RELIABILITY OF SAR PREDICTIONS FOR TTC RISK ASSESSMENT OF NEW - PowerPoint PPT Presentation

EMGS RISK ASSESSMENT SIG – TUESDAY 24 TH SEPT 2013 RELIABILITY OF SAR PREDICTIONS FOR TTC RISK ASSESSMENT OF NEW INGREDIENTS DIANA SUAREZ-RODRIGUEZ, PAUL FOWLER AND ANDREW SCOTT Safety & Environmental Assurance Centre 1

OVERVIEW  Thresholds of Toxicological Concern – their development and use  Role of in silico prediction models  Summary and Conclusions 2

WHAT IS THE THRESHOLD OF TOXICOLOGICAL CONCERN (TTC)?  A “pragmatic” risk assessment tool based on the principle of establishing human exposure threshold values below which there is no appreciable risk to human health for a chemical where specific toxicity data may be limited  Originally derived for food contact materials (Frawley 1967)  Cramer, Ford and Hall (1978) developed a decision tree that classifies chemicals on the basis of their chemical properties - Cramer Rules Bar Chart Class III  Three classes: Class II  Class I - Low concern chemicals Class I  Class II - Substances less innocuous than Class I, but don’t contain structural features suggestive of toxicity  Class III - High concern chemicals 3 Binned log(NOEL) (1)

EXPOSURE-DRIVEN RISK ASSESSMENT AND USE OF THE TTC  For food additives, a Threshold of Regulation was derived (1995) - 1.5 m g/person/day provided there are no structural alerts for genotoxicity/carcinogenicity  Munro (1996) developed generic thresholds for non-cancer endpoints using a data set of 613 compounds and their related systemic exposure data 5 th Percentile of the NOEL Cramer Class Human exposure threshold ( m g/person/day) I 1800 II 540 III 90 4

CURRENT USE OF THE TTC  Threshold of regulation adopted by FDA on food contaminants (food contact materials)  The TTC approach can be applied to low concentrations in food of chemicals with insufficient toxicity data – Adopted by JECFA on flavouring substances  TTC being investigated for cosmetics ingredients (Blackburn et al 2007, Kroes et al 2007)  Drivers:  Exposure-based risk assessment  Chemicals with insufficient data  Unable to carry out in vivo testing 5

TTC DECISION TREE - COSMETICS Decision tree taken from Kroes et al, 2007, Food Chem. Toxicol. , 45 , 2533-2562 6

EFSA 2012 & SCCS/SCHER/SCENIHR 2012 Removal of the Threshold of Expressed in Regulation (1.5 terms of kg bw/day m g/person/day) Re evaluation of Cramer class 2 EFSA (2012). Available from: http://www.efsa.europa.eu/en/efsajournal/pub/2750.htm SCCS/SCHER/SCENIHR (2012). Available from: http://ec.europa.eu/health/scientific_committees/consumer_safety/docs/sccs_o_092.pdf 7

ASSESSMENT OF IN SILICO TOOLS  TTC approach relies on in silico structural alerts to identify genotoxic or carcinogenic potential of an unknown material  In general, in silico tools such as Derek are known to perform well for mutagenicity  No guidance provided by EFSA or SCCS/SCHER/SCENIHR on what approach should be adopted to determine structural alerts  This study aimed to assess the utility of a suite of in silico prediction models as predictive tools for genotoxicity and carcinogenicity using two data sets containing Ames, in vivo MN and CARC data 8

ASSESSMENT OF IN SILICO TOOLS  A data set was compiled from publicly available (ISS) and proprietary data sets (Leadscope Enterprise)  A total of 399 compounds with data across the three endpoints ILSI DATA SET COMBINED ISS DATA SET (FDA data extracted ISS + ILSI http://www.iss.it from Leadscope) CARC CARC CARC Ames Ames Ames in vivo in vivo in vivo 381 399 214 MN MN MN SMILES SMILES cpds cpds SMILES cpds Overall Overall Overal l call call call TD 50 TD 50 TD 50 9

GENOTOXICITY DATA SET: DETAILS Endpoint Positives Negatives Equivocal Inconclusive Carcinogenicity 265 134 Mutagenicity 160 238 1 in vivo MN 151 241 2 5 Carcinogens Non-carcinogens 48% are +ve in the Ames 75% are – ve in the Ames 44% are +ve in the in vivo MN 72% are – ve in the in vivo MN 56% +ve in either the Ames or in vivo MN 10

