Fish experiments: plan variation alongside standardisation

Define the population your result should describe

Reducing differences between tanks can make an effect easier to detect in an aquatic research facility. Yet a precise estimate from one setting does not necessarily describe what will happen elsewhere. Age, strain, production batch, microbiota and housing conditions may influence the response to the treatment under study.

Internal validity concerns the credibility of the comparison within an experiment. External validity concerns the conditions to which its conclusions can be extended. Both should be considered during planning. Reproducibility therefore involves more than repeating a procedure: it also requires a clear account of which animals and environments the result represents.

Evidence from simulated preclinical experiments

Voelkl and colleagues published a methodological analysis in 2018 based on 440 preclinical studies across 13 interventions. The source data concerned animal models of stroke, myocardial infarction and breast cancer. This was not a fish experiment and did not compare zebrafish facilities.

The researchers simulated studies conducted in a single laboratory and studies distributed across several laboratories while holding total sample size constant. Parameters derived from the published studies defined the distributions used for sampling. The analysis examined how between-study heterogeneity influenced estimates and statistical inference.

In these simulations, including several laboratories generally increased the probability that a confidence interval covered the reference effect. A substantial improvement occurred when moving from one laboratory to two. This is not evidence that two facilities are sufficient for every biological question. The finding depends on the underlying datasets, sample sizes and models.

The distinction matters when discussing fish research. The study offers evidence about a methodological problem and a possible strategy, rather than direct validation of a particular tank layout or husbandry schedule. Any application to fish needs its own justification.

Heterogeneity does not solve inadequate power

Better coverage can come with wider confidence intervals. The additional uncertainty reflects variation that a narrowly standardised sample did not include. A less precise estimate may therefore represent the intended range of conditions more honestly.

However, when statistical power was insufficient, the simulations also showed that greater heterogeneity could increase the probability of missing a real effect. Adding contexts is not an automatic remedy for an unsuitable sample size. Allocation and analysis need to be planned together before animals enter the experiment.

The experimental unit is equally important. If treatment is assigned to a tank, several fish within that tank do not necessarily provide independent treatment replicates. Increasing the number of individuals in one unit cannot substitute for independent tanks when replication at that level is required by the design.

Translate the question cautiously to zebrafish

For a study using Danio rerio, begin by defining the intended population. Is the question restricted to a particular strain and developmental stage, or should the finding extend beyond one production batch? That decision helps identify which differences are relevant and which should remain controlled.

A team might consider blocks representing several days or batches, with comparison groups represented within each block. The aim is not to vary everything simultaneously. It is to distribute identified sources of variation without making them inseparable from treatment. The exact allocation requires appropriate statistical input.

For example, studying every control on Monday and every treated fish on Friday confounds treatment with day. Distributing both groups across days produces a more interpretable comparison. This example illustrates a design principle; it is not a fish protocol validated by the 2018 simulation study.

A planned change should also have a stated purpose. Recording that several batches were used is less informative than explaining which population those batches represent and how the analysis handles their differences. Without that link, diversity in the dataset may remain difficult to interpret.

Preserve care standards while documenting variation

Planned heterogeneity is not permission to compromise water quality, neglect care or impose avoidable stress. Conditions necessary for animal health remain essential. Any scientifically relevant variation must fit the authorised project and its welfare requirements.

The protocol should distinguish factors held constant from factors deliberately varied. Records can connect each experimental unit with its date, batch, life stage, strain and relevant housing conditions. Feed changes, hydraulic incidents and operator changes should remain traceable when datasets are combined.

Randomisation and blinded outcome assessment remain important. They address sources of bias that broader sampling does not remove. Likewise, knowing the microbiological or health status of a batch may help interpret an unexpected difference without establishing that it caused the difference.

Standardised recording is particularly useful here. Deliberately varying selected conditions does not imply abandoning consistency in measurement, identification or reporting. Comparable records allow investigators to understand how a result changes across contexts instead of merely observing that it changes.

Limits of the methodological evidence

The analysis used simulations based on published studies with substantial differences between models and conditions. It did not directly test a heterogenisation scheme in an aquatic facility. Applying its reasoning to fish is therefore a methodological proposal to evaluate, not an established numerical benefit.

The authors discuss possible biases in the published evidence and the logistical demands of multi-laboratory work. Varying a small number of factors within one facility may fail to reproduce the differences found between laboratories. No universal combination of days, batches or housing conditions guarantees reproducibility.

These limits should appear in reporting alongside the result. An experiment may provide a credible answer for one defined setting while leaving broader generalisation unresolved. Recognising that boundary is more useful than claiming either universal relevance or complete failure.

Conclusion and support from Vetofish

A useful experimental plan explains both how groups will be compared and where the conclusions are intended to apply. For fish studies, planning relevant variation can support that discussion, provided replication, statistical power and animal care remain adequate.

Vetofish can help document housing conditions and review the aquatic health aspects of projects through our technical advice and support for research facilities. Statistical design should be developed with the project’s methodological specialists.