ECHA

The REACH regulation requires registrants to collect and assess all data available for the sub­stance they intend to register under REACH with regard to substance properties, exposure, use and risk man­agement measures. This information must be documented in a registration dossier submitted to ECHA.

After collecting existing information the registrants need to identify data gaps and consider whether these gaps can be filled by using non-standard data before any new tests are conducted. This means that all available information is collect­ed: in vivo and in vitro studies, informa­tion from human exposure, information from structurally-related substances (i.e. ‘read-across' and ‘chemical categories') and predictions from valid (Q)SARs*.

When collecting and assessing data, grouping (chemical categories) and read-across approaches have proven to be very useful for filling data gaps. These approaches rely on the fact that the substances in the group have physi­cochemical, toxicological and ecotoxi­cological properties that are likely to be similar or follow a regular pattern as a result of structural similarity. The QSAR Toolbox software developed by OECD and ECHA, in close cooperation, is a valuable tool for building meaning­ful categories and applying read-across approaches. It helps registrants to fill data gaps, and to assess the (eco)toxic­ity hazards without new testing under REACH. "The QSAR Toolbox aims to make (Q)SAR methodologies more ac­ceptable by exploiting the success of grouping and building categories", says Doris Hirmann, the QSAR Toolbox project manager at ECHA. "One of the objectives of REACH is to promote the development of alternative methods for the assessment of hazards of substances. The development of the QSAR Toolbox supports this objective. Using alternative approaches also lowers the costs", Ms Hirmann continues.

The Toolbox can be used not only by the chemical industry but also by authorities and other stake­holders to identify the substances most likely to be hazardous and to verify the quality of non-test data. "At ECHA, for example, the tool supports the evalua­tion process", Ms Hirmann says.

Together with the OECD

The QSAR Toolbox project was initiated by the OECD member countries with the aim of improving the regulatory use of (Q)SAR methodologies. The first ver­sion of the Toolbox which emphasises the technological proof-of-concept was released in March 2008 with an update following soon after in December 2008. Version 2.0 was released in October 2010 and the current version 2.1 in Feb­ruary 2011. "We are now in phase two of the development process. ECHA has decided to support the further develop­ment of the Toolbox; we have signed a joint development agreement with the OECD meaning that we are the co-owners of the tool", says Ms Hirmann. OECD has the scientific lead in the development of the project and does a lot of the day-to-day management. "We contribute a lot in the IT area, and also offer our scientific knowledge", Ms Hir­mann explains.

Chemical categories

A chemical category is a group of chemi­cals which can be expected to behave in a similar manner. To explain the concept of grouping, Ms Hirmann has used a real life example with people:"First you identify a certain aim to group people, let's say I want to find people who share my dancing hobby. Then you look for certain characteris­tics like outgoing, nice and sporty from the group of people and identify those individuals who meet these characteris­tics. You can basically do the same with chemicals. You look for certain charac­teristics that relate to a certain endpoint. For example if you look for an endpoint related to genotoxicity you may want to identify those chemicals that react with DNA. Then you look for the chemicals that have experimental data available and apply a read-across. Read-across means that information on an endpoint for an untested chemical is predicted from data from a similar tested chemi­cal. The basic idea is that entire catego­ries of chemicals can be assessed when only a few are tested.

The similarities for grouping may be based on a common functional group, common constituents or chemical class­es or similar carbon range numbers, an incremental and constant change across the category or the likelihood of com­mon precursors and/or the breakdown products which result in structurally similar chemicals.

Dealing with uncertainty

How sure can one be that predictions based on the similarities of substances reflect reality? Ms Hirmann says that dealing with uncertainty is the most interesting and challenging question. "It is also interesting in terms of what you compare with. We would like to predict toxicity to human health and to the en­vironment, but we don't have a golden standard for that, so we model. What we used to do was to model with experi­mental data. Now we try to compare the predicted values to the experimental val­ues generated for example with rats, fish and daphnia. But still it's not reality; it is again a comparison to another model."

Ms Hirmann thinks that the regula­tors feel more confident with the read-across and grouping approach than with the statistical (Q)SAR approach. "In the statistical (Q)SAR prediction you have quite a lot of data and you can relate the data to a descriptor. Then you can see a trend and make a trend analysis for your substance and predict. But at the end you don't understand why. And you have outliers. It gets even more dif­ficult when the endpoint is complex. The approach is considered to be a kind of black box, where you don't know exactly what is happening." In the read-across and categories approach one deals with fewer substances but there is more information available about each sub­stance. "More attention is paid to the mechanisms inside the human body, than to the understanding of why those ten or fifteen substances belong to this one group. This gives confidence in the result", says Ms Hirmann. Ms Hirmann emphasises that statistical (Q)SAR mod­els are not bad and that there are actu­ally some that work very well: "We have received dossiers with predictions that could be accepted."

The OECD has created a list of five key validation principles for (Q)SAR models, and several guidelines on the application of (Q)SARs. These five prin­ciples and the guidelines also provide a basis for ECHA guidance and other supporting documents about the basic concepts of validity, applicability and acceptance of (Q)SAR models.

How does the Toolbox work in practice?

The Toolbox incorporates information and tools from various databases into a logical workflow. "Basically it has a workflow with six steps. First you enter your chemical into the database and retrieve the characteristics for this sub­stance. This is called profiling. After that you retrieve any data that is avail­able on your substance and/or use the information available in the tool. Then you start to build your category based on the characteristics that you identi­fied before. Eventually you will find similar substances and hopefully these similar substances have experimental data available. The next step is that you apply either read-across or trend analy­sis to fill the data gap. The last step is the generation of reports", Ms Hirmann explains. The Toolbox software can be downloaded free of charge with the vari­ous databases, stored models, profiles and rule basis from the QSAR Toolbox website http://www.qsartoolbox.org.

Ms Hirmann says that the Toolbox can be of great assistance when regis­trants prepare for the upcoming 2013 and 2018 REACH deadlines. "For the first deadline there was a lot more data available on substances. For the 2013 and 2018 deadlines substances with less data will have to be registered. This means more experimental testing, in-vitro test­ing or using these non-test methods to fill the data gaps."

"You can do a lot with the Toolbox and I encourage registrants to use it. However, it's not the most simple soft­ware. In the end, the quality of informa­tion you get out of the Toolbox depends on how you build your prediction. This requires expertise and experience", Ms Hirmann says.

Further development

The QSAR Toolbox project is appreci­ated among the OECD member coun­tries, and both the OECD and ECHA are committed to further developing the tool in a four-year collaborative project. "The development is progressing very well. One of the future challenges is to improve the methodology of the tool to be able to reduce uncertainty", says Ms Hirmann. The release of the next version is scheduled for October 2012. "Version 3.0 is envisioned to have more databases incorporated, improved rule basis and profilers to identify characteristics of chemicals, especially profilers that iden­tify specific mechanisms or modes of action."

ECHA Information toolkit

Evaluation of non-test data   The non-standard information has to be equivalent to the information obtained from the standard test data. The key point is that the non-standard data must be suitable for an adequate risk assessment to ensure the substance can be used safely and also for adequate classification for hazard communication.   Registrants have to justify these adaptations of the standard information requirements in the registration dossier and provide scientific explanations why the non-standard data is nevertheless adequate. Within this context it should be noted that industry remains responsible for assessing the intrinsic properties for hazard and/or risk assessment and classification; hence they are responsible for making the technical and scientific judgments. However, ECHA can require missing information to be provided, including tests if the data waivers or non-standard data do not meet the information needed for registration, as an outcome of the dossier evaluation processes.