TEXTE 72/2013 Identification of potential PBT/vPvB-Substance by QSAR methods | TEXTE | 72/2013 Identification of potential PBT/vPvB- Substances by QSAR methods by Anna Böhnhardt Federal Environment Agency (Germany) On behalf of the Federal Environment Agency (Germany) UMWELTBUNDESAMT This publication is only available online. It can be downloaded from http://www.uba.de/uba-info-medien-e/4551.html ISSN 1862-4804 Study completed in: September 2010 Publisher: Federal Environment Agency (Umweltbundesamt) Wörlitzer Platz 1 06844 Dessau-Roßlau Germany Phone: +49-340-2103-0 Fax: +49-340-2103 2285 Email: [email protected] Internet: http://www.umweltbundesamt.de http://fuer-mensch-und-umwelt.de/ Edited by: Section IV 2.3 Chemicals Anna Böhnhardt Dessau-Roßlau, July 2013 Identification of potential PBT/vPvB Substances by QSAR methods 1 Introduction The identification and the assessment of potential PBT/vPvB substances are key tasks of the German Federal Environment Agency (UBA) under the REACH regulation. Potential PBT/vPvB substances can for example be identified using quantitative structure-activity relationships (QSARs). For a definitive assessment, experimental data are required. In this context, a QSAR screening study has been carried out in July 2010. The aim was to identify potential PBT/vPvB substances, which would afterwards undergo a further, refined assessment and for which data from registration dossiers could be checked. According to Article 23(1) of the REACH Regulation, the substances that were due to be registered until November 30th 2010 are mainly high production volume substances, but also substances that are classified due to their effects on human health and / or environment (European Union 2006). Due to their high production volume and / or their critical properties, these substances are of particular interest for PBT/vPvB assessment. Further, data from registration dossiers can be used for a refined assessment of selected substances. 2 Materials and Methods The substances examined in this study are taken from the “List of substances identified for registration in 2010” (ECHA 2010), which was compiled by ECHA based on feedback received from companies. This list of 4445 substances was downloaded from the ECHA website in July 2010 and is given in the appendix. QSAR results were generated applying the EPISUITE (US EPA 2008) and CATALOGIC (OASIS LMC 2009) software packages. 3 QSAR Models and Screening Strategy Results from the following QSAR models were calculated and used: KOWWIN (Meylan and Howard 1995;US EPA 2008), BIOWIN 2 (Howard et al. 1992;US EPA 2008), BIOWIN 3 (Boethling et al. 1994;US EPA 2008), BIOWIN 6 (Tunkel et al. 2000;US EPA 2008), OECD 301 C model (Jaworska et al. 2002;OASIS LMC 2009), OECD 301 F model (Dimitrov et al. 2007;OASIS LMC 2009). KOWWIN is a fragment-based QSAR model implemented in EPISUITE to estimate the decadic logarithm of the octanol-water partition coefficient (log KOW). The log KOW is a well-known indicator for the bioaccumulation potential of organic substances (ECHA 2008;Pavan et al. 2006;Pavan et al. 2008;Schüürmann et al. 2007) and is commonly used as a screening criterion for bioaccumulation. There are also several QSAR models available for estimation of bioaccumulation, most of which are essentially based on log KOW. If calculated log KOW values from KOWWIN are used as input for log KOW based QSARs, the results are subject to both the KOWWIN prediction error and the bioaccumulation model prediction error. For neutral nonpolar organic substances with 1 < log KOW < 6, the relationship between log KOW and log BCF is assumed to be nearly linear (Pavan et al. 2006;Pavan et al. 2008;Schüürmann et al. 2007). Consequently, both log KOW based QSARs and the log KOW itself are useful for estimating the bioaccumulation potential of these substances. For substances with large log KOW values, measurement of BCF data becomes technically difficult and the relationship between 2 Umweltbundesamt I Wörlitzer Platz 1 I 06844 Dessau-Roßlau I www.umweltbundesamt.de log Kow and log BCF is less significant because the available data scatter considerably (Müller and Nendza 2007;Pavan et al. 2006;Pavan et al. 2008;Schüürmann et al. 2007). Several mitigating factors like metabolism or molecular size were discussed to explain these findings (Müller and Nendza 2007;Pavan et al. 2006;Pavan et al. 2008;Schüürmann et al. 2007), and there are approaches to account for these factors (Dimitrov et al. 2005). However, as shown by Pavan et al., the use of current bioaccumulation QSARs is still limited by the uncertainty of these models and by the fact that for many regulatory purposes, a worst-case model is needed rather than a best-fit model (Pavan et al. 2006). This also holds true for this screening study and thus, it was decided to use log KOW as an indicator for bioaccumulation. While REACH allows to waive bioaccumulation studies for substances with a log KOW < 3 (European Union 2006), the