Track daily zebrafish care to strengthen reproducibility

Feeding, cleaning a tank, moving a unit or netting a fish are not outside the experiment: they form part of the animal’s recent environment. A 2026 opinion article proposes that human–animal interaction and the timing of husbandry may be hidden variables in zebrafish behavioural research. A recent experimental study provides converging signals, but does not establish a universal “caretaker effect”. For research facilities, the practical response is to record meaningful interventions, harmonise sensitive procedures and connect those records to study design.

Routine care may occur before a test without being neutral

Tsang and Gerlai focus on four ordinary activities: feeding, netting, tank cleaning and water changes. Each may cause a short disturbance. If repeated, or carried out shortly before behavioural measurement, it could shift a fish’s baseline state and interact with an experimental treatment. The authors also emphasise timing: feeding hour, staff rotations, weekly schedules and the interval between an intervention and testing.

Their paper is an opinion article rather than a controlled experiment. It proposes a plausible mechanism and a research agenda. It cannot quantify a bias attributable to one person, and it does not show that changing caretakers necessarily weakens a study. Its methodological value is simpler: a variable that is not recorded cannot be balanced prospectively or examined when an unexpected result appears.

This distinction also prevents an overly rigid response. Assigning one caretaker to every fish does not solve the problem if treatment groups are handled at different times or if the person knows group allocation. Conversely, several operators may improve generalisability when their contribution is planned, balanced and documented.

A 30-day experiment offers a signal, not a universal rule

In 2026, Al-Mohammad and colleagues studied mixed-sex adult zebrafish aged six to eight months in a 30-day tank-based experiment. They compared several husbandry variations: a shortened photoperiod, repeated netting and transfer, tank relocation and caregiver variability. Outcomes included the novel tank test, whole-body cortisol, brain catecholamines and expression of genes associated with stress and plasticity.

Under their conditions, exposed groups explored the upper zone less and showed more freezing. Repeated netting and caregiver variability produced the most pronounced changes across several outcome types. Behavioural responses were accompanied by differences in cortisol, norepinephrine and gene expression.

The study supports the possibility that pre-experimental husbandry can shift baseline state. It does not define a universal recovery interval or a maximum number of handling events. It was conducted at one site, used deliberately organised exposures to compare groups and covered several forms of disturbance. “Caregiver variability” may itself include differences in movement near tanks, capture technique, timing or intervention duration. The findings should inform study design, not become an assessment of individual staff members.

Record events that can plausibly alter fish state

FELASA-EUFishBioMed recommendations already state that housing and husbandry should be described in enough detail to support both welfare and reproducibility. They favour appropriate operating ranges over one fixed value imposed across facilities. The additional challenge is to make chronology usable without creating a log that staff cannot maintain.

A proportionate record can link each tank or cohort to a short list of timestamped events:

• feeding time and diet batch, including unusual intake;
• water change, cleaning, transfer or net capture;
• a coded operator or team identifier;
• alarms, technical work, or unusual temperature and light changes;
• atypical behavioural or health observations;
• test start time and the interval since the last potentially disturbing event.

Categories should remain separate. Reduced exploration may be compatible with a disturbance response, but it does not identify the cause. A photoperiod change is not equivalent to a caretaker change, and netting is not equivalent to a routine visual check. Combining all events into a single “husbandry score” would make the result difficult to interpret.

Build husbandry into the design before data collection

The first decision is whether a protected window is needed before a sensitive measurement. Non-urgent interventions can be scheduled outside that window. Care required for welfare still takes priority and should be recorded as a deviation when its timing changes. The interval needs to be justified for the assay, life stage and intervention; it should not be copied from an unrelated protocol.

Teams can then choose among three approaches. They may standardise a procedure when the action must be comparable, such as netting and transfer. They may balance or randomise operators across groups when several people participate. Where replication allows it, they may include operator, day, tank or experimental batch in the analytical model. These decisions belong in the protocol and statistical plan, not in an explanation assembled after results are known.

Training should use observable criteria: capture duration, number of attempts, time away from the system, abrupt movements, release technique and treatment-group order. A shared demonstration followed by periodic checks is more useful than an instruction to “handle gently”. The purpose is not to rate people. It is to reduce avoidable differences and make remaining variation interpretable.

Blinding also matters. If staff can see treatment labels, consistent technique does not remove the risk of differential handling. Coded tanks, a predefined order that does not align with treatment and separation between husbandry records and outcome scoring may be appropriate. The exact safeguards should reflect the study rather than becoming a universal checklist.

Avoid over-standardising the facility

Reproducibility does not require every facility to copy the same timetable and equipment. Excessive standardisation can hide dependence on one local routine and make a protocol fragile when a staff member is absent.

It is useful to distinguish three levels: invariants required for welfare and protocol integrity, variables kept within justified ranges, and events recorded without attempting to eliminate them. A coded operator identity often belongs to the third level. It becomes scientifically useful when a design includes several people, batches or sites, but it should not become a disciplinary metric or unnecessarily exposed personal information.

Post hoc analysis also needs restraint. Finding that one operator worked more often with one treatment group does not prove that the operator caused the outcome. It can reveal an allocation imbalance and justify a prospective study, but it cannot replace one.

Conclusion: what facilities can change now

A facility can begin with a one-week process audit: which activities reach the fish, when do they occur, how much do they vary and which data already exist? The team can then add only fields that support a decision, harmonise two or three sensitive procedures and define a pre-test window where justified. A pilot period will show whether the log is complete enough for analysis without obstructing animal care.

The central message is deliberately measured. Daily human interactions are credible candidates for experimental variables, but their effects are context dependent. Recording them improves interpretation; it does not turn every intervention into a demonstrated cause.

How Vetofish can help

Vetofish can support aquatic research facilities by auditing husbandry workflows, refining handling procedures, developing practical traceability records and aligning welfare safeguards with experimental design and deviation analysis. The aim is safe care, understandable data and procedures that the whole team can apply consistently.