Anyone who reads the aquafeed literature regularly will recognize a familiar conclusion: “Dietary supplementation of the additive had no significant effect on growth performance.”
Enzymes, acidifiers, probiotics and emulsifiers have all been tested, and many trials report nothing. The tempting conclusion, for readers and sometimes for the authors, is that the additive simply doesn’t work.
Before accepting that verdict, ask a more basic question: was there anything for the additive to fix? Very often, a null result reflects the design of the trial rather than the value of the product. And even a well-designed trial can be undermined by statistics that don’t match the question being asked.
The Ceiling Problem
Every animal has a performance ceiling set by its genetics, its environment and its nutrition. When a diet already meets the animal’s requirements for digestible protein, available phosphorus, essential amino acids and energy, the animal is growing close to that ceiling. At that point, there is little room left for improvement.
Feed additives such as enzymes don’t create nutrients. They release nutrients that are present but locked away. Phytase frees phosphorus bound in phytate. Proteases improve the digestion of poorly available proteins and break down some antinutritional factors. Carbohydrases open up the non-starch polysaccharides (NSP) in plant cell walls, making trapped nutrients accessible. If the diet already provides everything the animal needs in available form, releasing a little more achieves nothing measurable. The animal can’t grow faster than its ceiling allows.
A trial built on a nutritionally complete basal diet is therefore set up to fail. The additive may be working perfectly at the biochemical level, but performance has no reason to change. Such a result doesn’t prove that the additive is useless. It proves only that the trial was unable to detect its effect.
Building A Trial That Can Show an Effect
A sound efficacy trial creates a measurable gap and then asks whether the additive can close it. This requires three components. The positive control (PC) is a nutritionally complete diet. It defines the performance ceiling and serves as the benchmark for full recovery. The negative control (NC) is the same diet with the target nutrient deliberately reduced, so that performance drops in a measurable way. This is the heart of the design. Without it, there is nothing for the additive to recover.
The test diets are the NC supplemented with the additive. The question then becomes clear: how much of the gap between the NC and the PC does the additive recover? Partial recovery shows efficacy. Full recovery, where the supplemented NC is statistically equal to the PC, shows that the additive can fully replace the nutrient that was removed.
The negative control must be designed around how the additive works. For phytase, the logical approach is to reduce the inorganic phosphate source, such as monocalcium or dicalcium phosphate, by around a third, leaving phytate-bound phosphorus as a major source of dietary P. If phytase then restores growth, bone or whole-body mineral content and phosphorus retention to the level of the PC, the trial has shown both efficacy and a real phosphate-sparing value, with both economic and environmental benefits.
For a protease, the challenge is usually created by replacing part of the fishmeal with plant proteins such as soybean, rapeseed or cottonseed meal, or by reducing digestible protein. Lower protein digestibility and the presence of antinutritional factors then become limiting, and the protease has a real job to do.
For carbohydrases such as xylanase, mannanase and glucanase, the NC should contain enough plant ingredients for NSP to meaningfully limit nutrient access. An enzyme can’t act on a substrate that is barely present.
For acidifiers and gut-health additives, the challenge may come from a plant-heavy diet, high stocking density, poor water quality or a controlled pathogen challenge. A healthy animal under ideal conditions gives these products little opportunity to show their value.
Practical Points That Decide The Outcome
Several practical details strongly influence whether a trial can detect an effect. First, the model must be verified. The NC has to perform significantly worse than the PC. If it doesn’t, the challenge was too mild, and the trial cannot evaluate the additive. That outcome should be reported honestly as an insufficient challenge, not as evidence that the additive failed.
Second, the additive must be matched to its substrate. Phytase in a diet with little phytate, or xylanase in a diet with few xylans, has nothing to act on, and no effect should be expected.
Third, the enzyme activity actually present in the finished feed should be measured. Pelleting, extrusion and post-conditioning can destroy a large share of enzyme activity. The dose added at the mixer is not necessarily the dose the animal receives.
Fourth, the response variables should go beyond weight gain. Apparent digestibility of dry matter, protein and minerals, nutrient retention efficiency, digestive enzyme activity, whole-body or bone minerals, and intestinal histology such as villus height often respond earlier and more sensitively than growth. They also explain the mechanism behind any growth response.
Fifth, the trial must allow enough growth. Ideally, the animals should reach three to five times their initial weight, giving differences in nutrient utilization time to show up as differences in performance.
Sixth, feeding should be controlled. Feeding to excess can mask improvements in feed efficiency, because animals simply eat more to compensate for poorer nutrient availability. A standardized or restricted feeding protocol makes differences in FCR visible. Finally, testing more than one inclusion level, rather than a single dose, greatly strengthens the conclusions.
Choosing The Right Statistics Before The Trial Begins
The statistical approach should be chosen during planning, not after the data are collected, and it must match the question the trial asks. Many real effects go undetected, and some false ones are claimed, simply because the analysis didn’t fit the design.
