Name: Food colourings: what the evidence and the regulator splits actually show
Creator: Realised
Published: 2026-08-01

Summary

Almost every alarming claim about synthetic food colourings comes from rodents at doses far above any diet or from cell studies, almost every reassuring claim comes from a regulator rather than from human outcome data, the one real human trial mixed the dyes with a preservative so it cannot tell you which ingredient did anything, and the defensible practical conclusion is not that the molecules are hurting you but that a synthetic colouring is a reliable marker of a heavily formulated food.

Why Moderate

This entry sits at Moderate, and the tier is doing something specific here: it describes the confidence of the entry's conclusions, not the strength of any one study.

The regulatory record is well documented and directly verifiable, which is the entry's most solid material. The chemistry, the class distinctions, the exposure assessments, the statutory basis of the Red No. 3 revocation and the exact wording of the European warning are all high-confidence facts, but none of them is a health-effect finding, so none of them can raise the tier on its own.

The biological claims are weaker than they look. The immune finding, the 4-MEI tumour data and the Red No. 3 thyroid tumours are all rodent, all at doses far above dietary intake, and two of the three have known species- or mechanism-specific limits on extrapolation. The one human randomised trial is properly designed and carries a confound severe enough to prevent attributing the effect to the colours at all. The intolerance evidence is human and blinded, which is the right design, and it is largely negative, which supports a rarity conclusion rather than a harm conclusion — but the systematic review covering it declined firm conclusions on the grounds that there was too little of it, so the negative is a weak negative rather than a clearance.

That combination supports moderate confidence in the entry's negative conclusions, which is where most of its weight lies: we can say with reasonable confidence that the popular alarming claims are not supported. It does not support high confidence in a positive safety claim either, because the assessments rest on industry-supplied exposure data and two major regulators disagree. Emerging would understate the quality of the regulatory record and the Southampton trial. Strong Evidence would require human outcome data that does not exist. Moderate is the honest placement, and it is likely to stay there until the Southampton design is repeated with the colours isolated.

Practical takeaway

The evidence does not support treating colourings as toxins, and it does support treating them as a shopping signal. Both of those can be acted on without pretending to certainty you do not have.

Use the colouring as a marker, not as a hazard. If a product's ingredient list contains a synthetic colour, that is near-conclusive evidence it is a formulated product engineered to look like something, which puts it in the category the population-level diet evidence actually concerns. Acting on the category is defensible. Acting on the molecule is not.

Read the label for the class where it matters. Caramel colour appears as caramel colour, E150a, E150b, E150c or E150d. If you care about the class-specific questions above, the E number is the only place the class is disclosed, and outside Europe it usually is not disclosed at all. In practice this means the class-specific concern is largely unactionable for most shoppers, which is worth saying plainly rather than implying a vigilance that the label does not permit.

Treat the European warning as information about six specific colours. Sunset yellow, quinoline yellow, carmoisine, allura red, tartrazine and ponceau 4R carry the warning; other colours do not. If you have a child whose behaviour you are already tracking, removing those six for a defined period and watching is a cheap, low-risk personal experiment. Do it as an n-of-1 with a fixed window and an honest before-and-after, not as a permanent belief. The trial that motivated the label could not separate the colours from the preservative, so if you run the experiment, expect that a real-seeming response might be about the whole product rather than the dye.

Do not build an elimination diet on suspected dye intolerance without testing it. Only a small minority of suspected reactions reproduce when the patient is rechallenged blind. If you genuinely suspect a reaction, the useful step is a supervised challenge through a clinician, not a widening exclusion list. Elimination diets have real costs in nutrition, social life and food anxiety, and those costs land whether or not the intolerance is real.

The direction of travel is the same as everywhere else in this pillar. Whole food first, formulated food second, and the colouring question mostly dissolves as a side effect of that ordering rather than needing to be solved on its own terms. Note the parallel to the platform's position on drinks: a reformulated product that removes one ingredient is a step, not a destination, and the same logic that makes diet soda a genuine bridge off sugar without being the finish line applies to a naturally coloured version of an otherwise identical product.

Evidence detail

Why This Entry Exists

Food colourings sit at the exact point where a real regulatory disagreement gets laundered into a health claim in both directions. On one side, a genuine rodent finding about one class of caramel colour gets attached to a cola coloured with a different class, and a hazard listing in California gets reported as a cancer risk in your kitchen. On the other side, "approved and safe at permitted levels" gets used to close a conversation that regulators themselves have not closed, since the European food-safety assessor and the American one disagree in public about titanium dioxide, and the European legislature mandated a warning label on six colours that its own scientific panel declined to endorse.

The topic also has an unusually clean teaching case buried in it. In January 2025 the US Food and Drug Administration revoked the authorisation for Red No. 3 in food while stating plainly that the mechanism by which the dye produced tumours in male rats does not occur in humans, and that human exposures are far below the levels that produced the rat effect. It revoked anyway, because a 1960 statute forbids authorising any colour additive found to induce cancer when ingested by man or animal, or in tests appropriate to evaluating its safety. Whether the dose was realistic and whether the mechanism transfers to people are not part of that test. That is a legal automatism, not a new safety finding, and it was immediately reported as the agency admitting the dye causes cancer. If you can hold that one case correctly you can hold the whole topic.

