Thiocolchicine

Genetic toxicity: in vivo

Administrative data

Endpoint:

in vivo mammalian somatic cell study: cytogenicity / erythrocyte micronucleus

Type of information:

(Q)SAR

Adequacy of study:

weight of evidence

Study period:

2019

Reliability:

2 (reliable with restrictions)

Rationale for reliability incl. deficiencies:

results derived from a valid (Q)SAR model and falling into its applicability domain, with adequate and reliable documentation / justification

Justification for type of information:

Selected computational tools for QSAR predictions
Several computational tools are nowadays available for applying in silico (non-testing) QSAR approaches. A battery of computational tools, both commercial and publicly available, implementing scientifically valid QSAR models was selected. These tools were selected according to the OECD Guidance Document on the validation of (Q)SAR models3 as well as ECHA guidelines4,5, that describe generally accepted guidelines to evaluate if an in silico data is suitable for regulatory use.
As mentioned in the previous paragraph, for the prediction of the endpoint(s) of interest, the computational assessment documented in the present report is based on the most reliable prediction among the ones provided by different tools. The predictors selected for each endpoint are following described in details.

Leadscope
Leadscope Model Applier (Leadscope, Inc., version version 2.4, 2019) is a chemoinformatic platform that provides QSAR for the prediction of potential toxicity and adverse human clinical effects of pharmaceuticals, cosmetics, food ingredients and other chemicals. The Models are constructed at the Food and Drug Administration (FDA) by the Division of Drug Safety Research Staff (DDSR) based on both proprietary and non-proprietary data and are intended to support regulatory decision making processes. The models were built under a Research Collaboration Agreement (RCA) using the Leadscope Prediction Data Miner software. Leadscope models are developed with molecular descriptors that include structural features and seven calculated properties, which are molecular weight, LogP, polar surface area, hydrogen bond acceptors, hydrogen bond donors, number of rotational bonds and, Lipinski score (rule violation). The prediction results for each model are provided as the “prediction call” and the “positive prediction probability”. The prediction call can be “positive”, “negative”, and “not-in-domain”. The positive prediction probability is given as the likelihood value between 0 (non-toxic) and 1 (toxic). The higher the probability is, the greater chance there is of the test chemical being toxic in a particular endpoint. Within the FDA, a test chemical is evaluated as active if the probability is ≥ 0.5 and inactive if the probability is < 0.5. Leadscope uses two parameters to guide the applicability of model domain: 1) Model Features Count: parameter used to verify that the target compound contains a significant number of features that are present in the prediction model; 2) 30% Similarity Training Neighbours Count: number of training compounds structurally similar to the target (with at least 30 % similarity). If at least one model feature and one training neighbour (> 30% similarity) are found, then the target is included in the applicability domain of the model. The robustness of the prediction can be further evaluated by examining compounds similar to the target from the training set. Training set structural analogues are provided along with experimental and predicted data.
A brief description of the Leadscope predictors employed in the present study is provided in Table 3.1.

Data source

Reference

Reference Type:

study report

Title:

Unnamed

Year:

2019

Report date:

2019

Materials and methods

Test guideline

Guideline:

other: REACH Guidance on QSARs R.6

Deviations:

no

Principles of method if other than guideline:

QSAR methodology
(Quantitative) Structure-Activity Relationships, collectively referred to as (Q)SARs, are theoretical models based on a (quantitative/mathematical) relationship between a numerical representation of chemical structure and a biological effect/activity or property. (Q)SAR models can be used to predict in a qualitative or quantitative manner the physico-chemical, biological (e.g. toxicological) or environmental fate properties of compounds from knowledge of their chemical structure.
The presented study documents (Q)SAR predictions to fulfil the information requirements under REACH Regulation in compliance with conditions set out in REACH Annex XI. According to REACH Annex XI, the results of (Q)SARs may be used instead of testing when: i) the (Q)SAR models have been scientifically validated, ii) the substance falls within the applicability domain, iii) the results are adequate for the purpose of classification and labelling, and iv) adequate and reliable documentation of the applied method is provided. The scientific validity of the model is evaluated according to the internationally recognised OECD principles for the validation, for regulatory purposes, of QSARs (1: Organisation for Economic Co-operation and Development. Report on the Regulatory Uses and Applications in OECD Member Countries of (Quantitative) Structure-Activity Relationship [(Q)SAR] Models in the Assessment of New and Existing Chemicals. ENV/JM/MONO(2006)25, OECD, Paris, France 2007. Available from: http://www.oecd.org/.) i) a defined endpoint; ii) an unambiguous algorithm; iii) a defined domain of applicability; iv) appropriate measures of goodness-of-fit, robustness and predictivity; and v) a mechanistic interpretation, if possible.
The predictions generated by the employed QSAR models are evaluated in terms of their reliability, as required by OECD principles. The assessment of the reliability of a QSAR prediction is critical and is therefore a main requirement for the prediction to be accepted by regulatory authorities and initiatives. Therefore, each prediction is provided together with the information on the applicability domain of the model used to derive it. As a general rule, predictions are assigned to four levels of reliability:
• high reliable: target compound is included in the applicability domain of the model and the prediction is assessed as highly accurate (very limited uncertainty).
• moderate reliable: target compound is included in the applicability domain of the model and prediction is assessed as moderately accurate (rather limited uncertainty).
• borderline reliable: target compound is included in the applicability domain of the model, but the level of confidence of the prediction is limited (medium uncertainty).
• not reliable: target compound is outside the applicability domain of the model and/or high uncertainty is associated to the prediction.
Several criteria are considered to define the level of reliability (2: European Chemicals Agency - ECHA (2016) Practical Guide – How to use and report (Q)SARs. ECHA-16-B-09-EN. Available at: https://echa.europa.eu/documents/10162/13655/pg_report_qsars_en.pdf) including the applicability domain of the model (e.g. descriptor domain, structural fragment domain,…) as well as information on training set analogues (e.g., degree of similarity toward the target, consistency of experimental data, prediction accuracy). It is important to highlight that different predictors provide different parameters and information to support the reliability assessment of the generated predictions Different tools and/or approaches are used, when possible, and assessed for their reliability. The final computational assessment is based on the most reliable prediction among the ones provided by different tools.
Predictions are finally assessed for their adequacy for regulatory purposes, and their reliability in terms of Klimisch score is provided, together with the rationale for assigning it. The rationale for reliability of QSAR predictions is provided in line with the current IUCLID template:
• results derived from a valid QSAR model and falling into its applicability domain, with adequate and reliable documentation/justification (Klimisch 2).
• results derived from a valid QSAR model and falling into its applicability domain, with limited documentation/justification (Klimisch. 2, 3, or 4).
• results derived from a valid QSAR model, but not (completely) falling into its applicability domain, with adequate and reliable documentation/justification (Klimisch 2 or 3).
• results derived from a valid QSAR model, with limited documentation/justification, but validity of model and reliability of prediction considered adequate based on a generally acknowledged source (Klimisch 2, or 3)
• results derived from a valid QSAR model, but not (completely) falling into its applicability domain, and documentation/justification is limited (Klimisch 3 or 4).
• results derived from a valid QSAR model, with limited documentation/justification (Klimisch 4).

Results and discussion

Additional information on results:

Genotoxicity: Micronucleus in vivo rodent
Computational tool: Leadscope Model Applier
Genotoxicity as micronucleus in vivo on rodent predictions were generated employing Leadscope Model Applier. Leadscope model for in vivo (mouse) micronucleus test (Genotox Suite/Clastogenicity in vivo/Micronucleous mouse) estimates the probability that a compound will result positive in the experimental assay. Leadscope results include a genotoxicity prediction (positive, negative or not in domain), a positive prediction probability and two parameters which assess model applicability domain, i.e. Model Features Count and 30% Similarity Training Neighbours Count.

Any other information on results incl. tables

Tool	Prediction call	Positive Predictions probability	AD parameters: Model Features Count	AD parameters: 30% Sim. Training Neighbors Count	Realiability assessment
Leadscope	POSITIVE	0.76	16	3	HIGHLY RELIABLE

Applicant's summary and conclusion

Conclusions:

This study was designed to generate in silico (non-testing) data by means of QSAR (Quantitative Structure-Activity Relationship) methodology for Thiocolchicine to be used for its hazard assessment. Genotoxicity (Micronucleus in vivo rodent) result: Positive (from prediction call). The reliability assessment has: Highly Realiable