U SING READ - ACROSS TO PREDICT THE SENSITIZING POTENTIAL OF CHEMICALS An exercise by the FRANCOPA research working group (+ a glimpse into the CALEIDOS exercise on read-across) E NRICO M OMBELLI , FRANCOPA WG, CALEIDOS CONSORTIUM 1
Skin sensitization : the AOP PROTEIN BINDING LLNA http://www.oecd.org/ 2
The Initiating Event LYSINE ARGININE CYSTEINE HISTIDINE METHIONINE https://en.wikipedia.org/ http://www.oecd.org/ 3
Local Lymph Node Assay (LLNA) EC3 LLNA Category (% w/v) > 100 Non sensitizers ≥ 10 Weak 1-10 Moderate 0.1-1 Strong < 0.1 Extreme 4 https://en.wikipedia.org/
SET #1 (« Training set ») SET #2 (« Test set ») CAS RN SP CAS RN SP CAS RN SP CAS RN PS CAS RN SP CAS RN SP 151-21-3 * 112-05-0 * Negative 93-99-2 Weak 5392-40-5 Moderate 85-44-9 Strong Negative 106-50-3 Strong 71-36-3 Negative 39236-46-9 Weak 80-54-6 Moderate 121-79-9 Strong 99-96-7 Negative 123-31-9 Strong 108-90-7 Negative 97-90-5 Weak 818-61-1 Moderate 35691-65-7 Strong 57-55-6 Negative 5307-14-2 Strong 56-81-5 Negative 107-75-5 Weak 90-15-3 Moderate 100-39-0 Strong 121-32-4 Negative 27072-45-3 Strong 67-63-0 Negative 62-53-3 Weak 111-80-8 Moderate 1166-52-5 Strong 63-74-1 Negative 94-36-0 Extreme 50-21-5 Negative 106-24-1 Weak 591-27-5 Moderate 95-55-6 Strong 94-02-0 Negative 2682-20-4 Extreme 119-36-8 Negative 140-67-0 Weak 101-39-3 Moderate 108-31-6 Strong 91-64-5 Negative 94-13-3 Negative 103-95-7 Weak 108-46-3 Moderate 127-65-1 Strong 81-07-2 Negative 69-72-7 Negative 94-09-7 Weak 111-40-0 Moderate 55406-53-6 Strong 75-35-4 Negative 84-66-2 Negative 2426-08-6 Weak 552-30-7 Moderate 111-12-6 Strong 104-54-1 Weak 92-48-8 Negative 120-51-4 Weak 93-53-8 Moderate 97-00-7 Extreme 110-86-1 Weak 124-07-2 Negative 431-03-8 Weak 17369-59-4 Moderate 15646-46-5 Extreme 514-10-3 Weak 121-33-5 Negative 61-33-6 Weak 119-84-6 Moderate 106-51-4 Extreme 78-70-6 Weak 100-52-7 Negative 103-41-3 Weak 100-69-6 Moderate 886-38-4 Extreme 144-62-7 Weak 65-85-0 Negative 19317-11-4 Weak 5231-87-8 Moderate 100-11-8 Extreme 13706-86-0 Weak 109-65-9 Negative 140-88-5 Weak 6728-26-3 Moderate 1154-59-2 Extreme 107-22-2 Moderate 923-26-2 Negative 111-25-1 Weak 93-51-6 Moderate 93-91-4 Extreme 122-78-1 Moderate 9005-65-6 Negative 1118-71-4 Weak 106-47-8 Moderate 97-53-0 Moderate 100-06-1 Negative 107-15-3 Moderate 96-27-5 Moderate 109-55-7 Moderate 8001-54-5 Negative 101-86-0 Moderate 764-85-2 Moderate 137-26-8 Moderate 110-27-0 Negative 149-30-4 Moderate 5910-85-0 Moderate 122-40-7 Moderate 68-12-2 Negative 97-54-1 Moderate 112-67-4 Moderate 118-58-1 Moderate 3810-74-0 Negative 2634-33-5 Moderate 50-00-0 Strong 2111-75-3 Moderate 874-23-7 Negative 122-57-6 Moderate 104-55-2 Strong 2785-87-7 Moderate 99-76-3 Negative 141-05-9 Moderate 111-30-8 Strong 104-27-8 Moderate 5
• Protein Binding Profiler by OECD (PB-OECD) • Protein Binding Profiler by OASIS (PB-OASIS) The scope of the profilers is to identify the presence of structural alerts for the formation of covalent bonds with proteins within the structure of target molecules 6
(1) The algorithm Profile target chemical according to PB-OECD and PB-OASIS (2) Fetch structural analogs characterized by the same PB profile (OECD+OASIS) (3) Subcategorize according to structural similarity (atomic pathways) : Tanimoto index ≥ 60% (4) Identify closest analogs (2-5) as a function of Log P (5) Predict unknown sensitizing potential according to a majority rule 7
Interpolation vs. Extrapolation 1-Octanol LUMO Energy Water HOMO BIOAVAILABILITY REACTIVITY 8
Predictions with interpolation constraints “[…] Some residual level of uncertainty must be considered acceptable […] even study data have a degree of associated uncertainty” (Ball et al., 2016) 9
Predictions with interpolation constraints 1) The number of possible predictions is particularly reduced if interpolation constraints as a function of Log P and orbital energies are imposed 2) Grouping chemicals together on the only basis of the absence of alerts results in a set of very dissimilar structural analogs 10
