The Way Forward in Non-Animal Toxicity Testing Approaches: Towards Regulatory Acceptance of (Q)SAR Methodologies

Home / New Perspectives / Emerging Technologies / The Way Forward in Non-Animal Toxicity Testing Approaches: Towards Regulatory Acceptance of (Q)SAR Methodologies

Emerging Technologies

The Way Forward in Non-Animal Toxicity Testing Approaches: Towards Regulatory Acceptance of (Q)SAR Methodologies

Bob Diderich, Organisation for Economic Cooperation and Development (OECD)

Published: February 9, 2010

About the Author(s)
Bob Diderich, Principal Administrator, Environment Directorate, Organisation for Economic Cooperation and Development.

Bob Diderich has been involved in environmental hazard and risk assessment of chemical substances since 1992, when he joined the German Federal Environmental Agency. He was working in France between 1995 and 2002, first for the French Ministry of the Environment and then the French National Institute for Industrial Environment and Risks, where he was assessing the environmental risks of industrial chemicals and biocides. In 2002 he joined the Organisation for Economic Co-operation and Development where he is now in charge of the OECD Existing Chemicals Programme and the OECD Project on (Quantitative) Structure Activity Relationships.

Bob Diderich
2, rue André Pascal
75775 Paris Cedex 16

Disclaimer: The views expressed in this article are the sole responsibility of the author and do not necessarily reflect the views of the OECD and its member countries.

The regulatory acceptance of (Q)SAR methodologies has been the biggest challenge to (Q)SAR experts over the last 20 years and the issue continues to haunt the (Q)SAR community. Why is it that so many regulatory hazard assessors are not comfortable with using (Q)SAR models, many of which have a predictivity which is comparable to experimental results with standardized test methods. The following short essay tries to identify some of the hurdles for regulatory acceptance of (Q)SAR models and what can be learnt from them, especially within the context of the further development of the OECD (Q)SAR Application Toolbox.

1. Regulatory need

Regulatory acceptance is closely linked to the question of whether a regulatory decision can be made without the use of (Q)SAR methodology. Indeed if evaluators have the means to request experimental results for all the endpoints that they need to reach a regulatory decision, they are less inclined to investigate the usefulness of (Q)SAR methodologies. On the contrary, (Q)SAR methodologies are often used to fill data gaps which are not filled via legislative testing requirements.

This can be illustrated with the new chemicals notification schemes used by the US and the EU during the 1980s and 1990s. Under the US Toxic Substances Control Act, no experimental test results are required for submitting pre-manufacture notices (PMNs). Hence the US-Environmental Protection Agency (EPA) has to estimate the properties of the new chemical to perform a preliminary risk assessment of the chemical. To be able to do their work, the EPA developed and published their own (Q)SARs such as ECOSAR or Oncologic. By necessity, the regulatory acceptance in the US of these models is good.

On the contrary, the EU Dangerous Substances Directive required notifiers to test new chemicals for many endpoints allowing the authorities to perform a robust initial risk assessment. The EU Existing Substances Regulation implemented the same requirements for priority existing chemicals. The regulatory need for (Q)SAR models was therefore low and the evolution of regulatory acceptance of (Q)SAR methods was practically at a standstill while this legislation was in force.

This divide in regulatory acceptance between the US and the EU also limited the international cooperation at OECD on this subject. The OECD (Q)SAR project in the early ’90s focused on information exchange and stopped altogether in 1995.

This situation evolved dramatically towards the end of the ’90s as the focus of many governmental review programmes shifted from the assessment of a few priority chemicals towards the assessment of large numbers of chemicals, i.e. whole inventories. Examples are the categorisation of the Canadian Domestic Substances List, The US HPV (High Production Volume) Challenge Program and finally REACH (Registration Evaluation Authorisation and Restriction of Chemicals) in the EU. Testing systematically all chemicals for a fixed list of endpoints is not an option anymore, due to cost, time and animal welfare considerations. The use of (Q)SAR methodologies is therefore today an acceptable method for regulatory data gap filling in many OECD countries and the renewed vitality of the OECD (Q)SAR project since 2002 is witness to this interest.

2. Free public availability

Experience has shown that if (Q)SAR models are freely available and easy to use, they are used. It is no coincidence if ECOSAR and EPIWIN are the most used models for industrial chemicals. The free availability of models creates a level playing field between authorities and chemical industry.