There is a second reason this entry exists. Caramel colour is not one ingredient. It is four separately specified substances made by heating carbohydrates with different reactants, and the compound behind the immune-suppression story belongs to one of them and not to the one used in cola. That single correction resolves more bad internet content than anything else on this page.

What bad advice this protects against, in all directions:
• "Caramel colour in cola suppresses your immune system." Wrong by class and by dose. The compound that reduced lymphocyte counts is a constituent of Class III caramel colour (E150c). Cola-type soft drinks are coloured with Class IV (E150d). The finding is also rodent, at dietary levels around three percent of total food intake, and reversed by the end of the study. Three separate errors are stacked to produce the claim.
• "California proved 4-MEI causes cancer, that is why they had to reformulate." A Proposition 65 listing is a hazard listing with a disclosure duty attached. It states that a substance has produced cancer in some test system, not that a given product will cause cancer in a person. The underlying rodent data is split by species and sex, and the European assessor concluded the same compound is not genotoxic and raised no exposure concern. That is a live disagreement between regulators, and reporting one half of it as settled science is the most common failure on this topic.
• "The EU banned food dyes because they cause hyperactivity in children." The EU did not ban them and the label does not say that. It requires the words "may have an adverse effect on activity and attention in children" beside six named colours. The trial behind it found small effects that were not consistent across the two age groups or the two additive mixtures, and the drinks contained a preservative alongside the colours, so the trial cannot say the colours did it. Europe's own scientific panel said the findings could not justify changing the safety limits for any of the additives involved.
• "Approved at permitted levels, therefore a non-issue." This is the industry mirror of the same error, and it omits things the same regulators said. Europe's assessor set a lower individual safe-intake limit for Class III caramel colour specifically because of concern about the immunotoxicity of one of its constituents, and after refining its exposure estimates it still noted that toddlers and higher-intake adults could exceed that colour-specific limit. Approval is not the same as an absence of open questions.
• "Cut the dyes and you cut the risk." The exposure model is incomplete. 4-MEI forms during ordinary thermal browning in coffee roasting, roasted meat and soy sauce, so removing caramel colour does not remove the compound from a diet. And a manufacturer swapping to a natural colouring changes what the label says, not what kind of food the product is.
• "Tartrazine triggers asthma, especially if you react to aspirin." This was a plausible hypothesis that has repeatedly failed blinded testing. Double-blind challenge series have not reproduced dye-provoked asthma even in aspirin-sensitive people, and the systematic review of the question found the evidence too thin to draw a firm conclusion in either direction while noting that routine exclusion is unlikely to benefit anyone without proven individual sensitivity. Genuine intolerance exists and is rare; the broad version of the claim sells elimination diets and intolerance testing.

What this entry owns: the four caramel-colour classes and which compound carries which finding; the Southampton trial and the European warning label that followed it; the regulator splits on Red 3 and titanium dioxide; and the azo-dye intolerance question.

What it defers: additives in general, including preservatives, emulsifiers and the additive-avoidance question as a whole, to food_additives_to_avoid. Sweeteners of every kind to artificial_sweeteners_evidence and its siblings, including aspartame_and_the_iarc_classification, which carries the same hazard-versus-risk lesson in a better-known case. The general problem of reading a rodent or observational result as a human causal one to rct_vs_observational_evidence. Whole-diet strategy to whole_food_emphasis.

Evidence

Read the tiers, not the thesis. On this topic the pattern is almost mechanical: every alarming claim traces to a rodent study or a cell study, and every reassuring claim traces to a regulatory assessment rather than to human outcome data. Neither of those is human evidence of harm or safety in a diet. The one genuine human randomised trial in the whole literature has a confound severe enough that it cannot name the ingredient responsible. Each claim below is labelled by evidence type for that reason.

Sub-area one: caramel colour is four substances, and the classes matter

1. Caramel colour comprises four separately specified classes, and only the two ammonia-process classes carry the imidazole byproducts. (Regulatory and chemical assessment, not a health claim.) Europe's food-safety assessor re-evaluated E150a (plain), E150b (caustic sulphite), E150c (ammonia caramel, Class III) and E150d (sulphite ammonia caramel, Class IV) as four distinct substances with distinct specifications, produced by heating carbohydrates with different reactants. It estimated 4-MEI exposure specifically from Classes III and IV, using use-level data reported by industry: for caramel manufactured between 2008 and 2010, mean 4-MEI content of roughly 42 mg/kg in Class III with a reported range up to 140 mg/kg, and roughly 102 mg/kg in single-strength Class IV with a reported range up to 183 mg/kg. Both classes carry a specification ceiling for 4-MEI in EU law, and the two ceilings are not the same number, with the Class III ceiling the lower of the two; the exact figures should be read off the current specifications regulation rather than quoted from here. Caveat that matters: class membership is a manufacturing distinction, not a purity guarantee, levels vary by manufacturer and batch, the figures were supplied by industry rather than independently sampled, and the top of a reported range is routinely quoted as though it were the typical level. Citation: EFSA ANS Panel, Scientific Opinion on the re-evaluation of caramel colours (E 150a,b,c,d) as food additives, EFSA Journal 2011;9(3):2004. Cui bono: collapsing the four classes into one "caramel colour" is what lets a Class III finding be pointed at a Class IV product, and wellness publishers benefit from that collapse. But the same distinction is also used from the industry side to wave away class-specific findings without ever saying which class is in the product in front of you, which the label does not disclose.

