It is entirely machine driven, with no a priori human assumptions of what constitutes a nasty structural feature.
Supporting in silico analyses
There is an abundance of online databases to support the findings of structural alert and (Q)SAR analyses. A systematic review of the key resources [2] identified 57 toxicological databases, as well as 80 databases describing associations between bioactivities and physicochemical properties. That paper was published almost 5 years ago, and no doubt there are many more now! There is clearly a need to integrate such databases with in silico tools for predicting the properties or activities of compounds.
Read-across
Read-across approaches can also be used to support in silico assessments. These use existing toxicity data for one or more compounds to predict the toxic effects of compounds with similar structures. In this way, the experimental data for the comparators can be used to “read across” to the specific target compound. The identification of suitable read-across candidates requires expert knowledge and a justification of why one compound is similar to another (for example, in terms of expected reactivity, toxicokinetics, mechanism/mode of action, structure, physicochemical properties, and/or metabolic profile). Read-across is ideally used when there are sufficient experimental data from high-quality databases on a number of similar substances. It can be invoked in less data rich contexts when the spirit of the 3Rs is the guiding light, rather than strict adherence to regulatory texts, which often set very high bars to justify the use of read-across.
Which in silico method(s) should I use to assess the potential toxicity of my compound?
This can be a minefield! In some cases, in silico models can produce seemingly contradictory results, and it is always important to understand the reliability and limitations of each model and to apply the results appropriately. Proper curation of the structure of your compound is also essential because errors in the electronic representation entered into the in silico model will result in invalid predictions. Nothing can be done unfortunately for the often “black box” nature of some of the packages, or the probable dubious calls on the experimental data that they might be based upon. As such, an expert review of the selection of in silico assessments and their outputs is crucial.
Are in silico toxicity assessments accepted by regulatory agencies?
In silico methods are beginning to gain acceptance for regulatory data submission. Indeed, several guidance documents have been drafted to improve the standardisation, harmonisation and uptake of such methods by regulatory authorities – bibra has been proud to be part of several of these projects [3]. A notable example is the International Council for Harmonization (ICH) M7 guideline for the assessment and control of DNA reactive (mutagenic) impurities in pharmaceuticals to limit potential carcinogenic risk [4].
Several other groups have developed specific documents supporting the use of in silico tools, which bibra make use of in our work. Examples include reports from the Organisation for Economic Co-operation and Development (OECD) outlining principles for the validation of (Q)SARs and guidance on chemical grouping; reports from the European Chemicals Agency (ECHA) on (Q)SARs, chemical grouping and read-across; a report from the European Centre for Ecotoxicology and Toxicology of Chemicals (ECETOC) on read-across and (Q)SAR approaches; as well as several complementary peer-reviewed publications outlining processes for the implementation of such computational assessments.
Despite growing interest in the use of in silico toxicity assessments, some issues still hinder their general acceptance. Most notably, there is a lack of standardised procedures for the application of in silico tools and the interpretation of their outputs, and the combined use of experimental and in silico to support risk assessment varies across organisations [4]. Consequently, seeking expert advice from your favourite toxicologist is essential when considering in silico assessments, particularly for regulatory purposes.
How can bibra help with in silico toxicity assessments?
Bibra has substantial experience in selecting, applying and clarifying the results of in silico toxicity models, which we interpret in the context of existing literature identified in searches of databases such as PubMed and our own in-house database TRACE. Our toxicologists perform robust sanity checks on all in silico outputs to ensure that the predictions are sensible and consistent with the knowledge and expectations built up over many years.
Our team routinely evaluate(Q)SAR predictions using a range of in silico models, including Toxtree, QSAR Toolbox, Leadscope, Derek Nexus, Meteor Nexus and Sarah Nexus. Bibra is an expert consulting partner of Leadscope, which we use frequently to evaluate DNA-reactive compounds (particularly pharmaceutical impurities) in accordance with ICH guidance M7.
The toxicological assessment of impurities in drug products poses a particular challenge as, in most cases, no substance-specific data exist. At bibra, we have developed a workflow to combine our extensive expertise with in silico models and threshold of toxicological concern (TTC) concepts to establish the safety of pharmaceutical impurities.
We are also experienced in read-across and have routinely used this approach in our hazard and risk assessments for many years, including for regulatory submissions (following ECHA guidance).
References
[1] University of Oxford. In Silico Human Drug Safety and Efficacy. Accessed August 2023. https://www.cs.ox.ac.uk/insilicocardiotox/about/
[2] Pawar G, Madden JC, Ebbrell D, Firman JW, Cronin MTD (2019). In silico Toxicology Data Resources to Support Read-Across and (Q)SAR. Frontiers in Pharmacology 10, 561. doi: 10.3389/fphar.2019.00561. https://www.ncbi.nlm.nih.gov/pmc/articles/PMC6580867/
[3] Myatt GJ et al. (2018). In silico toxicology protocols. Regulatory Toxicology and Pharmacology 96, 1-17. doi: 10.1016/j.yrtph.2018.04.014. https://www.ncbi.nlm.nih.gov/pmc/articles/pmid/29678766/
[4] ICH (2023). ICH guideline M7(R2) on assessment and control of DNA reactive (mutagenic) impurities in pharmaceuticals to limit potential carcinogenic risk. Final version. Adopted on 3 April 2023. https://database.ich.org/sites/default/files/ICH_M7%28R2%29_Guideline_Step4_2023_0216_0.pdf