On the other hand, the use of commercial platforms can prove to be problematic, as it gives advantages to large companies that have more resources compared to small and medium enterprises. Nevertheless, workarounds are possible, as shown by the initiative of the Danish Environmental Protection Agency to create the Danish EPA (Q)SAR Database. With the help of a commercial platform, they developed numerous models and generated estimates for over 100000 industrial chemicals. These estimates were published in an internet database and are freely accessible, together with background information on the individual models.

3. Transparency

Another hurdle to the regulatory acceptance of (Q)SAR methodologies is the black box syndrome. A chemical ID is entered into a computer model and out comes the result, often without providing the background information necessary for the user to appreciate the applicability of the model for the intended regulatory purpose.

The Setubal principles and subsequently the adoption of the OECD principles for validation, for regulatory purposes, of (Q)SAR models have been crucial steps towards improving the transparency of (Q)SAR models.

The efforts of the European Commission to develop the “(Q)SAR Model Reporting Format”, based on the OECD principles as well as the “(Q)SAR Prediction Reporting Format,” have to be highlighted as important steps into the right direction. Also the database with (Q)SAR Model Reports of the European Commission will be a very valuable resource of information on (Q)SAR models which will help the users to decide whether or not a model can be used for a given purpose.

4. Communication between (Q)SAR experts and toxicologists

(Q)SAR models are developed by experts in chemo-informatics whereas regulatory decisions are proposed by (eco)toxicologists. Regulatory acceptance is therefore dependant on the ability between these two groups of experts to communicate and understand each other. This remains one of the most challenging aspects of the regulatory acceptance.

Explaining neural networks and how they can estimate complex regulatory endpoints like developmental toxicity can prove to be difficult. It is therefore crucial that these two groups of experts establish a common language and exchange views on scientific evidence needed to make a regulatory decision and the workflow used to reach such a decision. The fact that ECOSAR was actually developed by the ecotoxicologists within the US-EPA evaluating PMNs is an illustration that there is still a major divide between those two groups of experts.

It could even be considered that communication has to be extended to the judiciary experts. Indeed as Vince Nabholz (one of the developers of ECOSAR) never failed to remind us, a regulatory decision based on a (Q)SAR estimate has to withstand a challenge in court. In other words you have to be able to explain to a judge how you derived the estimate and why it is adequate for the intended regulatory purpose.

5. The chemical category approach: an illustration of all the above

The success of the chemical category approach is a good illustration of a solution to all the arguments developed above. Chemicals that are similar can be grouped into chemical categories and available experimental results from some members of the category can be used to fill data gaps of other members by read-across or trend analysis. The first guidance document on the chemical category concept was developed by the US-EPA in 1998 for the US HPV Challenge Program. It was subsequently used for the OECD HPV Chemicals Programme and substantially revised in later years based on the experience gained. The latest OECD version was published in 2007 and has been implemented in the EU REACH guidance documents.

The regulatory acceptance of the category approach – compared to classical (Q)SAR models – is phenomenal. Hundreds of datagaps including for high-end endpoints such as reproductive toxicity were filled by this approach in the OECD HPV Chemicals Programme over the last few years. The success in the US HPV Challenge Program was even more important.

Considering that building a chemical category is nothing else but building a (Q)SAR model (albeit a small and local one), there seems to be a paradox. It is easily explained by the fact that the chemical category concept overcomes all the hurdles outlined above:

  • The concept was introduced to respond to the regulatory needto assess more HPV Chemicals in a shorter timeframe.
  • It does not use any commercial tools (it is free).
  • It is fully transparentbecause the assessment report for the category contains robust study summaries for all the key test results available for all members of the category (including rationales for each test result that was rejected), detailed discussions on conclusions for each endpoint for each member of the category and descriptions of how the data gaps are filled, using the available test results.
  • The categories are developed by the (eco)toxicologists themselves, often without the help of (Q)SAR experts, thereby short-circuiting the communication problem between (eco)toxicologists and (Q)SAR experts.

The last point can also be viewed as a weakness of the practical application of the chemical category concept. Indeed for practical reasons, the chemical categories that are built for assessment in a regulatory context usually cover only a limited number of chemicals which are under scrutiny. Using (Q)SAR tools, such categories could be built in a more systematic way and cover many more chemicals. Thereby many more data gaps could be filled.

6. The OECD (Q)SAR Application Toolbox: how we should learn from the above.

The project for developing the OECD (Q)SAR Application Toolbox tries to reconcile (eco)toxicologists with (Q)SAR methodologies via the chemical category concept. Indeed the aim of the Toolbox is to allow the user to build chemical categories in a systematic manner, using freely available and transparent (Q)SAR methodologies.