2. The compound behind the immune story belongs to Class III, not to the class used in cola. (Rodent, high-dose dietary.) THI, a constituent of Class III caramel colour, is why Europe's assessor set a group safe-intake limit of 300 mg/kg body weight per day across all four caramel colours and then set a lower individual limit of 100 mg/kg body weight per day for E150c alone, explicitly citing concern about the immunotoxicity of a constituent of that colour. The underlying toxicology: Class III caramel colour reduced total white cell and lymphocyte counts and raised neutrophils in rats and mice fed high dietary levels, at or above roughly three percent of the diet, with THI identified as the responsible constituent, and the haematological changes reverted by the end of the study. There is no human study showing immune suppression from dietary caramel colour of any class. Evidence label: Moderate, and rodent. Citations: EFSA ANS Panel, EFSA Journal 2011;9(3):2004, for both limits and the immunotoxicity rationale; primary toxicology, Houben GF et al., Toxicity studies of Caramel Colour III and 2-acetyl-4(5)-tetrahydroxybutylimidazole in F344 rats, Food and Chemical Toxicology 1992; review, Houben GF and Penninks AH, Immunotoxicity of the colour additive Caramel Colour III, Toxicology 1994. Cui bono: the popular claim is manufactured by taking this rodent lymphocyte result, dropping the class, dropping the three-percent-of-diet dose, and aiming it at a cola coloured with a different class. Equally, "THI is not in cola" is the correct correction and is routinely used to close the entire caramel question, when the same regulator independently flagged that intake of this specific class may exceed its own lower limit in some groups.

3. 4-MEI is a real byproduct of both ammonia-process classes and is present in cola-type drinks. (Analytical chemistry and exposure assessment, not a health claim.) Europe's exposure work covers 4-MEI from Classes III and IV, and the compound has been repeatedly quantified in cola beverages, with a peer-reviewed quantitative exposure and risk assessment published in 2015. Note the precision that gets lost: unlike THI, 4-MEI is in both ammonia classes. Evidence label: Strong Evidence for presence and exposure, which is not a claim about harm. Citations: Smith TJS, Wolfson JA, Jiao D, et al., Caramel Color in Soft Drinks and Exposure to 4-Methylimidazole: A Quantitative Risk Assessment, PLOS ONE 2015;10(2):e0118138; EFSA Journal 2011;9(3):2004 for the class levels. Cui bono: exposure papers are routinely reported as harm papers, which benefits the alarm side. The counter-point that 4-MEI also forms in coffee roasting and roasted meats is true and useful, and it also works as a whataboutism that stops the reader asking whether total exposure matters at all.

4. 4-MEI produced tumours in rodent bioassays at high doses, with a split result across species and sex. (Rodent, two-year high-dose feeding.) The US National Toxicology Program's two-year feeding studies concluded there was clear evidence of carcinogenic activity in male and female mice, based on lung adenomas and carcinomas; equivocal evidence in female rats, based on increased leukaemia; and no evidence in male rats. There is no human epidemiology showing cancer from 4-MEI at dietary exposures. Evidence label: Moderate, and rodent. Citation: National Toxicology Program, Technical Report on the Toxicology and Carcinogenesis Studies of 4-Methylimidazole in F344/N Rats and B6C3F1 Mice (Feed Studies), NTP TR 535, 2007. Cui bono: "cancer-causing chemical in your Coke" is a headline, and the mouse-only, high-dose, non-genotoxic qualifiers are always the part that gets cut. From the other side, "not genotoxic, therefore fine" glides past the fact that a regulator did list it and that manufacturers did reformulate rather than carry a warning.

5. California listed 4-MEI as a carcinogen under Proposition 65 in 2011, and manufacturers reformulated. (Regulatory record.) The listing was made on the basis of the mouse data, and a no-significant-risk level was subsequently published. Major cola manufacturers changed their caramel colour for the California market to stay below the warning threshold. The framing that matters: Proposition 65 is a hazard listing with a disclosure duty. It says a substance can cause cancer in some test system; it does not say a given product will cause cancer in a person, and its listing threshold is deliberately low. Citation: OEHHA, No Significant Risk Level for the Proposition 65 Carcinogen 4-Methylimidazole, 2011, and the listing under California's Safe Drinking Water and Toxic Enforcement Act of 1986. Cui bono: "California says it causes cancer" is the single most quoted line on this topic and is a hazard statement being sold as a risk statement. The industry rejoinder that Proposition 65 warns about everything is a fair criticism of the statute that also conveniently discredits any individual listing.

