Laboratory fish: turn humane endpoints into decisions

A humane endpoint is not a general instruction to “euthanise if the animal suffers.” It must describe an observable state, state when to look, link the finding to an action and identify who can make the decision. This is challenging in laboratory fish because group behaviour, small body size and the aquatic environment can conceal deterioration in an individual.

Directive 2010/63/EU requires death as an endpoint to be avoided wherever possible and replaced by earlier, humane endpoints. It also requires prospective severity classification and reporting of severity actually experienced. A functional score sheet therefore supports both real-time refinement and retrospective learning.

Begin with expected protocol effects

Criteria should arise from the procedure, species, life stage and time course. Infection, surgery, tumour models and toxicity studies do not create the same hazards. Teams first describe plausible changes: reduced activity, isolation from the group, increased ventilation, loss of equilibrium, reduced feeding, lesions, swelling or inability to maintain a normal position.

These observations need time and context. A short reduction in swimming after handling is different from prolonged immobility on the tank floor. An ethogram should define each behaviour and observation window. Validated video examples used in training reduce observer disagreement more effectively than vague terms.

Water quality remains a major confounder. Temperature, oxygen, pH, nitrogen compounds and flow are checked before a sign is assigned to the experimental procedure. A technical failure may affect an entire rack and demand immediate system action, whereas a protocol effect may be treatment-specific. Both can occur together.

Convert signs into action levels

A short grid is often more reliable than an exhaustive catalogue. It may cover five domains: swimming and posture, ventilation, feeding, external appearance and response to the environment. Each level uses concrete descriptions rather than a global judgement. The total score is only an aid; some severe findings should trigger action regardless of the sum.

Persistent loss of equilibrium, convulsions, inability to feed, marked respiratory distress or extensive lesions may be such triggers. The FELASA/ECLAM/ESLAV report includes fish examples such as haemorrhage, skin lesions, gill defects, sustained abnormal swimming position and convulsions. These are examples, not a universal list to copy across models.

Every level needs a response and deadline: continue routine monitoring, increase frequency, apply an authorised supportive measure, request veterinary examination, remove the fish from the procedure or perform humane killing according to the approved protocol. “Check later” is not operational when no time or responsible person is named.

Group-level findings need their own escalation rules. One affected fish may require individual removal, while the same sign in several tanks may indicate a system failure or an unexpectedly severe model. The grid should therefore define both individual and population triggers, including mortality rates over a stated interval. Thresholds must be justified from pilot work, prior studies or veterinary experience and revised when the observed course differs from prediction.

Match frequency to risk

Daily observation may be suitable for a stable colony but insufficient during a period of expected rapid deterioration. Monitoring frequency increases around known critical times, after invasive procedures and after the first warning sign. Coverage must include periods with fewer staff; a limit likely to occur overnight or at weekends requires specific arrangements.

When small fish are not individually identified, re-identification may be impossible. The grid may need tank-level triggers, a gentle capture plan for individual examination and a count of affected fish. Images collected at defined intervals can support comparison but cannot replace direct welfare checks.

Do not equate immobility with absence of pain

The 2024 FELASA report on zebrafish pain identifies reduced activity, altered space use and decreased distance travelled among practical indicators. No single sign is specific. Staff seek converging changes against baseline behaviour, control groups and water conditions.

Analgesia should be planned where tissue injury is expected, with scientific justification when it cannot be used. Its effectiveness must also be assessed; administering a nominal dose does not prove relief. Anaesthesia during a procedure does not address all post-procedural effects and can itself temporarily alter swimming and ventilation.

Close the loop after the study

Actual severity and endpoint activations are compared with predictions. How many fish crossed each level? Which changes preceded deterioration? Was monitoring frequent enough? Was a trigger too late or overly sensitive? This review improves the next protocol and gives the animal-welfare body evidence for refinement.

Training should include agreement testing between observers using simulated or archived cases. Escalation rules, emergency contacts and approved humane-killing methods remain available at the work area. A technically excellent score sheet fails if the person present does not have authority to act.

Data integrity benefits from the same clarity. Record time, tank, observer, water variables, intervention and reassessment rather than overwriting an earlier score. If treatment or supportive care changes the scientific measurement, that consequence is addressed prospectively in the design; it is not a reason to leave avoidable suffering unmanaged. Deviations and exclusions should be reported transparently.

Conclusion

A good humane endpoint combines a defined sign, duration, threshold, action and responsibility. It is model-specific, reinforced during high-risk windows and updated from actual outcomes. This protects fish welfare and improves data quality by preventing uncontrolled extreme physiological states.

Vetofish can help facilities create protocol-specific grids, train observers, organise veterinary escalation and analyse retrospective severity data.