The first version released in 2008 has shown that this can be done. It also overcomes already some of the hurdles outlined above. It clearly responds to a regulatory need as many countries around the world are starting to assess thousands of chemicals. It is also freely available from the internet. Furthermore it transparently documents many of the (Q)SAR tools that can be used to build categories (Profilers), both in scientific terms (how were they developed) and chemoinformatic terms (how are they coded up in the Toolbox).

Version 2.0 of the Toolbox which is scheduled to be released in October 2010, implements mainly improvements from an IT point of view, as can be expected when moving from a proof-of-concept version towards a tool with a more definitive architecture. Improvements focus on such issues as communication with other tools (e.g. IUCLID), database structure (implementation of the OECD Harmonised Templates), automated reporting as well as development of a distributed server version in addition to the stand-alone workstation version. Nevertheless, this version will contain major improvements in terms of transparency and providing the necessary information as to whether a read-across is justified or not. Indeed, using the model of the European Commission’s “(Q)SAR Prediction Reporting Format”, the user of the Toolbox will be able to automatically generate a report for any read-across outlining the relationship of the target chemical with the source chemicals in terms of structural and mechanistic characteristics, thereby providing the user with all the necessary information needed to justify the read-across to regulatory authorities.

The biggest challenge remains the issue of establishing a common (eco)toxicology – (Q)SAR language. In terms of the workflow of the (Q)SAR Application Toolbox, the problem translates into: “can we provide (Q)SAR tools that can group chemicals into (eco)toxicologically meaningful categories and that (eco)toxicologists recognise as being relevant for the regulatory endpoint for which a read-across is proposed?”

Version 1.0 has shown that this can indeed be achieved for a number of endpoints. Using mechanisms of covalent binding to proteins as a tool to build categories to fill data gaps on skin sensitisation is a good example. Indeed it is widely accepted among toxicologists that chemicals or their skin metabolites have to bind to skin protein in order to provoke skin sensitisation (it is also called the Molecular Initiating Event)). This situation could be further improved by including into the Toolbox detailed information on the mechanistic link between covalent protein binding and skin sensitisation, i.e. the information on how protein binding causes biological effects at the sub-cellular, cellular, tissue, organ and whole animal level of observation (also called Adverse Outcome Pathway). Such information on the Adverse Outcome Pathway provides assurance to toxicologists that “protein binding” is the right “descriptor” to group chemicals for skin sensitisation.

For the further development of the Toolbox, the OECD has started to gather such mechanistically-based tools to build (eco)toxicologically meaningful categories together with the information on how they mechanistically relate to the endpoint used in risk assessment. Furthermore, to foster communication between (eco)toxicologists and (Q)SAR experts, the OECD is holding expert consultations on such tools. Expert consultations have already been held on (Q)SAR tools to estimate binding affinity to the estrogen receptor and the mechanistic relationship with reproductive toxicity in fish, as well as on estimating DNA-binding and the relationship with mutagenicity. Consultations on other endpoints relevant for risk assessment will be held over the following months and years.

While version 2.0 will contain more mechanistic information for many of the profilers, the implementation of the Adverse Outcome Pathway concept will be the focus of the subsequent versions of the Toolbox. A dedicated OECD workshop on this issue is scheduled to be held during the second half of 2010.

It is clear that this is a long-term project. Developing Adverse Outcome Pathways, identifying key events along the pathways for which (Q)SARs can be developed, is not going to happen overnight, especially for such complex endpoints like repeat dose toxicity, developmental and reproductive toxicity. This also means that the Toolbox will continue to contain traditional empirical (Q)SAR models that authorities and industry consider to be useful in their assessment work.

7. Conclusion

Regulatory acceptance of (Q)SAR methodologies has been and will continue to be a slow process. However, the fact that there is now a clear regulatory need for the use of (Q)SAR methodologies in many OECD countries drastically improves the chances for success over the next few years. Furthermore, the rise of the use of information on mechanisms and mode of action in regulatory (eco)toxicology presents a formidable opportunity to bridge the communication gap between (eco)toxicologists and (Q)SAR experts. The extensive work which is going on in many countries to further develop mechanistic understanding of (eco)toxicological effects seen in vivo is a reason for optimism and the OECD (Q)SAR project is ready to meet the challenge.

©2010 Bob Diderich