6. Europe reached the opposite operational conclusion on the same compound. (Regulatory assessment of rodent and genotoxicity data.) Europe's assessor concluded that 4-MEI does not appear to cause DNA mutations and that it had no concerns about European exposure to 4-MEI from caramel colours in food. This is a direct disagreement with California's listing and is worth reporting as a disagreement rather than resolving. The two bodies are answering different questions: one asks whether dietary exposure is likely to cause harm, the other asks whether cancer has been induced in any animal test. Both answers can be correct. Citations: EFSA ANS Panel, EFSA Journal 2011;9(3):2004; EFSA, Refined exposure assessment for caramel colours (E 150a, c, d), EFSA Journal 2012;10(12):3030. Cui bono: the European clearance is the industry headline and rarely travels with the caveat in the next claim. "That regulator is captured" is the standard dismissal from the other side and is asserted rather than evidenced.

7. The same regulator flagged possible exceedance of safe-intake limits, and one caveat survived the refinement. (Regulatory exposure assessment.) The 2011 opinion concluded that anticipated dietary exposure of children and adults may exceed the limits for E150a, E150c and E150d, with exposure to E150b below its limit. The 2012 refined assessment, which covers three of the four classes rather than all four, found combined exposure to the caramel colours considerably lower, with the group limit of 300 mg/kg body weight per day not exceeded for any population group, but noted that toddlers and adults with higher Class III intake could still exceed the 100 mg/kg body weight per day limit specific to that colour. Report both halves or neither. And note what an exceedance means: these limits carry a hundred-fold safety factor below the no-effect level, so exceeding one means the margin narrowed, not that an effect occurred. Citations: EFSA ANS Panel, EFSA Journal 2011;9(3):2004; EFSA, Refined exposure assessment for caramel colours (E 150a, c, d), EFSA Journal 2012;10(12):3030. Cui bono: "children exceed the safe limit" drops the refinement and the meaning of the limit. Quoting only the refined headline drops the surviving caveat on the one class whose limit was set on immunotoxicity grounds.

Sub-area two: the hyperactivity question

8. The Southampton trial was a genuine randomised, double-blind, placebo-controlled crossover trial in ordinary children, and it found small effects on activity and attention. (Human RCT.) One hundred and fifty-three three-year-olds and one hundred and forty-four eight and nine-year-olds were recruited from the general community, not from a diagnosed population. Children drank a challenge drink containing sodium benzoate plus one of two artificial-colour mixtures, or placebo. The colours were tartrazine (E102), quinoline yellow (E104), sunset yellow (E110), carmoisine (E122), ponceau 4R (E124) and allura red AC (E129). The primary outcome was a composite hyperactivity aggregate built from observed behaviour plus teacher and parent ratings, with a computerised attention test added for the older group, and the trial reported adverse effects on that aggregate. This is the strongest human evidence in the entire food-colouring literature. Evidence label: Moderate, and human randomised. The confound is decisive and is almost always omitted: sodium benzoate was in the active drinks alongside the colours, so the trial cannot attribute the effect to the dyes rather than the preservative. The effects were small and were not consistent across the two age groups or the two mixtures, and the outcome was a behavioural composite in normal children, not a clinical diagnosis. Citation: McCann D, et al., Food additives and hyperactive behaviour in 3-year-old and 8/9-year-old children in the community: a randomised, double-blinded, placebo-controlled trial, Lancet 2007;370(9598):1560–1567, funded by the UK Food Standards Agency. Cui bono: "proven to cause hyperactivity" oversells a small average shift on a composite score from a design that cannot separate two ingredients. "Small and inconsistent, therefore nothing" undersells a properly conducted randomised trial in ordinary children that a national regulator commissioned and a major journal published.

9. Europe's scientific panel was markedly more equivocal than Europe's legislature. (Regulatory assessment of a human RCT.) In its own words, the panel found limited evidence that the tested mixtures had a small effect on the activity and attention of some children, noted the effects were not consistent for the two age groups or the two mixtures, held that the significance of those effects was unclear because it was not known whether such small changes would interfere with schoolwork or other intellectual functioning, and concluded the findings could not be used as a basis for altering the safe-intake limits of the colours or of sodium benzoate. The legislature mandated a warning label anyway. That is the honest shape of the episode: the risk assessor declined to move any safety limit, and the lawmakers required a consumer warning regardless. Citation: EFSA, statement of the AFC Panel evaluating the Southampton study on food additives and child behaviour. Cui bono: treating a legislature's precaution as a scientist's finding is the alarm side's move. Reporting the panel as having said it was fine is the industry side's misquote, since the panel explicitly said a small effect on some children was observed.

10. The European warning wording is specific, conditional, and not what people quote. (Legal and regulatory text.) Annex V of Regulation (EC) No 1333/2008 requires foods containing sunset yellow, quinoline yellow, carmoisine, allura red, tartrazine or ponceau 4R to carry the additional information "may have an adverse effect on activity and attention in children" beside the name or E number of the colour. The requirement applied from 20 July 2010. The verb is "may", the wording is about activity and attention rather than hyperactivity, and it does not say "causes". Citation: Regulation (EC) No 1333/2008 on food additives, Annex V. Cui bono: reproducing the warning as though it read "causes hyperactivity" is the most common misquote on this topic. From the other direction, many manufacturers reformulated to natural colours rather than carry the label at all, which is the strongest available evidence that industry itself treats the warning as commercially damning. Note also that sodium benzoate, present in both active mixtures in the trial, carries no such warning.