IN SILICO PREDICTIVE TOOLS  TOXTREE version 2.5.4  DEREK NEXUS version 2.0.3  OECD (Q)SAR TOOLBOX version 3 11

IN SILICO PERFORMANCE - TOXTREE  Carcinogenicity and mutagenicity rulebase  A decision tree for estimating carcinogenicity and mutagenicity, based on the rules published in the document: “The Benigni / Bossa rulebase for mutagenicity and carcinogenicity – a module of Toxtree ”, by R. Benigni, C. Bossa, N. Jeliazkova, T. Netzeva, and A. Worth. European Commission Report EUR 23241 EN TOXTREE Positive Negative Total EXPERIMENTAL CARCINOGENICITY Positive 169 95 264* Sensitivity = 64% Negative 46 88 134 Specificity = 66% *1 carcinogen was not processed in Toxtree – Pb 2+ 12

IN SILICO PERFORMANCE - DEREK NEXUS  Knowledge-base expert system  Process against all genotoxicity endpoints: mutagenicity, chromosome damage, genotoxicity and carcinogenicity DEREK NEXUS Positive Negative Total EXPERIMENTAL CARCINOGENICITY Positive 174 91 265 Sensitivity = 66% Negative 64 70 134 Specificity = 52% 13

IN SILICO PERFORMANCE - OECD (Q)SAR TOOLBOX  Freely available tool developed by the OECD – not predicting the carcinogenicity  DNA binding profiling OECD TOOLBOX Positive Negative Total EXPERIMENTAL CARCINOGENICITY Positive 173 91 264* Sensitivity = 65% Negative 76 58 134 Specificity = 57% *1 carcinogen was not processed in OECD Toolbox – Pb 2+ 14

CONSENSUS MODELLING - PREDICTIONS  DEREK Nexus, OECD toolbox and TOXTREE  Integration of the predictions from the three models  25 carcinogens are not predicted (2 of these are metals – excluded from TTC approach)  A total of 23 carcinogens would be missed, i.e. 9% Number of Ames in vivo MN Carcinogenicity Compounds 12 7 3 Positive 1 Negative 15

SUMMARY OF CARCINOGENIC CHEMICALS MISSED BY THE IN SILICO APPROACH Number of Clastogenicity in vitro and in in silico predictivity Chemicals vivo compared with in vitro genotoxicity 2 Yes – clear positive in vitro and Would be predicted by in vitro in vivo genetic tox tests but not QSAR 1 No, but Ames positive Would be predicted by in vitro genetic tox tests but not QSAR 3 Negative in vitro assays. Negative in in vitro genetic tox Weak positive / questionable tests and also QSAR. in vivo MN assays. An evaluation of 2 of the chemicals indicated that these were negative in genotoxicity assays, which suggests they were falsely categorised. 16

CONCLUSIONS  If take worst case view 5 genotoxic carcinogens (positive in vivo MN data) were not predicted by in silico approaches Three of these were not detected by in vitro genetic tox methods   Additional 4 genotoxic and carcinogenic materials (positive Ames) with no alert in silico  9 in total  2% probability (based on this dataset) of supporting a genotoxic carcinogen (at exposures of 90 m g/person/day or above), based on Cramer classification 1.5 m g/kg bw/day 17

CONCLUSIONS  The TTC approach is a pragmatic exposure-driven risk assessment tool, and is of particular use where compound specific data may be limited  The presence of structural alerts for genotoxicity/carcinogenicity restricts the internal exposure to 0.15 m g/person/day 0.0025 m g/kg bw/day  An integrated suite of 3 in silico prediction models (DEREK Nexus, OECD (Q)SAR TOOLBOX and TOXTREE) could be useful as a screen for potential genotoxicity/carcinogenicity, with an absence of alerts used to support chemicals at higher exposure levels using the Cramer decision tree 18

ACKNOWLEDGEMENTS Nora Aptula Phil Carthew Catherine Clapp Claire Davies Paul Fowler David Mason Claire Moore Diana Suarez Rodriguez 19