Automated predictive welfare assessment in groups of fish

Fish are widely used as research models in many different areas of science, from helping to increase our understanding of traumatic brain injury, to assessing chemical toxicity. Given that, just like mammals, fish appear capable of experiencing pain, show avoidance of unpleasant events and actively seek out pleasurable rewards, it is our responsibility to identify better ways to measure, and therefore to improve, fish welfare. However, as well as being relatively understudied, work in this area presents a considerable challenge, and some of the currently used measures have limitations that restrict their practical use. For example, some behavioural measures of poor welfare are observable at such a late stage that considerable suffering is already likely to have taken place by the time that the behaviours are noticed. What is therefore needed is a far more sensitive method that gives a much earlier indication of welfare status, giving us the chance to act sooner and therefore maximise fish welfare. In order to achieve this, we propose to develop a new way of measuring the welfare of fish: the detailed, real-time quantification of the social interactions that take place between fish living in groups. The rationale underpinning this is that social behaviours are highly sensitive to any changes that the fish perceive in their environment, both negative and positive. Therefore, by recording any unexpected changes in who interacts with who, and how often, we can be alerted to any potential threats to welfare, allowing us the chance to intervene before significant suffering has taken place. Such detailed and extensive observations of fish are likely to be time consuming, and so we will fully automate the data capturing process using the latest imaging technologies, ensuring that we produce a system that is effective, reliable, easy to use, and practical in a range of industrial and academic research environments, with the potential to improve the welfare of large numbers of fish. Although our proposal will focus on two of the most widely used species of fish, zebrafish and rainbow trout, our approach, once developed, is likely to be applicable to assessing welfare in any group-housed species (both fish and non-fish).

Fish are increasingly popular as a research model and, when combined with research revealing putative pain experiences and aversive behaviour in fish, this highlights a clear and urgent need to identify ways in which we can refine the procedures used to assess fish welfare. Existing measures are of limited value, either due to invasiveness, ambiguity or late onset; we therefore propose a novel way of assessing welfare in fish - the use of social network analysis to reveal consistent changes in social interactions within groups of fish, predictive of subsequent welfare status. Social network analysis can be used to quantify the overall structure of an animal group based on the group-wide association patterns of its constituent members, by considering the frequency, strength, type and direction of associations. Temporal analysis of these group-level metrics allows the rapid detection of deviations from group norms (e.g. relating disruption to group structure caused by the abnormal behaviour of one or more group members). Computing these metrics automatically in real-time will let us detect early changes in intra-group social interactions that can reliably predict subsequent welfare status, and potentially identify the animal(s) responsible for this change. A social network analysis approach to assessing group behaviour can detect more detail and subtlety in the mechanisms that underpin changes in social behaviour in comparison to the conventional analysis of social behaviour, and the flexibility of this approach makes it suitable not only for the assessment of fish welfare in experimental studies (i.e. for assessing the impact of particular procedures on welfare) but also for use in general housing/breeding environments (i.e. for refining and monitoring husbandry to optimise welfare), extending its potential impact to all areas of aquaculture.

The aim of this project is to validate an improved tool for welfare assessment in laboratory fish, a group which constituted 14% of the 1.93 million scientific procedures carried out on animals in the UK during 2014 (Home Office 2014). Our tool - the automated application of social network analysis to assess fish welfare - will provide us with an early warning of compromised welfare in a range of situations, allowing refinement of testing protocols to reduce animal suffering (such as the identification of humane early endpoints in fish toxicity tests) as well as the introduction of targeted interventions (e.g. in the case of social cohesion breakdown) designed to improve welfare. In addition, with animal welfare being about the induction of positive welfare as well as the avoidance of negative welfare, this new approach will also help to determine when groups of fish are experiencing positive welfare, and so could be used to identify changes to fish housing environments and management practices that result in good welfare. The flexibility of this social network approach makes it suitable not only for the assessment of fish welfare in experimental studies (i.e. for assessing the impact of particular procedures on welfare) but also for use in general housing/breeding environments (i.e. for refining and monitoring husbandry to optimise welfare), extending its potential impact to all areas of aquaculture. Once developed, this approach has the potential to be translated to any other group-housed species (fish or non-fish).