Sub-area three: the regulator splits

11. Red No. 3 was withdrawn in the US under a statute, while the agency stated the mechanism does not apply to humans. (Rodent, mechanism argued non-transferable.) On 15 January 2025 the FDA issued an order revoking the colour additive listings for FD&C Red No. 3 in food and in ingested drugs, responding to a 2022 petition. The basis was two studies showing thyroid tumours in male laboratory rats exposed to high levels through a hormone-driven mechanism. The agency stated that the way the dye causes cancer in male rats does not occur in humans, and that relevant human exposures are typically far below the levels producing the rat effect. It revoked anyway, because the Delaney Clause of the 1960 Color Additive Amendment prohibits authorising any colour additive found to induce cancer in humans or animals. Compliance deadlines are 15 January 2027 for food and 18 January 2028 for ingested drugs. Evidence label: Moderate, and rodent, in male rats only. Citations: FDA, FDA to Revoke Authorization for the Use of Red No. 3 in Food and Ingested Drugs, 15 January 2025; Federal Register, 16 January 2025, Color Additive Petition From Center for Science in the Public Interest, et al. Cui bono: "the FDA finally admits Red 3 causes cancer" inverts the agency's actual statement, and campaign groups gain a headline that reads far stronger than the finding. "It was only Delaney, there is no real risk" is accurate about the mechanism and obscures that the dye was retained for decades after the rat data existed because of how the statute was applied.

12. Titanium dioxide is a live disagreement between two major regulators. (In-vitro and rodent genotoxicity concern; regulatory split.) On 6 May 2021 Europe's assessor concluded that a concern for genotoxicity could not be ruled out, that no safe level of daily intake could be established, and that E171 can no longer be considered safe as a food additive, while noting that evidence for general toxic effects was not conclusive. Commission Regulation (EU) 2022/63, adopted 14 January 2022, withdrew the authorisation; it entered into force on 7 February 2022 with a six-month transition, so products made under the old rules could be placed on the market until 7 August 2022, and stock lawfully placed on the market by that date could remain until its date of minimum durability or use-by. The FDA continues to permit titanium dioxide in food at no more than one percent by weight. Evidence label: Emerging. The concern rests largely on nanoparticle data from cell studies and rodents, not on human outcomes. Citations: EFSA FAF Panel, Safety assessment of titanium dioxide (E171) as a food additive, 6 May 2021; Commission Regulation (EU) 2022/63 of 14 January 2022; US 21 CFR 73.575. Cui bono: "Europe banned it, so America is poisoning you" treats an absence-of-reassurance verdict as proof of harm. "The FDA still allows it" treats a decades-old listing that has not been re-examined against modern nanomaterial standards as an active endorsement.

Sub-area four: intolerance

13. Azo-dye intolerance is real and rare, the asthma link has repeatedly failed blinded challenge, and the formal evidence base is thin in both directions. (Human, double-blind placebo-controlled challenge, plus a dated systematic review.) In a double-blind, placebo-controlled crossover challenge of twenty-six atopic adults with allergic rhinitis, asthma, urticaria or pseudo-allergic reactions to anti-inflammatory drugs, tartrazine given in escalating steps to a cumulative 35 mg produced no significant skin, respiratory or cardiovascular reactions versus placebo. On asthma specifically, double-blind challenge series in aspirin-sensitive and aspirin-tolerant asthmatics have essentially failed to provoke tartrazine-induced bronchoconstriction, with at most isolated equivocal reactors, and cross-sensitivity between aspirin and tartrazine has not been demonstrated. The Cochrane review of tartrazine exclusion in asthma is the formal statement of that position, and it is weaker than it is usually quoted: because of the paucity of evidence it declined to draw firm conclusions about the effect of tartrazine on asthma control, and then noted that routine exclusion may not benefit most patients other than the very few with proven sensitivity. It rests on six small randomised trials and dates from 2001. Two further figures circulate widely and are used here only as orders of magnitude, because this entry has not traced either to a primary source: that oral rechallenge reproduces the reaction in roughly one in a hundred patients with suspected tartrazine-induced urticaria or angioedema, and that population prevalence of tartrazine intolerance sits well under one percent. Evidence label: Moderate. Caveats: challenge studies deliver a defined cumulative dose in one supervised sitting, which may sit below real-world intake spread across many products in a day, and studies of this design are small and underpowered for rare responders, so "rare" is well supported and "nonexistent" is not, and the systematic review's own position is that the evidence is too thin for a firm conclusion in either direction. Citations: Pestana S, et al., Safety of ingestion of yellow tartrazine by double-blind placebo controlled challenge in 26 atopic adults, Allergologia et Immunopathologia 2010; Stevenson DD, Simon RA, Lumry WR, Mathison DA, Adverse reactions to tartrazine, Journal of Allergy and Clinical Immunology 1986;78:182–191; Ardern KD, Ram FSF, Tartrazine exclusion for allergic asthma, Cochrane Database of Systematic Reviews 2001, Issue 4, CD000460. Cui bono: elimination-diet practitioners, intolerance-testing companies and clean-label brands all benefit from a broad intolerance narrative that blinded challenge does not support. "Proven safe" overstates the other side, since a small number of genuinely sensitive individuals exist, are hard to identify in advance, and the review that is usually quoted as clearing tartrazine actually says the evidence is too sparse to conclude.

