Novel food safety substantiation

A novel food is defined by absence: no significant history of human consumption in the jurisdiction before a cut-off date. That absence is the whole problem. For a conventional food, safety is inferred from centuries of use by large populations; for a fermented protein, an insect flour, a cultured ingredient or a plant extracted in a new way, that inference is unavailable and has to be reconstructed from measurement.

Identity comes before toxicology

The first question is not “is it toxic” but “what is it, exactly, and will the next batch be the same thing”. A dossier stands on a compositional specification: macronutrients, the identity of the production organism where one is used, and — decisively — limits on what should not be there. For microbially produced ingredients that means the strain’s genome read for toxin biosynthesis clusters and transferable antimicrobial resistance determinants, and the fermentation screened for secondary metabolites. Batch-to-batch variability is treated as a safety parameter, not a quality one, because every downstream study was performed on one particular batch and only transfers to the others if the specification holds.

Why animal studies run out of room

The standard toxicological package — a genotoxicity battery, then a 90-day rodent feeding study yielding a no-observed-adverse-effect level, compared against estimated intake with an uncertainty factor of typically 100 — was designed for additives eaten in milligrams. It does not scale to a food eaten in grams. To apply a hundredfold factor to an ingredient someone might eat at 30 g a day, the rodent diet would have to consist almost entirely of the test material, at which point any effect observed is nutritional imbalance rather than toxicity. In practice, whole-food feeding studies are capped at a fraction of the diet, so the achievable margin of exposure is small, and the assessment leans correspondingly harder on characterisation, in vitro work and human tolerance data.

Allergenicity is predicted badly, and everyone knows it

A new protein entering the diet at scale carries the risk of becoming a new allergen. Prediction relies on bioinformatic comparison against known allergen sequences — the widely used criterion is 35 % or greater identity across any window of 80 amino acids — supported by resistance to pepsin digestion and, where a source is already allergenic, serum IgE binding tests. These tools are good at flagging a protein closely related to a known allergen and poor at anything else: de novo sensitisation cannot be predicted from sequence, and pepsin resistance correlates only loosely with allergenicity. Assessments say so, and post-market monitoring exists partly because of it.

Composition limits that only appear at volume

Some constraints are invisible at tasting scale and decisive at dietary scale. Single-cell protein is the classic instance: microbial biomass is rich in nucleic acid, and purine catabolism raises serum uric acid, which is why intake guidance for such products has long been framed around a nucleic acid ceiling rather than a protein target. Comparable arithmetic governs heavy metals in algae and seaweed, chitin and anti-nutrients in insect meals, and vitamin or mineral content that becomes an over-exposure question once an ingredient stops being a garnish and becomes a staple.