current REACH guidance recommends log KOW > 4.5 as a screening criterion for bioaccumulative or very bioaccumulative substances (ECHA 2008). However, this screening criterion is not protective; for example, estimated and measured log KOW values for Anthracene are 4.35 and 4.45, respectively, although the substance is known to be bioaccumulative. In this study, log KOW > 3.5 was chosen as a first, coarse screening criterion to reduce the substance list before starting the more elaborate CATALOGIC calculations. Finally, it was decided to apply a screening criterion of log KOW > 4.0 to the remaining substances. The BIOWIN program is implemented in EPISUITE (US EPA 2008) and contains seven different fragment-based QSAR models for biodegradability. BIOWIN models 1, 2, 5 and 6 yield estimates of ready biodegradability in aerobic tests and differ with respect to algorithm and training set. The linear BIOWIN 1 model and the non-linear BIOWIN 2 model are both based on the Syracuse Research Corporation data set (Howard et al. 1992;US EPA 2008). Analogously, the MITI1 data set was used to derive the linear BIOWIN 5 model and the non-linear BIOWIN 6 model (Tunkel et al. 2000;US EPA 2008). Both data sets are available in the EPISUITE software (US EPA 2008). BIOWIN 3 and 4 are models for ultimate and primary degradation in the aquatic environment, respectively, based on semi-quantitative expert’s estimates (Boethling et al. 1994;US EPA 2008). BIOWIN 7 is a linear model for ready biodegradability in anaerobic tests (Meylan et al. 2007;US EPA 2008). The BIOWIN models for ready biodegradability are based on qualitative data for ready biodegradability (“biodegrades fast” or “does not biodegrade fast”) (Howard et al. 1992;Tunkel et al. 2000;US EPA 2008). The REACH guidance (ECHA 2008) recommends a combination of BIOWIN 2 and BIOWIN 3 results, or a combination of BIOWIN 3 and BIOWIN 6 results. If one of these combinations yields the result that the substance “does not biodegrade fast (probability < 0.5)” and the “ultimate biodegradation timeframe prediction: ≥ months (value < 2.2)”, the substance is considered to fulfill the screening criterion for persistency (ECHA 2008). The OECD 301 C model and OECD 301 F model are both based on a library of documented biodegradation reactions and follow a probabilistic approach, but are derived from different training sets with experimental data from OECD 301 C and 301 F tests, respectively (Dimitrov et al. 2007;Jaworska et al. 2002). An advantage as compared to the BIOWIN models is that the models are based on quantitative data (values of the biological oxygen demand (BOD)). 1 Ministry of International Trade and Industry, Japan Umweltbundesamt I Wörlitzer Platz 1 I 06844 Dessau-Roßlau I www.umweltbundesamt.de 3 Furthermore, the algorithm is based on specific reaction sites and respective documented biodegradation reactions (Dimitrov et al. 2007; Jaworska et al. 2002). Consequently, these models additionally suggest possible biodegradation pathways and their mechanistic interpretation is more straightforward than for the BIOWIN models. The reliability of QSAR results strongly depends on whether or not the predicted substance is within the applicability domain of the model. The CATALOGIC software includes an automatic algorithm to recognize whether a given substance is within the applicability domain. For the EPISUITE models, however, the applicability domain is described in the user guidance only and has to be evaluated on a case-by-case basis, which is beyond the scope of this screening study. 4 QSAR Screening 4.1 Generation of SMILES Codes for QSAR Screening To allow a fast QSAR screening of the substance list, SMILES codes (Weininger 1988;Weininger et al. 1989) are used. Computer aided generation of SMILES codes from CAS numbers is possible for a large number of substances, using the SMILESCAS database implemented in the EPISUITE software package (US EPA 2008). CAS numbers were given for 3916 of the 4445 substances on the list. Most of the substances without a given CAS number were substances of unknown or variable composition (UVCB). The 3916 CAS numbers were used as input for the EPISUITE program. SMILES codes could be generated by EPISUITE for 2432 CAS numbers. Thus, from the 4445 substances present on the initial list, 2432 substances were chosen for the following QSAR screening study. For the remaining 2013 substances, a more elaborate, probably manual approach is required to enable further computer aided screening. 4.2 Coarse log KOW Screening For all 2432 substances with known SMILES codes, calculations were carried out using the EPISUITE software package, yielding results from the different EPISUITE programs, including estimation of octanol-water partition coefficient KOW (KOWWIN) and estimation of biodegradability (BIOWIN). If available, experimental data from the PHYSPROP database implemented in EPISUITE were also given. The data set was compiled in tables for further work.