Sub-area five: what actually survives

14. The strongest practical argument against synthetic colourings is not toxicity, it is what their presence tells you about the food. (Reasoning and synthesis, not a primary finding.) Every verified item above resolves the same way: rodent findings at doses far above dietary intake, a small and confounded human trial, an uncertainty-based withdrawal, and an intolerance that blinded challenge shows is rare. None of it supports "food dyes cause cancer" or "food dyes damage your immune system". What does survive is simpler. A synthetic colouring exists to make a food look like something it is not, which means it appears almost exclusively in formulated products, and the outcome evidence at population level lives in the ultra-processed food literature rather than in the colour-additive literature. That literature is human observational and carries its own confounding; see whole_food_emphasis and the platform's diet-foundations material for how it is sourced and how strongly it should be read. The marker is also imperfect: some minimally processed foods use natural colourings and some ultra-processed foods use none. Cui bono: this framing is the honest one and can be smuggled back toward "dyes are harmful" by implication, so the causal arrow must stay pointed at the food category rather than at the molecule. From the other side, "the dye is safe at permitted levels" is technically defensible and is precisely the sentence that lets a product avoid the question of what else is in it.

Mechanism

Three mechanisms are worth understanding, because each one explains why the corresponding scare is shaped the way it is.

The imidazoles. Heating carbohydrates with ammonia-containing reactants, which is how Classes III and IV caramel colour are made, produces imidazole byproducts as an unavoidable consequence of the chemistry. 4-MEI arises in both ammonia classes. THI arises in Class III. THI's biological action is genuinely interesting: it inhibits sphingosine-1-phosphate lyase, an enzyme in the pathway that governs lymphocyte trafficking out of lymphoid tissue. That is the same signalling pathway a licensed multiple sclerosis drug works on, though at a different molecular target — the drug modulates the sphingosine-1-phosphate receptors rather than inhibiting the enzyme. That pathway is why high dietary exposure in rodents drops circulating lymphocyte counts. It is also why the finding does not transfer: a real pharmacological mechanism operating at a pharmacological dose is not a dietary effect, and the rodent changes reversed within the study.

The rat thyroid pathway. Red No. 3 produced thyroid tumours in male rats through chronic stimulation of the thyroid by thyroid-stimulating hormone, a feedback loop that rodents are far more susceptible to than humans because of species differences in how thyroid hormone is bound and cleared. This is a well-characterised rodent-specific tumour mechanism, and the FDA said so explicitly while revoking the authorisation on statutory grounds.

Genotoxic versus non-genotoxic. This distinction decides whether "no safe dose" is a coherent statement. A genotoxic carcinogen damages DNA directly, so in principle any exposure carries some risk and no threshold can be set. A non-genotoxic one acts through some other route, typically requiring sustained exposure above a threshold before the effect appears. Europe judged 4-MEI non-genotoxic, which is why a threshold-based safe-intake figure is defensible for it. Europe could not rule out genotoxicity for titanium dioxide nanoparticles, which is exactly why it could not set a safe intake and withdrew the authorisation instead. The two conclusions look opposite and follow from the same logic.

Risks And Contraindications

The risk of over-restriction is real and is usually larger than the risk being avoided. Cutting an entire food category on the strength of a rodent finding at three percent of diet buys nothing and costs flexibility, and in children it can seed food anxiety that outlasts the belief.

Genuine dye intolerance exists. It is rare, it is hard to identify prospectively, and it should be confirmed by supervised challenge rather than assumed from an unblinded observation. Anyone with a history of urticaria, angioedema or reactions to anti-inflammatory drugs who suspects a dye should be assessed rather than self-managed.

This entry is not medical guidance for anyone with a diagnosed condition. Behaviour changes in a child that concern you are a reason to see a clinician, not a reason to run a diet experiment first. The trial behind the European label was about small average shifts on a behavioural composite in ordinary children, and it is not a diagnostic framework.

Do not read regulatory withdrawal as evidence you were previously harmed. The Red No. 3 revocation and the titanium dioxide withdrawal both happened without any new human harm data. Retrospective alarm about past consumption is not supported by anything on this page.

Watch the exposure model. Avoiding caramel colour does not avoid 4-MEI, which forms during ordinary roasting and browning. An avoidance strategy built on a single ingredient will feel more protective than it is.

Controversy

Nature of the disagreement. This is not a case of good evidence being ignored. It is a case of two regulators asking different questions and reaching different operational answers on the same data, and of a legislature acting where its own scientific panel declined to. Advocates on both sides then quote whichever body agreed with them.