Predictions w/o interpolation constraints 1) There is a general tendency towards a lower precision especially in the case of a categorization based only on PB profiling 2) Chemicals with the same alert can have different potencies/Similar chemicals can display rather different potencies 11
“One -to- one” read -across One-to-one read-across can be reliable. A selection based on an identical PB profile and structural similarity seems to serve as a gatekeeper against the selection of inappropriate analogs 12
FRANCOPA : position letter (Following the workshop on « read-across » 15/11/2013, Paris) • Substantiating read-across predictions is more time-consuming than usually expected. The role of the expert is crucial The CALEIDOS Exercise • Read-across approaches have to be especially supported by a mechanistic rationale that justifies the adopted criteria for similarity • They also have to be supported by information on toxicokinetics • Synergy between read-across and QSAR should not be overlooked RAAF, GRAP, CAAT-CEFIC-EU-ToxRisk • FRANCOPA both supports the emergence of guidelines and clear rules of acceptance (ECHA) and initiatives to enhance the read- across approaches through a synergy with other methods. 13
C hemical A ssessment according to L egislation E nhancing the I n silico Do cumentation and S afe use (coordinator : Emilio Benfenati, IRFMN) • A round-robin exercise was conducted within the CALEIDOS LIFE project (2013-2015). The participants were invited to provide read- across predictions for three endpoints: mutagenicity (Ames test), bioconcentration factor (BCF) and fish acute toxicity. • Nine chemicals were attributed to each endpoint • FOCUS ON : reproducibility and adopted strategies • 181 questionnaires from 40 participants: - Mutagenicity (93), BCF (47), Fish acute toxicity (41) 14
• General Remarks - Regulators prefer the OECD QSAR toolbox - Users adopting the software ToxRead tended to provide concordant predictions - Self-declared experts, do not say that they are not sure, however they also make mistakes http://www.toxread.eu/ http://www.oecd.org/ MUTAGENICITY - The agreement among participants ranged from 60 to 89% depending on the target chemical - High false-positive rate - Participants from industry/consultancy provided the highest rate of correct predictions. - Answers from regulators were characterized by the highest rate of “not sure” answers and a tendency to provide “positive predictions” 15
BCF Improved level of agreement with respect to mutagenicity. Disagreement only in two cases with respect to the nB – B threshold (3.3). FISH ACUTE TOXICITY The disagreement among participants is higher even when the same tool was adopted. Several “not sure” LD 50 answers and the spread of answers reaching levels as high as two orders of magnitude for the same chemical 16
Conclusions • This exercise showed that there are large areas of uncertainty in read- across evaluations • In the case of mutagenicity we have noticed a larger disagreement, in particular when participants used the OECD QSAR Toolbox • Some endpoints are more consistently evaluated among experts (BCF) • The case of fish acute toxicity indicates a large discrepancy among the assessments, larger than mutagenicity and BCF • Reproducibility is endpoint and software dependent • Robustness of read-across assessments is linked to the availability of tools facilitating the overall judgment by the assessor • Regulators tend to judge their conclusions from read-across evaluations less certain than their industrial and academic counterparts 17
Thank you for your attention! Acknowledgments : This presentation was made possible thanks to the work carried out by : - the FRANCOPA research group - The CALEIDOS consortium - The participants to the CALEIDOS exercise 18
Recommend
More recommend