When the trial compares a set of independent treatments, such as PC, NC and NC plus one or more additives, a one-way ANOVA is the appropriate starting point. Many studies stop at a significant P-value followed by superscript letters from a multiple-range test, but a stronger approach is to use planned contrasts that ask the questions the trial was designed for. Does the NC perform worse than the PC, confirming that the challenge worked? Does the supplemented NC perform better than the NC, showing that the additive works? Is the supplemented NC equal to the PC, showing full recovery?
When every treatment is compared against a single control, Dunnett’s test is the appropriate post-hoc procedure. Tukey’s test is a sound choice for all pairwise comparisons. Duncan’s multiple range test, although still widely used, is liberal and increases the risk of false positives.
When the trial asks about interactions, a factorial design analyzed with two-way or three-way ANOVA is required. Interactions can be the most informative part of a trial. A design that crosses the additive (with or without) against diet type (high-fishmeal or plant-based) can show a significant additive × diet interaction.
That is powerful evidence that the additive works when there is a nutritional gap and not when there isn’t. Similarly, a 2×2 or 2×2×2 design crossing several enzymes, such as phytase, protease and xylanase each present or absent, lets the interaction terms reveal whether combinations are synergistic, simply additive or redundant. Without a factorial structure, claims of synergy are little more than guesswork. A positive control that sits outside the factorial structure is perfectly acceptable, but the factorial treatments should be analyzed on their own, with the PC compared separately through contrasts.
When an additive is tested at three or more inclusion levels, regression analysis is more informative than ANOVA. Linear, quadratic or broken-line models can estimate the optimum inclusion level and the point at which the response plateaus. When the levels are equally spaced, orthogonal polynomial contrasts offer a neat way to test for linear and quadratic trends. A conclusion such as “500 U/kg performed better than the control” is weak. A fitted dose–response curve that identifies the optimum dose is strong.
Modern trials often generate many response variables at once, including digestive enzyme activities, gene expression, immune markers and gut microbiome profiles. Multivariate methods help make sense of these data, but each has its role. Principal component analysis (PCA) is exploratory. It reveals patterns and groupings but doesn’t formally test whether treatments differ. Redundancy analysis (RDA) and canonical correspondence analysis (CCA) relate community or gene expression profiles to treatment and dietary variables. PERMANOVA provides a formal statistical test for multivariate differences among treatments. These tools add depth and mechanistic insight, but they support, rather than replace, the primary analysis of growth, feed conversion and digestibility.
Some fundamentals are frequently overlooked. The experimental unit is the tank, cage or pond, not the individual animal. Ten fish sampled from a single tank are not ten replicates. Treating them as such is pseudoreplication, and it artificially inflates the chance of finding significance. Individual measurements should be averaged per tank, or analyzed with a mixed model that includes tank as a random effect. Measurements taken repeatedly over time call for repeated-measures or mixed-model analysis. Initial body weight can be included as a covariate (ANCOVA) to reduce noise. The assumptions of normality and homogeneity of variance should be checked, and percentage data such as survival and digestibility may need transformation or a generalized linear model. Most importantly, a power analysis should be done before the trial starts.
Commercially relevant additives often improve performance by 3–10%. Detecting a 5% difference in weight gain may require five or six replicates per treatment rather than the three that are so often used. An underpowered trial is likely to report “no significant difference” even when a real and economically meaningful effect exists. Reporting effect sizes and confidence intervals alongside P-values gives readers a much clearer picture of what the data actually show.
Factorial or NOT?
The answer depends on the question. If the goal is simply to show whether an additive works, a one-way design with a properly challenged NC, a PC and planned contrasts is enough. If the goal is to understand when it works, with which ingredients, or whether combinations add value, a factorial design is essential.
The Bottom Line
An additive trial asks one fundamental question: can this product close a nutritional gap? If the design contains no gap, or the statistics can’t detect the change, the answer will always appear to be no, and that tells us nothing about the product itself.
No negative control, no proof. No positive control, no context. Wrong statistics, no conclusion.
How do you design and analyze your additive trials? Do you rely on a challenged basal diet, a factorial design, dose–response regression, or another approach? I’d be glad to hear your experience in the comments.
#MaverickMiniBytes #AquacultureNutrition #FeedAdditives #Enzymes #ExperimentalDesign #Statistics #ShrimpFarming #Phytase Alltech Aquaculture Kemin AquaScience™ ICC Animal Nutrition Weifang Kenon Biotechnology Co., Ltd Jefo Novonesis Agriculture & Industrial Biosolutions dsm-firmenich Animal Nutrition & Health Jean Fontaine Abdalla Salah, PhD INVE Aquaculture Novus International, Inc. International Aquafeed Dominique Bureau Albert Tacon Zuridah Merican Kabir Chowdhury