Position A: synthetic colourings are an unacceptable and inadequately policed risk. Best evidence for it: the rodent findings are real, not fabricated. 4-MEI produced clear carcinogenic activity in mice in a two-year study. Class III caramel colour genuinely reduced lymphocyte counts in rodents, and Europe genuinely set a lower safe-intake limit for that class on immunotoxicity grounds. The Southampton trial was a properly conducted randomised, double-blind, placebo-controlled trial in general-population children, commissioned by a national regulator and published in a major journal, and it found adverse effects. Europe genuinely concluded that titanium dioxide could no longer be considered safe and banned it. And Europe's own 2011 exposure work said intake may exceed the limits for three of the four caramel classes, with the class-specific caveat surviving the later refinement. That is not nothing, and treating it as nothing is not scientific caution. Where it overreaches: every biological finding in that list is rodent or cell-based except one, and the one human trial cannot attribute its effect to the colours because a preservative was in the same drinks. Converting "clear evidence of carcinogenic activity in mice at high dose" into "causes cancer" is a category change, not a summary. Converting a hazard listing into a risk statement is the same move again. And "no safe level could be established" means reassurance was absent, not that harm was found.

Position B: these additives are approved, assessed and safe at permitted levels. Best evidence for it: the safe-intake limits carry a hundred-fold safety factor below the no-effect level, so even an exceedance is a narrowed margin rather than an effect. Europe judged 4-MEI non-genotoxic and raised no concern about European exposure. The FDA stated openly that the Red No. 3 rat mechanism does not occur in humans and that human exposures are far lower, and revoked only because a 1960 statute compelled it. Blinded challenge has repeatedly failed to provoke asthma with tartrazine, including in aspirin-sensitive people, and only a small minority of suspected dye urticaria reproduces on blinded rechallenge. The refined European exposure assessment cleared the group limit for all population groups. Where it overreaches: "approved and safe" is doing more work than the assessments support. The same regulator that cleared the group limit flagged that toddlers and higher-intake adults could still exceed the limit for the one class whose limit was set on immune grounds, and the exposure inputs behind those estimates were industry-supplied use levels rather than independent sampling, with the reported ranges wide enough that the top of a range is several times the mean. The systematic review usually cited as clearing tartrazine in asthma explicitly declined firm conclusions because the evidence was too sparse, which is not the same as a clean negative. "Not genotoxic, therefore fine" skips the fact that a regulator listed the compound and manufacturers reformulated. And approval says nothing about the food the additive is in.

The funding and bias dimension, cui bono, both ways. On the alarm side, food-dye fear is a durable content product. Wellness publishers, supplement sellers, intolerance-testing services, clean-label brands and additive-free product lines all convert this topic into revenue, and every one of them benefits from the dose, species and class qualifiers being stripped out. "Clean label" is a marketing category before it is a health category, and a naturally coloured version of a formulated food is a competitive position, not a nutritional improvement. On the industry side, the exposure figures underpinning the reassuring assessments were reported by the manufacturers whose products were being assessed, which is standard practice and is also a structural conflict worth naming. Beverage manufacturers reformulated for the California market to avoid a warning label, and European manufacturers reformulated to avoid the attention warning, both of which are commercially rational acts that get presented as safety improvements. And the counter-argument that Proposition 65 warns about everything is a legitimate criticism of a badly calibrated statute that also happens to discredit every listing under it, which is convenient for anyone listed. Note too that the trial which produced the European label was funded by a national food regulator rather than by industry or by an advocacy group, which is unusually clean provenance for this literature and is a point in favour of taking it seriously even while naming its confound.

Realised Position: The alarming claims about synthetic colourings do not survive contact with their own sources. The compound behind the immune story is in a different caramel class from the drink it is usually attached to, the cancer findings are rodent results at doses far above any diet with a split across species and sex, the one human trial cannot say which of two ingredients did anything, and the intolerance claim mostly fails under blinding. We will not repeat any of those as harm. We also will not say "approved, therefore fine", because the regulators disagree with each other in public, one of them set a lower limit for one caramel class specifically on immune grounds and then noted some groups may still exceed it, and the exposure inputs came from industry. What we hold is this: the honest case against synthetic colourings is not that the molecules are hurting you, it is that a synthetic colouring reliably marks a food built to look like something it is not, and the actual outcome evidence lives with that food category rather than with the dye. Act on the category. Leave the molecule as an open question, and keep every asterisk on it.

Cross-Pillar Connections

Diet. This entry is a specific case of the general additive question owned by food_additives_to_avoid, and the general strategy it points to belongs to whole_food_emphasis. The sweetener entries, particularly aspartame_and_the_iarc_classification, carry the same hazard-versus-risk structure in a case most readers already have opinions about, and reading them together makes the pattern legible. The bridge logic also transfers: a reformulated product that removes one ingredient can be a genuine step without being the destination, in exactly the way diet soda beats sugared soda as a step off sugar while water remains the finish line. And the same discipline applies to fruit: in human observational cohorts, with all the confounding that design carries, whole fruit is associated with lower type-2 diabetes risk while fruit juice is associated with higher risk, so a naturally coloured drink is not thereby a health drink.

Evidence literacy. The defining error on this topic, reading rodent or cell findings as human harm, is exactly what rct_vs_observational_evidence exists to teach. The Red No. 3 case is arguably the cleanest worked example of hazard versus risk available anywhere in the corpus and is worth using as a teaching case beyond this entry.

Immune function. The Class III caramel finding is the only immune-relevant claim here, and it is a rodent finding at roughly three percent of diet through a lymphocyte-trafficking mechanism, which places it nowhere near the practical levers described in immune_function_cross_pillar_optimisation. If a user arrives worried that their diet is suppressing their immune system, this entry's job is to close that specific worry and route them there for what actually moves immune function.

Mental. Food anxiety is a real cost of additive avoidance, and an entry that leaves a reader more afraid of an ingredient list than the evidence warrants has done net harm even if every sentence in it is true. The over-restriction risk noted above is the connection point.

What would change our mind

Falsifiability: explicit upgrade/downgrade criteria from source

• A human cohort or trial linking dietary exposure to a specific colouring, at realistic intake, to a hard outcome. Not exposure measurement, not a biomarker, not a rodent extrapolation. This would move the entry substantially and none currently exists.
• A replication of the Southampton design with the colours tested separately from sodium benzoate. This is the single most valuable missing study on the topic. If the colours alone reproduced the effect, the European label would be resting on evidence rather than on precaution, and the hyperactivity claim would move from a confounded signal to a real one. If the benzoate alone reproduced it, most of the public conversation is aimed at the wrong ingredient.
• Human data on titanium dioxide genotoxicity at dietary exposure. Europe withdrew authorisation on an absence of reassurance. Either direction of human evidence would resolve a split that is currently being argued from cell studies.
• Independent sampling of imidazole levels across caramel-colour products. The current figures are industry-reported use levels with wide ranges. Independent sampling showing substantially higher levels, or confirming that real products sit near the top of the reported range rather than the mean, would matter for the exposure side of the argument without touching the harm side.
• A properly powered modern challenge series on dye intolerance. The existing evidence is a handful of small trials and a systematic review that declined to conclude. A large blinded rechallenge series would settle a rate this entry currently reports only as an order of magnitude.
• A demonstration that the Class III immune effect appears at intakes achievable through a normal diet. The finding sits at roughly three percent of total diet in rodents. Evidence of an effect within a few multiples of realistic human intake would be a genuine change.

What would NOT move us: another regulator adding a colouring to a hazard list, since hazard listings are not risk determinations and this entry already reports two of them. A new rodent study at high dose, since we already have several and they do not resolve the extrapolation problem. Another country banning a dye, since bans follow precaution and politics as often as they follow data, and the Red No. 3 case shows an agency can revoke while stating the mechanism does not apply to humans. Mechanistic cell studies showing a colouring does something to a cell line at concentrations no gut ever sees. Testimonial or unblinded elimination-diet reports, given that blinded rechallenge reproduces only a small minority of suspected reactions. And a manufacturer reformulating, which is a commercial decision, not evidence.

Industry bias note

Structural incentives the evidence base may reflect

Who profits from alarm. Wellness publishing runs on additive fear because it is evergreen, emotionally available and requires no expertise to write. Supplement and detox sellers benefit from a generalised toxin narrative that positions their product as a remedy. Food-intolerance testing companies sell tests with poor predictive validity into a market that broad dye-intolerance messaging creates. Clean-label and additive-free brands compete on the absence of an ingredient rather than on the nutritional content of the food, which is a cheaper competitive position and one that colour fear directly subsidises. Advocacy organisations gain funding and legitimacy from regulatory wins, which creates pressure to report a statutory revocation as a scientific vindication.

Who profits from reassurance. Beverage and confectionery manufacturers are the direct beneficiaries of the approval framing, and they supplied the use-level data on which the European exposure assessments were built. That is standard regulatory practice, not misconduct, and it is a structural conflict that should be stated whenever those exposure figures are quoted. Trade associations fund and amplify the "safe at permitted levels" line. And there is a quieter beneficiary: any manufacturer whose product survives the colouring question intact avoids the more consequential question of what the rest of the formulation is.

The tell that runs through both. Almost every alarming claim on this topic is rodent or in-vitro, and almost every reassuring claim is regulatory. Neither category is human evidence about a diet. When a source presents a rodent result without the dose, a range maximum as though it were a typical level, or a regulatory clearance without the caveats the same regulator attached, that omission is the signal to check the original.

What Realised gains and does not gain. This platform sells nothing that depends on either answer. There is no supplement to move, no additive-free product line, no elimination protocol. The one bias worth naming on our own side is that the whole-food framing is the platform's general position, and an entry could drift into using a weak toxicity story to reinforce a conclusion we already hold. We have tried to avoid that by making the ultra-processed argument explicitly a piece of reasoning rather than a finding, and by sourcing that argument from the entries that carry its evidence rather than asserting it here.

Food colourings: what the evidence and the regulator splits actually show